基于BP神经网络的左右手击键动作的意识任务识别

目的研究基于神经网络的左右手运动的意识任务识别方法,探讨神经网络在脑机接口中的作用。方法在特征提取的基础上,设计3层BP神经网络。选用对数Sigmoid函数,实现输入到输出的非线性映射;采用梯度最速下降算法训练神经网络。结果应用BP神经网络和线性分类器分别对测试样本进行意识任务识别。以脑电信号两个频段的功率谱以及击键前-100￣-50ms和-50￣0ms均值组成特征向量。应用线性分类器,对测试样本的识别正确率为71%,采用本文设计的BP神经网络,识别正确率为84%。结论BP神经网络是意识任务识别的有效方法,在基于脑电信号的脑机接口中有良好的应用前景。

Recognition of mental task of pressing key about left-right hand based on BP neural networks

[Objective] To study the algorithms of recognition of mental task of left-right hand based on BP neural network, and to explore the function of neural network in brain -computer interface (BCI). [Methods] After extracting the features of electroencephalogram (EEG), a BP neural network of three layers BP was designed. The nonlinear mapping from input to output was realized by choosing logarithmic Sigmoid function in the hidden layer of the network. Steepest descent backpropagation (SDBP) was adopted to train the network. The network and linear discriminant analysis (LDA) were applied respectively to recognize the mental task. The input feature vector consisted of two frequency bands of power spectrum and the average value of time bands from -100 to -50ms and from -50 to 0ms. [Result] The recogniton accuracy was up to 84% by using the BP neural network as the classifier, while the accuracy was 71% by using the LDA. [Conclusion] BP neural network is an efficient method for the recognition of the mental task, and has good prospect in the application of BCI.