Fast Image Reconstruction in Forward-Looking GPR Using Dual $\ell _1$ Regularization

In the forward-looking ground-penetrating radar (GPR) imaging scenario, detection and classification performance is limited due to inherent constraints placed on the sensor configuration as well as limitations associated with the conventional image formation process. Image reconstruction using sparsity inducing regularization has been demonstrated as a promising technique to overcome these limitations but at a significant computational cost. In this paper, we present a dual ℓ
₁
regularization technique, which is consistent with the expectation that objects are sparsely populated but extended with respect to the pixel grid. Furthermore, we propose a series of preprocessing steps, which produce a greatly reduced data size and computational cost. Demonstrations of the proposed approach are provided using experimental data and results are given in terms of signal to background ratio, image resolution, and relative computation time.