CAJ | 학술논문

叶面积指数(LAI)是作物长势诊断及产量预测的重要参数。通过对冬小麦采样点的高光谱曲线进行连续小波变换(CWT)，然后利用小波系数与LAI建立支持向量机回归(SVR)模型，实现冬小麦不同生育时期的叶面积指数估算。通过对所研究方法与选取的植被指数、偏最小二乘(PLS)回归等5种方法的反演结果进行统计分析。结果表明：利用连续小波变换确定的 LAI的敏感波段为680、739、802、895 nm，对应尺度分别为8、4、9和8，对应小波系数的LAI回归确定系数(R2)明显高于冠层反射率的回归确定系数；利用小波系数与LAI建立的SVR模型的反演精度最高，模型实测值与预测值的检验精度(R2)为0.86，均方根误差(RMSE)为0.43；而常用植被指数(归一化植被指数，NDVI；比值植被指数，RVI)建立的估测模型对冬小麦多个生育时期LAI反演精度最低(R2<0.76，RMSE>0.56)。因此利用连续小波变换进行数据预处理，能更好地筛选出对叶面积指数敏感的信息，LAI回归方法比较结果表明，SVR比PLS更适合于LAI的估测，通过将CWT与SVR结合(CWT-SVR)能实现不同生育时期冬小麦叶面积指数的遥感估算。
협면적지수(LAI)시작물장세진단급산량예측적중요삼수。통과대동소맥채양점적고광보곡선진행련속소파변환(CWT)，연후이용소파계수여LAI건립지지향량궤회귀(SVR)모형，실현동소맥불동생육시기적협면적지수고산。통과대소연구방법여선취적식피지수、편최소이승(PLS)회귀등5충방법적반연결과진행통계분석。결과표명：이용련속소파변환학정적 LAI적민감파단위680、739、802、895 nm，대응척도분별위8、4、9화8，대응소파계수적LAI회귀학정계수(R2)명현고우관층반사솔적회귀학정계수；이용소파계수여LAI건립적SVR모형적반연정도최고，모형실측치여예측치적검험정도(R2)위0.86，균방근오차(RMSE)위0.43；이상용식피지수(귀일화식피지수，NDVI；비치식피지수，RVI)건립적고측모형대동소맥다개생육시기LAI반연정도최저(R2<0.76，RMSE>0.56)。인차이용련속소파변환진행수거예처리，능경호지사선출대협면적지수민감적신식，LAI회귀방법비교결과표명，SVR비PLS경괄합우LAI적고측，통과장CWT여SVR결합(CWT-SVR)능실현불동생육시기동소맥협면적지수적요감고산。
Leaf area index (LAI) is an important parameter of crop diagnosis and yield prediction. The LAI of winter wheat obtained from Beijing city had been estimated successfully by support vector machine regression (SVR) model built with LAI and wavelet coefficients of hyperspectral reflectance. The inversion results of this paper method and other five methods, such as selected vegetation indices and partial least-square (PLS) regression models, were analyzed. It was found that the sensitive bands to assess LAI were 680 nm, 739 nm, 802 nm, and 895 nm, and the corresponding wavelet decomposition scales were 8, 4, 9, and 8 determined by continuous wavelet transform(CWT), respectively. The decision coefficient (R2) of regression equation between LAI and wavelet coefficient was significantly higher than that of between LAI and canopy reflectance. The SVR model based on wavelet coefficients performed best with R2 of 0.86, and RMSE of 0.43, while the regression models based on two common spectral vegetation indices (NDVI and RVI) performed poor in estimating LAI of winter wheat's multiple birth period (R2<0.76, RMSE>0.56). It can conclude that the pretreatment method of CWT is better effective for selecting sensitive spectral characteristics to LAI. Meanwhile, SVR is more suitable for developing model in LAI estimation than PLS regression. The combination of CWT and SVR is feasible to realize remote sensing inversion of LAI in the whole growth period of winter wheat.