作者:陈宇，胡秀秀，王胜
Authors:CHEN Yu,HU Xiuxiu,WANG Sheng
摘要:目前脑电信号(EEG) 的抑郁症识别方法主要采用单一特征提取方法,无法覆盖多域特征信息,导致现有模型分类性能不高 ,因此提出了一种多域特征结合CBAM 模型(CNN-BiLSTM-attention mechanism) 的抑郁症识别算法 。首先利用连续小波变换(CWT)提取时频域特征,并结合脑电电极空间信息构成2D特征图像 ,共同保留脑电的空间、时间和频率信息 ;然后使用卷积神经网络(convolutional neural network,CNN)提取空间和频域特征 ,再输入双向长短时记忆网络(bidirectional long and short-term memory,BiLSTM) 以捕获时间信息 ;最后结合注意力机制 (attention mechanism,AM) ,对网络提取的多域特征赋予不同的权重,以筛选出更具代表性的抑郁特征,从而提高识别抑郁症的准确性 。实验表明,本文提出的基于 CBAM 模型的抑郁症识别算法在公共数据集上取得了 99. 10% 的准确率,为脑电信号抑郁症识别研究提供了一种有效的新方法。
Abstract:At present, the electroencephalogram (EEG) identification method for depression mainly uses a single feature extraction method, which cannot cover multi-domain feature information, resulting in poor classification performance of the existing model. Therefore, this paper proposes a depression recognition algorithm based on multi-domain features combined with CBAM model (CNN- BiLSTM-Attention Mechanism). Firstly, the continuous wavelet transform (CWT) is used to extract time-frequency domain features, and combined with the spatial information of EEG electrodes to form a 2D feature image, which jointly retains the spatial, time and frequency information of EEG; then the convolutional neural network (CNN) is used) to extract spatial and frequency domain features, and then input bidirectional long and short-term memory ( BiLSTM) to capture time information; finally combined with attention mechanism (AM) , different weights are assigned to the multi-domain features extracted from the network, enabling the selection of more representative depressive features, thereby improving the accuracy of identifying depression. Experiments show that the depression recognition algorithm based on the CBAM model proposed in this paper has achieved an accuracy rate of 99. 10% on the public data set, which provides an effective new method for the research on depression recognition of EEG signals.

PDF全文下载地址:

可免费Download/下载PDF全文