Deep Learning-Based Plasma Parameter Diagnostics Using Microwave Reflection

This paper presents a plasma parameter diagnostic method integrating microwave reflection and deep learning techniques. First, the Finite Element-Boundary IntegralMultilevel Fast Multipole Algorithm (FE-BI-MLFMA) is employed to generate high-accuracy training datasets. Then, a Plasma-Electromagnetic wave coupling network (PENet) is proposed to reconstruct plasma electron density efficiently by converting antenna reflection coefficients into image-based input features. This approach enhances the robustness of feature extraction and significantly improves inversion accuracy. Compared with traditional microwave diagnostic methods, the proposed method overcomes limitations in diagnosing electron density in complex plasma environments, substantially enhancing the understanding of electromagnetic phenomena around hypersonic vehicles and effectively supporting reliable operation of critical communication and detection systems.