Emotion Recognition from Multi-Channel EEG through Parallel Convolutional Recurrent Neural Network-2018年EI会议
EEG信号的预处理方法(DEAP数据集)
DEAP数据集【2】
背景
大多数基于EEG的情绪识别研究直接使用EEG trials信号,没有考虑到基线(无刺激的EEG信号)的作用
创新点
针对DEAP数据集提出了一种EEG脑电数据预处理新方法:baseline signals are considered.
方法
BaseMean为基线数据平分为N个segment的均值
BaseRemoved为trial 数据按照BaseMean数据减去BaseMean后的叠加
实验参数设置
节律信号:4 band
window length :1s
segment:60/trial
length:128
sample:40x60=2400/subject
label one-hot-code:阈值5
结论与不足
Experimental results have shown that the pre-processing approach can improve the accuracy by 32% approximately.
我的疑问:虽然从结果上面看,能取得较好的实验效果,但是这样做有什么依据?或者数学原理。不过,已经有很多人在此基础上进行应用和实践。
Continuous Convolutional Neural Network with 3D Input for EEG-Based Emotion Recognition-2018
window length :1s
segment:60/trial
length:128
sample:40x60=2400/subject
label one-hot-code:阈值5
EEG-based emotion recognition using 4D convolutional recurrent neural network
节律信号:4 band
window length:0.5s
segment:120/trial
length:64
sample:40x120=4800/subject->40x(T/window length)x30
T :2s ->sample:4800/4=1200
label one-hot-code:阈值5
三篇论文同样介绍了脑电数据转变为适用于Deep learning的2D map
参考
参考文献
[1]: Continuous Convolutional Neural Network with 3D Input for EEG-Based Emotion Recognition
[2]: Emotion Recognition from Multi-Channel EEG through Parallel Convolutional Recurrent Neural Network
[3]: EEG-based emotion recognition using 4D convolutional recurrent neural network
[4]: DEAP数据集申请地址
[5]: DEAP数据论文:DEAP: A Database for Emotion Analysis using Physiological Signals