基于高光谱数据的地表水化学需氧量反演  

Inversion of Chemical Oxygen Demand in Surface Water Based on Hyperspectral Data

作　　者：王雪映 刘适搏 朱继伟[1,3] 马婷婷 WANG Xue-ying;LIU Shi-bo;ZHU Ji-wei;MA Ting-ting(Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China;University of Chinese Academy of Sciences,Beijing 100049,China;Changchun Changguang Gerui Optoelectronic Technology Co.,Ltd.,Changchun 130102,China;Jilin Provincial Bureau of Hydrology and Water Resources,Changchun 130022,China)

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033 [2]中国科学院大学,北京100049 [3]长春长光格瑞光电技术有限公司,吉林长春130102 [4]吉林省水文水资源局,吉林长春130022

出　　处：《光谱学与光谱分析》2024年第4期997-1004,共8页Spectroscopy and Spectral Analysis

基　　金：吉林省科技发展计划项目(20210203015SF);国家重点研发计划项目(2018YFF01011102)资助。

摘　　要：化学需氧量(COD)是地表水质量评价的重要指标。传统的COD检测方法存在需使用有毒试剂、易造成二次污染等缺点,高光谱法可避免上述缺点,在COD检测方面有着广阔的应用前景。为了探索在室内利用高光谱技术反演地表水COD浓度的可行性方法,以吉林省内流域的129个地表水样本为研究对象,将样本集以3∶1划分为训练集和测试集,使用高光谱成像系统收集样本的DN值并计算相应的水体光谱反射率,采用导数法进行数据预处理,通过Pearson相关性分析判断光谱数据与COD浓度实测值间的相关程度并提取特征谱数据。利用全谱数据和特征谱数据分别建立基于粒子群算法优化的最小二乘支持向量机(PSO-LSSVM)反演模型,通过决定系数R^(2)、均方差RMSE和相对偏差RPD分析比较这几种模型的预测精度和可靠性。研究结果表明:经过导数法预处理后,地表水COD浓度与光谱反射率的相关性明显增强;利用导数光谱数据建模的预测结果优于用原始光谱数据建模的预测结果;提取特征谱数据所建立的模型比利用全谱数据建立的模型有更好的预测效果。其中采用一阶导数预处理方法并利用特征谱建立的地表水COD浓度反演模型预测结果最好,验证集决定系数R^(2)=0.8567,均方差RMSE=3.8229,相对偏差RPD=2.6414。以上研究初步证实了在室内基于高光谱数据对地表水COD浓度进行反演的可行性,为高光谱技术用于地表水COD检测提供了新的方法和思路。Chemical oxygen demand(COD)is an important surface water quality evaluation index.The traditional COD detection method needs to use toxic reagents,easily cause secondary pollution and other shortcomings;hyperspectral method can avoid the above shortcomings so that it has a broad application prospect in COD detection.In order to explore the feasibility of indoor inversion of COD concentration of surface water by hyperspectral technology,this paper takes 129 surface water samples in Jilin Province as research objects,divides the sample set into the training set and test set with sample number ratio of about 3∶1,and uses hyperspectral imaging system to collect DN values of samples and calculate the corresponding spectral reflectance of water bodies.The derivative method is used for data preprocessing.Pearson correlation analysis is used to judge the correlation degree between spectral data and measured COD concentration,and the characteristic spectral data is extracted.A least square support vector machine(PSO-LSSVM)inversion model optimized by particle swarm optimization was established using full and characteristic spectrum data respectively.These models prediction accuracy and reliability were compared by analyzing the coefficient of determination R^(2),root mean square error RMSE,and relative percent deviation RPD.The results show that the correlation between COD concentration and spectral reflectance of surface water is significantly enhanced after derivative pretreatment.The prediction results based on derivative spectral data are better than those based on original spectral data.The model based on extracting characteristic spectral data has a better prediction effect than the model based on full spectral data.Among them,the inversion model of surface water s COD concentration established using the first derivative preprocessing method and the characteristic spectrum has the best prediction results.The determination coefficient of verification set R^(2)=0.8567,the root mean square error RMSE=3.8229,and the r

关 键 词：高光谱 化学需氧量 导数法 最小二乘支持向量机 粒子群算法 

分 类 号：O433.4[机械工程—光学工程]