Abstract

In the context of emotion recognition, Artificial Intelligence technology has demonstrated several functions in people's lives. Computing research is now focused on Electroencephalogram (EEG) signals to identify emotional states. The connection and interaction between multichannel EEG signals give important information about emotional states. However, most existing emotion identification techniques perform poorly in practical applications by preventing their advancement. The main objective of this paper is to design an efficient model for emotion recognition based on deep learning technology by EEG signals. The proposed model for emotion recognition collects the EEG signals from the standard benchmark datasets. Then, the signal decomposition is performed using the Tunable Q-factor Wavelet Transform with the collected EEG signals. The decomposed signals are taken for the optimal feature selection phase, where the significant features of the emotion are selected through the hybrid optimization algorithm named Aquila Fireworks Optimization Algorithm (AFOA). Finally, the EEG emotion classification is performed using Hybrid Variational Autoencoder and Block Recurrent Transformer Network. The tuning of the parameter is made through the same AFOA to improve the efficiency of classification. The experimental analysis is conducted to analyze the efficiency of the recommended emotion recognition model with the comparison over the traditional techniques.

中文翻译：

---

基于智能最优特征选择的混合变分自动编码器和块循环变压器网络，用于使用脑电信号的准确情绪识别模型

摘要

在情感识别的背景下，人工智能技术在人们的生活中展现了多种功能。计算研究现在主要集中在脑电图 (EEG) 信号上，以识别情绪状态。多通道脑电图信号之间的联系和相互作用提供了有关情绪状态的重要信息。然而，大多数现有的情绪识别技术在实际应用中表现不佳，阻碍了其进步。本文的主要目的是设计一种基于脑电信号深度学习技术的有效情绪识别模型。所提出的情感识别模型从标准基准数据集中收集脑电图信号。然后，使用可调谐 Q 因子小波变换对收集的 EEG 信号进行信号分解。分解后的信号被用于最佳特征选择阶段，其中通过名为 Aquila Fireworks Optimization Algorithm (AFOA) 的混合优化算法选择情感的重要特征。最后，使用混合变分自动编码器和块循环变压器网络执行脑电图情绪分类。通过相同的AFOA进行参数的调整，以提高分类效率。通过实验分析来分析推荐情感识别模型的效率并与传统技术进行比较。