AUTHOR=Zhou Hongying , Liu Suhong , Zhou Chengkai , Ma Zhiguo 

TITLE=Scale-adaptive and mask refinement modules for accurate alluvial fan boundary detection in remote sensing data

JOURNAL=Frontiers in Earth Science

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2025.1685685

DOI=10.3389/feart.2025.1685685

ISSN=2296-6463

ABSTRACT=IntroductionAlluvial fans are crucial geomorphic features in arid regions, playing key roles in geomorphic evolution, hydrological modeling, and land-use planning. However, their irregular morphology and multi-scale characteristics make accurate boundary delineation challenging for conventional remote sensing methods.MethodsTo overcome these limitations, this study proposes a multi-module enhanced Mask R-CNN framework that integrates topographic and spectral information for precise alluvial fan recognition. The model consists of a Topographic–Spectral Fusion (TSF) module, a Scale-Adaptive Module (SAM), and a Mask–Boundary Refinement (MBR) module, jointly designed to improve recognition accuracy and structural detail preservation.ResultsExperiments based on multi-source remote sensing imagery and terrain data show that the proposed model achieves an accuracy of 91.7%, precision of 89.8%, recall of 88.5%, and F1-score of 89.1% in full-region classification. For segmentation, the model attains a mean intersection over union (mIoU) of 81.5% and a boundary F1-score of 80.4%. Ablation experiments confirm that the TSF module enhances spatial–structural modeling, while the MBR module improves boundary fitting.DiscussionThe results demonstrate that the proposed framework provides robust and transferable performance across different fan size categories, achieving a minimum false negative rate of 3.9%. The method offers both theoretical value and practical applicability for accurate alluvial fan recognition in arid regions.