Research On Insulator Defect Detection Based On Deep Learning

With the basic realization of Chinese "13th Five-Year" power plan and the smooth development of the "14th Five-Year" power plan,the requirements for the transmission voltage,transmission distance and transmission capacity of my country’s transmission lines have also become higher.As one of the important components on the transmission line,the insulator bears the burden of increasing the creepage distance and preventing the current from returning to the ground.Since insulators are exposed to harsh environments for a long time and are prone to failures such as string drop,this article summarizes the relevant research at home and abroad,and gives the insulator defect detection method divided into three parts: target recognition,image segmentation and defect detection.Firstly,aiming at the problem of fast and accurate identification of insulators under high resolution,a target identification algorithm for insulators based on improved YOLOv3 is proposed.For the aerial insulator image data set,the K-means++algorithm is used to generate a new sight frame to reduce the number of convolutional layers of the feature extraction network Darknet-53 and the number of times the residual unit is used to reduce the network depth and the amount of calculation,and add a scale prediction.Comparative experiment results show that the average detection time of the improved algorithm is 30.975 ms,and the accuracy rate reaches 95.99%.Secondly,due to the low efficiency and low precision of existing insulator image segmentation methods,an insulator image segmentation algorithm based on improved DeepLabv3+ is proposed.In view of the importance of the special texture features of the insulator image,the Xception backbone network is optimized to reduce the number of spatial convolutional layers,replace the global average pooling layer with the global maximum pooling layer,and integrate the attention mechanism into the backbone network and decoder.And increased multi-scale prediction.The comparative experiment proves that the improved algorithm can better restore the characteristics of the insulators and realize the accurate segmentation of the insulators.Furthermore,in view of the problem of mutual concealment of insulator strings,a least squares ellipse fitting method based on RANSAC is proposed,which better fits the contours of insulators.Finally,a defect detection method for cascaded insulators is proposed aiming at the string drop defect of insulators,According to the recognition results of the insulator target recognition model based on the improved YOLOv3,it is cropped and sent to the insulator image segmentation model based on the improved DeepLabv3+,and then the segmented image is fitted by the least squares ellipse fitting method based on RANSAC and the defect location is judged.Return to the original Figure.The experimental results prove that the cascaded insulator defect detection algorithm improves the positioning accuracy of the insulator and the defect block.