基于高光谱的土壤不同颗粒含量预测分析

以典型黄河下游冲积平原区的土壤为研究对象，分析土壤高光谱特征，探讨土壤质地不同粒级颗粒含量的统一估测途径，为土壤质地快速监测评价提供技术支持。选择原始光谱，及其倒数、对数、标准正交变换、多元散射变化、一阶微分、二阶微分共7种光谱变换形式，首先主成分降维，然后分别建立土壤黏粒、粉粒和砂粒含量的支持向量机预测模型，采用决定系数、均值绝对误差、均方根误差3种精度指标来衡量模型的预测能力。结果表明:原始光谱的对数为最佳光谱变换形式，具有最佳的土壤不同颗粒含量估测能力，决定系数R2 ≥ 0.6853，均值绝对误差MAE ≤ 0.1193，均方根误差RMSE ≤ 0.1683；黏粒含量的变化范围相对集中，预测能力整体表现的相对较强，R2 = 0.8127，MAE = 0.0820，RMSE = 0.1248。通过筛选最佳光谱变换处理，主成分降维，支持向量机预测，建立了土壤中黏粒、粉粒和砂粒含量的统一估测途径，实现了简单快捷的高光谱估测。

Different Soil Particle Contents Prediction Based on Hyperspectral Data

The hyperspectral characteristics of different particle contents of soil in the typical alluvial plain of the lower Yellow River were investigated in order to provide technical support for rapid monitoring and evaluation of soil texture. Seven spectral transformation forms were selected, including the original spectrum and its reciprocal, logarithm, standard orthogonal transformation, multivariate scattering change, the first derivative and the second derivative. First, principal component analysis was applied to reduce dimension. Then, predictive models of the contents of clay, silt and sand were established with support vector machines. Three accuracy indices were selected including determination coefficient, mean absolute error and root mean squared error. The results showed that the logarithm of the original spectrum was the best spectral transformation form due to the best prediction with the R2 ≥ 0.6853, the MAE ≤ 0.1193 and the RMSE ≤ 0.1683. The variation range of clay content was relatively concentrated, showing the best prediction by the R2 of 0.8127, MAE of 0.0820 and RMSE of 0.1248. The soil spectrum was reduced dimension with principal component by selecting the best spectral transformation. Whereafter, the support vector machine modeling was used to predict the contents of clay, silt and sand in soil, which was realized a simple and fast hyperspectral estimation on soil texture.