Adaptable parallel strategy to extract polygons from massive classified images on multi-core clusters

作者：Chen, Zhenjie^*; Zhou, Chen; Li, Manchun^*; Zhu, A-xing; Huang, Qiuhao; Pian, Yuzhe

来源：Concurrency and Computation: Practice and Experience (CCPE) , 2017, 29(4): e3861.

DOI：10.1002/cpe.3861

• adaptable parallel strategy
• improved heuristic partitioning
• hierarchical task scheduling
• raster-to-polygon conversion
• geographical information system

It is always computing intensive and time consuming to extract polygons from massive classified images. Although parallel computing can improve the efficiency of geographical data processing, the performance of the conversion suffers from the trade-off between data decomposition and result stitching. In this paper, we present an adaptable parallel strategy that accelerates the conversion process on a multi-core cluster. The strategy improves the method of data decomposition and optimizes task scheduling. In our strategy, an adaptable decomposition method is used to partition raster data according to the data complexity of the raster dataset and computing capability of the computing nodes. Moreover, the hierarchical task scheduling optimizes task allocation and load balance among computing nodes for the procedure of parallel conversion and stitching. We implemented parallel algorithms based on a boundary linking algorithm using the adaptable parallel strategy and compared the performance with the performances of conventional parallel strategies. The results show that the processing time of experimental raster data was reduced from 1362.36 to 165.78 s and that a desirable speedup with the maximal value of 8.23 was achieved.

• 出版日期2017-2
• 单位南京大学