Intelligent defect detection method for transmission lines integrating UAV image recognition

Abstract

Unmanned aerial vehicles (UAVs) equipped with optical cameras have become important sensing platforms for automated inspection of transmission-line corridors. However, practical defect detection remains difficult because small damaged parts occupy only a few pixels, component scales vary sharply with flight distance, and blurred or overexposed frames easily cause false alarms. This paper proposes a quality-aware intelligent defect detection method that integrates UAV image recognition, adaptive multi-scale fusion, attention-guided defect localization, and severity-oriented maintenance scoring. The method first filters unreliable optical frames according to blur, exposure and valid-component indicators. A compact convolutional backbone with coordinate-channel attention is then used to extract local component features from insulators, vibration dampers, fittings and conductors. A weighted feature pyramid preserves small-defect cues, and multi-task heads jointly predict component category, defect type, bounding box and severity class. Finally, detection confidence, defect area and component criticality are mapped to a practical risk score for maintenance scheduling. Experiments on a UAV transmission-line image set containing five defect categories show that the proposed model reaches 94.2% mean average precision at an intersection-over-union threshold of 0.5 (mAP@0.5), 92.1% recall and 132 frames per second (FPS) on the test graphics processing unit (GPU), while maintaining higher robustness under low-light, foggy and motion-blur conditions than representative baselines. The results demonstrate that optical UAV inspection can be connected with automated fault diagnosis and edge-cloud maintenance workflows for critical power infrastructure.