首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 125 全文下载次数: 182
引用本文:

DOI:

10.11834/jrs.20211146

收稿日期:

2021-03-20

修改日期:

2021-06-02

PDF Free   EndNote   BibTeX
时间序列SAR图像不可移动文物水域淹没监测
吴樊, 李娟娟, 张波, 王超, 张红, 陈富龙, 李璐, 许璐
中国科学院空天信息创新研究院 中国科学院数字地球重点实验室
摘要:

全球气候变化引起的极端降雨、洪涝灾害是导致不可移动文物受损的重要破坏因素。合成孔径雷达(Synthetic aperture radar, SAR)具有全天时、全天候、大范围周期性对地观测的优势,是大区域水体监测的重要手段,对不可移动文物水域淹没及风险监测具有重要意义。本文利用时间序列SAR图像提出基于残差U-Net的不可移动文物水域淹没及风险监测框架。首先,基于双峰阈值分割法结合专家知识辅助进行水体样本生成,提高样本制作效率;其次,引入残差模块建立U型卷积网络,综合残差结构及U-Net的特点,缓解梯度更新时的弥散、消失等现象,通过卷积层之间累加和跳跃链接,保留多尺度的地物特征信息,以实现水体快速、高精度的语义分割;最后通过将结果与不可移动文物点位进行空间叠加分析,实现对不可移动文物水淹状况的监测。选取鄱阳湖及南昌市昌邑北垱遗址作为试验研究对象。获取了21景覆盖鄱阳湖区域不同时相Sentinel-1SAR图像,并结合Sentinel-2光学图像进行结果分析与评价。实验结果表明:本文方法在鄱阳湖试验区对水体提取总精度优于95%,相较于FCN(Fully Convolutional Networks)与U-Net方法具有更好的精度。利用不同时相SAR图像获得时间序列水体分布范围变化图,与昌邑北垱遗址点位进行空间叠加分析,得到不可移动文物水域淹没长时间序列监测结果。实验结果表明本文提出的方法可以有效提取水体范围,对不可移动文物水域淹没及风险监测可以提供有力支撑。

Inundation monitoring of immovable cultural relics with time-series SAR images
Abstract:

Immovable cultural relics are an important part of material cultural heritage. In recent years, meteorological and hydrological disasters such as flood storms and other disasters have been a huge threat to immovable cultural relics due to global climate change, torrential weather and other extreme weather. It is of great significance to carry out the extensive mapping and dynamic monitoring of water bodies in time. Synthetic aperture radar (SAR) has the advantages of all-day, all-weather and large-scale periodic earth observation, and plays a key role in the application of large-scale water body monitoring. In this paper, a framework for monitoring inundation and risk of immovable cultural heritage based on residual U-Net is proposed by using time series SAR images. Firstly, based on the bimodal threshold segmentation method and combined with expert knowledge, water sample generation was carried out to improve the efficiency of sample production. Secondly, the U-shaped convolutional network was established by introducing the residual module, which combined the characteristics of the residual structure and U-Net to alleviate the phenomena of gradient dispersion and disappearance during the value updating. By accumulating and jumping links between convolutional layers, more feature information of objects is retained to achieve rapid and high-precision semantic segmentation of water bodies. Finally, by superimposing the water extraction results with the area of immovable cultural relics, the waterlogging status of immovable cultural relics can be monitored. Poyang Lake and the relic of Changyi Beidang in Nanchang were selected for the experiments. Twenty-one Sentinel-1 images covering Poyang Lake at different dates were obtained for water body extraction, the results were analyzed and evaluated by combining corresponding Sentinel-2 optical images. The experimental results show that the overall accuracy of the proposed method in the Poyang Lake experimental area is better than 95%, and the proposed method outperforms the methods of Fully Convolutional Networks (FCN) and U-Net. Spatial superposition analysis was conducted combining the site of relic of Changyi with the monitoring results of water bodies from SAR images in long-term series. The experimental results show that the method proposed in this paper can effectively extract water bodies and has great potential for inundation and risk monitoring of immovable cultural relics.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮