一种改进的超体素与区域生长点云分割方法

点云分割作为识别地理场景的空间特征、探索和记录空间信息的关键处理步骤,其分割精度直接影响后续三维场景重建、地物特征提取等应用的效果。针对传统区域生长点云分割算法的不稳定等问题,本文结合超体素和区域生长算法对点云数据进行分割,并利用点云自身的色彩信息进一步改进分割结果。试验结果表明,相较于传统的区域生长和已有的结合超体素与区域生长的分割算法,本文算法对点云数据分割的效果更好,且其精确率与召回率均有提高。     

航空与卫星影像数据多尺度分割对比研究

本文从遥感影像多尺度分割的角度分析了同一地区的航片与QuickBird、IKONOS等卫星数据的分割效果,讨论了不同地物和不同影像的最佳分割尺度,以增强对目标物的检测与识别能力,提高现有航空遥感数据及卫星影像数据应用的精度和效率,并对最终分类结果进行了比较。结果表明航片、QuickBird、IKONOS的最佳分割尺度分别为125、100、75,QuickBird的分类精度最高,航片和IKONOS的分类精度次之。因此可认为,航片在实际应用中,可以代替高分辨率卫星影像。     

通过磁共振物理参数估计实现磁共振图像的分解

近年来,单谱与多谱磁共振图像的分割方法研究已经取得了很大进展,并应用于正常年龄的脑发育和脑疾病的诊断研究,例如老年痴呆综合征、脑损伤、脑肿瘤等的临床研究等。据此可以通过多谱MR图像获得多种对比度信息,更加准确表达人体组织及其病理情况。现已提出的大多数方法,把组织分割问题考虑成统计决策、模式识别和聚类、图像处理和分析等问题。在这些方法中用于组织分割的特征主要是单谱/多谱图像的灰度值/矢量,它们不能直接反映组织的物理特征。而且,这些方法将组织分割问题表述成组织磁共振图像中像素不同成分的有限集合。所以,这些组织分割的方法所取得的结果在某种意义上是不够合理的。这篇论文提出了一种基于磁共振图像谱分解的新的组织分割方法,该方法将组织分割问题考虑成组织的磁共振物理参数的估计问题。这个方法不仅用于磁共振成像中脂肪信号的抑止,也用于核磁共振谱中的水信号抑止。因此,这个方法可以称为空间和谱MR成像。     

基于直方图区域生长的遥感图像阈值分割算法

传统阈值分割算法从单阈值扩展到多阈值的过程中,时间复杂度会大幅度增加,并且由于遥感图像信息复杂,会导致分割效果降低.为了解决这些问题,本文提出了基于直方图区域生长的遥感图像阈值分割算法.在本文算法中,每一个灰度级均作为1个初始阈值,用256个阈值将直方图分割成256个原始小区域.为了减少阈值数目,本文将小区域合并成大区...     

基于双注意力残差网络的高分遥感影像道路提取模型

高分辨率遥感影像中,道路光谱信息丰富,且空间几何结构更清晰。但是,基于高分遥感影像的道路提取面临道路尺寸变化大、容易受树木、建筑物及阴影遮挡等因素影响,导致提取结果不完整。此外,高分遥感影像中同物异谱和异物同谱现象较为严重,从而影响道路提取结果连续性及细小道路信息完整性,而且难以区分道路和非道路不透水层。因此,本文提出基于双注意力残差网络的道路提取模型DARNet,利用深度编码网络,获取细粒度高阶语义信息,增强网络对细小道路的提取能力,通过嵌入串联式通道-空间双重注意力模块,获取道路特征图逐通道的全局语义信息,实现道路特征的高效表达及多尺度道路信息的深层融合,增强阴影和遮挡环境下网络模型的鲁棒性,改善道路提取细节缺失现象,实现复杂环境下高效、准确的道路自动化提取。本文在3个实验数据集对DARNet和DLinkNet、DeepLabV3+等5个对比模型进行对比试验和定量评估,结果表明,本文DARNet模型的F1分别为77.92%、67.88%和80.37%,高于对比模型。此外,定性比较表明,本文提出模型可以有效克服由于物体阴影、遮挡和高分影像光谱变化导致道路提取不准确与不完整问题,改善细...     

太湖流域漕桥河小流域水环境容量估算及污染物削减分配

太湖流域漕桥河区域水环境质量日益恶化,成为影响该区域可持续发展的主要问题之一,为此基于漕桥河小流域和水质污染分段的特征,计算了研究区域的水环境容量和污染物削减量,提出河流污染分段控制和流域管理的水环境容量控制模式,建立河流氮磷污染控制的可操作性措施,从而确定重点工程和措施实施的年度计划,形成污染源年度削减方案,可为小流域污染控制管理提供科学依据.     

Analyse multi-échelles de la variabilité spatiale de l’équivalent en eau de la neige (EEN) sur le territoire de l’Est du Canada

La présente étude a pour objectif d’analyser la variabilité spatiale multi-échelles de l’équivalent en eau de la neige (EEN). Dans un premier temps, la variabilité spatiale de l’EEN par rapport à la latitude et à la longitude a été analysée. Des indices locaux ont été utilisés pour caractériser les différentes structures spatiales. Par la suite, les structures spatiales homogènes ont été délimitées à l’aide de l’approche de segmentation spatiale multi-résolutions en utilisant des méta-variables physiographiques. La segmentation a été validée à l’aide du test non paramétrique de Kruskal-Wallis appliqué aux données de l’EEN de chaque paire de zones adjacentes. À l’échelle régionale, la segmentation spatiale a permis d’identifier six zones géographiques différenciées par leur position par rapport aux modes de circulations atmosphériques et la disposition du relief. À l’échelle locale, la segmentation spatiale montre le rôle de la pente, de la courbure, etc. dans la variabilité spatiale du couvert nival.

Editeur Z. W. Kundzewicz; Editeur associé E. Gargouri     

ANALYSIS OF EVOLUTION OF THE RIYUESHAN FAULT SINCE LATE PLEISTOCENE USING STRUCTURAL GEOMORPHOLOGY

The northeastern margin of Tibetan plateau is an active block controlled by the eastern Kunlun fault zone, the Qilian Shan-Haiyuan fault zone, and the Altyn Tagh fault zone. It is the frontier and the sensitive area of neotectonic activity since the Cenozoic. There are widespread folds, thrust faults and stike-slip faults in the northeastern Tibetan plateau produced by the intensive tectonic deformation, indicating that this area is suffering the crustal shortening, left-lateral shear and vertical uplift. The Riyueshan Fault is one of the major faults in the dextral strike-slip faults systems, which lies between the two major large-scale left-lateral strike-slip faults, the Qilian-Haiyuan Fault and the eastern Kunlun Fault. In the process of growing and expanding of the entire Tibetan plateau, the dextral strike-slip faults play an important role in regulating the deformation and transformation between the secondary blocks. In the early Quaternary, because of the northeastward expansion of the northeastern Tibetan plateau, tectonic deformations such as NE-direction extrusion shortening, clockwise rotation, and SEE-direction extrusion occurred in the northeastern margin of the Tibetan plateau, which lead to the left-lateral slip movement of the NWW-trending major regional boundary faults. As the result, the NNW-trending faults which lie between these NWW direction faults are developed. The main geomorphic units developed within the research area are controlled by the Riyueshan Fault, formed due to the northeastward motion of the Tibet block. These geomorphic units could be classified as:Qinghai Lake Basin, Haiyan Basin, Datonghe Basin, Dezhou Basin, and the mountains developed between the basins such as the Datongshan and the Riyueshan. Paleo basins, alluvial fans, multiple levels of terraces are developed at mountain fronts. The climate variation caused the formation of the geomorphic units during the expansion period of the lakes within the northeastern Tibetan plateau. There are two levels of alluvial fans and three levels of fluvial terrace developed in the study area, the sediments of the alluvial fans and fluvial terraces formed by different sources are developed in the same period. The Riyueshan Fault connects with the NNW-trending left-lateral strike-slip north marginal Tuoleshan fault in the north, and obliquely connects with the Lajishan thrust fault in the south. The fault extends for about 180km from north to south, passing through Datonghe, Reshui coal mine, Chaka River, Tuole, Ketu and Xicha, and connecting with the Lajishan thrusts near the Kesuer Basin. The Riyueshan Fault consists of five discontinuous right-step en-echelon sub-fault segments, with a spacing of 2~3km, and pull-apart basins are formed in the stepovers. The Riyueshan Fault is a secondary fault located in the Qaidam-Qilian active block which is controlled by the major boundary faults, such as the East Kunlun Fault and the Qilian-Haiyuan Fault. Its activity characteristics provide information of the outward expansion of the northeastern margin of Tibet. Tectonic landforms are developed along the Riyueshan Fault. Focusing on the distinct geomorphic deformation since late Pleistocene, the paper obtains the vertical displacement along the fault strike by RTK measurement method. Based on the fault growth-linkage theory, the evolution of the Riyueshan Fault and the related kinetic background are discussed. The following three conclusions are obtained:1)According to the characteristics of development of the three-stage 200km-long steep fault scarp developed in the landforms of the late Pleistocene alluvial fans and terraces, the Riyueshan Fault is divided into five segments, with the most important segment located in the third stepover(CD-3); 2)The three-stage displacement distribution pattern of the Riyueshan Fault reveals that the fault was formed by the growths and connections of multiple secondary faults and is in the second stage of fault growth and connection. With CD-3 as the boundary, the faults on the NW side continue to grow and connect; the fault activity time on the SE side is shorter, and the activity intensity is weaker; 3)The extreme value of the fault displacement distribution curve indicates the location of strain concentration and stress accumulation. With the stepover CD-3 as the boundary, the stress and strain on NW side are mainly concentrated in the middle and fault stepovers. The long-term accumulation range of stress on the SE side is relatively dispersed. The stress state may be related to the counterclockwise rotation inside the block under the compression of regional tectonic stress.     

融合随机森林和超像素分割的建筑物自动提取

建筑物是城市空间的重要部分,建筑物信息的提取对基础地理空间数据库更新、城市规划、城市动态监测等具有重要意义.基于遥感影像数据提取建筑物信息具有非常广泛的应用,本文提出了一种基于随机森林和超像素分割算法,并从机载激光点云和数字航空影像数据中自动提取建筑物的方法.试验选取广州市海珠区某处为研究区域,结果表明:在一般的城市区...     

多尺度SLIC-GMRF与FCNSVM联合的高分影像建筑物提取

遥感影像建筑物提取具有重要的应用价值。然而,高分辨率遥感影像中细节信息繁多、特征复杂,增加了建筑物提取难度。针对这一问题,本文提出一种基于多尺度SLIC-GMRF和FCNSVM的建筑物提取方法,一定程度上提高了高分辨率遥感影像建筑物提取能力。首先,利用多尺度SLIC-GMRF分割算法确定初始建筑物区域,然后,充分利用FCN神经网络在语义分割中的优势抽取建筑物特征,最后,结合提取出的建筑物特征训练SVM分类器细化建筑物提取结果,通过3种控制实验,两种对比方法得出以下结论:SLIC分割算法影响初始分割结果;SVM分类器影响建筑物细部提取;FCN特征影响SVM分类器性能。对于特征清晰、遮挡干扰较少的研究区,本文方法能够较好提取影像中的建筑物,查准率、查全率、质量指标均优于对比方法,对建筑物复杂分布的研究区同样能够取得较好的提取效果。