田安红, 赵俊三, 张顺吉, 付承彪, 熊黑钢. 基于分数阶微分的盐渍土电导率高光谱估算研究[J]. 中国生态农业学报(中英文), 2020, 28(4): 599-607. DOI: 10.13930/j.cnki.cjea.190865
引用本文: 田安红, 赵俊三, 张顺吉, 付承彪, 熊黑钢. 基于分数阶微分的盐渍土电导率高光谱估算研究[J]. 中国生态农业学报(中英文), 2020, 28(4): 599-607. DOI: 10.13930/j.cnki.cjea.190865
TIAN Anhong, ZHAO Junsan, ZHANG Shunji, FU Chengbiao, XIONG Heigang. Hyperspectral estimation of saline soil electrical conductivity based on fractional derivative[J]. Chinese Journal of Eco-Agriculture, 2020, 28(4): 599-607. DOI: 10.13930/j.cnki.cjea.190865
Citation: TIAN Anhong, ZHAO Junsan, ZHANG Shunji, FU Chengbiao, XIONG Heigang. Hyperspectral estimation of saline soil electrical conductivity based on fractional derivative[J]. Chinese Journal of Eco-Agriculture, 2020, 28(4): 599-607. DOI: 10.13930/j.cnki.cjea.190865

基于分数阶微分的盐渍土电导率高光谱估算研究

Hyperspectral estimation of saline soil electrical conductivity based on fractional derivative

  • 摘要: 传统电导率的反演模型采用整数阶微分(1阶或2阶)的预处理方法,忽略位于分数阶微分处的高光谱反射率信息。因此,本研究提出一种基于分数阶微分的盐渍土电导率高光谱估算方法,以新疆昌吉回族自治州境内的盐渍化土壤为研究靶区,于2017年5月采集0~20 cm的表层土壤样品,利用FieldSpec®3 Hi-Res光谱仪测量盐渍土的野外高光谱,并在实验室化验土壤的电导率理化参数。在Matlab 2019a软件中编程实现0阶-2.0阶的Grünwald-Letnikov分数阶微分计算(阶数间隔为0.1)。分析土壤高光谱与电导率的相关系数曲线在21种微分处的变化规律,选择每阶微分的最大相关系数大于0.5时对应的波长为敏感波长,采用逐步多元线性回归模型对电导率进行预测。结果表明:分数阶微分预处理方法能够把相关系数曲线位于不同分数阶时的变化细节呈现出来,在全波段范围内出现更多的波峰和波谷信息。电导率的8个敏感波长为400 nm、418 nm、567 nm、1 667 nm、2 132 nm、2 193 nm、2 257 nm和2 258 nm。估算电导率的最佳模型位于分数阶1.5阶,其验证集的RPD值为1.99,R2为0.81,RMSE为1.08,该模型因RPD值大于1.8对电导率的估算能力好。本研究探索了电导率在不同分数阶微分处的差异信息,为电导率的估算提供一种新的研究思路,对新疆干旱区盐渍土的改良提供了科学可靠的依据。

     

    Abstract: The integer-order differential (first-order or second-order) preprocessing method is often used in traditional electrical conductivity inversion models, but it ignores the hyperspectral reflectance information at the fractional-order differential. In this paper, a hyperspectral method based on fractional differential to estimate the electrical conductivity of saline soil was proposed. The salinized soil in Changji, Xinjiang was used as the research subject. The surface soil samples of 0-20 cm were collected in May 2017, the field hyperspectral of the saline soil was measured by a FieldSpec®3 Hi-Res spectrometer, and physical and chemical parameters of soil electrical conductivity were tested in the laboratory. Next, the Grünwald-Letnikov fractional derivative calculation between 0.0-order and 2.0-order was programmed in MATLAB 2019a software (order interval is 0.1). Then, the variation law of the correlation coefficient curves between soil hyperspectral and electrical conductivity under 21 kinds of differentials was analyzed. When the maximum correlation coefficient of each fractional derivative was greater than 0.5, the corresponding wavelength was selected as the sensitive wavelength. Finally, the stepwise multiple linear regression model was used to predict the electrical conductivity. The results showed that the fractional derivative preprocessing method could display the variation details of the correlation coefficient curve under different fractional orders, and more peaks and troughs appeared in the whole band. The eight sensitive wavelengths of electrical conductivity were 400 nm, 418 nm, 567 nm, 1 667 nm, 2 132 nm, 2 193 nm, 2 257 nm, and 2 258 nm. The best model for estimating electrical conductivity was located at the 0.5th-order. The relative percent difference (RPD) value of the verification set was 1.99, the determination coefficient (R2) was 0.81, and the root mean square error (RMSE) was 1.08. This model had the ability to estimate the electrical conductivity because the RPD value was greater than 1.8. This study explored the difference in electrical conductivity estimates under different fractional derivatives and provided a new method for electrical conductivity estimation, which could be of considerable value for research into improvement of saline soils in the arid regions of Xinjiang.

     

/

返回文章
返回