A Project Presentation for the Degree of Bachelor in Computer Engineering
Wildlife Conservation and Analysis using
Machine Learning
Group Member Name
1) Manvi Gawande 3) Vedant Angane
2) Vaidehi Pawar 4) Gaurav Suryavanshi
Under the Guidance of
Dr. Mahendra Salunke
Pimpri Chinchwad College Of Engineering & Research
A.Y 2023-2024 1
OUTLINE
➔ Introduction ➔ Requirement Analysis
➔ Objectives ➔ Work Breakdown Structure
➔ Challenges ➔ Class Diagram
➔ Literature Survey ➔ Results
➔Problem Statement ➔ Conclusion
➔ Proposed System ➔ Future Work
➔ Methodologies ➔ References
2
INTRODUCTION
● Wildlife plays a vital role in maintaining the balance of ecosystems.
For example, pollinators such as bees and butterflies are essential
for plant reproduction, and predators help to control populations of
prey animals. If wildlife populations decline, it can have a
cascading effect on the entire ecosystem.
● Wildlife is a vital part of biodiversity, which refers to the variety of
life on Earth. Biodiversity is important for healthy ecosystems and
for human well-being. So, there is a need to design smart or
intelligent systems through advanced technology that can monitor
wildlife conservation.
3
INTRODUCTION
● The limitations of existing approaches is that they only focus on tagged animals.
● The project focuses on classifying animals as endangered or extinct using images,
videos and summarizes animal information from across the world, including
migration and biodiversity of animals in various national parks.
● In our project, we develop a detection method that involves tracking the location of
the animals and classification of the animals.
4
OBJECTIVES
● For accurate wildlife population counts from images and videos, it is
important to develop tools that incorporate computer vision and machine
learning.
● To create species profiles that compile detailed information on population
trends, habitat characteristics, identified threats, and distribution data.
● To make informed conservation decisions, it is important to categorize species
according to the IUCN Red List criteria for endangerment classification.
5
● To make tailored conservation strategies are aided by data collected on
animal migration and biodiversity in national parks, providing insights
into biodiversity
● To develop a regularly maintained unified platform to effortlessly access
and combine data.
CHALLENGES
● Manually labelling large datasets was time consuming.
● Wildlife images for certain animals were difficult to find.
● Animals images needed to captured in different poses as affects how their bodies
appear in the image.
● Busy backgrounds with cluttered environments can made it difficult for the model.
7
LITERATURE SURVEY
AUTHOR TITLE METHODOLOGY LIMITATIONS
M.Mangale- swaran and M. A Comparison of Different CNN Effective wildlife detection
Azhagiri [2023] Learning Algorithms for SVM method in conservation remains
RandomForestClassifier
Wildlife Detection and unclear.
Classification in Animal
Conservation Applications.
Miao, Z., Gaynor, K.M., Wang, Insights and approaches using CNNs Overlooked Object Focus: The
J. et al.[2019]. deep learning to classify VGG-16 CNN's tendency to focus on
wildlife. EUCLIDEAN DISTANCE background elements rather than
target objects or animals may
limit its ability to accurately
classify species.
8
AUTHOR TITLE METHODOLOGY LIMITATIONS
N. K. Jana, A. Classification of Modified CNN Model's high accuracy may not generalize well
Chowdhury, D. Saha, Animal Images with to all animal species and may require careful
C. Dutta, and A. Maiti "MODIFIED CNN" parameter tuning and could potentially
Oct. [2022]. model. introduce noise or artifacts into the
classification process.
UAVs
Tuia, D., Kellenberger, Perspectives in Analyzing large, diverse sensor data
Remote Sensors
B., Beery, S. et machine learning for efficiently demands significant computational
al.[2022] wildlife conservation. resources.
9
PROBLEM STATEMENT
● To classify wildlife using images and videos and to visualize information about
wildlife from various regions around the globe.
10
PROPOSED SYSTEM
11
FEASIBILITY STUDY
Software Backend
● Jupyter Notebook ● Python
● Streamlit Python ● Numpy
Framework ● Matplotlib
● Tensorflow
Hardware ● Keras
● CPU Speed: 3.2 GHz ● YoloV5
● RAM: 16GB ● InceptionV3
● HardDisk: 1TB 12
METHODOLOGIES
Data collection and pre-processing
● Crowdsourcing: Gather images and videos of wildlife from various
contributors around the globe using a dedicated platform.
● Data Storage: Store the collected data in a structured format for easy access
and processing.
● Image Resizing: Resize images to a uniform size compatible with Inception
V3, typically 299x299 pixels.
13
METHODOLOGIES
Annotation
● Labeling: Annotate images with relevant labels indicating the
species or other pertinent information. This step provides ground
truth for supervised learning.
● Bounding Boxes: Mark regions of interest within the images
(optional, depending on the dataset and task).
14
METHODOLOGIES
Feature Extraction Layer:
Use the convolutional layers of Inception V3 as a feature extractor. These layers
will process the input images to generate feature maps that capture important visual
characteristics.
● Activation Maps: Extract activation maps from intermediate layers of the model.
These maps represent different levels of features, such as edges, textures, and
shapes.
● Global Average Pooling: Apply global average pooling to reduce the spatial
dimensions of the feature maps while preserving the most important features.
15
METHODOLOGIES
Model Training and Testing:
● Train and fine-tune both Inception V3 and YOLOv5 models on the annotated
dataset.
● Evaluate the models on separate validation and test sets to monitor performance and
adjust hyperparameters.
16
METHODOLOGIES
Visualization:
● Create interactive maps and dashboards to visualize wildlife data, showing species
distribution, population trends, and other relevant metrics.
Animal Profiling:
● Integrate data from the IUCN Red List to profile the conservation status of detected
species.
● Include information such as species vulnerability, population trends, and threats.
17
REQUIREMENT ANALYSIS
● Use of GPS collar locators/satellite imagery/camera traps.
● Python for implementing Algorithm
● Python Streamlite for the app
18
SOFTWARE REQUIREMENT:
● Operating System : ANY
● IDE : Google Colab
● Language : Python
● Libraries : Opencv, scipy, tensorflow,sklearn
HARDWARE REQUIREMENT:
● Any Laptop having configuration above 4 GB RAM
19
ALGORITHMS/METHODS
InceptionV3
1. Advanced CNN Architecture: Builds on InceptionV1 and V2 for superior image
classification.
2. Efficient Feature Extraction: Utilizes parallel convolutions with varying filter sizes.
3. Diverse Feature Capture: Captures a wide range of spatial and spectral features for
better classification.
4. Optimized Training: Incorporates batch normalization and dimensionality reduction to
prevent overfitting.
5. Pre-trained Weights: Ready-to-use for various image classification tasks or fine-tuning.
20
ALGORITHMS/METHODS
YOLOv5
6. Real-time Object Detection: Detects objects in images and videos
swiftly and accurately.
21
ALGORITHMS/METHODS
7. Speed-Accuracy Balance: Offers different model sizes (s, m, l, x) to
balance speed and accuracy.
8. CNN Backbone: Uses a powerful CNN backbone to extract and combine
features for detection and classification.
22
WORK BREAKDOWN STRUCTURE
Activity Time required (days) Description
A 5 Data collection (Behavioral, Vehicular, Physiological)
B 10 Implementing Dataset Preprocessing
C 10 Implementing Feature Extraction
D 20 Implementing visualization of Animals
E 10 Implementing Detection and Classification of Animals
F 15 Integrating the ML model with the web application
G 5 Evaluation and improvements based on Testing
23
H 5 Report Writing and Plagiarism Check
CLASS DIAGRAM
24
RESULTS
● Screenshot of UI
25
RESULTS
● Screenshot of UI
26
RESULTS
● Screenshot of UI
27
RESULTS
● Screenshot of UI
28
CONCLUSION
Summarization of the research findings, highlighting the
effectiveness of the proposed technique in significantly improving
wildlife conservation tasks such as monitoring animal
populations, tracking the movement of animals, and detecting
threats to their survival.. Emphasis on the system's potential to
revolutionize wildlife conservation, helping in maintain a count of
animals
29
FUTURE WORKS
● Understanding Social Interactions among the animals
● Identifying the species using sound
● Detecting disease Outbreak
● Monitoring the habitat to detect habitat degradation or resource scarcity
30
REFERENCES
● M. A. Thalor, R. Nagabhyrava, K. Rajkumar, A. Chakraborty, R. Singh and U. Singh Aswal, "Deep
learning insights and methods for classifying wildlife," 2023 3rd International Conference on Advance
Computing and Innovative Technologies in Engineering (ICACITE), Greater Noida, India, 2023, pp.
403-407, doi: 10.1109/ICACITE57410.2023.10183057.
● M. Mangaleswaran and M. Azhagiri, "A Comparison of Different Learning Algorithms for Wildlife
Detection and Classification in Animal Conservation Applications," 2023 Third International
Conference on Artificial Intelligence and Smart Energy (ICAIS), Coimbatore, India, 2023, pp.
920-923, doi: 10.1109/ICAIS56108.2023.10073833.
● Duhart, Clément & Dublon, Gershon & Mayton, Brian & Davenport, Glorianna & Paradiso, Joseph.
(2019). Deep Learning for Wildlife Conservation and Restoration Efforts.
● Aarju, R. Bahuguna, S. Pandey, R. Singh, H. Kaur and G. Chhabra, "Enabling Technologies for
Wildlife Conservation," 2023 IEEE Devices for Integrated Circuit (DevIC), Kalyani, India, 2023, pp.
31
217-220, doi: 10.1109/DevIC57758.2023.10134561.
REFERENCES
● Miao, Z., Gaynor, K.M., Wang, J. et al. Insights and approaches using deep
learning to classify wildlife. Sci Rep 9, 8137
(2019).https://doi.org/10.1038/s41598-019-44565-w
● Otto, S. P., & Day, T. (2011). A biologist's guide to mathematical modeling in
ecology and evolution. Princeton University Press
THANK YOU!
33
