YOLOv8-MFD: An Enhanced Detection Model for Pine Wilt Diseased Trees Using UAV Imagery.

Area of Science:

• Forestry
• Plant Pathology
• Remote Sensing
• Computer Vision

Background:

• Pine Wilt Disease (PWD) poses a significant threat to global pine forests and economies.
• Accurate and early detection of PWD is critical for effective forest management and outbreak prevention.
• Existing remote sensing models face challenges with performance in complex environments and real-time efficiency.

Purpose of the Study:

• To develop an improved object detection model for accurate and efficient detection of PWD-infected trees using UAV imagery.
• To enhance feature representation and model robustness in complex forest backgrounds for PWD detection.
• To provide a reliable solution for early-stage PWD monitoring across large forested areas.

Main Methods:

• Proposed an improved object detection model, YOLOv8-MFD, for detecting PWD-infected trees from UAV imagery.
• Incorporated a MobileViT-based backbone for fusing CNNs and Transformers to enhance feature representation.
• Integrated a Focal Modulation mechanism and a Dynamic Head to improve robustness, precision, and multi-scale perception.

Main Results:

• YOLOv8-MFD achieved high performance metrics: 92.5% precision, 84.7% recall, 88.4% F1-score, and 88.2% mAP@0.5.
• The model demonstrated superior accuracy compared to baseline YOLOv8 and YOLOv10.
• Maintained acceptable computational cost (11.8 GFLOPs), compact model size (10.2 MB), and suitable inference speed for real-time deployment.

Conclusions:

• The proposed YOLOv8-MFD model offers a reliable and efficient solution for early-stage PWD monitoring in extensive forest areas.
• Enables timely disease intervention and enhances forest resource protection strategies.
• The generalizable architecture shows potential for broader applications in forest health and agricultural disease detection.