Epilepsy is a common chronic neurological disorder marked by seizures of unknown origin. Electroencephalography (EEG) is crucial for diagnosing and managing epilepsy due to its cost-effectiveness and portability, compared to advanced neuroimaging methods like MEG, fMRI, MRS, and SPECT. This study proposes a novel method to categorize six types of EEG votes using deep learning techniques. We employed 1D convolutional neural networks (1D CNN) and vision transformers (ViT), enhancing model performance with pseudo-labeling (PL) and ensemble learning. The dataset, annotated by clinical experts and scientific collaborators, was divided into high-quality and low-quality categories based on expert participation. The EEGNet-ViT model with ensemble and pseudo-labeling achieved a Kullback-Leibler divergence (KLD) of 0.3362 and an accuracy of 74.22%. This approach leverages the strengths of both 1D CNN and ViT models, reducing variance and improving the accuracy of EEG signal analysis.