A Machine Learning Method for 2-D Scattered Far-Field Prediction Based on Wave Coefficients

Abstract

In this letter, a machine learning method is presented to evaluate the scattering by 2D conducting objects. First, the scattered far field is expressed by angular harmonics with weighted wave coefficients (WCs), which are distinctive to the cross-section of the scatterer. Then, a neural network (NN) is trained to learn the WCs from a range of objects. Finally, the NN is used to extract the WCs for a given object, and the scattered far field or radar cross-section is readily computed by using the WCs. Numerical examples show that the proposed approach can be a viable choice for fast online prediction.