基于高光谱成像反射和透射技术的雨生红球藻叶绿素含量研究

Chlorophyll content research of Haematococcus pluvialis based on hyperspectral reflection and transmission technology

通过高光谱成像仪采集雨生红球藻(Haematococcus pluvialis)反射和透射的光谱信息,结合化学计量学方法,对雨生红球藻叶绿素a、叶绿素b含量进行预测。比较了5种预处理方法[原始光谱(raw)、基线校正(baseline correction,BO)、卷积平滑(Savitzky-Golay smoothing,SG)、多元散射校正(multiplicative scatter correction,MSC)、变量标准化(standard normal variate,SNV)],并结合偏最小二乘回归(partial least square,PLS)建模的结果,选择较好的预处理方法后采用连续投影算法(successive projections algorithm,SPA)提取特征波长,并比较3种建模算法[PLS、多元线性回归(multiple linear regression,MLR)、最小二乘支持向量机(least square support vector machine,LS-SVM)]的预测结果,选择效果较好的建模算法用于预测叶绿素含量。其中,高光谱成像反射法对叶绿素a含量的预测,显示SNV预处理算法和SPA-MLR建模算法的效果较好,预测的剩余预测偏差(residual predictive deviation,RPD)达到3.429 6;高光谱成像透射法对雨生红球藻叶绿素a含量的预测显示,BO预处理算法和SPA-LS-SVM建模算法的效果较好,预测的RPD值达到3.156 3;同样地,对于高光谱成像反射法和透射法雨生红球藻叶绿素b含量的预测,显示均是采用BO预处理算法和SPA-MLR建模算法的效果较好,预测的RPD值分别为1.822 1和2.013 2。研究表明,采用高光谱成像反射和透射系统,通过一定的预处理结合建模算法可以对雨生红球藻叶绿素a、叶绿素b含量进行预测,为叶绿素a、叶绿素b含量的检测提供了1种新的方法。

The reflection and transmission spectral information of Haematococcus pluvialis were collected through hyperspectral imager, the contents of chlorophyll a and b of H. pluvialis were predicted combined with chemometrics method. For predicting the chlorophyll content of H. pluvialis, we compared five pretreatment methods [-（Raw）, Baseline Off （BO）, Savitzky-Golay smoothing（SG）, multiplication scatter correction （MSC）, standard normal variate （SNV）] combined with partial least square （PLS） modeling, and then chosen better pretreatment and used successive projections algorithm （SPA） Then we compared three modeling algorithm [-PLS, multiple linear regression （MLR）, least square support vector machine （LS-SVM）], and chosen better modeling algorithm. For predicting the chlorophyll a content of H. pluvialis by hyperspectral reflection technology, the result shows that SNV pretreatment and SPA-MLR modeling algorithm work best （RPD value reached 3. 429 6）. For predicting the chlorophyll a content of Haematococcus pluvialis by hyperspectral transmission technology, the result shows that t30 pretreat- ment and SPA-LS-SVM modeling algorithm work best （RPD value reached 3. 156 3）. Both for predicting the chlorophyll b content of H. pluvialis by hyperspectral reflection and transmission technology, the result shows that IN） pretreatment and SPA- MLR modeling algorithm are best （RPD value reached 1. 822 1 and 2. 013 2）. The research has shown that it can predict the contents of chlorophyll a and chlorophyll b through some pretreatment algorithms and modeling algorithms by hyperspeetral reflection and transmission technology and it can provide a new approach for detecting the content of chlorophyll a and chlorophyll b.