Abstract: The automatic fault recognition function for TEDS (Trouble of moving EMU Detection System) of singlepoint operation has the problems of insufficient recognition accuracy and high misjudgment rate. This paper proposeda fault image recognition method for EMU based on multi-source data. Based on TEDS data of network operationand combined with traditional difference detection method, the paper carried out the multi-source data fusion andweight difference calculation for the same train image collected by TEDS in different space and time, and implementedabnormal parts detection of EMU car body. Experiments show that the proposed method establishes a more accuratecomparative reference source, reduces the impact of the environment on the image content, improves the automaticrecognition rate of EMU operation fault, and reduces the false alarm rate.