特征变量选择结合SVM的耕地土壤Hg含量高光谱反演

为探讨应用高光谱数据反演耕地土壤重金属汞(Hg)含量，对原始光谱进行10 nm重采样和SG平滑处理，用不同光谱变换数据与土壤重金属Hg含量进行相关性分析，采用IRIV、Random Frog和PCC提取光谱特征波段，分别建立SVM与GWO-SVM土壤Hg含量高光谱反演模型，获取Hg含量最优反演路径。研究表明，一阶微分变换光谱后土壤光谱特征更明显；上述特征提取方法在不同程度上减少光谱数据冗余，保留有效变量信息；经灰狼算法优化后支持向量机模型反演精度提高，IRIV结合GWO-SVM预测精度更高，其验证集R²为0.894,RMSE为0.082,MAE为0.016。研究成果可为类似土壤重金属含量的反演提供借鉴。

Feature variable selection combined with SVM for hyperspectral inversion of cultivated soil Hg content

In order to explore the application of hyperspectral data to invert the content of heavy metal mercury(Hg)in cultivated soils,the original spectrum is resampled and SG smoothed at 10nm,and the correlation between different spectral transformation data and soil heavy metal Hg content is analyzed.IRIV,Random Frog and PCC extracts spectral characteristic bands,and establishes SVM and GWO-SVM soil Hg content hyperspectral inversion models,respectively,to obtain the optimal Hg content inversion path.Research shows that the soil spectral characteristics are more obvious after the first-order differential transformation spectrum;the above-mentioned feature extraction methods reduce spectral data redundancy to varying degrees and retain effective variable information;the support vector machine model inversion accuracy is improved after the gray wolf algorithm is optimized.IRIV combined with GWO-SVM has higher prediction accuracy.Its verification set R2 is 0.894,RMSE is 0.082,and MAE is 0.016.The research results can provide reference for the inversion of similar soil heavy metal content.