基于优选波长的复合肥总氮含量可见/近红外光谱分析

利用可见/近红外光谱分析仪得到复合肥的原始光谱，经过MSC和一阶导数预处理后，建立了复合肥中总氮含量的PLS模型，实现了复合肥中总氮含量的快速准确无损测量。通过选取不同范围波长建模，取得了预测决定系数（R²）、预测标准差（S_RMSEP）、相对分析误差（K_RPD）最好的基础波段。在基础波段的基础上，采用优选波长算法，获得了加入波长后的模型的预测决定系数和预测标准差图。通过分析，最终确定加入42个优选波长。实验结果表明，加入优选波长后的模型的预测决定系数由不加优选波长模型的0.760 4提高到了0.991 1，S_RMSEP降低为原来的1/5，K_RPD提高到原来的5倍。

Analysis of Total Nitrogen in Compound Fertilizers by VIS-NIR Spectroscopy

The original spectrum of compound fertilizer was obtained by VIS-NIR spectroscopy analyzer. After pre-treatment with multiplicative scatter correction(MSC) and first derivative, a partial least squares(PLS) model of total nitrogen(TN) content in compound fertilizer was established, which achieved rapid, accurate and non-destructive measurement. By selecting different wavelength models, the best fundamental band for predictive determination coefficient(R²), prediction standard deviation(S_RMSEP) and relative analysis error(K_RPD) were obtained. On the basis of the basic band, the optimal wavelength algorithm was used to obtain the predictive deterministic coefficient and the predictive standard deviation of the model after adding the wavelength. Through analysis, this article finally determined to add 42 preferred wavelengths. The experimental results show that the prediction deterministic coefficient of the model after adding the preferred wavelength is increased from 0.760 4 without the preferred wavelength model to 0.991 1, the S_RMSEP is reduced to 1/5, and the K_RPD is increased by 5 times.