论文部分内容阅读
对光谱数据进行预处理是提升高光谱建模精度十分必要和有效的途径。为了研究分数阶微分预处理方法在高光谱数据估算荒漠土壤有机碳含量中的应用,该研究以艾比湖流域为研究靶区,利用2015年5月采集的103个表层土壤样本的实测有机碳数据和室内测定的高光谱数据,以0.2阶为步长对原始光谱反射率及对应的倒数变换、对数变换、对数倒数变换、均方根变换的0-2阶微分进行分数阶运算预处理并研究其与土壤有机碳含量相关性,基于通过0.01显著性检验的特征波段对土壤有机碳含量进行偏最小二乘回归建模并进行精度分析。结果表明:1)分数阶微分预处理可以细化土壤有机碳及其光谱反射率相关性的变化趋势;2)各阶微分预处理后的相关系数通过显著性检验波段的数量均呈现先增后减的趋势,但波段数量最多的对应阶数并不统一;3)对数变换的1.6阶微分所建立的模型为最优模型,该模型的RMSEC=2.433 g/kg,R2c=0.786,RMSEP=2.263 g/kg,R2p=0.825,RPD=2.392。说明了分数阶预处理过后的模型精度和稳健性较整数阶微分有了大幅提升,并且运用在高光谱反演土壤有机碳含量上是可行的。
Pretreatment of spectral data is a very necessary and effective way to improve the accuracy of hyperspectral modeling. In order to study the application of fractional differential preprocessing method in estimating soil organic carbon content in desert soil by using hyperspectral data, this study took the Lake Aibi Basin as the target of study. Based on the data of 103 organic soil carbon samples collected in May 2015, Data and indoor hyperspectral data, fractional order operation of the 0-2 order differential of the original spectral reflectance and corresponding reciprocal transformation, logarithmic transformation, logarithmic reciprocal transformation, and root mean square transformation is performed in 0.2 steps The relationship between soil organic carbon content and soil organic C content was studied and studied. Partial least squares regression was used to model the soil organic C content based on the characteristic bands through 0.01 significance test. The results show that: 1) fractional differential preprocessing can refine the trend of correlation of soil organic carbon and its spectral reflectance; 2) the correlation coefficients of differential preprocessing by various orders show the first increase after passing the significant test band 3) the model established by the 1.6th order differential of logarithm transform is the optimal model, RMSEC = 2.433 g / kg, R2c = 0.786, RMSEP = 2.263 g / kg, R2p = 0.825, RPD = 2.392. It shows that the accuracy and robustness of the model after fractional preprocessing are greatly improved compared with integer differential and it is feasible to use the hyperspectral inversion of soil organic carbon content.