IWD-Hankel-SVD模型下玉米叶片光谱铜污染信息预测

Predicting the Copper Pollution Information of Corn Leaves Spectral Based on an IWD-Hankel-SVD Model

近年来在工业化和城镇化快速发展的地区,由重金属污染导致的环境问题尤为突出,特别是农业重金属污染更为社会所关注,因此,探索快速便捷的重金属污染甄别与监测方法极为重要。高光谱遥感作为新兴的重金属污染监测技术已有了深入研究。提出了固有波长尺度分解(IWD)概念和方法,并结合Hankel矩阵和奇异值分解(S VD)等建立了植被重金属污染程度预测的IWD-Hankel-SVD模型,该模型分为单变量模型和多变量模型。单变量模型主要是通过重金属污染的植被光谱IWD处理来获取光谱信息固有旋转分量(PRC)以提取最佳PRC的有效特征波段;在对各特征波段所构建的Hankel矩阵进行奇异值分解(SVD)基础上,依据获得该模型的奇异熵实现重金属污染信息预测。多变量模型是以植物叶绿素浓度相对值、单变量模型奇异熵作为参数实现重金属污染的信息预测。根据不同重金属Cu^(2+)胁迫梯度下玉米植株污染的叶片光谱和叶绿素浓度以及叶片中Cu^(2+)含量测定的数据,首先对不同浓度Cu^(2+)胁迫下玉米叶片光谱进行IWD分析,获得能够较好保留原始输入光谱信息的最佳PRC,并从中提取到有效特征波段553~680,681~780,1266~1429,1430~1631,1836~1913和1914~2111 nm;然后对每一个特征波段构造其Hankel矩阵并进行SVD处理,以求取单变量的IWD-Hankel-SVD模型奇异熵;最后通过各特征波段所对应模型奇异熵与玉米叶片中Cu^(2+)含量的相关分析,得到依据1266~1429和1836~1913 nm特征波段计算出奇异熵与玉米叶片中Cu^(2+)含量的决定系数R 2均高达0.9左右,说明这两个特征波段用于IWD-Hankel-SVD模型的Cu污染程度预测更具优越性和解释能力。同时,再把玉米叶片中叶绿素浓度相对值、1266~1429和1836~1913 nm特征波段相应模型奇异熵作为参数,采用偏最小二乘回归分析,得出多变量IWD-Hankel-SVD模型的玉米叶片Cu污染程度预测能力更强,决定系数R ^(2)达到0.9476,证明了多变量模型更具有鲁棒性和稳健性。

The environmental problems caused by heavy metal pollution have been particularly prominent in the regions with the rapid industrialization and urbanization development,especially the heavy metal pollution in agriculture is more concerned by the society.Therefore,it is very important to explore some fast and convenient methods on screening and monitoring heavy metal pollution.As a new technology of monitoring heavy metal pollution,the hyperspectral remote sensing has been paid attention and researched deeply by many scholars.A concept and method of inherent wavelength-scale decomposition(IWD)was proposed in the paper,and an IWD-Hankel-SVD model was established for predicting heavy metal pollution degree of vegetation combined with the Hankel matrix and the singular value decomposition(SVD),here the model was divided into single-variable model and multi-variable model.The single-variable model was mainly used to obtain the intrinsic rotation components(PRC)of spectral information of vegetation polluted by heavy metal through IWD processing and to extract the effective characteristic bands of the best PRC,then it could be realized to predict the heavy metal pollution according to the singular entropy of the model acquired by using SVD to decompose the Hankel matrix constructed based on each characteristic band.But the multi-variable model was used to realize the prediction of heavy metal pollution information by taking the relative values of vegetation chlorophyll concentration and the singular entropy acquired by the single-variable model as parameters.According to the data of leaf spectra,measured chlorophyll concentrations and Cu^(2+)contents in corn leaves polluted by heavy metal Cu^(2+)under different stress gradients,firstly the spectra of corn leaves stressed by the different Cu^(2+)concentrations were analyzed by IWD,the best PRC was obtained which could well retain the original spectral information,and some effective characteristic bands were extracted to be 553~680,681~780,1266~1429,1430~1631,1836~1913 and 1914~2111 nm from the PRC,then the Hankel matrix of each characteristic band was constructed and processed by SVD to obtain the singular entropy of the single-variable model,finally through the correlation analysis between the singular entropy of the model corresponding to each characteristic band and the Cu^(2+)contents in corn leaves,it was found that the determination coefficients R2 of the singular entropy and the Cu^(2+)contents in the leaves were all about 0.9 computed based on the 1266~1429 and 1836~1913 nm characteristic bands,the result shows that the two characteristic bands had more advantageous and interpretable for the IWD-Hankel-SVD model on predicting the Cu pollution degrees.At the same time,it was concluded that the multi-variable IWD-Hankel-SVD-model had stronger prediction ability of Cu pollution degrees in corn leaves by using the partial least square regression analysis based on taking the relative values of chlorophyll concentration in corn leaves and the singular entropy of the single-variable model corresponding to 1266~1429,1836~1913 nm characteristic bands as parameters,and the determination coefficient R2 reached 0.9476,so the multi-variable model was proved to be more robustness and steadiness.