EEG Cross-Subject Taste Classification Method: A Meta-Learning Wavelet Graph Convolutional Neural Network Under Sweet...

Area of Science:

• Neuroscience
• Sensory Science
• Machine Learning

Background:

• Traditional taste evaluation is subjective and inefficient, hindering industrial flavor detection.
• Developing objective and generalized taste recognition methods is crucial for industrial applications.
• Electroencephalogram (EEG) signals offer a potential objective measure for taste perception.

Purpose of the Study:

• To propose and validate an electroencephalogram (EEG) classification method for accurate cross-subject taste recognition.
• To develop a meta-learning wavelet graph convolutional neural network (ML-WGCNet) for sweet and bitter taste stimuli.
• To overcome the limitations of manual sensory analysis in industrial flavor detection.

Main Methods:

• EEG signals were collected from 20 subjects exposed to varying concentrations of sucrose (sweet) and quinine (bitter).
• Morlet wavelet transform decomposed EEG signals, extracting energy features from five frequency bands.
• A graph convolutional neural network (GCN) modeled spatial brain region correlations, enhanced by meta-learning for rapid subject adaptation.

Main Results:

• The ML-WGCNet achieved average accuracies of 76.03% for sweet and 77.01% for bitter taste classification.
• The model demonstrated strong performance with precision, recall, and F1-scores exceeding 75% for both taste types.
• The proposed method significantly outperformed existing mainstream EEG classification techniques in cross-subject generalization.

Conclusions:

• The proposed ML-WGCNet method provides an objective and accurate approach for cross-subject taste recognition.
• This technique has the potential to revolutionize industrial flavor detection by overcoming the subjectivity of traditional methods.
• The meta-learning framework enables rapid adaptation, making the model practical for real-world applications.