一种近红外混合光谱定性模型构建方法

A Qualitative Model Construction Scheme for Hybrid NIR Spectroscopy

【目的】为了提高模型的通用性,确保在实际的近红外定性模型的构建过程中能够提取近红外光谱更多的有效信息,建立一套稳定的、能够适应外界光谱变化的近红外定性模型。【方法】研究小组采用基于加速遗传算法的投影寻踪方法构建光谱的定性模型。通过混合多批次、多产地、多等级、多部位的光谱作为训练样本集,来提高模型的通用性。然后计算投影特征向量和其他光谱集的投影值,并分别进行比较。【结果】建模集与同批次光谱集的均值的变异系数为0.63%,与不同批次光谱集的均值的变异系数为1.54%,并且,混合不同批次样本前后,不同批次样本集投影值均值的变异系数从1.54%降至0.26%。【结论】通过混合不同批次的光谱,可明显提高模型在不同批次光谱的定性表现。因此,使用投影寻踪方法建立光谱的定性模型具备很好的可行性。

[Objective]To improve the versatility of the model,ensure that more effective information from the near infrared spectrum can be extracted during the construction of the actual near infrared qualitative model,and establish a stable near infrared qualitative model that can adapt to external spectral changes.[Method]The research team used the projection pursuit method based on accelerated genetic algorithm to construct a qualitative model of the spectrum.The versatility of the model is improved by mixing multiple batches,multiple origins,multiple levels,and multiple parts of the spectrum as a training sample set.Then calculate the projection value of the projection feature vector and other spectral sets,and compare them respectively.[Result]The coefficient of variation of the mean values of the modeling set and the same batch spectral set was 0.63%,and the coefficient of variation of the mean values of the different batch spectral sets was 1.54%.Moreover,before and after mixing different batch samples,the coefficient of variation of the mean values of the projection values of the different batch sample sets decreased from 1.54%to 0.26%.[Conclusion]By mixing different batches of spectra,the qualitative performance of the model in different batches of spectra can be significantly improved.Therefore,using projection pursuit method to establish qualitative models of spectra has good feasibility.