2017 2nd International Conference on Information Technology and Management Engineering (ITME 2017) ISBN: 978-1-60595-415-8 Detection of Rail Fastener Based on Wavelet Decomposition and PCA Ben-yu XIAO 1, Yong-zhi MIN 1,* and Hong-feng MA 2 1 School of Automation and Electrical Engineering, Lanzhou Jiao tong University, Gansu Lanzhou, China 2 School of Electronical Information Engineering, Lanzhou Institute of Technology University, Gansu Lanzhou, China *Corresponding author Keywords: Wavelet decomposition, Rail fastener, deletion detection, Principal component analysis. Abstract. A new method based on wavelet decomposition and principal component analysis (PCA) is proposed to solve the problem that the traditional method cannot effectively and quickly detect the rail fastening nut. Through the wavelet decomposition of the rail fastener image, the high frequency component of the image can be removed, and the noise can be reduced and the running time of the algorithm can be reduced. The principal component analysis method is used to reduce the dimension of the image, and the minimum distance classifier is used to detect the rail fastener. The experimental results show that the proposed algorithm can effectively detect the missing state of rail fastener, and the algorithm is robust to the occlusion of the noise image and rail fastener image. Introduction The fastener plays an important role in ensuring the safe operation of the train [1]. However, due to the vibration of the train and other reasons, the fixed fastening nut is often missing, bring a security risk to the normal running of trains [2]. France, the United States, Germany, Japan and other countries have long been committed to the use of computer graphics technology to achieve visual inspection of track structure components [3]. Among them, the Italy researchers Mazzeo and his partner have designed a fastener system nut missing detection device, which has achieved some results by now [4-6]. However, this research uses the neural network to classify the feature, it is easy to fall into the local optimum, besides, it needs a lot of training sample [2]. In view of the shortcomings of the existing detection methods, this paper provides a method for the detection of rail fastener based on wavelet decomposition and PCA. By using wavelet transform to decompose image and to remove the image noise, using principal component analysis to extract image features to reduce dimension of image, a minimum distance classifier is used to detect the missing of the fastening nut. The experimental results show that the algorithm of this paper can achieve a better detection of the missing of the rail fastening nut, and it has a certain robustness. Process Flow Image Acquisition In this paper, the image is collected by a high speed CCD camera mounted on a rail car which was designed by ourselves(as shown in Figure 1).The rail surface defect detection system including image acquisition module, image processing module, display module and power module. The image acquisition module uses the CCD GE680 camera with a resolution of 640 480Pixel and the focal length of the 8mm lens, the image processing module uses the NI industrial computer. Using the external circuit to trigger the camera to capture the rail images and then transfer them to the industrial computer. 163
Figure 1. Rail inspection car. In the track inspection car, CCD camera and 4 LED strip light source are placed in a closed box, these LED strip light source are arranged in the closed box to provide a light source to the CCD camera(as shown in Figure 2).In this way, the interfere of external light source can be reduced. Wavelet Transform Figure 2. The interior design of box. One characteristic of the image data is that the gray value of the neighboring pixels is largely related to each other. In the frequency domain of the image, when the information in a wide range gradually change we call it the low frequency information, while the information in a small range changes quickly are called the high frequency information. Among them, the low frequency information reflects the whole information of the image, and the high frequency information contains the details of the image and the noise signal. Image quality will declines for it is easy be polluted by noise pollution in the transmission process. Considering that the edge details are not so important in the rail fastener detection, so We can have wavelet decomposition on image to filter out the high frequency information in the image and achieve the purpose of removing the image noise. At the same time, because some of the details of the image are filtered out, The amount of computation of the algorithm is also reduced. The result of rail fastener image decomposed by wavelet is shown in Figure 3. (a) (b) Figure 3. Wavelet decomposition of the fastener image. Figure 3(a) refers the rail image containing noise, after one layer of wavelet decomposition, the rail fastener with noise is be decomposed the 4 parts as shown in figure 3(b). From the Figure 3(b), we can see that after wavelet decomposition, the interfere of the noise bring to the low frequency part of the image is reduced[7]. Although the low frequency part of the image lose some edge information and make the fastener edge become blurred, but the overall characteristics of rail fastenings are still intact. In our a large number of simulation experiments, when we use DB4 as wavelet base and select the layer of wavelet decomposition as 3, the algorithm can get the best effect. 164
PCA Feature Extraction The PCA method is used to get M features to replace the original N features (M is much less than N), the M new features are the linear combination of the old features, and these new features are independent of each other. By using PCA, we can reduce the computational complexity and the identification error caused by information redundancy. By calculating the Euclidean distance of the fastener to be measured to the training fastener set to complete the detection of rail fastener state. Some of the characteristic images of the rail fastener are shown in Figure 4. (a) Figure 4. Some of the characteristic images of the rail fastener. (b) Algorithm Implementation The steps of wavelet decomposition and PCA based detection algorithm for rail fastener are as follows: (1) Selecting 60 images from 280 acquired rail fastener images as the image of a sample set. The sample images contain 12 rail fastener missing images and 48 normal rail fastener images. The remaining 220 images contain 17 rail fastener missing images and 203 normal rail fastener images are used as the test image set of this experiment. (2) Wavelet decomposition of both the sample set and the test image set to get the low frequency samples image set and the low frequency test image set which are constituted by the low frequency sub band image. (3) Using the PCA algorithm to extract the features of the low frequency sample image set and the low frequency test image set separately. (4) Using the minimum classifier to classify the image of the normal rail fastener and the missing of the nut of the rail fastener, and realize the detection of the missing of the rail fastening nut. Simulation Experiment and Analysis We selected 280 images form the images collected by the rail image acquisition car as the object of simulation. In these 280 images, there are 29 images absence of fastener and in the rest of the image, there are no lack of fastener. We select 12 rail fastener missing images and 48 complete rail fastener images as the sample image set of the experiment. The rest of the images are used as the experimental test image set. In order to reduce the complexity of the operation, all image sizes are set to 256 341, the gray level is 256.we used the R2014a MATLAB as the software platform of this simulation experiment. Through the simulation experiment, the final recognition rate of the algorithm is 94.64%. Among them, All the errors are due to the integrity of the rail fastener images are judged to the lack of a fastener image. There is also need to consider this situation that rail fasteners are covered by stone ballast, food bag and some other things. In the simulation experiment of this paper, the rail fastening images are artificially add a continuous block to simulate the condition of the rail fastener being obscured [9]. Partially occluded part rail fastener images are shown in Figure 5. 165
(a) (b) (c) Figure 5. Partially occluded part rail fastener image. For rail fastener image with continuous occlusion, the recognition rate of the proposed algorithm is 89.29%. Among them, the wrong part includes not only the integrity of the fastener image is judged to the missing fastener image, but also the lack of fastener image is judged to the integrity of the fastener image. The reason is due to the majority part of fastener is be obscured, making the algorithm cannot extract the correct feature space, so the algorithm cannot make the right judgment. Summary In this work, a combination of wavelet transform and PCA feature extraction rail fastening nut missing detection method is proposed, by using wavelet transform of the image of decomposition to remove the image noise, by using PCA algorithm to image for dimension reduction reducing the complexity of the image processing. Simulation results show that the proposed algorithm of this paper is effective and has robustness to noise and occlusion of rail fastening images. In this paper, the algorithm is mainly aimed at the detection of the overall lack of rail fastener, for the existence of the fastene elastic bar, but the nut is missing and other possible situations are needed to be further studied. Acknowledgement This research was supported by the National Natural Science Fund (Grant No. 61663022 and No. 61461023) and The Open Subject of the Key Laboratory of high altitude traffic information engineering and control in Gansu Province (Grant No. 20161105). References [1] Meng W, Jianzheng C, Shen L. Track fastener state detection based on image processing [J].China Computer and Comunication: Theory Edition, 2013(1). [2] Ling W, Bing Z, Xiai C. Inspection system for loss of rail fastening nut based on computer vision [J]. Computer Engineering and Design, 2011, 32(12):4147-4150. [3] Hailang L. High speed integrated detection system for train video monitoring [J]. Railway Technical Innovation, 2012(1):20-22. [4] Ruvo G D, Ruvo P D, Marino F, et al. A FPGA-based architecture for automatic hexagonal bolts detection in railway maintenance[c].proceedings of the 7th International Workshop on Computer Architecture for Machine Perception, 2005:219-224. [5] Marino F, Distante A, Mazzeo P L, et al. A real time visual inspection system for railway maintenance: automatic hexagonal headed bolts detection[j].ieee Transactions on Systems Man and Cy-bernetics Part C, 2007, 37(3):418-428. [6] Yella S, Dougherty M, Gupta N K. Condition monitoring of wooden railway sleepers [J]. Transportation Research Part C, Emerging Technologies, 2008, 17(1):38-55. 166
[7] Wang Xiaohua, Zhao Zhixiong. PCA face recognition algorithm combined with gamma transform and wavelet transform. Computer Engineering and Applications, 2016, 52(5):190-193. [8] Jianming Z, Pei L, Honglin W, et al. A mixed denoising algorithm based on sparse representation and noise distribution prior knowledge[j].computer Engineering and Science, 2015, 37(10):1917-1923. [9] Cheng C, Ming Z, Junping Z. Weed seeds classification and recognition based on compressed sensing theory[j] Science China: Information Science, 2010(S1):160-172. 167