基于多维度光谱特征波长提取大豆冠层SPAD值估算模型

Estimation model of SPAD value for soybean canopy based on multi-dimensional spectral characteristic wavelength extraction

应用多种光谱特征波长段提取技术,研究了全生育期大豆冠层的近红外光谱曲线特性,提出了基于多维度光谱特征波长提取的大豆冠层叶绿素相对含量(SPAD值)估算模型.在大豆冠层原始近红外光谱曲线中,优选多元散射校正预处理和偏最小二乘回归法建模.经多元散射校正预处理后,应用竞争性自适应重加权采样算法(CARS)、主成分分析法(PCA)和连续投影算法(SPA),提取大豆冠层的光谱特征波长数量分别为22,51和12个.以此特征波长为基础,利用偏最小二乘回归法和多元线性回归法(MLR),分别建立大豆冠层SPAD值估算模型.结果表明CARS-MLR模型试验效果较好,该模型校正样本和预测样本SPAD值的均方根误差分别为5.67和5.94,平均值约为5.81.

The extraction techniques of multiple spectral characteristic wavelength band were applied,and the near-infrared spectral characteristics of soybean canopy during the whole growth period were investigated.An estimation model of SPAD value for soybean canopy was proposed based on the multi-dimensional spectral characteristic wavelength extraction.For the original near-infrared spectral curve of soybean canopy,the multiple scattering correction preprocessing and the partial least squares regression modeling were preferred.Preprocessed with multiple scattering corrections,the competitive adaptive reweighted sampling algorithm(CARS),the principal component analysis and the successive projections algorithm were applied to extract the spectral characteristic wavelengths of soybean canopy,which were respective 22,51 and 12.Based on the characteristic wavelengths,the partial least squares regression and the multiple linear regression(MLR)methods were used to establish the estimation models of soybean canopy SPAD values.The results show that the CARS-MLR model has good experimental effect,and the root mean square errors of CARS-MLR model for corrected and predicted samples are respective 5.67 and 5.94 with the average error of about 5.81.