Research On Orbital Edge Detection Algorithm Based On Deep Learning

Urban trams are an important component of the urban public transportation system,and when mixed with other means of transportation,they are easily affected by foreign objects on the track,which can affect driving safety.With the rapid development of the new generation of fully automated driving technology and video surveillance technology,intelligent detection of track foreign object intrusion has become a new research hotspot,and track edge detection is gradually developing towards intelligence and automation.By using technical means such as computer vision,automatic recognition,analysis and determination of track edges can be achieved,thus greatly improving the efficiency and accuracy of detection.This not only reduces the cost of manual operation,but also reduces the impact of human factors on the test results,so as to better guarantee the safety and stability of the railway train operation.Therefore,it is of great significance to develop the research of track edge detection method.This article focuses on the problem of low accuracy in track edge detection in complex environments.By using research methods such as image enhancement algorithms and edge detection algorithms,we conduct research on track edge detection algorithms based on deep learning.By improving the Retinex image enhancement algorithm and RCF edge detection algorithm,we propose a track edge detection algorithm based on deep learning.The main research content is as follows:(1)Aiming at the problem of significant information loss caused by the influence of light and environment on track images captured under complex environmental conditions,an improved Retinex based track edge image enhancement algorithm is proposed.On the basis of Retinex algorithm,first design an improved Bilateral filter algorithm to process the reflected image,remove the interference of noise while maintaining the edge information,then introduce gamma correction to rescale the nonlinear illumination image,enhance the contrast and perception characteristics of the image,and finally design an improved wavelet transform to fuse the enhanced reflected image with the illumination image to obtain a clear and more detailed image.The simulation results show that the improved Retinex algorithm can enhance the track edge image more effectively than histogram equalization and Retinex algorithm.(2)Aiming at the problem of the problem of existing edge detection models,such as the low track edge detection accuracy in complex environments,a track edge detection model based on improved RCF is proposed.On the basis of the RCF model,the deep deconvolution operation is removed,the feature fusion module is added to enhance the ability of deep feature network to express trackedge features.The multi-receptive field module is designed,and it is used to replace the last concat layer to increase the effective receptive field.The efficient attention module is introduced to extract channels conducive to feature detection,reduce the noise of edge detection,and improve the loss function.The simulation results show that compared with HED and RCF algorithms,the improved RCF algorithm can detect track edges more efficiently.