测绘学报 ›› 2026, Vol. 55 ›› Issue (10): 1852-1866.doi: 10.11947/j.AGCS.2026.20260228

• 摄影测量学与遥感 • 上一篇    下一篇

LG-SSDA-MSIFNet:LiDAR数据引导的复垦矿区多光谱影像超分辨率融合模型

冯媛媛1,2(), 李朝奎2, 李婷1,2(), 雷敏1,3, 吴柏燕2   

  1. 1.湖南科技大学资源环境与安全工程学院,湖南 湘潭 411201
    2.湖南科技大学测绘遥感信息工程湖南省重点实验室,湖南 湘潭 411201
    3.湖南省第二测绘院,湖南 长沙 410118
  • 收稿日期:2026-06-12 修回日期:2026-10-12 出版日期:2026-11-16 发布日期:2026-11-16
  • 通讯作者: 李婷 E-mail:1298791880@qq.com;155405544@qq.com
  • 作者简介:冯媛媛(1993—),女,博士生,研究方向为矿山环境保护与时空智能建模。E-mail:1298791880@qq.com
  • 基金资助:
    自然生态系统碳汇监测湖南省工程研究中心重点开放课题(2026HNNCS01);安徽省重点实验室开放课题(2024PGE001)

LG-SSDA-MSIFNet: a LiDAR-guided super-resolution fusion model for multispectral imagery of reclaimed mining areas

Yuanyuan Feng1,2(), Chaokui Li2, Ting Li1,2(), Min Lei1,3, Baiyan Wu2   

  1. 1.School of Resource Environment and Safety Engineering, Hunan University of Science and Technology, Xiangtan 411201, China
    2.Hunan Provincial Key Laboratory of Surveying, Remote Sensing and Information Engineering, Hunan University of Science and Technology, Xiangtan 411201, China
    3.The Second Surveying and Mapping Institute of Hunan Province, Changsha 410118, China
  • Received:2026-06-12 Revised:2026-10-12 Online:2026-11-16 Published:2026-11-16
  • Contact: Ting Li E-mail:1298791880@qq.com;155405544@qq.com
  • About author:Feng Yuanyuan (1993—), female, PhD candidate, majors in mining environmental protection and spatio-temporal intelligent modeling. E-mail: 1298791880@qq.com
  • Supported by:
    Hunan Engineering Research Center of Natural Ecosystems Carbon Sink Monitoring(2026HNNCS01);Open Research Fund of Anhui Provincial Key Laboratory(2024PGE001)

摘要:

针对复垦矿区地物混杂、常规多光谱影像难以支撑精细化监测的问题,本文提出了一种LiDAR引导的超分融合模型(LG-SSDA-MSIFNet)。模型将LiDAR五维特征按物理生成机制解耦为空间几何与地物辅助组,通过双分支编码(BFE)与空间-光谱解耦注意力(SSDA)实现零初始化加性残差融合,使各分支贡献可区分并定位至对应物理分组,增强模型可解释性。以湘乡白云石矿(WLS)和湘潭锰矿(XTMK)为研究区,试验结果表明:WLS上模型在SAM(0.006 9 rad)、PSNR(50.91 dB)、SSIM(0.998 0)等6项指标上均显著优于GF、GLP、HSV等传统方法及EDSR、RCAN、PanNet-L 3种深度模型;迁移至XTMK后,PSNR仍达49.63 dB(较深度方法提升约8.9 dB),保持高精度与稳定性。宏观统计与微观光谱验证相互印证,为复垦矿区遥感融合探索了一种高保真、跨域稳定的物理语义引导思路。

关键词: 复垦矿区, 多光谱影像, LiDAR, 超分辨率重建, 空间-光谱解耦注意力, 跨区域泛化

Abstract:

To address mixed ground objects and the difficulty of conventional multispectral imagery in supporting fine monitoring in reclaimed mining areas, a LiDAR-guided super-resolution fusion model (LG-SSDA-MSIFNet) is proposed. The model decouples 5D LiDAR features into spatial-geometric and auxiliary groups based on physical mechanisms. Via bifurcated feature encoding (BFE) and spatial-spectral decoupled attention (SSDA), zero-initialized additive residual fusion is achieved, enabling distinguishable branch contributions localized to physical groups, thus enhancing interpretability. Experiments on the Xiangxiang Dolomite Mine (WLS) and the Xiangtan Manganese Mine (XTMK) show that on WLS, the model significantly outperforms traditional methods (GF, GLP, HSV) and three deep learning models (EDSR, RCAN, PanNet-L) across six metrics including SAM (0.006 9 rad), PSNR (50.91 dB), and SSIM (0.998 0). When transferred to XTMK, PSNR reaches 49.63 dB (an improvement of approximately 8.9 dB), maintaining high accuracy and stability. Macroscopic and microscopic validations corroborate each other, providing a high-fidelity, cross-domain stable, and physically semantic-guided approach for remote sensing fusion in reclaimed mining areas.

Key words: reclaimed mining areas, multispectral imagery, LiDAR, super-resolution reconstruction, spatial-spectral decoupled attention, cross-domain generalisation

中图分类号: