<![CDATA[Snakebites remain a major global health concern, with over 4.5 million cases annually, primarily affecting rural populations in tropical regions. Accurate snake species identification is critical for proper treatment, yet challenges persist due to morphological similarities, particularly among visually similar green snake species. We test five Vision Transformer (ViT)-based models to see how well they can classify snakes based on pictures of their heads and bodies. The models are ViT-B16, DeiT, PoolFormer, Swin-T, and CaiT. Results indicate that head structure classification achieved higher accuracy than body pattern classification due to more distinct morphological features. CaiT outperformed other models, achieving 87% accuracy, particularly when trained on RGB images. These findings highlight the importance of model selection and dataset characteristics in improving snake species classification, especially for species with high visual similarity.]]>
Copyrights © 2025
