Table 7 Layer wise architecture details of the employed four models along with total trainable and non-trainable parameters.
Layer type | MobileNetV2 | EfficientNetB0 | ResNet-50 | Inception V3 |
|---|---|---|---|---|
Input layer | 224 × 224 × 3 RGB image | 224 × 224 × 3 RGB image | 224 × 224 × 3 RGB image | 299 × 299 × 3 RGB image |
Base layers (non-trainable) | - Initial Conv2D (3 × 3, 32 filters, stride = 2) - Bottleneck blocks (MobileNetv2) | - Initial Conv2D (3 × 3, 32 filters, stride = 2) - MBConv blocks (EfficientNetB0) | - Initial Conv2D (7 × 7, 64 filters, stride = 2) - Residual blocks (ResNet-50) | - Initial Conv2D (3 × 3, 32 filters, stride = 2) - Inception blocks (Inceptionv3) |
Global pooling | Global average pooling | Global average pooling | Global average pooling | Global average pooling |
DenseLayer1 | 512 units, ReLU | 1024 units, ReLU | 512 units, ReLU | 1024 units, ReLU |
DropOut | Rate = 0.5 | Rate = 0.5 | Rate = 0.5 | Rate = 0.5 |
DenseLayer2 | 256 units, ReLU | 256 units, ReLU | 256 units, ReLU | 256 units, ReLU |
DropOut | Rate = 0.5 | Rate = 0.5 | Rate = 0.5 | Rate = 0.5 |
Output layer | Dense (10 units, softmax) | Dense (10 units, softmax) | Dense (10 units, softmax) | Dense (10 units, softmax) |
Total parameters | Total: ~ 3.4 M Trainable: ~ 400 K Non-Trainable: ~ 3 M | Total: ~ 5.3 M Trainable: ~ 300 K Non-Trainable: ~ 5 M | Total: ~ 25 M Trainable: ~ 2 M Non-Trainable: ~ 23 M | Total: ~ 22 M Trainable: ~ 2 M Non-Trainable: ~ 20 M |
