Research On Insulator Detection Algorithm Based On Improved Faster RCNN

With the continuous increase of industry and city scale,the demand for electricity also increased extremely,which promoted the rapid development of the power system.The continuous expansion of the area covered by the power system increases the difficulty of the inspection work for the power personnel.Power workers need to process and analyze a large amount of data,which is time-consuming and inefficient.At the same time,there will be deviation caused by experience judgment and fatigue work.Insulators are common components in the line,which are exposed to wind and rain all the year round,and are prone to faults caused by loss,which is a great hidden danger to the safety of the power grid.Therefore,fault diagnosis needs to be carried out efficiently and quickly.In this paper,intelligent identification method is introduced for insulator detection,which mainly includes the following research contents:1)By improving the network structure of the target detection algorithm and optimizing the error algorithm,problems such as weak generalization ability of the traditional target detection algorithm and low detection accuracy for small targets can be solved.Firstly,the main structure and function of convolutional neural network are studied.According to the engineering requirements and the performance of experimental hardware,an appropriate network framework is selected to construct a network model.Then the insulator data set collected is expanded to improve the performance of the network model in the process of training.Finally,Tensor Flow is used as the framework to train the network model and constantly adjust the parameters of the model to improve the performance.The deep learning method is used to detect insulators,which greatly improves the accuracy of insulator detection.The accuracy of model training can reach94.72%.2)The method of multi-feature fusion and the optimization of the loss function are used to solve the shortcomings of the traditional target detection algorithm which has low detection accuracy for small target objects.In the original Faster RCNN network model,ROI Pooling directly processes the feature map output by the feature extraction layer to generate candidate regions.However,as the network model deepens,each feature extraction layer will lose part of the information.In this paper,the method of feature fusion is used to fuse the feature images on different convolution layers.The fused feature images can contain more information,so that the model can better detect the targets of small dimensions.Secondly,in the original calculation of regression loss function,when there is no intersection between the prediction box and the background box,the intersection ratio IOU cannot be calculated,and problems such as divergence will occur in the training process.Taking into account the distance between the target and anchor,overlap rate,scale and penalty term,the target box regression becomes more stable and the regression loss function can be better calculated.3)The optimized models were tested and verified and compared.Firstly,insulators of different backgrounds and types are collected.In order to make the model have better generalization,the data set is expanded to some extent.Then,the optimized algorithm and the pre-optimized algorithm,SSD YOLOV4 and YOLOV5 were used to predict the data set respectively.From the simulation results,it can be seen that the optimized Faster RCNN is not only more accurate than other algorithms,but also can accurately identify insulators of small targets in complex situations and with low brightness.After verification,the optimized Faster RCNN has a certain application prospect.There are 46 figures,11 tables and 56 references in this paper.