基于SVR算法的苹果叶片叶绿素含量高光谱反演

为实现苹果叶片叶绿素含量的高光谱反演,分析了多种光谱参数与实测SPAD值的相关性,并将归一化光谱参数值及SPAD值进行多项式回归及支持向量回归。其中以归一化植被指数为变量的SVR(Support vector regression)反演模型在建模及模型检验中决定系数分别为0.741 0、0.891 4,均方根误差分别为0.133 2、0.125 6,具有较高的精度及良好的预测能力。与多项式回归相比,SVR具有更好的反演效果,可以作为叶绿素高光谱反演的优选算法。

Chlorophyll Content Inversion with Hyperspectral Technology for Apple Leaves Based on Support Vector Regression Algorithm

To realize the chlorophyll content inversion with hyperspectral technology for apple leaves, the spectral data and SPAD values were obtained by SVC HR-1024i full band spectrometer and SPAD-502 portable chlorophyll analyzer, respectively. The correlation of original spectral data, the first derivative spectral data and measured SPAD values were analyzed, various spectral parameters were selected based on the sensitive wave bands and models between spectral parameters and measured SPAD values were established. The original spectra and the SPAD value were significantly negatively correlated in visible bands, and significantly positively correlated in NIR bands. The first order derivations of spectra and the SPAD value were negatively correlated in blue and green light bands, and positively correlated in yellow and red light bands. The SPAD inversion model based on NDVI and R565 fitted well. The first derivative spectral data and measured SPAD values had improved correlation coefficient compared with the original spectral data, and the determination coefficient R in the inspection process of models establishment based on normalized difference vegetation index (NDVI) and R565 were 0.8896 and 0.8524, which showed better prediction ability than other models. To avoid the difference of order of magnitude, the spectral parameters and measured SPAD values were normalized, and polynomial regression and support vector regression (SVR) were carried out by using the normalized spectral parameters and SPAD values. The R in the modeling process and inspection process of SVR inversion model were 0.7410 and 0.8914 with root mean square error of 0.1332 and 0.1256, respectively, which indicated that the SVR inversion had high precision and good prediction ability. Compared with polynomial regression, the SVR algorithm had better inversion effect, thus it can be used as an optimization algorithm for chlorophyll content inversion.