基于长短时间记忆网络与集成学习的多通道脑电情感识别

Multi-Channel EEG Emotion Recognition Based on Long-Short-Term Memory Network and Ensemble Learning

情感识别是人机交互中一个比较关键的问题,脑电作为人生理号中重要的组成部分,是识别人情感的关键因素。由于大脑中复杂的神经元活动会导致信号的非平稳行为,利用脑电进行情绪识别是一项具有挑战性的研究,并且多通道的分析是处理脑电信号时需要考虑的重要问题。本文提出了一种基于长短时间记忆网络与集成学习的多通道脑电情感识别模型,通过研究选取对情绪反馈响应较大的脑电通道,将不同通道的脑电数据分别放入长短时记忆网络中进行训练,再将各通道的训练模型通过集成学习策略进行优化,从而能整合各通道的信息进行情感分类。此研究在DEAP基准数据集上进行了情绪识别实验,我们在效价和唤醒两个情感维度的的情感识别结果均有明显的提升,准确率分别达到了87.6%,90.52%,验证了多通道集成方法的有效性。

Emotion recognition is a key issue in human-computer interaction, and EEG, as an important part of human physiological signals, is also a key factor in recognizing human emotions. Emotion recognition using EEG is a challenging study due to the complex neuronal activity in the brain leading to non-stationary behavior of signals, and multi-channel analysis is an important issue to consider when processing EEG signals. This paper proposes a multi-channel EEG emotion recognition model based on long-short-term memory network and ensemble learning. Through research, the EEG channel that responds more to emotional feedback is selected, and the EEG data of different channels are put into the long-term memory network for training. Then, the training model of each channel is optimized through the integrated learning strategy, so that the information of each channel can be integrated for emotion classification. This study conducted emotion recognition experiments on the DEAP benchmark dataset. Our emotion recognition results in the two emotional dimensions of valence and arousal have been significantly improved, and the accuracy rates have reached 87.6% and 90.52%, which validates the effectiveness of the ensemble multi-channel approach.