基于特征层融合的EEG-NIRS识别方法研究

Research on EEG-NIRS Recognition Method Based on Feature Layer Fusion

针对目前卷积神经网络对时序序列识别率较低、单模态数据信息量不充足等问题,提出了一种基于特征层融合的卷积神经网络(CNN)和双向长短期记忆(BiLSTM)网络组合的方法。首先在毒品成瘾患者的脑电图和近红外光谱数据上,利用全新的CNN和BiLSTM组合网络分别对双模态数据进行特征提取,将最后一层BiLSTM的输出作为特征并进行特征串联,然后对串联特征进行分类识别。特征融合实验结果表明,文章提出的CNNBiLSTM模型的分类效果最高准确率达到97.3%,并且双模融合方法进一步提高了分类准确率。

Aiming at the low recognition rate of temporal sequences by convolutional neural networks and insufficient information in single mode data,this paper proposes a method of combining Convolutional Neural Networks(CNN)and Bidirectional Long Short-Term Memory(BiLSTM)networks based on feature layer fusion.Firstly,on the EEG(Electroencephalogram)and near-infrared spectral data of drug addiction patients,the new CNN and BiLSTM combination network are used to extract features from the bimodal data.Then,the output of the last layer of BiLSTM is used as features for feature concatenation.Then,the concatenated features are classified and recognized.The experimental results of feature fusion show that the proposed CNN-BiLSTM model achieves the highest classification accuracy of 97.3%,and bimodal fusion method further improves the classification accuracy.