Remote sensing monitoring of open-pit mine's year-to-year change in the earthwork volume using AI data reconstruction

Reconstructing the 3D landform directly from planar RGB imagery enables substantial cost savings in related downstream applications, such as estimate of the yearly earthwork volume. We proposed a novel deep learning algorithm, termed P_DEMNet and an effective post-processing workflow in this study. Our innovations focus on: (1) a just-right integration of deep supervision, fractal characteristics and estimate of the uncertainties; and (2), development of mining-GIS-remote sensing joint rulesets. Our estimation over Yichun is 27,764,507 ± 1,932.3011 m3 (from 1998 to 2014) and 530,676,971 ± 6,319.411 m3 (from 2014 to 2024), and the yearly yield of metal (Li) in the Yifeng mine is at least 2410.2 tons (±51.5)​, and the Yichun Mine (open-pit parts)'s contribution of Li is 172.30 tons (±0.51), of Nb is 4.59 tons (±0.013) and of Ta is 7.72 tons (±0.023). A series of comparative/ablation experiments have demonstrated its advantages over other state-of-the-art modules with nontrivial improvement.