Visual Defect Detection for Substation Equipment based on Joint Inspection Data of Camera and Robot

Abstract

The basic principle of joint inspection with video and robot is that the deep learning algorithm can extract specific defect features from mass images and video data. However, the defect information of substation equipment is often masked by label noise and complex backgrounds. Furthermore, there are many different types of equipment in the substation and the defect varies from equipment to equipment, which increases the difficulty of detection. Previous work mostly focuses on defect features of single frame image, which ignores the association between typical defect characteristics and the substation equipments. Besides, it requires manual selection for preliminary image annotation. To solve the above-mentioned problems, this paper proposes a visual defect detection strategy for substation equipments using cascade deep learning model. The efficacy of the proposed method is verified through an experiment based on electrical equipment of substation.