ABSTRACT

Applying machine learning to hyperspectral remote sensing is a new way to solve practical problems. A new method that combines spectral indexes and Naive Bayes, named SINB, was proposed to distinguish heavy metal pollution in crops. First, the CLPI_OR, CLPI_CR, and CLPI_FOD (Copper-Lead Pollution Index of the original spectrum, continuum removed spectrum, and first-order differential spectrum) were constructed based on the processed crop spectra, and they were used as input variables for Bayes discrimination. Then, the discriminant functions were constructed using samples from the training group, and the discriminant rule was formulated. Finally, the samples from the validation group were used to test the universality and robustness of the SINB. The results showed that the CLPI_OR, CLPI_CR, and CLPI_FOD were sensitive to distinguish Cu (Copper) and Pb (Lead) pollution, and the discrimination accuracy of SINB was 100% in the training group and 100% in the validation group. This method fully used the crop leaf spectral characteristics and the advantages of machine learning and achieved the discrimination of crop heavy metal pollution.

摘要

将机器学习应用于高光谱遥感是解决实际问题的一种新方法。提出了一种将光谱指数和朴素贝叶斯相结合的新方法，称为SINB，以区分农作物中的重金属污染。首先，CLPI _OR，CLPI _CR和CLPI _FOD（原始光谱的铜铅污染指数，连续谱去除光谱和一阶微分光谱）基于处理后的作物光谱构建，并将它们用作贝叶斯判别的输入变量。然后，使用训练组的样本构造判别函数，并制定判别规则。最后，验证组的样本用于测试SINB的通用性和鲁棒性。结果表明，CLPI _OR，CLPI _CR和CLPI _FOD对区分铜（铜）和铅（铅）的污染敏感，SINB的辨别准确性在训练组中为100％，在验证组中为100％。该方法充分利用了农作物叶片的光谱特征和机器学习的优势，实现了对农作物重金属污染的判别。