A Lightweight Deep Learning Model for EEG Classification Across Visual Stimuli

Visual stimuli have a multifaceted impact on brain activity, yet the nuanced differences in how various types of stimuli affect electroencephalogram (EEG) signals are still under investigation. This study endeavors to classify EEG signals in response to a range of visual stimuli by crafting a lightweight deep learning model. Utilizing the 170 EEG dataset from the ERP core, which encompasses recordings from 40 healthy participants exposed to roughly 10-minute sessions of randomly presented sets of normal and scrambled photographs. Each set consisted of images portraying either normal or scrambled representations of faces and cars, encapsulating four unique visual stimuli. By harnessing the EEG data from the 40 participants, our Reset18-based model attained an impressive average classification accuracy of 98.13% for face images and 97.81% for car images, significantly surpassing the performance of traditional machine learning models. otably, this marks the inaugural application of the Reset18 model to the 170 dataset classification experiment within the ERP core. The findings of this study enrich our comprehension of the brain’s distinct cognitive responses to these stimuli and the manifestation of these differences in EEG signals. The successful deployment of this model paves the way for furthering the exploration and development of brain-computer interface technologies.