39 7 20147 GeomaticsandInformationScienceofWuhanUniversity Vol.39No.7 July2014 DOI:10.13203/j.whugis20130202 :1671-8860(2014)07-0862-05 1 2 3 1,,430079 2,,518060 3,,430079 : 基于江苏省宜兴市 100 个土样的可见光 - 近红外高光谱反射率 (400~2450nm) 数据, 结合包络线去 除 (continuumremoval,cr) 与偏最小二乘回归 (partialleastsquaresregression,plsr), 构建了土壤重金属砷 (As) 和有机质 (OM) 含量的反演模型结果表明, 相比普通 PLSR 模型 ( 模型决定系数 R 2 和预测根均方误差 RMSEP 分别为 0.512,3.090 和 0.621,5.934),CR-PLSR 构建的模型预测能力有明显的改善 (R 2 和 RMSEP 分别为 0.763,2.323 和 0.911,4.599) CR 有效增强了 550 900 1420 1900 和 2200nm 等波段处的反射光谱特征, 根据模型回归系数分析,CR 有效突出的波段正是 As 和 OM 的 CR-PLSR 模型所共用的重要波段 研究表明,CR 能够协助 PLSR 模型重要波段的选择, 利用遥感技术结合 CR-PLSR 能够有效提高土壤重金属 As 和 OM 含量的反演精度, 从而为土壤质量的遥感监测提供参考 : 高光谱 ; 土壤质量 ; 包络线去除 ; 偏最小二乘回归 :P237 :A (OM),, (As) 180 [9], [1] OM, [10],-,CR PLSR AsOM, [2-7], CR PLSR (CR)(CR-PLSR) AsOM,, As PLSR [3,7] OM ; CR PLSR,, CR PLSR (PCR) (PLSR)BP [2,7-8], 1 PCR, CR-PCR OM [7],, 1.1,,,, 30 (30 m 30 m),,, 3~4 (0~10cm), 100, [7] PLSR,, OM [2-3 ],As ClarkRoush, :2013-06-03 : (41171290) :,, E-mail:pxiaoting91@gmail.com
39 7 : 863 1.2 ASDFieldSpec?3 2) (350~2500 : nm) 50 W, 25, 15, (root mean square error of 40cm 15cm 10,cross-validation,RMSECV) (R 2 cv) ; (350~399nm 2451~2500nm), (rootmeansquareerrorofprediction,rmsep) Savitzky-Golay 1.3,RMSECV RMSEP, PLSR,, [11] 100 2 4 4, 92 CR AsOM 2.1, 23, 69 92 CR-PLSR (1), (400~780 PLSR AsOM, nm), ; (800~2450nm), 1400 1) PLSR CR CR 19002200nm, (400~2450nm),, (latentvariables, LVs), PLSR (coeficientsofdetermination,r 2 ) (R 2 pre) R 2 R 2 cvr 2 pre, OH - [12] 1 Fig.1 OriginalSoilReflectanceSpectrum 2.2 2100~2300nm CR,, 0~1 (2(b)) CR ( 1420 1900 2200 nm 550 900 1420 1900 2200 [7] As OM, ),CR AsOM 450 1000 1400 1900 2050 2200 2250 2400 2400nm 2470nm CR 550nm OM -, CR As 600~930nm OM,OM CR 2.3,CR CR-PLSR PLSR As
864 20147 OM, 1 CR-PLSR RMSE PLSR,R 2 PLSR, OM R 2 0.621 0.911,RMSEP 22.5%, R 2 cvr 2 pre 0.581 0.683, nm),cr 6 PLSR, AsOM 3,AsOM CR-PLSR PLSR (regressioncoeficient, RC), PLSR, CR-PLSR RC, (correlationcoeficient,cc)(3(a) -20~20-110~110( PLSR RC -10~5-10~10 ) RC, -0.6074~0.1023),CR ( RC As,OM CR- -0.5154~0.4564-0.5085~0.6650) PLSR RC PLSR 10, OM RC, OM, CR-PLSR RC,As CR,CR CR OM PLSR, As, CR-PLSR PLSR RC (550 700 900 1420 1900 2200nm), AsOM, [5,7] AsOM 2.2 CR (550 900 1420 1900 2200 2400 AsOM, RC, CR-PLSR RC, AsOM CR 3(f)) CR AsOM ( -0.4499~0.343 PLSR RC 2 Fig.2 CurveofAverageReflectanceSpectrum Tab.1 1 AsOM CR-PLSR PLSR ModelingandPredictionResultsofCR-PLSRandPLSR ModelsforAsandOM CR-PLSR PLSR R 2 LVs RMSECV RMSEP R 2 cv R 2 pre R 2 LVs RMSECV RMSEP R 2 cv R 2 pre As OM 3 0.763 0.911 8 11 1.736 5.943 2.323 4.599 0.531 0.581 0.522 0.683 0.512 0.621 7 6 1.760 6.005 3.090 5.934 0.500 0.564,CR-PLSR AsOM CR-PLSR 0.240 0.474 AsOM CR PLSR, OM RC,CR
39 7 : 865 3 AsOM CR-PLSR PLSR RC CR Fig.3 RegressionCoeficientsofCR-PLSRandPLSR ModelsofAsandOM ContentandPearson s CorrelationCoeficientwithSpectrum BeforeCRandAfterCR :550 700 900 1420 1900 2200 rounda Mining AreabyReflectanceSpectroscopy: nm, A CaseStudy[J].Pedosphere,2009,19(6):719- (OM), 726 [5] GomezC,Lagacherie P,Coulouma G.Continuum RemovalVersusPLSR MethodforClayandCalci- CR-PLSR um CarbonateContentEstimationfrom Laboratory and Airborne Hyperspectral Measurements [J]. Geoderma,2008,148(2):141-148 [6] XieBocheng,Xue Xuzhang,Liu Weidong,etal. Hul-curve-methodBasedExtractionandAnalysisof [1] LiFuyan,LiXuming,WuPengfei,etal.Correla- SoilSpectralCharacteristics[J].Acta Pedologica tionbetween Heavy MetalPolutionandBasicProp- Sinica,2005,42(1):171-175 (,, ertiesofagriculturalsoilsin HainanProvince[J].,. Soils,2009,41(1):49-53(,, [J].,2005,42(1):171-175),. [7] Ren Hongyan,ZhuangDafang,QiuDongsheng,et ph [J].,2009,41(1):49-53) al.analysisofvisibleandnear-infraredspectraof [2] ZhuDengsheng,WuDi,Song Haiyan,etal.De- As-Contaminated Soilin Cropl-ands Beside Mines termination of Organic Mater Contents and ph ValuesofSoil Using NearInfrared Spectroscopy [J].TransactionsoftheChineseSocietyof Agri- dictionofasinsoilwithreflectancespectroscopy [J].Spectroscopy and Spectral Analysis,2011,31 (1):173-176(,,. [J].,2011, [9] ClarkR N,RoushT L.ReflectanceSpectroscopy: 31(1):173-176) [4] Ren H Y,ZhuangDF,SinghA N,etal.Estimation ofasandcucontaminationinagriculturalsoilsa- [J].Spectroscopy and Spectral Analysis,2009, 29(1):114-118(,,,. - 28(5):1160-1164(,,. BP [J].,2008,28(5):1160-1164) culturalengineering,2008,24(6):196-199( [J].,2009,29(1):114-118),,,. [8] ZhengLihua,LiMinzan,PanLuan,etal.Estima- ph [J].,2008,24 tionofsoilorganic MaterandSoilTotalNitrogen (6):196-199) BasedonNIRSpectroscopyandBP NeuralNetwork [3] ZhengGuanghui,ZhouShenglu,WuShaohua.Pre- [J].Spectroscopy and Spectral Analysis,2008, QuantitativeAnalysisTechniquesforRemoteSens- ingapplications[j].j GeophysRes,1984,89(B7): 6329-6340
866 20147 cationpress,2006:145-148 (,,. fenceindustrypress,2006:119-125 (. [M]. :,2006:119-125) [10]Tong Qingxi,ZhangBing,ZhengFenlan.Hyper- spectralremotesensing[m].beijing:higheredu- [M]. :,2006: [12] Xu Binbin.The Reflection Spectrum Researchof 145-148) SoilProfile[J].Soils,2000,32(6):281-287 ( [11] Wang Huiwen.Partial Least Squares Regression. [J].,2000,32 MethodandApplication[M].Beijing:NationalDe- (6):281-287) InversionofSoilParametersfrom HyperspectraBasedonContinuum RemovalandPartialLeastSquaresRegression PENG Xiaoting 1 GAO Wenxiu 2 WANG Junjie 3 1 StateKeyLaboratoryofInformationEngineeringinSurveying,MappingandRemoteSensing, WuhanUniversity,Wuhan430079,China 2 SchoolofArchitectureandUrbanPlanning,ShenzhenUniversity,Shenzhen518060,China 3 SchoolofResourceandEnvironmentalScience,WuhanUniversity,Wuhan430079,China Abstract:Inthestudy,wecombinedcontinuumremoval(CR)withapartialleastsquaresregression (PLSR)methodtobuildinversionmodelsofsoilAsconcentrationandorganicmater(OM)content basedonhyperspectraldatabetween400-2500nmof100soilsampleswhichwerecolectedinyixing regionofjiangsuprovince.theresultsshowthat,comparedwiththecommonplsr models(model's determinationcoeficients(r 2 )androotmeansquareerrorofprediction(rmsep)are0.512,3.090and 0.621,5.934,respectively),CR-PLSR modelspresentanimprovementofasandominversionmod- els(r 2 andrmsepare0.763,2.323and0.911,4.599,respectively).inaddition,crefectively strengthensthespectralcharacteristicsof550 900 1420 1900and2200nm wavelengths.theanaly- sisofregressioncoeficientsofcr-plsrandplsr modelsofasandom demonstrateseveralimpor- tantwavelengthsenhancedbycrandaresimultaneouslyusedforcr-plsr modelsofasandom. ThusCRisabletohelpPLSRconductwavelengthselectionandenhancestheinversionprecisionof soilasandomcontentefectively.thisfindingcanbeusedasareferenceforremotesensingmonito- ringofsoilquality. Keywords:hyperspectra;soilquality;continuum removal(cr);partialleastsquaresregression (PLSR) Firstauthor:PENG Xiaoting,postgraduate,specializesininversionresearchofsoilandvegetationparametersbasedonremotesensing technology.e-mail:pxiaoting91@gmail.com Foundationsupport:TheNationalNaturalScienceFoundationofChina,No.41171290.