Algorithm Research On Fault Detection Of EMU's Bottom Based On Image Features

Transportation is the lifeblood of a country's economy,which plays the important impetus in economic development.As a new means of transportation,high-speed EMU(Electric Multiple Units)is very convenient to human travel,so the safety of EMU has attracted much attention.Because of complex structure and too many parts of EMU bottom,the traditional manual train inspection method is not only inconvenient,but also more likely to cause economic losses and even casualties due to manual omissions.The Trouble of moving EMU Detection System(TEDS)with image acquisition,image transmission,digital image processing and computer vision technology is to achieve the purpose of intelligent trouble detection.At present,TEDS uses image comparison recognition technology to detect trouble of EMU images,and then combines the manual recheck to determine the EMU trouble.However,the false alarm rate of image recognition technology based on gray difference is up to 80%,which greatly increases the workload of manual recheck.Therefore,according to the different parts of EMU bottom image,this thesis designs corresponding trouble detection algorithm to reduce the false alarm rate.The bottom plate of EMU has simple structure and widely distributed bolts.In view of the loss of bolts,this paper proposes the characteristics of dark edge of bolts,that is,the gray level of bolt edge pixels is lower than that of center pixels.Combined with the FAST(features from accelerated segment test)corner detection algorithm,the recognition and location of bolts are realized.Then,based on the bright edge effect caused by the bolt loss of the bottom plate,that is,the gray level of the edge pixel of the bolt lost is higher than that of the center pixel,the bolt loss detection is realized.The structure of axle,drawbar and other parts at the bottom of EMU is complex,and there are many other parts of EMU.In view of the deep learning method based on image classification,which can only detect the key parts of EMU,this thesis designs an algorithm to detect the bottom image of EMU in an all-round way through the idea of image comparison.In view of the problem of image block size,this thesis proposes a strategy of segmenting image to make image block of EMU contain enough information without information redundancy.In view of traditional image registration which has low accuracy,this thesis combines the feature-based registration method and template-based matching method to propose the secondary registration algorithm to achieve highprecision registration of EMU image blocks.For the feature comparison of image blocks,this thesis proposes Haar and histogram of oriented gradients feature extraction method of EMU image blocks,and realizes the image block comparison by several different feature similarity measurement methods.Finally,this thesis achieves the fault detection algorithm based on the direction gradient histogram feature and the standardized European distance measurement.Finally,the algorithm is simulated by the existing image database.The experimental results show that the bolt loss detection algorithm proposed in this paper can recognize the bolt loss accurately,with low error recognition rate and high recognition efficiency.The fault detection algorithm proposed in this thesis for the axle and other areas can significantly reduce the error detection rate on the basis of ensuring the fault recognition rate.Therefore,the research in this paper provides an effective theoretical basis and algorithm support for the improvement of TEDS image recognition technology,and has a good application prospect.