Visual Inspection Models And Algorithms For Rail Surface Defects

High-speed railway has grown to be the major mode of transportation in China.With the sustained development of railway network,the safe operation has become increasing-ly prominent.Thus,rail daily maintenance is required to guarantee the efficiency,security and longevity of the whole system.Visual inspection has attracted much attention,due to the merits of automatism,rapidity,non-destructivity and objectivity.However,espe-cially in the field of rail defect inspection,it still faces challenges(eg,various visual appearances).To deal with the complicated imaging environment,complex evolution-ary process,slight surface defects with arbitrary scale as well as sparsity,the dissertation conducts a research including "model construction "mechanism analysis","theoretical analysis",and "performance evaluation",and focuses on increasing the accuracy,versa-tility,security,stability,objectivity and intelligence of the inspection system.Besides,the research in this dissertation is of certain significance to enrich the theoretical framework of visual inspection.The innovation of the dissertation mainly includes:1)The dissertation proposes a novel method named as dynamic-background and structure-sparsity decomposition(DSD)model,assuming that the defect image can be represented as the superposition of background,defect and noise components.Specifically,a linear dynamic depicting matrix instead of commonly used low rank is applied to construct the dynamic background and address the decomposition problem of high coherence between the sparse and background components.In the structure sparsity regularization term,the tree structure of superpixel is adopted to constrain the sparse components and obtain more compact and complete defect regions.The experimental results show that the proposed model can achieve better decomposition performance,as compared with other typical decomposition models.2)Based on the prior information that pixel intensity appears relatively consistency a-long longitudinal direction to form some main clusters,the dissertation explores the information of these main denser regions to locate defects and further transforms the track surface defect inspection problem into the background distribution modeling.Accordingly,we propose a hierarchical inspection framework containing coarse ex-tractor and fine extractor.Specifically,the coarse extractor handles this defect inspec-tion problem by exploring characteristics of rail background instead of defect itself,and focuses on finding the background modes.The fine extractor,which integrates the longitudinal context information and transversal prior information,is proposed to largely reduce the impacts of other irregulars.The experimental results demonstrate that our method outperforms the state-of-the-art works.3)This dissertation reformulates the inspection task by considering specified charac-teristics of the track and proposes a background-oriented defect inspector(BODI).specifically,a stochastic strategy is utilized to obtain the background model to reduce the computational cost,assuming that it is more reliable to estimate the statistical distribution of a background pixel with a small number of close values than with a majority of all values.Then,a sufficient number of the random selections generates adequate and diverse background statistics,and defect-determination and a fusion of procedures determine whether current pixel belongs to the background.Finally,a background update mechanism and parallelism ensure real-time applicability.The experimental results show that the proposed method meets the requirements of real-time detection while satisfying the accuracy detection.