A Method for Detecting Temperature Anomalies of Disc Suspension Insulators on Power Transmission Lines

Coupled with the extensive application of infrared thermal imaging technology in intelligent fault diagnosis of insulators on the transmission line, current image processing algorithms struggle to effectively identify the abnormal hot spots in infrared images of insulators. To cope with this problem, this paper proposes a lightweight model based on the ShuffleNet V2 network, which is based on the key point detection algorithm of the Gaussian heat map and embedded with efficient channel attention (ECA) mechanism to accurately locate and identify the iron cap area in the infrared image of insulators. After detecting the position of the insulator's iron cap, Hough straight line detection algorithm is used to combine the attitude of the insulator string, and then extract the thermal information of the insulator iron cap area. The model takes floating-point operations per second (FLOPs), memory consumption, model size, average precision (object keypoint similarity)-AP (OKS), and inference time as evaluation indexes. The experimental results demonstrate that the FLOPs size of the model is 4.65×107, the model size is 11.72 M, and the average precision reaches 93.39%. The memory consumption is 22.98 M and the inference time is 1.60 s when deployed and tested in Raspberry Pi. In comparison with the existing methods, the model achieves a sound balance between detection precision and inference velocity, meets the needs of edge equipment deployment, and provides a new solution for improving the inspection efficiency of the transmission line system.