茶是一种让人喜爱的健康饮品, 不同品种的茶叶其功效和作用是不相同的。 研究出一种可靠、 简单易行、 分类速度快的茶叶品种鉴别方法具有重要的意义。 在模糊非相关判别转换(FUDT)算法和模糊C均值聚类(FCM)算法的基础上提出了一种模糊非相关鉴别C均值聚类(FUDCM)算法。 FUDCM可以在聚类过程中动态提取光谱数据的模糊非相关鉴别信息。 用FTIR-7600型傅里叶红外光谱分析仪分别采集优质乐山竹叶青、 劣质乐山竹叶青和峨眉山毛峰三种茶叶的傅里叶中红外光谱, 波数范围为4 001.569～401.121 1 cm-1。 先用多元散射校正(MSC)进行光谱预处理, 然后用主成分分析法(PCA)将光谱数据降维到20维, 再利用线性判别分析(LDA)提取光谱数据中的鉴别信息。 最后分别运行FCM和FUDCM进行茶叶品种鉴别。 实验结果表明: 当权重指数m=2时, FCM的聚类准确率为63.64%, FUDCM的聚类准确率为83.33%； FCM经过67次迭代计算实现了收敛, 而FUDCM仅需17次迭代计算就可以实现收敛。 用傅里叶红外光谱技术结合主成分分析、 线性判别分析和FUDCM的方法能快速、 有效地实现茶叶品种的鉴别分析, 且鉴别准确率比FCM更高。

Tea, as a kind of healthy drink, is loved by many people. But its function and effect vary from different varieties. Therefore, it is of great significance to find a fast, easy and simple method for the identification of tea varieties. In order to classify different tea varieties quickly and accurately, fuzzy uncorrelated discriminant c-means clustering algorithm (FUDCM) was proposed based on the fuzzy uncorrelated discriminant transformation (FUDT) algorithm and fuzzy c-means clustering (FCM) algorithm in this paper. FUDCM can extract the fuzzy uncorrelated discriminant information from spectral data dynamically in the process of fuzzy clustering. To start with, Fourier transform infrared spectroscopy (FTIR) data of three kinds of tea samples (i. e. Emeishan Maofeng, high quality Leshan trimeresurus and low quality Leshan trimeresurus) was collected using FTIR-7600 spectrometer in the wave number range of 4 001.569～401.121 1 cm-1,. Secondly, multiple scattering correction (MSC) was applied to preprocess these spectra. Thirdly, principal component analysis (PCA) was employed to reduce the dimensionality of spectral data from 1 868 to 20 and linear discriminant analysis (LDA) was used to extract the identification information of the spectral data. Finally, FCM and FUDCM were performed to identify the tea varieties respectively. The experimental results showed that when the weight index m=2, the clustering accuracy rate of FCM was 63.64% and that of FUDCM was 83.33%. After 67 iterations, FCM achieved convergence while FUDCM did that after only 17 iterations. Tea varieties could be quickly and efficiently identified by combining FTIR technology with PCA, LDA and FUDCM, and the identification accuracy of FUDCM was higher than that of FCM.