Aiming at the important item of safety maintenance in the manual daily maintenance process of metro trains, high-frequency vision sensing technology is used to image the real-time high-definition appearance of key components and systems of trains passing through the inspection area, including the walking part, traction, receiving flow, brake air supply, doors and windows, air conditioning, etc. Through artificial intelligence technologies such as computer vision, deep learning and data-driven graded warning, the real-time warning of component abnormalities is realized to achieve the purpose of improving operational efficiency, reducing the amount of manual maintenance and lowering operation and maintenance costs.