An Insulator Defect Detection Method in Distribution Lines Based on Vector Dot Product

Target detection algorithms for current deep learning all require a large number of defect samples to train the network model. As a consequence, they are unable to accurately defect defects when only a small number of samples are provided. Especially in the situation of occlusion, the recognition accuracy is further decreased. Based on the above, this paper proposes an insulator defect detection method using deep learning for initial localisation and vector features for fine localisation. Firstly, the deep learning network model YOLO and a small number of samples are used to locate the insulator string. Then image segmentation and feature extraction are performed on the insulator string area. Finally, the method of vector dot product sum is used to extract insulator defects. The above method can solve the problem that the current deep learning insulator defect detection requires a large number of defect sample training data sets and the over-fitting problem caused by too many parameters. The experimental results show that the method can realize a small number of training samples to complete the localization of insulators. And combine image segmentation with vector dot product sum to complete the rapid detection of defects. It also provides a new defect detection idea and method for identify detection in distribution network equipment.