SCPA-Net: Text-Enhanced Cross-Platform Framework with Semantic Consistency Enhancement for Pine Wilt Detection

Area of Science:

• Forestry
• Remote Sensing
• Deep Learning
• Plant Pathology

Background:

• Accurate detection of pine wilt disease (PWD) is crucial for forest health management.
• Existing remote sensing methods struggle with PWD detection due to varied sensor platforms, subtle disease symptoms, and complex forest backgrounds.
• Traditional methods limited by reliance on general visual features, failing to capture nuanced PWD indicators.

Purpose of the Study:

• To develop a high-precision detection network for pine wilt disease (PWD) applicable to both UAV and satellite imagery.
• To address challenges of cross-platform data integration, semantic gap reduction, and adaptation to varying disease stages.
• To enhance the robustness of PWD detection in complex forest environments.

Main Methods:

• Proposed SCPA-Net: a cross-platform, semantic-consistent, and phenotype-adaptive detection network.
• Integrated remote sensing images with disease-related text descriptions for semantic prior knowledge.
• Implemented target-context relational modeling and a staged adaptation strategy for progressive learning.

Main Results:

• Achieved superior detection accuracy and generalization performance across multiple datasets (self-built, satellite, PDT, Roboflow).
• Demonstrated effective PWD detection in challenging forestry remote sensing data.
• Successfully reduced semantic gap and improved cross-platform consistency.

Conclusions:

• SCPA-Net offers a robust solution for high-precision pine wilt disease detection using multimodal remote sensing data.
• The proposed framework enhances forest health monitoring capabilities by overcoming limitations of traditional methods.
• The staged adaptation strategy improves model generalization for PWD detection across different disease severities.