Table 8.

Average performance metrices of the cutting edge approaches for the thermal image in detecting CG subjects

Algorithms	Performance metrics
	Accuracy	Precision	Recall	F1-score	Computational cost (sec/image)
TransUNet	0.832	0.86	0.81	0.86	0.45
Swin-UNETR	0.83	0.70	0.82	0.733	0.52
MedT (Gated Axial Transformer)	0.88	0.812	0.88	0.825	0.49
DE-ResUNet	0.81	0.85	0.84	0.89	0.58
Hybrid CNN-ViT (ResNet-ViT)	0.82	0.80	0.89	0.89	0.51
ViT-Caps	0.90	0.85	0.82	0.845	0.43
Proposed Model	0.958	0.932	0.98	0.97	0.37