Dual-branch network for threat analysis of external-force hazards on power transmission lines

External-force hazards have become a major cause of faults in overhead transmission lines. Existing studies focus on hazard detection, but do not localize the transmission-line safety region or analyze whether a detected target is threatening the line. To address this issue, we propose a dual-branch network that integrates hazard detection, transmission-line segmentation, and pixel-level spatial discrimination for line-aware threat analysis. In the detection branch, a multiscale multihead convolutional gated attention mechanism and a multiscale adaptive spatial attention gate are introduced to enhance multiscale hazard perception, and suppress corridor background interference while maintaining a lightweight design. In the segmentation branch, context anchor attention is embedded into TransUNet to improve the extraction of elongated transmission-line structures and generate reliable line masks. The pixel-level positioning and discrimination module fuses the outputs of the two branches and performs line-aware threat screening. Experimental results show that the detection model improves mAP 50 by 5.5 percentage points over the baseline model, while the segmentation branch improves IoU by 1.16 percentage points over the baseline TransUNet. More importantly, after pixel-level screening, only 2.08% of the images require further manual verification, demonstrating the effectiveness of the proposed method in improving warning precision in practical transmission-line inspection.