EEG Signals Classification Related to Visual Objects Using Long Short-Term Memory Network and Nonlinear Interval...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Understanding brain function relies on decoding neural activity.
• Electroencephalography (EEG) signals offer insights into brain states but are complex and noisy.
• Classifying visual objects from EEG data presents challenges due to signal nonlinearity and individual variability.

Purpose of the Study:

• To develop and evaluate a new method for classifying EEG signals associated with visual objects.
• To address uncertainties in EEG data, including nonlinearity, noise, and limited sample sizes.
• To improve the accuracy of mapping image features to EEG signal features for better brain activity decoding.

Main Methods:

• Utilized ResNet for image feature extraction.
• Employed a Long Short-Term Memory (LSTM) network for EEG signal feature extraction.
• Applied nonlinear interval type-2 fuzzy regression (NIT2FR) to map image features to EEG signal features.
• Evaluated performance using the Stanford database and metrics like accuracy, precision, recall, and F1 score.

Main Results:

• The LSTM network alone achieved 55.83% accuracy in categorizing images from raw EEG data.
• NIT2FR combined with an SVM classifier achieved 68.05% accuracy, outperforming other regression and classification methods.
• NIT2FR demonstrated superior performance in handling high-uncertainty environments, showing a 6.03% accuracy improvement over prior studies.
• NIT2FR achieved precision, recall, and F1 scores of 68.93%, 68.08%, and 68.49%, respectively.

Conclusions:

• NIT2FR is highly effective for classifying EEG signals in environments with significant uncertainty.
• The proposed method enhances the understanding of how brain activity encodes visual information.
• This approach offers a robust solution for brain-computer interfaces and neurological studies.