Research progress on tobacco pest and disease detection: from traditional methods to deep learning

Abstract

Tobacco pests and diseases cause significant yield losses and quality deterioration in global tobacco production. Therefore, accurate and timely detection of tobacco pests and diseases is essential for supporting field management and reducing production losses. This review traces the methodological development of tobacco pest and disease detection, from conventional image analysis and machine learning to deep learning-based detection and UAV-based field-scale monitoring. Furthermore, this review summarizes the major challenges in current research, including the scarcity of tobacco-specific datasets, the difficulty of detecting small lesions and pests, interference from complex field backgrounds, limited crossscenario generalization, and practical deployment constraints. Future research directions are also discussed from the perspectives of dataset construction, model robustness, multi-source information fusion, and efficient field deployment. This review provides a useful reference for the development and practical application of intelligent tobacco pest and disease detection systems.