基于卷积神经网络的脑电信号上肢运动意图识别

Recognition of upper limb motion intention of EEG signal based on convolutional neural network

为了在脑机交互中能够对运动意图进行识别,使设备能够预判人的行为动作并提前作出反应,脑电(EEG)信号运用学习过程去解码,并建立识别机制.针对传统生物信号模式识别模型中手动提取特征可能会产生信息损失的问题,引入深度学习的卷积神经网络(CNN),并和目前广泛使用的两种特征提取方法使用BP神经网络分类进行对比.结果显示,CNN在左、右手2分类动作和单手3分类动作中,提高识别精度分别约为4%和8%,增加了动作预测的可靠性.通过对上肢运动意图识别的讨论,可以更好地进行脑机交互控制,并加深对中枢神经信号与手部动作关系的理解.

Electroencephalogram（EEG）signal was decoded using the learning process and the recognition mechanism was established in order to identify the motion intention in the brain-computer interaction and make device predict the behavior of the person and respond in advance.In the traditional biological signal pattern recognition model,the manual extraction of features may cause the problem of information loss.Convoluted neural network（CNN）of depth learning was introduced to identify the motion intention based on the EEG signal and compared with the two feature extraction methods widely used at present that use BP neural network.Results showed that CNN improved the recognition accuracy about 4%and 8%in the experiments of left-right hand 2classification and one-hand 3classification,and increased the reliability of the prediction.The brain-computer interaction control can be better conducted through the discussion of the upper limb motor intention recognition.The understanding of the relationship between central nervous signals and hand motions was further enhanced.