-
无人机技术在过去的十年中迅速发展,提供了一个高效的高分辨率图像采集平台,并被以各种方式应用于地貌学研究中。应用摄影测量技术获取的高精度地面和地形数据,开展了包括监测河床地形和形态[1-3],河岸[4],冲沟侵蚀[5]等的研究。其中近景摄影测量的应用还包括直接量化土壤侵蚀和实验室尺度景观的地貌动力学演化模型[3, 6]。在时间、劳动力、成本等因素的影响下,通过无人机航拍图像来补充或取代野外大面积调查已经成为越来越多地貌学研究者的选择。通过机载图像完成的粒度制图已经显示出利用无人机航拍图像来分析粒径的前景[7]。
戈壁是干旱或极端干旱环境中的一种独特地貌景观[8],广泛分布于我国西北地区[9-10]。典型的戈壁通常表面由砾石覆盖[11-12],这也是其区别于其他荒漠景观的主要特征。戈壁表面砾石的覆盖度和粒径与地貌特征的关系,可以用来推测戈壁的形成演化规律[13-14]。现有的大面积粒度估算的研究通常使用卫星遥感图像,而戈壁表面砾石的粒级往往在厘米级,因此在利用卫星遥感影像来估算粒径时不可避免地存在精度过低的问题。因此,本研究通过无人机平台和运动结构建模(Structure from Motion, SfM)技术获取了较大范围的高空间分辨率的戈壁表面正射图像和数字地面模型(Digital Terrain Model, DTM),以探讨利用无人机和图像处理技术进行大范围戈壁砾石研究的适用性,为戈壁形成演化研究提供新的数据来源及技术支撑。
-
经过照片对齐,3个样区对齐的照片数量分别为253张,346张,357张。利用地面控制点对相片对齐的结果进行了优化,得到地面控制点的误差结果如表 1所示。样区控制点总体误差最大的是样区1,误差为28 cm,整体误差最小的样区是样区2,误差为8.19 cm。
表 1 各样区控制点误差分析
Table 1. The overall error analysis of control point in sample zones
样区
Sample zone东西方向误差
X error /cm南北方向误差
Y error/cm垂直方向误差
Z error /cm水平方向误差
XY error /cm误差总计
Total /cm1 9.79 26.22 1.06 27.99 28.01 2 7.67 2.83 0.31 8.18 8.19 3 52.94 8.43 5.46 8.93 10.47 通过以DSM作为表面模型进行正射校正,生成了正射图像。3个样区正射图像的地面分辨率依次是6.98,7.61,9.46 mm·像元-1。根据PhotoScan的结果报告显示,样区1到样区3的正射拼接图像的地面覆盖面积分别为0.022 8,0.034 7,0.038 6 km2。
-
如图 3内放大图所示,洪积扇表面主要包含砾石、土壤以及植被3种地物类型。大多数砾石颜色为深蓝色,包含少部分黄色、红色、白色砾石。因此在进行砾石训练数据采集时,对各个颜色的砾石都进行了采样,而把土壤和植被的训练数据都归入背景数据集。根据构建的决策树分类模型,分别得到了3个样区的砾石分类图像。如图 3中砾石分类结果图所示,图像中砾石呈白色显示,植被和土壤呈黑色显示,决策树分类模型较好的区分出了砾石和背景像元。
分别对3个样区分类后的图像进行了图像分割,按照确定的砾石筛选规则,提取出了形状较为完整且粒径大于32 mm极粗砾。如图 4中放大图所示,提取出来的砾石与正射图像中砾石形状和大小较为一致。
-
应用PhotoScan依据地面点点云生成了样区DTM。3个样区DTM的地面分辨率分别为1.4,1.5,1.9 cm·像元-1。如图 5所示,样区1的平均海拔为1 735.21 m,样区2的平均海拔为1 436.43 m,样区3的平均海拔为1 208.94 m。整体而言,3个样区的海拔变化均在25 m左右,海拔变化方向都是自东北方向向西南方向递减。
图 5 样区数字地面模型(从左到右分别为样区1、样区2、样区3)
Figure 5. DTM of sample zones. Sample zone 1, sample zone 2, sample zone 3 is arranged from left to right.
应用ArcGIS 10.2以DTM数据为基础,生成了坡度和坡向数据。如图 6所示,3个样区的地形大体平坦,基本坡度都小于10°。样区1坡度以2°~10°为主,样区2坡度以2°~5°为主,样区3坡度以小于2°为主。如图 7所示,样区1和样区2的坡向都是以西南坡为主;而样区3中各个方向的坡向分布较为均匀。在样区1和样区2中坡向变化还较为连续,而在样区3中由于地形起伏较小,因此坡向变化杂乱。由此可知,从洪积扇扇心到扇缘,地形起伏越来越平缓,冲沟和地垄逐渐消失。
-
经过计算各样区的总体砾石覆盖度(表 2),样区1的砾石覆盖度为34.22%,样区2的砾石覆盖度为26.85%,样区3的砾石覆盖度为21.88%。通过样区间砾石覆盖度的变化可知,随着海拔下降,砾石覆盖度逐渐下降且下降幅度减小。
表 2 各样区砾石特征统计
Table 2. Statistics of gravel characteristics in sample zones
样区
Sample zone覆盖度
Coverage/%砾石数量
Amount粒径均值
Mean/mm中位数
Median/mm众数
Mode/mm标准差
STD/mm1 34.22 74 053 130 128 56 125 2 26.85 314 599 95 85 61 73 3 21.88 206 246 78 69 51 57 如表 2所示,通过统计3个样区的砾石数量,粒径均值、中位数、众数以及标准差,发现:样区1内大砾石较多而砾石数量最少,粒径均值130 mm,中值128 mm,以56 mm粒径砾石最多,标准差最大,说明砾石粒径差异最大;样区2砾石粒径居中而砾石数量最多,粒径均值95 mm,中值85 mm,以61 mm粒径的砾石最多;样区3砾石普遍较小而砾石数量居中,粒径均值78 mm,中值69 mm,以51 mm粒径砾石最多,标准差最小,说明砾石粒径差异最小。
整体而言,从洪积扇扇心到靠近扇缘,砾石覆盖度逐渐降低,砾石粒径均值也逐渐减小,而且变化幅度都逐渐减小,该趋势与海拔变化趋势基本一致,说明在洪积扇尺度上砾石覆盖度与粒径的变化主要受海拔影响。
-
以南北方向上约30 m为间隔将单个样区划分为样方,计算了每个样方内的砾石覆盖度,砾石Feret最大粒径的均值、中位数、众数、方差。
在单个样区内,如图 8所示,从北向南,样区1内砾石覆盖度先升后降,而样区2和样区3内砾石覆盖度在样区靠下位置上升明显。砾石覆盖度变化与海拔关系不明显。
如图 9所示,在单个样区内,自北向南,样区1内砾石粒径均值、中位数和标准差整体呈上升趋势;样区2内,砾石粒径均值和中位数略有下降;样区3内,砾石粒径均值和中位数变化不大。在靠近洪积扇扇心区域砾石粒径相比其他区域变化较大。
整体而言,在单个样区内砾石覆盖度和粒径的变化与海拔关系不大。
基于无人机图像的戈壁表面砾石特征变化研究
Spatial Distribution of Gravel Characteristics on Gobi Desert Surface Based on Image Acquired by Unmanned Aerial Vehicle
-
摘要:
目的 通过无人机平台和运动结构建模技术获取了覆盖较大范围的戈壁表面正射图像和数字地面模型, 以位于新疆维吾尔自治区哈密市天山南坡的洪积扇为研究对象, 进行戈壁表面砾石覆盖度和粒径的测量及分析, 探讨利用无人机和图像处理技术进行大范围戈壁砾石研究的适用性。 方法 通过自洪积扇扇心到靠近扇缘选取3个典型样区, 利用无人机获取连续覆盖的戈壁表面高分辨率图像, 生成了戈壁表面的高清正射拼接图像以及高精度的地形数据, 并提取了样区内砾石(粒径≥ 7 mm)的覆盖度及砾石(粒径≥ 32 mm)的粒径。 结果 自扇心到靠近扇缘, 3个样区的砾石覆盖度分别为34.22%, 26.85%, 21.88%;砾石粒径均值分别为130, 95, 78 mm。随海拔下降, 样区总体砾石覆盖度和粒径均呈下降趋势, 且自扇中到扇缘相比自扇心到扇中下降趋势有所减缓; 而在各个样区内部, 砾石覆盖度及粒径与海拔关系不明显。 结论 在洪积扇尺度上, 砾石覆盖度与粒径变化主要受海拔影响; 在样区尺度上, 砾石覆盖度和粒径变化则可能受到植被分布和局部地形的影响。利用无人机和图像处理技术, 可以高效且准确地评估大范围戈壁表面砾石特征及分布, 为戈壁形成演化研究提供基础数据及技术支撑。 Abstract:Objective To explore the applicability of using unmanned aerial vehicle (UAV) and image processing technique for large scale research of gravels on Gobi surface. Method Taking the pluvial fan at south slope of Tianshan Mountains in Hami of Xinjiang Uygur Autonomous Region as study area, the orthographic images and digital terrain models (DTM) covering a large range of Gobi were obtained by UAV platform and the structure from motion (SfM) technology, and the gravel coverage and diameter were measured and analyzed. From the center to near the edge of the pluvial fan, three typical sample zones were selected, and high resolution images of Gobi surface were shot by camera on UAV. High resolution mosaic image and terrain data of Gobi surface were produced, and the coverage of gravels (diameter ≥ 7 mm) and diameter of gravels (diameter ≥ 32 mm) were calculated. Result From center to near the edge of the fan, the gravel coverage of the three sample zones was 34.22%, 26.85% and 21.88%, and the mean gravel diameter was 130, 95 and 78 mm. As elevation decreased, the gravel coverage and diameter dropped. The gravel coverage and diameter dropped more slowly from the middle to near the edge of the fan than from the center to the middle of the fan. However, within each sample zone, the relationship between gravel characteristics (coverage and diameter) and elevation was not obvious. Conclusion On the scale of pluvial fan, the changes of gravel coverage and diameter are mainly affected by elevation. On the scale of sample zone, the change of gravel coverage and diameter may be affected mainly by the vegetation distribution and local topography. Using UAV and image processing technology can efficiently and accurately estimate a wide range of the gravel characteristics and their spatial distribution, which provides basic data and methodological support for the research on Gobi evolution. -
Key words:
- UAV
- / gravel coverage
- / gravel diameter
- / Gobi desert
-
表 1 各样区控制点误差分析
Table 1. The overall error analysis of control point in sample zones
样区
Sample zone东西方向误差
X error /cm南北方向误差
Y error/cm垂直方向误差
Z error /cm水平方向误差
XY error /cm误差总计
Total /cm1 9.79 26.22 1.06 27.99 28.01 2 7.67 2.83 0.31 8.18 8.19 3 52.94 8.43 5.46 8.93 10.47 表 2 各样区砾石特征统计
Table 2. Statistics of gravel characteristics in sample zones
样区
Sample zone覆盖度
Coverage/%砾石数量
Amount粒径均值
Mean/mm中位数
Median/mm众数
Mode/mm标准差
STD/mm1 34.22 74 053 130 128 56 125 2 26.85 314 599 95 85 61 73 3 21.88 206 246 78 69 51 57 -
[1] LANE S N, JAMES T D, CROWELL M D. Application of digital photogrammetry to complex topography for geomorphological research[J]. The Photogrammetric Record, 2000, 16(95):793-821. doi: 10.1111/phor.2000.16.issue-95 [2] CHANDLER J. Effective application of automated digital photogrammetry for geomorphological research[J]. Earth Surface Processes and Landforms, 1999, 24(1):51-63. doi: 10.1002/(ISSN)1096-9837 [3] BRASINGTON J, SMART R M A. Close range digital photogrammetric analysis of experimental drainage basin evolution[J]. Earth Surface Processes and Landforms, 2003, 28(3):231-247. doi: 10.1002/(ISSN)1096-9837 [4] BARKER R, DIXON L, HOOKE J. Use of terrestrial photogrammetry for monitoring and measuring bank erosion[J]. Earth Surface Processes and Landforms, 1997, 22(13):1217-1227. doi: 10.1002/(SICI)1096-9837(199724)22:13<>1.0.CO;2-E [5] BETTS H D, DEROSE R C. Digital elevation models as a tool for monitoring and measuring gully erosion[J]. International Journal of Applied Earth Observation and Geoinformation, 1999, 1(2):91-101. doi: 10.1016/S0303-2434(99)85002-8 [6] HANCOCK G, WILLGOOSE G. The production of digital elevation models for experimental model landscapes[J]. Earth Surface Processes and Landforms, 2001, 26(5):475-490. doi: 10.1002/(ISSN)1096-9837 [7] ARIF M S M, GüLCH E, TUHTAN J A, et al. An investigation of image processing techniques for substrate classification based on dominant grain size using RGB images from UAV[J]. International Journal of Remote Sensing, 2016, 10:1-23. [8] GOUDIE A. Arid and semi-arid geomorphology[M]. Cambridge:Cambridge University Press, 2013. [9] 申元村, 王秀红, 程维明, 等.中国戈壁综合自然区划研究[J].地理科学进展, 2016(1):57-66. [10] 冯益明, 吴波, 姚爱冬, 等.戈壁分类体系与编目研究[J].地理学报, 2014(3):391-398. [11] MCFADDEN L D, MCDONALD E V, WELLS S G, et al. The vesicular layer and carbonate collars of desert soils and pavements:formation, age and relation to climate change[J]. Geomorphology, 1998, 24(2-3):101-145. doi: 10.1016/S0169-555X(97)00095-0 [12] QUADE J. Desert pavements and associated rock varnish in the Mojave Desert:How old can they be?[J]. Geology, 2001, 29(9):855-858. doi: 10.1130/0091-7613(2001)029<0855:DPAARV>2.0.CO;2 [13] ADELSBERGER K A, SMITH J R. Desert pavement development and landscape stability on the Eastern Libyan Plateau, Egypt[J]. Geomorphology, 2009, 107(3-4):178-194. doi: 10.1016/j.geomorph.2008.12.005 [14] DIETZE M, KLEBER A. Contribution of lateral processes to stone pavement formation in deserts inferred from clast orientation patterns[J]. Geomorphology, 2012, 139-140:172-187. doi: 10.1016/j.geomorph.2011.10.015 [15] MU Y, WANG F, ZHENG B, et al. The gravel coverage and size of Gobi desert analyzed by a rapid image-based method[C]. IUFRO Regional Congress for Asia and Oceania 2016, Beijing, China: 2016: 364. [16] 钱广强, 董治宝, 罗万银.基于数字图像的中国西北地区戈壁表面砾石形貌特征研究[J].中国沙漠, 2014, 34(3):625-633.