Multiple Targets Echo Separation on Radar Range–Doppler Maps via Dual Decoupling Perception

Abstract

Under the scenario of multiple persons in the radar field of view, the researchers consider the separation of multiperson echo signals as the primary task for subsequent detection and identification. However, the task is challenged by the complex superimposition of echo signals when multiple persons move simultaneously at similar speeds and distances. In this case, the existing clustering methods lead to incorrect separation which destroys the semantic information of signals. In this article, we address the issue by transforming the task into a generative problem and propose a novel generative adversarial network (GAN)-based method for the multiperson echo signal separation using range–Doppler (r-D) maps. In the method, we introduce the dual decoupling network (DDN) with dual branches to separate the coupling signals of multiple persons. We also adopt a perceptual-based training strategy to maintain the semantic integrity of the generated r-D maps using an additional constraint via a pretrained classifier. Experimental results demonstrate that the proposed method can effectively separate the superimposed signals from multiple persons and outperform the compared separation methods qualitatively and quantitatively.