A feature recognition and detection algorithm for pine wilt disease trees based on FLMP-YOLOv8.

Area of Science:

• Forestry
• Plant Pathology
• Computer Vision

Background:

• Pine wilt disease, caused by pine wood nematodes and transmitted by pine sawyer beetles, is a major forest threat.
• Accurate detection of infected trees is critical for effective disease management and prevention.
• Complex terrain and uneven forest distribution in regions like China's Qinba Mountains challenge traditional detection methods.

Purpose of the Study:

• To develop a novel, efficient, and accurate method for detecting pine wilt disease-infected trees.
• To address the challenges of feature extraction in complex forest environments.
• To improve upon existing deep learning models for forest disease detection.

Main Methods:

• Proposed a novel FLMP-YOLOv8 algorithm integrating FasterBlock modules for enhanced feature extraction and reduced complexity.
• Incorporated a Large Separable Kernel Attention (LSKA) mechanism into the SPPF module to improve fine detail perception and reduce interference.
• Utilized the MPDIoU loss function for precise bounding box regression and localization.

Main Results:

• The FLMP-YOLOv8 model achieved 92.0% precision, 80.8% recall, and 87.0% mAP@0.5.
• Demonstrated a significant detection speed of 81.79 FPS.
• Showed improvements over the original YOLOv8, including a 2.2% increase in precision, 0.6% in recall, and 2.0% in mAP@0.5, with a 65.48 FPS speed increase.

Conclusions:

• The FLMP-YOLOv8 algorithm offers a more reliable and cost-effective solution for detecting pine wilt disease.
• The study successfully addressed feature extraction challenges in complex forest terrains.
• The developed method provides a valuable tool for forest health monitoring and disease management.