Note
This is the second iteration of the model.
To set up and run the model, follow these steps:
- Clone this repository:
git clone https://github.com/therajsekharsaha/pestdetection.git cd pestdetection pip install -r requirements.txt python manage.py
This model was developed as part of our thesis project, "GroPro: Grow and Protect", focused on detecting and mitigating plant pests in urban gardens using both object and audio detection technologies. It enables real-time pest detection in images, videos, and other media formats.
The model uses the YOLOv8 Nano architecture, a compact and efficient variant of the YOLOv8 object detection model, optimized for edge devices like the Raspberry Pi 4. The model was trained on a custom dataset of plant pest images, collected via web scraping from various online sources. YOLOv8 Nano is designed for real-time, low-power pest detection in urban gardens.
The model's performance was evaluated based on mean Average Precision (mAP) across various Intersection over Union (IoU) thresholds, ranging from 0.5 to 0.95 (with a step size of 0.05). The overall mAP achieved was 0.195 on the validation set. Here's a breakdown of mAP for each pest class:
- Aphid: 0.0899
- Fruit Fly: 0.292
- Scale Insect: 0.202
- Preprocessing: 0.3 ms per image
- Inference: 33.7 ms per image
- Postprocessing: 4.3 ms per image
While the model shows promising results, there is still room for improvement, especially in detecting aphids, which currently have lower accuracy.
Below is a demonstration of the model’s performance on a test video. The model successfully detects and labels various plant pests in real-time:
