基于EEGNet的脑电情绪分类应用研究

An Application Research of EEGNet for Emotion Classification Based on Electroencephalography

针对特定脑电信号数据集的情绪分类问题,研究紧凑型的卷积神经网络EEGNet在不同脑电数据集上的能力与效果,并通过在不同的脑电数据集上对EEGNet进行训练与调试,实现单模态脑电数据集的情绪分类。首先,介绍紧凑轻量型卷积神经网络EEGNet结构在时空数据集上的强大处理能力,提出在对EEG信号特征进行编码时的有效性假设。其次,介绍两种经典的脑电公开数据集SEED和SEED-IV,设计针对性的预处理方法、基于EEGNet的情绪分类实验并与其他经典分类方法进行了比较分析。最终,经过在SEED和SEED-IV数据集上的多轮测试,分别得到了85.3%和73.3%的分类准确率,验证了EEGNet在基于脑电信号的情绪分类任务中具有较好的健壮性与准确率。

This study investigates the capabilities and effectiveness of EEGNet,a compact convolutional neural network,for the task of emotion classification using specific EEG datasets.By training and fine-tuning EEGNet on different EEG datasets,the classification of emotions from single-modal EEG data is achieved.Firstly,the study introduces the powerful processing capability of the compact convolutional neural network EEGNet in handling spatio-temporal data and proposes the hypothesis of its effectiveness in encoding EEG signal features.Secondly,two well-known publicly available EEG datasets,SEED and SEED-IV,are introduced.Targeted preprocessing methods are designed,and emotion classification experiments based on EEGNet are conducted,comparing them with other classical classification methods.Finally,through multiple rounds of testing on the SEED and SEED-IV datasets,classification accuracies of 85.3%and 73.3%are achieved,respectively,demonstrating the robustness and accuracy of EEGNet in emotion classification tasks based on electroencephalography signals.