基于深度神经网络的脑控技术研究

Research on brain control technology based on deep neural network

提出一种通过对脑机接口数据进行时域、频域划分的运动想象信号识别和分类系统。实验采集了4名被试者的前进、停止、左转、右转运动想象的脑电信号数据作为实验数据集,提出了两种深度学习模型,门控循环单元神经网络(GRU)和一种混合深度学习框架1DCNN-GRU模型来进行信号识别准确性对比。并对未处理的脑电信号进行快速傅里叶变换提取数据重要特征值,对实验数据集进行6∶2∶2训练-验证-测试分割。

The subject proposes a system for recognizing and classifying motor imagery signals by dividing the brain-computer interface data into time and frequency domains.The EEG signal data of forward,stop,left-turn,and right-turn motor imagery of four subjects were collected as the experimental dataset,and two deep learning models were proposed to compare the signal recognition accuracy,the two deep learning models are:gated recurrent unit neural network(GRU)and a hybrid deep learning framework 1DCNN-GRU model.The unprocessed EEG signals were also subjected to Fast Fourier Transform to extract the important feature values of the data,and the experimental dataset was subjected to 6:2:2 train-validate-test segmentation.