【摘要】 随着人工智能与深度学习的发展，基于深度学习的多通道脑电信号的情绪识别研究逐渐受到关注，但多通道脑电情绪识别信号复杂且各通道重要性一致，并不能高效且有针对性地进行脑电情绪识别。为此，该文提出一种基于缩放卷积层和脑电通道增强模块的情绪识别方法，能直接在脑电物理通道上进行增强学习。首先，通过缩放卷积层提取多通道脑电情绪信号的类时频特征；然后，通过脑电通道增强模块对所有脑电物理通道重新赋予不同的重要性；最后，利用卷积神经网络对情绪进行分类。该方法能够融合多通道脑电信号的时间和频率信息，同时，通过输出各脑电通道的重要性，探究不同情绪维度与脑电通道之间的关系。在DEAP数据集上进行了实验验证，不同脑电通道对情绪识别任务的重要性存在差异，其中，额叶区和枕叶区的C4、 P4、 P3、 PO4、 F7 5个脑电通道重要性相对较高，该情绪识别方法在愉悦度、唤醒度和支配度3个情绪维度上的识别准确率也均有提升。更多还原

【Abstract】 With the development of artificial intelligence and deep learning, deep learning-based research on multi-channel EEG signals for emotion recognition is gaining attention. However, multi-channel EEG emotion recognition signals not only have redundant information but also are complex, which means they can not efficiently enhance the performance of EEG emotion recognition. Therefore, an emotion recognition method based on the scaling convolution layer and the EEG channel enhancement modules is proposed to perform enhanced learning directly on the physical channels of EEG. Firstly it extracts the time-frequency-like features of multi-channel EEG emotional signals by the scaling convolution layer module. Then it reassigns different importance to all EEG channels by the EEG enhancement module. Finally it uses the convolutional neural network to classify the emotions. The method is able to fuse temporal and frequency information of multi-channel EEG signals, and it explores the relationship between different emotional dimensions and EEG channels by outputting the importance of each EEG channel. Experiments conducted on the DEAP dataset showed that the importance of different EEG channels for the emotion recognition task are different and the five EEG channels(C4, P4, P3, PO4 and F7) in the frontal and occipital regions are of relatively high importance. Meanwhile, the method also improved the accuracy of the three emotional dimensions of valance, arousal and dominance respectively.更多还原