GTMH-TasteNet: Advanced Deep Learning for sEMG-Based Taste Sensation Recognition

Medical diagnostics and advanced virtual reality recognize and categorize human taste sensations. This research will examine the recognition of five fundamental human taste sensations—sour, sweet, bitter, salty, and umami—alongside a no-stimulation physiological condition using GTMH-TasteNet, an advanced deep learning framework. Data was obtained through the sEMG device, which can capture signals from muscle activity during taste perception. After collecting the dataset, it was preprocessed involving four steps: detrending, high-pass filtering, adaptive notch filtering, and segmenting the signal into 1-second intervals. Data augmentation methods were employed to improve the model's resilience and generalizability. The GTMH-TasteNet architecture was applied to the processed dataset to extract features, improve interpretability, and emphasize the most significant data elements. The model demonstrated commendable performance, achieving an accuracy of 85.2% in categorizing taste experiences. The results validate that GTMH-TasteNet is highly proficient in detecting taste from sEMG, with potential implications for customized medicine, virtual reality, and multisensory interaction. It presents a scalable, interpretable, and accurate system for identifying taste sensations, advancing human-computer interaction, and taste-driven healthcare.