TABLE 2.

Basic learning parameters of designed neural Network’s architecture different training phases.

Parameter	First phase (initial training)	Second phase (fine-tuning)	Third phase (fine-tuning on new data)
Data Used	Jeanray-2015 Selected Training Data	Jeanray-2015 Selected Training Data	Our training data
Base Model	ResNet50 (ImageNet weights network’s)	Same as First Phase	Same as First Phase
Attention Mechanism	CBAM applied to base model output	Same as First Phase	Same as First Phase
Frozen Layers	All base model layers	First 1/3 of base model layers frozen	None; all layers are trainable
Unfrozen Layers	Only new layers added on top	Last 2/3 of base model layers and new layers	Entire model is trainable
Optimizer	Adam optimizer	Adam optimizer	Adam optimizer
Initial Learning Rate	1e-4	1e-5	1e-5
Batch Size	16	16	16
Epochs	Up to 100 (with early stopping)	Up to 100 (with early stopping)	Up to 100 (with early stopping)
Loss Function	Categorical Crossentropy	Categorical Crossentropy	Categorical Crossentropy
Metrics	Accuracy	Accuracy	Accuracy