Feature Fusion Based on Mutual-Cross-Attention Mechanism for EEG Emotion Recognition
arXiv: 2406.14014
Springer: MICCAI 2024 (Will be updated once available)
MCA is a purely mathematical method applying Attention Mechanism from each directions of two features. In the field of EEG emotion analysis, we are the first to propose a pure mathematical fusion method, coupled with customized 3D-CNN, to accomplish the task of feature fusion.
To acquire the original datasets, please refer to the official website of DEAP. We used the preprocess method from open-source repository DEAP_MNE_preprocessing. And we finally got the preprocessed datasets, which can be obtained from Google Drive.
Installation
conda env create -f environment.yml
conda activate MCA-EEGTrain
Download the preprocessed datasets and put them under the implementation directory. Then, open the mca_experiment.ipynb and modify the path to your data directory. Run the notebook to train the model.
Test
Specify the validation type in the mca_validations.ipynb and run the notebook to test the model. You can get confusion matrix and accuracy.
# The default validation type is 'valence', change to validate others
# ['valence', 'arousal', 'dominance', 'liking']
validation_type = 'arousal'The model is licensed under the Apache 2.0 license
Our paper has been accepted by MICCAI 2024. Remenber to cite the paper if you find this work useful.
@inproceedings{Zhao2024MCA,
title = {Feature Fusion Based on Mutual-Cross-Attention Mechanism for EEG Emotion Recognition},
booktitle = {Medical Image Computing and Computer Assisted Intervention –- MICCAI 2024},
author = {Zhao, Yimin and Gu, Jin},
year = {2024},
month = jun,
number = {arXiv:2406.14014},
eprint = {2406.14014},
publisher = {arXiv},
urldate = {2024-06-21},
archiveprefix = {arxiv},
}