Image Recognition Algorithms Of The Typical Failures Of The Freight Train

Along with the railway transportation enterprise of our country rapidly grows, the journey speed and freight volume of freight train has been a sharp increase. The traditional safeguard system can't meet demand. Therefore, that desigining the automactic recognition system by analyzing the images collected form the freight train based on the pattern recognition theory, has the extremely important practicability and the broad fundamental research prospect.Image recognition algorithms on the failures of freight train, are reseached in the intial stage. In this paper, four kinds of recognition algorithms on those failures of the freight train in the image is respectively researched on the issues as: losing failure of coupler yoke bolts screw, losing failure of train bogie center plate screw, dropping failure of safety-chain, and bending failure of across bar. This paper focuses on the feature extraction, feature descriptors and the classification for recognizing those feature of the failure. The main contents of the dissertation are as follows:Recognition algorithm on the lost failure of coupler yoke bolts screw is researched. The segmentation feature of the image of coupler yoke bolts screw is analyzed. The two steps segmentation strategy is proposed. Based on the gray-scale projection algorithm(GSP) and its rapidity, the installation region of the screw is extracted by the strategy as follow: the process region is selected based on the priori knowledge, then the region is processed by GSP at different levels to obtain the region. The screw region in the installation region is extracted by using the texture template match method with the variable-weight. For describing the feature of the screw region, which has the similar background feature, this proposed a threshold segmentation algorithm based on the area of background region(TSAB). Compared with the Ostu Method, the reliability and advantage are discussed. Based on the TSAB, the linear segmentation is used to identify the failure.Recognition algorithm on the lost failure of bogie center plate screw is researched. All images is divided into two kinds according to the type of bogie. Five kinds of segmentation feature is defined for each kind. For extracting a linear segmentation feature in all images,an algorithm on the fast extraction of the linear segmentation feature based on the gradient dependent of gray projection is proposed. The gray distribution of image is processed by the gray projection with rectangle window, the linear segmentation feature is extracted by analyzing the gradient dependent of these extreme points of the gray distribution curve. The center plate region is extracted by using the algorithm combined with the conventional method, then the screw region is extracted by using the TSAB. For describing the shape feature of the center plate screw, a descriptor is designed for the contour of the irregular-closed graphic. The Constrain Rectangle(CR) and Contour Measurement Matrix(CMM) is built to analyze the graphic shape, and a series of Eccentric-Rate is defined to describe the contour. Compared to the Fourier Descriptor(FD), the practicality of the descriptor proposed in this paper is demonstrated. By combining the descriptor with Hough transformation, the characteristical parameter set for the pattern of center plate screw is built. Based on the fuzzy theory and fuzzy reasoning, the characteristical parameter set is converted to the fuzzy set. The decision tree is designed on the basis of the fuzzy set, the if-then rule is designed for identifying the losing failure of train bogie center plate screw.Recognition algorithm on the dropped failure of bogie safty-chain is researched. The feature of the dropping failure of bogie safty-chain formed by non-rigid mechanical structure is analyzed, the method for analyze the feature is proposed as followed: the histogram equalization and the self-adapting thresholding method are applied to pre-process the image, then the edges of the image is extracted by using the Sobel operators. On the analysis of the edges distribution, a descriptors for the Convex Polygon(CP) is proposed. The principle is discussed in details. Compared with conventional method, the detection method has advantage of overcoming the problem of detecting the CVP shape with discontinuity and broken edges, and thus worthy of being promoted.The safety-chain region is extracted by using the descriptors with Hough transformation with angle constraint and the priror knowledge. For analyzing the feature of dropping failure of bogie safty-chain, detection model on the normal pattern is applied on the classification based on the ratio between the area of foreground and background.Recognition algorithm on the bending failure of across bar is researched. The feature of the region of the across bar is analyzed. The segmentation feature is extracted by using the vertical projection of the horizontal gradient. For the feature of the bending failure, a descriptor for the failure is proposed. The image is cut two parts, then each parts is processed by the Hough transformation, the difference between the angle average of each parts is defined as the core parameter of the classification, the sub-section discriminant function is designed for the bending failure.The threshold value of the discriminant function is optimized by using the Bayes decision with minimum error probability.The overall structure,classificaiton and pre-process module for the images and the failure information database of the automatic recognition system of the failure of freight car is researched. Based on the theory of design pattern, the integrated strategy of automatic recognition algorithm by using the Strategy and Singleton pattern is studied. The separability and expansibility of the automatic recognition system can be guaranteed by using the integrated strategy. Based on the statistical analysis, the recognition rate and operation time of four kinds of automatic recognition algorithm is studied.