Classification of Ceremonial Plants with Vision Transformer

Abstract

The rapid advancement of computer vision technology has accelerated the adoption of artificial intelligence in agriculture, particularly for plant image classification tasks. However, the identification of ceremonial plants remains challenging due to the high visual similarity among species and the continued reliance on manual identification methods, which are time-consuming and require expert knowledge. Unlike previous studies that primarily focused on general crop species or plant disease classification, this study specifically investigates the application of the Vision Transformer (ViT) model for the classification of ceremonial plants, which represent culturally significant plant species with distinctive yet visually similar characteristics. An experimental approach was employed using a dataset of 1,244 ceremonial plant images representing seven classes, with the data divided into training, validation, and testing sets at proportions of 70%, 15%, and 15%, respectively. A pretrained Vision Transformer model was fine-tuned by adapting its classification head to the target classes and evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. The experimental results demonstrate that the proposed model achieved a test accuracy of 97.33% and an average class accuracy of 97.01%, indicating its effectiveness in learning complex visual representations and accurately distinguishing visually similar ceremonial plant species. These findings demonstrate the feasibility of Vision Transformer for culturally specific plant recognition and provide a reliable baseline for the development of intelligent ceremonial plant identification systems, contributing to the digital preservation of traditional botanical knowledge and AI-based plant recognition applications.

Keywords

Deep Learning; Image Classification; Ceremonial Plants; Vision Transformer