基于GST-ECNN的运动想象脑电信号识别方法

在对脑电信号的解码研究中,存在着现有时频分析方法对高频信号处理能力有限,多通道信号信息冗余,常用卷积神经网络分类器ReLU激活函数受学习速率的影响较大,对不同层采用相同的正则化很难获得满意结果等问题。为此,提出了一种基于广义S变换特征提取和增强卷积神经网络分类相结合的方法,同时提出一种结合Relief算法和向前选择搜索策略的包裹式方法进行通道选择。结果表明,提出的方法利用较少的信号通道,具有更强的特征提取和分类的能力,在第Ⅳ届BCI的数据集I上取得最高98.44±1.5%的分类准确率,高于其他现有算法。该方法良好的分类性能不仅减少了计算消耗,也有效提高了分类准确率,对脑电信号特征提取和分类具有一定的参考意义。

Motor Imagery EEG Signal Recognition Method Based on GST-ECNN

In the research on the decoding of EEG signals, there are existing time-frequency analysis methods that have limited high-frequency signal processing capabilities, multi-channel signal information redundancy, and the ReLU activation function of the commonly used convolutional neural network classifier is greatly affected by the learning rate. It is difficult to obtain satisfactory results with the same regularization for different layers. To solve the above problems, a method based on the combination of generalized S-transform feature extraction and enhanced convolutional neural network classification is proposed. At the same time, a wrapping method combining Relief algorithm and forward selection search strategy is proposed for channel selection. The results show that the proposed method uses less signal channels and achieves better ability of feature extraction and classification. The highest classification accuracy of 98.44±1.5% is obtained in the fourth BCI dataset I, which is higher than other existing algorithms. The good classification performance of this study not only reduces the calculation consumption, also effectively improves the classification accuracy, which has a certain reference significance for EEG feature extraction and classification.