Fault Identification And Positioning Of Insulators Based On Multi-angle Information Fusion And YOLO V4

At present,the number of insulators in China has reached billions of pieces.The importance of insulators makes it a key detection object for the safe operation status of the contact network.In China,drones have been gradually used to collect insulator images and use deep learning algorithms to identify and detect insulators.However,due to insulator masking and improper acquisition angle when the actual acquisition of images,the insulator fault identification algorithm based on multi-angle information fusion model combined with the improved YOLO v4 network can solve the problem of insulator leakage due to improper masking and acquisition angle,and greatly improve the efficiency of intelligent inspection of transmission line insulators.There is insufficient masking and database acquisition information for multi-target insulator images,and there are problems such as leakage detection and low accuracy in traditional convolutional neural networks.Compared with the traditional fault identification method,the multi-angle information fusion model can accurately achieve multi-target detection of insulators.Therefore,a multi-angle information evaluation model is proposed for the information loss caused by the acquisition of insulator images from different angles.Firstly,the optimal observation angle range is calculated by the insulator angle information entropy,in order to obtain the relatively complete data characteristics of the insulator,secondly,the insulator multi-angle image is fused by Cutmix algorithm to ensure that the feature information remains unchanged while reducing the training data,thereby improving the efficiency of network operation,Finally,in view of the leakage detection caused by insufficient extraction of insulator information,the insulator image recognition based on the yolo v4 network of improved feature pyramid is applied to evaluate multi-angle information,the extraction of advanced semantic information of insulator is realized through the feedback of recursive feature pyramid,and the identification of insulator images using different hollow convolution rates is effectively solved.By comparing the Loss values before and after the improvement of the network,and comparing the accuracy with YOLO v4 network,SSD algorithm and DSSD algorithm,the insulator recognition accuracy reached 92.4% and the real-time rate reached 43 frames/s.Through data analysis,the fault identification method of insulator can solve the problem of leakage of insulator due to improper masking and acquisition angle.