An optimized railway fastener detection method based on modified Faster R-CNN

Abstract

Accurate fastener positioning and state detection form the prerequisite for ensuring the safe operation of rail track. The demands for intelligent, fast and accurate detection cannot be satisfied by traditional methods using image processing and fastener classification. In view of this, a two-stage classification model based on the modified Faster Region-based Convolution Neural Network (Faster R-CNN) and the Support Vector Data Description (SVDD) algorithms is proposed in the paper for fastener detection. Firstly, the data set of detection images is built with the images being labeled, and the classification and detection model based on Faster R-CNN is constructed according to the characteristics of practical fastener images. The anchor box optimization function is established by labeled data set to optimize the box of region proposal network in the model, to enhance the detection rate and accuracy of detection. Then, according to the detection result by Faster R-CNN, the SVDD algorithm is applied for the second stage classification of deviated fasteners, which avoids inaccurate classification caused by different deviated angles of fasteners. Through the verification and analysis of practical detection case, it is verified that the proposed method can improve the efficiency and precision of fastener detection with higher detection rates and accuracy in comparison with other baseline detection methods, making it suitable for fast and accurate detection of fastener states.