Technological University(Mandalay)
Department of Mechatronics Engineering
Fruits Detection
With Raspberry Pi & Camera
First Seminar
20.2.2025
Presented By : Supervised By :
Group-I Dr. Saint Saint Pyone
Group-I member list
Outlines of presentation
Introduction
Aim and Objectives
Block Diagram
Methodology
Types of Detection
Conclusion
Future Work
References
Introduction
In
Aim and Objectives
Aim
The target of this project is to create an automated, efficient, and cost-effective system for
identifying and classifying fruits by using computer vision and machine learning techniques in
real-time.
Objectives
The objectives of this project is as following:
•To detect and classify fruits (orange, line, apple) in real-time.
•To use Raspberry Pi for fruits classification and control robotic arm.
•To implement an object detection model (Slingshot detector, SSD) for fruit recognition.
•To integrate a robot arm with the object detection system.
•To achieve high accuracy in fruit detection and classification.
Hardware Components
Raspberry Pi 4
Pi camera module 3
Hardware Components
Raspberry Pi 4
• The Raspberry Pi 4 is a powerful yet affordable single-board computer
• User-friendly operating system and suitable for hobby and DIY project.
• Support AI models like SSD and camera interface.
• Raspbian (Raspberry Pi OS) supports Python and AI frameworks like TensorFlow and
Pytorch, which are commonly used for object detection .
• Include memory option - 2GB / 4GB / 8GB .
• Contain microSD card storage minimum 16GB (Recommended 32GB or higher).
• Involve 40 GPIO pins and available ports : USB, HDMI, camera, ethernet, audio and others.
Raspberry Pi Camera Module 3
1. The Raspberry Pi Camera Module 3 is an upgraded official camera for Raspberry Pi boards,
offering major improvements over earlier versions.
2. It features a 12MP Sony IMX708 sensor, replacing the previous 8MP IMX219, for sharper
and more detailed images. The larger 1.4μm pixel size improves light capture, enhancing low-
light and nighttime image quality.
3. Available in both standard (75°) and wide-angle (120°) versions, it captures more of the scene
in a single frame. It also supports HDR for better detail in shadows and highlights. NoIR
variants are ideal for night vision and security applications.
4. Its small and lightweight design makes it perfect for robotics, drones, and compact projects. It
also works well with AI and computer vision applications like object detection and facial
recognition.
5. These upgrades make the Camera Module 3 a powerful choice for photography, video
streaming, security, and AI projects.
6. Raspberry Pi camera module captures the fruit images and sends them to a cloud server for
processing. The system uses convolutional neural network (CNN) to extract the features of the
fruit images and SoftMax classifier to classify them into fresh or rotten classes.
This system leverages the capabilities of Raspberry Pi, coupled with a camera module, to capture
and process images of fruits in real-time.
Python programming, machine learning algorithms, and Convolutional Neural Networks (CNNs)
are the backbone of the methodology, enabling the system to perform tasks such as fruit detection.
These datasets serve as the foundation for training and testing our system, allowing it to recognize
and classify various fruit types effectively.
By employing both training and testing concepts, we have not only fine-tuned the system but also
demonstrated its consistent and precise results.
In conclusion, the project embodies the fusion of cutting-edge technology, precision, and a
commitment to continuous improvement.
By harnessing Raspberry Pi, Python, machine learning, and CNNs, a robust fruit detection system
which is capable of delivering accurate and insightful results have been created.
Block Diagram
Pi Camera Raspberry Pi Preprocessing Run SSD
Detection &
Prediction
Robotic Arm
control
Start
Flowchart Initial Raspberry Pi &
Pi Camera
Capture Image from Pi Camera
Preprocess Image[Resize,tensor
from Load Pretrained
Model[SSD]
Load Pretrained Model[SSD]
Pass the preprocessed image to
Model
The model identify and detect
the image
If Fruit detected
Produce Coordinate of bounding
box
Sent signal to Robotic Arm
END
Conclusion
• However, the project also faces challenges. The limited processing power of the Raspberry Pi makes it
difficult to run deep learning models efficiently, requiring optimizations such as model quantization or
hardware accelerators like Coral TPU for faster inference. Additionally, factors like lighting, camera quality,
and dataset accuracy affect detection performance. Improving the dataset with more diverse images and better
annotation can enhance accuracy.
• For future improvements, the system can be optimized for faster and more accurate detection, integrated with
hardware accelerators for better performance, and enhanced with a user-friendly interface for practical
applications. Overall, this project highlights the potential of AI-powered fruit detection and opens doors for
further advancements in smart agriculture and automation.
Future Work
References
