GPU-based Parallel Computing for Processing Sentinel-1 SAR Image

Abstract

In this paper, we discuss about Graphics Processing Unit(GPU) based parallel computing for processing high resolution Synthetic Aperture Radar(SAR) image. Especially, we focus on Sentinel-1 dual-polarization image. Since Sentinel-1 SAR Ground Range Detected(GRD) image has about 0.5billion pixels, processing the image is very time consuming. Hence we adopt parallel computing technique to shortening the processing time. To do that, we adopt Compute Unified Device Architecture(CUDA), which is GPU-based parallel computing platform. Compared by CPU, GPU has simple but enormous processing cores, so it is fast when process huge amount of data with same algorithms. Hence in this paper, we introduce CUDA and the structure of CUDA-based image processing. Then we implement CUDA-based Sentinel-1 SAR image land masking algorithm and Pauli RGB color composite operator. Finally, we evaluate how GPU processing is efficient compared by CPU with Sentinel-1 SAR GRD dual-polarization image.nd Range Detected(GRD) image has about 0.5billion pixels, processing the image is very time consuming. Hence we adopt parallel computing technique to shortening the processing time. To do that, we adopt Compute Unified Device Architecture(CUDA), which is GPU-based parallel computing platform. Compared by CPU, GPU has simple but enormous processing cores, so it is fast when process huge amount of data with same algorithms. Hence in this paper, we introduce CUDA and the structure of CUDA-based image processing. Then we implement CUDA-based Sentinel-1 SAR image land masking algorithm and Pauli RGB color composite operator. Finally, we evaluate how GPU processing is efficient compared by CPU with Sentinel-1 SAR GRD dual-polarization image.