基于高光谱的大叶女贞叶片水分定量测定

Determination of Ligustrum Leaf Water Content Based on Hyperspectral

较传统水分化学检测法相比,高光谱是一种新兴的简单、快速、无损、原位的绿色检测方法.大叶女贞是一种具有经济、医药、欣赏价值的树种,研究价值较高,然而有关高光谱对其的研究鲜有报道,将其作为研究对象,利用可见近红外高光谱采集40个叶片样本的高光谱数据,利用ENVI 4.7从中提取出反射光谱数据,在MATLAB7.0平台对光谱数据进行直接正交信号校正(DOSC)预处理后,采用共生矩阵法(SPXY)选取30个样本作为建模集,其余作为预测集,构建支持向量回归(SVR)模型,然后在此全波段建模基础上采用自适应权重采样法(CARS)、连续投影变换(SPA)等波段筛选法进行特征波段建模分析,结果表明DOSC-SPA-SVR模型效果最优,预测平方相关系数R2p=0.974 3,预测均方根误差(RMSEP)为4.4×10-5,较全波段模型相比,精度相当甚至有所提升,模型简单、稳定,能够用于对女贞叶片水分进行快速无损的定量分析,为其他作物营养诊断和精准灌溉提供一定的参考价值,减少水资源的浪费.

Compared with traditional chemical detection methods for moisture, hyperspectral is a simple,rapid,nonde-structive ,new,green,and in-situ detection method. Ligustrum lucidum is a kind of tree with economy,medical and apprecia-tion value,and it is of great value in this field,however,there are few reports about its research in hyperspetral. This paper choosed its leaves as the subject of study, 40 leaf Hyperspectral datas with visible near infrared spectrometer were collected and the reflectance spectral data were obtained by using ENVI 4.7. The direct signal correction preprocessing for hyperspectral datas were performed in the MATLAB 7.0. The co-occurrence matrix method （SPXY） was used to devide the sample set, 30 of those were chose as the modeling set and the other as prediction set, then establishing support vector regression （SVR） model we compared and selected the optimal model. Adaptive weighted sampling method （CARS）,successive projection algorithm （SPA） , and the other characteristic wavelength were used to build model after screening of the whole band spectrum based on the optimal model. Results show that DOSC-SPA-SVR is optimum and the precision of model is equivalent to optimal full band m）del, squared correlation coefficient of prediction set is 0.^14 3, root mean square error of prediction is 4 .4X 10 -5 . This mod-el is single and stable, its accuracy is equivalent and even improved compared with the whole band model. It can be used in the rapid and nondestructive quantitative analysis for lisgustrum lucidum leaf moisture and can provide some reference value for oth-er nutrition diagnosis and precision irrigation to reduce the waste of water resources.