Research On Deep Learning-Based Detection Of Power Grid Insulator Defects

Insulator defects ser IoUsly affect the safety and reliable operation of power grids.In order to quickly and accurately identify insulator defects,this article proposes an insulator detection algorithm based on YOLO v3(You Only Look Once),which can achieve precise positioning and defect detection of insulators.The main contents of the work are as follows:(1)Improvement of the prior box calculation method in the YOLO v3 algorithm.A prior box optimization algorithm based on K-means clustering algorithm is proposed.The genetic algorithm is used to generate the initial center point of the K-means algorithm,and then K-means clustering algorithm is used to generate anchor boxes that are closer to the real boxes.Experimental results show the improved algorithm has higher real-time performance and accuracy,effectively improving the recall rate and IoU of the YOLO v3 algorithm.(2)To solve the problem of imbalanced positive and negative samples in the YOLOv3 algorithm,a DIoU location loss function is used to replace the original IoU location loss function and label smoothing and weighted cross-entropy functions are added to optimize classification.Finally,Quality Focal Loss(QFL)is introduced in the confidence classification step to improve the model’s generalization ability.Experimental results show that the proposed algorithm is better than the original YOLOv3 algorithm in accuracy.(3)By integrating improvements in the prior boxes and optimization of the loss function in YOLO v3,a new algorithm for insulator detection called Q-YOLO v3 has been proposed.The experimental results show that the Q-YOLO v3 model improves the accuracy of detecting the parts of insulators and insulator defects.The average precision(mAP)of this model has also increased,and the confidence level in detecting insulators is improved to achieve more accurate localization.The model can effectively distinguish the potential defect areas on the insulators improving the accuracy of target recognition.This helps to achieve more accurate positioning and classification during subsequent processing.Therefore,the proposed algorithm is advanced and meets the practical requirements of insulator defect detection in power grids.