Research On Key Technologies Of Image Intelligent Recognition In High-speed Rail Automatic Inspection System

With the development and progress of society,people’s quality of life is getting higher and higher,and people’s travel is more and more convenient.Railway transportation has become an important means of transportation for the masses because of its safety,convenience,affordable and all-weather transportation.The high-speed railway has the advantages of high speed and comfort.The development of high-speed rail brings convenience to our travel,and we must also pay attention to the safety of railway transportation.The traditional method of railway safety detection is through manual detection.Its shortcomings are low efficiency and a lot of manpower,and manual detection has different evaluation criteria.Therefore,it is imperative to use technology to realize railway safety monitoring.Since the 1960 s,image processing detection technology has gradually attracted attention and become a hot spot.Traditional target recognition detection algorithms and edge detection algorithms have problems to some extent,so this paper improves on these issues.This paper first discusses the status quo of the image target detection at home and abroad and its basic processing steps.The high-speed rail image is experimentally tested using traditional algorithms to obtain experimental results,and then the new algorithm is improved or studied and applied to high-speed rail image detection.The main work of this thesis is as follows:(1)The traditional algorithms including feature extraction algorithm,classification algorithm and edge detection algorithm are applied to the high-speed rail image to obtain experimental results and summarize their advantages and disadvantages.(2)Improve it against defects in traditional algorithms,or increase the functions that are missing in traditional algorithms.The traditional algorithms studied in this paper include HOG feature extraction algorithm,SVM classification algorithm and Canny edge detection algorithm.The improvements made in this paper for different algorithms are:(1)Adding PCA to reduce the dimension based on the original HOG and SVM;(2)Performing GPU-based parallelization processing on HOG features;(3)Proposing a probability based on The target detection algorithm of the topic model;(4)The Canny algorithm is improved by adding short-term memory threshold and tracking algorithm.The first two points are to improve the speed of abnormal detection of high-speed rail fasteners.The third point is for high-speed rail contact network detection.