Abstract

Since the introduction of CUDA (Compute Unified Device Architecture), GPU (Graphics Processing Unit) was used in various fields rapidly. Some researchers used the GPU computing technology in remote sensing image processing, and revealed that one hundred times speedup could be obtained. Current GPU-based approaches need to load all the image data at a time prior to image processing. However, the current computer memory and GPU memory are limited, and are not big enough for loading the remote sensing image data which are always massive. Hence, current GPU-based image processing approaches cannot be directly applied in remote sensing image processing. Under this situation, this paper proposes a dual-parallel processing mechanism, which is based on GPU and POSIX thread technologies, in massive remote sensing image data processing. Experimental results illustrate that our methodology can not only deal with massive remote sensing image data, but also improve the processing efficiency greatly.

Paper available at IEEE.