ESE-PWDNet: an efficient early-stage pine wilt disease detection network

Area of Science:

• Forestry
• Remote Sensing
• Computer Vision
• Plant Pathology

Background:

• Pine Wilt Disease (PWD) poses a significant threat to forest ecosystems.
• Early detection of PWD is crucial for effective management and mitigation.
• Recognizing small, early-stage disease indicators using remote sensing presents a substantial challenge.

Purpose of the Study:

• To develop an efficient Unmanned Aerial Vehicle (UAV) remote sensing detection model for early Pine Wilt Disease (PWD) identification.
• To enhance the recognition of small and subtle PWD targets in complex forest scenes.
• To provide a robust technical solution for precise and efficient forestry pest monitoring.

Main Methods:

• Construction of a multi-temporal, multi-view high-resolution dataset of early-stage PWD using a DJI Air3 UAV.
• Design of the Efficient Visual Linear Unit (EFVLU) as a foundational module for the backbone network.
• Development of a novel neck network incorporating the Attention State Space Block (ASSB) for high-resolution image processing.
• Integration of Efficient Multi-scale Attention (EMA) and Lightweight Shared Detail Enhanced Convolutional Detection Head (LSDECD) in the prediction head.

Main Results:

• The proposed ESE-PWDNet model significantly improves the recognition performance of tiny PWD targets in complex environments.
• The model achieves a Precision (P) of 75.9% and a Recall (R) of 75.1%.
• ESE-PWDNet demonstrates high inference efficiency with low computational complexity (6.5 GFLOPs, 2.6M parameters), outperforming comparative models.

Conclusions:

• ESE-PWDNet offers a reliable and efficient technical solution for early and precise UAV remote sensing monitoring of PWD.
• The developed dataset and model provide a strong foundation for future research in automated forestry pest detection.
• This approach facilitates timely intervention strategies to combat the spread of Pine Wilt Disease.