fabalcu97 / emotioncaps-fabalcu97

Code for Multi-level Features Guided Capsule Network （MLF-CapsNet）

This repository contains the Keras implementation for the paper: "Multi-channel EEG-based Emotion Recognition via a Multi-level Features Guided Capsule Network"

About the paper

• Title: Multi-channel EEG-based Emotion Recognition via a Multi-level Features Guided Capsule Network
• Authors: Yu Liu, Yufeng Ding, Chang Li, Juan Cheng, Rencheng Song, Feng Wan, Xun Chen
• Institution: Hefei University of Technology
• Published in: 2020 Computers in Biology and Medicine (CBM)

Instructions

• Before running the code, please download the DEAP dataset, unzip it and place it into the right directory. The dataset can be found here. Each .mat data file contains the EEG signals and consponding labels of a subject. There are 2 arrays in the file: data and labels. The shape of data is (40, 40, 8064). The shape of label is (40,4).
• Please run the deap_pre_process.py to Load the origin .mat data file and transform it into .pkl file.
• Using capsulenet-multi-gpu.py to train and test the model (10-fold cross-validation), result of 10 folds will be saved in a .xls file.
• count_accuracy_deap.py is used to calculate the final accuracy of the model.
• The usage on DREAMER dataset is the same as above. The DREAMER dataset can be found here.

Requirements

• Pyhton3.5
• tensorflow (1.3.0 version)
• keras (2.2.4 version)

If you have any questions, please contact [email protected]

Reference

• XifengGuo/CapsNet-Keras
• ynulonger/ijcnn