基于优化植被指数组合的多品种辣椒叶片叶绿素数值估测

Numerical estimation of chlorophyll in pepper leaves based on optimized vegetation index combination

叶片叶绿素与植被生长状况息息相关,SPAD值能够反映作物叶片叶绿素含量,不同品种辣椒外形和生理生态参数具有明显差异,因此,准确、快速地估算SPAD值具有重要意义。以4个不同品种辣椒为研究对象,测量其SPAD值,对原始光谱进行倒数、对数、倒数对数、一阶微分和二阶微分变换,通过将变换光谱替换原始光谱来优化植被指数,对比优化植被指数和经典植被指数搭建模型的差异,最终得到不同品种辣椒SPAD值和高光谱之间的关系,寻找SPAD值的最优反演模型。结果表明:不同品种辣椒冠层光谱特性存在差异;辣椒叶片建模集、验证集和全样本SPAD值的变化趋势均为线椒大于朝天椒;基于倒数对数光谱优化的植被指数除了CI_(red edge)外,其余植被指数的相关系数均高于经典植被指数;基于lg1/R-VI搭建的随机森林模型无论是建模集还是验证集精度均较好,适合于不同品种辣椒SPAD值的估算,其中全样本模型测试集决定系数(R^(2))为0.83,平均绝对误差(MAD)为1.90,验证集R^(2)和MAD分别为0.45和1.26。

Leaf chlorophyll is closely related to vegetation growth,and it is an important parameter that directly affects vegetation growth and health status.SPAD value can reflect the chlorophyll content of crop leaves.The larger the SPAD value,the higher the chlorophyll content,and the smaller the SPAD value,the lower the chlorophyll content.The shape and physiological and ecological parameters of different varieties of pepper are significantly different,so it is of great significance to accurately and quickly estimate the SPAD value.The SPAD values of 4 different varieties of peppers,Qianjiao 8,Hongla 18,Layan 101,and Red Globe,were measured.The reciprocal,logarithm,reciprocal logarithm,first-order differential and second-order differential transformation of the original spectrum were performed.The vegetation index was optimized by replacing the original spectrum with the transformed spectrum,and the difference between the optimized vegetation index and the classical vegetation index was compared.Finally,the relationship between SPAD values and hyperspectral of different varieties of peppers was obtained,and the optimal inversion model of SPAD values was found.The results showed that there were differences in canopy spectral characteristics of different pepper varieties.SPAD values of modeling set,validation set and full sample of pepper leaves showed that the change trend of line pepper was greater than that of Chao tian pepper.The correlation coefficients of vegetation indices based on reciprocal logarithmic spectral optimization were higher than classical vegetation indices except CI_(red edge).The random forest model based on lg1/R-VI had good accuracy in both the modeling set and the validation set,which was suitable for the estimation of SPAD values of different varieties of pepper leaves.The coefficient of determination(R^(2))of full sample model test set was 0.83,mean absolute deviation(MAD)was 1.90,and R^(2)of validation set was 0.45,MAD was 1.26.