面向矿区沉降监测的InSAR技术及应用研究.pdf

博士学位论文 面向矿区沉降监测的InS AR技术及应用研究 A S tudy on InS AR for S ubsidence Monitoring in Mining Area 作 者 陈炳乾 导 师 邓喀中教授 中国矿业大学 二〇一五年五月 中图分类号 P237 学校代码 1029 0 UDC___________________ 密 级 公开 博士学位论文 面向矿区沉降监测的InS AR技术及应用研究 A S tudy on InS AR for S ubsidence Monitoring in Mining Area 作 者 陈炳乾 导 师 邓喀中 申请学位 工学博士 培养单位 环境与测绘学院 学 科 专 业 大地测量与测量工程 研 究 方 向InSAR技术及变形监测 答辩委员会主席________________评 阅 人_______________________ 二〇一五年五月 学位论文使用授权声明 本人完全了解中国矿业大学有关保留、使用学位论文的规定，同意本人所撰 写的学位论文的使用授权按照学校的管理规定处理 作为申请学位的条件之一， 学位论文著作权拥有者须授权所在学校拥有学位 论文的部分使用权，即 ①学校档案馆和图书馆有权保留学位论文的纸质版和电 子版，可以使用影印、缩印或扫描等复制手段保存和汇编学位论文；②为教学和 科研目的，学校档案馆和图书馆可以将公开的学位论文作为资料在档案馆、图书 馆等场所或在校园网上供校内师生阅读、浏览。另外，根据有关法规，同意中国 国家图书馆保存研究生学位论文。 保密的学位论文在解密后适用本授权书） 。 作者签名 年月日 导师签名 年月日 论文审阅认定书 研究生 陈 炳 乾 在规定的学习年限内， 按照研究生培养方案 的要求，完成了研究生课程的学习，成绩合格；在我的指导下完成本 学位论文， 经审阅，论文中的观点、数据、表述和结构为我所认同， 论文撰写格式符合学校的相关规定，同意将本论文作为学位申请论文 送专家评审。 导师签字 年月日 致谢 门前老树长新芽，院里枯木又开花，一切都昭示着春天到了。 弹指一挥间，二十四年的求学生涯即将结束，有苦涩，有欣喜，但更多的是 感谢。我不是一个天生聪明伶倒的孩子，更确切的说我算的上反应愚钝，能一路 跌跌撞撞走到现在，要感谢的人实在太多。 首先要感谢我的导师邓喀中教授， 作为国际知名的“三下”采煤领域的专家， 您在百忙之中抽出大量时间对论文的选题、研究内容的确定、实验设计、初稿修 改、定稿等方面都给予了极大的指导和帮助。师从五载，您深厚的学术造诣、严 谨的治学态度、积极的人生态度和不断进取的科学精神一直不断激励着我，使学 生终身受益。每次与您的交谈都会让学生我有种醍醐灌顶的感觉，虽是如此，无 奈学生鲁钝，悟性较差，总不能领会恩师您深刻思想的精髓，每想至此不由如芒 刺背，惭愧汗颜。同时感谢您给我提供去英国诺丁汉访问交流的机会，让学生增 长了见识，拓展了视野。在此，谨向我的导师表示衷心的感谢和崇高的敬意。 特别感谢中国矿业大学谭志祥教授。您在论文的选题、撰写、定稿、修改过 程中给予的指导和帮助，每一次都让学生深受感动，从您的身上，我学到了很多 东西。您勤奋、乐观、宽厚的待人风格都给我留下了深刻的印象，是我学习的榜 样。 感谢我的师兄范洪冬副教授，是您让我对“师兄”这个词产生温馨的感觉， 在矿大五年里，科研上您是我的老师，生活中您如大哥，科研学习，接人待物， 您是我的榜样。 感谢中国矿业大学汪云甲教授、张书毕教授、顾和和教授、张华副教授、郑 南山副教授、顾伟老师在论文选题过程中给予的宝贵意见。 感谢中国矿业大学郭广礼教授、雷少刚老师、周鸣老师、杨化超老师、张宏 贞老师、刘涛健老师、蒋冬梅老师给予的关心和帮助。 感谢我的师兄剧成宇和薛继群， 正是你们最初在我学习InSAR时给予的帮助 和指导才使我少走了很多弯路，并最终完成了本文的撰写。 感谢407全体成员，我们是一个团结的集体，你们和我一起见证了五年来科 研之路的酸甜苦辣，同时感谢你们在实验过程中给予的无私帮助。 感谢同窗好友郝明同学，你积极、乐观、进取的精神是我学习的榜样，与你 在一起每天都是积极向上的。同时感谢郝明同学在科研上无数次的援助，向 “ 郝 指导”致敬。 感谢环测学院博士 2012级、大地测量硕士 2010级的所有同学在学习、工作 和生活上给予的帮助。 感谢舍友杨立峰、石学军、吴海、李崇茂、贾佳佳、张卫强，田丰，同宿三 年你们对我的宽容与理解让我感到家的温暖，我会记住这段岁月。 特别感谢香港理工大学张磊博士、西南交通大学张瑞博士、中国国土资源航 空物探遥感中心葛大庆师兄在数据处理、形变分析等方面给予我的帮助。虽然我 们未曾见面，但一封封的电子邮件使我感受到你们就在我的身旁给予我指导，再 次感谢你们慷慨和无私的帮助。 感 谢De丨f t大学Andrew Hooper博士提供了 StaMPS软件及相关论文、De I ft 大学提供了 DORIS开源软件、GAMMA公司提供了 GAMMA软件、SARMAP公司提供了 SARscape软件，正是有了这些软件和数据，我才能在此基础上进行处理分析， 并对相关算法进行改进，对你们慷慨和无私的帮助表示深深的谢意。 在多年的求学过程中，我的家人给了我最大的鼓励和支持，他们是我学习和 生活上的精神支柱。在论文完成之际，谨向我的父母、爷爷奶奶、姑姑、姐姐表 示最衷心的感谢，愿你们永远身体健康。 最后，向在百忙之中对本论文审阅的各位专家致以诚挚的感谢 陈炳乾 二零一五年四月于徐州 摘 要 煤炭是我国的主要能源之一，由于煤炭资源高强度、大面积的开采，使得开 采所引起的地表沉降及环境灾害问题日益突出。 对矿区地面沉降进行持续有效监 测，可及时掌握沉降规律和地表破坏程度，为合理开采地下矿产资源，控制地面 沉降提供决策依据。传统地表沉降监测方法和技术覆盖面小、作业强度大、效率 低、不能适应生产发展的需要。近年来发展的InSAR技术具有全天候、全天时、 髙分辨率和连续空间覆盖的优势，已发展成为新型监测矿区开采沉陷的技术。 然 而由于煤矿区自身地理环境的特殊性、 开采沉陷形变的复杂性（ 快速、 形变量大 以及InSAR自身成像条件的苛刻性（ 大气效应、时间失相关和空间失相关 ，使 得 InSAR技术在监测煤炭开采引起的矿区地表沉降应用上受到诸多限制。因此 本文针对InSAR技术在煤矿区进行地面沉降监测中的存在的问题及其相关算法 进行研宄，以期能够在一定程度上解决InSAR技术应用于矿区开采沉陷监测存 在的问题。本文的研究工作和成果如下 1 总结了 InSAR技术的研宄现状， 指出了国内外在InSAR技术及相关算 法研宄中的不足，阐述了 SAR、InSAR、D-InSAR及时序InSAR技术的基本原 理，分析了 InSAR系统中的干涉相位、干涉相干性和干涉相位对髙程的灵敏度 几个关键参数。 2 提出一种基于多级匹配策略的干涉SAR影像自动配准新方法。 该方法 首先利用改进后的SIFT 算法提取SAR影像上的特征点并完成特征点初匹配， 之 后基于RCM算法快速剔除相关性较低的匹配点对，最后采用RANSAC算法进 行匹配点对的二次提纯，最终完成SAR影像的精确匹配。该方法无需任何先验 卫星轨道信息和外部DEM，且整个配准过程为自动实现。实验证明该方法能满 足不同波长、不同分辨率及不同相干条件下SAR影像的高精度配准要求。 3 提出一种基于InSAR数据和地面LiDAR点云数据融合的方法来实现 矿区地表大梯度形变监测。该方法利用地面LiDAR技术点云数据的高密度及其 三维建模后单点定位精度高的特点，利用IDW 算法，一方面对InSAR形变场中 的大梯度形变和失相关区域进行填补，另一方面对地面LiDAR点云形变场和 InSAR形变场的公共覆盖区域进行加权平均融合。该方法一定程度上解决了 InSAR技术应用于矿区大梯度形变中所遇到的问题，为利用InSAR技术实现矿 区内大梯度形变监测提供了新的途径和手段。 4 提出一种综合利用SAR影像幅度和相位信息获取矿区地表时间序列沉 降的新方法。 该方法一方面利用基于幅度信息的ABFT 技术进行大梯度形变区域 的监测，同时采用基于相位信息的干涉测量技术进行微小变形区域的监测， 最后 将二者得到的形变监测结果进行融合得到形变区域的完整监测结果。 同时该方法 属于时间序列模型，能反映开采沉陷时序变化特征。 5 引入一种超短基线干涉测量技术新方法进行老采空区沉降监测。该方 法相比传统的干涉差分SAR技术具有无需外部DEM的优势，避免了外部DEM 的引入所带来的误差。 利用该方法获取了老采空区沉降速率和形变时间序列， 在 此基础上建立了地表残余下沉速度循环周期与采厚、 下沉速度循环峰值与深厚比 的经验关系式，为预测和评价老采空区残余形变提供基础。 6 构建了 InSAR技术监测与预计一体化模型。 该模型利用InSAR技术的 全天候、髙精度、大区域等优势进行开采沉陷监测，获取开采沉陷的影响范围与 发展趋势，得到其时空演化规律。在此基础上将监测结果作为SVR算法的训练 与学习样本建立已观测数据与未来沉降之间的函数，并引入PSO算法进行SVR 函数最优参数选取，同时采用滚动预测方法进行开采沉陷动态预计， 最终实现开 采沉陷监测与预计的一体化。 该论文有图77幅，表 23个，参考文献208篇。 关键词开采沉陷；沉降监测；沉降预计；InSAR;地 面 LiD A R;大梯度形变; 数据融合 Abstract Coal is one the major s ou rces of energy of China. T he problems of grou nd s u bs idence and env ironmental hazards cau s ed by high intens ity and large areas exploiting coal res ou rces becom e m ore and m ore s eriou s . In order to redu ce mining s u bs idence dis as ters , we s hou ld carry ou t a continu ou s and efficient monitoring on mining s u bs idence and obtain the ru les of grou nd s u bs idence and the degree of dam age, which help u s m ake plans of coal mining and control grou nd s u bs idence dis as ters . T raditional monitoring m ethods requ ire large am ou nts of hu man and material res ou rces and carry the dis adv antage of being bas ed on point-wis e m eas u rem ents and lacking area ination, which can not m eet the produ ction needs of dev elopm ent. T he appearance of interferom etric s ynthetic apertu re radar InSAR technology prov ides a new for grou nd s u bs idence monitoring. Du e to its ou ts tanding adv antages of fas t， high-precis ion, all-tim e， all-weather and large areas , it m akes u p for the deficiencies of the traditional monitoring m ethods to a large extent. Howev er, InSAR technology has many limits when applying in mining s u bs idence monitoring du e to the s pecial characteris tics of geographical env ironment of coal mining areas and the complexity of mining s u bs idence rapid，large gradient deation. Moreov er, atm os pheric effect， temporal and s patial decorrelation als o influ ence the imaging qu ality. T herefore, this paper focu s es on the problem s that exis t in mining s u bs idence monitoring when u s ing InSAR technology and its relev ant algorithms . T he main contribu tions are des cribed as follows . 1 T he res earch s tatu s of InSAR techniqu e were s u m marized. T he ins u fficiency of InSAR and its related techniqu es s tu dies were pointed ou t. T he bas ic principles of SAR, InSAR, D-InSAR and time s eries InSAR technology were des cribed, and s ev eral important param eters of InSAR s ys tem were als o analyzed, inclu ding inteferom etric phas e， inteferom etric coherence and the s ens itiv ity of interferom etric phas e to elev ation change. 2 A new au tomatic regis tration approach bas ed on a m u lti-s tep matching s trategy was propos ed. In the firs t s tep， key points were detected and m atched u s ing an improv ed s cale inv ariant featu re trans SIFT operator. In the s econd s tep， region correlation matching RCM algorithm was u s ed to exclu de matched points with low correlations . In the third s tep，random s am ple cons ens u s RANSAC algorithm was u s ed to condu ct pu rifying of matched points once again and finally an hi exact match points were obtained. T he propos ed does not any priori ination, s u ch as s atellite orbit ination and external DEM. T he experim ent res u lts s how that this cou ld m eet the requ irem ent of high accu racy regis tration at different conditions , inclu ding different wav elengths ， different s patial res olu tions , different correlations . 3 A was propos ed that integrates InSAR data and point clou d data obtained by terres trial LiDAR to improv e the detectable deation gradient of InSAR technology. T he propos ed takes adv antage of high-dens ity of terres trial LiDAR point clou d data and its high precis ion of point pos itioning after 3D modeling. Bas ed on inv ers e dis tance weighting IDW algorithm, on the one hand, large gradient deation and decorrelation area exis ting in InSAR deation field was filled with point data, on the other hand, the deation field of com mon cov erage area monitored by InSAR and terres trial LiDAR was taken a weighted av erage to generate a higher accu racy monitoring res u lts . T he propos ed can s olv e the problem of large gradient deation to s om e extent and it prov ides new ways and m eans for obtaining large gradient deation monitoring by InSAR technology in mining area. 4 A time s eries deation model was propos ed that com bines SAR image phas e m eas u rem ent and amplitu de m eas u rem ent to monitor time s eries mining s u bs idence. O n the one hand， this u s ed the am plitu de-bas ed featu re tracking ABFT technology to monitor large gradient deation, on the other hand it u s ed interferom etric phas e m eas u rem ent to condu ct monitoring in s mall deation area, followed by the integrating of the two deation res u lts . Com pared with m erely u s ing phas e m eas u rem ent technology, the integrated res u lts can obtain the deation ination in the large gradient deation area. Likewis e， com pared with m erely u s ing intens ity tracking technology, the integrated res u lts can get m ore accu rate s mall deation ination. 5 A new u ltras hort bas eline USB InSAR m ethod was introdu ced to monitor the s u bs idence of old goaf. Com pared to the traditional differential interferom etric technology, it has the u niqu e adv antage, i.e. with no need for external DEM 5 which av oids the introdu ction of new errors . Us ing the obtained s u bs idence v elocity and time s eries s u bs idence res u lts of old goaf， m oreov er， the relations hip between cyclic period of s u bs idence v elocity and H/M/c and relations hip between flu ctu ant peak of s u bs idence v elocity and mining thick were es tablis hed， which prov ides the IV bas is for the prediction and ev alu ation of old goaf res idu al deation. 6 A model was propos ed that integrates InSAR technology and the s u pport v ector regres s ion SVR algorithm to monitor and dynamically predict mining s u bs idence. InSAR technology was firs t u s ed to monitor the range of influ ence and the dev elopm ent trend of mining s u bs idence, thu s obtaining the law of s u rface s u bs idence. Bas ed on the monitoring res u lts obtained by InSAR technology, the SVR algorithm was u s ed to des cribe the nonlinear fu nction correlativ ity between the monitored data and fu tu re s u bs idence. As the per ance of the SVR algorithm depends largely on the choice of relev ant param eters , the particle s warm optimization PSO algorithm was introdu ced to s elect the optimal param eters for the SVR algorithm. Finally， a of rolling prediction bas ed on the optimized SVR param eters was adopted to u pdate the training and learning s am ples of SVR, thu s allowing the algorithm to u s e the lates t monitored data to dynamically predict fu tu re mining s u bs idence. T here are 77 figu res , 23 tables and 208 references in this dis s ertation. Keywords mining s u bs idence; s u bs idence monitoring; s u bs idence predition; InSAR; T erres trial LiDAR; large gradient deation; data integration Extended Abstract Coal is the China’s primary energy s ou rce and an indis pens able material bas is for econom ic and s ocial dev elopm ent. Grou nd s u bs idence and other env ironmental hazards cau s ed by high intens ity exploitation of coal res ou rces ov er large areas are becoming increas ingly m ore s eriou s . In China, ov er 700 thou s and hectares hav e reportedly been affected by mining s u bs idence， res u lting in the los s of m ore than 50 billion Yu an. In order to redu ce mining s u bs idence dis as ters ， dom es tic and foreign s cholars hav e condu cted extens iv e res earch and achiev ed fru itfu l achiev em ents . Du e to the complexity of the mining s trata， m os t s tu dies in mining s u bs idence are cu rrently bas ed on obs erv ed data, and als o coal mining s u bs idence is traditionally monitored u s ing triangu lation, lev eling and GPS m eas u rem ents , i.e., es tablis hing the obs erv ation lines on the main s ectional plane of the s u bs idence bas in， obs erv ing regu larly and obtaining deation laws throu gh data proces s ing and analys is . Su ch m ethods requ ire large am ou nts of hu man and material res ou rces and carry the dis adv antage of being bas ed on point-wis e m eas u rem ents and lacking area ination. T he appearance of interferom etric s ynthetic apertu re radar InSAR techniqu e prov ides a new for grou nd s u bs idence monitoring. T he techniqu e u tilizes the relations hip between the phas e difference and s patial dis tance difference of com plex data obtained in two obs erv ations to extract three-dim ens ional ination and elev ation change ination of grou nd s u rface. According to the working principle of InSAR, it can get s mall deation of grou nd res olu tion cell and reach millimeter accu racy in theory. Du e to its ou ts tanding adv antages of fas t, high-precis ion, all-tim e， all-weather and large areas ，it m akes u p for the deficiencies of the traditional monitoring m ethods to a large extent. Howev er, InSAR technology has many limits when applying in mining s u bs idence monitoring du e to the s pecial characteris tics of geographical env ironment of coal mining areas and the complexity of mining s u bs idence rapid, large gradient deation. Moreov er, atm os pheric effect, temporal and s patial decorrelation als o influ ence the imaging qu ality. T herefore, this paper focu s es on the problem s that exis t in mining s u bs idence monitoring when u s ing InSAR technology and its relev ant algorithms . T he main contribu tions are des cribed as follows . 1 T he res earch s tatu s of InSAR techniqu e were s u m marized. T he ins u fficiency of InSAR and its related techniqu es s tu dies were pointed ou t. T he bas ic principles of VI s ynthetic apertu re radar SAR, InSAR, differential interferom etry s ynthetic apertu re radar D-InSAR and time s eries InSAR technology inclu ding perm anent s catterer interferom etry technology, s mall bas eline s u bs et technology and interferom etric point target analys is technology were des cribed，and s ev eral important param eters of InSAR s ys tem were als o analyzed. 2 A new au tomatic regis tration approach bas ed on a m u lti-s tep matching s trategy was propos ed. In each s tep, different matching algorithms were s elected to concentrate on the giv en objectiv e. T he s trategy to improv e the matching was incremental， mov ing one s tep at a time， coars e to fine. In the firs t s tep, key points were detected and matched u s ing an improv ed s cale inv ariant featu re trans SIFT operator. Here， becau s e of the need for compu tational efficiency and the influ ence of s peckle nois e， the original algorithm of key point detection was adapted to s kip the s earch for key points in the firs t octav e of the s cale-s pace pyramid. And partitioning s trategy was adopted for large image. Becau s e of the pres ence of s peckles and problem s with the matching s trategy o