GEFA-YOLO: Lightweight Weed Detection with Group-Enhanced Fusion Attention.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Weed management in cotton is crucial for yield and quality.
• Existing weed detection models face challenges with complex weed morphologies and environmental factors.
• Current attention mechanisms in models have limitations in balancing performance and computational cost.

Purpose of the Study:

• To develop an efficient and accurate weed detection model for cotton fields.
• To address the limitations of existing attention mechanisms in computer vision models for agriculture.
• To propose a novel attention mechanism that reduces complexity while enhancing feature expression.

Main Methods:

• Proposed a grouped enhanced fusion attention mechanism (GEFA) combining grouped convolution and local spatial attention.
• Developed the GEFAY detection model integrating the GEFA mechanism.
• Evaluated the model on CottonWeedDet12, VOC, and COCO datasets.

Main Results:

• The GEFAY model achieved a good balance between efficiency, accuracy, and complexity.
• GEFA demonstrated a smaller increase in parameters and computational costs compared to classic attention methods.
• The model significantly improved detection accuracy, making it suitable for edge device deployment.

Conclusions:

• The GEFA mechanism effectively enhances feature expression while reducing computational load.
• The GEFAY model offers a practical and scalable solution for intelligent weed detection in precision agriculture.
• The developed system provides real-time weed detection capabilities for edge devices, supporting smart farming applications.