A CNN-Bigru-Based Defect Detection Method for Transmission Line Insulators in Edge Computing Environment

Aiming at the problem of leakage and misdetection caused by the small target when detecting defects of transmission line insulators under edge computing environment, this paper proposes a CNN-BiGRU-based defect detection method for transmission line insulators under edge computing environment. First, a convolutional neural network (CNN) combined with a bidirectional gated recurrent unit (BiGRU) is used to extract the long sequence-dependent feature relationships of the data and fully extract the multi-scale information. Secondly, the model is lightweighted with the help of maximum pooling operation to reduce the number of parameters. Finally, the Coordinate Attention (CA) mechanism is introduced to effectively capture the global features of direction and position sensing to better localize insulator defects. Experiments show that the P R, and F1 values of the proposed method are 0.938, 0.911, and 0.924, respectively, which are optimal compared with the comparison methods. Therefore, the method is of good guiding significance for improving the operation and maintenance efficiency of power inspection business.