铜胁迫下玉米叶片光谱MRSVD特征与污染预测模型

Spectral MRSVD characteristics of corn leaves under copper stress and pollution prediction model

基于不同铜离子(Cu^2+)胁迫梯度下玉米盆栽实验,依据所测玉米叶片光谱数据和叶片中Cu^2+含量,通过多分辨奇异值分解(MRSVD)提取光谱弱信息,建立玉米叶片Cu^2+污染定性分析及含量预测模型.利用MRSVD对350~1 300 nm波段内玉米叶片光谱进行多层分解,依据模极大值理论和定义的敏感因子确定反映Cu^2+污染信息的最佳分量,在此基础上构建奇异系数(SC)、奇异指数(SI)、奇异面积(SA)和奇异谱熵(SSE)等奇异特征参数分析光谱奇异性,并采用逐步回归分析法建立奇异特征参数与玉米叶片Cu^2+含量之间的定量关系模型.结果表明:第9层MRSVD细节分量能较好地提取Cu^2+污染信息,奇异性主要集中在480~850 nm波段内,奇异特征参数与玉米叶片Cu^2+含量有较好的相关性,其中SI和SSE与玉米叶片Cu^2+含量存在较强的定量关系,其决定系数(R^2)为0.842 5,均方根误差为0.826 0,模型精度优于常规方法.奇异特征参数能够较好地诊断并定量分析玉米叶片铜污染水平,可为作物重金属污染监测提供参考.

Based on pot experiments of corn with different copper ion(Cu^2+) stress gradients, measured spectral data of com leaves and Cu^2+ content in corn leaves, and extracted weak spectral information by multi-resolution singular value decomposition(MRSVD), a qualitative analysis and content prediction model of Cu^2+ pollution in corn leaves was proposed. Decomposing the original spectrum of corn leaves in the range of 350-1 300 nm band by MRSVD, and ascertaining the optimal component that can reflect Cu^2+ pollution information is necessary. According to the modulus maximum theory and the defined sensitive factors, the singular characteristic parameters such as singular coefficient(SC), singular index(SI), singular area(SA) and singular spectral entropy(SSE) were constructed. And the quantitative relationship model between singular characteristic parameters and Cu2+ content in corn leaves through stepwise regression analysis was also proposed.The experimental results indicate that the ninth layer MRSVD detail component can best extract Cu^2+ pollution information. The singularity is mainly concentrated in the range of 480-850 nm band. The singular characteristic parameters have a good correlation with the Cu^2+ content in corn leaves, among which SI and SSE have a strong quantitative relationship with the Cu^2+ content in corn leaves. The coefficient of determination(R^2) between the predicted value and the measured value of the model is 0.842 5, and the root mean square error is 0.826 0. Model accuracy is superior to conventional method. Singular characteristic parameters can better diagnose and quantitatively analyze the copper pollution level in corn leaves, which can provide reference for monitoring heavy metal pollution in crops.