At present, most studies rely on a large number of annotated samples to describe the feature distribution of known categories when solving the open set recognition problem of image classification models, which may lead to two problems: when the annotated samples are insufficient, the model training is easy to fall into the local extreme value; The decision boundary constructed according to the feature difference of known classes only reflects the feature distribution among known classes and lacks the generalization of open sets. In this paper, we propose a separate independent classification network structure, in which each category contains an independent linear feature layer. The neural nodes designed in the feature layer can capture the category features more accurately under limited data samples. At the same time, a class of negative samples without labeling is introduced in the model training, so that the model not only relies on the feature difference of the known categories when constructing the decision boundary, but also increases the open set generalization of the model decision boundary without adding additional labeled samples. The experimental results show that the open set recognition algorithm based on independent classification network and adaptive open set training on FDS and Imagenet-Crop data sets has better open set recognition performance than the existing open set recognition algorithm.