高光谱图像组合光谱特征研究.docx

高光谱图像组合光谱特征研究摘要：在此分析了典型植被、岩矿、土壤和水体地物的光谱特性曲线及其一阶和二阶导数的特点与异同。研究表明通过光谱导数技术可以快速分析出高光谱特性曲线的反射峰和吸收谷谱带的相关特征且不同类别地物目标在原始光谱曲线和各阶光谱导数曲线上存在不同敏感波段。因此提出将高光谱原始数据和其各阶导数曲线的敏感波段综合起来构成组合光谱特征以此达到优化高光谱特征性和降低分类难度从而进一步提高分类性能的方法。实验结果表明组合光谱特征可以实现高光谱图像有效分类。关键词：高光谱分类；组合光谱特征；光谱导数特征；导数光谱技术中图分类号：TN919?34；TP751.1文献标识码：A文章编号：1004?373X（2013）14?0104?04ResearchoncombinationalspectralfeaturesofhyperspectralimagesCHANGWen?juanHEMing?yi（EarthObservationResearchCenterShaanxiKeyLabofInformationAcquisitionandProcessingNorthwesternPolytechnicalUniversityXi’an710129China）Abstract：Thespectralcharacteristiccurvesoftypicalvegetationmineralssoilandwaterbodyandsimilaritiesanddifferencesoftheirfirst?orderandsecond?orderderivativesareanalyzed.Theresearchdemonstratesthattherelevantcharacteristicsofthereflectionpeakandtheabsorptionvalleybandsofhyperspectralspectralcharacteristiccurvescanbequicklyobtainedbyanalysisandspectrumderivativetechnologyandtherearedifferentsensitivebandsbetweenoriginalspectralcurveandvariousorderspectralderivativecurvesofdifferentcategoriesobjectives.Thereforeitisproposedthatthecompositionofthesesensitivebandsofhyperspectraloriginaldataandeachorderderivativecurvecanconstitutecombinationalspectralfeaturestooptimizetheperformanceofhyperspectralfeaturesandreducetheclassificationdifficulty.Theexperimentalresultsshowthatthiscombinationalspectralfeaturescanrealizetheeffectiveclassificationofhyperspectralimages.Keywords：hyperspectralclassification；combinationalspectralfeature；spectrumderivativefeature；derivativespectrumtechnology0引言近年来高光谱遥感数据处理受到特别关注IEEE高光谱遥感图像与信号处理?遥感进化（IEEEWorkshoponHyperspectralImageandSignalProcessing?EvolutionalRemoteSensingWHISPERS1）2009在法国召开首届大会第4届大会于2012年6月在上海召开IEEE遥感学报、传感器学报及图像处理学报等近来先后出版了有关高光谱遥感数据处理的多个专辑[1?4]。高光谱遥感影像具有光