Transmission Line Fittings And The Defects Detection Method Based On Relation Reasoning

As a very important and widely distributed metal accessory on transmission lines,the fittings are prone to defects such as deformation,damage,corrosion and other defects,due to the complex and harsh field environment for a long time.Therefore,the safe and efficient realization of automatic inspection of fittings and its defects has become basic work to ensure the safe and normal operation of power system.Using computer vision methods and combining the prior knowledge of fittings to identify fittings and its defects,it can effectively improve the efficiency of line inspection,operation and maintenance while saving manpower and material resources.This paper has mainly completed the following works on the detection of transmission line fittings and its defects based on relation reasoning:The existing object detection methods are prone to misdetection and missed detection without considering the connection relationship information between the fittings and scene information.This paper proposes a method based on occlusion relation reasoning for fittings detection.The fitting feature,scene-related information and occlusion relationship information are extracted and learned by Faster R-CNN,and the occlusion relationship inference module is constructed to complete the joint reasoning detection of category and location of fitting.The experiment results show that the m AP of the proposed method is 84.15%,the m AP has been improved by2.85%,in particular,the accuracy of small target fittings have been improved more obviously,such as U-type hanging ring and yoke plate.Aiming at the problem of insufficient fitting defect samples and diversification of defect target shapes,only using the deep model leads to the low accuracy of classification,a causal classification method of fitting defects combined with deep features is proposed.Based on the expanded dataset,the VGG16 model is used to extract deep features,and constructed the feature set that conforms to the causal relationship learning,and a causal classification model is constructed that conforms to the characteristics of fitting defects through global confounder balancing.The experiments prove that the proposed method can effectively improve the performance of fitting defect classification model.Among them,the accuracy of the shockproof hammer intersection and deformation are 92.99% and 91.18%,and the accuracy of the shielded ring corrosion and grading ring damage respectively are 95.67% and96.69%.