A Primitive Scatterer Dictionary for Semantic Representation of Radar Target Images

High-resolution radar images convey rich information about targets and their geometric characteristics. It remains a challenge to extract the semantic information of target geometry from the speckle-like radar images. We argue that semantic-centric scattering modeling is crucial in this regard. This article proposes to establish a primitive scatterer dictionary (PSD) following the criteria of atomicity, completeness, extensibility, and pixel-wise independence. By sorting out the geometric structure of the smallest atomic geometric unit in common high-frequency approximation methods, the proposed scatterer dictionary consists of primitive 0-D point, 1-D line, and 2-D surface scatterers, being either straight or curved. A unified bidirectional scattering distribution function (BSDF) is derived to analytically model the multidimensional scattering dependences on target attributes, position parameters, and sensor observation configurations. It is further extended to model most typical double and triple scattering structures. The efficacy and applicability of PSD are demonstrated by comparing it to numerical methods in various cases including the commonly seen attributed scattering center (ASC) models.