This repository is a part of EEG-Emotion Recognition Research. It manifests models used in our experiments.
There are 4 CNN architectures (3Conv - 6Conv). You can see Keras implementation in /typicalModels. Also, we have tested 3D physical electrode placement. In this experiment, we had to adjust the models to fit the new input's size. You can find these modified models in /3DModels.
| 3Conv | 4Conv | 5Conv | 6Conv | ||||
|---|---|---|---|---|---|---|---|
| Conv2D (5x5)x32 | Conv2D (5x5)x32 | Conv2D (5x5)x32 | Conv2D (5x5)x32 | ||||
| Conv2D (3x3)x32 | Conv2D (3x3)x32 | Conv2D (2x2)x32 | Conv2D (2x2)x32 | ||||
| MaxPooling2D 2x2 | MaxPooling2D 2x2 | Conv2D (2x2)x32 | Conv2D (2x2)x32 | ||||
| Conv2D (3x3)x64 | Conv2D (2x2)x64 | MaxPooling2D 2x2 | MaxPooling2D 2x2 | ||||
| Dropout 0.5 | Conv2D (2x2)x64 | Conv2D (2x2)x64 | Conv2D (2x2)x64 | ||||
| FC 128x1 | FC 128x1 | Dropout 0.5 | Conv2D (2x2)x64 | Conv2D (2x2)x64 | |||
| Dropout 0.5 | Dropout 0.5 | FC 128x1 | FC 128x1 | Dropout 0.5 | Conv2D (2x1)x64 | ||
| FC 2x1 | FC 2x1 | Dropout 0.5 | Dropout 0.5 | FC 128x1 | FC 128x1 | Dropout 0.5 | |
| FC 2x1 | FC 2x1 | Dropout 0.5 | Dropout 0.5 | FC 128x1 | FC 128x1 | ||
| FC 2x1 | FC 2x1 | Dropout 0.5 | Dropout 0.5 | ||||
| FC 2x1 | FC 2x1 | ||||||
| 3Conv | 4Conv | 5Conv | 6Conv | ||||
|---|---|---|---|---|---|---|---|
| Conv3D (9x2x3)x32 | Conv3D (9x2x3)x32 | Conv3D (9x2x3)x32 | Conv3D (9x2x3)x32 | ||||
| Conv3D (3x2x3)x32 | Conv3D (3x2x3)x32 | Conv3D (3x2x3)x32 | Conv3D (3x2x3)x32 | ||||
| MaxPooling3D 4x1x1 | MaxPooling3D 4x1x1 | Conv3D (3x1x1)x64 | Conv3D (3x1x1)x64 | ||||
| Conv3D (3x1x1)x64 | Conv3D (3x1x1)x64 | MaxPooling3D 4x1x1 | MaxPooling3D 4x1x1 | ||||
| Dropout 0.5 | Conv3D (3x1x1)x64 | Conv3D (3x1x1)x64 | Conv3D (3x1x1)x64 | ||||
| FC 128x1 | FC 128x1 | Dropout 0.5 | Conv3D (3x1x1)x64 | Conv3D (3x1x1)x64 | |||
| Dropout 0.5 | Dropout 0.5 | FC 128x1 | FC 128x1 | Dropout 0.5 | Conv3D (3x1x1)x64 | ||
| FC 2x1 | FC 2x1 | Dropout 0.5 | Dropout 0.5 | FC 128x1 | FC 128x1 | Dropout 0.25 | |
| FC 2x1 | FC 2x1 | Dropout 0.5 | Dropout 0.5 | FC 128x1 | FC 128x1 | ||
| FC 2x1 | FC 2x1 | Dropout 0.5 | Dropout 0.5 | ||||
| FC 2x1 | FC 2x1 | ||||||
If you find the code useful for your research, please cite our paper:
@article{keelawat2019,
title={Spatiotemporal Emotion Recognition using Deep CNN Based on EEG during Music Listening},
author={Panayu Keelawat, Nattapong Thammasan, Masayuki Numao and Boonserm Kijsirikul},
journal={arXiv preprint arXiv:1910.09719},
year={2019}
}