Pine wilt disease detection algorithm based on improved YOLOv5.

Area of Science:

• Forestry Science
• Remote Sensing Technology
• Artificial Intelligence

Background:

• Pine wilt disease (PWD) is a severe forest threat due to high infectivity and lethality.
• Effective PWD management relies on timely detection and isolation of infected trees.
• Current deep learning methods for PWD detection using UAV imagery often require extensive expertise and lack efficiency.

Purpose of the Study:

• To develop a highly accurate, lightweight, and efficient deep learning model for real-time detection of PWD-infected pine trees.
• To improve the practical applicability of UAV-based remote sensing for forest disease monitoring.

Main Methods:

• Integration of ShuffleNetV2, SimAM attention module, and ASFF for feature fusion within the YOLOv5 architecture, creating the YOLOv5L-s-SimAM-ASFF model.
• Utilizing unmanned aerial vehicle (UAV) RGB images for detecting diseased pine trees under natural conditions.
• Evaluation of model performance using metrics such as mean average precision (mAP) and frames per second (FPS).

Main Results:

• The proposed YOLOv5L-s-SimAM-ASFF model achieved a mAP of 95.64% and a recall rate of 91.28% for PWD detection.
• The model demonstrated a high processing speed of 95.70 FPS, enabling real-time detection.
• Significant reduction in model size and parameter count compared to the original YOLOv5-Lite, enhancing efficiency.

Conclusions:

• The YOLOv5L-s-SimAM-ASFF model is highly suitable for real-time, accurate, and lightweight detection of PWD-infected trees.
• This model provides a valuable tool for precise localization and quantification of infected trees, aiding forest management.
• The developed model offers practical guidance for effective PWD management and eradication strategies.