Triple-Link Fusion Decision Method for Through-the-Wall Radar Human Motion Recognition

To better solve the accuracy degradation of human motion recognition due to low signal-to-clutter-plus-noise ratio (SCNR) and low resolution of through-the-wall radar (TWR) imaging, a triple-link fusion decision human motion recognition method for through-the-wall radar is proposed in this paper. This method combines the physical information, visual local information and visual global information in imaging. Specifically, the idea of complementarity of three weak models, including empirical modal decomposition (EMD) algorithm based on statistic signal detection, visual gradient-level based kernel method and visual regionalized macro-level based shuffle attention improved residual neural network (SA-Inception-ResNet) algorithm are introduced in the method, and the Dempster-Shafer (D-S) synthesis theory is used to achieve decision level fusion recognition. The final results are inferred by an adaptive boosting method on the trained weak models and the fused strong model. Experiments are carried out to demonstrate that the accuracy of the algorithm exceeds 99.54%, while the prediction performance and robustness are significantly improved compared with previous methods.