The trouble of moving electric multiple-unite (EMU) detection system (TEDS), can monitor comprehensively the status of China railway high-speed (CRH) train, by using the high-speed linear cameras located besides the railway. We fulfill the automatic identification and inspection of the target in the high-quality image by using the machine learning and pattern recognition technology. However, the image taken by the linear camera is susceptible to the speed of vehicle, and there is a geometric deformation in the horizontal direction of image, which brings difficulties to the automatic recognition and detection of the subsequent target. To solve this problem, we set up a set of reference images, register and re-divide the images obtained in other time, according to the corresponding reference image, so as to minimize the influence of train′s speed on imaging deformation. Combined with TEDS, we use multi-resolution image fast registration method to achieve the rapid segmentation and alignment of subsequent target images. Hence, we propose an improved image subtraction technique, in order to quickly and automatically locate and detect the fault region by comparing and analyzing the aligned image pairs.