Crackle-Image of Swing Bolster Detection Algorithm Based on Visual Perception

Abstract

A novel method based on the visual perception mechanism was proposed for solving the problem of fault-image detection of the running train. The detection of swing bolster crack could be achieved in high efficiency with small samples. Firstly, the receptive field of simple cells in the primary visual cortex was obtained from the image sequence by using the ICA model. Secondly, the neuron response of normal and fault image was calculated, then the neuron responding stronger to the stimulus could be found out, and its corresponding content for fault detecting could be output as well. Experimental results demonstrated that this novel method had a high fault-detecting rate.