深度森林DF21模型在土壤镉含量高光谱反演中的性能评价  

Performance Evaluation of the Deep Forest 2021(DF21)Model in Retrieving Soil Cadmium Concentration Using Hyperspectral Data

作　　者：张梓浩 郭飞 吴坤泽[1] 杨馨玉 许镇 ZHANG Zi-hao;GUO Fei;WU Kun-ze;YANG Xin-yu;XU Zhen(Department of Electronic Information Engineering,College of Engineering,Shantou University,Shantou 515063,China;Department of Mechanical Engineering,College of Engineering,Shantou University,Shantou 515063,China;Institute of Geophysical and Geochemical Exploration,China Academy of Geological Sciences,Langfang 065000,China;Research Center of Geochemical Survey and Assessment on Land Quality,China Geological Survey,Langfang 065000,China)

机构地区：[1]汕头大学工学院电子信息工程系,广东汕头515063 [2]汕头大学工学院机械工程系,广东汕头515063 [3]中国地质科学院地球物理地球化学勘查研究所,河北廊坊065000 [4]中国地质调查局土地质量地球化学调查评价研究中心,河北廊坊065000

出　　处：《光谱学与光谱分析》2023年第8期2638-2643,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(42101398,62006096);汕头大学卓越人才计划科研启动经费项目(NTF20023);中国地质科学院地球物理地球化学勘查研究所所长基金项目(AS2019J02)资助。

摘　　要：高光谱技术可提供近乎连续的地物光谱曲线,对土壤组分定量反演具有极大的潜力。针对受污染土壤,通过综合分析实验室测定的土壤镉(Cd)含量和同期获取的高光谱数据,提出了一种基于深度森林2021(Deep Forest 2021,DF21)算法的高光谱土壤Cd含量反演模型。采用原始光谱数据(OS)和其经主成分分析(PCA)降维处理后的数据作为模型输入参数,分别构建了基于原始光谱数据的DF21(OS-DF21)模型和基于PCA降维数据的DF21(PCA-DF21)模型。随后,基于相同训练样本,OS-DF21和PCA-DF21分别建立了土壤Cd含量和输入参量之间的关系,并据此对检验样本土壤Cd含量进行了定量反演和对比分析。选取决定系数(R^(2))、均方根误差(RMSE)和相对分析误差(RPD)对模型反演性能进行了评估。分析结果表明:OS-DF21模型反演精度最佳,其R^(2)、RMSE和RPD分别可达0.873、0.120和2.892。相比而言,PCA降维处理虽可降低光谱数据的冗余度,但PCA-DF21模型的预测能力有所下降。尽管如此,PCA-DF21模型仍表现出较好的土壤Cd含量反演能力,其R^(2)、RMSE和RPD分别为0.779,0.159和2.190。因此,DF21算法可作为研究区及类似环境区域的土壤重金属快速勘测的补充手段。Hyperspectral technology can provide nearly continuous spectral curves of ground objects,which has excellent potential for retrievingthe soil s components.This paper investigates components retrieval from contaminated soil by hyperspectral technology.By so doing,it analyzes thesoil cadmium(Cd)concentration measured in the laboratory and the corresponding hyperspectral curvature data obtained at the same period,following whichthe retrieval model for the soil Cd concentration from the hyperspectral data in light with the(Deep Forest 2021,DF21)model is developed.In this study,the original spectrum(OS)data and the data processed by the Principal Component Analysis(PCA)are used as the model s input parameters.Subsequently,two models,i.e.,the OS-DF21 model based on the original spectral data and the PCA-DF21 model based on the PCA processed data,are established.The relationships between the input parameters and soil Cd concentration are respectively obtained by the OS-DF21 model and PCA-DF21 model.Then the soil Cd concentrationis estimated from the testing samples accordingly.To evaluate the retrieval performance,three indices,namely the coefficient of determination(R^(2)),Root Mean Square Error(RMSE),and Residual Predictive Deviation(RPD)applied in this study.It is found that the OS-DF21 model has the best performance for the retrieval of soil Cd concentration,whose R^(2),RMSE,and RPD are 0.873,0.120,and 2.892,respectively.In contrast,the PCA-DF21 model has arelatively lower retrieval accuracy,with R^(2),RMSE,and RPD being 0.779,0.159,and 2.190,though the PCA can reduce the dimensionality of the spectral data.In this regard,the DF21 shows good retrieval performance and can be an essential supplementary method for soil heavy metal surveys in the study area and similar environmental regions.

关 键 词：土壤Cd含量 深度森林DF21 主成分分析(PCA) 高光谱技术 反演模型 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]