基于脉冲神经网络的近红外信号的分类研究

Classification of fNIRS Based on Spiking Neural Network

脉冲神经网络(Spiking Neural Network,SNN)是研究脑机接口(Brain-Computer Interface,BCI)的新方向。与传统人工神经网络不同,SNN使用离散脉冲信号传递和处理信息,模拟神经元的生物电活动,目前主要应用在脑电图(Electroencephalogram,EEG)中,而近红外信号的分类大多数以传统机器学习和卷积神经网络(Convolutional Neural Network,CNN)为主。受SNN在EEG分类应用的启发,将SNN应用于功能近红外光谱(Functional Near-Infrared Spectroscopy,fNIRS)分类任务,并采用频谱图SNN适应fNIRS数据特性。与传统方法和CNN相比,频谱图SNN在三元分类任务中表现出卓越性能,传统方法和CNN的分类准确率分别为75.7%和72.8%,频谱图SNN模型在同一任务上取得了显著的成功,分类准确率达80.49%。这表明,SNN模型可显著提高fNIRS脑机接口性能,为神经科学研究和BCI的发展提供了新的可能性。

SNN(Spiking Neural Network)is a new direction in BCI(Brain-Computer Interface)research.Unlike conventional artificial neural networks,SNNs use discrete pulse signals to transmit and process information,simulating the bioelectric activity of neurons.Currently,they are primarily applied in EEG(Electroencephalogram)analysis,while the classification of NIR(Near-InfraRed)signals mostly relies on conventional machine learning and CNN(Convolutional Neural Networks).Inspired by the application of SNN in EEG classification,this paper applies SNN to the classification task of fNIRS(functional Near-InfraRed Spectroscopy).The spectral SNN is adapted to the characteristics of fNIRS data.Compared to conventional methods and CNN,the spectral SNN demonstrates outstanding performance in a ternary classification task.Conventional methods and CNN achieve classification accuracies of 75.7%and 72.8%,respectively,while the spectral SNN model achieves significant success in the same task with a classification accuracy of 80.49%.This indicates that SNN models can significantly enhance the performance of fNIRS-based BCI,providing new possibilities for neuroscience research and BCI development.