论文部分内容阅读
提出一种利用可见/近红外光谱技术进行杉木林土壤全氮测定的方法。利用不同方法实现了土壤光谱的预处理,并以偏最小二乘回归算法(PLS)建立土壤氮含量估测模型对其进行比较分析,发现小波除噪结合多远散射校正能最有效地消除原始光谱的噪声与背景信息,此时PLS模型校正集与预测集R2分别为0.891与0.885。为优化模型,对预处理后的光谱数据采用主成分分析法(PCA)降维,以最小二乘支撑向量机回归算法(LS-SVR)建立了土壤氮含量估测模型,其校正集与预测集R2分别提高至0.921与0.917,具有比PLS算法更高的精度。结果表明:以可见/近红外光谱技术进行林地土壤氮含量快速监测是可行的,其中小波去噪结合多元散射校正系光谱预处理的优选方法,而LS-SVR则是建模的优选方法。
A method for determination of total nitrogen in Chinese fir plantation by visible / near infrared spectroscopy was proposed. Soil spectra were preprocessed by different methods and compared with partial least-squares regression (PLS) to establish a soil nitrogen content estimation model. It was found that wavelet denoising combined with multi-far scatter correction could eliminate the original Spectral noise and background information, then the PLS model calibration set and prediction set R2 were 0.891 and 0.885 respectively. In order to optimize the model, principal component analysis (PCA) dimensionality reduction was applied to the preprocessed spectral data to establish a soil nitrogen content estimation model based on least square support vector machine regression (LS-SVR). The calibration set and prediction Set R2 to 0.921 and 0.917, respectively, with higher accuracy than the PLS algorithm. The results show that it is feasible to monitor nitrogen in forest soils by visible / near infrared spectroscopy. Among them, wavelet denoising combined with multivariate scatter correction is the preferred method for spectral pretreatment, while LS-SVR is the preferred method for modeling.