Enhanced YOLOv8 with lightweight and efficient detection head for for detecting rice leaf diseases

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate rice leaf disease detection is crucial for agricultural stability.
• Existing object detection models like YOLO face limitations in complex environments and computational demands.
• Challenges include disease diversity, uneven distribution, and intricate field conditions.

Purpose of the Study:

• To develop an optimized object detection model for precise, multi-scale rice leaf disease detection.
• To address the limitations of current YOLO algorithms in terms of feature extraction and computational efficiency.
• To enhance the deployment of disease detection systems on resource-constrained agricultural devices.

Main Methods:

• Introduction of G-YOLO, a novel architecture integrating a Lightweight and Efficient Detection Head (LEDH) and Multi-scale Spatial Pyramid Pooling Fast (MSPPF).
• LEDH simplifies network structure for reduced computational load while preserving accuracy.
• MSPPF fuses multi-level feature maps to improve capture of disease details across scales.

Main Results:

• G-YOLO demonstrated superior performance on the RiceDisease dataset compared to YOLOv8n.
• Achieved 4.4% higher mAP@0.5 and 3.9% higher mAP@0.75.
• Showcased a 13.1% increase in Frames Per Second (FPS), indicating enhanced processing speed.

Conclusions:

• G-YOLO offers a significant advancement in automated rice leaf disease detection.
• The model's efficiency and accuracy make it suitable for resource-constrained environments.
• This research contributes to improved crop health management through advanced AI.