以乐山产正品竹叶青、劣质竹叶青和峨眉山毛峰为研究对象, 提出了一种基于近红外光谱的不同茶叶品种分类识别算法。该算法采用多元散射校正(Multiplicative Scatter Correction, MSC)对3种茶叶的近红外光谱数据进行预处理, 最大限度地扣除光谱数据中的随机变异; 再采用主成分分析算法(Principal Component Analysis, PCA)对预处理后的光谱数据进行降维, 去除冗余; 接下来进行线性判别分析(Linear Discriminant Analysis, LDA), 进一步提取特征; 最后采用K_近邻算法(K_Nearest Neighbor, KNN)对LDA结果的前两个特征进行分类, 从而达到对茶叶进行定性分类的目的。实验结果表明, 该算法能有效地对3种茶叶进行分类, 正确识别率达到100%。本研究为不同品种茶叶的分类识别提供了一种新思路。

Taking the real Zu Yeqing tea produced in Leshan, the inferior Zu Yeqing tea and the Maofeng tea produced in Emei Mountain as the research objects, a classification algorithm for different kinds of tea based on near infrared spectroscopy is put forward. The algorithm uses Multiplicative Scatter Correction (MSC) to preprocess the near infrared spectral data of the above three kinds of tea for removing the random variation in the spectral data maximally. Then, it uses Principal Component Analysis (PCA) to reduce the dimensionality of the spectral data for removing redundant. Next, it carries out Linear Discriminant Analysis (LDA) for further feature extraction. Finally, it uses the K_Nearest Neighbor algorithm to classify the first features in the LDA result so as to realize the qualitative tea classification. The experimental results show that this algorithm can classify the above three kinds of tea effectively. Its correct recognition rate is up to 100%. This study provides a new idea for the classification of tea.