空间数据多尺度建模关键技术研究

发布时间：2018-06-05 05:33

本文选题：地理信息系统 + 空间数据　；参考：《国防科学技术大学》2014年博士论文

【摘要】：地理信息系统(GIS)经过半个世纪的发展,已广泛应用于地图导航、城市规划、公共卫生、国防、资源监测、物流等众多领域,在这些GIS应用中,空间数据是各种服务的基础,随着空间信息获取平台的不断增多,空间数据已成为海量,同时,移动GIS已逐渐发展为主要的应用,导致了数据量与频带、有限计算资源之间的巨大矛盾,建立空间数据的多尺度模型是解决这一矛盾的一个行之有效的方法。小波分析具有固有的多分辨率特性,在多尺度建模上具有明显优势,但目前基于小波的空间矢量数据多尺度建模方法仍存在很多问题,妨碍了实际应用中多尺度建模统一框架的建立。因此,本文重点对基于小波的空间数据多尺度建模相关关键技术进行研究,以便形成以小波为基础的统一且实用的多尺度建模方法。论文的主要工作如下:针对现有线矢量数据小波分析方法不便于控制局部误差的问题,提出了具有局部误差修正的线矢量数据小波变换算法。算法以高频分量定位低频表示误差过大区域,对这些区域利用设计的特定插值函数进行局部插值,以降低对应区域的局部低频表示误差;算法可利用计算式恢复出修正前的小波系数,从而实现精确重构。该算法可有效降低低频表示曲线的局部大误差,结果数据易于组织,数据量增加少。针对现有面向空间网格的Loop细分小波算法低频表示误差大、滤波器系数不易解算的问题,提出了平滑Loop细分小波算法。算法首先构建了一个通用提升细分小波的结构,进而通过分析修正点确定了平滑Loop细分小波的具体形式,最后依据逆细分结果随扰动变化最小的准则求解出滤波器系数。算法可得到细分小波滤波器系数随顶点度变化的解析解,并且显著降低了小波低频表示的误差。针对现有Loop细分小波变换计算方法效率较低的问题,提出了基于相邻位置计算的Loop细分小波快速算法(FLSWTMNPC)。该算法由四部分组成:建立适用于细分/逆细分的网格分片方法、统一形式的面片顶点环状组织方式、用于支持细分/逆细分计算的相邻顶点位置关系计算式、解决面片边界处顶点计算正确性的边界拼接方法。该算法能适用于Loop细分小波的分解和重构计算,数据组织精炼,计算效率高。研究了空间网格数据、栅格数据小波变换的GPU加速方法:针对现有细分网格的GPU加速方法只能适用细分计算、算法整体计算效率不高的问题,提出了FLSWTMNPC的GPU加速方法,使得细分小波变换可以完全在GPU上完成,提高了计算效率。对空间栅格数据小波变换的GPU加速,解决了分块小波系数的边界拼接问题,提出了多相形式的小波分解GPU加速方法,提高了数据的组织效率和计算效率。
[Abstract]:Geographic Information system (GIS) has been widely used in many fields such as map navigation, urban planning, public health, national defense, resource monitoring, logistics and so on. In these GIS applications, spatial data is the basis of various services. With the increasing of the spatial information acquisition platform, the spatial data has become a mass. At the same time, the mobile GIS has gradually developed into the main application, resulting in a huge contradiction between the amount of data, the frequency band, and the limited computing resources. Establishing multi-scale model of spatial data is an effective method to solve this contradiction. The wavelet analysis has the inherent multi-resolution characteristic and has obvious superiority in the multi-scale modeling. However, there are still many problems in the wavelet based multi-scale modeling method of space vector data. It hinders the establishment of a unified framework for multi-scale modeling in practical applications. Therefore, this paper focuses on the research on the key technologies of wavelet based multi-scale modeling of spatial data in order to form a unified and practical multi-scale modeling method based on wavelet. The main work of this paper is as follows: aiming at the problem that the existing wavelet analysis method of line vector data is not convenient to control the local error, a wavelet transform algorithm of line vector data with local error correction is proposed. The algorithm uses the high frequency component to locate the large region of the low frequency representation error, and makes use of the special interpolation function to carry on the local interpolation to these regions, in order to reduce the local low frequency representation error of the corresponding region. The modified wavelet coefficients can be recovered by using the computational formula, so that the accurate reconstruction can be realized. The algorithm can effectively reduce the local large error of the low frequency representation curve, the result is easy to organize and the amount of data increases less. A smooth Loop subdivision wavelet algorithm is proposed to solve the problem that the low frequency representation error is large and the filter coefficients are difficult to solve in the existing spatial grid-oriented Loop subdivision wavelet algorithm. The algorithm first constructs the structure of a general lifting subdivision wavelet, and then determines the concrete form of smooth Loop subdivision wavelet by analyzing the correction points. Finally, the filter coefficients are calculated according to the criterion that the inverse subdivision results vary least with the disturbance. The algorithm can obtain the analytical solution of the coefficient of subdivision wavelet filter with vertex degree, and reduce the error of wavelet low frequency representation. Aiming at the low efficiency of existing Loop subdivision wavelet transform, a fast Loop subdivision wavelet algorithm based on adjacent position calculation is proposed. The algorithm is composed of four parts: a mesh segmentation method suitable for subdivision / inverse subdivision, a uniform form of planar vertex ring organization, which is used to calculate the position relationship of adjacent vertices in support of subdivision / inverse subdivision. A boundary splicing method to solve the correctness of the vertex calculation at the boundary of the surface. The algorithm can be applied to the decomposition and reconstruction of Loop subdivision wavelets, and the data organization is refined, and the calculation efficiency is high. The GPU acceleration method of spatial grid data and raster data wavelet transform is studied. Aiming at the problem that the existing GPU acceleration method of subdivision grid can only be applied to subdivision calculation and the overall computational efficiency of the algorithm is not high, the GPU acceleration method of FLSWTMNPC is proposed. The subdivision wavelet transform can be completed on GPU, and the computational efficiency is improved. The GPU acceleration of spatial grid data wavelet transform solves the boundary splicing problem of block wavelet coefficients. A multiphase wavelet decomposition GPU acceleration method is proposed to improve the efficiency of data organization and calculation.
【学位授予单位】：国防科学技术大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P208

【相似文献】