A Sentinel-2 Image Dataset for Mining Detection Across Mining Proportion Ranges in the Brazilian Legal Amazon

The recent expansion of mining in the Amazon is a major driver of environmental degradation. Machine learning and satellite imagery are effective tools to monitor this problem, but most datasets treat mining detection as a strictly binary task. This ignores how the actual mining proportion inside an image affects classification results. We present a Sentinel-2 dataset designed to explore this behavior. It contains 2,600 image patches collected across the Brazilian Legal Amazon in 2024 using Google Earth Engine and MapBiomas Collection 10 data. The dataset includes metadata detailing geographic locations, mining proportion ranges in 10% intervals, and a suggested practical split for experiments.