Abstract: The processing efficiency of optical satellite remote sensing image orthographic fusion method based on coprocessing of CPU and GPU in domestic cloud platform is discussed systematically and is improved by data flow configuration and the intermediate data storage access optimization. The Phytium S2500 and NVIDIA A100 are used in the cloud platform to do the orthographic fusion experiment. The experiment results show that the method can greatly improve the fusion efficiency of optical satellite remote sensing image, and the acceleration ratio is more than 14.3 times of the traditional X86 architecture CPU and GPU collaborative orthographic fusion algorithm., and the corresponding processing time is reduced to less than 8.4 s, and the GPU operation time is only 1 s, which can meet the requirements of rapid orthographic correction of the large data of optical satellite remote sensing image.