A Lightweight YOLOv4-Based Forestry Pest Detection Method Using Coordinate Attention and Feature Fusion.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Forestry Science

Background:

• Traditional pest detection methods in forestry suffer from low accuracy and speed, hindering effective management.
• Complex forest environments pose significant challenges for current detection technologies.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for pest detection in forestry.
• To improve upon existing object detection models for enhanced performance in complex environments.

Main Methods:

• Proposed the YOLOv4_MF model, integrating MobileNetv2 with depth-wise separated convolution for reduced parameters.
• Incorporated a coordinate attention mechanism within MobileNetv2 to enhance feature representation.
• Implemented a symmetric three-layer spatial pyramid pool and an improved feature fusion structure.
• Utilized focal loss to improve the detection of small pest targets.

Main Results:

• The YOLOv4_MF model demonstrated a 4.24% increase in mAP, 4.37% in precision, and 6.68% in recall compared to the YOLOv4 model.
• The model size was reduced to 1/6 of the original YOLOv4, indicating significant efficiency gains.
• Achieved 38.62% mAP on the COCO dataset, showing competitive performance against state-of-the-art algorithms.

Conclusions:

• The YOLOv4_MF model offers a substantial improvement in pest detection accuracy and efficiency for forestry applications.
• The model's reduced size and enhanced performance make it a viable solution for real-time pest monitoring in complex environments.