Integrated Deep Learning Pipeline for Tomato Plant Disease Recognition and Severity Estimation

DOI: https://doie.org/10.10399/JBSE.2026858643

Prakriti Saxena, Aditi Sharma, Divyanshu Chauhan

Keywords:

Leaf Disease Detection, Convolutional Neural Network, YOLO, Tomato Leaf Diseases

Abstract:

Diseases affecting tomato leaves can decrease the yield and quality of crops. Therefore, there is an urgent need for timely and accurate identification of tomato leaf diseases in order to implement precision agriculture management practices. This research reports the implementation of a comparative deep learning framework for monitoring tomato leaf diseases using three different architectural approaches: Fine-tuned Mobile Net V2, CNN+GRU hybrid model; and our new proposed multi-output CNN hybrid model. The study used a dataset from Plantcity which contained images of 11 different types of diseases affecting tomato leaves taken under numerous environmental conditions. The images used for the experiments were pre-processed by resizing, normalizing, augmenting, generating variations based on single input, and splitting the dataset into three parts - training, validation, and testing set aside for future experiments. The Multi-Output CNN Hybrid was identified as the primary model for evaluating the models due to its capacity to classify disease and assess severity and locate infected regions at once, by having a shared convolutional backbone with varied prediction heads for multiple predictions. This model uses a composite loss function that combines sparse categorical cross-entropy and MSE to allow for joint optimization and is able to make better predictions than any of the other models tested, with nearly 98.93% classification accuracy and a strong precision, recall, and F1 Score performance across all test reports. Multi-task learning was used in the creation of this architecture to improve feature sharing, reduce overfitting, and improve generalization. This architecture offers a reliable, scalable and interpretable approach to automated tomato disease monitoring, and is suitable for precision agriculture, as well as real-time crop health assessments.

PDF