🚀 Are you passionate about leveraging cutting-edge technology to solve real-world problems? Look no further! I’m excited to introduce Fruits Detection, a TensorFlow Lite object detection project built on the foundational work by Edje Electronics and designed to detect fruits with precision using Google Colab.
This project utilizes TensorFlow Lite and reinforcement learning techniques to detect and classify different types of fruits such as apples, bananas, and oranges. Initially based on Edje Electronics' coin detector project, I customized the model by training it on a new fruit dataset, enabling efficient fruit detection for practical applications.
- Source: Kaggle - Fruit Images for Object Detection
- Author: Muhammed Buyukkinaci, Kaggle Expert, Istanbul, Turkey
- Details: The dataset includes images of various fruits, providing a robust foundation for building a reliable object detection model. Images were annotated using LabelImg to create bounding boxes, ensuring accurate model training.
- Platform: Google Colab
- TensorFlow Lite Object Detection API: Adapted from the TensorFlow Lite Object Detection API by Edje Electronics.
- Customizations: This model was retrained with a custom fruit dataset, leveraging TensorFlow Lite’s lightweight architecture to enable quick and efficient fruit recognition on mobile and embedded devices.
The model is trained to recognize the following fruits:
- 🍎 Apple
- 🍌 Banana
- 🍊 Orange
By replacing the original dataset with fruit images and tuning the model parameters, the system achieves a detection accuracy of 95-97%. This makes it an excellent tool for applications in:
- Smart Grocery Sorting: Identify fruits in real-time for inventory management.
- Agriculture Monitoring: Detect and classify fruits in the field.
- Dietary Apps: Use smartphone cameras to identify fruit types for nutritional tracking.
-
Visit The Collab Master File:
-
Set up Google Colab for Training: Open the Colab notebook and execute it according to your dataset.
-
Install Dependencies: Make sure you have the required libraries by running:
- Upload your dataset to Google Colab.
- Set up TensorFlow Lite and start the training with custom fruit images.
- Download the trained model and deploy it to your desired device.
The model’s accuracy ranges from 95% to 97%, based on testing with varied lighting and backgrounds. This can verified on my LinkedIn post mentioned at media/ section of Project Strucure below.
src/: Main scripts for training and detection "https://colab.research.google.com/github/EdjeElectronics/TensorFlow-Lite-Object-Detection-on-Android-and-Raspberry-Pi/blob/master/Train_TFLite2_Object_Detction_Model.ipynb".docs/: Technical documentation and annotated dataset details."https://youtu.be/XZ7FYAMCc4M?si=-f3ZH17552Y3lxSn"media/: Contains images and screenshots of detected fruits. "https://www.linkedin.com/posts/uzaif-talpur_fruitsdetection-technology-computervision-activity-7084097366751948801-O1IZ?utm_source=share&utm_medium=member_desktop"models/: Pre-trained models and trained weights. "https://github.com/tensorflow/models/tree/master/research/object_detection"
Thanks to Edje Electronics for their open-source contributions to TensorFlow Lite object detection, and Muhammed Buyukkinaci for the Kaggle dataset that made this project possible. You can access the Edje Electronics Main Github Repository Branch via this link: https://github.com/EdjeElectronics
Use these tags to boost discoverability on GitHub and social platforms:
#FruitDetection#TensorFlowLite#ComputerVision#ObjectDetection#ReinforcementLearning#GoogleColab#MachineLearning#Python#OpenSource#TechnologyInnovation
Contributions are welcome! Feel free to open issues or submit pull requests for improvements.