基于DBN结合SVM的脑电信号识别研究

EEG Signal Identification Based on DBN and SVM

针对多类脑电信号识别率不高的问题,提出一种深度信任网络（Deep Belief Network,DBN）与支持向量机（Support Vector Machine,SVM）结合的方法。将采集的想象左手、右手、双脚以及舌头运动的四类脑电信号数据作为训练样本训练DBN网络,以得到其最优参数值。用训练好的DBN网络进行特征提取,采用SVM对提取的特征进行分类,在MATLAB上对该算法进行仿真实验测试。实验结果表明,使用该方法分析四类运动想象脑电信号具有较高的识别率,证明了该方法的有效性。

The paper proposes a method of combining deep belief network（DBN） with support vector machine（SVM） for analyzing many kinds of electroencephalogram（EEG） signals against the existing issue of low recognition rate. To get the optimal parameter values, the EEG data collected through the imagination of the left hand, right hand, foot and tongue movement of four types of EEG signals are used as the training sample for training the DBN. This study uses the trained DBN network for feature extraction, and these features extracted will be classified by the SVM classification. The algorithm is tested through the simulation experiment by using MATLAB. The experimental results show that using this method to analyze four types of imagination EEG signals has higher recognition rate, which proves the effectiveness of the proposed method.