基于GPU混合反演的隧道电阻率超前探测成像研究

隧道施工期超前探测对于避免突涌水灾害的发生具有重要作用,为满足隧道三维电阻率超前探测快速化解译与成像的要求,本文提出了一种基于GPU并行的蚁群算法与最小二乘方法相结合的混合反演算法.该方法结合线性反演与非线性反演的优点,利用蚁群算法全局搜索能力强的优点为最小二乘反演提供较优的初始模型,以克服最小二乘算法容易陷入局部最优的缺点,提高了隧道三维电阻率反演成像的精度.同时,基于蚁群算法的天然并行性,提出了CUDA环境下的GPU并行策略,实现了三维电阻率反演的快速化成像.其次,开展了基于GPU混合反演的数值算例,与传统最小二乘线性反演进行了对比,基于GPU并行计算的混合反演计算效率得到了显著提高,对含水构造的位置、形态有较好的反映,压制了三维反演的多解性.最后开展了物理模型试验,结果表明基于GPU混合反演探测的低阻异常体与实际含水构造的位置较为相符,发现基于GPU加速的混合反演方法在提高探测精度与加快反演速度方面具有显著优势,为三维电阻率混合反演方法在隧道超前探测实际工程中的应用奠定了基础.     

基于GPU并行的时间域全波形优化共轭梯度法快速GPR双参数反演

探地雷达（GPR）时间域全波形反演计算量巨大,内存要求高,在微机上计算难度大.本文中作者基于GPU并行加速的维度提升反演策略,采用优化的共轭梯度法,避免了Hessian矩阵的计算,在普通微机上实现了时间域全波形二维GPR双参数（介电常数和电导率）快速反演.论文首先推导了二维TM波的时域有限差分法（FDTD）的交错网格离散差分格式及波场更新策略.然后,基于Lagrange乘数法,将约束问题转化为无约束最小问题,构建了共轭梯度法反演目标函数,采用Fletcher-Reeves公式与非精确线搜索Wolfe准则,确保了梯度方向修正因子及迭代步长选取的合理性.而GPU并行计算及维度提升反演策略的应用,数倍地提升了反演速度.最后,开展了3个模型的合成数据的反演实验,分别从观测方式、梯度优化及天线频率等方面,分析了这些因素对雷达全波形反演的影响,说明双参数的反演较单一的介电常数反演,能提供更丰富的信息约束,有效提高模型重建的精度.     

Implementation of the parallel mean shift-based image segmentation algorithm on a GPU cluster

The mean shift image segmentation algorithm is very computation-intensive. To address the need to deal with a large number of remote sensing (RS) image segmentations in real-world applications, this study has investigated the parallelization of the mean shift algorithm on a single graphics processing unit (GPU) and a task-scheduling method with message passing interface (MPI)+OpenCL programming model on a GPU cluster platform. This paper presents the test results of the parallel mean shift image segmentation algorithm on Shelob, a GPU cluster platform at Louisiana State University, with different datasets and parameters. The experimental results show that the proposed parallel algorithm can achieve good speedups with different configurations and RS data and can provide an effective solution for RS image processing on a GPU cluster.     

自适应分块加权Wallis并行匀色

针对区域范围内多幅待镶嵌影像之间的色彩差异问题,提出一种基于GPU的分块加权Wallis并行匀色算法。首先,根据变异系数对影像自适应分块并利用双线性插值确定每一个像素的变换参数,利用加权Wallis变换消除影像间的色彩差异。然后,为了控制区域整体的匀色质量,利用Voronoi图和Dijkstra算法确定影像间的处理顺序。最后,利用GPU技术进行并行任务设计并从配置划分、存储器访问和指令吞吐量等方面进行优化,提高算法运算效率。实验结果表明,本文方法既能有效地消除影像间色彩差异,又能消除影像间的对比度差异。与CPU串行算法相比,GPU并行算法显著减少了计算时间,加速比最高达到60倍以上。     

一种基于GPU的大规模地形实时生成算法

大规模地形实时可视化一直以来都是计算机图形学、虚拟现实和仿真领域研究的重点和热点。利用Geometry Clipmap算法的基本思想,改进了数据的组织方法,简化了算法的裂缝消除技术,对视景体的裁切算法进行了改进,实现了海量数据的实时可视化。     

一种基于分块策略的SIFT特征快速提取与匹配方法

通过研究基于GPGPU的SIFT特征匹配,提出一种通过分块处理实现大面阵遥感影像快速匹配的方法。首先采用鲁棒性强的随机采样一致性算法来估计待匹配影像间的单应关系并剔除误匹配点;然后通过单应关系找到待匹配影像间的重叠区域,进而实现大面阵遥感影像的分块SIFT特征提取和匹配。试验证明该算法可较快提取相应航摄测区的同名像点,为后续进行的光束法区域网平差提供了鲁棒的数据源。     

The effects of topographic depressions on multiscale overland flow connectivity: A high‐resolution spatiotemporal pattern analysis approach based on connectivity statistics

In watershed modelling, the traditional practice of arbitrarily filling topographic depressions in digital elevation models has raised concerns. Advanced high‐resolution remote sensing techniques, including airborne scanning laser altimetry, can identify naturally occurring depressions that impact overland flow. In this study, we used an ensemble physical and statistical modelling approach, including a 2D hydraulic model and two‐point connectivity statistics, to quantify the effects of depressions on high‐resolution overland flow patterns across spatial scales and their temporal variations in single storm events. Computations for both models were implemented using graphic processing unit‐accelerated computing. The changes in connectivity statistics for overland flow patterns between airborne scanning laser altimetry‐derived digital elevation models with (original) and without (filled) depressions were used to represent the shifts of overland flow response to depressions. The results show that depressions can either decrease or increase (to a lesser degree and shorter duration) the probability that any two points (grid locations) are hydraulically connected by overland flow pathways. We used macro‐connectivity states (Φ) as a watershed‐specific indicator to describe the spatiotemporal thresholds of connectivity variability caused by depressions. Four states of Φ are identified in a studied watershed, and each state represents different magnitudes of connectivity and connectivity changes (caused by depressions). The magnitude of connectivity variability corresponds to the states of Φ, which depend on the topological relationship between depressions, the rising/recession limb, and the total rainfall amount in a storm event. In addition, spatial distributions of connectivity variability correlate with the density of depression locations and their physical structures, which cause changes in streamflow discharge magnitude. Therefore, this study suggests that depressions are “nontrivial” in watershed modelling, and their impacts on overland flow should not be neglected. Connectivity statistics at different spatial scales and time points within a watershed provide new insights for characterizing the distributed and accumulated effects of depressions on overland flow.     

GPU/CPU协同粗粒度并行计算及在城市区域震害模拟中的应用

采用精细结构模型和动力时程分析以提高城市区域建筑震害预测精度已经成为一重要研究方向,而传统的CPU计算平台成本过高,难以推广。本文提出采用基于GPU/CPU协同粗粒度并行计算的方法来实现城市区域建筑震害的高效精细化动力时程计算,可以显著提高效率并降低成本。简述了所采用的程序架构、计算模型、参数选取,对并行计算的效率进行了详细的讨论,并通过一个中等大小城市的案例展示了该方法的优势。     

三维逆时偏移GPU/CPU机群实现方案研究

叠前逆时偏移是当前最为准确的地震成像方法,由于计算量大、存储量大等原因需要合适的实现策略和高效的计算平台.本文以高阶有限差分逆时偏移为基础,重点讨论了在GPU上实现需要解决的显存不足问题和人工边界问题.利用区域分解技术可以在当前GPU上高效地实现任意生产规模的三维逆时偏移成像,不会受到GPU显存规模的制约.常规最佳匹配层边界条件边界区域控制方程与内部区域差异较大,不适于GPU高速运算.本文在GPU上实现近似最佳匹配层(NPML)边界条件,使得高阶有限差分计算不需要分支判断,边界区域辅助波场的存储量也较低,保证了在GPU上进行波场传播的高效性.三维理论数据和实际资料成像结果表明了本文方法的正确性.     

GPU提速叠前时间体偏移技术

为了进一步提高叠前时间体偏移的计算效率,实现了在GPU＼CPU协同并行计算模式下Kirchhoff叠前时间体偏移技术,并进行优化。经在Nvida Tesla C1060GPU上的测试表明,GPU（Graphic Processing Unit）的处理速度是CPU（单核）的四十倍左右。同时表明,CUDA（Cornpute Unified Device Architectarc）编程为CPU向GPU的转化提供了一个较为方便的语言环境。