Vision transformers (ViTs) have recently gained traction in plant disease classification due to their strong performance in visual recognition tasks. However, their application to tomato leaf disease recognition remains challenged by two factors, namely the need for models that can generalise across diverse disease conditions and the absence of a unified framework for systematic comparison. Existing ViT-based approaches often yield inconsistent results across datasets and disease types, limiting their reliability and practical deployment. To address these limitations, this study proposes the ViT-Based Robust Framework (ViT-RoT), a novel benchmarking framework designed to systematically evaluate the performance of various ViT architectures in tomato leaf disease classification. The framework facilitates the systematic classification of state-of-the-art ViT variants into high-, moderate-, and low-performing groups for tomato leaf disease recognition. A thorough empirical analysis is conducted using one publicly available benchmark dataset, assessed through standard evaluation metrics. Results demonstrate that the ConvNeXt-Small and Swin-Small models consistently achieve superior accuracy and robustness across all datasets. Beyond identifying the most effective ViT variant, the study highlights critical considerations for designing ViT-based models that are not only accurate but also efficient and adaptable to real-world agricultural applications. This study contributes a structured foundation for future research and development in vision-based plant disease diagnosis.
In the agricultural domain, the availability of labelled data for disease recognition tasks is often limited due to the cost and expertise required for annotation. In this paper, a novel semi-supervised learning framework named TOM-SSL is proposed for automatic tomato leaf disease recognition using pseudo-labelling. TOM-SSL effectively addresses the challenge of limited labelled data by leveraging a small labelled subset and confidently pseudo-labelled samples from a large pool of unlabelled data to improve classification performance. Utilising only 10% of the labelled data, the proposed framework with a MobileNetV3-Small backbone achieves the best accuracy at 72.51% on the tomato subset of the PlantVillage dataset and 70.87% on the Taiwan tomato leaf disease dataset across 10 disease categories in PlantVillage and 6 in the Taiwan dataset. While achieving recognition performance on par with current state-of-the-art supervised methods, notably, the proposed approach offers a tenfold enhancement in label efficiency.