Table 2.

Experimental hyper-parameters.

Model parameters	Loss functions	Categorical cross-entropy loss and Categorical smooth loss
	Optimizers	Adam and SGD
	Learning rates	0.0001 and 0.001
	Input size	224 × 224
	Epoch	100
	Batch size	8
Multi-head attention model Parameters	Patch size	(2, 2)
	Window size	Window Size//2
	Number of heads	8
	Number of MLP	256
	Embed_dim	64
	Drop rate	0.01
Learning rate selection regulating parameters	Reduce learning rate	0.2
	Verbose	1
	Epsilon	0.001
	Es-Callback (Patience)	10
	Clip value	0.2
	Patience	10