基于PCA的土壤含水量高光谱反演模型比较研究

Comparison of Hyperspectral Soil Water Content Inversion Models Based on PCA

摘要: 土壤含水量(SMC)是农作物生长过程中的重要指标之一，快速测定土壤水分状况对于农业发展极为重要。针对高光谱波段信息冗余和共线性特点以及光谱测定中无关信息和噪声带来的影响，对原始光谱进行倒数对数一阶微分变换预处理，利用主成分分析(PCA)对光谱信息进行降维，提取出12个主成分作为自变量；利用偏最小二乘、支持向量机、随机森林、BP神经网络建立SMC预测模型并进行综合对比分析。结果表明：利用PCA对光谱信息降维后的主成分建立的模型均有良好的预测能力，选取的主成分累积贡献率达到92%以上；通过对光谱变换前后模型的综合对比可以发现PCA-RF模型精度最高(R2分别为0.8362，0.9630，RPD分别为2.4229、3.7019)。

Abstract: Soil moisture content (SMC) is one of the important indexes in the process of crop growth. How to quickly measure soil moisture is very important. In the previous SMC water content inversion research, the characteristic band and linear empirical model are mostly used to construct the inversion model, and the characteristics of hyperspectral band information redundancy and collinearity are not fully considered, so the prediction ability of the model needs to be improved. In order to reduce the influence of irrelevant information and noise in spectral determination, the original spectrum was preprocessed by reciprocal logarithm first-order differential transformation; In order to weaken the redundancy and collinearity of hyperspectral band information, principal component analysis (PCA) is used to reduce the dimension of the original spectral information, and 12 principal components are extracted as independent variables; The SMC prediction model is established by using partial least squares, support vector machine, random forest and BP neural network. The results show that PCA has good prediction ability for the principal component model after spectral dimensionality reduction, and the cumulative contribution rate of the selected principal component is 99.99%; Through the comprehensive comparison of the models before and after spectral transformation, it can be found that the accuracy of PCA-RF model is the highest (R2 is 0.8362 and 0.9630 respectively, RPD is 2.4229 and 3.7019 respectively).