Research On The Detection Method Of Railway Locomotive Speed Sensor Based On Machine Learning

With the increase in the use of high-speed trains,the number of maintenance tasks for trains has also increased.The inspection of traditional locomotive equipment is done through manual methods.The maintenance workers need to go to the scene every day to inspect the locomotive equipment through the naked eye to see if there is any missing or damaged equipment.This day-to-day work is exhausting and cumbersome.The quality assurance of the inspection work also depends on the personal professional quality and psychological and physiological status of the workers.Therefore,this traditional manual detection method is no longer suitable for on-site needs,but also has a certain security risk.In recent years,digital image processing and machine learning technologies have been successfully applied to all walks of life.Pedestrian detection,face recognition,vehicle detection and other applications are relatively mature.The locomotive speed sensor in this article belongs to the photoelectric speed sensor.Its role is to send the speed change of the locomotive through the pulse signal to the locomotive monitoring device.Its image feature is expressed as the outer inner circle,which is composed of four bolts and connecting coils.Peripheral squares,inside a square with a raised circular device.Therefore,this paper proposes an automatic detection method for locomotive speed sensor equipment based on image processing and machine learning.It uses the method of positioning after detection first.The main work is as follows:(1)Firstly,the speed sensor installation area is located.This paper introduces three kinds of regional positioning algorithms: regional positioning algorithm based on Hough circle transform,positioning algorithm based on HOG feature and SVM algorithm(HOG-SVM),and positioning algorithm based on Haar feature and Adaboost algorithm(Haar-Adaboost).Among them,the Haar-Adaboost algorithm and the HOG-SVM algorithm based on machine learning theory detect 1000 test pictures that do not overlap with the training sample under the same training set of 6659 positive samples and 15273 negative samples.The experimental results verify the adoption of this article.The effectiveness and feasibility of the positioning algorithm.Among the three methods,the HOG-SVM algorithm is more effective.(2)The device detection is performed on the correctly positioned speed sensor installation area to detect whether there is a speed sensor device in the area.Three kinds of detection algorithms are also introduced: SSIM algorithm and histogram comparison algorithm based on image similarity,and SVM classifier detection algorithm based on HOG feature.In this paper,500 test samples are used to test the three detection algorithms.The test results show the effectiveness of the device detection algorithm used in this paper and its good practical value,and the detection performance of the SVM classifier detection algorithm based on HOG features is better than the other two algorithms.(3)Based on OpenCV and VS2010,an experimental platform was built to implement the above positioning and detection algorithms.By comparing and analyzing the experimental results of each algorithm,this paper finally optimizes the HOG-SVM algorithm to solve the positioning and detection problems of the speed sensor area in the locomotive image,and integrates it into the system for application.After testing,the results show that the detection accuracy has reached the requirements of the application requirements.