Aiming at the problem of mixed assembly of parts in the production process of enterprises, the background separation algorithm based on Gaussian mixture model is designed to achieve stable and flexible background separation effect. The feature extraction algorithm based on gray level, similar shape and simple contour is used to achieve effective and stable feature data extraction effect. The multi-layer neural network algorithm based on XML data storage is used to realize the background separation effect The effect of dynamic change of item type. Through the image separation, extraction, recognition, to achieve the effect of classification, the results show that the system can not only achieve the stable and intelligent classification and recognition effect of industrial environment, but also ensure more than 99% of the high classification accuracy at the same time, quickly complete the classification task, and provide a simple and effective solution for the demand of market diversification.