A multi-feature and multi-scale fusion CNN with attention mechanism for EEG emotion recognition
- 28 April 2026
- journal article
- Published by IOP Publishing in Physica Scripta
- Vol. 101 (19)
Abstract
Emotion is a psychological state closely related to subjective experience, physiological response and behavioral performance, and accurate recognition is of great significance in the fields of auxiliary psychological diagnosis, autism treatment and health education. Based on the advantages of low cost and strong accuracy of EEG, a multi-feature and multi-scale deep convolutional neural network (MF-MSCNN) fused with Multi-Head Self-Attention (MHSA) is proposed. MF-MSCNN first extracts two types of features that reflect different characteristics of EEG: one is the statistical feature that characterizes the overall amplitude level and dispersion, and the other is the time-domain feature that focuses on the change trend and dynamic characteristics. Subsequently, MF-MSCNN uses multi-scale convolutional kernels to extract features in parallel, in which the small convolutional kernel focuses on local details and the large convolutional kernel captures global information. Combined with MHSA, the features are adaptively weighted to enhance the representation ability of key features, and at the same time suppress noise interference, so as to further improve the ability of MF-MSCNN to discriminate emotions. Experimental results show that MF-MSCNN is superior to comparison models.Keywords
Funding Information
- National Key Research and Development Program of China (2024YFF0618301)
This publication has 30 references indexed in Scilit:
- A multi-scale deep CNN based on attention mechanism for EEG emotion recognitionJournal of Neuroscience Methods, 2025
- CBSAtt: a CNN-BiLSTM network with multi-head self-attention for EEG emotion recognitionSignal, Image and Video Processing, 2025
- Mifnet: a MamBa-based interactive frequency convolutional neural network for motor imagery decodingCognitive Neurodynamics, 2025
- Transformer-based fusion model for mild depression recognition with EEG and pupil area signalsMedical & Biological Engineering & Computing, 2025
- Emotion recognition using spatial-temporal EEG features through convolutional graph attention networkJournal of Neural Engineering, 2023
- Human emotion recognition from EEG-based brain–computer interface using machine learning: a comprehensive reviewNeural Computing and Applications, 2022
- EEG-based Emotion Recognition with Feature Fusion NetworksInternational Journal of Machine Learning and Cybernetics, 2021
- Multi-channel EEG-based emotion recognition via a multi-level features guided capsule networkComputers in Biology and Medicine, 2020
- Recognizing Emotions Evoked by Music Using CNN-LSTM Networks on EEG SignalsIEEE Access, 2020
- Learning CNN features from DE features for EEG-based emotion recognitionPattern Analysis and Applications, 2019