论文部分内容阅读
常见的近红外光谱分析技术,一般将欧式距离作为相似性判据,但是在很多情况下并不能真实体现样本间的相似性;同时,线性回归模型无法克服校正样本集光谱数据中非线性以及样本差异大而导致的精度降低问题。针对上述问题,本文首次将光谱信息散度引入到局部建模算法中,以未知样本光谱与校正样本光谱间的光谱信息散度作为样本相似性判据,选取一定数量与待测样本最相似的校正样本组成局部校正子集,建立局部偏最小二乘模型。为了验证算法的有效性,将现有的全局建模算法、基于样本光谱间欧式距离的局部建模算法与本文提出的基于光谱信息散度的局部建模算法应用于猪肉近红外光谱标准数据集。实验结果表明:本文新方法的预测均方根误差(RMSEP)分别比现有的两种算法降低了22.8%与48.7%,克服猪肉近红外光谱的非线性和差异性,在近红外光谱定量分析领域具有良好的应用前景。
Common near-infrared spectroscopy techniques generally use the Euclidean distance as the criterion of similarity, but in many cases the similarity between the samples can not be truly reflected. At the same time, the linear regression model can not overcome the nonlinearity of the sample data in the calibration sample set, Differences caused by the accuracy of the problem. In view of the above problems, the paper first introduces the divergence of spectral information into the local modeling algorithm. Taking the spectral information divergence between the spectrum of the unknown sample and the spectrum of the calibration sample as the criterion of similarity of the sample, a certain quantity is selected to be the most similar to the sample to be tested The calibration samples form a local calibration subset and a partial partial least squares model is established. In order to verify the effectiveness of the proposed algorithm, the existing global modeling algorithms, local modeling algorithms based on Euclidean distance between sample spectra and the local modeling algorithm based on spectral information divergence proposed in this paper are applied to the standard data set of pork near-infrared spectroscopy . The experimental results show that the RMSEP of this new method is reduced by 22.8% and 48.7% respectively compared with the existing two algorithms, overcomes the nonlinearity and difference of near-infrared spectroscopy of pork, and the quantitative analysis by near-infrared spectroscopy The field has good application prospects.