If the part manufacturer installs the wrong type of parts in the equipment, it will cause the equipment to work abnormally and directly affect the performance and function of the equipment. With the development of the manufacturing industry, when parts manufacturers produce parts products, in order to adapt to various types of equipment, it is unavoidable that many parts with very close appearances and sizes will be mixed together on the site, which is very easy to mix. Three different models with similar appearance If the part manufacturer installs the wrong type of parts in the equipment, it will cause the equipment to work abnormally and directly affect the performance and function of the equipment. In addition, defective parts may also appear in the production process, and it is necessary to sort the parts. At this stage, the customer's site uses manual screening and classification, screening products are qualified, but the work intensity, poor accuracy, low efficiency, so Xinwei visually according to market demand introduced an online classification screening system to complete this work. The system is combined with an automated mechanism and is detected on-line, with an average detection time of no more than 0.5 seconds per sample. The equipment first arranges the parts into a queue through a conveyor belt and a guide rail, and then pushes the parts into the visual inspection area through a material-dividing cylinder to take a photo inspection. When the detection is completed, the I/O signal is output to the motion control system and the reject/pass action is performed. Mixed visual online classification screening system (qualified product inspection effect chart) Mixed visual online classification screening system (non-conforming product inspection effect chart) Due to the large number of product models on the customer's site and unknown models that will be added in the future, the system uses machine learning algorithms for feature extraction and classification, as well as sampling training and updating parameters.