A

Project Report On

Seed Germination Prediction using

Machine Learning
(SDG-12)
SUBMITTED TO THE SAVITRIBAI PHULE PUNE UNIVERSITY, PUNE IN
THE PARTIAL FULFILLMENT FOR THE AWARD OF THE DEGREE

OF

BACHELOR’S OF ENGINEERING IN
INSTRUMENTATION ENGINEERING
BY
Dattaram Gawade (B190254610)
(B19025610)
(B190254609)
Akhsat Fulfagar (B19025609)
Under the Guidance of
Dr.A.A.Shinde

Department fo Instrumentation Engineering

AISSMS Institute of Information Technology, Pune.
2023-2024
 Department of Instrumentation

Institute Vission
“To Uplift The Common Masses By Rendering Value Added Education”.

Institute Mission
“Empowering Society Through Dynamic Education”.

Vision Of Department
To Be Recognized As One Of The Best Instrumentation Engineering Programs
By Developing Globally Competent Engineers, Researchers And Entrepreneurs To
Solve Real Life Problems Through Skill-Based Education.

Mission Of Department

M1 1: To promote learning for skill-based education and emerging technologies

to make students globally competent.

M1 2: To create conducive environment for research, innovations and entrepreneur-

ship.
 Program Education Objectives (PEOs)
1. Solve Real Life Problems By Applying The Knowledge Of Instrumentation
Technology.

2. Pursue Higher Education Or Be Researcher Or Be Entrepreneur.

3. Contribute As A Socially Responsible Citizen For The Development Of Na-

tion.

Program Specific Outcomes (PSOs)

1. Graduates Will Be Able To Apply Their Knowledge Of Measurement And
Control To Solve The Problems Related To Environment, Safety, Health And
Agriculture Sectors.

2. Graduates Will Be Able To Demonstrate Their Skills On Programmable

Logic Controller, LabView, Distributed Control System And Internet Of
Thing.
 CERTIFICATE

This is to certify that the Project Entitled

Seed Germination Prediction using Machine

Leanring
Submitted by

Dattaram Gawade (B190254610)

(B19025610)
Akshat Fulfagar (B190254609)
(B19025609)

It is a bonafide work carried out by them under the supervision of Dr. A.A.
Shinde and it is approved for the partial fulfilment of the requirement of Sav-
itribai Phule Pune University for the award of the Degree of Bachelor of
Engineering(Instrumentation), in the Acadamic year 2023-24. This project
report has not been earlier submitted to any other or university for the award of
degree or diploma.
The SDG Goal Responsible Consumption and Production focus on ensuring
sustainable consumption and production patterns. Our Seed Germination Pre-
diction project using Machine Learning falls under this SDG because it helps to
reduce waste by predicting which seeds are most likely to germinate successfully.
This ensures that resources such as water, fertilizers, and land are used more effi-
ciently.

Dr. A.A.Shinde
Project guide and Head of Department
Instrumentation Engineering

Dr. P.B. Mane

Principal
AISSMS’s IOIT, Pune
 Acknowledgement

It is our great pleasure in expressing sincere and deep gratitude towards our
guide Dr.A.A.Shinde,Associate Professor and Head of the Department of In-
strumentation Engineering Department for her stimulating guidance, continuous
encouragement and supervision throughout the course of present work. We take
the opportunity to thank, project coordinator Dr.B.N.Mohapatra and all fac-
ulty members and staff of Department of Instrumentation Engineering AISSMS’s
IOIT, Pune, for their cooperation provided to us in many ways. We would like
to thank Mr.R.S.Wadkar, Deputy Director of Agriculture Office, Pune for the
project idea, constant support, continuous guidance and interaction definitely
helped us in many ways. We are extremely thankful to our Honourable Princi-
pal, Dr.P.B.Mane, for providing us infrastructural facilities to work in, without
which this work would not have been possible. Lastly, we are thankful to all those
who have directly or indirectly supported our project work.

-Dattaram Gawade
-Akshat Fulfagar

i
 Abstract

For seed researchers, evaluating seed germination is a crucial step in assessing the
effectiveness and quality of seeds. Seed evaluations are often completed by hand,
which is a laborious, time-consuming, and prone to mistake procedure. Because
of the tremendous advancements in photography, electronics, and computer sci-
ence over the last several decades, infrared thermal imaging technology has seen
several technical advancements, clever uses for quality control and non-destructive
testing of many agro-food products. Thermal imaging is a potential non-contact
imaging technique for evaluating different quality attributes because it is based
on the infrared radiation released from target foods. Since large-scale germination
investigations often rely on human color-based threshold adjustments, they are
not well suited for conventional picture analysis methods.
Here, we are providing a machine learning method for high-throughput seed
germination tests and precise seed germination detection.

ii
Contents
Acknowledgement i

Abstact ii

1 Introduction 1
1.1 Seed Germination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Types of Seed Germination . . . . . . . . . . . . . . . . . . 2
1.2 Germination Process . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Germination Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Background of Project . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Literature Survey 5

3 Methodology 8
3.1 Hardware Requirements . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1.1 Infrared Thermal Imager . . . . . . . . . . . . . . . . . . . . 8
3.1.2 Germination Paper . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.3 Germination Chamber . . . . . . . . . . . . . . . . . . . . . 10
3.2 Software Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.1 Jupyter Notebook . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.2 Streamlit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4 Block Diagram 12

5 Design and Implementation 14

5.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.1.1 Thermal Imaging System . . . . . . . . . . . . . . . . . . . . 14
5.1.2 Active and passive thermal imaging systems . . . . . . . . . 15
5.1.3 Processing of thermal images . . . . . . . . . . . . . . . . . 16
5.1.4 Factors Influencing Thermal Image Quality . . . . . . . . . . 16
5.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2.1 Germination Test . . . . . . . . . . . . . . . . . . . . . . . . 18

6 Experimentation and Results 22

6.1 Experimentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

iii
 6.2.1 Experiment Analysis . . . . . . . . . . . . . . . . . . . . . . 24
6.2.2 Model Dashboard . . . . . . . . . . . . . . . . . . . . . . . . 26
6.2.3 Model Code . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.2.4 Code Explanation . . . . . . . . . . . . . . . . . . . . . . . . 93

7 Conclusion 110

References 110

Certification 113

Log Book 116

PO Mapping 120

Self Evaluation 123

Critical Thinking 124

SOCME 136

Paper Publication 138

Plagiarism Reports 139

Powerpoint Presenatation 156

CD 162

iv
List of Figures
1.1 Seed Germination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

3.1 Infrared Thermal Imager . . . . . . . . . . . . . . . . . . . . . . . . 8

3.2 Germination paper . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Germinatio chamber . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4 Jupyter Notebook Interface . . . . . . . . . . . . . . . . . . . . . . 10

4.1 Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5.1 Thermal Imaging System . . . . . . . . . . . . . . . . . . . . . . . . 14

5.2 Seeds are placed on germination paper . . . . . . . . . . . . . . . . 18

6.1 Day 1 Initial Thermal images . . . . . . . . . . . . . . . . . . . . . 22

6.2 Day 6 Thermal images . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.3 Day 7 Thermal images . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.4 Day 8 Thermal images . . . . . . . . . . . . . . . . . . . . . . . . . 23

v
List of Tables
6.1 Seeds Count . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

vi
Chapter 1
Introduction

Seed germination prediction using machine learning leverages advanced algorithms

to analyze and forecast the growth potential of seeds under various conditions.
By integrating vast datasets on environmental factors, seed characteristics, and
historical germination outcomes, machine learning models can accurately predict
germination rates and identify optimal conditions for seed development. This
approach not only enhances agricultural productivity but also aids in resource
management and sustainability efforts, providing a data-driven foundation for
improving crop yields and ensuring food security.
1.1 Seed Germination
The biological process of a seed turning into a sprout or a budding plant is called
germination. Everything begins when a seed, buried in the earth, comes into con-
tact with the proper elements—moisture, warmth, and occasionally light. Water
is absorbed by the seed, which causes it to grow and release stored enzymes. These
enzymes set off metabolic processes that decompose the seed’s stored nutrients.
The seed coat rips apart as a consequence, and a little root appears to look for
nutrients in the soil.

Figure 1.1: Seed Germination

A shoot simultaneously thrusts forward, toward the sun. Eventually, this sprout
will begin to unfurl its leaves, propelled by the stored nutrients. A plant’s au-
tonomous existence begins with germination, when it stops depending on the seed
and starts producing its own energy through photosynthesis.It marks the beginning
1
of a new plant’s journey towards development and maturity, much like nature’s
alarm clock.
1.1.1 Types of Seed Germination
There are two primary processes by which seeds can germinate: epigeal and hy-
pogeal. Plants have evolved several kinds of germinations as adaptations to max-
imize their development in various conditions.

Epigeal
When a seed germinates epigeally, the growing stalk emerges above the soil’s sur-
face. The cotyledons, or seed leaves, are raised into the air during this phase and
develop into the young plant’s first true green leaves. This kind of germination is
typical of plants like sunflowers and beans. The cotyledons are exposed to sunlight
for photosynthesis as the shoot rises and carries them along with it. With this
technique, the plant may effectively use soil nutrients and sunshine for its early
growth and development.

Hypogeal
When a seed germinates hypogeally, the growing branch remains below the soil’s
surface. During this phase, the stem elongates to force the first genuine leaves
above earth, leaving the cotyledons (seed leaves) buried. Plants that frequently
display hypogeal germination are peas and peanuts. Using this strategy, the cotyle-
dons act as nutrient stores for the growing seedling. This tactic shields the tender
branch from damage during the early phases of growth while enabling the young
plant to obtain vital resources.

1.2 Germination Process

The germination process is a critical phase in the life cycle of a seed, marking its
transition from dormancy to active growth. This process involves several stages,
each essential for the successful development of a new plant. Here is a detailed
look at the germination process:

1. Imbibition: The first step in germination is imbibition, where the dry seed
absorbs water through its seed coat. This water uptake causes the seed to
swell and the seed coat to soften, making it permeable to gases and soluble
nutrients.

2. Activation of Metabolism: As the seed absorbs water, its metabolic processes

are reactivated. Enzymes that were previously inactive during dormancy
become active, leading to the breakdown of stored food reserves in the seed.
These reserves, typically starches, proteins, and lipids, are converted into

2
 simpler molecules that provide energy and building materials for the growing
embryo.
3. Respiration: During germination, the seed’s respiration rate increases signif-
icantly. Oxygen is consumed, and carbon dioxide is produced as the seed
converts its stored food into energy. Adequate oxygen supply is crucial dur-
ing this stage, which is why well-aerated soil is important for successful
germination.
4. Growth of the Radicle: The radicle, or embryonic root, is the first part of the
seedling to emerge from the seed. It grows downward into the soil, anchoring
the plant and beginning the process of water and nutrient absorption. The
emergence of the radicle is often considered the definitive sign that germina-
tion has occurred.
5. Emergence of the Shoot: Following the radicle, the shoot (or plumule) begins
to grow upwards towards the soil surface. This shoot consists of the stem
and the first leaves (cotyledons). The growth of the shoot is facilitated by
the elongation of cells in the hypocotyl (the stem-like part of the seedling
below the cotyledons) and the epicotyl (the part above the cotyledons).
6. Development of True Leaves: Once the shoot reaches the surface and is
exposed to light, the cotyledons open up and photosynthesis begins. This
process provides the seedling with the energy required for further growth.
The true leaves, which will be the plant’s main photosynthetic organs, start
to develop after the cotyledons.
7. Establishment of the Seedling: As the seedling continues to grow, it becomes
less reliant on the seed’s food reserves and more dependent on photosynthesis
and nutrient uptake from the soil. The root system expands, the shoot
elongates, and the plant establishes itself as a young seedling capable of
independent growth.
Throughout these stages, various environmental factors such as temperature, light,
moisture, and soil conditions play crucial roles in influencing the success and speed
of germination. Optimal conditions vary among different plant species, making
understanding these requirements essential for effective cultivation and seedling
establishment.

1.3 Germination Rate

The germination rate is a measure of the proportion of seeds that successfully
sprout and begin to grow over a specific period. It is an important indicator

3
of seed viability and quality, influencing agricultural productivity and efficiency.
The percentage of seeds that successfully germinate in a given sample, typically
expressed over a specific time frame.
Germination rate can calculate as suppose 100 seed is taken for germination
test from a batch of 1000 seeds and that 100 seeds are put to germination. After
germination process from that 100 seeds 70 seed are germinated and 30 are not
germination , so the germination rate of that particular batch will be 70%.

1.4 Background of Project

The assessment of seed quality is typically performed by professional technicians,
relying on their individual expertise and taking considerable time. To simplify seed
grading for producers, researchers have developed machine vision-oriented methods
to monitor seed quality attributes such as texture, color, and form. These models
enable automated and rapid acquisition of relevant variables. However, the results
of seed grading, whether through automated computer classification or technician
assessment, must be validated by the conventional germination test.
This germination test usually takes two to ten days, depending on the seed type,
and requires expert personnel to manually count germinated seeds. If the results of
the samples fall outside the acceptable range, the test must be repeated, demanding
significant effort from skilled experts. Due to the unique nature of germination
tests and their dependence on skilled professionals for accurate classification, there
is a pressing need for a reliable, automated system to examine seed germination
with greater stability, independence, and accuracy across species.
The automated germination test monitoring system in the study now employs
traditional image analysis techniques. The conventional method for testing seed
vigor is labor-intensive and challenging, leading to its infrequent use. Additionally,
many seed tests established by the International Seed Testing Association (ISTA)
are manually assessed using protocols that vary by crop. Researchers have devel-
oped automatic methods to minimize errors and reduce manual steps in the seed
testing process. Modern image analysis techniques, being automated, produce
quantitative, unbiased measures with minimal errors and have thus been utilized
to detect seeds.

1.5 Objectives
1. To reduce the number of manual steps in seed testing, which is highly error-
prone.
2. To automate the process of seed germination testing.

4
Chapter 2
Literature Survey

In this chapter, a brief overview of the current germination classification model

is presented.

1. Here Nikita Genze et al. [1] introduce a machine learning approach for
high-throughput seed germination studies and reliable detection using mod-
ern artificial neural networks with region proposals. They compiled tagged
image data on the germination stages of over 2,400 seeds from three dif-
ferent crops: Zea mays (maize), Secale cereale (rye), and Pennisetum glau-
cum (pearl millet), utilizing approximately 23,000 images in total. Transfer
learning was employed to train various state-of-the-art convolutional neural
network (CNN) architectures with region proposals to automatically iden-
tify seedlings in petri dishes and predict their germination potential. There
suggested machine learning-based approach can facilitate the evaluation of
seed germination. Experiments with several seed varieties. In comparison
to traditional and manual approaches, it has reduced error rates and higher
performance, resulting in more precise germination indices and seed quality
evaluations.

2. In this study Srinath Yasam , et al. [2] the creation of the Automated
Germination Classification utilising ORB Feature Extractor with Machine
Learning (AGCORBML) model is the main emphasis of this study, according
to Srinath Yasam, et al. [2]. For the purposes of determining seed quality
and detecting
germination, the proposed AGCORB-ML approach employs hand-crafted
features and ML classifiers. The recently developed AGCORB-ML approach
uses the ORB technique to extract a suitable set of feature vectors from
the germination images. To categorize germination, two machine learning
models—the support vector machine (SVM) and the extreme learning ma-
chine (ELM)—are used. An comprehensive range of trials were conducted
to show the AGCORB-ML technique’s effectiveness in germination catego-
rization. According on the simulation results, the AGCORB-ML approach

5
 outperforms existing ML models in a promising manner.

3. In this study, Amjan Shaik, et al.[3] designed an MLA to extract pertinent

attributes of a seed that results in knobging the inconsistent targets, i.e.,
it minimises cardinality of features and uniqueness is kept, which helps the
farmer choose that specific seed. The seed’s chosen characteristics are then
used for training and testing. 40% of the data is used for testing, while
60% is used for training. To show the effectiveness of the chosen features of
a seed, three MLA classifiers—Decision Tree, SVM, and KNN—were used
in total. To test these algorithms, a big, well-known dataset of benchmark
seeds is employed in the experiment. The seed used in this study is from
a dataset of three different wheat varieties: kama, rosa, and Canadian. On
these datasets, the MLA has been used to control the pertinent features that
describe the growth of seed, and a classification technique has subsequently
been used to show the accuracy of the features that have been chosen. There
goal in using this effective strategy is to assist farmers in choosing quality
seeds for the development of the nation’s agriculture and economy while
saving them money, time, and energy.

4. 4.The development of technologies for managing the germination process

and enhancing seed quality is driven by the aim to increase agricultural out-
put.The three-stage simulated method utilized in this study assigns a good,
poor, or neutral grade to germination. A machine learning-based approach
for classifying seed germination is provided by M. R. Kumar et al. [4]. Every
seed chosen for the study is dynamically analyzed using an evaluation tech-
nique known as PBSGA (Pivot Based Seed Germination Evaluation), which
is based on statistical methods. Two scenarios are used to train and test the
model. Before comparing different baskets from the KAGGLE dataset, a ma-
chine learning method for determining the quality of rice seeds (RSGA) is
used. It demonstrates the effectiveness of the PBSGA recommended method
in selecting the suitable model category and conceptualizing successive im-
provements, such as the difference in quality in testing batches that were
conducted sooner and later. The experimental study’s findings provide an
indication of the model’s potential and the direction that multiclass labeling
research will take going forward.

5. 5.Nehoshtan, Y., et al. present the first-ever general germination predic-

tion method here [5]. It makes it simpler to categorize seeds based on their
usefulness and capacity to germinate—two characteristics that influence the
outcome of germination—and is based on deep learning and RGB image data.
By making hundreds of rejected seed batches from seven different vegetable
crops—representing different genotypes and production pipelines—industrially

6
acceptable, they exhibit their technological expertise. We also demonstrate
the ability to accurately categories lots using crop-level picture data rather
than lot-specific data. These successes mark a significant turning point in
the application of this technique for industrial seed sorting by germination
outcome for various crops. There suggested method builds on prior attempts
to use optics to forecast seed germination efficiently and without harm, but
unlike them, it makes use of colour imaging and neural network classifica-
tion. In a cohort of seven distinct crops, dispersed across a wide phylogeny
and represented by at least two lots, bred and produced by twelve distinct
enterprises, we used it to robustly enrich seed germinability and usability,
two distinct features of germination, to 90% precision. This new capability is
anticipated to allow seed companies to separate seeds of various crops based
on their germination potential for the first time in high-fidelity, in a single
separation process, and without obligatory seed priming. This will stream-
line the process of seed germination enrich- ment and make lots that, using
modern separation techniques, would have failed enrichment attempts, such
as the lots used in this study, industrially useful. Unfortunately, they only
know the cause of poor germination for a small number of the lots, which
was either genetics production, or prolonged storage. Those lots encompass
the potential causes of mal-germination, and their classification satisfied the
requirements of 90% precision and 80% recall, proving the method’s adapt-
ability to various causes of mal-germination.

7
Chapter 3
Methodology
To predict seed germination using machine learning, we integrate advanced hard-
ware and software tools to create a comprehensive and efficient methodology. The
hardware setup includes an infrared thermal camera to monitor temperature vari-
ations during germination, germination paper to provide a consistent medium, and
a controlled germination chamber to maintain optimal environmental conditions.
These tools enable the precise collection of thermal images and environmental data,
which are essential for accurate predictions. The software aspect involves using
Jupyter Notebook for data preprocessing, feature extraction, and model training,
leveraging Python libraries like OpenCV, pandas, and scikit-learn. Additionally,
Streamlit is employed to develop an interactive web application, allowing users to
input new data and visualize predictions in real-time.
3.1 Hardware Requirements
3.1.1 Infrared Thermal Imager

Figure 3.1: Infrared Thermal Imager

8
Features:

ˆ TFT full view color screen display.

ˆ Dual MCU processors, refresh faster.

ˆ Users may navigate to the hot and cold spots in a thermal picture by using
the hot and cold point temperature indicators.

ˆ Selectable palette

ˆ hybrid imaging using both infrared and visible data.

ˆ Images capture saved in SD.

ˆ Connecting PC with USB can read the image stored in instrument directly.

3.1.2 Germination Paper

Figure 3.2: Germination paper

This specific kind of paper is used as a substratum in the germination test. This
paper is ideal for checking the ideal moisture content of the different types of seeds
and germination forms. Its unique structure and exceptional wet strength assist
to inhibit the growth of fine seed roots. The provided seed germination paper has
a high capacity for absorption and a pH range of 5-7 with strong bursting power.
The primary criterion used to assess a seed’s viability, vigour, and outward appear-
ance is its germination count. With seed germination paper, the optimal moisture
content is ensured for the greatest range of seeds and germination forms.

9
3.1.3 Germination Chamber

Figure 3.3: Germinatio chamber

An enclosed space with adjustable humidity and temperature is called a germi-

nation chamber. Often, trays are stacked so that it is easy to add and remove
trays and to see which ones have germinated. Without entering the field, precise
environmental conditions are produced within the chamber to conduct various
experiments.

3.2 Software Requirements

3.2.1 Jupyter Notebook

Figure 3.4: Jupyter10Notebook Interface

Interactive notebook papers with real-time code, equations, graphics, video, and
other computational outputs can be created with Jupyter Notebook. Program-
mers, data scientists, and students frequently use Jupyter Notebook to record and
illustrate coding processes or just to play around with code.
 One kind of integrated development environment (IDE) is Jupyter Notebook.
IDEs are code writing environments with certain auxiliary capabilities. Debugging,
code completion, and syntax highlighting are features found in almost all IDEs.

3.2.2 Streamlit
With the help of the open-source Python module Streamlit, developers can quickly
create eye-catching user interfaces. The simplest method for integrating code into
a web site, particularly for those without any prior front-end experience, is to use
Streamlit: There is no need for prior front-end (html, js, css) expertise or experi-
ence. A free and open-source framework called Streamlit makes it easy to create
and distribute stunning online applications for data science and machine learning.
It is a Python-based library created with machine learning developers in mind. on-
line developers are neither data scientists or machine learning engineers, and they
have little interest in spending weeks learning how to create online applications
using these frameworks. As long as the tool can present data and gather the neces-
sary parameters for modelling, they would prefer something simpler to understand
and operate. Streamlit helps you to develop a stunning-looking application with
only a few lines of code.

11
Chapter 4
Block Diagram
The block diagram for seed germination prediction using machine learning outlines
a systematic workflow integrating both hardware and software components.

12
Figure 4.1: Block Diagram
 Initially, the hardware setup includes an infrared thermal camera capturing
temperature profiles of seeds on germination paper within a controlled germina-
tion chamber, ensuring consistent environmental conditions. The captured thermal
images and environmental data are then fed into the preprocessing block, where
image processing techniques extract relevant features and synchronize them with
environmental parameters. These preprocessed features are input into the machine
learning model development block, where algorithms like decision trees, random
forests, and neural networks are trained and validated using labelled data. The
final block involves deploying the trained model via a Streamlit-based interac-
tive web application, allowing users to input new data, visualize predictions, and
make informed decisions about seed germination. This comprehensive diagram
illustrates the seamless integration of hardware data acquisition with advanced
machine learning techniques and user-friendly software interfaces.
To conduct the seed germination process, the methodology involves several
steps. Firstly, three germination papers are acquired and thoroughly wetted to
ensure adequate moisture for seed germination. Next, soybean seeds are care-
fully placed onto two of the wetted germination papers, with appropriate spacing
between each seed. These seeds are then covered with the third germination pa-
per, forming a sandwich-like structure to promote germination. Subsequently, the
prepared seeds are left undisturbed for a duration of 8 days to initiate the ger-
mination process under controlled environmental conditions. After the incubation
period, thermal images of the soybean seeds are captured using specialised thermal
imaging equipment. These thermal images serve as valuable data to discern the
germination status of the seeds.
The next step involves organising the thermal images into a comprehensive
dataset, meticulously labelling each image according to the germination status of
the corresponding seed—whether germinated or dead. This dataset compilation
is crucial for training and testing deep learning algorithms. Finally, deep learning
algorithms, such as convolutional neural networks (CNNs), are employed to classify
the seeds based on the thermal images. Through meticulous training, validation,
and testing processes, the deep learning model can effectively distinguish between
germinated and dead seeds with high accuracy. This comprehensive methodology
ensures a systematic approach to analyse and classify soybean seeds’ germination
status, offering valuable insights for agricultural research and practice.

13
Chapter 5
Design and Implementation
The design and implementation of a seed germination prediction system using
machine learning involves integrating thermal imaging systems, both active and
passive, to capture comprehensive temperature profiles of seeds during the germi-
nation process. In the design phase, we employ a thermal imaging system where
infrared cameras continuously monitor the seeds on germination paper within a
controlled environment. Both active thermal imaging, which involves external
heat sources, and passive thermal imaging, relying on natural heat emission from
the seeds, are utilized to ensure accurate data collection. The processing of ther-
mal images includes extracting relevant features that reflect germination activity,
such as temperature patterns and heat distribution. Factors affecting the quality of
thermal images, such as camera resolution, environmental stability, and seed place-
ment, are meticulously managed to maintain data integrity. In the implementa-
tion phase, these processed thermal images, along with environmental parameters,
are fed into machine learning models to predict germination outcomes. The sys-
tem leverages software tools like Jupyter Notebook for data processing and model
training, and Streamlit for developing an interactive application, thus providing a
robust and user-friendly platform for accurate seed germination prediction.
5.1 Design
5.1.1 Thermal Imaging System

Figure 5.1: Thermal Imaging System

14
 The traditional infrared thermal imaging system for assessing seed quality com-
prises two primary components: (1) a thermal camera with optical units, including
a focused lens, collimating lens, and filters, and (2) a thermal detector array that
features microbolometers with temperature stabilization or cooling mechanisms,
along with a control and display unit equipped with signal and image processing
tools. To accurately record the infrared radiation emitted by the seeds, the system
must function with high precision and resolution, requiring further refinement of
its components.
It’s also important to remember that the thermal imaging system lacks a light-
ing element, in contrast to the other color or spectrum imaging systems. The
signal processing unit’s job is to translate the incoming electrical signals into ther-
mal records that show the target object’s surface temperature distribution in the
scene. As a result, the thermal image created by this technique is a matrix of sev-
eral color levels, each of which represents a distinct temperature. Thermal imaging
technology scans at a pace of around 30 images per second with a sensitivity range
of -20 to 1500°C while evaluating seed quality. By utilizing certain optical filters,
its sensitivity can be increased even further.

5.1.2 Active and passive thermal imaging systems

In order for thermal imaging systems to be used practically, the target object’s
temperature has to vary from its surroundings. This means that before the imaging
begins, the item must be either an energy source or a sink, or it must be heated
or cooled. Furthermore, it is crucial.
To take into account the possibility that the target items’ temperature may
fluctuate while being observed. Essentially, there are two methods for employing
thermal imaging to discern objects: either using the varying heat conductivities
or capacities of the object materials, or using differing emissivities. Infrared light
emitted from two objects can be used to directly differentiate them if their inherent
emissivity coefficients differ. As an alternative, covering the items with a heat pulse
and because of the variations in their heat conductivities, it is possible to
distinguish between things by monitoring the radiation emitted from them.

As a result, the two fundamental designs for developing an infrared thermo-

graphic imaging system in seed quality monitoring are active thermal imaging and
passive thermal imaging, which both rely on the emissivity concept.
Imaging method (which is based on the theory of heat conductivity). When
the system is in a passive state, no external energy is used to monitor the tar-
get body’s thermal radiation emissions under ambient circumstances. When the
items being evaluated naturally produce heat, passive thermography is typically
employed. Thus, for non-contact temperature and to characterize the surface

15
thermal characteristics of the items under research, the passive-based approach is
typically utilized.
Measurement in the course of processing. Additionally, as insects in seed lots
produce a lot more heat through their respiration than do sound seeds, it might
be utilized to identify insect infestation in seeds.
In the second method of thermal imaging, emissivity often fails to produce
infrared images with sufficient contrast. Therefore, it is advisable to heat or cool
the samples before capturing thermal image frames. It is advised to use an in cases
when it is not feasible to collect enough thermal signals.
Outside energy source in order to provide a sufficient thermal contrast. In
these situations, a thermal imaging technique known as active thermal imaging
is employed, in which the target items are heated or rapidly cooled in order to
deposit external energy before taking an image.

5.1.3 Processing of thermal images

The true potential of any imaging system lies in its ability to provide quantitative
data on the spatial variation of the measured quantity. Thermographs can inde-
pendently describe various characteristics of the imaged objects or indirectly link
thermal data to hidden attributes. To accurately evaluate surface temperature
data captured in thermal imaging, one must understand all the processes that can
affect an object’s temperature. A thermograph taken by a thermal camera can be
analyzed using various image processing techniques. The choice of technique for
thermal imaging data analysis depends on the application’s primary goal and the
required speed of decision-making. In industrial settings, the processing method
for thermal image analysis must be quick and straightforward to enable real-time
task completion. Current thermal imaging cameras come pre-programmed with
processing methods to estimate the temperature of target objects in the scene.

5.1.4 Factors Influencing Thermal Image Quality

When using thermal imaging for real-time, contactless temperature measurement,
objects within the camera’s field of view must emit radiation that reaches the
detector at a detectable intensity. Infrared radiation must travel from the object’s
surface to the detector inside the camera housing. However, this radiation is often
weakened during its journey due to environmental factors and the materials of the
focusing optics and filters it passes through. Understanding the scattering and
absorption processes in the atmosphere is crucial for calculating the attenuation
of infrared radiation, which depends on the concentration of absorbing gas species
and the relative humidity. Several variables significantly influence both the capture
and interpretation of infrared thermal images. Although it might be challenging

16
to regulate every one of these elements in reality, understanding them is crucial
for many application schemes.
Since the main objective of thermal imaging is to record the surface tempera-
ture of every object in the camera’s field of view based on the thermal radiation
emitted from these objects, factors that attenuate or augment thermal radiation
before it reaches the detector should be carefully considered. For instance, one
way that specific air parameters, including temperature, relative humidity, and
wind speed, affect the transmission of infrared light from the target object to the
thermal imaging system is by attenuation through absorption or scattering.

17
5.2 Implementation
5.2.1 Germination Test
The warm germination test serves as the standard method for assessing the viabil-
ity of seed batches. In compliance with the Nebraska State Seed Law, commercially
marketed seed must display its germination percentage on the seed tag. This test
subjects the seed to optimal conditions, including ideal light, temperature, and
relative humidity, which vary depending on the plant species under examination.
For conducting the warm germination test on large-seeded crops like soybean and
corn, the necessary supplies include a seed counter tray, a seed-retaining container,
and two moistened paper towels.

Figure 5.2: Seeds are placed on germination paper

The seed counter tray is filled with seeds, ensuring each hole is filled and excess
seed is drained. Upon opening a sliding drawer, fifty seeds are dispensed onto the
paper towel, ensuring uniform placement after removing the counter tray. The
seeds are then covered with a thick and a thin, moistened paper towel, with their
edges sealed together to prevent drying out. The bottom edges of the towels are
folded to prevent seeds from rolling out, and the towels are rolled up for easier
handling. The paper towel is marked with the seed lot number using pencils
designed for writing on damp paper.

18
The thick paper towel and the thin, moist paper towel are layered over the seeds,
and the edges of the paper towels are squeezed together to seal them.

To prevent the seed from rolling out, the towels’ bottoms are folded. I’ve rolled
up the towel. In addition to making handling easier, this prevents the towel from
drying out.

19
For identifying purposes, the towel has the seed lot number in writing. Pencils
designed specifically to write on damp paper are used.

A germination chamber, equipped with a humidifier to control relative humidity,

regulates light, temperature, and humidity levels. The temperature settings within
the germination chamber are adjusted based on the optimal germination tempera-
tures for the specific plant species being tested. Temperature and humidity levels
inside the chamber are monitored and recorded throughout the germination pro-
cess.

20
The rolled paper towels are placed plastic sacks to prevent drying, placed in a
germination chamber with the proper temperature, humidity and light.

The towels are taken out of the chamber after the designated period of time, which
is usually one week.

The paper towels are unrolled, completely uncovering the germinating seedlings.
Ungerminated seeds and germinated seedlings are separated and counted, and the
germination percentage is calculated.

21
Chapter 6
Experimentation and Results
The Experimentation and Result phase of seed germination prediction using
machine learning involves evaluating the performance of the developed models
and interpreting their predictions. By analyzing the data collected from thermal
imaging systems and environmental sensors, we assess the model’s accuracy,
precision, and recall in predicting seed germination outcomes. The evaluation
metrics indicate how well the machine learning algorithms, such as decision
trees, random forests, and neural networks, have learned from the training data.
Detailed analysis of the results helps identify patterns and correlations between
thermal image features and germination success.
6.1 Experimentation
Soybean seeds were positioned on germination paper so that various germination
stages could be photographed with infrared thermal imaging. To provide a sharp
contrast between the developing radicle and the background, the seeds were put on
a black fabric during the thermal picture capture. To lessen the impact of water
evaporation and the consequent drying out of the seeds, the cloth was moistened
with tap water and the germination paper was covered with butter paper. Reflec-
tions from the camera setup were among the reflections that the lid occasionally
produced. Each picture was taken in an office at room temperature using the same
artificial light source. For this purpose, in a single batch of germination 10 seed
samples were put for germination. Seed Germination required total 8 days to get
fully germinated but on-wards 6th day we are able to see that seed is germinating.
Seed Germination was captured over a time span of 1 days with a resolution of
33x33 (1089 pixels). First we have captured initial images of seed on day 1 before
putting for germination than on-wards 6th day we have captured images everyday
till 9th day.
Day 1

22

Figure 6.1: Day 1 Initial Thermal images

Day 6

Figure 6.2: Day 6 Thermal images

Day 7

Figure 6.3: Day 7 Thermal images

Day 8

23

Figure 6.4: Day 8 Thermal images

Day 9

Figure 5.5: Day 9 Thermal images

These samples were germinated at a temperature of 20 degree Celsius to 25 degree

Celsius this played a major role in the time required for germination.

ˆ Temperature = 20 to 25 degree Celsius

ˆ Emissivity = 0.96

ˆ Light Intensity = 50 percent

Additionally, the analysis highlights the model’s robustness in different environ-

mental conditions and its ability to generalize to new, unseen data. The findings
from this phase provide insights into the effectiveness of the predictive system,
guiding further optimization and practical application in agricultural settings.

6.2 Results
6.2.1 Experiment Analysis

Total numbers of Dead

Germinated
Seeds Seeds

360 350 10

Table 6.1: Seeds Count

24
Preprocessing:
1. Data Collection: Gathered data on various parameters affecting seed germi-
nation such as moisture, temperature, light exposure, and seed type.
2. Data Cleaning: Removed duplicate records and handled missing values either
through imputation or deletion.
3. Data Transformation:
ˆ Normalization/Standardization: Features are scaled to maintain uni-
formity and prevent the dominance of particular features.
ˆ Encoding Categorical Variables: Categorical variables are transformed
into numerical representations using methods such as one-hot encoding.
4. Data Splitting: The dataset is divided into training and testing sets, ensuring
a balanced representation of both classes.

Model Training:
1. Support Vector Machine (SVM):
ˆ Initialized an SVM model with suitable hyperparameters (e.g., linear
or RBF kernel).
ˆ Trained the SVM model on the preprocessed training data.
2. Random Forest:
ˆ Created a Random Forest model, a robust ensemble learning method.
ˆ Trained the Random Forest model on the preprocessed training data.

Evaluation:
1. Accuracy: Achieved an accuracy of 97.14%, indicating the model’s effective-
ness in predicting seed germination outcomes.
2. Epoch Iterations: Iterated over multiple epochs to fine-tune the model pa-
rameters and improve performance.
3. Seed Distribution:
ˆ Germinated Seeds: 350
ˆ Dead Seeds: 10
ˆ The model’s ability to predict both outcomes demonstrates its robust-
ness and effectiveness.

25
6.2.2 Model Dashboard
1. Write code in notepad and save as app.py
2. Open Command Prompt
3. Type the following

4. Streamlit Dashboard will open on localhost 8501

26
6.2.3 Model Code

27
6.2.4 Code Explanation

93
The `!pip install opencv-python` command is used to install the OpenCV (Open Source Computer
Vision) library in a Python environment using the pip package manager. Let's break down the
command:

- `!`: In a Jupyter Notebook or similar interactive environment, the exclamation mark (`!`) is used to
run shell commands from within the notebook.

- `pip`: Pip is the package installer for Python, used to install and manage Python packages.

- `install`: This is the command used with pip to install packages.

- `opencv-python`: This is the name of the package being installed. OpenCV is a popular computer
vision library, and `opencv-python` is a Python wrapper for the OpenCV library.

When you run `!pip install opencv-python`, you are instructing pip to download and install the
OpenCV package, making the library and its functionalities available for use in your Python
environment. Once the installation is complete, you can import and use OpenCV in your Python
scripts or notebooks for tasks related to computer vision, image processing, and more.

The `import cv2` statement is used to import the OpenCV library into a Python script or environment.
OpenCV (Open Source Computer Vision) is a popular computer vision library that provides a wide
range of tools and functionalities for image and video processing.

Once you have imported OpenCV using `import cv2`, you can use its functions and classes in your
code.
from PIL import Image

import os

import matplotlib.pyplot as plt

1. Importing necessary libraries:

- `from PIL import Image`: Importing the `Image` module from the Python Imaging Library (PIL),
which is commonly used for working with images.

- `import os`: Importing the `os` module for interacting with the operating system.

- `import matplotlib.pyplot as plt`: Importing the `pyplot` module from Matplotlib for creating plots
and visualizations.

# Set the path to your image folder

folder_path = r"D:\Project\images1\after"

2. Setting the path to the image folder:

- `folder_path`: The variable is set to the path where the images are stored. Make sure to replace
this path with the actual path to your image folder.

# Get a list of all files in the folder

files = os.listdir(folder_path)

3. Getting a list of all files in the specified folder:

- `os.listdir(folder_path)`: Using the `os` module to obtain a list of all files in the specified directory
(`folder_path`).

# Filter out only the image files (you can customize this based on your image file extensions)

image_files = [file for file in files if file.endswith(('.jpg', '.jpeg', '.png', '.bmp'))]

4. Filtering out only the image files:

- `[file for file in files if file.endswith(('.jpg', '.jpeg', '.png', '.bmp'))]`: Using a list comprehension to
filter out only the files with specified image file extensions (`.jpg`, `.jpeg`, `.png`, `.bmp`).

# Display the images (up to the first 10)

axs = plt.subplots(nrows=2, ncols=5, figsize=(12, 6))

axs = axs.flatten()

for i in range(min(10, len(image_files))): # Ensure you don't go beyond the number of available
images

filepath = os.path.join(folder_path, image_files[i])

img = Image.open(filepath)

axs[i].imshow(img)

axs[i].axis('off') # Turn off axis labels for better visualization

plt.show()

5. Displaying the images using Matplotlib:

- `plt.subplots(nrows=2, ncols=5, figsize=(12, 6))`: Creating a 2x5 grid of subplots for displaying the
images.

- `axs = axs.flatten()`: Flattening the 2D array of axes into a 1D array for easier indexing.

- The subsequent loop iterates over the images, loads each one using `Image.open()`, displays it in
the corresponding subplot using `axs[i].imshow(img)`, and turns off axis labels for better visualization
with `axs[i].axis('off')`.

- `plt.show()`: Displaying the entire figure with the images.

In summary, this code reads and filters images from a specified folder, and then displays the first 10
images (or fewer if there are fewer than 10 images) in a 2x5 grid using Matplotlib.

Perform the same code as ‘after’ images for ‘dead’ images as well

The code `cv2.imread(r'D:\Project\images1\after\1 (1).bmp')` uses the OpenCV library to read an

image file from the specified file path. Let's break down the code:

- `cv2.imread`: This is a function in the OpenCV library used to read an image from a file.

- `r'D:\Project\images1\after\1 (1).bmp'`: This is the file path to the image you want to read. The `r`
before the string indicates a raw string, which is used to treat backslashes (`\`) as literal characters.

This line uses the `cv2.imread` function to read an image from the specified file path. The file path
points to a BMP image file (`1 (1).bmp`) located in the `D:\Project\images1\after\` directory. The
function returns a NumPy array representing the image.
It's important to note that the resulting image, when read using `cv2.imread`, will be represented as
a NumPy array with pixel values. This array can be further processed or used for various computer
vision tasks using the OpenCV library. If you want to display the image, additional steps such as using
`cv2.imshow` may be required.

This code is setting up an image data generator using TensorFlow's Keras API for training a neural
network. The image data generator is configured to perform various data augmentation techniques.
Let's go through each part of the code:

import tensorflow as tf

from tensorflow.keras.preprocessing.image import ImageDataGenerator

1. Importing TensorFlow and ImageDataGenerator:

- `import tensorflow as tf`: Imports the TensorFlow library.

- `from tensorflow.keras.preprocessing.image import ImageDataGenerator`: Imports the

`ImageDataGenerator` class from the Keras API within TensorFlow. This class is used for generating
batches of augmented images during training.

# Other relevant configurations

img_height = 224

img_width = 224

batch_size = 32

2. Configuration Parameters:

- `img_height` and `img_width`: Define the height and width to which the images will be resized
during training.

- `batch_size`: Determines the number of images processed in each iteration (batch) during
training.

# Create an instance of ImageDataGenerator for training

train_datagen = ImageDataGenerator(

rescale=1./255,

shear_range=0.2,

zoom_range=0.2,

horizontal_flip=True

3. Creating an ImageDataGenerator for Training:

- `train_datagen = ImageDataGenerator(...)`: Creates an instance of `ImageDataGenerator` for

augmenting training images.

- `rescale=1./255`: Scales pixel values to the range [0, 1].

- `shear_range=0.2`: Applies shear transformations.

- `zoom_range=0.2`: Applies random zoom transformations.

- `horizontal_flip=True`: Randomly flips images horizontally.

# Flow training images from the specified directory

train_dataset = train_datagen.flow_from_directory(

r"D:\Project\train1",

target_size=(img_height, img_width),

batch_size=batch_size,

class_mode='binary'

4. Flow from Directory:

- `train_dataset = train_datagen.flow_from_directory(...)`: Uses the `flow_from_directory` method

to generate batches of augmented data from images in the specified directory (`D:\Project\train1`).

- `target_size=(img_height, img_width)`: Specifies the dimensions to which all images will be

resized.

- `batch_size=batch_size`: Sets the number of images in each batch.

- `class_mode='binary'`: Specifies the type of label array generated by the generator. In this case,
it's set to `'binary'`, indicating binary labels (e.g., for a binary classification task).

In summary, this code sets up an image data generator for training a neural network, applying
various data augmentation techniques to the images from the specified directory. The generated
batches are then typically used for training a model, such as a convolutional neural network (CNN).
This code is similar to the previous one, but it sets up a separate image data generator for validation
using a different directory. Let's break down the code:

validation_dataset = train_datagen.flow_from_directory(

r"D:\Project\validation1",

target_size=(img_height, img_width),

batch_size=batch_size,

class_mode='binary'

1. Flow from Directory for Validation:

- `validation_dataset = train_datagen.flow_from_directory(...)`: Uses the same `train_datagen`

instance (which was previously configured for training) to generate batches of augmented data from
images in the specified validation directory (`D:\Project\validation1`).

- `target_size=(img_height, img_width)`: Specifies the dimensions to which all images in the

validation set will be resized. The dimensions should match the ones used for training.

- `batch_size=batch_size`: Sets the number of images in each batch for validation.

- `class_mode='binary'`: Specifies the type of label array generated by the generator. In this case,
it's set to `'binary'`, indicating binary labels (e.g., for a binary classification task).

This code essentially creates a new data generator (`validation_dataset`) that applies the same data
augmentation settings as the one used for training (`train_datagen`). The validation set, located in
the directory specified, will be processed similarly to the training set, allowing for consistency in
preprocessing steps.

Typically, having separate data generators for training and validation helps ensure that the model
generalizes well to new, unseen data. The validation dataset is used to monitor the model's
performance during training and assess its ability to generalize beyond the training set.

The `train_dataset.class_indices` code refers to accessing the class indices assigned by the
`ImageDataGenerator` during the setup of the training dataset. When you use `flow_from_directory`
to generate batches of images, the generator automatically assigns numerical indices to the different
classes based on the subdirectories in the specified training directory.

# Accessing class indices from the training dataset generator

class_indices = train_dataset.class_indices
1. `train_dataset`: This is an instance of the `ImageDataGenerator` that was created for the training
dataset using `flow_from_directory`.

2. `.class_indices`: This attribute of the `ImageDataGenerator` object provides a dictionary where the
keys are the class names (or subdirectory names in the training directory), and the values are the
corresponding numerical indices assigned to each class.

The `train_dataset.classes` code refers to accessing the class labels assigned to the images in the
training dataset by the `ImageDataGenerator`. When you use `flow_from_directory` to generate
batches of images, the generator assigns numerical labels to the images based on the subdirectories
in the specified training directory.

# Accessing class labels from the training dataset generator

class_labels = train_dataset.classes

1. `train_dataset`: This is an instance of the `ImageDataGenerator` that was created for the training
dataset using `flow_from_directory`.

2. `.classes`: This attribute of the `ImageDataGenerator` object provides an array of numerical labels
corresponding to the classes of the images in the training dataset. The order of the labels
corresponds to the order in which the images are loaded by the generator.

This code defines a Convolutional Neural Network (CNN) model for binary classification using the
TensorFlow and Keras library. Let's break down each part of the code:

import tensorflow as tf

# Define the Sequential model

model = tf.keras.models.Sequential([

tf.keras.layers.Conv2D(16, (3, 3), activation='relu', input_shape=(200, 200, 3)),

 tf.keras.layers.MaxPooling2D(2, 2),

tf.keras.layers.Conv2D(32, (3, 3), activation='relu'),

tf.keras.layers.MaxPooling2D(2, 2),

tf.keras.layers.Conv2D(64, (3, 3), activation='relu'),

tf.keras.layers.MaxPooling2D(2, 2),

tf.keras.layers.Flatten(),

tf.keras.layers.Dense(512, activation='relu'),

tf.keras.layers.Dense(1, activation='sigmoid')

])

1. Sequential Model Definition:

- `tf.keras.models.Sequential`: This line creates a sequential model, which is a linear stack of layers.

2. Convolutional Layers:

- `tf.keras.layers.Conv2D(16, (3, 3), activation='relu', input_shape=(200, 200, 3))`: The first

convolutional layer with 16 filters of size (3, 3), Relu activation, and input shape (200, 200, 3)
representing a 200x200 image with 3 color channels (RGB).

- `tf.keras.layers.MaxPooling2D(2, 2)`: Max-pooling layer with a pool size of (2, 2) to reduce spatial
dimensions.

3. Additional Convolutional Layers:

- Similar architecture with increasing filter count and decreasing spatial dimensions.

4. Flatten Layer:

- `tf.keras.layers.Flatten()`: Flatten layer to convert the 3D output to 1D before passing to the dense
layers.

5. Dense Layers:

- `tf.keras.layers.Dense(512, activation='relu')`: A dense layer with 512 units and Relu activation.

- `tf.keras.layers.Dense(1, activation='sigmoid')`: The output layer with a single unit and sigmoid
activation for binary classification.

# Compile the model

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

6. Compile the Model:

- `model.compile`: Configures the model for training.

- `optimizer='adam'`: The Adam optimizer, an extension to stochastic gradient descent.

- `loss='binary_crossentropy'`: Binary crossentropy loss for binary classification.

- `metrics=['accuracy']`: Evaluation metric to be monitored during training.

# Display the model summary to confirm the architecture and parameter counts

model.summary()

7. Model Summary:

- `model.summary()`: Prints a summary of the model architecture, including layer names, output
shapes, and parameter counts.

from keras.optimizers import RMSprop

8. **Import RMSprop Optimizer :

- `from keras.optimizers import RMSprop`: Importing the RMSprop optimizer, it is potentially

intended for future use.

In summary, this code defines a simple CNN model for binary classification, compiles it with specified
settings, displays the model summary, and includes an unnecessary import of the RMSprop
optimizer. The model consists of convolutional layers followed by max-pooling layers, a flatten layer,
and dense layers, culminating in an output layer with a sigmoid activation for binary classification.

1. Snippet 1: Adding MaxPooling2D Layer

model = tf.keras.models.Sequential([

# ... your other layers

tf.keras.layers.MaxPooling2D(2, 2),

# ... your other layers

])

In this snippet, a MaxPooling2D layer with a pool size of (2, 2) is added to the model. Max pooling
reduces the spatial dimensions of the input data by selecting the maximum value within a specified
window. Note that you should have other layers (like Conv2D) before and after this MaxPooling2D
layer for a complete model.

2. Snippet 2: Adding MaxPooling2D and Flatten Layers

model = tf.keras.models.Sequential([

# ... your other layers

 tf.keras.layers.MaxPooling2D(2, 2),

tf.keras.layers.Flatten(),

# ... your other layers

])

In this snippet, after the MaxPooling2D layer, a Flatten layer is added. The Flatten layer is used to
convert the 3D output of the previous layer to a 1D array before connecting it to dense layers. This is
a common practice when transitioning from convolutional layers to fully connected layers in a CNN.

3. Snippet 3: Adding Flatten and Dense Layers

model = tf.keras.models.Sequential([

# ... your other layers

tf.keras.layers.Flatten(),

tf.keras.layers.Dense(512, activation='relu'),

# ... your other layers

])

In this snippet, a Flatten layer is added, followed by a Dense layer with 512 units and ReLU
activation. The Flatten layer is used to convert the 3D output of the previous layer to a 1D array
before connecting it to the dense layer.

It's essential to have a well-defined sequence of layers in a `Sequential` model, starting with input
layers, followed by convolutional layers, pooling layers, flattening layers, and then dense layers for
the architecture to make sense. The exact structure depends on the specific requirements of your
task (e.g., image classification, object detection).

The code involves compiling a model and training it using TensorFlow and Keras. Let's break down
each part of the code:

model.compile(loss='binary_crossentropy',

optimizer=RMSprop(lr=0.001),

metrics=['accuracy'])

1. Compilation of the Model:

- `model.compile`: Configures the model for training.

 - `loss='binary_crossentropy'`: Specifies the loss function for binary classification. Binary
crossentropy is commonly used for binary classification tasks.

- `optimizer=RMSprop(lr=0.001)`: Specifies the optimizer to be used during training. Here, RMSprop

is chosen with a learning rate (`lr`) of 0.001. The optimizer is responsible for updating the model's
weights based on the computed gradients.

- `metrics=['accuracy']`: Specifies the metric(s) to be monitored during training. In this case, it

monitors accuracy, which is a common metric for classification tasks.

model_fit = model.fit(train_dataset,

steps_per_epoch=3,

epochs=10,

validation_data=validation_dataset)

2. Model Training:

- `model.fit`: Trains the model on the specified dataset.

- `train_dataset`: The training dataset generated by an `ImageDataGenerator` (assumed from the

previous code snippets).

- `steps_per_epoch=3`: Number of batches to be processed per epoch. This value is set to 3, but in
practice, it should be adjusted based on the size of your dataset.

- `epochs=10`: Number of times the entire training dataset is passed through the model. In this
case, the training will proceed for 10 epochs.

- `validation_data=validation_dataset`: The validation dataset generated by another

`ImageDataGenerator`. The model's performance on this dataset will be monitored during training.

The `model_fit` variable will store information about the training process, including loss and accuracy
values at each epoch. After training completes, we can analyze this information to evaluate how well
our model has learned from the data.

This code demonstrates the process of defining, compiling, and training a neural network for binary
classification using TensorFlow and Keras. Adjustments can be made based on the specific
requirements of your task and dataset.
This code defines a simple Sequential model for binary classification using TensorFlow and Keras. The
model consists of two Dense layers. Let's break down each part of the code:

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense

1. Importing Libraries:

- `Sequential`: A linear stack of layers for building a deep learning model layer by layer.

- `Dense`: A fully connected layer where each neuron is connected to every neuron in the previous
layer.

model = Sequential()

2. Creating a Sequential Model:

- `model = Sequential()`: Initializes an empty Sequential model.

model.add(Dense(512, activation='relu', input_shape=(200, 200, 3)))

3. Adding the First Dense Layer:

- `model.add(Dense(512, activation='relu', input_shape=(200, 200, 3)))`: Adds a Dense layer with

512 units, Relu activation, and an input shape of (200, 200, 3). Adjust the `input_shape` to match the
size of your input images.

model.add(Dense(1, activation='sigmoid'))

4. Adding the Output Dense Layer:

- `model.add(Dense(1, activation='sigmoid'))`: Adds the output Dense layer with 1 unit and a
sigmoid activation function. For binary classification, a sigmoid activation is common in the output
layer.

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

5. Compiling the Model:

- `model.compile`: Configures the model for training.

- `optimizer='adam'`: Specifies the Adam optimizer.

- `loss='binary_crossentropy'`: Specifies binary crossentropy as the loss function for binary

classification.

- `metrics=['accuracy']`: Monitors accuracy during training.

model.summary()

6. Printing the Model Summary:

- `model.summary()`: Prints a summary of the model architecture, including layer names, output
shapes, and parameter counts.
In summary, this code defines a basic feedforward neural network with one hidden layer for binary
classification. The architecture can be adjusted based on the specific requirements of your task.
Remember to customize the input shape, activation functions, and other parameters based on your
dataset and problem.

This code performs image classification using a pre-trained model to predict whether seeds in the
given images are germinated or not. Let's break down each part of the code:

Explanation:

1. File Iteration:

- The outer loop iterates through the files in the directory specified by `dir_path`, assuming it
contains subdirectories with seed images.

2. Inner Loop for Image Files:

- The inner loop iterates through the files in the "after" subdirectory (presumably containing images
of germinated seeds).

3. Image Loading and Display:

- `image.load_img`: Loads the image from the specified file path.

- `plt.imshow` and `plt.show`: Displays the loaded image using Matplotlib.

4. Image Preprocessing for Prediction:

- `image.img_to_array`: Converts the image to a NumPy array.

 - `np.expand_dims`: Adds an extra dimension to represent the batch size.

- Normalizes pixel values to be between 0 and 1.

5. Prediction using the Model:

- `model.predict`: Obtains the model's prediction for the preprocessed image.

6. Germination Check:

- Checks if the predicted probability is greater than 0.2. If yes, it's considered as germinated;
otherwise, it's considered not germinated or dead.

This code assumes that the model (`model`) has been previously defined and trained for seed
germination classification. The threshold of 0.2 for considering germination is arbitrary and can be
adjusted based on the model's performance and specific requirements.

This code demonstrates the use of a pre-trained VGG16 model as a feature extractor, followed by
training a Support Vector Machine (SVM) classifier on the extracted features. Let's break down each
part of the code:

from tensorflow.keras.applications import VGG16

from tensorflow.keras.models import Model

1. Importing Libraries:

- `VGG16`: Importing the VGG16 model from Keras applications.

- `Model`: Importing the Model class to create a modified version of the VGG16 model.

base_model = VGG16(weights='imagenet', include_top=False, input_shape=(img_height, img_width,

3))

2. Loading Pre-trained VGG16 Model:

- `VGG16(weights='imagenet', include_top=False, input_shape=(img_height, img_width, 3))`:

Loading the VGG16 model pre-trained on the ImageNet dataset. The `include_top=False` parameter
excludes the fully connected layers at the top of the model, and `input_shape` is set to the desired
image input shape.

# Extract features using the pre-trained model

train_features = base_model.predict(train_dataset)

train_features = train_features.reshape(train_features.shape[0], -1)

3. Extracting Features:

- `base_model.predict(train_dataset)`: Extracting features from the pre-trained VGG16 model using

the training dataset (`train_dataset`). The resulting features are flattened to create a 1D array for
each sample.

from sklearn.svm import SVC

from sklearn.model_selection import train_test_split

from sklearn.metrics import accuracy_score

4. Importing Scikit-Learn Libraries:

- Importing necessary modules from scikit-learn for SVM classification.

# Split the data into training and testing sets

X_train, X_test, y_train, y_test = train_test_split(train_features, train_dataset.classes, test_size=0.2,

random_state=42)

5. Train-Test Split:

- Splitting the data into training and testing sets using scikit-learn's `train_test_split` function.

# Create an SVM classifier

svm_classifier = SVC()

# Train the classifier

svm_classifier.fit(X_train, y_train)

6. Creating and Training SVM Classifier:

- Creating an SVM classifier using scikit-learn's `SVC` class.

- Training the SVM classifier using the training data.

# Make predictions on the test set

svm_predictions = svm_classifier.predict(X_test)

# Evaluate accuracy

svm_accuracy = accuracy_score(y_test, svm_predictions)

print(f"SVM Accuracy: {svm_accuracy}")

7. Making Predictions and Evaluating Accuracy:

- Making predictions on the test set using the trained SVM classifier.

- Evaluating the accuracy of the SVM classifier using scikit-learn's `accuracy_score` function.

- Printing the accuracy of the SVM classifier on the test set.

This code demonstrates a transfer learning approach where the pre-trained VGG16 model is used to
extract features from images, and an SVM classifier is trained on these features for a specific
classification task.

This code demonstrates the use of a Random Forest classifier from scikit-learn for a classification
task. Let's break down each part of the code:

from sklearn.ensemble import RandomForestClassifier

1. Importing the Random Forest Classifier:

- `RandomForestClassifier`: Importing the Random Forest classifier from scikit-learn's ensemble

module.

# Create a Random Forest classifier

rf_classifier = RandomForestClassifier(n_estimators=100, random_state=42)

2. Creating the Random Forest Classifier:

- `RandomForestClassifier(n_estimators=100, random_state=42)`: Creating a Random Forest

classifier with 100 decision trees (`n_estimators=100`) and setting the random seed for
reproducibility (`random_state=42`).

# Train the classifier

rf_classifier.fit(X_train, y_train)

3. **Training the Random Forest Classifier:**

- `rf_classifier.fit(X_train, y_train)`: Training the Random Forest classifier using the training data
(`X_train` features and `y_train` labels).

# Make predictions on the test set

rf_predictions = rf_classifier.predict(X_test)
4. Making Predictions on the Test Set:

- `rf_classifier.predict(X_test)`: Making predictions on the test set using the trained Random Forest
classifier.

# Evaluate accuracy

rf_accuracy = accuracy_score(y_test, rf_predictions)

print(f"Random Forest Accuracy: {rf_accuracy}")

5.*Evaluating Accuracy:

- `accuracy_score(y_test, rf_predictions)`: Calculating the accuracy of the Random Forest classifier

on the test set using scikit-learn's `accuracy_score` function.

- Printing the accuracy of the Random Forest classifier.

In summary, this code creates a Random Forest classifier, trains it on a training dataset (`X_train` and
`y_train`), makes predictions on a test dataset (`X_test`), and evaluates the accuracy of the classifier.
Random Forests are an ensemble learning method that builds multiple decision trees during training
and outputs the mode of the classes as the prediction. The number of decision trees (`n_estimators`)
is a hyperparameter that can be adjusted based on the dataset and desired performance.
Chapter 7
Conclusion

In this project, we have studied a conventional method of seed germination

testing. We observed various stages of seed germination. The experimental out-
comes stated thermal imaging system and the processing of thermal image, factors
affecting thermal images.
We have developed a machine learning-based seed germination prediction model.
We aimed to automate the process of seed germination testing using a machine
learning model. Seed evaluations are often completed by hand, which is a laborious,
slow, and prone to mistake procedure. Owing to the tremendous advancements in
imaging, electronics, and computer science over the past several decades, infrared
thermal imaging has seen many scientific advancements and clever uses in non-
destructive testing and quality monitoring of many agro-food products. Thermal
imaging provides a non-invasive and nondestructive method to assess the germina-
tion process by monitoring temperature variations in seeds. Making ensuring the
camera is calibrated correctly and that the data gathering procedure is systematic
and thoroughly recorded are crucial.

110
 References

1. Genze, N., Bharti, R., Grieb, M. et al. ”Accurate machine learning-based

germination detection, prediction and quality assessment of three grain crops”,
Plant Methods 16, 157 (2020).
2. Srinath Yasam , Dr.S. Anu H Nair and Dr.K.P. Sanal Kumar,”Automated
Germination Classification using ORB Feature Extractor with Machine Learn-
ing Model,” Journal of Pharmaceutical Negative Results., Volume. 13, Spe-
cial Issue 9, (2022).
3. Amjan Shaik, Nishath Ansari, M. Neelakantappa, Amtul Nimra, K. Pur-
nachand, Saikumar Tara; Machine learning approach for seed analysis using
feature selection technique and evolutionary algorithms. AIP Conf. Proc.
(4 May 2023); 2477 (1): 030064. https://doi.org/10.1063/5.0125198
4. M. R. Kumar, A. Sharma, K. Sreenivasulu and G. Ramesh, ”Pivot Based
Seed Germination Assessment (PBSGA) Pattern for Germination Quality
Analysis,” 2022 International Conference on Inventive Computation Tech-
nologies (ICICT), Nepal, (2022), pp. 230-237, doi: 10.1109/ICICT54344.2022.
9850837.
5. Nehoshtan, Y., Carmon, E., Yaniv, O. et al. Robust seed germination pre-
diction using deep learning and RGB image data. Sci Rep 11, 22030 (2021).
https://doi.org/10.1038/s41598-021-01712-6
6. Enyang A. Grace, Jackline Kemboi and Jesse Owino,”Seed germination,
seedling growth and biomass prediction on pre treated seeds of Acacia sene-
gal”,Academia Journal of Agricultural Research 8(6): 187-192, June (2020).
7. T. T. Nguyen, V. -N. Hoang, T. -L. Le, T. -H. Tran and H. Vu, ”A Vi-
sion Based Method for Automatic Evaluation of Germination Rate of Rice
Seeds,” 2018 1st International Conference on Multimedia Analysis and Pat-
tern Recognition (MAPR), Ho Chi Minh City, Vietnam, (2018), pp. 1-6,
doi: 10.1109/MAPR.2018.8337511.
8. Baranowski, Piotr & Mazurek, Wojciech & Tadeusz, Walczak. ”The use of
thermography for pre-sowing evaluation of seed germination capacity”, Acta
horticulturae. 6004. 459-465. 10.17660/ActaHortic.2003.604.50.
9. Enyang A. Grace, Jackline Kemboi and Jesse Owino,”Seed germination,
seedling growth and biomass prediction on pre treated seeds of Acacia sene-
gal”,Academia Journal of Agricultural Research 8(6): 187-192, (June 2020).

111
10. Abram J, Bicksler P . Testing Seed Viability Using Simple Germination
Tests; A Regional Supplement to ECHO Development Notes. Unpublushed.(2011),
Pp 1-4.

11. Dell’Aquila, A. Digital imaging information technology applied to seed ger-

mination testing. A review. Agron. Sustain. Dev. 29, 213–221 (2009).

12. Saffariha, M., Jahani, A. & Potter, D. ”Seed germination prediction of Salvia
limbata under ecological stresses in protected areas: an artificial intelligence
modeling approach”, BMC Ecol 20, 48 (2020). https://doi.org/10.1186/s12898-
020-00316-4

13. Chitra, H. S. H., Suguna, S., & Sujatha, S. N. A survey on image analysis
techniques in agricultural product. Indian Journal of Science and Technol-
ogy, 9(12), 1-13, (2016).

14. Tiziana Lombardi, Tiziana Fochetti, Andrea Bertacchi, Antonino Onnis,

Germination requirements in a population of Typha latifolia, Aquatic Botany,
Volume 56, Issue 1, (1997), Pages 1-10, ISSN 0304-3770, https://doi.org/10.1016/
S0304-3770(96)01096-0.

15. Chaerle L, Van Der Straeten D. Seeing is believing: imaging techniques to

monitor plant health. Biochim Biophys Acta. (2001 Jun 28);1519(3):153-
66. doi: 10.1016/s0167-4781(01)00238-x. PMID: 11418181

112
 Certification

Project Participation Certificate

113
114
115
 Log Book

Department fo Instrumentation Engineering

B.E. PROJECT MONITORING SHEET (SEM I/II)

Name of Project : Seed Germination Prediction using Machine Learning
Name of Students : 1) Dattaram Gawade 2) Akshat Fulfagar
Internal Guide : Dr. A. A. Shinde

Sr. Internal
Date Work Done
No Guide

Project discussion
1 14/07/2023
with Guide.

Project Discussion
with domain expert
2 21/07/2023
officer Mr. Rawindra
Wardkar sir

Prepared Action plan

3 28/07/2023
of project.

Started the process of

4 04/08/2023
creating a dataset.

Created the first batch

5 13/08/2023
of thermal images.

116
 Created a second
6 25/08/2023 batch of thermal
images.

1. Meeting with
Guide

2. Searching and
7 15/09/2023 collection of
research papers
related to the
project.

Attended Principal sir

seminar on discussion
8 21/09/2023
of guidelines for BE
project.

Read 15 to 20 Re-
search papers and col-
9 29/09/2023 lected relevant infor-
mation for review pa-
pers.

Making of review pa-

10 06/10/2023
per.

Meeting with guide re-

garding checking and
11 19/10/2023
correction in review
paper.

1. Created a
dataset of
approx 500
thermal images.
12 03/11/2023
2. . Analysing
the dataset and
sorting of im-
ages.

117
 1. . Importing
dataset

2. Checking accu-
racy of dataset
13 10/11/2023
using SVM and
Random forest
machine learn-
ing algorithm.

Project Stage I
14 23/11/2023
Demonstration.

Meeting with the

15 14/12/2023
guide.

Studying various pre-

processing techniques
16 28/12/2023
to improve the accu-
racy of output

Applying the tech-

17 11/01/2024 niques to improve
accuracy

Getting the overall

18 18/01/2024
targeted output

Project Stage II
19 02/02/2024
progress presentation.

Research paper work

20 22/02/2024
started

UGC Journal search

21 07/03/2024
for publication

Pariyojna Pratiyogita
2024 project com-
22 08/04/2024 petition at Sinhgad
academy of engineer-
ing.

118
 Project Competition
23 15/04/2024 and Exhibition in
AISSMS IOIT.

1. Project Stage II
Examiantion
24 25/05/2024
2. BlackBook Sub-
mission

119
 PO Mapping

Department of Instrumentation Engineering

PSO Mapping of Project

Project Title: Seed Germination Prediction Using Machine

Learning

(Tick the POs/PSOs which are mapping)

ENGINEERING GRADUATES WILL BE ABLE TO

PO 1: Apply the knowledge of mathematics, science, engineering fun-

2
damentals, and an engineering specialization to the solution of
complex engineering problems. [Engineering Knowledge]

PO 2: Identify, formulate, research literature, and analyse complex en-

gineering problems reaching substantiated conclusions using first

2
principles of mathematics, natural sciences, and engineering sci-
ences. [Problem analysis]

PO 3: Design solutions for complex engineering problems and design

system components or processes that meet the specified needs with
appropriate consideration for the public health and safety, and the

2
cultural, societal, and environmental considerations.[Design/ de-
velopment of solutions]

120
PO 4: Use research-based knowledge and research methods including
design of experiments, analysis and interpretation of data, and syn-

2
thesis of the information to provide valid conclusions. [Conduct in-
vestigations of complex problems].

PO 5: Create, select, and apply appropriate techniques, resources, and

modem engineering and IT tools including prediction and mod-

2
elling to complex engineering activities with an understanding of
the limitations. [Modern tool usage]

PO 6: Apply reasoning informed by the contextual knowledge to assess

societal, health, safety, legal and cultural issues and the consequent

2
responsibilities relevant to the professional engineering practice.
[The engineer and society]

PO 7: Understand the impact of the professional engineering solu-

tions in societal and environmental contexts, and demonstrate the
knowledge of, and need for sustainable development. [Environment
and sustainability)

PO 8: Apply ethical principles and commit to professional ethics and

2
responsibilities and norms of the engineering practice. [Ethics]

PO 9: Function effectively as an individual, and as a member or leader

2
in diverse teams, and in multidisciplinary settings. [Individual and
team work]

PO 10: Communicate effectively on complex engineering activities with

the engineering community and with society at large, such as, being
able to comprehend and write effective reports and design docu-

2
mentation, make effective presentations, and give and receive clear
instructions. [Communication]

PO 11: Demonstrate knowledge and understanding of the engineering

and management principles and apply these to one’s own work, as a

2
member and leader in a team, to manage projects and in multidisci-
plinary environments. [Project management and finance]

121
PO 12: Recognize the need for, and have the preparation and ability to

2
engage in independent and lifelong learning in the broadest context
of technological change. [Life-long learning]

PSO 1: Graduates will be able to apply their knowledge of measure-

2
ment and control to solve the problems related to the environment,
safety, health and agriculture sectors.

PSO 2: Graduates will be able to demonstrate their skills on Program-

2
ming, Programmable logic controller, LabView, Distributed con-
trol system and Internet of thing.

Dr.A.A.Shinde
Project Guide

122
 Self Evaluation

ALL INDIA DHRI SHIVAJI MEMORIAL SOCIETY’S

INSTITUTE OF INFORMATION TECHNOLOGY,

PUNE

Self-Evaluation sheet

Observation and comments of Guide:

Sign of Guide: ...................

Name of Student:

1. Dattaram Gawade ..................

2. Akshat Fulfagar ..................

123
 Critical Thinking

Department of Instrumentation Engineering

Project Critical Thinking Report

Q1: Who benefits from this?
Ans: The project can offer several benefits to various stakeholders:

1. Farmers and Agricultural Industry: Farmers can benefit greatly from accu-
rate seed germination predictions. By knowing in advance which seeds are
likely to germinate successfully, farmers can optimize planting schedules, re-
duce waste of seeds, and increase overall crop yields. This leads to improved
efficiency and profitability in the agricultural industry.

2. Seed Producers and Distributors: Seed producers and distributors can use
machine learning models to predict seed germination rates more accurately.
This allows them to better manage inventory, ensure timely deliveries, and
maintain customer satisfaction by providing high-quality seeds.

3. Research Institutions: Research institutions studying plant biology, genetics,

and agriculture can utilize the insights gained from machine learning models
to understand the factors influencing seed germination. This can lead to
advancements in breeding programs, crop improvement, and the development
of new seed varieties that exhibit better germination rates under various
conditions.

4. Environmental Conservation: Predicting seed germination rates accurately

can also aid in ecological restoration efforts. By understanding how different
species respond to environmental factors, conservationists can make informed
decisions about which seeds to plant in restoration projects, thus promoting
biodiversity and ecosystem health.

124
 5. Education and Outreach: Projects like this can also be valuable for educa-
tional purposes. Students and enthusiasts interested in machine learning,
agriculture, or environmental science can learn about the practical applica-
tions of data science in addressing real-world challenges such as optimizing
agricultural practices and enhancing food security.

6. Food Security and Sustainability: By improving seed germination predic-

tion, this project contributes to ensuring food security and sustainability.
With more efficient use of seeds and resources, as well as increased crop
yields, it helps to meet the growing demand for food while minimizing the
environmental impact of agriculture.
Q2: Who is most directly affected?
Ans: The stakeholders most directly affected by the project include:
1. Farmers: Farmers are at the forefront of utilizing the predictions gener-
ated by the machine learning models. They directly benefit from optimized
planting schedules, reduced seed wastage, and increased crop yields. By ac-
curately predicting seed germination, farmers can make informed decisions
about when and where to plant their seeds, thereby maximizing their chances
of a successful harvest. This directly impacts their income, livelihood, and
overall agricultural productivity.

2. Seed Producers and Distributors: Seed producers and distributors are di-
rectly impacted by the project’s outcomes as they rely on accurate germi-
nation predictions to manage their inventories and meet customer demands
effectively. By incorporating machine learning-generated insights into their
operations, they can ensure the timely availability of high-quality seeds with
predictable germination rates. This enhances their competitiveness in the
market and strengthens customer satisfaction, leading to increased sales and
revenue.

3. Researchers and Breeders: Researchers and breeders in the agricultural sec-

tor benefit directly from the project by gaining valuable insights into the
factors influencing seed germination. By analyzing the data produced by the
machine learning models, they can identify genetic markers and environmen-
tal variables associated with optimal germination outcomes. This knowledge
informs their breeding programs, enabling the development of new crop vari-
eties with improved germination rates, resilience to environmental stressors,
and desirable agronomic traits.

4. Conservationists: Conservationists involved in habitat restoration and bio-

diversity conservation initiatives are directly affected by the project’s out-

125
 comes. Accurate predictions of seed germination enable them to select ap-
propriate seed species for restoration projects, ensuring the successful estab-
lishment of native plant communities. By leveraging machine learning-based
insights, conservationists can enhance the effectiveness of their efforts to re-
habilitate degraded ecosystems, preserve endangered species habitats, and
promote landscape-scale conservation.
5. Educational Institutions: Educational institutions, including universities and
research centers, are directly impacted by the project as it contributes to the
advancement of knowledge in fields such as plant biology, ecology, and data
science. Students and researchers have access to the data generated by the
project, allowing them to study the underlying mechanisms of seed germina-
tion and explore innovative applications of machine learning in agriculture
and environmental science. This fosters interdisciplinary collaboration and
prepares future generations of scientists and professionals to address complex
challenges in food security and sustainability.
Q3: What are the strengths and weaknesses?
Ans:
Strengths:
1. Data-Driven Insights: One of the primary strengths of using machine learn-
ing for seed germination prediction is its ability to analyze vast amounts
of data. Machine learning algorithms can identify complex patterns and
relationships within datasets that may not be apparent through traditional
statistical methods. This enables researchers to extract valuable insights into
the factors influencing seed germination, such as environmental conditions,
seed characteristics, and genetic traits.
2. Improved Accuracy: Machine learning models have the potential to signifi-
cantly improve the accuracy of seed germination predictions. By training on
large and diverse datasets, these models can learn to recognize subtle cues
and patterns associated with successful germination outcomes. This leads
to more reliable predictions, allowing farmers, seed producers, and conserva-
tionists to make informed decisions with greater confidence.
3. Customization and Adaptability: Machine learning techniques offer flexibil-
ity in model development and customization. Researchers can tailor mod-
els to specific crop species, environmental conditions, or geographic regions,
ensuring relevance and applicability to diverse agricultural and ecological
contexts. Moreover, machine learning algorithms can adapt and refine their
predictions over time as new data becomes available, leading to continuous
improvement in accuracy and performance.

126
 4. Resource Efficiency: By predicting seed germination outcomes in advance,
machine learning models help optimize resource allocation in agriculture and
conservation. Farmers can minimize the use of seeds, water, fertilizers, and
other inputs by strategically planning planting schedules based on predicted
germination probabilities. Similarly, conservationists can prioritize the se-
lection and deployment of seeds in habitat restoration projects, maximizing
the efficiency and effectiveness of restoration efforts.

Weakness:

1. Data Limitations and Quality: A key challenge in developing machine learn-

ing models for seed germination prediction is the availability and quality of
data. Obtaining comprehensive and reliable datasets encompassing a wide
range of environmental conditions, seed types, and germination outcomes
can be challenging. Biases or inaccuracies in the data can lead to biased
predictions or unreliable model performance.

2. Complexity and Interpretability: Machine learning models, particularly deep

learning algorithms, can be highly complex and opaque, making them dif-
ficult to interpret and understand. This lack of interpretability poses chal-
lenges in identifying the underlying factors driving germination predictions
and may hinder the adoption of these models by stakeholders who require
transparent decision-making processes.

3. Overfitting and Generalization: Machine learning models may face challenges

related to overfitting, where the model learns to memorize the training data
rather than capturing underlying patterns. This can result in poor gener-
alization performance on unseen data, leading to inaccurate predictions in
real-world settings. Addressing overfitting requires careful model selection,
regularization techniques, and validation strategies to ensure robust perfor-
mance across diverse scenarios.

4. Ethical and Social Implications: The use of machine learning in agriculture

and environmental management raises ethical and social considerations re-
lated to data privacy, equity, and environmental justice. Concerns may arise
regarding the ownership and control of data, as well as the potential for al-
gorithmic biases to disproportionately impact marginalized communities or
exacerbate existing inequalities in access to resources and decision-making
processes.

127
Q4: What is the best and worst case scenario?
Ans: The best and worst-case scenarios for the project include:
Best-Case Scenario:
1. High Accuracy and Reliability: In the best-case scenario, the machine learn-
ing models exhibit exceptional accuracy and reliability in predicting seed
germination outcomes. These models successfully capture the complex rela-
tionships between environmental factors, seed characteristics, and germina-
tion rates, leading to highly accurate predictions across diverse conditions
and crop species.
2. Optimized Resource Allocation: Farmers, seed producers, and conservation-
ists benefit from optimized resource allocation based on the predictions gen-
erated by the machine learning models. Planting schedules are meticulously
planned to coincide with optimal germination conditions, minimizing seed
wastage and resource usage while maximizing crop yields and restoration
success rates.
3. Improved Agricultural Productivity: The project contributes to significant
improvements in agricultural productivity and sustainability. Farmers achieve
higher yields and profitability through enhanced germination rates and effi-
cient use of inputs, leading to increased food security and economic prosper-
ity in agricultural communities.
4. Enhanced Conservation Efforts: Conservationists leverage the predictive
power of machine learning to prioritize seed selection and habitat restora-
tion projects effectively. Native plant species are reintroduced into degraded
ecosystems with a high likelihood of successful establishment, resulting in
improved biodiversity, ecosystem resilience, and ecological restoration out-
comes.
5. Scientific Advancements: The project drives scientific advancements in the
fields of plant biology, ecology, and data science. Researchers gain valuable
insights into the mechanisms underlying seed germination and ecosystem
dynamics, fostering interdisciplinary collaboration and innovation in agri-
cultural and environmental research.
Worst-Case Scenario:
1. Inaccurate Predictions: In the worst-case scenario, the machine learning
models produce inaccurate predictions of seed germination outcomes. Fac-
tors such as limited or biased data, inadequate model training, or unforeseen
environmental variability contribute to unreliable predictions, leading to sub-
optimal decision-making and resource allocation by stakeholders.

128
 2. Resource Misallocation: Farmers and conservationists may experience re-
source misallocation due to unreliable predictions generated by the machine
learning models. Planting schedules may be poorly timed, resulting in de-
creased crop yields, wasted seeds, and inefficient use of resources. Conser-
vation efforts may suffer setbacks as seeds fail to germinate in restoration
projects, undermining biodiversity conservation and ecosystem restoration
goals.

3. Loss of Trust and Adoption: Stakeholders may lose trust in the predictive ca-
pabilities of machine learning models if they consistently produce inaccurate
results. This loss of trust could lead to reluctance or resistance in adopting
machine learning technology for seed germination prediction, hindering its
potential to improve agricultural practices and conservation outcomes.

4. Negative Environmental Impact: Inaccurate predictions may inadvertently

contribute to negative environmental impacts, such as the spread of invasive
species or the disruption of fragile ecosystems. Improperly selected seeds may
fail to establish in restoration sites, leading to the loss of native vegetation
and exacerbating ecological degradation in sensitive habitats.

5. Stagnation of Scientific Progress: The failure of machine learning models to

deliver reliable predictions could impede scientific progress in understanding
seed germination processes and ecosystem dynamics. Research efforts may be
hampered by data inconsistencies, model limitations, and a lack of actionable
insights, hindering efforts to address pressing challenges in agriculture and
conservation.

Q5: Where would we see this in the real world?

Ans: In the real world, applications could be observed in agricultural settings,
conservation projects, and research institutions. Farmers may use predictive mod-
els to optimize planting schedules and resource allocation, while seed producers
could enhance inventory management and customer satisfaction. Conservationists
might leverage these models to guide habitat restoration efforts, and researchers
could gain insights into plant biology and ecosystem dynamics. Ultimately, these
applications aim to improve agricultural productivity, biodiversity conservation,
and scientific understanding.

Q6: Where is there the most need for this?

Ans: The greatest need for ”Seed Germination Prediction Using Machine Learn-
ing” is evident in regions grappling with a convergence of environmental, socio-
economic, and food security challenges. Particularly, areas characterized by harsh
environmental conditions, such as arid or drought-prone regions, stand to benefit

129
significantly. In these locales, where unpredictable weather patterns and water
scarcity pose constant threats to agriculture, accurate prediction of seed germina-
tion offers a lifeline. By pinpointing optimal planting times and identifying seeds
resilient to local conditions, farmers can mitigate risks and optimize resource usage,
ultimately bolstering agricultural productivity and resilience. Moreover, the tech-
nology holds immense promise in supporting marginalized or vulnerable commu-
nities, including rural farmers, smallholders, and indigenous groups. Often lacking
access to modern agricultural practices, these communities face heightened vul-
nerability to climate change impacts and resource constraints. Implementing seed
germination prediction models empowers them with valuable insights and tools
to navigate uncertain agricultural landscapes, enhancing their ability to sustain
livelihoods and ensure food security. Furthermore, regions characterized by bio-
diversity hotspots, degraded ecosystems, or undergoing habitat restoration efforts
also present significant opportunities for applying predictive modeling. By guid-
ing conservationists in selecting appropriate seeds and restoration strategies, these
models can accelerate ecosystem recovery, promote biodiversity conservation, and
strengthen ecosystem resilience in the face of ongoing environmental challenges.

Q7: Where would this idea take us?

Ans: The idea of ”Seed Germination Prediction Using Machine Learning” has
the potential to take us towards a future characterized by enhanced agricultural
productivity, sustainable resource management, and improved environmental con-
servation. Here are some directions in which this idea could lead us:

1. Precision Agriculture: Implementing machine learning models for seed germi-

nation prediction enables farmers to adopt precision agriculture practices. By
accurately predicting germination outcomes and optimizing planting sched-
ules, farmers can tailor their agricultural practices to specific environmental
conditions, soil types, and crop varieties. This leads to more efficient use of
resources such as water, fertilizers, and pesticides, reducing environmental
impact and enhancing crop yields.

2. Climate Resilience: In the face of climate change and increasingly unpre-

dictable weather patterns, the ability to predict seed germination becomes
essential for building climate-resilient agricultural systems. By anticipating
shifts in growing seasons, extreme weather events, and water availability,
farmers can adapt their planting strategies to mitigate risks and maintain
productivity in changing climates.

3. Sustainable Land Management: Predictive models for seed germination can

inform sustainable land management practices, particularly in ecologically
sensitive areas and degraded landscapes. Conservationists can use these

130
 models to guide habitat restoration efforts, selecting native plant species with
high germination probabilities to restore biodiversity, improve soil health,
and enhance ecosystem resilience.

4. Innovation in Crop Breeding: Machine learning-driven insights into seed ger-

mination processes provide valuable information for crop breeders seeking to
develop new varieties with improved germination rates and resilience to en-
vironmental stressors. By identifying genetic markers associated with desir-
able germination traits, breeders can accelerate the development of climate-
resilient crops capable of thriving in diverse agro-ecological conditions.

5. Data-Driven Decision-Making: The integration of machine learning tech-

nology with seed germination prediction fosters a culture of data-driven
decision-making in agriculture and conservation. Stakeholders across the
supply chain, from farmers and seed producers to policymakers and re-
searchers, can leverage predictive analytics to make informed choices, op-
timize resource allocation, and address pressing challenges related to food
security, environmental sustainability, and ecosystem health.

6. Interdisciplinary Collaboration: The convergence of agricultural science, data

science, and environmental research facilitated by seed germination predic-
tion using machine learning promotes interdisciplinary collaboration and
knowledge exchange. Researchers, practitioners, and policymakers from di-
verse fields work together to develop innovative solutions to complex chal-
lenges at the intersection of agriculture, ecology, and climate change.
Q8: When would this cause a problem?
Ans: While this project holds immense potential for improving agricultural prac-
tices and conservation efforts, there are several instances where its implementation
might encounter challenges or even cause problems:
1. Data Bias and Quality Issues: Machine learning models heavily rely on data
for training and validation. If the data used to train these models is biased,
incomplete, or of poor quality, it can lead to inaccurate predictions and
unreliable outcomes. Biases in the data, such as overrepresentation of certain
regions or crop varieties, may result in models that generalize poorly to
diverse contexts, leading to suboptimal decision-making.

2. Environmental Variability and Uncertainty: Environmental conditions can

vary widely over time and space, posing challenges for predictive modeling.
Unforeseen environmental events, such as extreme weather events or sudden
changes in soil composition, may disrupt established patterns and render
predictive models ineffective. Moreover, the inherent uncertainty associated

131
 with ecological systems makes it difficult to anticipate all potential factors
influencing seed germination accuratel

3. Ethical and Social Implications: The use of machine learning in agricul-

ture and conservation raises ethical and social considerations related to data
privacy, equity, and power dynamics. Issues such as ownership of data,
transparency in decision-making, and equitable access to technology must
be carefully addressed to prevent the exacerbation of existing inequalities
and ensure responsible and inclusive implementation of predictive modeling
solutions.

4. Overreliance on Technology: There is a risk of overreliance on technology,

where stakeholders may become overly dependent on predictive models with-
out considering other sources of knowledge or expertise. Blindly following
model predictions without critical assessment or local contextual understand-
ing can lead to suboptimal outcomes and undermine traditional agricultural
knowledge or conservation practices.

5. Unintended Consequences: Predictive modeling efforts may inadvertently

lead to unintended consequences or perverse incentives. For example, if farm-
ers prioritize seeds predicted to have the highest germination rates without
considering other factors such as genetic diversity or local adaptation, it could
result in monoculture cropping systems or the displacement of indigenous
crop varieties, leading to long-term agricultural and ecological consequences.

6. Regulatory and Policy Challenges: The adoption of predictive modeling tech-

nologies in agriculture and conservation may encounter regulatory and pol-
icy challenges related to data governance, intellectual property rights, and
liability issues. Policymakers must navigate complex legal and regulatory
landscapes to ensure that the benefits of predictive modeling are realized
while mitigating potential risks and safeguarding public interests.

Q9: Why is this a problem or challenge?

Ans: The use of this project presents several challenges and potential problems
due to various factors:

1. Data Limitations: One significant challenge is the availability and quality of

data required to train machine learning models effectively. Obtaining com-
prehensive and representative datasets encompassing diverse environmental
conditions, seed varieties, and germination outcomes can be challenging. In-
adequate or biased data can lead to inaccurate model predictions and unre-
liable results.

132
 2. Environmental Complexity: Environmental conditions influencing seed ger-
mination are complex and dynamic, making it challenging to capture all rel-
evant factors accurately. Variables such as soil moisture, temperature, light
intensity, and nutrient availability interact in intricate ways, and predicting
their combined effects accurately can be difficult. Moreover, environmental
conditions may vary significantly across different geographical regions and
cropping systems, further complicating predictive modeling efforts.

3. Model Interpretability: Machine learning models, particularly complex deep

learning algorithms, often lack interpretability, making it challenging to un-
derstand how predictions are generated. This lack of transparency can hinder
stakeholders’ ability to trust and interpret model outputs, raising concerns
about accountability, bias, and the reliability of model prediction

4. Generalization and Transferability: Machine learning models trained on spe-

cific datasets may struggle to generalize their predictions to new or unseen
environments. Models that perform well in one region or under certain condi-
tions may fail to generalize to other contexts, leading to poor performance in
real-world applications. Achieving robustness and transferability across di-
verse environmental conditions is a significant challenge in seed germination
prediction.

5. Ethical and Social Implications: The use of predictive modeling in agricul-

ture and conservation raises ethical and social considerations related to data
privacy, equity, and decision-making. Issues such as data ownership, con-
sent, and the potential for algorithmic biases to perpetuate or exacerbate
existing inequalities must be addressed to ensure responsible and equitable
implementation of predictive modeling solutions.

6. Regulatory and Policy Frameworks: The adoption of predictive modeling

technologies in agriculture and conservation may encounter regulatory and
policy challenges related to data governance, intellectual property rights, and
liability issues. Policymakers must develop appropriate regulatory frame-
works to address these challenges and ensure that the benefits of predictive
modeling are realized while safeguarding public interests and mitigating po-
tential risks.
Q10: Why is there a need for this today?
Ans: The need for this project is particularly pressing today due to several factors:
1. Climate Change: Climate change is causing increased variability in weather
patterns, including changes in temperature, precipitation, and extreme weather
events. These shifts can disrupt traditional planting schedules and impact

133
 seed germination rates, necessitating more accurate and adaptable predictive
models to optimize agricultural practices and mitigate risks.

2. Food Security: With a growing global population and increasing pressure on

food systems, ensuring food security is a critical priority. Predictive models
for seed germination can help farmers maximize crop yields and minimize
losses by optimizing planting schedules, selecting resilient crop varieties, and
managing resources efficiently.

3. Resource Scarcity: Scarce resources such as water, land, and seeds require
careful management to meet the demands of a growing population while
minimizing environmental impact. Predictive modeling enables farmers to
make informed decisions about resource allocation, optimizing seed usage,
irrigation practices, and land management to achieve sustainable agricultural
production.

4. Biodiversity Conservation: Preserving biodiversity and restoring degraded

ecosystems are essential for maintaining ecosystem services, supporting wildlife
habitats, and mitigating climate change. Predictive models for seed germi-
nation can guide habitat restoration efforts by selecting appropriate native
plant species with high germination probabilities, enhancing restoration suc-
cess rates and promoting ecosystem resilience.

5. Technological Advancements: Advances in machine learning algorithms, com-

puting power, and data analytics have made it feasible to develop sophisti-
cated predictive models for seed germination. By harnessing these technolo-
gies, researchers and practitioners can unlock new insights into seed biol-
ogy, environmental interactions, and crop performance, driving innovation
in agriculture and conservation.

6. Sustainable Agriculture: There is a growing recognition of the need to tran-

sition towards more sustainable and regenerative agricultural practices. Pre-
dictive modeling supports this transition by enabling precision agriculture
techniques, optimizing inputs, reducing waste, and promoting resilience to
environmental stressors, ultimately contributing to more sustainable food
production systems.
Q11: How does this benefit us/others?
Ans: ”Seed Germination Prediction Using Machine Learning” offers a multitude
of benefits to various stakeholders, thereby contributing to agricultural sustainabil-
ity, environmental conservation, and societal well-being. For farmers, the ability to
accurately predict seed germination outcomes translates into optimized planting
schedules, reduced wastage of seeds and resources, and increased crop yields. This

134
not only enhances farm profitability but also promotes food security and economic
resilience in agricultural communities. Seed producers and distributors similarly
benefit from improved inventory management and customer satisfaction, leading
to enhanced market competitiveness and profitability. Conservationists leverage
predictive models to guide habitat restoration efforts, selecting native plant species
with high germination probabilities to restore biodiversity and ecosystem resilience.
Moreover, researchers gain valuable insights into seed biology and environmental
interactions, driving scientific advancements and interdisciplinary collaboration.
Ultimately, the widespread adoption of seed germination prediction using machine
learning fosters technological innovation, sustainable development, and positive so-
cietal impacts, thereby benefiting individuals, communities, and ecosystems alike.

Q12: How will we approach this safely?

Ans: Approaching ”Seed Germination Prediction Using Machine Learning” safely
involves careful consideration of ethical, social, and environmental implications,
as well as proactive measures to address potential risks and ensure responsible
implementation. Firstly, ensuring data privacy and security is paramount, with
stringent measures in place to protect sensitive information and adhere to data
governance standards. Transparent and inclusive decision-making processes must
be established, involving stakeholders from diverse backgrounds to promote equity,
accountability, and trust. Moreover, ongoing monitoring and evaluation are essen-
tial to assess the performance and impact of predictive models, identify potential
biases or unintended consequences, and iterate on model development accordingly.
Collaborative partnerships between researchers, practitioners, policymakers, and
affected communities foster knowledge exchange and collective problem-solving,
enhancing the effectiveness and legitimacy of predictive modeling efforts. Addi-
tionally, promoting education and awareness about the benefits and limitations of
machine learning in agriculture and conservation empowers stakeholders to make
informed decisions and participate meaningfully in the decision-making process.
By prioritizing safety, ethics, and stakeholder engagement, we can harness the
potential of seed germination prediction using machine learning to drive positive
outcomes while minimizing risks to individuals, communities, and ecosystems.

135
 SOCME

Department of Instrumentation Engineering

SOCME (System-Oriented Concept Map Extension)

The SOCME (System Oriented Concept Map Extension) for seed germination
prediction using machine learning involves several key components:
Data Collection and Preprocessing:

ˆ Gather a dataset with relevant features such as seed type, environmental

conditions (temperature, humidity, light), soil type, and possibly genetic
information.

ˆ Preprocess the data by handling missing values, encoding categorical vari-

ables, and normalising numerical features.

Feature Engineering:

ˆ Identify relevant features that may influence seed germination.

ˆ Extract additional features if needed, such as derived environmental variables

or genetic markers.

Model Selection and Training:

ˆ Choose appropriate machine learning algorithms (e.g., decision trees, random

forests, support vector machines) for seed germination prediction.

ˆ Split the dataset into training and testing sets for model evaluation.

ˆ Train multiple models and tune hyperparameters to optimise performance.

136
 Model Evaluation:
ˆ Assess the performance of each model using metrics like accuracy, precision,
recall, and F1-score.
ˆ Validate the models using cross-validation techniques to ensure robustness.
Integration and Deployment:
ˆ Integrate the trained model into a user-friendly interface or application.
ˆ Ensure scalability and efficiency for real-time or batch prediction of seed
germination.
ˆ Deploy the system in relevant environments such as agriculture labs or seed
production facilities.
Continuous Improvement:
ˆ Monitor the performance of the deployed system and collect feedback.
ˆ Fine-tune the model periodically using additional data or updated techniques
to improve accuracy and reliability.
Throughout the SOCME process, collaboration between domain experts, data
scientists, and stakeholders is essential to ensure the successful development and
deployment of the seed germination prediction system.

Sign of Guide: ................

137
Paper Publication

138
Plagiarism Reports

139
 Seed Germination Prediction using Machine Learning

Abstract : For seed researchers, evaluating seed germination is a crucial step in assessing the
effectiveness and quality of seeds. Seed evaluations are often completed by hand, which is a
laborious, time-consuming, and prone to mistake procedure. Because of the tremendous
advancements in photography, electronics, and computer science over the last several decades,
infrared thermal imaging technology has seen several1 technical advancements . clever uses for
quality control and non-destructive testing of many agro-food products. Thermal imaging is a
potential non-contact imaging technique for evaluating different quality attributes because it is
1
based on the infrared radiation released from target foods. Since large-scale germination
investigations often rely on human color-based threshold adjustments, they are not well suited
for conventional picture analysis methods.
Here, we provide a machine learning method for high-throughput seed germination tests and
precise seed germination detection.

1. Introduction : Seed germination prediction using machine learning leverages

advanced algorithms to analyze and forecast the growth potential of seeds under various
conditions. By integrating vast datasets on environmental factors, seed characteristics,
and historical germination outcomes, machine learning models can accurately predict
germination rates and identify optimal conditions for seed development. This approach
not only enhances agricultural productivity but also aids in resource management and
sustainability efforts, providing a data-driven foundation for improving crop yields and
ensuring food security.

1.1 Seed Germination : The biological process of a seed turning into a sprout or a
budding plant is called germination. Everything begins when a seed, buried in the earth,
comes into contact with the proper elements—moisture, warmth, and occasionally light.
Water is absorbed by the seed, which causes it to grow and release stored enzymes. These
enzymes set off metabolic processes that decompose the seed's stored nutrients. The seed
coat rips apart as a consequence, and a little root appears to look for nutrients in the soil.
A shoot simultaneously thrusts forward, toward the sun. Eventually, this sprout will begin
to unfurl its leaves, propelled by the stored nutrients. A plant's autonomous existence
begins with germination, when it stops depending on the seed and starts producing its own
energy through photosynthesis.It marks the beginning of a new plant's journey towards
development and maturity, much like nature's alarm clock.

1.1.1 Types of Seed Germination : There are two primary processes by which seeds can
germinate: epigeal and hypogeal. Plants have evolved several kinds of germinations as
adaptations to maximize their development in various conditions.

Epigeal

When a seed germinates epigeally, the growing stalk emerges above the soil's surface.
The cotyledons, or seed leaves, are raised into the air during this phase and develop into
the young plant's first true green leaves. This kind of germination is typical of plants like
sunflowers and beans. The cotyledons are exposed to sunlight for photosynthesis as the
shoot rises and carries them along with it. With this technique, the plant may effectively
use soil nutrients and sunshine for its early growth and development.

Hypogeal

When a seed germinates hypogeally, the growing branch remains below the soil's surface.
During this phase, the stem elongates to force the first genuine leaves above earth, leaving
the cotyledons (seed leaves) buried. Plants that frequently display hypogeal germination
are peas and peanuts. Using this strategy, the cotyledons act as nutrient stores for the
growing seedling. This tactic shields the tender branch from damage during the early
phases of growth while enabling the young plant to obtain vital resources.

1.1.2 Germination Process : The germination process is a critical phase in the life cycle
of a seed, marking its transition from dormancy to active growth. This process involves
several stages, each essential for the successful development of a new plant. Here is a
detailed look at the germination process:
 Imbibition: The first step in germination is imbibition, where the dry seed absorbs
water through its seed coat. This water uptake causes the seed to swell and the seed
coat to soften, making it permeable to gases and soluble nutrients.
 Activation of Metabolism: As the seed absorbs water, its metabolic processes are
reactivated. Enzymes that were previously inactive during dormancy become active,
leading to the breakdown of stored food reserves in the seed. These reserves,
typically starches, proteins, and lipids, are converted into simpler molecules that
provide energy and building materials for the growing embryo.
 Respiration: During germination, the seed's respiration rate increases significantly.
Oxygen is consumed, and carbon dioxide is produced as the seed converts its stored
food into energy. Adequate oxygen supply is crucial during this stage, which is why
well-aerated soil is important for successful germination.
 Growth of the Radicle: The radicle, or embryonic root, is the first part of the seedling
to emerge from the seed. It grows downward into the soil, anchoring the plant and
beginning the process of water and nutrient absorption. The emergence of the radicle
is often considered the definitive sign that germination has occurred.
 Emergence of the Shoot: Following the radicle, the shoot (or plumule) begins to grow
upwards towards the soil surface. This shoot consists of the stem and the first leaves
(cotyledons). The growth of the shoot is facilitated by the elongation of cells in the
hypocotyl (the stem-like part of the seedling below the cotyledons) and the epicotyl
(the part above the cotyledons).
 Development of True Leaves: Once the shoot reaches the surface and is exposed to
light, the cotyledons open up and photosynthesis begins. This process provides the
 seedling with the energy required for further growth. The true leaves, which will be
the plant’s main photosynthetic organs, start to develop after the cotyledons.
 Establishment of the Seedling: As the seedling continues to grow, it becomes less
reliant on the seed's food reserves and more dependent on photosynthesis and nutrient
uptake from the soil. The root system expands, the shoot elongates, and the plant
establishes itself as a young seedling capable of independent growth.

Throughout these stages, various environmental factors such as temperature, light,

moisture, and soil conditions play crucial roles in influencing the success and speed of
germination. Optimal conditions vary among different plant species, making
understanding these requirements essential for effective cultivation and seedling
establishment.

1.3 Germination Rate : The germination rate is a measure of the proportion of seeds that
successfully sprout and begin to grow over a specific period. It is an important indicator
of seed viability and quality, influencing agricultural productivity and efficiency. The
percentage of seeds that successfully germinate in a given sample, typically expressed
over a specific time frame.
Germination rate can calculate as suppose 100 seed is taken for germination test from a
batch of 1000 seeds and that 100 seeds are put to germination. After germination process
from that 100 seeds 70 seed are germinated and 30 are not germination , so the
germination rate of that particular batch will be 70%.
1.4 Background of Project : The assessment of seed quality is typically performed by
professional technicians, relying on their individual expertise and taking considerable
time. To simplify seed grading for producers, researchers have developed machine vision-
oriented methods to monitor seed quality attributes such as texture, color, and form. These
models enable automated and rapid acquisition of relevant variables. However, the results
of seed grading, whether through automated 5computer classification or technician
assessment, must be validated by the conventional germination test.

This germination test usually takes two to ten days, depending on the seed type, and
requires expert personnel to manually count germinated seeds. If the results of the
samples fall outside the acceptable range, the test must be repeated, demanding significant
effort from skilled experts. Due to the unique nature of germination tests and their
dependence on skilled professionals for accurate classification, there is a pressing need for
a reliable, automated system to examine seed germination with greater stability,
independence, and accuracy across species.

The automated germination test monitoring system in the study now employs traditional
image analysis techniques. The conventional method for testing seed vigor3 is labor-
intensive and challenging, leading to its infrequent use. Additionally, many seed tests
established by the International Seed Testing Association (ISTA) are manually assessed
using protocols that vary by crop. Researchers have developed automatic methods to
 5
minimize errors and reduce manual steps in the seed testing process. Modern image
analysis techniques, being automated, produce quantitative, unbiased measures with
minimal errors and have thus been utilized to detect seeds.

1.5 Objectives :
3
1. To reduce the number of manual steps in seed testing, which is highly error-prone.
9
2. To automate the process of seed germination testing.

2. Literature Survey :
5
In this chapter, a brief survey of the present germination classification model has
been mentioned.
1. Here Nikita Genze et al. [1] introduce a machine learning approach for high-
3
throughput seed germination studies and reliable detection using modern artificial
neural networks with region proposals. They compiled tagged image data on the
3
germination stages of over 2,400 seeds from three different crops: Zea mays (maize),
Secale cereale (rye), and Pennisetum glaucum (pearl millet), utilizing approximately
3
23,000 images in total. Transfer learning was employed to train various state-of-the-
9
art convolutional neural network (CNN) architectures with region proposals to
automatically identify seedlings in petri dishes and predict their germination
potential.
There suggested machine learning-based approach can facilitate the evaluation of
seed germination. Experiments with several seed varieties. In comparison to
traditional and manual approaches, it has reduced error rates and higher performance,
resulting in more recise germination indices and seed quality evaluations.

2. In this study Srinath Yasam , et al. [2] the creation of the Automated Germination
Classification utilising ORB Feature Extractor with Machine Learning
(AGCORBML) model is the main emphasis of this study, according to Srinath
Yasam, et al. [2]. For the purposes of determining seed quality and detecting
germination, the proposed AGCORB-ML approach employs hand-crafted features
and ML classifiers. The recently developed AGCORB-ML approach uses the ORB
technique to extract a suitable set of feature vectors from8the germination images. To
categorize germination, two machine learning models—the support vector machine
(SVM) and the extreme learning machine (ELM)—are used. An comprehensive
range of trials were conducted to show the AGCORB-ML technique’s effectiveness
in germination categorization. According on the simulation results, the AGCORB-
ML approach outperforms existing ML models in a promising manner.

3. In this study, Amjan Shaik, et al.[3] designed an MLA to extract pertinent

attributes of a seed that results in knobging the inconsistent targets, i.e., it minimises
cardinality of features and uniqueness is kept, which helps the farmer choose that
specific seed. The seed’s chosen characteristics are then used for training and testing.
40\% of the data is used for testing, while 60\% is used for training. To show the
effectiveness of the chosen features of a seed, three MLA classifiers—Decision Tree,
SVM, and KNN—were used in total. To test these algorithms, a big, well-known
dataset of benchmark seeds is employed in the experiment. The seed used in this
study is from a dataset of three different wheat varieties: kama, rosa, and Canadian.
On these datasets, the MLA has been used to control the pertinent features that
describe the growth of seed, and a classification technique has subsequently been
used to show the accuracy of the features that have been chosen. There goal in using
this effective strategy is to assist farmers in choosing quality seeds for the
development of the nation’s agriculture and economy while saving them money,
time, and energy.

4. The development of technologies for managing the germination process and

enhancing seed quality is driven by the aim to increase agricultural output.The three-
stage simulated method utilized in this study assigns a good, poor, or neutral grade to
4
germination. A machine learning-based approach for classifying seed germination is
4
provided by M. R. Kumar et al. [4]. Every seed chosen for the study is dynamically
4
analyzed using an evaluation technique known as PBSGA (Pivot Based Seed
Germination Evaluation), which is based on statistical methods. Two scenarios are
used to train and test the model. Before comparing different baskets from the
KAGGLE dataset, a machine learning method for determining the quality of rice
4
seeds (RSGA) is used. It demonstrates the effectiveness of the PBSGA recommended
method in selecting the suitable model category and conceptualizing successive
improvements, such as the difference in quality in testing batches that were
conducted sooner and later. The experimental study's findings provide an indication
of the model's potential and the direction that multiclass labeling research will take
going forward.

5. Nehoshtan, Y., et al. present the first-ever general germination prediction method
here [5]. It makes it simpler to categorize seeds based on their usefulness and
capacity to germinate—two characteristics that influence the outcome of
2
germination—and is based on deep learning and RGB image data. By making
2
hundreds of rejected seed batches from seven different vegetable crops—representing
different genotypes and production pipelines—industrially acceptable, they exhibit
their technological expertise.
We also demonstrate the ability to accurately categories lots using crop-level picture
2
data rather than lot-specific data. These successes mark a significant turning point in
the application of this technique for industrial seed sorting by germination outcome
for various crops. There suggested method builds on prior attempts to use optics to
forecast seed germination efficiently and without harm,2 but unlike them, it makes use
of colour imaging and neural network classification. In a cohort of seven distinct
crops, dispersed across a wide phylogeny and represented by at least two lots, bred
 7
and produced by twelve distinct enterprises, we used it to robustly enrich seed
germinability and usability, two distinct features of germination, to 90\% precision.
2
This new capability is anticipated to allow seed companies to 2separate seeds of
various crops based on their germination potential for the first time in high-fidelity, in
a single separation process, and without obligatory seed priming. This will streamline
2
the process of seed germination enrich- ment and make lots that, using modern
separation techniques, would have failed enrichment attempts, such as the lots used in
this study, industrially useful. Unfortunately, they only know the cause of poor
germination for a small number of the lots, which was either genetics production, or
prolonged storage. Those lots encompass the potential causes of mal-germination,
and their classification satisfied the requirements of 90% precision and 80% recall,
proving the method’s adaptability to various causes of mal-germination.

3. Methodology
To predict seed germination using machine learning, we integrate advanced hardware
and software tools to create a comprehensive and efficient methodology. The
hardware setup includes an infrared thermal camera to monitor temperature variations
during germination, germination paper to provide a consistent medium, and a
controlled germination chamber to maintain optimal environmental conditions. These
tools enable the precise collection of thermal images and environmental data, which
are essential for accurate predictions. The software aspect involves using Jupyter
Notebook for data preprocessing, feature extraction, and model training, leveraging
Python libraries like OpenCV, pandas, and scikit-learn. Additionally, Streamlit is
employed to develop an interactive web application, allowing users to input new data
and visualize predictions in real-time.

3.1 Hardware Requirements :

3.1.1 Infrared Thermal Imager

10
 TFT full view color screen display.
 Dual MCU processors, refresh faster.
 Users may navigate to the hot and cold spots in a thermal picture by using the hot and
cold point temperature indicators.
 Selectable palette
 hybrid imaging using both infrared and visible data.
6
 Images capture saved in SD.
 Connecting PC with USB can read the image stored in instrument directly.

3.1.2 Germination Paper

This specific kind of paper is used as a substratum in the germination test. This paper
is ideal for checking the ideal moisture content of the different types of seeds and
germination forms. Its unique structure and exceptional wet strength assist to inhibit
 the growth of fine seed roots. The provided seed germination paper has a high
capacity for absorption and a pH range of 5-7 with strong bursting power.

The primary criterion used to assess a seed's viability, vigour, and outward
appearance is its germination count. With seed germination paper, the optimal
moisture content is ensured for the greatest range of seeds and germination forms.

3.1.3 Germination Chamber

An enclosed space with adjustable humidity and temperature is called a germination
chamber. Often, trays are stacked so that it is easy to add and remove trays and to
see which ones have germinated.
Without entering the field, precise environmental conditions are produced within the
chamber to conduct various experiments.

3.2 Software Requirements:

3.2.1 Jupyter Notebook:

Interactive notebook papers with real-time code, equations, graphics, video, and
other computational outputs can be created with Jupyter Notebook. Programmers,
data scientists, and students frequently use Jupyter Notebook to record and illustrate
coding processes or just to play around with code.

One kind of integrated development environment (IDE) is Jupyter Notebook. IDEs

are code writing environments with certain auxiliary capabilities. Debugging, code
completion, and syntax highlighting are features found in almost all IDEs.

3.2.2. Streamlit
With the help of the open-source Python module Streamlit, developers can quickly
create eye-catching user interfaces. The simplest method for integrating code into a
web site, particularly for those without any prior front-end experience, is to use
Streamlit: There is no need for prior front-end (html, js, css) expertise or experience.
A free and open-source framework called Streamlit makes it easy to create and
distribute stunning online applications for data science and machine learning. It is a
Python-based library created with machine learning developers in mind. online
developers are neither data scientists or machine learning engineers, and they have
little interest in spending weeks learning how to create online applications using
these frameworks. As long as the tool can present data and gather the necessary
parameters for modelling, they would prefer something simpler to understand and
operate. Streamlit helps you to develop a stunning-looking application with only a
few lines of code.
4. Block Diagram:

The block diagram for seed germination prediction using machine learning outlines a
systematic workflow integrating both hardware and software components.
Initially, the hardware setup includes an infrared thermal camera capturing
temperature profiles of seeds on germination paper within a controlled germination
chamber, ensuring consistent environmental conditions. The captured thermal images
and environmental data are then fed into the preprocessing block, where image
processing techniques extract relevant features and synchronize them with
environmental parameters. These preprocessed features are input into the machine
learning model development block, where algorithms like decision trees, random
forests, and neural networks are trained and validated using labelled data. The final
block involves deploying the trained model via a Streamlit-based interactive web
application, allowing users to input new data, visualize predictions, and make
informed decisions about seed germination. This comprehensive diagram illustrates
the seamless integration of hardware data acquisition with advanced machine
learning techniques and user-friendly software interfaces.

To conduct the seed germination process, the methodology involves several steps.
Firstly, three germination papers are acquired and thoroughly wetted to ensure
adequate moisture for seed germination. Next, soybean seeds are carefully placed
onto two of the wetted germination papers, with appropriate spacing between each
seed. These seeds are then covered with the third germination paper, forming a
sandwich-like structure to promote germination. Subsequently, the prepared seeds are
left undisturbed for a duration of 8 days to initiate the germination process under
controlled environmental conditions. After the incubation period, thermal images of
the soybean seeds are captured using specialised thermal imaging equipment. These
thermal images serve as valuable data to discern the germination status of the seeds.

The next step involves organising the thermal images into a comprehensive dataset,
meticulously labelling each image according to the germination status of the
corresponding seed—whether germinated or dead. This dataset compilation is crucial
for training and testing deep learning algorithms. Finally, deep learning algorithms,
such as convolutional neural networks (CNNs), are employed to classify the seeds
based on the thermal images. Through meticulous training, validation, and testing
processes, the deep learning model can effectively distinguish between germinated
and dead seeds with high accuracy. This comprehensive methodology ensures a
systematic approach to analyse and classify soybean seeds' germination status,
offering valuable insights for agricultural research and practice.
5. Design and Implementation:
The design and implementation of a seed germination prediction system using
machine learning involves integrating thermal imaging systems, both active and
passive, to capture comprehensive temperature profiles of seeds during the
germination process. In the design phase, we employ a thermal imaging system
where infrared cameras continuously monitor the seeds on germination paper within a
controlled environment. Both active thermal imaging, which involves external heat
sources, and passive thermal imaging, relying on natural heat emission from the
seeds, are utilized to ensure accurate data collection. The processing of thermal
images includes extracting relevant features that reflect germination activity, such as
temperature patterns and heat distribution. Factors affecting the quality of thermal
images, such as camera resolution, environmental stability, and seed placement, are
meticulously managed to maintain data integrity. In the implementation phase, these
processed thermal images, along with environmental parameters, are fed into
machine learning models to predict germination outcomes. The system leverages
software tools like Jupyter Notebook for data processing and model training, and
Streamlit for developing an interactive application, thus providing a robust and user-
friendly platform for accurate seed germination prediction.

5.1 Design :

5.1.1 Thermal Imaging System :

1
The traditional
1
infrared thermal imaging system for assessing seed quality comprises
two primary components: (1) a thermal camera with optical units, including a focused
lens, collimating lens, and filters, and (2) a thermal detector array that features1
microbolometers with temperature stabilization or cooling mechanisms, along with a
control and1 display unit equipped with signal and image processing tools. To
accurately record the infrared radiation emitted by the seeds, the system must
function with high precision and resolution, requiring further refinement of its
components.
1
It's also important to remember
1
that the thermal imaging system lacks a lighting
element, in contrast to the other color or spectrum imaging systems. The signal
processing unit's 1job is to translate the incoming electrical signals into thermal
records that show the target object's surface temperature distribution in the scene. As
a result, the thermal image created by this technique is a matrix of several color
levels, each of which represents
1
a distinct temperature. Thermal imaging technology
scans at a pace of around 30 images per second with a sensitivity range of -20 to
1500°C while evaluating seed quality. By utilizing certain optical filters, its
sensitivity can be increased even further.
5.1.2 Active and Passive Thermal Imaging System :

In order for thermal imaging systems to be used practically, the target object's
temperature has to vary from its surroundings. This means that before the imaging
begins, the item must be either an energy source or a sink, or it must be heated or
cooled. Furthermore, it is crucial

To take into account the possibility that the target items' temperature may fluctuate
while being observed. Essentially, there are two methods for employing thermal
imaging to discern objects: either using the varying heat conductivities or capacities
of the object materials, or using differing emissivities.
Infrared light emitted from two objects can be used to directly differentiate them if
their inherent emissivity coefficients differ. As an alternative, covering the items
with a heat pulse and because of the variations in their heat conductivities, it is
possible to distinguish between things by monitoring the radiation emitted from
them.

As a result, the two fundamental designs for developing an infrared thermographic

imaging system in seed quality monitoring are active thermal imaging and passive
thermal imaging, which both rely on the emissivity concept.Imaging method (which
is based on the theory of heat conductivity). When the system is in a passive state,
no external energy is used to monitor the target body's thermal radiation emissions
under ambient circumstances. When the items being evaluated naturally produce
heat, passive thermography is typically employed. Thus, for non-contact temperature
and to characterize the surface thermal characteristics of the items under research,
the passive-based approach is typically utilized.Measurement in the course of
processing. Additionally, as insects in seed lots produce a lot more heat through their
respiration than do sound seeds, it might be utilized to identify insect infestation in
seeds.In the second method of thermal imaging, emissivity often fails to produce
infrared images with sufficient contrast. Therefore, it is advisable to heat or cool the
samples before capturing thermal image frames.outside energy source in order to
provide a sufficient thermal contrast. In these situations, a thermal imaging
technique known as active thermal imaging is employed, in which the target items
are heated or rapidly cooled in order to deposit external energy before taking an
image.
1
5.1.3 Processing of thermal images :

The true potential of any imaging system lies in its ability to provide quantitative data
on the spatial variation of the measured quantity. Thermographs can independently
describe various characteristics of the imaged objects or indirectly link thermal data
to hidden attributes. To accurately evaluate surface temperature data captured in
thermal imaging, one must understand all the processes that can affect an object's
temperature. A thermograph taken by a thermal camera can be analyzed using various
 image processing techniques. The choice of technique for thermal imaging data
analysis depends on the application's primary goal and the required speed of decision-
making. In industrial settings, the processing method for thermal image analysis must
be quick and straightforward to enable real-time task completion. Current thermal
imaging cameras come pre-programmed with processing methods to estimate the
temperature of target objects in the scene.

5.1.4 Factors Influencing Thermal Image Quality

When using thermal imaging for real-time, contactless temperature measurement,

objects within the camera's field of view must emit radiation that reaches the detector
1
at a detectable intensity. Infrared radiation must travel from the object's surface to the
detector inside the camera housing. However, this radiation is often1
weakened during
its journey due to environmental factors1 and the materials of the focusing optics and
filters it passes through. Understanding1the scattering and absorption processes in the
atmosphere is crucial for calculating the attenuation of infrared radiation, which
depends on the concentration of absorbing gas species and the relative humidity.
Several variables significantly influence both the capture and interpretation of
infrared thermal images. Although it might be challenging to regulate every one of
these elements in reality, understanding them is crucial for many application
schemes.
1
Since the main objective of thermal imaging 1
is to record the surface temperature of
every object in the camera's field of view based on the thermal radiation emitted from
these objects, factors that attenuate or augment thermal radiation before it reaches the
detector should be carefully considered. 1
For instance, one way that specific air parameters, including temperature, relative
humidity, and wind speed, affect the transmission of infrared light from the target
object to the thermal imaging system is by attenuation through absorption or
scattering.

5.2 Implementation

5.2.1 Germination Test :

The warm germination test serves as the standard method for assessing the viability
of seed batches. In compliance with the Nebraska State Seed Law, commercially
marketed seed must display its germination percentage on the seed tag. This test
subjects the seed to optimal conditions, including ideal light, temperature, and
relative humidity, which vary depending on the plant species under examination. For
conducting the warm germination test on large-seeded crops like soybean and corn,
the necessary supplies include a seed counter tray, a seed-retaining container, and
two moistened paper towels.
 The seed counter tray is filled with seeds, ensuring each hole is filled and excess
seed is drained. Upon opening a sliding drawer, fifty seeds are dispensed onto the
paper towel, ensuring uniform placement after removing the counter tray. The seeds
are then covered with a thick and a thin, moistened paper towel, with their edges
sealed together to prevent drying out. The bottom edges of the towels are folded to
prevent seeds from rolling out, and the towels are rolled up for easier handling. The
paper towel is marked with the seed lot number using pencils designed for writing on
damp paper.

The thick paper towel and the thin, moist paper towel are layered over the seeds, and
the edges of the paper towels are squeezed together to seal them.To prevent the seed
from rolling out, the towels' bottoms are folded. I've rolled up the towel. In addition
to making handling easier, this prevents the towel from drying out. For identifying
purposes, the towel has the seed lot number in writing. Pencils designed specifically
to write on damp paper are used.A germination chamber, equipped with a humidifier
to control relative humidity, regulates light, temperature, and humidity levels. The
temperature settings within the germination chamber are adjusted based on the
optimal germination temperatures for the specific plant species being tested.
Temperature and humidity levels inside the chamber are monitored and recorded
throughout the germination process.

The rolled paper towels are placed plastic sacks to prevent drying, placed in a
germination chamber with the proper temperature, humidity and light.The towels are
taken out of the chamber after the designated period of time, which is usually one
week.The paper towels are unrolled, completely uncovering the germinating
seedlings. Ungerminated seeds and germinated seedlings are separated and counted,
and the germination percentage is calculated.

6. Experimentation and Results :

The Experimentation and Result phase of seed germination prediction using machine
learning involves evaluating the performance of the developed models and
interpreting their predictions. By analyzing the data collected from thermal imaging
systems and environmental sensors, we assess the model's accuracy, precision, and
recall in predicting seed germination outcomes. The evaluation metrics indicate how
well the machine learning algorithms, such as decision trees, random forests, and
neural networks, have learned from the training data. Detailed analysis of the results
helps identify patterns and correlations between thermal image features and
germination success.

6.1 Experimentation :
Soybean seeds were positioned on germination paper so that various germination
stages could be photographed with infrared thermal imaging. To provide a sharp
contrast between the developing radicle and the background, the seeds were put on a
black fabric during the thermal picture capture. To lessen the impact of water
evaporation and the consequent drying out of the seeds, the cloth was moistened with
tap water and the germination paper was covered with butter paper. Reflections from
the camera setup were among the reflections that the lid occasionally produced. Each
picture was taken in an office at room temperature using the same artificial light
source. For this purpose, in a single batch of germination 10 seed samples were put
for germination. Seed Germination required total 8 days to get fully germinated but
on-wards 6th day we are able to see that seed is germinating. Seed Germination was
captured over a time span of 1 days with a resolution of 33x33 (1089 pixels). First we
have captured initial images of seed on day 1 before putting for germination than on-
wards 6th day we have captured images everyday till 9th day.
For this purpose, in a single batch of germination 10 seed samples were put for
germination. Seed Germination required total 8 days to get fully germinated but on-
wards 6th day we are able to see that seed is germinating. Seed Germination was
captured over a time span of 1 days with a resolution of 33x33 (1089 pixels). First we
have captured initial images of seed on day 1 before putting for germination than on-
wards 6th day we have captured images everyday till 9th day.

These samples were germinated at a temperature of 20 degree Celsius to 25 degree

Celsius this played a major role in the time required for germination.
 Temperature = 20 to 25 degree Celsius
 Emissivity = 0.96
 Light Intensity = 50 percent

Additionally, the analysis highlights the model's robustness in different

environmental conditions and its ability to generalize to new, unseen data. The
findings from this phase provide insights into the effectiveness of the predictive
system, guiding further optimization and practical application in agricultural settings.

6.2 Results:
6.2.1 Experiment Analysis:
Total numbers of Seeds Germinated Dead Seeds

360 350 10

Preprocessing:

 Data Collection: Gathered data on various parameters affecting seed germination

such as moisture, temperature, light exposure, and seed type.
 Data Cleaning: Removed duplicate records and handled missing values either
through imputation or deletion.
 Data Transformation:
Normalization/Standardization: Features are scaled to maintain uniformity and
prevent the dominance of particular features.
 Encoding Categorical Variables: Categorical variables are transformed into numerical
representations using methods such as one-hot encoding.

 Data Splitting: The dataset is divided into training and testing sets, ensuring a
balanced representation of both classes.
Model Training:
1. Support Vector Machine (SVM):
 Initialized an SVM model with suitable hyperparameters (e.g., linear or RBF kernel).
 Trained the SVM model on the preprocessed training data.
2. Random Forest:

 Created a Random Forest model, a robust ensemble learning method.

 Trained the Random Forest model on the preprocessed training data.

Evaluation:

1. Accuracy: Achieved an accuracy of 97.14\%, indicating the model's effectiveness

in predicting seed germination outcomes.
2. Epoch Iterations: Iterated over multiple epochs to fine-tune the model parameters
and improve performance.
3. Seed Distribution:
 Germinated Seeds: 350
 Dead Seeds: 10
 The model's ability to predict both outcomes demonstrates its robustness and
effectiveness.

7. Conclusion :

In this project, we have studied a conventional method of seed germination testing.

We observed various stages of seed germination. The experimental outcomes stated
thermal imaging system and the processing of thermal image, factors affecting
thermal images.

We have developed a machine learning-based seed germination prediction model.

We aimed to automate the process of seed germination testing using a machine
learning model. Seed evaluations are often completed by hand, which is a laborious,
slow, and prone to mistake procedure. Owing to the tremendous advancements in
imaging, electronics, and computer science over the past several decades, infrared
thermal imaging has seen many scientific advancements and clever uses in non-
destructive testing and quality monitoring of many agro-food products. Thermal
imaging provides a non-invasive and nondestructive method to assess the
germination process by monitoring temperature variations in seeds. Making ensuring
the camera is calibrated correctly and that the data gathering procedure is systematic
and thoroughly recorded are crucial.
 Similarity Report ID: oid:3618:59919957

9% Overall Similarity
Top sources found in the following databases:
8% Internet database 8% Publications database
Crossref database Crossref Posted Content database
0% Submitted Works database

TOP SOURCES
The sources with the highest number of matches within the submission. Overlapping sources will not be
displayed.

hdl.handle.net
1 4%
Internet

ncbi.nlm.nih.gov
2 1%
Internet

link.springer.com
3 1%
Internet

M. Rudra Kumar, Avinash Sharma, K Sreenivasulu, G. Ramesh. "Pivot B...

4 <1%
Crossref

ijisae.org
5 <1%
Internet

mextech.co.in
6 <1%
Internet

Yuval Nehoshtan, Elad Carmon, Omer Yaniv, Sharon Ayal, Or Rotem. "R...
7 <1%
Crossref

mdpi.com
8 <1%
Internet

Sources overview
 Similarity Report ID: oid:3618:59919957

Nikita Genze, Richa Bharti, Michael Grieb, Sebastian J. Schultheiss, Do...

9 <1%
Crossref

htcinstrument.com
10 <1%
Internet

Sources overview
Powerpoint Presenatation

156
157
158
159
160
161
CD

162