茶作为世界最受欢迎的三大饮料之一, 不仅能够提神醒脑, 而且还有帮助消化和降低血压等作用。 随着人们对茶叶品质要求的日益提高, 需要对不同品种的茶叶实现准确的鉴别分析以防止茶叶市场里茶叶品牌名不副实和以次充好等现象的发生。 为实现对茶叶快速精准的鉴别分析, 设计了一种综合采用傅里叶近红外光谱和新的模糊极大熵聚类(FEC)分析算法的茶叶品种鉴别系统。 传统模糊极大熵聚类分析在聚类含噪声数据时, 聚类结果往往容易出现错误, 即FEC对噪声数据敏感。 为解决这个问题, 在FEC分析算法的基础上引入可能C均值聚类分析(PCM), 提出了一种混合模糊极大熵聚类(MFEC)分析算法。 MFEC可通过迭代计算得到模糊隶属度值, 能实现对含噪声的茶叶傅里叶近红外光谱数据的准确聚类分析。 首先, 使用傅里叶近红外光谱仪(Antaris Ⅱ型)采集岳西翠兰、 六安瓜片、 施集毛峰三种安徽茶叶的傅里叶近红外光谱数据, 光谱波数范围为10 000~4 000 cm-1。 其次, 对采集到的光谱数据使用多元散射校正(MSC)进行预处理, 预处理后先用主成分分析(PCA)将光谱数据维数降至10维, 然后再用线性判别分析(LDA)对降维后的近红外光谱数据进行特征提取。 最后, 通过混合模糊极大熵聚类分析和传统的模糊极大熵聚类分析对三种茶叶的光谱数据进行聚类分析, 并对两种聚类分析算法得到的聚类准确率、 收敛速度等进行对比分析。 实验结果表明: 混合模糊极大熵聚类(MFEC)分析算法与传统的模糊极大熵聚类(FEC)分析算法相比较, 在相同的权重指数m下MFEC具有更高的聚类准确率。 在m=2条件下, MFEC的聚类准确率达到了100%, 而传统的模糊极大熵聚类在相同条件下聚类准确率仅为37.98%。 MFEC收敛过程中仅需迭代10次即可达到收敛, 而FEC需要迭代100次, 因此MFEC可以更高效的进行模糊聚类分析, MFEC相比于FEC聚类性能具有明显的优越性。 通过傅里叶近红外光谱技术, 混合模糊极大熵聚类分析结合PCA与LDA算法构建的茶叶品种鉴别系统能够高效快速的完成对岳西翠兰、 六安瓜片、 施集毛峰三种茶叶的准确分类, 为茶叶检测领域提供了一种创新的方法与设计思路, 具有一定的理论价值和良好的市场应用前景。

Tea is one of the three most popular drinks in the world. It can not only refresh the mind, but also help digestion and lower blood pressure. With the increasing advance of requirements of tea quality by people, it is necessary to achieve accurate identification of different varieties of tea to prevent the false tea brands and adulteration in the tea market from happening. In order to identify tea varieties quickly and accurately, a tea variety identification system was designed with a combination of Fourier transform near-infrared spectroscopy (FT-NIR) and a novel fuzzy maximum entropy clustering. When traditional fuzzy maximum entropy clustering (FEC) clusters the data with noise, clustering results are often prone to errors, that is to say, FEC is sensitive to noise. To solve this problem, a mixed fuzzy maximum entropy clustering (MFEC) was proposed by introducing possibilistic c-means (PCM)clustering into traditional FEC. MFEC has fuzzy membership and typicality values by iterative computing, and it can cluster FT-NIR data mixed with noise accurately. Firstly, three kinds of Anhui tea samples (i. e. Yuexi Cuilan, Lu’an Guapian and Shiji Maofeng) were prepared for FT-NIR data collection with Antaris Ⅱ spectrometer in the wave number range of 10 000～4 000 cm-1. Secondly, spectral data were preprocessed by multiple scattering correction (MSC), and then the dimensionality of the data was reduced to 10 by principal component analysis (PCA), and then the discriminant information of the data was extracted by linear discriminant analysis (LDA). Finally, MFEC and FEC were applied to perform clustering analysis on the data, respectively, and they were compared in the clustering accuracy and convergence speed. The results of this study indicated that in the condition of m=2, the clustering accuracy rate of MFEC was 100%, while that of FEC was 37.98%. MFEC achieved convergence after four iterations while FEC converged after 100 iterations. Therefore, MFEC could cluster spectral data more efficiently than FEC, and MFEC had the obvious superiority. Three types of Anhui tea samples could be classified correctly and efficiently by combining FT-NIR technology with PCA, LDA and MFEC. This method provided an innovative method and design idea for the identification analysis in the tea testing field, and it has certain theoretical value and good market application prospect.