Extraction of loess shoulder-line based on the parallel GVF snake model in the loess hilly area of China
Extraction of loess shoulder-line based on the parallel GVF snake model in the loess hilly area of China
复制标题
基于平行GVF蛇模型的我国黄土丘陵区黄土肩线提取
DOI：
10.1016/j.cageo.2012.08.014
复制
发表时间：
2013-03-01
影响因子：
4.4
通讯作者：
Qjan, Kejian
中科院分区：
文献类型：
--
作者：
Song, Xiaodong;Tang, Guoan;Qjan, Kejian
黄土肩线是表征和模拟黄土高原地貌最关键的地形特征。现有算法在复杂曲面的连续肩线提取、DEM质量和算法限制等方面存在不足。本文提出了一种新的方法，利用梯度矢量流（GVF）蛇形模型生成可以连接肩线断续片段的一体化轮廓。在此基础上，提出了一种新的蛇形模型初始种子选择准则，即取局部邻域的中值平滑值。这样就可以从肩线带中提取出黄土正负地形的相邻边界，为利用梯度矢量流求出真实的肩线奠定了基础。然而，对于大型DEM数据集，该方法的计算量仍然很大。本文提出了一种用于肩线自动提取的聚类并行计算方案，并利用并行GVF蛇形模型实现了该方案。在分析该方法原理的基础上，提出了一种集单程序多数据（SPMD）和主/从（M/S）编程模式于一体的有效并行算法。基于DEM数据的区域分解，对各分区进行规则分解，同时进行计算。在不同DEM数据集上的实验结果表明，与顺序模型相比，并行编程可以在不损失精度的前提下实现显著缩短执行时间的主要目标。本文混合算法的平均肩线偏移量为15.8 m，与已有的提取方法相比，在精度和效率上都取得了令人满意的结果。(c) 2012 Elsevier Ltd.版权所有。
Loess shoulder-lines are the most critical terrain feature in representing and modeling the landforms of the Loess Plateau of China. Existing algorithms usually fail in obtaining a continuous shoulder-line for complicated surface, DEM quality and algorithm limitation. This paper proposes a new method, by which gradient vector flow (GVF) snake model is employed to generate an integrated contour which could connect the discontinuous fragments of shoulder-line. Moreover, a new criterion for the selection of initial seeds is created for the snake model, which takes the value of median smoothing of the local neighborhood regions. By doing this, we can extract the adjacent boundary of loess positive-negative terrains from the shoulder-line zones, which build a basis to found the real shoulder-lines by the gradient vector flow. However, the computational burden of this method remains heavy for large DEM dataset. In this study, a parallel computing scheme of the cluster for automatic shoulder-line extraction is proposed and implemented with a parallel GVF snake model. After analyzing the principle of the method, the paper develops an effective parallel algorithm integrating both single program multiple data (SPMD) and master/slave (M/S) programming modes. Based on domain decomposition of DEM data, each partition is decomposed regularly and calculated simultaneously. The experimental results on different DEM datasets indicate that parallel programming can achieve the main objective of distinctly reducing execution time without losing accuracy compared with the sequential model. The hybrid algorithm in this study achieves a mean shoulder-line offset of 15.8 m, a quite satisfied result in both accuracy and efficiency compared with published extraction methods. (c) 2012 Elsevier Ltd. All rights reserved.