Research And Application Of State Recognition Algorithms For Railway Fasteners Based On Image Processing

With the rapid development of China’s railway transportation industry,the railway mileage is increasing year by year,and the pressure of railway line inspection is increasing.It has become an urgent need for railway engineering departments to design a system that can automatically inspect the condition of railway line.Rail fastener system is one of the important components of railway track surface infrastructure and an indispensable part of the whole track system.It plays the role of fixing rails on sleepers or coagulation base to ensure the safety of train operation.Therefore,it is of great theoretical and practical significance to study the track fastener state recognition algorithm based on image processing.Firstly,in order to recognize the status of fasteners normally,according to the characteristics of actual track images,a fastener location method based on morphological processing is constructed.The locating coordinates are quickly obtained by using the edge features of sleepers,rails and shoulders,and the fastener sub-images are located from the original image according to coordinate information.Aiming at locating the sub-image of fastener with deviation,fully mining the feature information of fastener area and non-fastener area,an algorithm of automatically adjusting the location coordinate of fastener sub-image based on super-pixel processing is constructed.The sub-image of fastener is segmented into several super-pixel areas by using super-pixel processing method,and then the five features of super-pixel are used to aggregate the super-pixel areas into several larger ones.The clustering region is merged according to the color features of the fastener region,and the fastener region is segmented from the sub-image.Finally,the coordinates of the original fastener sub-image are adjusted according to the coordinate position of the fastener region in the sub-image.This method improves the positioning accuracy of the fastener sub-image.Secondly,in order to obtain the local features of the fastener image,a Markov model of the fastener image is established.The model is used to prove the relationship between the state of the key area and the state of the fastener image.The LBP uniform pattern feature of the key area is taken as the local feature.The global feature of fastener image is HOG feature.Firstly,the same fastener image is processed several times with different bandwidth of Gauss kernel function,retaining the external shape feature of fastener and eliminating local details.Then,the HOG features of all images processed by Gauss kernel function are extracted,and all HOG features of the same fastener image are cascaded sequentially to form the global feature of the fastener image.Thirdly,in view of the small number and lack of diversity of the fastener images in the sample library,all the fastener images are divided into two parts according to the symmetrical axis,and then the fastener images are reconstructed by mirror image and mosaic method.This method will increase the number of images in the training sample library and improve the diversity of the sample library.Finally,a hybrid identification algorithm for fastener States is proposed,which combines two training models.The global feature is used to train the support vector machine and the local feature is used to train the BP neural network.The two training models are used to identify the state of the fastener image.Their discriminant results are obtained through logic and operation.The results show that the recognition rate of the fastener state based on the two models is higher than that of the single model.