Research On Mosaic Technology Remote Sensing Image Based On CPU-GPU

In recent years, as the frequently outbreak of natural disasters, emergencysupport of surveying and mapping work has gradually become an importantresponsibility of the government.With the development of space remote sensingtechnology, the emergency of remote sensing image data increased tremendously. Inthe face of universal image data mosaic and high complexity of computation task,traditional image mosaic technology become powerless. It is the key point to improvethe efficiency of batch image mosaic in emergency.This paper study general calculation of GPU and parallel stitching algorithmbased on CUDA programming, Aiming at the limitation of traditional image mosaicalgorithm efficiency, put forward to accelerate processing by using GPU hardwarestrategy, Finally, analyse the performance and quality of effort. This thesis mainlyincludes the following contents:(1)Firstly, analyzing GPU architecture and CUDA programming technology.Programming technology is the research priorities, optimization strategy of CUDAprogram performance, the realization process of the parallel imaging algorithm basedon CUDA. All of these are the basis technology of the paper.(2) Aiming at the defects of the traditional SIFT feature matching algorithmefficiency underground, This paper proposed the accelerated processing based onGPU hardware strategy. The key steps of the SIFT algorithm analysis, on the basisof CUDA programming strategy proposed the overall design of the SIFT parallelalgorithm and Realization, Finally, the correctness of the test performance analysis ofparallel algorithm and feature matching results.(3)The efficiency of the algorithm and the fusion quality two aspects,comparative analysis of several commonly used image fusion algorithm, tshe parallelimplementation of multiband fusion algorithm based on CUDA. The core module ofmultiband fusion algorithm, o find out the suitable for data parallel computing link,the parallel algorithm design and Implementation Finally, test and analysis theperformance of the parallel algorithm and the quality of fused image precision.(4) Based on the above research content, this paper presents a fast stitchingimage solutions. The design and implementation of emergency rapid splicing remotesensing image system. Mainly includes image feature matching, fast image fusion,image sequence registration error and so on. Finally from the image mosaic algorithm performance quality, image mosaic method compared with the traditional serial mode,comprehensive analysis.Experiment results show that CPU-GPU remote sensing image mosaic systemcould ensure the quality of mosaic image with shorten time obviously and realizereal-time requirements to emergency safeguard.