/DE_CNN

This repository contains the tensorflow implementation for our ICONIP-2018 paper: "Continuous Convolutional Neural Network with 3D Input for EEG-Based Emotion Recognition" (To appear...)

Primary LanguagePython

Code for ICONIP 2018 submission

This repository contains the tensorflow implementation for our ICONIP-2018 paper: "Continuous Convolutional Neural Network with 3D Input for EEG-Based Emotion Recognition"

About the paper

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.
  • Please run the get_1D_data.py to compute the Differential Entropy for each original .mat file. DE features of each .mat file will be stored in 1D_dataset folder.
  • 1D_to_3D.py is used to transform the 1-dimentional data into 3-dimentional format, which will be used to train the proposed model.
  • Using cnn.py to train and test the model (10-fold cross-validation), result of each fold will be saved in a .xls file (you can find these .xls files in ./result folder).
  • count_accuracy.py is used to summarize the final accuracy of the model. The generated .xls files can be found in ./result/summary folder.

Requirements

  • Pyhton 3
  • scipy
  • numpy
  • pandas
  • sk-learn
  • tensorflow (1.4 version)
  • import xlrd
  • import xlwt

If you have any questions, please contact yilongyang@stu.xmu.edu.cn