Intelligent Prioritisation of Tomato Leaf Disease Diagnosis Using Symptom Hierarchies and Computer Vision

Abstract: Tomato Leaf Diseases (TLD) pose a significant threat to crop productivity and fruit quality, as they can spread rapidly if not detected and treated at an early stage. Manual inspection of leaves is time-consuming and prone to human error, particularly because different diseases often exhibit similar visual symptoms. With the advancement of Computer Vision (CV), automatic detection of TLD has become possible; however, most existing methods rely solely on image-based classification and fail to consider the relationships among symptoms that agricultural experts typically use for accurate diagnosis.
This study introduces a Multi-Modal Diagnosis (MMD) system that integrates CV techniques with a Symptom Hierarchy (SH) to enhance both the accuracy and interpretability of TLD detection. The proposed system employs a pre-trained ResNet-18 (RN18) model to extract visual features from leaf images while simultaneously identifying relevant symptom tags. These symptoms are organized hierarchically—from general indicators such as spots and discoloration to more specific patterns—allowing the system to rank potential diseases based on their severity and likelihood.
Additionally, Data Augmentation (DA) and class balancing techniques are applied to improve model reliability and minimize bias. Experimental results demonstrate that the proposed model achieves an accuracy of 92.5% and an F1- score of 91.8%, outperforming single-modality approaches. By combining CV with hierarchical symptom reasoning, the system provides early, reliable, and interpretable disease detection, empowering farmers to make informed and timely management decisions.

Keywords: Tomato Leaf Disease (TLD), Symptom Hierarchy (SH), ResNet-18 (RN18), Multi-Modal Diagnosis (MMD), Data Augmentation (DA), Leaf Disease Detection(LDD).