CS-BASED HRRP EXTRACTION METHOD FOR THROUGH-WALL DETECTION

Feature extraction is of significant importance for final results of the through-wall detection procedure. High resolution range profile (HRRP) is related to target reflectivity coefficients which can be used as a new feature for object detection. Compressive sensing (CS) is an emerging technique which enables a sparse signal to be recovered using much fewer measurements. This method can provide a novel way for achieving the HRRP since the target reflectivity coefficients are often known to be sparsely distributed in range cells. In this paper, after a set of input-output patterns that consist of target position and HRRP are obtained, through-wall detection problem is reformulated into a nonlinear regression one, which can be solved by support vector machine (SVM). Numerical simulations demonstrate that the prediction accuracy of target position is related to the number of range cells, the number of observations, and signal-to-noise ratio (SNR). Furthermore, the proposed method performs better than the one using signal amplitude as a feature in terms of smaller estimation error and shows better robustness against noise.