基于小波变换和连续投影算法的黑土有机质含量高光谱估测

Hyperspectral estimation of black soil organic matter content based on wavelet transform and successive projections algorithm

为实现黑土有机质含量更准确的估测,提出了基于小波变换和连续投影算法的高光谱估测方法。以典型黑土区采集的土壤样品为研究对象、分析光谱设备(analytical spectral devices,ASD)光谱仪获取的可见光—近红外区间光谱数据和经化学分析得到的土壤有机质含量为数据源,首先采用小波变换提取1~7层小波低频系数,然后利用连续投影算法分别对土壤全谱和1~7层小波低频系数进行变量筛选,最后分别基于土壤全谱、1~7层小波低频系数、连续投影算法选择的变量,利用偏最小二乘和支持向量机两种方法建立估测模型。结果表明:经小波变换和连续投影算法处理后,不仅变量数目得到了大幅降低,而且模型精度也进一步得到了提高,采用偏最小二乘法时,R2由土壤全谱的0.79提高至第6层小波低频系数的0.93,RMSE由6.06 g·kg^(-1)降低至3.48 g·kg^(-1);采用支持向量机方法时,R2由土壤全谱的0.75提高至第3层小波低频系数的0.91,RMSE由7.46 g·kg^(-1)降低至4.12 g·kg^(-1),说明提出的方法能有效用于黑土有机质含量高光谱估测。

Black soil is a valuable land resource,and the content of organic matter is an important index reflecting soil fertility,state and degradation degree.In order to estimate black soil organic matter content more accurately,this paper proposes a hyperspectral estimation method based on wavelet transform and successive projections algorithm.In this paper,the soil samples collected in the typical black soil region were used as the research object,and the Vis-NIR spectral data of the soil obtained from analytical spectral deviees(ASD)spectrometer and the organic matter content through chemical analysis were used as the data sources.Firstly,wavelet transform was used to extract the wavelet coefficients of 1 to 7 levels,and then successive projections algorithm was used to select the variables from soil original spectrum and the wavelet coefficients of 1 to 7 levels respectively.Finally,based on the soil original spectrum,the wavelet coefficients of 1 to 7 levels and the selected variables based on successive projections algorithm respectively,partial least squares and support vector machine were used to build the estimation models.The results show that,by using wavelet transform and successive projections algorithm,not only the number of variables is reduced greatly,but also the accuracies of the models are improved.When using partial least squares method,R2 increases from 0.79 of the soil original spectrum to 0.93 of the wavelet coefficient of the sixth level,and RMSE decreases from 6.06 g·kg^(-1) to 3.48 g·kg^(-1).When support vector machine method is used,R2 increases from 0.75 of the soil original spectrum to 0.91 of the wavelet coefficient of the third level,and RMSE decreases from 7.46 g·kg^(-1) to 4.12 g·kg^(-1).The results indicate that the proposed method can be effectively used for the hyperspectral estimation of black soil organic matter content.