Open Set Image Classification Using Normalized Main Class Distance of Feature Space in Intelligent Retail

Image classification technology is often used to identify goods in intelligent retail scenarios.However,not all the objects in the scene are known by classification system,and unknown objects will interfere with the performance of the system in the scene.Aiming at the problem of unknown objects classification in intelligent retail scenarios,this paper proposes a classification prediction method for unknown objects by calculating and modifying the classification features of closed data sets with known classes.We construct a feature space of known categories,and design normalized main class distance of image classification feature space to fit the boundary probability model of the feature space.Finally,the boundary probability model is used to modify the original classification features,and the classification prediction of unknown classes of objects is obtained.Experiments are conducted on datasets of intelligent retail scenario.Experimental results show that the rejection rate of open set predicted by our method has increased by 14.20%,reaching 44.85%.