Key points are not available for this paper at this time.
ABSTRACT Intensive mining activities in ion‐adsorption rare earth areas have resulted in widespread bare surface land, aggravating soil erosion and ecological degradation. Accurate extraction of bare surface land from remote sensing imagery remains challenging because these areas are often fragmented, vary greatly in size, and have unclear boundaries under complex background conditions. To address this problem, this study proposes a modified U‐Net framework for fine‐scale extraction from high‐resolution unmanned aerial vehicle imagery. The model strengthens multi‐scale feature representation and improves the identification of key target regions, thereby enhancing the segmentation of fragmented bare land in complex mining environments. Experiments were conducted using high‐resolution imagery collected from the Gutian rare earth mining area in Jiangxi Province, China. The proposed method achieved about 90% F1‐score, a balanced measure of precision and recall, 81% intersection over union, and 94% recall, and outperformed several widely used semantic segmentation models under the same experimental conditions. The results show that the proposed framework can more effectively delineate bare surface land with blurred boundaries and scattered distribution, providing reliable technical support for environmental monitoring, ecological assessment, and restoration management in rare earth mining areas.
Cai et al. (Fri,) studied this question.