基于VMD-MSE的玉米铜污染信息提取与预测模型

摘要:

重金属污染会引起作物光谱畸变，本文通过挖掘光谱信息中微弱的畸变信息诊断玉米受污染程度。将变分模态分解（VMD）运用到高光谱弱信息探测中，并结合多尺度熵（MSE）构建VMD-MSE光谱弱信息探测模型，同时利用模型值VM进行Cu2+含量回归分析与建模。结果表明：对原始光谱数据进行3次VMD分解后，可有效提取光谱奇异特征；计算VMD结果的MSE值，可获取5个尺度的模型值。各尺度模型值VM与玉米叶片中Cu2+含量呈现显著负相关，其中第一尺度模型值（VM1）与叶片中Cu2+相关性最好。对各尺度VM构建的Cu2+含量预测模型应用结果进行比较，证明VM1线性回归模型预测效果最优。表明VMD-MSE模型可为作物污染信息提取、污染诊断及Cu2+含量预测提供思路与方法。

Abstract:

Spectral reflectance of crop will be changed slightly when crop is stressed by heavy metal. The changes of crop spectral reflectance have considerable significance for crop contamination diagnosis. However, vegetation photosynthetic components are complex, which means that there may be no visible symptoms in leaf spectral reflectance when the crop is stressed by heavy metal. And therefore the object was to develop a weak information extraction method to excavate the vegetative stress signals through minimizing the effects of background materials, such as those caused by nonphotosynthetic components. A VMD-MSE model was built to excavate and measure the weak information in corn leaves spectrum by introducing the variational mode decomposition (VMD) into hyperspectral weak information detection and combining with multiscale entropy (MSE). The model value could be obtained after treating corn leaves spectrum by VMD-MSE model. In addition, linear regression models between model values of corn leaves spectrum under different stress concentrations and Cu2+ contents in corn leaves were established. The results showed that the spectrum singular features of the original spectrum of corn leaves can be extracted effectively after three times decomposition of variational mode decomposition. Model values of five scales were obtained by calculating the multiscale entropy of the result of threetime variational mode decomposition. And VM, the model value at five scales, had a significant negative correlation with Cu2+ contents in corn leaves, and the most significant correlation was between the firstscale model value (VM1) and Cu2+ contents in leaves. The linear regression model established based on VM1 and Cu2+ contents in corn leaves was proved to be optimal by comparing the application results of five Cu2+ contents prediction models. Therefore, the VMD-MSE model can provide a new method for pollution information extraction, crop contamination diagnosis and Cu2+ contents prediction.