基于高光谱数据的荔枝SPAD值快速估测

Estimation Model for Litchi Leaf SPAD-Reading Based on Hyper-Spectral Data

通过分析荔枝叶片的反射光谱曲线与荔枝秋稍老熟阶段叶片SPAD-502叶绿素计读数（SPAD值）之间的关系,探索高光谱数据进行荔枝叶片SPAD值估算的能力。研究4类光谱变量,第1类变量是其他文献中提出的光谱指数,第2类变量是基于波段迭代算法进行波段优化后的新指数,第3类变量是蓝边、黄边和红边（＂三边＂）的面积、位置、斜率,第4类变量是二阶导数光谱波峰、波谷区间区间内光谱曲线的面积、最值、最值处波长。结果表明：一阶导数的蓝边斜率和光谱二阶导数提取的490~520 nm范围内光谱数值之和与SPAD值的相关程度也很高（r分别为-0.834和-0.856）;对于3种光谱数据处理形式SNV光谱（Standard Normal Variate,SNV）、FD光谱（First Derivative,FD）和SD光谱（Second Derivative,SD）,波段迭代算法优化后的最佳四波段双差值指数与SPAD值相关性最好,其中（FD_（516）-FD_（474））-（FD_（684）-FD_（660））与SPAD值的相关系数最高达到0.875,预测模型的确定系数达到0.747,RMSECV为2.375。利用高光谱数据监测荔枝叶片SPAD值是可行的。

The aim of this study was to analyze the relationship between spectral characteristics of litchi leaf spectra and SPAD reading and to explore the capacity of hyper-spectral data in litchi leaf SPAD value estimation. The relationship between four types of spectral variables and the SPAD values were investigated. The first type was the spectral indices proposed in literature for estimating chlorophyll content. The second was the optimized spectral indices derived by band iteration. The third was the area, location and the slope derived by the red-edge, blue-edge and yellow-edge of the first derivative spectra. The last was the spectral feature variables constructed by the peaks and valleys of the second derivative spectra. The results showed that the slope of blue-edge and the sum of the second derivative spectra in 490-520 nm presented high correlation with SPAD values（the correlation coefficients were-0.834 and-0.856, respectively）. For the standard normal variate spectra（SNV）, the first derivative spectra（FD） and the second derivative spectra（SD）, the four band double difference indices presented the best relationship with SPAD values. The correlation coefficient between（FD516-FD474）-（FD684-FD660） and SPAD values was the highest, which was 0.875. The determination coefficient and RMSE of the prediction model developed by this variable and SPAD values were 0.747 and 2.375, respectively. The result indicated that taking advantage of hyper-spectral data for monitoring of litchi SPAD values would be effective.