Real-time True-color Synthesis of Remote Sensing Image with CUDA

This paper presents a CUDA-based real-time solution for true-color synthesis of remote sensing image. The solution reduces total execution time in three aspects. 1. Apply Pinned Memory to reduce data transfer time between host memory and GPU global memory. 2. Use look-up table to reduce computing time. 3 Overlap kernel executions and data transfers, besides register pages and execute kernels in parallel. The experimental results showed that our solution cost 345.22ms in average to composite a true-color image of a single WRS-2 located region on the GPU gtx970m. The speedup ratio of 3.96 proved our solution was useful and efficient, which satisfied the need of real-time synthesis on both PC and laptop.