APPLICATION OF THE OPTIMISED YOLOv3 ALGORITHM IN SUBSTATION POWER EQUIPMENT DEFECT IDENTIFICATION AND DETECTION, 235-243.

Ti Liu, Jia Feng, Dahong Fu, and Wenhan Chen

Keywords

You only look once version 3 (YOLOv3) algorithm, substation, recognition, detection

Abstract

With the operation of the substation, defects and faults will gradually appear in the power equipment; therefore, timely identification and detection of these defects are necessary to ensure the reliable operation of the substation. In this paper, for the identification and detection of power equipment defects, the you only look once version 3 (YOLOv3) algorithm was improved. After the backbone network was improved using the cross stage partial darknet53, and the focal loss replaced the cross-entropy loss, the optimised yolov3 algorithm was obtained. Experiments were then conducted on both the PASCAL VOC dataset and the actual defect dataset. It was found that, compared with the faster region-based convolutional neural network and YOLOv3 algorithm, the detection performance of the optimised YOLOv3 algorithm was significantly better, with a mean average precision of 82.55% for the PASCAL VOC dataset and 97.05% for the actual defect detection dataset, and frames per second of 61. The experimental results demonstrate the reliability of the optimised YOLOv3 algorithm designed in this paper for defect recognition and detection of substation power equipment, which provides a new idea for the optimisation of the YOLOv3 algorithm and is also helpful to improve the efficiency and quality of substation inspection.

Important Links:

Go Back