Applying Vision Transformers for Herbal Medicine Classification: A Novel Approach to Plant Identification

The medicinal plants are particularly important in traditional health care systems, yet the right identification of medicinal plants is one of the great challenges because of diversity and similarity among species. In this paper, we introduce a new approach to herbal medicine classification using Vision Transformers, or simply ViTs. This deep learning architecture is at the front edge of approaching complex patterns within plant images through leveraging self-attention mechanisms. Unlike the classical CNN, Vision Transformers are significantly better than their counterparts on tasks related to classification as they concentrate on global relationships between image patches. We apply and test our model on a comprehensive dataset of medicinal plants. It is demonstrated that Vision Transformer works better in accuracy and generalization than existing CNN-based approaches. In addition, the model demonstrates robustness under different environmental conditions related to lighting, variations of backgrounds, and also other probable challenges in identification with the real world. This study also brings out the possibility of directly integrating the classification system based on Vision Transformers into applications that may be used in mobile devices, which will now enable real-time identification in fieldwork settings for herbal medicines. Our findings show that the proposed Vision Transformers push medicinal plant classification in new directions in traditional and modern medical settings.