Adaptive Identification of Food Sweetness Concentration: An Electroencephalogram Feature Classification Network Under...

Area of Science:

• Neuroscience
• Food Science
• Machine Learning

Background:

• Consumer perception of food sweetness is crucial for food formulation optimization.
• Electroencephalogram (EEG) signals reflect brain activity changes in response to taste stimuli.
• Accurate identification of taste perception from EEG data remains a challenge.

Purpose of the Study:

• To develop a novel network, the EEG Feature Calculation and Classification Network (EFCC-Net), for recognizing taste EEG signals.
• To analyze brain region activation patterns associated with different sweetness concentrations using EEG topographic maps.
• To enhance the accuracy and stability of classifying EEG data corresponding to varying sweetness levels.

Main Methods:

• Collected taste-related EEG data from human subjects exposed to varying sweetness concentrations.
• Proposed an EEG Feature Calculation Module (EFCM) using convolutional kernels for local feature extraction (temporal and spatial).
• Implemented a lightweight self-attention mechanism for global feature computation and a multi-branch approach for enhanced feature extraction, forming the EFCC-Net.

Main Results:

• EFCC-Net achieved high classification performance: 96.57% accuracy, 96.58% precision, and 96.53% recall.
• Qualitative analysis of EEG topographic maps revealed distinct brain region activation patterns for different taste concentrations.
• Ablation experiments and comparisons confirmed EFCC-Net's superior performance and stability over existing methods.

Conclusions:

• The proposed EFCC-Net effectively classifies EEG signals related to sweetness perception.
• The EFCM module enhances the extraction of relevant temporal and spatial features from EEG data.
• This approach offers a robust and accurate method for objective assessment of taste perception in food science and neuroscience applications.