Classification of Eastern Black Sea Tea Using E-Nose: A Deep Learning-Based CNN Approach

The Black Sea tea is of significant cultural and economic importance, with its distinct characteristics influenced by factors like soil, climate, and cultivation practices, but its classification often faces challenges due to taster variability and sensory subjectivity. This paper presents a classification approach for different tea types from the Black Sea Region using an electronic nose (e-nose) system. Sensor data obtained from eight Metal Oxide Semiconductor (MOS) sensors were analyzed and classified using a Convolutional Neural Network (CNN) model. The study explores the performance of raw sensor data and data transformed through Fast Fourier Transform (FFT). The results indicate that the CNN model achieves high classification accuracy, with the highest accuracy reaching 90.64% using the sensor combination of 2, 4, 6, 7, and 8 with 5-fold cross-validation. Additionally, the 7th sensor alone demonstrated high performance with an accuracy of 90.51%. The use of FFT-transformed data further enhanced classification accuracy, with the highest being 92.06% for 3-fold cross-validation. The results suggest that the selected sensors, especially the 7th sensor, offer effective and economically viable solutions for tea classification.