高光谱数据预处理对大豆叶绿素密度反演的作用  

Effects of Hyperspectral Data Pretreatment on Model Inversion of Soybean Chlorophyll Density

作　　者：杨峰[1,2,3] 张勇[1,2,3] 谌俊旭 范元芳 杨文钰[1,2,3] YANG Feng ZHANG Yong CHEN Junxu FAN Yuanfang YANG Wenyu(College of Agronomy, Sichuan Agricultural University, Chengdu 611130, China Key Laboratory of Crop Ecophysiology and Farming System in Southwest, Ministry of Agriculture, Chengdu 611130, China Sichuan Engineering Research Center for Crop Strip Intercropping System, Chengdu 611130, China)

机构地区：[1]四川农业大学农学院,成都611130 [2]农业部西南作物生理生态与耕作重点实验室,成都611130 [3]四川省作物带状复合种植工程技术研究中心,成都611130

出　　处：《遥感信息》2017年第4期64-69,共6页Remote Sensing Information

基　　金：国家重点研发计划项目(2016YFD030060203);国家自然科学基金(31571615)

摘　　要：针对高光谱数据异常值影响叶绿素密度反演精度的问题,以大豆叶片为研究材料,利用马氏距离和蒙特卡洛交叉验证法(Monte Carlo cross validation,MCCV)剔除异常样本,探讨13种高光谱数据预处理方法对叶绿素密度偏最小二乘法(partial least square,PLS)建模的影响。结果表明,马氏距离法和MCCV剔除异常样本能提高校正模型的精度,在权重系数为1时剔除异常样本数3个,模型精度最高,校正集决定系数和均方根误差分别为0.821和0.112。微分处理能大幅度提高模型的预测精度,二阶微分处理效果最好,校正集决定系数和均方根误差分别为0.998和0.012,验证集决定系数和均方根误差分别为0.961和0.139,具有比原始光谱更高的精度。因此,适宜的高光谱数据预处理可有效提高大豆叶绿素密度估测精度。The outlier samples directly affect the precision of estimating chlorophyll density. This work aimed to discuss the effect of 13 treatment methods of hyperspectral data on the partial least square （PLS）model of assessing soybean chlorophyll density after removing the outlier spectral samples using Mahalanobis distance and the Monte Carlo cross validation （MCCV） methods. The results showed that the Mahalanobis distance method and MCCV eliminating abnormal samples could improve the accuracy of calibration model. The precision of PLS model was higher after extracting three samplings, and the coefficient of determination and root mean square error was 0. 821 and 0. 112, respectively. Compared with other methods, differential treatment can greatly improve the prediction accuracy of PLS model in soybean chlorophyll density. Especially for the second derivative treatment, the calibration coefficient of determination and root mean square error was 0. 998 and 0. 012, respectively. The validation set the coefficient of determination and root mean square error was 0. 961 and 0. 139 ,respectively. Therefore,appropriate treatment methods of hyperspeetral data can improve the precision of the precision of estimating chlorophyll density in soybean.

关 键 词：高光谱 预处理 大豆 叶绿素密度 偏最小二乘法 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]