Defect Detection of Railway Fasteners Based on Improved Pyramid Histogram of Gradients Characteristics

Aiming at the problem of low recognition rate and slow speed caused by the small proportion of key area information in feature vectors of original Pyramid Histogram of Gradients (PHOG) features, an improved feature extraction method of PHOG is proposed. The PHOG feature extraction method is combined with edge feature enhancement method based on Census transform to extract feature vectors of fasteners, and dimensionality reduction is processed by Kernel Principal Component Analysis (KPCA) method to reduce the interference of redundant information. The vector is inputted into the support vector machine for training in order to get the classifier model and realize the automatic identification of the fastener's state. The simulation results show that compared with the traditional PHOG method, this feature extraction method improves the false detection rate by 2.7%, and the complexity of the algorithm is greatly reduced.