油茶叶片叶绿素含量高光谱估测的偏最小二乘模型参数选择  

Parameter Selection of Partial Least Squares Model for Hyperspectral Estimation of Chlorophyll Content in Camellia oleifera Leaves

作　　者：王芳东[1,2] 严志雁 赵小敏 郭熙[1] 周洋[1] 国佳欣 WANG Fangdong;YAN Zhiyan;ZHAO Xiaomin;GUO Xi;ZHOU Yang;GUO Jiaxin(Key Laboratory of Agricultural Resources and Ecology in Poyang Lake Basin of Jiangxi Province,Jiangxi Agricultural University,Nanchang 330045,China;Base Management Center of Jiangxi Academy of Agricultural Sciences,Nanchang 330200,China;Institute of Agricultural Economics and Information,Jiangxi Academy of Agricultural Sciences,Nanchang 330200,China)

机构地区：[1]江西农业大学江西省鄱阳湖流域农业资源与生态重点实验室,江西南昌330045 [2]江西省农业科学院基地管理中心,江西南昌330200 [3]江西省农业科学院农业经济与信息研究所,江西南昌330200

出　　处：《江西农业大学学报》2022年第1期86-96,共11页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家重点研发计划项目(2020YFD1100603-02,2017YFD0301603);江西省重点研发计划项目(20202BBF62004)。

摘　　要：【目的】利用高光谱技术估测植物叶片叶绿素含量时,如何在高维的光谱数据中选择有效的高光谱参数作为估测模型的输入矢量是估测叶绿素含量精度的关键。【方法】以南方丘陵地区油茶为试验材料,收集了182份油茶叶片光谱反射率及叶绿素含量样本,综合分析了敏感波段、光谱指数、高光谱特征参数和全波段(400~1350 nm)4个不同高光谱参数与叶绿素含量的相关性,筛选出较优高光谱参数分别作为估测模型输入矢量,建立估测油茶叶片叶绿素含量的偏最小二乘回归(partial least square regression,PLSR)模型。【结果】表明:(1)基于全波段建立PLSR模型,其建模集与验证集的R^(2)分别为0.84和0.85,RPD值分别为2.52和2.59,估测效果最优;(2)与全波段参数相比,基于高光谱特征参数建立的PLSR模型预测结果略低,建模与验证R^(2)分别为0.81和0.80,RPD值分别为2.31和2.28,模型输入矢量的数量较全波段模型降低了98.73%,减少了模型运算量,提高了模型运算速度。【结论】利用PLSR模型对油茶叶片叶绿素含量进行估测时,全波段和高光谱特征为估算油茶叶片叶绿素含量的有效高光谱参数。研究结果对快速准确获取油茶叶片叶绿素含量的相关研究有重要的技术指导意义。[Objective]When estimating chlorophyll content of plant leaves by hyperspectral technology,how to select effective hyperspectral parameters from high-dimensional spectral data as the input vector of the model is the key to determine the accuracy of estimating chlorophyll content.[Method]182 samples of leaf spectral reflectance and chlorophyll content of Camellia oleiferas were collected from the hilly regions of south China.The correlation between chlorophyll content and four different hyperspectral parameters was analyzed,which included sensitive band,spectral index,hyperspectral characteristic parameters and full band(400-1350 nm).Selecting optimal hyperspectral parameters as the input variables,a partial least square regression(PLSR)model was established to estimate the chlorophyll content of Camellia oleifera leaves.[Result]The results showed that:(1)The estimation effect of the PLSR model established based on the whole band was optimal,when the R^(2) of the modeling set and verification set were 0.84 and 0.85,and the RPD were 2.52 and 2.59,respectively.(2)Compared with the full-band parameters,the prediction results of PLSR model based on hyperspectral characteristic parameters were slightly lower,the R2 of modeling set and validation set were 0.81 and 0.80 respectively,and the RPD value were 2.31 and 2.28 respectively,however,the number of input vectors of the model established by this method reduced by 98.73%.The reduction of calculation amount improved the speed of model computation.[Conclusion]When PLSR model was used to estimate the chlorophyll content of Camellia oleifera leaves,the full-band and hyperspectral characteristics were effective hyperspectral parameters of estimating the chlorophyll content of Camellia oleifera leaves.The results provide important technical guidance for obtainingthe chlorophyll content in Camellia oleifera leaves in a quick and accurate way.

关 键 词：油茶 高光谱参数 偏最小二乘回归 叶绿素含量 

分 类 号：S794.4[农业科学—林木遗传育种]