Transmission Lines Insulator State Detection Method Based on Deep Learning

Aerial images are commonly used for detecting insulators in transmission lines to ensure their safe operation. However, each capture session generates thousands of insulator images, requiring manual collection, organization, and analysis. Therefore, to achieve automation in insulator state detection, this paper proposes a method based on deep learning for insulator state detection in transmission lines. Firstly, an insulator state detection model is built based on YOLOv7, and the model is improved using a bi-level routing attention mechanism and a content-aware up-sampling operator. Then, combined with dataset augmentation, including cropping, flipping, rotating, scaling, and splicing and a bounding box loss function incorporating a dynamic non-monotonic focus mechanism, 4000 visible images from different voltage levels of transmission lines are used for training. Finally, using a confusion matrix combined with comparative and ablation experiments, the results of insulator state detection are analyzed. Experimental results show that the proposed method achieves a detection accuracy of 97.1%. The detection accuracies for insulators exhibiting self-explosion, damage, flashover, and insulator strings are 93.5%, 98.6%, 97.5%, and 98.9%, respectively. Analysis results demonstrate that the proposed method can effectively realize insulator status detection.