The visual information in the brain is passed layer by layer. Almost all the visual signals from the retina go through the receptive field of the primary visual cortex (V1 area) and pass on to a more advanced visual cortex after processing. The receptive field of V1 is mainly responsible for extracting the image shape, direction, color and other information, with the spatial domain of locality, time and frequency domain direction and choice, as well as sparse response characteristics. From the view of natural image statistics, Independent Component Analysis (ICA) is one of the main methods to model early computational vision. However, the space arrangement of basic functions (independent components of natural image) decomposed by basic ICA is chaotic and their amplitudes are uncertainty. This decomposition result is contradicted with physiological mechanisms of vision. So, a new computational model is proposed to simulate two important mechanisms of vision which are visual cortex receptive field topology construct and synchronous oscillation among neuron group. To solve the problem of train image fault detection, a new algorithm was proposed based on above compute model. The experiment results show that, the new algorithm can increase fault detection rate effectively compared with traditional methods which absence of above two important mechanisms of vision.