P300拼写器是允许用户使用脑电图(Electroencephalogram,EEG)输入的脑机接口(Brain-Computer Interface,BCI)系统,高传输、高准确率地检测P300信号对于提高P300拼写系统的性能非常重要.针对P300脑电信号特征提取方式信噪比低、识别困难等特点,提出一种基于批量归一化和残差块的卷积神经网络(Convolutional Neural Network,CNN)模型,可以在模型训练中保留重要特征的同时加快模型损失的收敛速度.采用分类精度、AUC(Area under Curve)、准确率、召回率、F1-score等指标来验证所提模型的有效性,并与其他方法进行了对比实验.实验结果显示,与传统的CNN算法相比,所提模型的分类精度提升6％,损失函数的收敛速度也有提升;与传统机器学习方法相比,所提模型的各项评价指标都优于传统算法.证明该算法是提高P300拼写器性能的有效方法.
P300 EEG signal decoding based on a convolutional neural network
The P300 speller,a brain-computer interface(BCI)system enabling users to input via electroencephalogram(EEG),plays a vital role in achieving high transmission and accuracy in detecting P300 signals to enhance the performance of P300 speller systems.To address such challenges as low signal-to-noise ratio and difficulty in feature extraction from P300 EEG signals,this study introduces a convolutional neural network(CNN)model incorporating batch normalization and residual blocks.This model aims to preserve critical features during training while expediting the convergence of the model's loss function.Performance evaluation of the proposed model and comparative algorithms is conducted using metrics such as classification accuracy,AUC,precision,recall and F1-score.Results indicate that compared to traditional CNN algorithms,the proposed model achieves a 6％increase in classification accuracy and exhibites improved convergence speed of the loss function.Furthermore,when compared to traditional machine learning methods,the proposed model outperformes across all evaluation metrics.Thus,this algorithm presents an effective approach for enhancing the performance of P300 speller systems.
万方数据
维普
