Predicting RCS of 2D Objects with Arbitrary Shapes using Gaussian Process Regression

This paper focuses on the RCS prediction of a 2dimensional object with Gaussian Process Regression (GPR) trained based on a set of shapes of 2D objects and their relevant RCS data. RCS (or shape) of the object is predicted for a given shape (or RCS) of the object. GPR is one of the commonly known machine learning (ML) algorithms that becomes prominent among other algorithms due to its capability to insert physical prior knowledge into the algorithm and construct a 2 -sided prediction mechanism (i.e. shape <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\leftrightarrow$</tex> RCS). The results show that for a smooth enough object the RCS can be predicted with a small error while a large error occurs at the side-lobs of the RCS data when the object has sharp changes in shape (e.g. a star-like contour). The works in the literature show that the error at the side-lobs of the RCS can be improved by using neural network algorithms. The authors are working in this direction to improve the error in RCS and shape prediction of the objects for a given RCS data.