Table 3.

Description of the proposed model architecture.

Model Content	Details
Input Image Size	224 × 224 × 3, with 5120 training images and 1280 images in each class
Feature extraction	Using EfficientNet with 1280 features
First Convolution Layer	32 filters; size = 3 × 3; ReLu; Padding = ‘Same’
First Max Pooling Layer	Pooling Size: 2 × 2
Second Convolution Layer	64 filters; size = 3 × 3; ReLu; Padding = ‘Same’
Second Max Pooling Layer	Pooling size: 2 × 2
Third Convolution Layer	128 filters; size = 3 × 3; ReLu; Padding = ‘Same’
Third Max Pooling Layer	Pooling size: 2 × 2
Fourth Convolution Layer	256 filters; size = 3 × 3; ReLu; Padding = ‘Same’
Fourth Max Pooling Layer	Pooling Size: 2 × 2
Fifth Convolution Layer	512 filters; size = 3 × 3; ReLu; Padding = ‘Same’
Fifth Max Pooling Layer	Pooling Size: 2 × 2
Fully Connected Layer	4096 nodes; ReLU
Dropout Layer	50% Neurons dropped randomly
Dense_1 Layer	8320 nodes; ReLu
Dense_2 Layer	516 nodes; ReLu
Output Layer	Four nodes; Softmax activation
Optimization Function	Adam optimization
Learning Rate	0.001
Loss Function	Categorical cross entropy