Application Of Vision Transformer For Identifying Indonesian Herbal Plants Based On Visual Images

Abstract

Indonesia has vast biodiversity, including herbal plants that have been used for generations as traditional medicinal ingredients. However, the many types of herbal plants that have similar shapes, colors, and textures often make it difficult for people to identify them accurately. To overcome this challenge, this research develops a visual image-based herbal plant identification system using the Vision Transformer (ViT) model, an artificial intelligence approach that is able to understand visual patterns more effectively than conventional methods. This research went through several stages, including the collection of herbal plant image datasets from public platforms, data preprocessing and image dimension adjustment, and training of the ViT model. The model was evaluated using metrics such as accuracy, precision, recall, and F1-score to ensure optimal performance. The results show that the ViT model is able to identify herbal plants with an accuracy of 92% and consistent performance of other evaluation metrics. This system is also implemented into the web, thus helping users in recognizing herbal plants quickly and accurately