Attributed Scattering Center Extraction With Genetic Algorithm
Maoqiang Jing, Guo Zhang
Abstract
An attributed scattering center (ASC) model provides concise and physically relevant features for complex targets and plays an important role in inverse scattering problems. Extracting ASC features from a radar image is designed to solve a highly nonlinear and nonconvex optimization problem, which requires an accurate yet complex parameter initialization process and generates intensive computation and storage cost. This article presents an ASC feature extraction method with a genetic algorithm (GA) (AFEM-GA). The proposed AFEM-GA utilizes an improved real-coded GA to search for the global optimum for each ASC feature in the radar frequency-aspect angle domain in parallel, without complex parameter initialization and with low computation and storage costs. Moreover, an improved relaxation algorithm is also proposed for multiple ASC feature extractions to avoid error accumulation. Experiments based on both numerical computed data and real data verify the accuracy and robustness of the proposed AFEM-GA and show its potential in radar image interpretation.