海洋测绘2025，Vol.45Issue(5)：16-20,5.DOI:10.3969/j.issn.1671-3044.2025.05.004
基于深度学习的航空重力浅海海域地形反演
Bathymetric inversion of shallow sea regions using aerial gravity based on deep learning
摘要
Abstract
To address the limitations of satellite altimetry gravity data in shallow water regions and improve the accuracy of bathymetric inversion,this study proposes a method based on deep learning using airborne gravity data.The approach is built upon the UNET model,enhanced with a Convolutional Block Attention Module(CBAM),and incorporates high-frequency loss and perceptual loss into the loss function to better preserve terrain details.The model is trained on synthetic topographic data and gravity anomalies derived from prism-based forward gravity modeling.It is then tested on both simulated and real airborne gravity datasets.For simulated data,the inversion yielded a mean error of 0.63 m and a standard deviation of 19.75 m,with a reasonable error distribution.In the real data test,the inverted results were compared with shipborne bathymetric data,yielding a mean error of 48.53 m and a root mean square error(RMSE)of 83.36 m.Compared with the BedMachine-v5 model,the proposed method reduced the mean absolute error by 65.21％and the RMSE by 46.26％.The findings demonstrate that the deep learning-based inversion method offers high accuracy and holds significant potential for improving bathymetric mapping in shallow coastal areas.关键词
地形反演/机载航空重力/深度学习/CBAM模块/UNET模型Key words
terrain inversion/aerial gravity/deep learning/CBAM module/UNET model分类
天文与地球科学引用本文复制引用
付天朔,叶周润,梁星辉,边少锋,柳林涛..基于深度学习的航空重力浅海海域地形反演[J].海洋测绘,2025,45(5):16-20,5.基金项目
国家自然科学基金(42430101 （）
41904010) （）
国家重点研发计划(2024YFB3908104) （2024YFB3908104）
国家自然科学青年科学基金(42204052) （42204052）
大地测量与地球动力学重点实验室开放基金(SKLGED2022-1-4). （SKLGED2022-1-4）
