As a famous green tea from China, Longjing tea produced in different geographical regions exhibits similarities in appearance, aroma, and taste. Traditional identification methods face challenges such as high equipment costs, cumbersome operations, and time consumption. This article proposes a novel method for the rapid detection of Longjing tea origin by using an electronic tongue (E-tongue) combined with Denoising Diffusion Probabilistic Model (DDPM) and Convolutional Neural Network (CNN)-Transformer hybrid model. First, an E-tongue device is utilized to collect the taste fingerprint information of Longjing tea samples. To explore richer information features in E-tongue signals, the Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), and Markov Transition Field (MTF) transformations are applied to convert the original E-tongue signals into time-frequency matrices and then embedded into a 3-channel composite spectrogram. Subsequently, DDPM is used for data augmentation on the composite spectrograms to increase both the training sample volume and diversity of the dataset. A CNN-Transformer hybrid model is then proposed to extract comprehensive features from spectrograms and realize pattern recognition. Experimental results show that the proposed model demonstrates superior recognition accuracy and stronger generalization ability. The model achieves an accuracy, precision, recall, and F1-score of 98.20%, 98.22%, 98.20%, and 0.9819, respectively. This study provides a new method for the rapid identification of Longjing tea origins and has broad potential applications for other tea varieties.