Table 2 Parameters of four base models.
CNN | AlexNet | ResNet34 | ResNet50 |
|---|---|---|---|
Input Data (64, 64, 3) | Input Data (64, 64, 3) | Input Data (64, 64, 3) | Input Data (64, 64, 3) |
Conv2D_1 (16 filters, 3 × 3, ReLU) | Conv2D: 96 filters, kernel size (11, 11), strides of 4 MaxPool (2,2), strides of 2 | ResNet34 (pre-trained) | ResNet50 (pre-trained) |
Conv2D_2 (32 filters, 3 × 3, ReLU) MaxPool (2,2) | Conv2D: 256 filters, kernel size (3, 3) MaxPool (2,2), strides of 2 | GlobalAveragePooling2D | GlobalAveragePooling2D |
Conv2D_2 (64 filters, 3 × 3, ReLU) MaxPool (2,2) | Conv2D: 384 filters, kernel size (3, 3) MaxPool (3,3), strides of 3 | Dropout (50%) | Dense (256, ReLU) |
Conv2D_2 (128 filters, 3 × 3, ReLU) MaxPool (2,2) | Dense (4096 units, ReLU) | Dense (256, ReLU) | Dense (5, Softmax) |
Dropout (25%) | Dropout (50%) | Dropout (50%) | Output (5 classes) |
Flatten | Dense (4096 units, ReLU) | Dense (5, Softmax) | |
Fully Connected Layer (64 units, ReLU) | Dropout (50%) | Output (5 classes) | |
Dropout (25%) | Dense (5 units, Softmax) | ||
Fully Connected Layer (5 units, Sigmoid) | Output (5 classes) | ||
Output (5 classes) |
