Structured Comparative Review of Remote Sensing, Hyperspectral Imaging, and Deep Learning for Agricultural Monitoring and Crop Classification
Abstract
Climate-resilient agriculture increasingly relies on remote sensing, hyperspectral imaging (HSI), machine learning, and deep learning to support crop classification, disease and stress detection, yield-related assessment, and field monitoring. This structured comparative review synthesizes 50 core empirical studies drawn from an 80-reference evidence base and examines their sensing platforms, learning strategies, reported performance measures, deployment constraints, and reproducibility characteristics. The evidence is organized by multispectral and time-series sensing, HSI and band selection, unmanned aerial vehicle (UAV) and high-resolution monitoring, crop and land-cover classification, disease and weed detection, yield and soil assessment, and Internet of Things (IoT)/cloud-enabled monitoring. Numerical results are reported only when explicitly stated in the source study; unreported metrics are marked as not reported, and no metric is inferred from a related method or dataset. Because datasets, sensors, crops, class definitions, geographic regions, and validation protocols differ substantially, performance values are interpreted within their original experimental context rather than used to rank methods across studies. The synthesis identifies recurring limitations involving annotation scarcity, spectral-domain shift, class imbalance, external validation, computational cost, and incomplete reproducibility reporting, and it outlines research priorities for transferable, interpretable, and operational agricultural intelligence.