Table 4.
Layers and parameters of the proposed TrAlexNet + LSTM.
| Layer Name | Layer Type | Filter Size | Stride | Padding | Output Channel | Activation Function |
|---|---|---|---|---|---|---|
| conv1 | convolution 2d | 11,11 | 4,4 | 0 | 96 | relu |
| pool1 | max pooling 2d | 3,3 | 2,2 | 0 | 96 | – |
| conv2 | grouped convolution 2d | 5,5 | 1,1 | 2 | 256 | relu |
| pool2 | max pooling 2d | 3,3 | 2,2 | 0 | 256 | – |
| conv3 | convolution 2d | 3,3 | 1,1 | 1 | 384 | relu |
| conv4 | grouped convolution 2d | 3,3 | 1,1 | 1 | 384 | relu |
| conv5 | grouped convolution 2d | 3,3 | 1,1 | 1 | 256 | relu |
| pool5 | max pooling 2d | 3,3 | 2,2 | 0 | 256 | – |
| fc6 | fully connected | – | – | – | 4096 | relu |
| fc7 | fully connected | – | – | – | 4096 | relu |
| fc8 | fully connected | – | – | – | 1000 | relu |
| flatten | flatten | – | – | – | – | – |
| lstm | lstm | – | – | – | – | – |
| fc_optimized | fully connected | – | – | – | 4 | softmax |