Table 2.
Hyperparameters configuration of the proposed classification models.
| Model | Hyperparameters | Value |
|---|---|---|
| Convolutional Neural Network | Input shape (layer 1) | (64, 64, 1) |
| Number of filters in Conv2D layers (layers 2–3, 6) | 64, 128, 256 | |
| Strides in Conv2D layers (layers 2–3, 6) | (2, 2) | |
| Kernel Size in Conv2D layers (layers 2–3, 6) | 3 | |
| Activation function in Conv2D layers (layers 2–3, 6) | Relu | |
| Pool size in MaxPool2D layers (layers 4, 7) | (1, 1) | |
| Dropout [48] rate in Dropout [48] layers (layers 5, 8) | 0.2 | |
| Units in Dense layer (layers 10) | 2 | |
| Activation function in Dense layer (10) | Sigmoid | |
| Convolutional and Denoising Variational Autoencoder(Encoder) | Input shape (layer 1) | (64, 64, 1) |
| Number of filters in Conv2D layers (layers 2–3) | 32, 64 | |
| Strides in Conv2D layers (layers 2–3) | 2 | |
| Kernel Size in Conv2D layers (layers 2–3) | 3 | |
| Activation function in Conv2D layers (layers 2–3) | Relu | |
| Padding in Conv2D layers (layers 2–3) | same | |
| Units in Dense layers (layers 5, 6–7) | 10, 5 | |
| Activation function used in Dense layer (layer 5) | Relu | |
| Units in Lambda layer (layer 9) | 5 | |
| Convolutional and Denoising Variational Autoencoder(Decoder) | Input shape (layer 1) | 5 |
| Units used in Dense layer (layer 2) | (16 × 16 × 64) = 16,384 | |
| Target Shape used in Reshape layer (layer 3) | (16, 16, 64) | |
| Number of filters used in Conv2DTranspose layers (layers 4–6) | 64, 32, 1 | |
| Kernel size used in Conv2DTranspose layers (layers 4–6) | 3 | |
| Strides used in Conv2DTranspose layers (layers 4–6) | 2, 2, 1 | |
| Padding used in Conv2DTranspose layers (layers 4–6) | same | |
| Activation function used in Conv2DTranspose layers (layers 4–6) | Relu, Relu, Sigmoid |