Wine discrimination based on multi-sensor fusion of GASF and Mel spectrogram features using an enhanced...

This study presents a novel multi-sensor fusion strategy for discriminating wines made from eight different raw materials using identical brewing processes. Aroma and taste signals were collected using a broad-spectrum electronic nose and noble metal electrodes, forming olfactory and gustatory fingerprints. To enhance feature representation, one-dimensional time-series sensor data were transformed into two-dimensional RGB images using Gramian Angular Summation Field (GASF) and Mel spectrogram algorithms, preserving temporal and spectral information. These image features were then input into an Enhanced-EfficientNet-B0 deep learning model, achieving a validation accuracy of 92.17 %. The results demonstrate that the integration of multi-sensor data with advanced signal transformation and deep learning enables accurate and efficient classification of wine raw materials.