Prestack Least-Squares Reverse Time Migration With an Exact Adjoint Operator for Ground-Penetrating Radar

This study develops prestack least-squares reverse time migration (LSRTM) for high-resolution ground-penetrating radar (GPR) imaging and compares its performance to standard RTM. We use the Born approximation to derive the LSRTM forward operator and we use matrix operations to determine its exact adjoint. Tests on synthetic data confirm that LSRTM outperforms both prestack and poststack RTM in terms of resolution and thus significantly improves the visibility of fine-scale structures. Furthermore, the exact adjoint operator is shown to facilitate faster convergence and lead to a lower data misfit. Although LSRTM entails higher computational costs than standard RTM, we demonstrate that accelerated computing with graphic processing unit (GPU) devices can make it remarkably affordable for 2-D GPR imaging, thereby opening avenues for future applications in 3-D.