自然资源遥感, 2023, 35(3): 145-152 doi: 10.6046/zrzyyg.2022197

技术方法

联合改进U-Net模型和D-InSAR技术采矿沉陷提取方法

林佳惠,1,2,3, 刘广1,2,3, 范景辉,4, 赵红丽4, 白世彪5,6, 潘宏宇1,2,3

1.中国科学院空天信息创新研究院数字地球重点实验室,北京 100094

2.可持续发展大数据国际研究中心,北京 100094

3.中国科学院大学,北京 100049

4.中国自然资源航空物探遥感中心,北京 100083

5.南京师范大学海洋科学与工程学院,南京 210023

6.中国科学院、水利部成都山地灾害与环境研究所,中国科学院山地灾害与地表过程重点实验室,成都 610041

Extracting information about mining subsidence by combining an improved U-Net model and D-InSAR

LIN Jiahui,1,2,3, LIU Guang1,2,3, FAN Jinghui,4, ZHAO Hongli4, BAI Shibiao5,6, PAN Hongyu1,2,3

1. Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

2. International Research Center of Big Data for Sustainable Development Goals, Beijing 100094, China

3. University of Chinese Academy of Sciences, Beijing 100049, China

4. China Aero Geophysical Survey and Remote Sensing Center for Natural Resources, Beijing 100083, China

5. College of Marine Sciences and Engineering, Nanjing Normal University, Nanjing 210023, China

6. Key Laboratory of Mountain Hazards and Earth Surface Processes, Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu 610041, China

通讯作者: 范景辉(1978-),男,博士,教授级高级工程师,主要从事InSAR技术应用研究以及3S技术在自然资源领域的应用。Email:jhfan2004@qq.com。

责任编辑: 陈昊旻

收稿日期: 2022-05-16   修回日期: 2022-10-31  

基金资助: 国家重点研发计划项目“高亚洲和北极积雪-冰川与地质灾害监测技术及示范应用”(2021YFE0116800)
中欧龙计划5期合作项目“Integration of multisource remote sensing data to detect and monitoring large and rapid landslides and use of artificial intelligence for cultural heritage preservation”(56796)
可持续发展大数据国际研究中心创新研究计划(CBAS2022IRP02)
国家自然科学基金项目“青藏高原露天煤矿排土场地形-土壤-植被响应机理及地貌重塑研究”(41977415)

Received: 2022-05-16   Revised: 2022-10-31  

作者简介 About authors

林佳惠(1998-),女,硕士研究生,主要从事InSAR数据处理与应用。Email: linjiahui20@mails.ucas.ac.cn。

摘要

矿产资源开采导致的地表沉陷不仅是矿区国土空间开发利用需要考虑的重要因素,而且对地下非法开采的区域具有明显的指征作用。矿产资源开采一般具有分布范围较广、分布不均且较分散的特点,因此快速、准确地识别并提取大区域内采矿沉陷的空间分布非常必要。本研究基于合成孔径雷达差分干涉测量技术(differential interferometric synthetic aperture Radar,D-InSAR)得到矿区多时相差分干涉相位图,并使用深度学习FCN-8s,PSPNet,Deeplabv3和U-Net模型训练网络开展采矿沉陷智能识别,结果显示U-Net模型具有较高的检测精度且用时较短。为提高采矿沉陷的语义分割提取精度,在传统U-Net模型中引入高效通道注意力模块进行训练。结果表明改进的U-Net模型与传统模型相比,在测试集上采矿沉陷对应的交并比提升2.54百分点,为大范围采矿沉陷时空分布提取问题提供新的解决方法。

关键词： U-Net模型; D-InSAR; 采矿沉陷提取; 语义分割; 注意力模块

Abstract

Surface subsidence caused by the exploitation of mineral resources must be considered during the development and utilization of land and space in mining areas. Furthermore, it serves as a significant indication of underground areas subjected to illicit mining. The exploitation of mineral resources is generally conducted in widespread, uneven, and dispersed areas, making it necessary to quickly and accurately identify and extract the spatial distribution of mining subsidence in large areas. This study determined the multitemporal differential interferometric phase diagram of mining areas using the differential interferometric synthetic aperture Radar (D-InSAR) technique. Furthermore, it trained networks for the intelligent identification of mining subsidence by employing deep-learning FCN-8s, PSPNet, Deeplabv3, and U-Net models. The results show that the U-Net model enjoys a high detection accuracy and a short detection time. To improve the semantic segmentation and extraction accuracy of information about mining subsidence, this study introduced the efficient channel attention (ECA) module into the conventional U-Net model during the training. Compared with the conventional model, the improved U-Net model increased the intersection over union (IOU) corresponding to mining subsidence by 2.54 percentage points.

Keywords： U-Net model; D-InSAR; extraction of mining subsidence; semantic segmentation; attention module

PDF (3020KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

林佳惠, 刘广, 范景辉, 赵红丽, 白世彪, 潘宏宇. 联合改进U-Net模型和D-InSAR技术采矿沉陷提取方法[J]. 自然资源遥感, 2023, 35(3): 145-152 doi:10.6046/zrzyyg.2022197

LIN Jiahui, LIU Guang, FAN Jinghui, ZHAO Hongli, BAI Shibiao, PAN Hongyu. Extracting information about mining subsidence by combining an improved U-Net model and D-InSAR[J]. Remote Sensing for Land & Resources, 2023, 35(3): 145-152 doi:10.6046/zrzyyg.2022197

0 引言

矿产资源的开采导致地表发生沉陷,这会造成土地损毁[1]、 生态环境修复困难[2]、矿区水环境遭到破坏[3]、房屋和基础设施受损[4],甚至会引起滑坡等地质灾害[5],对人民的生产和生活造成较大的危害。除此之外,矿区非法开采的监测也是各级政府面临的一个难题,一方面非法开采给国家造成了财产损失,同时非法开采的不规范性往往给开采人员带来更大的安全风险。对大范围地区的采矿沉陷进行快速、准确监测,识别并提取其范围能为国土资源管理提供重要依据,而国土资源的合理规划和利用有助于保障农业发展和粮食安全,这与可持续发展目标(sustainable development goals,SDG)中目标2的宗旨相符。对矿区全方位监测管理可减少采矿对水环境和生态环境的影响,有利于SDG目标6水资源和目标15生态系统的可持续发展和管理。

传统的采矿沉陷监测技术包括水准仪、全站仪以及GPS-RTK技术[6]等,但这些方法只能监测到点状数据、监测周期长且偏远山区不利于安装测量站、测量成本高[7]。合成孔径雷达差分干涉测量技术(differential interferometric synthetic aperture Radar,D-InSAR)能够大范围、高精度监测地表形变,且具有全天时全天候工作、穿云透雾、获取成本较低和空间分辨率较高的特点,在区域上快速识别采矿沉陷具有独特优势[8-9]。

Carnec等[10]首次利用D-InSAR技术监测由煤矿开采引起的地表缓慢沉降。之后有许多研究人员将D-InSAR技术应用于采矿沉陷监测[7,11⇓-13]和滑坡监测[14]中,时序InSAR技术如永久散射体合成孔径雷达干涉测量技术(persistent scatterer interferometric synthetic aperture Radar technology,PS-InSAR)、小基线集合成孔径雷达干涉测量技术(small baseline subset interferometric synthetic aperture Radar technology,SBAS-InSAR)和融合分布式散射体合成孔径雷达干涉测量技术(distributed scatterers interferometric synthetic aperture Radar technology,DS-InSAR)也得到了广泛的应用,监测精度被证实可以达到厘米级,甚至毫米级[15⇓⇓⇓⇓-20]。在应对快速响应需求时,对D-InSAR 监测结果采用人工解译的方法圈定采矿沉陷范围所需时间较长[21],因此半自动或自动提取采矿沉陷区域具有重要的研究意义。

传统的提取方法主要依据采矿沉陷呈圆形或椭圆形的形状特征,采用代数距离法[22]、Hough变换[23]、Circlet变换 [24]和Gabor滤波器[25]进行识别。但是这几种方法都存在各自的不足之处。其中,代数距离法能实现对椭圆形状的检测,然而这种方法效率较低[22]。Circlet变换检测的采矿沉陷识别率高于Hough变换,但这2种方法都存在漏检和误检的情况[24]。基于Gabor滤波器方法识别率在30%~53%之间,该方法只能完成圆形沉陷的初步识别,无法实现完全自动化的可靠检测[25]。相较之下,深度学习提取的特征不仅仅局限于目标的形状特征,还包括颜色、大小、位置和纹理等。利用语义分割模型实现精确、高效、自动化的采矿沉陷识别和提取是研究趋势之一。目前已有少量研究验证了将深度学习应用于小区域内采矿沉陷[26⇓-28]和单个火山沉陷[29⇓-31]监测中的可行性,且能达到较高的分类精度,这为大区域的采矿沉陷空间分布提取提供了新思路。Wu等[32]通过语义分割DDNet网络发现山西省大同市134个采矿沉陷区域,但是未评价模型的识别精度。考虑到当前较少开展不同深度学习模型提取采矿沉陷的性能比较研究,本文通过像素准确率(pixel accuracy,PA)和平均交并比(mean intersection over union,MIoU)对比各语义分割模型的应用差异,并针对模型优选和改进做进一步研究。

本文首先利用DInSAR 监测技术联合FCN[33],PSPNet[34],Deeplabv3[35]和U-Net[36]模型获取采矿沉陷范围的空间分布并进行对比分析; 为减少采矿沉陷小样本漏分及分割边缘粗糙现象的发生,本文提出一种引入高效通道注意力(efficient channel attention,ECA)模块[37]的U-Net模型,注意力模块能使网络有效地关注目标信息的同时忽略无关信息。将ECA模块引入U-Net模型可以在较少增加模型计算量的同时提高U-Net模型精度。

1 研究区概况及其数据源

哨兵一号卫星(Sentinel-1A)于2014 年4月发射,该卫星包含上升轨道数据和下降轨道数据,IW模式数据幅宽为250 km,这对于获取大区域数据提供了基础,重复轨道周期为12 d,时间失相干相对较小,所获得的干涉图相干性较好。本文采用的Sentinel-1A数据为Path113_Frame116和Path113_Frame121这2个图幅的数据,像幅中心的距离向和方位向像元尺寸分别为2.3 m和13.9 m,该数据位于中国山西省(如图1,底图来自天地图影像(墨卡托,WGS1984),网址: http://lbs.tianditu.gov.cn/),考虑到冬季植被相对其他季节较少,数据相干性较高,因此选定的数据时间跨度为2019年10月11日—12月22日。

图1

新窗口打开| 下载原图ZIP| 生成PPT

图1   研究区地理位置示意图

Fig.1   Geographical location of the study area


数字高程模型(digital elevation model,DEM)是D-InSAR数据处理的必要基础资料。本研究采用由美国国家航空航天局通过SRTM计划获取的空间分辨率约为30 m的DEM数据。

2 数据处理与技术流程

联合使用深度学习模型和D-InSAR技术进行采矿沉陷空间分布智能提取,包括利用FCN,PSPNet和Deeplabv3这3种传统的模型以及U-Net模型。采用二轨法生成差分干涉图,技术流程如图2所示。

图2

新窗口打开| 下载原图ZIP| 生成PPT

图2   技术流程图

Fig.2   Technical flow chart


2.1 数据预处理

首先利用干涉数据处理软件GAMMA对数据进行图像预处理,包括影像配准、多视和滤波等。本文所处理的干涉像对如表1所示。配准将同一地区的2景雷达影像变换到同一坐标系下,在像素层上得到最佳匹配。多视能有效地减少椒盐噪声,提高图像信噪比,增加在低相干区域的高相干点个数,本文所采用的距离向、方位向多视比为5∶1。由于时空失相干和大气效应的影响,干涉图通常具有相位噪声,为了减少相位噪声对卷积神经网络训练的影响,需要对图像进行滤波处理。采用的滤波为自适应滤波,滤波的指数参数设置为0.6,滤波窗口大小设置为64,其他参数为默认值。

表1   实验所用的差分干涉像对

Tab.1  Differential interference image pairs used in the experiment

Path113_Frame116Path113_Frame121
20191116,2019112820191011,20191104
20191116,2019121020191104,20191128
20191128,2019121020191116,20191128
20191128,2019122220191128,20191210
20191210,2019122220191210,20191222

新窗口打开| 下载CSV


2.2 制作语义分割数据集

采用Labelme软件标记沉陷矿区和背景区域,一般采矿沉陷区呈钟形或椭圆形,沉陷量从中心到边缘逐渐减小。通过人工目视解译方法对相位图进行样本标注,对于无法确定的区域,采用地理编码后的相位图与实际地物图进行对比最终标定沉陷区域范围。由于遥感影像尺寸较大,规范输入影像的尺寸大小可有效减少训练时长,因此采用滑动窗口裁剪的方法对标记的大幅影像进行裁剪,影像大小为256像素×256像素。最后选取1 625张包含沉陷矿区的影像,本文将样本随机划分为训练集和测试集,训练集影像共1 463张,测试集影像共162张。采矿沉陷数据集部分样本如表2所示。标记后背景颜色为黑色,沉陷矿区颜色为白色。

表2   采矿沉陷相位图和样本标记图

Tab.2  Mining subsidence phase diagram and labeled diagram

样本1样本2样本3
相位图
标记图

新窗口打开| 下载CSV


2.3 模型训练及实验环境

使用PyTorch作为深度学习框架搭建实验环境,并进行模型训练、调试参数及测试。计算机配置: CPU为E5-2630 v2,GPU为NVIDIA GeForce GTX TITAN,显存大小为6.0 GB,编程语言为Python3.7,深度学习框架为Pytorch1.2,CUDA为10.0。使用FCN,PSPNet和Deeplabv3这3种传统的模型以及U-Net模型进行采矿沉陷提取。语义分割的首个模型是FCN模型,其实现了端到端的逐像素分类[33]。Deeplabv3模型和PSPNet模型分别用了空洞空间金字塔池化模块和金字塔池化模块,都具有较高的分割精度[29-30]。

实验所用的批大小为4,迭代周期epoch最大值设置为50,实验所用的模型可在最大epoch范围内达到收敛。所使用的损失函数为交叉熵损失函数。每个epoch计算出MIoU,实验程序记录MIoU值最大时对应的模型,选择此模型作为测试集评价模型。为使损失函数达到最小值,使用带动量的随机梯度下降法加快收敛速度。初始学习率设置为0.001,学习率随epoch动态衰减,本实验学习率衰减策略为poly指数变换策略[38],学习率lr的公式为:

lr=lr0(1-epochmax_epoch)0.9

式中: lr0为初始学习率; max_epoch为最大迭代周期。

2.4 语义分割精度评价

PA和MIoU是常见的语义分割精度评价指标,常用于评价FCN[33],PSPNet[34]和Deeplabv3+[39]等模型的识别精度。其中PA的计算公式为:

PA=∑inii∑iti

式中: nii为类别i被预测正确的像素个数,则Σinii为所有类别被预测正确的像素个数; ti为类别i的像素个数,则Σiti为图像总像素个数。

MIoU的计算公式为:

MIoU=1ncl∑iniiti+∑jnji-nii

式中: nji为类别j被预测成类别i的个数; ncl为目标类别的个数。

3 改进的U-Net模型

3.1 U-Net模型

U-Net模型属于对称的编码-解码结构,其中编码结构对应模型左侧部分,也被称为收缩路径,这部分结构实现了特征提取和下采样的功能。右侧部分是扩展路径,作为解码结构通过4次上采样得到最终的分割图。3×3卷积层的步距为1,没有使用填充。卷积后的图形尺寸大小N计算公式为:

N=W-F+2PS+1

式中: W为输入图像的尺寸大小; F为卷积核大小; P为填充像素的个数; S为步距。由式(4)可知经过3×3卷积后图像长和宽减少2个像素。经过步距为2的池化下采样层图像尺寸大小会减少一半。原U-Net模型左侧特征图需要进行中心裁剪使得其尺寸大小与右侧上采样图像尺寸大小一致,再进行跳跃连接操作即concat拼接,使得图像分割结果的空间位置信息损失减小。输出图像的类别个数与使用的1×1卷积核个数一致。

原U-Net模型由于没有加入填充导致输入图像的大小与分割结果图大小不一致。为使输入图像能与分割尺寸一致,使用3×3的卷积层且填充值为1,这使得每次卷积后特征层的高度和宽度不发生改变,在进行拼接时也不需要对下采样特征图进行中心裁剪。上采样时使用的双线性插值操作不改变通道数。

3.2 引入ECA模块的U-Net模型

通道注意力模块的核心思想是不同通道的特征图对于识别任务的重要程度是不相同的,为了让网络更加关注有效通道的特征图,对每个通道的重要程度进行权重赋值。

ECA模块作为通道注意力模块的一种,采用一维卷积实现不降维的局部跨通道交互,通过自适应选择一维卷积核大小k来确定跨通道交互的范围。k的取值计算公式[37]为:

k=ψ(C)=lb(C)γ+bγodd

式中: C为通道维度; |t|odd为t的取值为最接近的奇数; γ和b分别为2和1。 ECA模块首先对输入维度[W,H,C]的特征图(H为图像高度)进行全局平均池化操作(global average pooling,GAP),减少参数数量后将参数存储在大小为1×1×C的矩阵中,进行自适应卷积核大小为k的一维卷积运算得到特征图各通道权重,用Sigmoid函数将权重进行了归一化,再将归一化权重与原特征图进行逐通道相乘,得到的结果作为下一级的输入特征图。

本文模型结构的改进在于在U-Net模型编码部分加入4个ECA模块[37],通过计算各个通道的权重重新调整了编码部分的特征。ECA-UNet模型结构如图3所示。

图3

新窗口打开| 下载原图ZIP| 生成PPT

图3   ECA-UNet模型结构

Fig.3   ECA-UNet model structure


4 结果与分析

本文使用的模型包括FCN,PSPNet,Deeplabv3和U-Net及引入ECA模块的U-Net模型。其中FCN的主干采用的是VGG16网络[40]。PSPNet和Deeplabv3所使用的主干是残差网络(Resnet50),深度残差网络的提出解决了深度网络退化问题[41]。

模型训练后需通过测试对比不同模型在差分干涉相位图中采矿沉陷提取性能的差异。不同模型在测试集上的精度评价如表3所示。精度评价将人工目视解译标记的样本图作为真值与各模型预测结果进行比较。由于遥感影像上背景像元数量远远多于采矿沉陷像元数量,受背景被正确分类的影响,各模型的PA值和交并比(intersection over union,IoU)(背景)都较高,PA值和IoU(背景)值不能较好地评价模型精度差异。本文关注的是采矿沉陷范围提取,因此采矿沉陷的交并比即IoU(采矿沉陷)更能反映模型间的差异。从表3可知,针对FCN的3种模型,FCN-8s的各项指标均优于FCN-16s和FCN-32s,这和FCN模型首次被提出时的结论一致[33]。由于FCN使用的特征提取网络VGG16层数较少,因此FCN模型在所有模型中所需的训练时长最少。PSPNet和Deeplabv3模型均采用Resnet50作为特征提取网络,由于Resnet50包含49层卷积层和一层全连接层,网络层数较多,所以训练时长较长。Deeplabv3 模型虽然在采矿沉陷的IoU上略高于U-Net模型,但其消耗的训练时间也更多。

表3   不同模型在测试集上的精度评价

Tab.3  Accuracy evaluation of different models on test set

模型PA/%MIoU
/%		IoU
(背景)
/%		IoU
(采矿
沉陷)/%		训练
时间/h
FCN-32s97.7173.6898.2149.151.84
FCN-16s98.1676.6198.3854.841.88
FCN-8s98.2978.1998.5157.872.11
PSPNet98.1679.2198.6559.776.50
Deeplabv398.2079.7198.5760.857.50
U-Net98.3179.2498.2760.205.08
本文模型98.5580.5898.4162.746.36

新窗口打开| 下载CSV


本文使用的方法在IoU(采矿沉陷)指标上均优于其他模型,且相比传统U-Net网络提升了2.54百分点。说明本文提出的模型能较好地应用于采矿沉陷提取。从测试集中选取3张影像做对比分析如表4所示。针对场景1,本文模型相对于传统U-Net模型边缘细节信息更加明显,与目视解译的采矿沉陷标记图相比更相近。针对场景2,背景不复杂的小采矿沉陷,各个模型识别的差异较小,都能较好地完成识别任务。针对噪声较大的场景3,U-Net模型会将部分噪声错分为沉陷区域,而改进的模型能有效避免这种情况。

表4   不同模型的结果

Tab.4  Results of different models

场景相位图标记图FCN-8sPSPNetDeeplabv3U-Net本文模型
场景1
场景2
场景3

新窗口打开| 下载CSV


根据实验结果发现本文方法在训练时长和分割效果上性能较好,因此应用本文模型对大范围差分干涉图进行采矿沉陷区提取,差分干涉相位图和提取结果如图4所示。通过观察对比发现本文方法在预测上较少出现采矿沉陷区漏分现象,对于不同尺度的采矿沉陷区都具备识别能力。由于本文监测的是沉陷矿区,当时间跨度内地表形变发生较大变化时,会导致地面失相干。而水体在差分干涉图中通常表现为低相干地物,因此当卷积神经网络提取到采矿沉陷的低相干性特征后,部分低相干水体及邻近区域会被错分为沉陷矿区,如图4中的椭圆标记1和2均是被错分的水体,图5中展示了部分被错分为采矿沉陷的水体地理编码后的相位图和实景图。

图4

新窗口打开| 下载原图ZIP| 生成PPT

图4   采矿沉陷相位图及提取结果

Fig.4   Mining subsidence phase diagram and extraction result


图5

新窗口打开| 下载原图ZIP| 生成PPT

图5   被错分为采矿沉陷区的水体

Fig.5   Water bodies wrongly divided into mining subsidence areas


5 结论

对采矿沉陷形变区域快速提取有利于在InSAR遥感大数据中及时识别采矿沉陷并缩小后续形变解算的范围,从而降低计算量。目前国内外利用D-InSAR真实干涉结果图做深度学习提取采矿沉陷的应用研究缺少各语义分割网络模型间的精度比较。在此前提下,本文构建了Sentinel-1A的D-InSAR采矿沉陷数据集,建立了一种联合使用深度学习模型和D-InSAR技术进行采矿沉陷空间分布智能提取的方法。发现本文提出的ECA-UNet模型在PA和MIoU指标分别达到98.55%和80.58%,均高于FCN8s,PSPNet,Deeplabv3和U-Net模型,分割的结果更加精细化,这将为大范围采矿沉陷时空分布提取问题的解决提供新方法。但在对大范围采矿沉陷提取时,发现部分水体相位噪声区域被误分为采矿沉陷的情况,未来将在数据集的构建上进行改善。

参考文献

胡振琪, 龙精华, 张瑞娅, 等.

中国东北多煤层老矿区采煤沉陷地损毁特征与复垦规划

[J]. 农业工程学报, 2017, 33(5):238-247.

[本文引用: 1]

Hu Z Q, Long J H, Zhang R Y, et al.

Damage characteristics and reclamation planning of coal mining subsidence land in old mining areas in northeast China

[J]. Transactions of the Chinese Society of Agricultural Engineering, 2017, 33(5):238-247.

[本文引用: 1]

刘辉, 朱晓峻, 程桦, 等.

高潜水位采煤沉陷区人居环境与生态重构关键技术:以安徽淮北绿金湖为例

[J]. 煤炭学报, 2021, 46(12):4021-4032.

[本文引用: 1]

Liu H, Zhu X J, Cheng H, et al.

Key technology of human environment and ecological reconstruction in high submersible level coal mining subsidence area:A case study from Lyujin Lake,Huaibei

[J]. Journal of China Coal Society, 2021, 46(12):4021-4032.

[本文引用: 1]

Ignacy D.

Comprehensive method of assessing the flood threat of artificially drained mine subsidence areas for identification and sustainable repair of mining damage to the aquatic environment

[J]. Water Resources and Industry, 2021, 26:100153.

DOI:10.1016/j.wri.2021.100153      URL     [本文引用: 1]

李佳洺, 余建辉, 张文忠.

中国采煤沉陷区空间格局与治理模式

[J]. 自然资源学报, 2019, 34(4):867-880.

DOI:10.31497/zrzyxb.20190415      [本文引用: 1]

大面积的采煤沉陷区引发严重的社会和环境问题,得到政府和学术界的广泛关注。与传统以自然条件为基础的沉陷区复垦研究不同,考虑采煤沉陷区自然生态因素和区域经济发展条件,从综合治理的角度出发,分析中国采煤沉陷区整体格局和面临的社会经济风险,深入研究各地采煤沉陷区综合治理路径。结果表明：中国采煤沉陷区面积预计超过60000 km<sup>2</sup>,其中与城乡建设用地和耕地叠压的面积分别达到4500 km<sup>2</sup>和26000 km<sup>2</sup>,涉及人口达2000万左右,其中山西和山东两省采煤沉陷区的影响最为严重;从区域特征来看,中国采煤沉陷区有开发利用、环境修复、民生保障、异地搬迁四大主要治理导向,进一步结合社会经济和空间特征,可以将沉陷区分为环境适应发展型、基础设施完善型、特色产业带动型、环境修复型、民生保障型、异地搬迁型六个治理类型。

Li J M, Yu J H, Zhang W Z.

Spatial distribution and governance of coal-mine subsidence in China

[J]. Journal of Natural Resources, 2019, 34(4):867-880.

DOI:10.31497/zrzyxb.20190415      URL     [本文引用: 1]

Donnelly L J, De La Cruz H, Asmar I.

The monitoring and prediction of mining subsidence in the Amaga,Angelopolis,Venecia and Bolombolo Regions,Antioquia,Colombia

[J]. Engineering Geology, 2001, 59(1-2):103-114.

DOI:10.1016/S0013-7952(00)00068-5      URL     [本文引用: 1]

Liu C, Zhou F, Gao J X, et al.

Some problems of GPS RTK technique application to mining subsidence monitoring

[J]. International Journal of Mining Science and Technology, 2012, 22(2):223-228.

DOI:10.1016/j.ijmst.2012.03.001      URL     [本文引用: 1]

Wang Z, Wang Z, Liu G, et al.

Monitoring the coal mining subsidence in Jibei mine area using D-InSAR technique

[C]// 2009 International Conference on Information Engineering and Computer Science.IEEE, 2009:1-4.

[本文引用: 2]

刘广, 郭华东, Ramon H, 等.

InSAR技术在矿区沉降监测中的应用研究

[J]. 国土资源遥感, 2008, 20(2):51-55.doi:10.6046/gtzyyg.2008.02.13.

[本文引用: 1]

Liu G, Guo H D, Ramon H, et al.

The application of InSAR technology to mining area subsidence monitoring

[J]. Remote Sensing for Land and Resources, 2008, 20(2):51-55.doi:10.6046/gtzyyg.2008.02.13.

[本文引用: 1]

Guéguen Y, Deffontaines B, Fruneau B, et al.

Monitoring residual mining subsidence of Nord/Pas-de-Calais coal basin from differential and persistent scatterer interferometry (Northern France)

[J]. Journal of Applied Geophysics, 2009, 69(1):24-34.

DOI:10.1016/j.jappgeo.2009.02.008      URL     [本文引用: 1]

Carnec C, Massonnet D, King C.

Two examples of the use of SAR interferometry on displacement fields of small spatial extent

[J]. Geophysical Research Letters, 1996, 23(24):3579-3582.

DOI:10.1029/96GL03042      URL     [本文引用: 1]

Ge L, Chang H C, Rizos C.

Mine subsidence monitoring using multi-source satellite SAR images

[J]. Photogrammetric Engineering and Remote Sensing, 2007, 73(3):1742-1745.

[本文引用: 1]

Wang Z, Zhang J, Liu G.

Measuring land subsidence by PALSAR interferometry in Yanzhou coal mine area

[C]// International Conference on Image Processing and Pattern Recognition in Industrial Engineering.SPIE, 2010, 7820:815-822.

[本文引用: 1]

白泽朝, 汪宝存, 靳国旺, 等.

Sentinel-1A数据矿区地表形变监测适用性分析

[J]. 国土资源遥感, 2019, 31(2):210-217.doi:10.6046/gtzyyg.2019.02.29.

[本文引用: 1]

Bai Z C, Wang B C, Jin G W, et al.

Applicability analysis of ground deformation monitoring in mining area by Sentinel-1A data

[J]. Remote Sensing for Land and Resources, 2019, 31(2):210-217.doi: 10.6046/gtzyyg.2019.02.29.

[本文引用: 1]

张腾, 谢帅, 黄波, 等.

利用Sentinel-1和ALOS-2数据探测茂县中部活动滑坡

[J]. 国土资源遥感, 2021, 33(2):213-219.doi:10.6046/gtzyyg.2020206.

[本文引用: 1]

Zhang T, Xie S, Huang B, et al.

Detection of active landslides in central Maoxian County using Sentinel-1 and ALOS-2 data

[J]. Remote Sensing for Land and Resources, 2021, 33(2):213-219.doi:10.6046/gtzyyg.2020206.

[本文引用: 1]

He Q, Zhang Y, Wu H, et al.

Mining subsidence monitoring with modified time-series SAR interferometry method based on the multi-level processing strategy

[J]. IEEE Access, 2021, 9:106039-106048.

DOI:10.1109/ACCESS.2021.3099633      URL     [本文引用: 1]

Xing X M, Zhu J J, Wang Y Z, et al.

Time series ground subsidence inversion in mining area based on CRInSAR and PSInSAR integration

[J]. Journal of Central South University, 2013, 20(9):2498-2509.

DOI:10.1007/s11771-013-1762-x      URL     [本文引用: 1]

Lesniak A, Porzycka S, Graniczny M.

Subsidence analysis in mining area of Dabrowskie coal basin using PSInSAR technique

[C]// 13th European Meeting of Environmental and Engineering Geophysics.European Association of Geoscientists and Engineers, 2007.

[本文引用: 1]

Han S, Zhao B, Bai Y, et al.

Mining subsidence research based on SBAS-InSAR in Yaojie coal mine

[J]. Mine Surveying, 2019, 47:1-5.

[本文引用: 1]

李梦梦, 范雪婷, 陈超, 等.

徐州矿区2016—2018年地面沉降监测与分析

[J]. 自然资源遥感, 2021, 33(4):43-54.doi:10.6046/zrzyyg.2020137.

[本文引用: 1]

Li M M, Fan X T, Chen C, et al.

Monitoring and interpretation of land subsidence in mining areas in Xuzhou City during 2016—2018

[J]. Remote Sensing for Natural Resources, 2021, 33(4):43-54.doi:10.6046/zrzyyg.2020137.

[本文引用: 1]

史珉, 宫辉力, 陈蓓蓓, 等.

Sentinel-1A京津冀平原区2016—2018年地面沉降InSAR监测

[J]. 自然资源遥感, 2021, 33(4):55-63.doi:10.6046/zrzyyg.2020341.

[本文引用: 1]

Shi M, Gong H L, Chen B B, et al.

Monitoring of land subsidence in Beijing - Tianjin - Hebei plain during 2016—2018 based on InSAR and Sentinel-1A data

[J]. Remote Sensing for Natural Resources, 2021, 33(4):55-63.doi:10.6046/zrzyyg.2020341.

[本文引用: 1]

刘金龙, 郭华东, 宋瑞, 等.

多模式雷达在矿区沉降监测中的应用研究

[J]. 遥感技术与应用, 2012, 27(4):584-590.

[本文引用: 1]

Liu J L, Guo H D, Song R, et al.

The application of multi-mode Radar to mining area subsidence monitoring

[J]. Remote Sensing Technology and Application, 2012, 27(4):584-590.

[本文引用: 1]

Bała J, Porzycka-Strzelczyk S, Strzelczyk J.

Subsidence troughs detection for SAR images:Preliminary results

[J]. 15th International Multidisciplinary Scientific GeoConferences(SGEM), 2015:18-24.

[本文引用: 2]

Klimczak M, Bała J.

Application of the Hough transform for subsidence troughs detection in SAR images

[J]. 17th International Multidisciplinary Scientific GeoConference(SGEM), 2017:819-826.

[本文引用: 1]

Bała J, Dwornik M, Franczyk A J S.

Automatic subsidence troughs detection in SAR interferograms using circlet transform

[J]. Sensors, 2021, 21(5):1706.

DOI:10.3390/s21051706      URL     [本文引用: 2]

This article presents the results of automatic detection of subsidence troughs in synthetic aperture radar (SAR) interferograms. The detection of subsidence troughs is based on the circlet transform, which is able to detect features with circular shapes. Compared to other methods of detecting circles, the circular transform takes into account the finite data frequency. Moreover, the search shape is not limited to a circle but identified on the basis of a certain width. This is especially important in the case of detection of subsidence troughs whose shapes may not be similar to circles or ellipses but to their fragments. The transformation works directly on the image gradient; it does not require further binary segmentation or edge detection as in the case of other methods, e.g., the Hough transform. The entire processing process can be automated to save time and increase reliability compared to traditional methods. The proposed automatic detection method was tested on a differential interferogram that was generated based on Sentinel-1A SAR images of the Upper Silesian Coal Basin area. The test carried out showed that the proposed method is 20% more effective in detecting troughs that than the method using Hough transform.

Porzycka-Strzelczyk S, Rotter P, Strzelczyk J.

Automatic detection of subsidence troughs in SAR interferograms based on circular Gabor filters

[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(6):873-876.

DOI:10.1109/LGRS.2018.2815782      URL     [本文引用: 2]

张堯. 基于InSAR技术大同煤田沉陷区的监测与识别[D]. 北京: 中国地质大学(北京), 2020.

[本文引用: 1]

Zhang Y. Monitoring and identification of subsidence area in Datong coalfield based on InSAR technology[D]. Beijing: China University of Geoscience (Beijing), 2020.

[本文引用: 1]

Wu Z, Zhang H, Wang Y, et al.

A deep learning based method for local subsidence detection and InSAR phase unwrapping:Application to mining deformation monitoring

[C]// 2020 IEEE International Geoscience and Remote Sensing Symposium.IEEE, 2020:20-23.

[本文引用: 1]

Anantrasirichai N, Biggs J, Kelevitz K, et al.

Detecting ground deformation in the built environment using sparse satellite InSAR data with a convolutional neural network

[J]. IEEE Transactions on Geo-science and Remote Sensing, 2020, 59(4):2940-2950.

DOI:10.1109/TGRS.2020.3018315      URL     [本文引用: 1]

Valade S, Ley A, Massimetti F, et al.

Towards global volcano monitoring using multisensor sentinel missions and artificial intelligence:The MOUNTS monitoring system

[J]. Remote Sensing, 2019, 11(13):1528.

DOI:10.3390/rs11131528      URL     [本文引用: 2]

Most of the world’s 1500 active volcanoes are not instrumentally monitored, resulting in deadly eruptions which can occur without observation of precursory activity. The new Sentinel missions are now providing freely available imagery with unprecedented spatial and temporal resolutions, with payloads allowing for a comprehensive monitoring of volcanic hazards. We here present the volcano monitoring platform MOUNTS (Monitoring Unrest from Space), which aims for global monitoring, using multisensor satellite-based imagery (Sentinel-1 Synthetic Aperture Radar SAR, Sentinel-2 Short-Wave InfraRed SWIR, Sentinel-5P TROPOMI), ground-based seismic data (GEOFON and USGS global earthquake catalogues), and artificial intelligence (AI) to assist monitoring tasks. It provides near-real-time access to surface deformation, heat anomalies, SO2 gas emissions, and local seismicity at a number of volcanoes around the globe, providing support to both scientific and operational communities for volcanic risk assessment. Results are visualized on an open-access website where both geocoded images and time series of relevant parameters are provided, allowing for a comprehensive understanding of the temporal evolution of volcanic activity and eruptive products. We further demonstrate that AI can play a key role in such monitoring frameworks. Here we design and train a Convolutional Neural Network (CNN) on synthetically generated interferograms, to operationally detect strong deformation (e.g., related to dyke intrusions), in the real interferograms produced by MOUNTS. The utility of this interdisciplinary approach is illustrated through a number of recent eruptions (Erta Ale 2017, Fuego 2018, Kilauea 2018, Anak Krakatau 2018, Ambrym 2018, and Piton de la Fournaise 2018–2019). We show how exploiting multiple sensors allows for assessment of a variety of volcanic processes in various climatic settings, ranging from subsurface magma intrusion, to surface eruptive deposit emplacement, pre/syn-eruptive morphological changes, and gas propagation into the atmosphere. The data processed by MOUNTS is providing insights into eruptive precursors and eruptive dynamics of these volcanoes, and is sharpening our understanding of how the integration of multiparametric datasets can help better monitor volcanic hazards.

Anantrasirichai N, Biggs J, Albino F, et al.

A deep learning approach to detecting volcano deformation from satellite imagery using synthetic datasets

[J]. Remote Sensing of Environment, 2019, 230:111179.

DOI:10.1016/j.rse.2019.04.032      URL     [本文引用: 2]

Anantrasirichai N, Biggs J, Albino F, et al.

Application of machine learning to classification of volcanic deformation in routinely generated InSAR data

[J]. Journal of Geophysical Research:Solid Earth, 2018, 123(8):6592-6606.

DOI:10.1029/2018JB015911      URL     [本文引用: 1]

Recent improvements in the frequency, type, and availability of satellite images mean it is now feasible to routinely study volcanoes in remote and inaccessible regions, including those with no ground‐based monitoring. In particular, Interferometric Synthetic Aperture Radar data can detect surface deformation, which has a strong statistical link to eruption. However, the data set produced by the recently launched Sentinel‐1 satellite is too large to be manually analyzed on a global basis. In this study, we systematically process &gt;30,000 short‐term interferograms at over 900 volcanoes and apply machine learning algorithms to automatically detect volcanic ground deformation. We use a convolutional neutral network to classify interferometric fringes in wrapped interferograms with no atmospheric corrections. We employ a transfer learning strategy and test a range of pretrained networks, finding that AlexNet is best suited to this task. The positive results are checked by an expert and fed back for model updating. Following training with a combination of both positive and negative examples, this method reduced the number of interferograms to ∼100 which required further inspection, of which at least 39 are considered true positives. We demonstrate that machine learning can efficiently detect large, rapid deformation signals in wrapped interferograms, but further development is required to detect slow or small deformation patterns which do not generate multiple fringes in short duration interferograms. This study is the first to use machine learning approaches for detecting volcanic deformation in large data sets and demonstrates the potential of such techniques for developing alert systems based on satellite imagery.

Wu Z, Zhang H, Wang Y, et al.

A deep learning based method for local subsidence detection and InSAR phase unwrapping:Application to mining deformation monitoring

[C]// 2020 IEEE International Geoscience and Remote Sensing Symposium.IEEE, 2020:20-23.

[本文引用: 1]

Long J, Shelhamer E, Darrell T.

Fully convolutional networks for semantic segmentation

[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015:3431-3440.

[本文引用: 4]

Zhao H, Shi J, Qi X, et al.

Pyramid scene parsing network

[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017:2881-2890.

[本文引用: 2]

Chen L C, Papandreou G, Schroff F, et al.

Rethinking atrous convolution for semantic image segmentation

[J/OL]. arXiv, 2017(2017-12-5). http://arxio.org/pdf/1706.05587.pdf.

URL     [本文引用: 1]

Ronneberger O, Fischer P, Brox T.

U-net:Convolutional networks for biomedical image segmentation

[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention.Springer,Cham, 2015:234-241.

[本文引用: 1]

Wang Q, Wu B, Zhu P, et al.

ECA-Net:Efficient channel attention for deep convolutional neural networks

[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).IEEE, 2020:11531-11539.

[本文引用: 3]

Xu Z, Wu C, Zheng E, et al.

Semantic segmentation of buildings in remote sensing images based on dense residual learning and channel adaption

[C]// 2019 4th International Conference on Electromechanical Control Technology and Transportation (ICECTT).IEEE, 2019:117-123.

[本文引用: 1]

Chen L C, Zhu Y, Papandreou G, et al.

Encoder-decoder with atrous separable convolution for semantic image segmentation

[C]// Proceedings of the European Conference on Computer Vision (ECCV), 2018:801-818.

[本文引用: 1]

Simonyan K, Zisserman A.

Very deep convolutional networks for large-scale image recognition.

[C]// 3rd International Conference on Learning Representations (ICLR), 2015:1-14.

[本文引用: 1]

He K, Zhang X, Ren S, et al.

Deep residual learning for image reco-gnition

[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016:770-778.

[本文引用: 1]

/

〈 〉

京ICP备05055290号-2
版权所有 © 2015 《自然资源遥感》编辑部
地址:北京学院路31号中国国土资源航空物探遥感中心 邮编：100083
电话:010-62060291/62060292 E-mail:zrzyyg@163.com
本系统由北京玛格泰克科技发展有限公司设计开发