Pu’er Tea Aroma Discrimination Based on Electronic Nose

The tea-growing areas in Yunnan are widely distributed. The unique climatic conditions and geographical environments have given rise to distinctive regional tea aromas. In view of this, this paper takes Pu’er tea as the research object. By generating the measurement data of the electronic nose, it first conducts a comparative analysis of the aroma characteristics of Pu’er tea from three different perspectives of tea variety, state, and region through radar charts. It can be known that the aroma substances of various tea samples have both commonalities and differences. Subsequently, using analytical methods such as Principal Component Analysis (PCA), decision trees, Neural Networks, Naive Bayes, and Ensemble Learning, an in-depth study is conducted on unfermented Pu’er tea and fermented Pu’er tea in three forms: dry tea leaves, tea liquid, tea residue. The aim is to construct a classification model that can accurately distinguish Pu’er tea from different regions. The aroma components of regional samples are dimensionality reduced by using the PCA method, and then discriminant analysis is conducted. The research results show that for unfermented Pu’er tea, the Random Forest method performs best in distinguishing dry tea leaves, tea liquid and tea residue, among which tea liquid is the easiest to distinguish. The Naive Bayes classifier is particularly effective in identifying dry tea leaves samples, followed by tea liquid and tea residue. For fermented Pu’er tea, the Random Forest method also yields the best classification results. Among the three forms, tea residue of fermented tea is the easiest to distinguish, especially when using Decision Trees, Bagging, and AdaBoost ensemble methods. Neural Networks show relatively consistent performance in the three forms. Overall, Random Forest proves to be a highly practical and effective method for distinguishing Pu’er tea from the three study regions.