基于深度学习的单体新月形沙丘多视图三维重建研究  

Multi-view 3D reconstruction of single barchan dune based on deep learning

作　　者：李明[1] 杨泽坤 燕洁华 李浩然 叶汪忠 LI Ming;YANG Zekun;YAN Jiehua;LI Haoran;YE Wangzhong(Energy and Transportation Engineering College,Inner Mongolia Agricultural University,Hohhot 010018,China)

机构地区：[1]内蒙古农业大学能源与交通工程学院,呼和浩特010018

出　　处：《干旱区资源与环境》2025年第1期189-199,共11页Journal of Arid Land Resources and Environment

基　　金：内蒙古自治区自然科学基金项目(2024LHMS03069);内蒙古自治区直属高校基本科研业务费基础研究项目(BR220138);内蒙古自治区高等学校科学研究项目(NJZY22520)资助。

摘　　要：沙丘形态学是风沙地貌学研究的重要内容,传统的三维重建技术在处理弱纹理和复杂几何形态的沙丘表面时存在局限性,图像特征匹配失去表征能力,直接进行三维形态重建效果不佳。文中通过无人机五向倾斜摄影技术,结合PatchmatchNet深度学习框架,实现了对新月形沙丘的高精度、高完整度三维重建。通过与传统三维重建软件的主观目视评估比较,验证了所提方法在沙丘形态完整性和结构准确性方面的优势。利用实地勘测的RTK精度评价点集进行精度误差计算,结果表明重建沙丘在X、Y、Z各单个方向上均方根误差均处于7cm以内,最大为6.587cm,最小仅为0.653cm,平均误差均处于2cm以内,标准差最大为8.838cm,均达到了厘米级精度,证明文中方法在捕捉新月形沙丘整体结构和恢复三维形态方面具有较高的精度和可靠性,对新月形沙丘形态特征参数测量、形态发育演变过程等研究具有重要指导意义。The three-dimensional(3D)morphology of barchan dunes is an essential aspect of aeolian geomorphology research.There are limitations in traditional 3D reconstruction techniques when dealing with dune surfaces with weak textures and complex geometries,which leads to a loss of representational capacity in image feature matching,and unsatisfactory outcomes in direct 3D morphological reconstruction.This study utilizes an unmanned aerial vehicle(UAV)with five-directional oblique photography technology,combined with the PatchmatchNet deep learning framework,to achieve high-precision and high-integrity 3D reconstruction of barchan dunes.By comparing subjective visual assessments with traditional 3D reconstruction software,the advantages of the proposed method in terms of dune morphology integrity and structural accuracy have been validated.Using the field survey's RTK precision evaluation point set to calculate the accuracy error,the results indicate that the root mean square errors of the reconstructed dunes in the X,Y,and Z individual directions is within 7cm,with a maximum of 6.587cm and a minimum of only 0.653cm.The average errors are within 2cm,and the maximum standard deviation is 8.838cm,achieving centimeter-level precision.This proves that the method presented in this paper has high accuracy and reliability in capturing the overall structure of barchan dunes and restoring their 3D morphology,which is of significant guidance for the researches in measurement of morphological feature parameters and morphological development and evolution processes of barchan dunes.

关 键 词：新月形沙丘 深度学习 三维重建 无人机多视图 

分 类 号：P931.3[天文地球—自然地理学]