Journal — Smart Agri. Tech. — 2025
Beyond Classification: Benchmarking Object Detection Models for Efficient Tomato Leaf Disease Identification on a Real-World Dataset
Fahim Mahafuz Ruhad, Md Fahim, Mir Sazzat Hossain, Md Fahad Monir, Ashraful Islam, M Ashraful Amin
Smart Agricultural Technology, Volume 12, 2025, 101336, ISSN 2772-3755.
Abstract
Tomato is one of the most economically significant horticultural crops worldwide, yet its cultivation is frequently hindered by foliar diseases such as Late Blight, Bacterial Spot, and Septoria Leaf Spot. These diseases not only reduce yield but also result in substantial financial losses, particularly for smallholder farmers. Traditional manual inspection methods are labor-intensive and error-prone, highlighting the need for automated and accurate disease detection systems. While image classification models have been widely explored for plant disease identification, they often fail to capture spatial symptom details, leading to reduced interpretability and accuracy—especially for visually ambiguous diseases. In this study, we propose an object detection-based approach to localize and identify tomato leaf diseases more effectively. To facilitate this, we introduce a novel, regionally diverse dataset comprising nine common tomato diseases collected across three Bangladeshi regions over three months. We benchmark a range of classification and object detection models, demonstrating that YOLO-v9 achieves superior performance (F1-score: 97.86, mean Average Precision (mAP): 79.44). Furthermore, for real-time deployment on resource-constrained edge devices, we present YOLO-v8-CSwin, a lightweight model with competitive accuracy and significantly fewer parameters (11.12M). Our findings highlight the advantages of object detection over classification for plant disease diagnosis and offer practical insights for deploying AI-driven solutions in real-world agricultural settings.
Topics
Cite
@article{RUHAD2025101336,
title = {Beyond classification: Benchmarking object detection models for efficient tomato leaf disease identification on a real-world dataset},
journal = {Smart Agricultural Technology},
volume = {12},
pages = {101336},
year = {2025},
issn = {2772-3755},
doi = {https://doi.org/10.1016/j.atech.2025.101336},
url = {https://www.sciencedirect.com/science/article/pii/S2772375525005672},
author = {Fahim Mahafuz Ruhad and Md Fahim and Mir Sazzat Hossain and Md. Fahad Monir and Ashraful Islam and M. Ashraful Amin},
keywords = {Classification, Detection, Leaf detection, Edge computing, Leaf disease detection},
abstract = {Tomato is one of the most economically significant horticultural crops worldwide, yet its cultivation is frequently hindered by foliar diseases such as Late Blight, Bacterial Spot, and Septoria Leaf Spot. These diseases not only reduce yield but also result in substantial financial losses, particularly for smallholder farmers. Traditional manual inspection methods are labor-intensive and error-prone, highlighting the need for automated and accurate disease detection systems. While image classification models have been widely explored for plant disease identification, they often fail to capture spatial symptom details, leading to reduced interpretability and accuracy—especially for visually ambiguous diseases. In this study, we propose an object detection-based approach to localize and identify tomato leaf diseases more effectively. To facilitate this, we introduce a novel, regionally diverse dataset comprising nine common tomato diseases collected across three Bangladeshi regions over three months. We benchmark a range of classification and object detection models, demonstrating that YOLO-v9 achieves superior performance (F1-score: 97.86, mean Average Precision (mAP): 79.44). Furthermore, for real-time deployment on resource-constrained edge devices, we present YOLO-v8-CSwin, a lightweight model with competitive accuracy and significantly fewer parameters (11.12M). Our findings highlight the advantages of object detection over classification for plant disease diagnosis and offer practical insights for deploying AI-driven solutions in real-world agricultural settings.}
}