Research Article
Multiclass Classification of Imagined Speech Vowels and Words of Electroencephalography Signals Using Deep Learning
Table 1
Summary of the model architectural parameters.
| ā | Parameter | Values |
| Input layer | Signal dimension | (6, 2048) |
| TCN layer | No. of blocks | 5 | No. of filters | 256 | conv1d | Causal | Dilation factor | 1,2,4 | Kernel | 3 | Activation | Gated activation unit | Spatialdropout1d | 0.2 |
| CNN layer | No. of blocks | 5 | Convolution | conv1d | No. of filters | 256 | Activation | ReLU | Dropout | 0.1 | Pooling | maxpool1d (2) |
| Transformation layer | Fully connected | 1024 | Activation | ELU |
| Classification layer | Fully connected | 11 | Activation | Softmax | Loss function | Cross entropy |
| Training | Optimizer | Adam | Learning rate | 0.00001 | Epochs | 300 | Batch size | 256 |
|
|