基于卷积自注意力机制的KAN神经网络对脑机接口视觉电刺激信号分类OA

Objective To analyze the numerous challenges encountered in the classification of P300 visual evoked potential signals and to investigate novel prospective solutions.Methods This study introduced a novel KAN neural network model,augmented with a convolutional self-attention mechanism.This innovative approach proficiently captured the global feature information of P300 signals.Furthermore,the incorporation of the KAN layer enhanced its capability to manage nonlinear data effectively.To validate the efficacy of this model,experiments were executed utilizing the widely recognized brain computer interface Competition Ⅲ Challenge 2004 dataset.The performance of our proposed model was subsequently juxtaposed with modern P300 VEP classification techniques.Results The proposed model exhibited superior classification accuracy on the validation set,achieving an impressive 100.0%accuracy in P300 signal classification.This performance surpassed that of VGG-16,which achieved 98.9%,and ResNet-18,which achieved 99.0%.Furthermore,in experiments involving fast gradient sign methed attacks,the model maintained an accuracy of 82%.Conclusion This study presents a novel methodology for the classification of P300 VEP signals,which can also be applied to similar tasks in brain signal processing.This offers fresh research perspectives and contributes to the progression of brain signal research.