Advancements in Tobacco Crop Monitoring: A Survey of Deep Learning Technologies

The challenge of tobacco monitoring is a technical and political issue, given that tobacco production is a cross-cutting issue that involves agricultural management, regulatory compliance and illegal harvesting. Tobacco acreage estimation and tobacco classification have been expanded through the use of remote sensing and deep learning has enhanced the capacity to extract spatial, spectral, and temporal patterns from the timeseries data. This survey summarizes the literature on tobacco monitoring and tobacco-relevant crops in a new format, a critical comparison between the various sensing platforms and learning paradigms. It reviews the multispectral, Synthetic Aperture Radar (SAR), unmanned aerial vehicles (UAV) and hyperspectral data sources, as well as traditional machine learning, convolutional, recurrent, attention-based and hybrid models, and points out the experimental conditions in which several studies have achieved high accuracy. The most significant advances have been made in multi-temporal optical classification and UAV-based field-scale mapping, while operational deployment remains hindered by the lack of benchmark datasets, cross region validation, and few SAR–optical fusion studies, as well as the lack of mature decision support pipelines. Future progress will be possible with open tobacco-specific benchmarks, multi-sensor fusion, transferable models, explainable AI and scalable monitoring systems, which will facilitate agricultural and policy decision-making.