Abstract:

Remote sensing image fusion is an important processing step of the application of remote sensing images. With the scale of remote sensing image data and complexity of fusion algorithm increasing, the remote sensing image fusion is facing a challenge on the processing speed. In recent years,the power of the computing of GPU has been greatly improved,which results that using it for the generalpurpose computing has a rapid development. In this paper,based on GPU programming mode and its hardware features,the parallel accelerated algorithm of remote sensing image fusion is studied,and a parallel mapping model for the fusion execution stream is proposed.The IHS and waveletbased fusion algorithm with high accuracy and complexity of calculation is selected to design the parallel processing method on GPU,also some optimizations on data transfer,loop unrolling,thread setting,et al are done for the mainstream GPU hardware.Finally, the results of experiment on the GPU of nVIDIAGTX 460are given,which shows that our proposed parallel mapping model and the optimization strategy can be well applied to the field of remote sensing image fusion.In our experiment,the maximum speedup is up to 114X compared with the serial CPU program.This study also shows that the general computing technology of GPU has broad application prospects in the field of remote sensing image processing.

Key words: GPU;remote sensing image fusion;IHS;wavelet;parallel optimization;CUDA