End-to-End Differentiable RCS Optimization on 3D Geometry Based on Physical Optics Method

The optimization of radar cross section (RCS) is now a significant issue in the designation of military and civilian equipment. Comparing with the expensive material approaches, changing the geometry of an object is a relatively flexible and low-cost way. However, the RCS optimization of large-scale models often faces two major problems: too large optimization space and slow RCS calculation, which caused by increasing geometry parameters and iterative numerical computation, respectively. In addition, secondary problems such as geometric information loss and RCS results lacking of gaurantees always remain even if dimensionality reduction has been carried out for alleviating these two problems. In this paper, we propose a novel end-to-end differentiable RCS optimization framework based on the physical optics (PO) method. The proposed framework utilize the differentiability of the PO method, and realize an efficient and interpretable RCS optimization without dimension reduction. The innovation of this paper lies in the combination of PO method and gradient-based optimization to achieve RCS optimization of large-scale complex 3D geometries. Experiments show that in ordinary 2D scenarios, our method achieves at least 16 times higher efficiency than the mainstream optimization method. Meanwhile, the optimization error of RCS has been reduced by 75.29% compared to traditional methods (0.0515 vs. 0.2084). We further validate the performance of the framework on more complex tasks such as 3D plane model and analyzed the effectiveness of the overall framework. The proposed optimization method is expected to be widely used in applications such as stealth and aircraft designs.