Table 1 Detail of layers of proposed LSTM network.
Type | Learnable | Activation |
|---|---|---|
Feature input layer | – | 7 |
Long-short-term memory layers × 5 | Recurrent weight: 512 × 128 | 128 |
Input weight: 512 × 7 | ||
Bias 512 × 1 | ||
BN × 5 | Scale:128 × 1 | 128 |
Offset:128 × 1 | 128 | |
ReLU × 5 | – |  |
Fully connected layer 1 | Bias:22 × 1 | 22 |
Weight: 22 × 128 | ||
Fully connected layer 2 | Bias:22 × 1 | 22 |
Weights:22 × 22 | ||
SOFTMAX | – | 22 |
Class output | – | 22 |
