近红外光谱技术与偏最小二乘法及模糊聚类法相结合的糖品种分类方法

Discrimination of Varieties of Sugar Based on Partial Least Squares and Fuzzy Clustering Methods

提出了一种近红外光谱技术与偏最小二乘法及模糊聚类法相结合的可用于快速无损鉴别糖品种的新方法。采用近红外光谱仪获取了白砂糖、木糖醇、麦芽糖和葡萄糖等四种糖类别各30个样本的光谱漫反射特征曲线。运用偏最小二乘法提取了糖分类与特征值,并将提取到的经过归一化处理的11种主成分结果作为模糊聚类模型的建模参数。设定聚类数为4,建立模糊聚类模型,并对40个未知样本进行了预测。预测结果的准确率达到100%,说明本文提出的方法对于糖类别具有很好的分类和鉴别能力,同时也为光谱分析技术在对品种的快速、无损分类与识别中的应用提供了新的思路。

A new method which combines Partial Least Squares （PLS） with a near infrared spectroscopy is proposed for the nondestructive discrimination of the varieties of sugar. A near infrared spectrometer is used to obtain the diffusion spectral characteristic curves from the samples of white granulated sugar, xylitol, maltose and glucose. Then, the PLS is used to derive the variety and characteristic values of the sugar. The derived eleven main components which are normalized are used as the parameters for establishing a fuzzy clustering model. By setting four clusters, the fuzzy clustering model is established and is used to predict forty unknown sugar samples. The prediction accuracy is up to 100 %. This shows that the new method has a good ability to fast discriminate the variety of sugar.