Deep Learning for Tomato Disease Detection and Severity Assessment: A Systematic Analytical Review of Methods, Datasets, and Challenges

Abstract

Abstract Tomato diseases significantly affect crop productivity and food security, necessitating accurate and timely detection methods. This paper presents a systematic and analytical review of deep learning approaches for tomato disease detection and severity assessment, based on 76 research studies. Existing methods are categorized into classification, detection, segmentation, and emerging multi-task frameworks. The analysis shows that convolutional neural networks achieve high accuracy on controlled datasets but exhibit limited generalization in real-world conditions. Advanced architectures, including transformer-based and hybrid models, improve performance but increase computational complexity. A key finding is the limited focus on disease severity assessment, which remains underexplored despite its importance for precision agriculture. The review identifies major challenges, including dataset limitations, lack of standardized benchmarks, and deployment constraints. Future directions emphasize multi-task learning, real-world dataset development, lightweight models, and explainable AI. This study provides a foundation for developing robust and practical tomato disease detection systems.

Code and data availability

This is a systematic review of tomato disease detection literature; the supplied blocks contain no public phenotype datasets, images, author code, models, or supplements specific to this paper's own measurements or analysis.

No evidence-backed public reproduction asset is currently recorded.