VAE and Feature Fusion Based Open-Set Point Generation Network for Open-Set Radar Models Recognition

Zheng Yang,Xiyu Fan,Yifang Zhang,Caijin Xu,Guanchun Wang,Xiangrong Zhang

DOI: https://doi.org/10.1109/ICC59986.2023.10420926

2023-01-01

Abstract:Open-set recognition aims to classify known classes while identifying unknown classes as unknown. Based on this, we try to utilize known categories to maintain an open space and further enhance the accuracy of the open space with the help of generation models. Concretely, for each category we try to find the point in the feature space that is furthest away from its features, and consider the region where this point is located as the open space maintained by the category, which is called the open set point. In the process of finding open set points, we start from two perspectives, the original sample and the generated sample, and first calculate the mean of all sample features that do not contain samples of the current category, which is called the original open set point; In addition, the pseudo-unknown samples are generated using the feature fusion technique based on the attention mechanism and the VAE model to simulate the unknown category samples that may be faced in the real scenario, and such pseudo-unknown samples are referred to as the generated open set points. The original open-set points maintain the open-set space, and the generated open-set points realize the refinement of the open-set space, and realize the precise division of the unknown categories based on identifying the known categories, to solve the radar model open-set identification problem. Experimental results show that our method achieves state-of-the-art results.