GanSystem Weed Detection is a computer vision project developed to identify weeds within bean farms using machine learning and image recognition techniques.
The project was created as part of the broader GanSystem Smart Agriculture initiative, which focuses on leveraging Artificial Intelligence, IoT, and automation to improve agricultural productivity and decision-making.
Manual weed identification is labor-intensive, time-consuming, and often inconsistent across large agricultural environments.
This project explores the use of AI-powered object detection models to automatically identify weeds and bean crops from farm images, providing a foundation for future precision agriculture and automated weed management systems.
- Detect weeds within bean farms
- Distinguish crop plants from unwanted vegetation
- Improve farm monitoring through computer vision
- Support future autonomous spraying and precision farming systems
- Build an annotated agricultural dataset for machine learning applications
| Metric | Value |
|---|---|
| Images | 213 |
| Annotated Objects | 2,334 |
| Classes | 2 |
| Average Image Size | 12.19 MP |
| Median Resolution | 3024 × 4032 |
| Class | Annotations |
|---|---|
| Bean | 296 |
| Weed | 2,038 |
The dataset contains 213 agricultural images with 2,334 manually annotated objects distributed across two classes: Bean and Weed.
Annotation heatmaps were used to analyze object distribution across images and identify dataset coverage patterns.
- Auto-orientation
- Contrast enhancement
- Image resizing to 640×640
- Horizontal Flip
- Random Rotation
- Blur
- Motion Blur
- Noise Injection
- Zoom Augmentation
These augmentation techniques improved dataset diversity and model robustness.
The weed detection model was trained using the Roboflow ecosystem and deployed as a computer vision inference pipeline.
- Roboflow
- YOLO-based Object Detection
- Python
- OpenCV
| Metric | Score |
|---|---|
| mAP@50 | 65.1% |
| Precision | 42.2% |
| Recall | 81.8% |
| F1 Score | 55.7% |
The model achieved strong recall performance, demonstrating effectiveness in identifying weed instances within agricultural imagery.
- Python
- OpenCV
- Roboflow
- YOLO
- Computer Vision
- Machine Learning
- Deep Learning
- Image Processing
- Object Detection
- Computer Vision
- Deep Learning
- Model Evaluation
- Dataset Creation
- Data Annotation
- Data Augmentation
- Performance Analysis
- Python Development
- Image Processing
- Agricultural AI Applications
- Model Deployment Preparation
gansystem-weed-detection/
│
├── screenshots/
│ ├── cover.png
│ ├── dataset-overview.png
│ ├── annotation-heatmap.png
│ └── model-performance.png
│
├── weed_detection_live.py
├── dataset-overview.md
├── results.md
└── README.md
- Real-time camera integration
- Edge deployment on Jetson Nano
- Mobile weed detection interface
- Automated spray recommendations
- Integration with GanSystem Smart Farm Platform
- Weed density analytics dashboard
This project is part of the larger GanSystem ecosystem, which includes:
- Smart Irrigation
- IoT Farm Monitoring
- Fish Farm Automation
- Environmental Monitoring
- Agricultural AI Systems
Repository:
https://github.com/codeandbe/gansystem
This repository serves as a portfolio showcase demonstrating practical experience in:
- Computer Vision
- Machine Learning
- Agricultural AI
- Dataset Annotation
- Model Evaluation
- Python Development
Iyobosa Amaddin
GitHub: https://github.com/codeandbe
LinkedIn: https://linkedin.com/in/codeandbe
This repository is intended for educational, research, and portfolio demonstration purposes.
The repository showcases the dataset preparation, training workflow, evaluation results, and prototype implementation of an agricultural weed detection system.