A Diffusion Model-Based Open Set Identification Method for Specific Emitters
Wenyan Wang, Zheng Dou, Jiangzhi Fu, Yun Lin
Abstract
This paper proposes an open set identification architecture for Specific Emitters, which consists of four modules: a diffusion module, a denoising module, an out-of-distribution (OOD) detection module, and a classifier. The diffusion module destroys the input signal into a Gaussian prior distribution, while the denoising module restores the corresponding Gaussian distribution to the original data. The OOD detection module evaluates whether a sample belongs to a known class, and a classifier is used to identify emitters within the known class. Experimental results on the ADS-B dataset demonstrate that the proposed method outperforms the OpenMax algorithm with a 0.09 improvement in macro-F1 score at 13.4% openness. These results show that the proposed method is a promising solution for SEI in the open set scenario.