Research Insulator Detection And Defect Recognition Based On Deep Convolutional Neural Network

With the continuous construction of transmission lines and the innovative development of power grid technology,intelligent power inspection technology based on robots and unmanned aerial vehicles(UAV)has been widely used.As an important power component that supports and insulates power transmission lines,power insulators are mainly made of ceramics,glass or composite materials.However,they are prone to failure when exposed to the field for a long time.Once the failure occurs,they will seriously affect stable power transmission.In order to ensure the stable operation of the power grid,it is particularly important to detect the fault insulators in the transmission line inspection,among which the insulator defect is one of the common fault categories.The traditional inspection method is difficult to meet the actual demand,and the development of UAV technology has provided hardware support for intelligent power inspection.In addition,the continuous proposal of deep convolutional neural network algorithm and the continuous upgrading of hardware equipment have solved a large number of problems in power image processing.In recent years,the location and identification of power components have attracted more attention from scholars.Based on the deep convolutional neural network,this thesis selects insulators as representative power components and combines their characteristics to study the insulator detection and defect recognition,which is of great significance for the intelligent inspection of transmission lines.The main work and achievements of this thesis are as follows:(1)For the problems of low detection accuracy and poor real-time performance of existing methods,this thesis proposes a real-time insulator detection method based on the improved YOLOv3 model.Firstly,the K-means++ algorithm is used to obtain new anchor points for the data clustering class.Then,the label smoothing regularization constraint model and the improved non-maximum suppression algorithm are introduced to improve the model accuracy.Through experimental analysis,the proposed method outperformed other models with an AP value of 93.8% and a shorter detection time for insulator images.(2)Aiming at the problems of poor feature extraction ability and inaccurate identification and location of defective insulators by traditional detection methods,a multi-scale defective insulator detection algorithm based on attention mechanism is proposed.This method integrates the channel attention mechanism SENet into the feature extraction network Darknet53 to enhance the feature extraction capability of the network.Multiple detection scales are added to improve the detection accuracy of defective insulators.In addition,combined with multi-scale detection,the defective insulator dataset is re-clustered.The experimental results show that the method achieves an average accuracy of 94.42% and a recall rate of 95.74% while meeting the requirement of real-time detection.