Answers are grounded in this paper and cite their sources.
The remarkable progress in the areas of Computer Vision (CV), Artificial Intelligence (AI) and Deep Learning (DL), combined with the rapid democratization of Unmanned Aerial Vehicles (UAVs), has propelled Optical Remote Sensing (ORS) into a central role within the field of Earth Observation (EO). Among the available sensing modalities, the visible (Red-Green-Blue, RGB) bands remain the most accessible and widely adopted, as RGB sensors are inexpensive, robust, and produce imagery that is intuitively interpretable without specialized geophysical expertise. The corresponding body of literature is vast and rapidly expanding, spanning diverse tasks, architectures, and methodologies. In this context, the present work provides a holistic, systematic, and in-depth review of the most recent advances in deep-learning-based ORS, focusing exclusively on RGB satellite, aerial, and drone imagery. Following a structured literature review methodology, the evolution of the field is first delineated through a sequence of distinct research phases and the corresponding milestone works. Subsequently, a multi-criteria taxonomy of the literature is formed, examining the addressed tasks, the underlying neural network architectures, the adopted learning strategies, the employed acquisition platforms, the overall design objectives, and the selected application domains; for each criterion, a methodical comparative analysis of the various categories of approaches and critical insights are provided. Moreover, the publicly available ORS datasets are reviewed, accompanied by key observations. Furthermore, a comprehensive and structured discussion of the current challenges and the most promising future research directions in the field is provided.
No takes yet. Share an insight, caveat, or question.
Koletsis et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: