改进孪生网络的脑电信号处理方法

EEG Signal Processing Method Based on Improved Siamese Network

摘要:

针对运动想象脑机接口系统中分类准确率低的问题，提出一种改进孪生网络的脑电信号分类方法，把原孪生网络中的两个子网络扩充成三个子网络，并设计了新的学习样本采集方法和距离函数。脑电信号经过小波变换及经验模态分解，利用自相关系数筛选得到预处理后的小波分量，然后随机分割成训练集和测试集，从训练集中按照新的学习样本采集方法获得学习样本集，将其输入三个权重共享的子网络进行训练，使用新的距离函数进行相似度的对比，最后计算测试样本特征与训练集中标签为1和标签为0样本特征相似度，选择最高相似度样本标签作为该待测样本的类别。通过对国际公开BCI Competition II Data set III和The largest SCP data of Motor-Imagery数据集进行仿真，此算法分类准确率高达94.29%。与现有性能较高的算法进行对比，其有效的提高了分类准确率，能更好的进行脑电信号分类识别。

Abstract:

In order to solve the problem of low classification accuracy in motor imagination brain-computer interface system, an improved EEG signal classification method based on siamese network was proposed. Two subnetworks in the original siamese network were expanded into three subnetworks, and a new learning sample collection method and distance function were designed. After wavelet transform and empirical mode decomposition, EEG signals are screened by auto-correlation function threshold to obtain the pre-processed wavelet component. Then it is divided randomly into training set and test set, and the learning sample set is obtained from the training set according to the new learning sample collection method, and the learning sample set is input into three sub-networks with shared weights for training, and the new distance function is used to compare the similarity. Finally, the feature similarity between the test sample features and the samples labeled 1 and 0 in the training set is calculated, and the sample label with the highest similarity is selected as the category of the samples to be tested. Through The simulation of international open BCI Competition II Data Set III and The largest SCP Data of motor-imagery Data set, the classification accuracy of this algorithm is up to 94.29%.Compared with the existing algorithms with higher performance, it effectively improves the classification accuracy and can better classify and recognize eeg signals.