Research And Design Based Deep Learning Of Switch Position Recognition Device For High Speed Railway

With the rapid development of domestic high-speed railway,more and more attention has been paid to safety issues.The density between switch rail and basic rail is the direct representation of many faults in high-speed railway.The current density monitoring equipment is based on mechanical structure.With the increase of service life,the reliability of this equipment will be lower and lower.Therefore,in this thesis,machine vision technology is proposed to solve this problem.In order to solve the problem that image features are not easy to extract under complex weather conditions,a deep learning algorithm is proposed.Through the analysis of the existing monitoring technology of high-speed railway turnout,it can be determined that it is feasible to use machine vision to monitor the tightness between switch rail and basic rail.At the same time,through the analysis of the application of deep learning,we can confirm that it is feasible to use the deep learning algorithm to extract the features of fuzzy image.In addition,through the analysis of the switch system fault,the necessity of studying the switch rail and basic rail tightness identification device is determined.In this thesis,the key technologies of image processing are analyzed,among which the traditional image processing technologies such as gray-scale transformation and gray-scale histogram equalization are used in the image preprocessing stage,and these technologies are used to achieve the purpose of image processing in the early stage according to the data simulation results;in view of the problem that image features are not easy to extract under complex meteorological conditions,convolution neural network is used in this thesis The deep learning model is used to solve the problem of inaccurate feature extraction.In this thesis,the maximum pooling algorithm is optimized by using significant similarity constraints.From the data simulation results,the image feature extraction accuracy of the optimized deep learning model reaches the expected goal.In this thesis,through the comparison between the traditional computer hardware scheme and the embedded hardware scheme,the embedded hardware scheme is finally selected to realize the recognition device.In this thesis,the core processor,camera,power supply,communication circuit and extended memory of the recognition device are selected and designed.At the same time,the software program is designed with the idea of layering.The emphasis is to design the program flow chart of image preprocessing,image feature extraction,density calculation and other application tasks.Finally,this thesis tests the identification device from three aspects of calculation accuracy,communication performance and stability.From the test results,the high-speed railway turnout identification device studied in this thesis meets the expected effect.