Enhanced hybrid vision transformer for zero-shot plant leaf disease classification

Abstract

Plant leaf disease classification is essential in the agriculture industry, and recent advances in deep learning (DL) and machine learning (ML) have resulted in numerous approaches for detecting and classifying diseases using plant images. However, traditional diagnosis methods rely on human expertise and remain time-consuming and labor-intensive. To address this issue, the proposed system introduces a novel approach, Enhanced Hybrid Vision Transformer with Zero-shot learning classification (EHVZSC), for plant species identification. The method combines the strengths of Vision Transformers (ViT) and Zero-Shot Learning, enabling accurate classification of unseen classes without additional training data. The proposed approach leverages ViT to learn robust image representations, which are then used to generate a set of prototypes for zero-shot classification. Evaluated on the Plant Village dataset The model was trained with 50 epochs, a batch size of 32, and a learning rate of 0.0001 because these parameters provided stable convergence without overfitting, the proposed EHVZSC model achieves state-of-the-art performance with reduced data requirements, improving testing accuracy by up to 15 %, sensitivity by up to 10 %, specificity by up to 10 %, F1-score by up to 12 %, and ROC performance by up to 25 % over existing methods, while attaining 95.7 % accuracy, 97.8 % sensitivity, 95.0 % specificity, and a 96.45 % F1-score, demonstrating its superior ability to capture fine-grained disease features through attention-enhanced representation learning and robust zero-shot adaptability.

Plant phenotyping relevance

植物葉画像から病害状態を分類する新規Vision Transformer／ゼロショット手法の開発と性能評価が中心であり、植物病害フェノタイピング手法に該当する。

Code and data availability

The paper uses the public PlantVillage dataset, but this is a pre-existing third-party dataset, not a paper-specific deposit. No author code, trained models, or supplementary data availability statements appear in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.