Study On Insulator Image Segmentation Method Based On Deep Learning

In recent years,with the development of machine vision and unmanned aerial vehicle technology,more and more applications have been adopted in the detection of transmission lines.The traditional manual transmission line detection method is inefficient and high in danger.the current patrol inspection of transmission lines based on drones has gradually become mainstream.Insulator as one of the important components of transmission line,is exposed to the natural environment for a long time,which leads to the problem of component damage and aging,which seriously restricts the security of power grid operation and the stability of the system.Therefore,in order to strengthen the entire process management of the transmission line,it is particularly important to conduct periodic inspections of the insulator status.However,most of the insulators in transmission lines are in complex backgrounds with many false targets,which makes it difficult to segment insulators from complex drone aerial images.Therefore,seeking a general and efficient segmentation method is the focus of current research.This paper focuses on the research of insulator segmentation technology in transmission line inspection,in order to improve the accuracy and efficiency of insulator segmentation in complex background,and lay the foundation for the realization of transmission line automation.The main research work of this paper is as follows:Firstly,it discusses the background and research significance of aerial insulator image segmentation,and summarizes and analyzes the current research status and existing problems of insulator segmentation at home and abroad.Secondly,the aerial imagery of insulators is susceptible to interference from mixed noise,making subsequent insulator segmentation more difficult.Therefore,it is necessary to pre-process the aerial insulator image to reduce the influence of noise on the accuracy of subsequent image segmentation.This chapter first analyzes the characteristics of mixed noise in aerial insulator images,improves the disadvantage of the fixed value of the synaptic link strength of the traditional PCNN model,and proposes an adaptive synaptic link strength to achieve better interaction between the central neuron and its neighboring neurons.Function,and realize the accurate positioning of the noise point through the time matrix,reducing the phenomenon of missed judgment and misjudgment of the noise point by the traditional method;secondly in the time matrix,the use of the central neuron and the surrounding 8 neighboring neurons To determine the type of noise points;finally,select the corresponding filtering algorithm for different types of noise points to complete the filtering of noise points.Experimental results show that this method has better peak-to-noise ratio,signal-to-noise ratio improvement factor and mean square error than traditional denoising methods.Thirdly,in view of the characteristics of aerial imagery insulators with complex background and multiple pseudo targets,this paper proposes an adaptive insulator image segmentation method based on FSLIC.Starting from the perspective of superpixels,this paper first analyzes the influence of the number of superpixels K in the traditional SLIC algorithm on the segmentation results,and proposes an FSLIC method for adaptively determining the K value to eliminate errors caused by manual selection of the K value;,Extract the three types of features in color,texture and shape in the superpixel area and fuse to obtain composite features,eliminating the problem of false target interference caused by improper feature selection;finally using the feature similarity to construct the similarity matrix between superpixel areas,Using the clustering of complex network communities to obtain the segmented insulator image through similarity matrix.Fourthly,because the traditional segmentation method is based on the low-level features or shallow features of the vision,it is easy to cause problems such as incomplete feature expression and time-consuming artificial feature extraction.With the development of deep learning,the application of deep learning technology in aerial transmission line inspections effectively solves the problems of poor universality of traditional methods and incomplete feature expression.This paper proposes a S-Deeplab semantic segmentation method based on deep learning.Firstly,based on the DeeplabV3 network,the shortcomings of the Xception model in the traditional encoder structure are analyzed,and the lightweight Mobilenet model is used to replace the traditional Xception model;Secondly,the FSLIC superpixel is introduced and fused with the segmentation map sampled on the network to refine the insulator edge and improve the segmentation accuracy.Finally,through the experiment and the traditional DeepLabV3 network,the precision of semantic segmentation of aerial insulator images is compared to achieve the accurate segmentation of insulators.Finally,summarize all the work of this paper,point out the unsolved problems in insulator division and look forward to the next research direction.