REP-YOLOX: An Efficient Model for Defect Detection in Gas Insulated Switchgear Equipment

Abstract

Gas Insulated Switchgear (GIS) equipment is an important substation device in the power system and plays an indispensable role in maintaining the operation of the power system. However, GIS equipment is prone to ablation defects and some foreign objects during long-term operation, thus affecting the normal operation of the power system. Aiming at the problems of low efficiency and easy-to-miss detection in manual inspection.This paper proposes a single-stage object detection model REP-YOLOX, which is based on structural re-parameterization and attention mechanisms technology. Firstly, we combine dilated convolution and standard convolution to extract features and obtain larger receptive fields. Then, in the Neck layer of the model, the Mask Conv block module is used to enhance the feature extraction ability under occlusion. Finally, the Transformer module is used in the prediction layer of the model to weigh the global features and predict the results through the full connection layer. Experimental results show that the model can achieve 99.14% mAP50 in the GIS equipment defect detection task, which indicates it can complete GIS equipment defect detection task well.