Research On Recognition Algorithm Of Transmission Line Small Target Based On Deep Learning

The With the development of deep learning,the inspection of UAV circuit components based on deep learning has become a mainstream trend of social development.By collecting a large number of transmission line images,training,to achieve the effect of classification detection.In the actual image,because the insulator,shock hammer,spacer and other objects are relatively small pixel image,the semantic information is less,the traditional convolution neural network method for these three types of transmission line typical small parts detection effect is not good,so this paper proposes a transmission line small target detection method based on deep learning network.The main research contents are as followsFirstly,aiming at the problem that the existing open source data set does not contain the data set of transmission line detection,the UAV self inspection photography and web search are used to collect the data set,and the collected data set is expanded by rotation and brightness adjustment,and then annotated to generate pascalvoc2007 format XML file,which provides the data basis for transmission line inspection.Secondly,in the transmission line detection,due to the aerial angle problem,the detection components are easily blocked,including the transmission line components are blocked by each other and the tower or other objects.With the deepening of deep learning network,the occluded components are prone to semantic loss in high-scale feature layer.To solve this problem,this paper uses RESNET residual structure to adjust the yolov3 module to improve the receptive field of each feature layer.Finally,aiming at the problem that the detection target proportion is too small in the detection of transmission line components,the convolution layer,feature layer and activation function of yolov4 algorithm are optimized,and the adjusted yolov4 algorithm is used to detect the transmission line components,so as to improve the detection speed and accuracy,and the GUI interface is created by MATLAB.The experimental results show that the experimental speed of the improved algorithm is 53.62 FPS,which can achieve the effect of real-time detection,and the detection accuracy is improved compared with the original algorithm and the algorithm based on RESNET.Using the target detection algorithm to identify and locate the typical components in the transmission line image,the real-time monitoring of transmission line condition is realized,and the intelligent degree of transmission line inspection is improved.Research on small target recognition algorithm of transmission line based on genetic algorithm...