Electronic Nose Coupled with Deep Learning Techniques for Tea Quality Assessment

Currently, the evaluation of tea quality primarily relies on sensory methods such as the observation of tea appearance, smelling the aroma of the tea liquor, and tasting it. Alternatively, instrumental quantitative methods like chromatography and spectroscopy are also employed. However, these methods tend to be highly subjective, prone to fatigue in the evaluator’s sensory organs, lack real-time capabilities, and are susceptible to measurement errors. Given the crucial role of the aroma emitted by the tea liquor in assessing its quality, this study employs a gas sensor arrays, commonly known as electronic noses, with deep learning techniques to assess tea quality. Specifically, a stacked sparse autoencoder was utilized to build the neural network. The entire network was trained using (Multi-Layer Perceptron) MLP with labeled data. Additionally, a Softmax regression function was integrated into the output layer of the MLP. The proposed method achieved a classification accuracy of approximately 90%, outperforming the average accuracy of the SVM method for each tea type.