Research On The Detection Algorithm Of Railway Fastener Based On Information Entropy

With the rapid development of China’s high-speed railway technology,the maintenance of railway lines is becoming more and more important.As the fastener connecting rail and sleeper,rail fastener is the important foundation to ensure the safe and stable of railway transportation.In recent years,image processing technology,as the main means to realize the automatic detection of railway lines,has been increasing applied to the detection of fasteners.In this paper,the image processing technology is used to study and improve the automatic detection method of railway fasteners.The main contents are as follows:(1)Low-level single features in present railway fastener detection algorithms have poor distinguishing ability of fasteners under different condition and high false positive rate.Aiming at these problems,this paper proposes a fastener detection algorithm which fuses potential semantic topics based on the combine feature of LBP and HOG.Firstly,the algorithm extracts the LBP and HOG features of fastener images and calculated information entropy of them.Secondly,the LDA semantic topic distribution of these two features is adaptive weighted fused by using information entropy.Finally,it used these fused theme distribution to train classifier and detect fasteners.The algorithm by using LDA reduces the redundancy of underlying features and can merge the advantages of the two characteristics;to make the theme of the distribution has stronger descriptive ability of fastener.Theoretical analysis and experiments show that this algorithm effectively reduces the residual rate and false detection rate of fasteners.(2)For the original BOW model ignores the disadvantage of the words location and structure information,this paper proposes a BOW model construction method based on the weighted information entropy.Firstly,According to the structural characteristics of the fastener,the image is divided into four local regions,which prevented the defect that the original BOW model ignored the word location information.Then,the BOW model is weighted fused by using the information entropy of the local region,which highlights the characteristics of the BOW model in the local area,so as to improve the BOW model differentiation of the fastener image.Finally,the BOW semantic topic vector is extracted from the LDA model and sent to the classifier to complete the detection of the fastener state.The theoretical analysis and experiments show that this algorithm for fault fastener significantly reduces the misdetection rate and improves the detection accuracy.