Substation Equipment Abnormity Detection Based on Improved YOLOv5

Substation equipment inspection is the basic work contents of equipment management and security. However, the substation equipment inspection is facing the contradiction between the rapid increase in the number of substation and the shortage of inspection personnel. We propose a deep learning-based method for detecting substation equipment abnormity to enhance the efficiency of operations. Based on YOLOv5, we add a small object detection head to solve the large scale variation of the equipment images. Besides, the K-means clustering algorithm is adopted to generate suitable anchors for training process, and we also perform data augmentation to increase the data complexity and numbers. Last, considering the overlapping area, center point distance and aspect ratio of the bounding box and ground truth, we use CloU loss function to replace the GIoU loss function. The experiments are executed on the industrial dataset, and the results reach 0.677 for ten types of equipment abnormities by the mAP. The experiment results show that the improved YOLOv5 algorithm is feasible and effective.