基于GPU和分块技术的巨幅影像快速傅里叶变换算法研究.pdf

《基于GPU和分块技术的巨幅影像快速傅里叶变换算法研究.pdf》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。

1、第３４卷，第２期光谱学与光谱分析Vol畅３４，No畅２，pp４９８‐５０４２０１４年２月SpectroscopyandSpectralAnalysisFebruary，２０１４ResearchonFastFourierTransformsAlgorithmofHugeRemoteSensingImageTechnologywithGPUandPartitioningTechnology１，３，４２，３倡１倡４YANGXue，LIXue‐you，LIJia‐guo，MAJun，２１２ZHANGLi，YANGJan

2、，DUQuan‐ye１．InstituteofRemoteSensingandDigitalEarth，ChineseAcademyofSciences，Beijing　１００１０１，China２．ChineseAcademyofSurveyingandMapping，Beijing　１００８３０，China３．BeijingSiweiSpatialDataTechnologyCo．，Ltd．，Beijing　１０００３９，China４．SchoolofComputerandInformationEnginee

3、ring，HenanUniversity，Kaifeng　４７５００４，ChinaAbstractFastFouriertransforms（FFT）isabasicapproachtoremotesensingimageprocessing．Withtheimprovementofcapacityofremotesensingimagecapturewiththefeaturesofhyperspectrum，highspatialres‐olutionandhightemporalresolution，ho

4、wtouseFFTtechnologytoefficientlyprocesshugeremotesensingimagebecomesthecriticalstepandresearchhotspotofcurrentimageprocessingtechnology．FFTalgorithm，oneofthebasicalgorithmsofimageprocessing，canbeusedforstripenoiseremoval，imagecompression，im‐ageregistration，e

5、tc．inprocessingremotesensingimage．CUFFTfunctionlibraryistheFFTalgorithmli‐brarybasedonCPUandFFTW．FFTWisaFFTalgorithmdevelopedbasedonCPUinPCplatform，andiscurrentlythefastestCPUbasedFFTalgorithmfunctionlibrary．Howeverthereisacommonproblemthatoncetheavailableme

6、moryormemoryislessthanthecapacityofimage，therewillbeoutofmemoryormemo‐ryoverflowwhenusingtheabovetwomethodstorealizeimageFFTarithmetic．Toaddressthisproblem，aCPUandpartitioningtechnologybasedHugeRemoteFastFourierTransform（HRFFT）algorithmisproposedinthispaper．

7、ByimprovingtheFFTalgorithminCUFFTfunctionlibrary，theproblemofoutofmemoryandmemoryoverflowissolved．Moreover，thismethodisprovedrationalbyexperimentcombinedwiththeCCDimageofHJ‐１Asatellite．Whenappliedtopracticalimageprocessing，itimproveseffectoftheimageprocess‐i

8、ng，speedsuptheprocessing，whichsavesthetimeofcomputationandachievessoundresult．KeywordsGPU；HRFFT；FFT；Remotesensingimage中图分类号：TP７５１　　文献标识码：ADOI：１０畅３９６４／j畅issn畅１０００‐０５９３（２０１４）０２‐０４９８‐０７datacomputat