连续小波变换的土壤有机质含量高光谱估测  

Hyperspectral Estimation of Soil Organic Matter Content Based on Continuous Wavelet Transformation

作　　者：玉米提·买明 王雪梅[1,2] Yumiti Maiming;WANG Xue-mei(College of Geographic Science and Tourism,Xinjiang Normal University,Urumqi 830054,China;Xinjiang Uygur Autonomous Region Key Laboratory“Xinjiang Arid Lake Environment and Resources Laboratory”,Urumqi 830054,China)

机构地区：[1]新疆师范大学地理科学与旅游学院,新疆乌鲁木齐830054 [2]新疆维吾尔自治区重点实验室“新疆干旱区湖泊环境与资源实验室”,新疆乌鲁木齐830054

出　　处：《光谱学与光谱分析》2022年第4期1278-1284,共7页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(41561051);新疆维吾尔自治区自然科学基金项目(2020D01A79)资助

摘　　要：土壤有机质含量的高光谱估测可快速、准确监测土壤肥力,对现代化农业生产进行精准施肥提供科学依据。以新疆渭干河-库车河三角洲绿洲耕层土壤为研究对象,对采集的98个土壤样品的原始光谱反射率R分别进行传统倒数对数lg(1/R)、一阶微分R′和倒数对数一阶微分[lg(1/R)]′数学变换,以及基于小波母函数Bior1.3不同尺度分解的连续小波变换(CWT),并与实测土壤有机质含量进行相关分析,从而筛选出各类变换下与土壤有机质含量密切相关的特征波段和小波系数(p<0.01)。分别以原始光谱反射率(R)以及不同变换处理下的特征波段反射率和敏感小波系数作为自变量,土壤有机质含量作为因变量,采用偏最小二乘回归和支持向量机回归方法构建土壤有机质含量的估测模型。结果表明:(1)各类光谱变换方法有效提升光谱与土壤有机质含量之间的敏感性,其中经CWT变换后的土壤光谱反射率与有机质含量的相关性得到显著提高,相关系数由0.39提高到0.54(p<0.01)。(2)传统的[lg(1/R)]′变换构建的支持向量机回归模型,其决定系数(R^(2))高于lg(1/R)和R′变换构建的模型,说明倒数对数一阶微分变换可有助于提高估测模型的精度,且支持向量机回归模型的精度和稳定性高于偏最小二乘回归模型。(3)经过CWT分解后,以原始光谱反射率在不同尺度上的敏感小波系数作为自变量建立的模型,估测精度和稳定性均有明显的提高,构建的R-CWT-2^(3)-SVMR模型的决定系数(R^(2))为0.84,均方根误差(RMSE)为1.48,相对分析误差(RPD)等于2.11,模型精度达到最高并拥有极好的预测能力。高光谱数据经多种变换处理后可有效去除白噪声,而连续小波变换处理比传统的数学变换方法更适合于挖掘土壤有效信息,实现光谱信号的近似特征和细节特征的有效分离,建立的反演模型可更加精准估测土壤有机质含量。The hyperspectral estimation of soil organic matter content can quickly and accurately monitor soil fertility and provide a scientific basis for proper fertilization of modern agricultural production.Taking the cultivated soil in the delta oasis of Weigan-Kuqa river,Xinjiang as the research object,the original spectral reflectance(R)of the collected 98 soil samples was subjected to the traditional reciprocal logarithm lg(1/R),the first-order differential(R′)and the reciprocal logarithm first-order differential[lg(1/R)]′mathematical transformations,and continuous wavelet transformation(Continuous Wavelet Transformation,CWT)processing based on Bior1.3 as the wavelet mother function through different scale decomposition.Correlation analysis was conducted between the treatment results and the measured soil organic matter content to screen out the characteristic bands and wavelet coefficients closely related to soil organic matter content under various transformations(p<0.01).With the original spectral reflectance,the characteristic band reflectance and the sensitive wavelet coefficient under different transformation treatments as independent variables and the soil organic matter content as dependent variables,partial least squares regression and support vector machine regression was used to estimate models of soil organic matter content.The results showed that:(1)Various spectral transformation methods can effectively improve the sensitivity between the spectrum and the content of soil organic matter.The correlation between the soil spectral reflectance and the organic matter content after continuous wavelet transformation has been significantly improved,and the correlation coefficient has been increased from 0.39 to 0.54(p<0.01).(2)The support vector machine regression model built by the traditional[lg(1/R)]′transformation has a higher coefficient of determination(R^(2))than the model built by lg(1/R)and R′transformation,showing the reciprocal logarithm first-order differential transformation can help improv

关 键 词：连续小波变换 分解尺度 高光谱估测 土壤有机质 渭干河-库车河三角洲绿洲 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]