Abstract

Illegal river sand mining is hard to control because extraction sites are scattered along long, remote river reaches that cannot be inspected continuously. This paper reviews twelve studies that use satellite imagery, unmanned aerial vehicles (UAVs), geographic information systems (GIS) and machine or deep learning to monitor sand mining and closely related riverbank or mining processes, together with two reviews of deep-learning change detection, and compares their data, methods and published results. The comparison shows that existing work concentrates on detecting or measuring mining that has already happened, through spectral-index mapping, segmentation of mining areas, dredging-boat detection, shoreline analysis and UAV volumetry, whereas forward-looking risk prediction, and the joint use of satellite and drone data in a learning-based pipeline, are rarely addressed. Building on this gap, we propose a conceptual five-layer framework that links multi-source data acquisition, preprocessing, deep-learning detection, machine-learning risk mapping and a GIS decision-support layer with a UAV verification loop, and we define an evaluation protocol that separates training and test sites and reports class-wise metrics. The paper is a review and design study: it does not implement the framework, train a model or report new experimental results, and every number quoted comes from the cited works, which use different datasets and are not directly comparable.

Keywords

How to Cite This Article

Rights