Overview
This project focuses on real-time rust disease detection in sugar beets using the YOLOv8 object detection model, integrated with a cross-platform mobile application developed using Flutter. The system processes images from the device's camera and applies deep learning techniques to identify rust disease in sugar beets.
Real-time disease detection using YOLOv8
Cross-platform mobile application built with Flutter
Camera-based image processing
User-friendly UI for displaying disease classification results
Efficient model inference for mobile deployment
Deep Learning Framework: YOLOv8 (Ultralytics)
Mobile Development: Flutter (Dart)
Deployment: ONNX / TensorFlow Lite (TFLite) for mobile inference
The dataset used for training the YOLOv8 model consists of annotated images of sugar beet plants affected rust disease. The dataset was preprocessed and augmented to improve model performance.
Dataset Preparation
Collected and annotated images.
Applied data augmentation techniques such as rotation, flipping, and contrast adjustments.
Model Training
Used the Ultralytics YOLOv8 framework for training.
Trained on a GPU ( Google Colab).
Optimized hyperparameters including batch size, learning rate, and number of epochs.
Model Export for Mobile
Converted the trained YOLOv8 model to ONNX / TensorFlow Lite for efficient mobile inference.
Quantized the model to reduce size and improve speed without compromising accuracy.
Project Setup
Created a Flutter project using flutter create.
Integrated the necessary dependencies (camera, image_picker, tflite_flutter, etc.).
Model Integration
Loaded the YOLOv8 model into the Flutter app using TFLite/ONNX runtime.
Implemented real-time image processing using the device's camera.
Processed model output to extract bounding boxes and classification results.
Developed an intuitive interface for capturing and analyzing images.
Displayed detection results with bounding boxes and confidence scores.
Implemented feedback options for users to report false positives/negatives.
Tested on Android devices.
Optimized model inference for low-latency performance.
Packaged and released the application via Google Play Store (if applicable).
Challenge: Handling real-time inference with a large model.
Solution: Model quantization and performance optimizations.
Challenge: Integrating YOLOv8 into a Flutter application.
Solution: Used ONNX/TFLite for mobile compatibility.
Challenge: Improving accuracy for rare disease cases.
Solution: Applied advanced augmentation and increased dataset diversity.
Enhancing model accuracy with a larger dataset.
Adding cloud-based inference for improved efficiency.
Implementing user feedback mechanisms for active learning.
This project successfully combines deep learning and mobile application development to create an efficient sugar beet disease detection system. By leveraging YOLOv8 and Flutter, we achieved real-time disease detection capabilities on mobile devices, making it accessible for farmers and agricultural experts.
To set up and run the application, follow these steps:
-
Clone the Repository:
git clone https://github.com/nisanurtas/Sugar-Beet-Disease-Detection-App-BeetApp-.git
-
Install Dependencies:
cd BeetApp ./gradlew build -
Open in Android Studio:
Open the project in Android Studio and run it.
- Android: Provides the platform for the mobile application.
- AI and Machine Learning: YOLOv8 algorithm is used for disease detection.
- Flutter: Used for developing the user interface.
If you'd like to contribute, please feel free to create pull requests or file issues. We welcome all contributions and feedback!
For any questions or feedback regarding the project, please contact nisaanurtas@gmail.com.