Combining Transformer and EEGNet for Image Perception Recognition Using Brain Signals

This work proposes a new method for visual image recognition using brain EEG data, aiming to improve the accuracy and efficiency of brain-machine interface (BMI) systems. EEG signals, captured during image recognition tasks, were processed using advanced neural network architectures, specifically Transformer and EEGNet. These models were applied to classify visual stimuli, with the EEG signals being analyzed in the context of four distinct image categories. Various preprocessing steps were implemented to ensure the highest signal quality, including noise reduction and signal normalization. The performance of the proposed method is evaluated based on classification accuracy, highlighting the strengths of EEGNet in recognizing visual patterns from brain activity. The results indicate that EEGNet significantly outperforms traditional approaches, which have implications for real-time BMI applications. This could potentially improve assistive technologies by enabling individuals with disabilities to interact with devices using only brain signals, thus contributing to developing more accessible and efficient interfaces for neuroprosthetics, communication aids, and rehabilitation technologies.