Research On Image Generation Technology Of Pantograph And Catenary Based On Segmentation Fusion And Generative Adversarial Network

The pantograph-catenary system is an important equipment in the traction system,and intelligent detection of it helps to ensure the safety and stability of the pantograph-catenary system.The image data of the pantograph-catenary is the basic condition for the intelligent detection of pantograph-catenary,so the generation and enhancement technology of image data has important research significance and research value.According to the different characteristics of the collected images of pantograph-catenary,this paper designs two ideas for the conversion of pantograph-catenary.According to the characteristics of single image information of pantograph-catenary in the dark night and clear outline of pantograph-catenary,this article first uses the U-net to segment pantograph-catenary in the original image,and then uses Poisson fusion to fuse the segmented pantograph-catenary with the new background image.When segmenting,this article uses 300 pantograph-catenary images and their annotation data.At the same time,several image transformation functions such as inverse scale transformation,arc transformation and three transformations are constructed to process the original pantograph images,and the processed images are corresponding to the annotation data.Train the U-net for comparison experiments;among them,the three-transformed image has the highest segmentation accuracy,and the segmentation result is used for Poisson fusion with the collected background image;a composite image with real visual effects,uniform overall tone,and smooth edge fusion is obtained.According to the characteristics of the main body of the image information and the single color of pantograph-catenary image in the daytime,this article uses the CycleGAN to convert the background of pantograph-catenary image,trains two models of depth transformation and color transformation through the collected data set,and uses the trained model to compare the original,The image processing of pantographcatenary image is used to obtain the image after the background transformation;the peak signal-to-noise ratio and the image information entropy are used to calculate the image before and after the transformation.The comparative analysis shows that the original image is the basis for preserving the main body information of the original image after the background is transformed by the CycleGAN.The information on the image is more enriched.Both of the two methods used in this article to generate pantograph-catenary background have obtained visually realistic images,while retaining the original pantograph-catenary information while adding new image information to pantograph-catenary data.This provides a data foundation for the follow-up research on pantograph-catenary problem,such as the identification and positioning of pantograph-catenary,and the detection of arcing.