Artisanal and Small-Scale Gold Mining Detection in the Amazon Forest Using Contextual Data

Artisanal and small-scale gold mining (ASGM) is a major driver of land cover change in the Amazon, often challenging to detect due to its spectral similarity with other surface features. This study investigates whether incorporating contextual geospatial data from OpenStreetMap (OSM) alongside Sentinel-2 imagery can improve ASGM detection in Venezuela's Bolivar state. Rasterized OSM-derived semantic mask are appended as additional input channels to the satellite imagery and processed through a CNN. This setup enables joint learning of spectral and contextual features, allowing for a more accurate and reliable distinction between ASGM sites and spectrally similar land uses.