AN INTELLIGENT ATTENTION-BASED DEEP LEARNING MODEL FOR TOMATO LEAF DISEASE RECOGNITION AND CLASSIFICATION

Abstract

Automated recognition of tomato leaf conditions can support crop monitoring by identifying visible symptoms from digital images. This paper presents a Vision Transformer (ViT) model and analyzes the eight-class evaluation documented in the supplied study. The evaluated categories comprise bacterial spot, early blight, late blight, leaf mold, Septoria leaf spot, two-spotted spider mite damage, tomato yellow leaf curl virus, and healthy leaves. The model combines image resizing and normalization, patch embeddings, positional encoding, transformer-based feature extraction, and a fully connected classification head. A balanced held-out test set contains 800 images, with 100 images per class. The source reports overall accuracy of 0.98 and macro-averaged precision, recall, and F1-score of 0.98. Class-level F1-scores range from 0.97 to 1.00. Early blight has the lowest recall (0.94), whereas Septoria leaf spot has the lowest precision (0.94). These findings indicate strong discrimination within the reported evaluation, while identifying the classes requiring further investigation. Average precision and feature-space visualization provide complementary evidence but do not establish calibration or field robustness. The analysis distinguishes documented results from proposed procedures and excludes unverified numerical comparisons with convolutional baselines. Missing dataset provenance, training configurations, prediction-level records, and external validation limit reproducibility and the strength of deployment claims. The model provides a promising basis for tomato leaf screening, subject to controlled replication and validation on independently collected field images.