A Morphology-Aware and Hard-Negative-Optimized Detection Framework for External Intrusion Monitoring in Transmission Corridors

External intrusion hazards such as construction machinery pose serious threats to the safe operation of transmission lines. However, reliable detection in transmission corridors remains challenging because hazardous targets usually exhibit articulated and elongated structures, while monitoring images are dominated by complex backgrounds and rare hard-negative samples. To address these challenges, this paper proposes CMHdet, a morphology-aware and hard-negative-optimized detection framework for external intrusion monitoring in transmission corridors. First, a Dynamic Deformable Transformer module is embedded into the feature extraction backbone to adaptively adjust spatial sampling positions and enhance the representation of irregular machinery structures under viewpoint changes and occlusion. Second, a dual-path multi-scale aggregation network with shifted-window attention is designed to preserve local structural details while strengthening cross-region contextual interaction for small and large-span targets. Third, a hard-negative-aware optimization strategy is developed by combining Gradient Harmonized Mining loss with a false-alarm-guided dynamic copy-paste augmentation mechanism, enabling the model to learn from confusing background regions frequently encountered in long-term monitoring. Experiments on a real-world transmission corridor dataset demonstrate that CMHdet achieves 93.4% mAP, outperforming the YOLOv10L baseline by 5.7 percentage points, with notable improvements under long-distance, occluded, and adverse-weather conditions. The results indicate that the proposed framework provides a reliable solution for intelligent external intrusion monitoring in transmission corridors.