Optical-SAR Fusion Enhanced by Lightweight Multiscale Attention for Intelligent Disaster Detection: Protecting Transmission Corridors from Foreign Object Intrusions

This paper proposes an optical-SAR fusion framework for intelligent hazard detection in transmission corridors. A dual-modal dataset is constructed and processed to address multiscale feature challenges. The method employs a twostage deep learning architecture: a lightweight optical detection model integrates multiscale attention (EPSA) and adaptive downsampling (AD) for efficient feature extraction, while SAR image slices are classified via a spatial-enhanced MBConv-CA network with global pooling attention. Experiments demonstrate that complementary multimodal features significantly improve detection accuracy and robustness against environmental interference, enabling reliable identification of hazards including floating objects, vegetation encroachment, and construction activities.