Acceleration of multiplicative regularized contrast source inversion algorithm using paralleled computing architecture

Nonlinear inversion algorithms of electromagnetic data have been widely used in many areas to reconstruct the electrical properties of unknown domains. However, their applications are still restricted by the computational cost when processing large datasets. Many efforts have been made on developing fast nonlinear inversion algorithms, but few of them was focused on the characteristics of the hardware platform. In this paper, we study an acceleration scheme for both 2D and 3D multi-frequency multiplicative regularized contrast source inversion (MR-CSI) algorithm using compute unified device architecture (CUDA) and graphic process unit (GPU). Reconstruction from experimental data shows that the optimized algorithm on paralleled computing device outperforms the one on CPU processors with a significant reduction on computing time (15 times for 2D, 75 times for 3D reconstruction) while keeping the same level of reconstruction quality. This study may provide us another possible path for real-time imaging using nonlinear inversion algorithms.