A Classification Model Based on SEB-CNN for P300 Signal Recognition of P300 Speller

The P300 speller is a canonical application of brain-computer interface technology, where the critical challenge lies in improving the classification performance of P300 signals - particularly under conditions of limited stimulus repetition. This study proposes an enhanced convolutional neural network architecture that incorporates Squeeze-and-Excitation Blocks (SEB-CNN) to address this challenge. Building upon the conventional CNN framework (C-CNN), our approach introduces two key innovations: (1) spatial convolution mechanisms for feature extraction and (2) temporal attention mechanisms for sequence modeling. This dual mechanism design significantly enhances the model's adaptive capability and cross-subject generalization performance when processing P300 EEG signals. Experimental results demonstrate that the SEB-CNN model achieves markedly superior classification accuracy compared to traditional CNN approaches, with particularly notable improvements observed in low-repetition stimulus conditions (p<0.01).