Open set classification method and system for network structure optimization

The invention provides an open set classification method and system for network structure optimization, and the method comprises the steps: obtaining an object image which comprises a known-class object and an unknown-class object; the object image is input into a trained MobileNet V2 model with an optimized structure; softmax is used for outputting probability distribution of known classes under the condition of closed set classification, and Openmax is used for outputting probability distribution of unknown classes under the condition of open set classification. The structure-optimized MobileNet V2 model is realized based on an original MobileNet V2 model, a network structure of the structure-optimized MobileNet V2 model is composed of two lines which are respectively used for closed set classification and open set classification processes, and the two lines share a backbone network. According to the open set classification method for network structure optimization, the recognition of known objects and the