Table 2.

The experimental setting of the CNN models for the training and validation process.

CNN	Input Size	Activation Function	Optimizer	Batch Size/Epoch/Dropout	Learning rate of Nadam/RMSProp/SGD
Xception	299 × 299 × 3	Mish, ReLU	Nadam, RMSProp, SGD	20/50/0.5	0.002/0.001/0.002
Inception-V3	299 × 299 × 3	Mish, ReLU	Nadam, RMSProp, SGD	20/50/0.5	0.002/0.001/0.002
ResNet-50	224 × 224 × 3	Mish, ReLU	Nadam, RMSProp, SGD	20/50/0.5	0.002/0.001/0.002
NasNetMobile	224 × 224 × 3	Mish, ReLU	Nadam, RMSProp, SGD	20/50/0.5	0.002/0.001/0.002
VGG-16	224 × 224 × 3	Mish, ReLU	Nadam, RMSProp, SGD	20/50/0.5	0.002/0.001/0.002
AlexNet	224 × 224 × 3	Mish, ReLU	Nadam, RMSProp, SGD	20/50/0.5	0.002/0.001/0.002

Stochastic gradient descent (SGD).