土壤有机质高光谱灰色关联度估测模型研究

发布时间：2018-04-08 22:28

本文选题：高光谱遥感　切入点：土壤有机质　出处：《山东农业大学》2017年硕士论文

【摘要】：有机质是土壤的重要物质组成,对于作物生长,土地保水保肥以及陆地生态系统的正常运行均有重要作用。传统的土壤有机质测定方法以采样--化验为主要手段,虽然精度较高,但费时耗力,不易实施。高光谱遥感可以精确捕捉常规遥感所观测不到的地物细微的反射光谱信息,从而实现对地物的准确识别与反演,对于作物长势监测、土地利用评价以及精准农业具有重要意义。本研究以山东省泰安市为研究区,以采集的92个土样的有机质含量及其室外反射光谱为研究对象,基于土壤有机质光谱反演中的灰色特性,利用灰色系统理论,建立了土壤有机质高光谱灰色估测模型,对土壤有机质含量进行估测,通过与经典估测方法进行对比,验证了灰色估测模型的有效性。主要研究内容及结论如下:(1)确定了泰安市潮棕壤有机质敏感波段和特征因子利用平方根、倒数、对数、一阶微分及其组合、包络线去除等10种光谱变换技术对光谱进行变换,通过原始及变换光谱与土壤有机质含量的相关分析确定了有机质敏感波段,并利用极大相关性原则提取了特征因子。结果表明,10种变换技术中,一阶微分及对数倒数的一阶微分变换技术能明显提高有机质与反射光谱在可见光与近红外的相关性,而平方根、倒数、对数变换无益于提高反射光谱与土壤有机质的相关性;有机质的反射光谱特征主要位于可见光485~760nm波段以及近红外1375~1382nm、2121~2133nm、2336-2347nm三个水分吸收波段附近;选取的5个特征因子分别位于原始光谱665nm处,一阶微分光谱575nm和2341nm处以及对数倒数的一阶微分1378nm和2128nm处,相关性均大于0.55。(2)建立了土壤有机质高光谱灰色估测模型基于土壤有机质估测中的灰色特性和光谱特征因子的非时间序列特性,借助灰色系统理论,利用特征因子与因变量的相关系数和特征因子的标准差构造权重,对关联度进行改进,得到加权距离灰色关联度和灰色加权关联度;而后利用识别残差建立残差修正模型,得到两种具有残差修正的灰色关联估测模型,并在此基础上利用修正值残差的标准差将点值估测拓展为区间估测;最后与多元线性回归、BP神经网络、支持向量机模型进行了比较。结果表明,加权距离灰色关联度和灰色加权关联度均可用于土壤有机质高光谱估测,残差修正模型能有效提高估测精度,而区间估测能隐涵部分不确定性的影响,反映有机质的动态变化特性;5种模型的点值估测中,具有残差修正的灰色加权关联度模型和具有残差修正的加权距离灰色关联度估测模型精度最高,平均相对误差分别为6.79%、7.94%,其次是支持向量机,平均相对误差为12.94%,BP神经网络和多元线性回归模型表现较差,平均相对误差均高于14%,说明灰色关联估测模型在土壤有机质高光谱估测方面拥有很大潜力。
[Abstract]:Organic matter is an important material composition of soil, which plays an important role in crop growth, soil moisture and fertilizer conservation and the normal operation of terrestrial ecosystem.The traditional method of soil organic matter determination is sample-test. Although the precision is high, it is time-consuming and difficult to carry out.Hyperspectral remote sensing can accurately capture the subtle reflectance spectrum information of ground objects that can not be observed by conventional remote sensing, thus realizing the accurate identification and inversion of ground objects, which is of great significance for crop growth monitoring, land use evaluation and precision agriculture.Taking Taian City, Shandong Province as the research area, the content of organic matter and its outdoor reflectance spectrum of 92 soil samples were studied. Based on the grey characteristics of soil organic matter spectral inversion, the grey system theory was used.The hyperspectral grey estimation model of soil organic matter was established and the content of soil organic matter was estimated. The validity of the grey estimation model was verified by comparing it with the classical estimation method.The main contents and conclusions are as follows: (1) 10 spectral transformation techniques, such as square root, reciprocal, logarithm, first-order differential and their combination, and envelope removal, are used to transform the spectrum of organic matter sensitive bands and characteristic factors of Chao Brown soil in Taian City.The sensitive bands of soil organic matter were determined by correlation analysis of original and transformation spectra and soil organic matter content, and the characteristic factors were extracted by the principle of maximum correlation.The results show that the first-order differential transformation technique and the logarithmic differential transformation technique can obviously improve the correlation between organic matter and the reflected spectrum in visible light and near infrared, while square root, reciprocal.The five characteristic factors are located at the original 665nm, the first order differential spectrum at 575nm and 2341nm, and the logarithmic reciprocal of the first order differential 1378nm and 2128nm, respectively.The hyperspectral grey estimation model of soil organic matter was established. Based on the grey characteristics of soil organic matter estimation and the non-time series characteristics of spectral characteristic factors, the grey system theory was used to estimate soil organic matter.By using the correlation coefficient of characteristic factor and dependent variable and the standard deviation of characteristic factor to construct weight, the correlation degree is improved, and the weighted distance grey correlation degree and grey weighted correlation degree are obtained, and then the residual correction model is established by identifying residual error.Two grey correlation estimation models with residual correction are obtained, and on this basis, the point value estimation is extended to interval estimation by using the standard deviation of the modified residual value, and finally the multivariate linear regression BP neural network is used.The support vector machine model is compared.The results show that the weighted distance grey correlation degree and grey weighted correlation degree can be used to estimate soil organic matter hyperspectral. The residual correction model can effectively improve the accuracy of estimation, while the interval estimation can cover the effect of partial uncertainty.Among the five models reflecting the dynamic characteristics of organic matter, the grey weighted relational degree model with residual correction and the weighted distance grey relational degree model with residual correction have the highest accuracy.The average relative error was 6.79 and 7.94, followed by support vector machine. The average relative error was 12.94 BP neural network and multivariate linear regression model.The average relative error is higher than 14, indicating that the grey correlation estimation model has great potential in the hyperspectral estimation of soil organic matter.
【学位授予单位】：山东农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S153.621

【参考文献】