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

It is known that nonlinear inversion algorithms have good accuracy in inverting electromagnetic data. However, their applications are still limited by their computational complexity. In this paper, we study an acceleration scheme for the two-dimensional multiplicative regularized contrast source inversion (MR-CSI) algorithm using compute unified device architecture (CUDA) and NVIDIA general purpose graphic process unit (GPGPU). We benchmark this algorithm using both synthetic and experimental data. Numerical results show that the optimized algorithm on paralleled computing device outperforms the one on traditional CPU processors with a significant reduction on CPU time while keeping the same level of reconstruction quality. By using this algorithm, we successfully invert the microwave experimental data (2D Fresnel dataset) in only a few minutes.