面向剖分面片模板的遥感影像并行处理方法

Research on remote sensing image parallel processing method for partition facet template

空间数据特别是遥感影像数据的快速增加和应用需求的扩大,其组织效率和处理速度已经成为制约技术应用的瓶颈,地球剖分理论和高性能计算为上述问题的解决提供了一种可能途径。针对上述问题,在遥感影像剖分面片数据模型的研究基础上,提出了剖分面片模板并行计算模式,设计并实现了一种面向剖分面片模板的遥感影像并行处理方法。该方法基于MPI(message passing interface)与Open MP(open multi-processing)混合并行计算框架,构建算法并行处理模型,形成算法并行化类库,通过调用其内部方法实现计算任务的并行执行。通过一个遥感影像剖分化并行分割处理实例,验证了该方法的有效性。实验结果表明,该方法具有较好的分割效果和加速比,有一定的示范意义,为进一步提高遥感影像应用能力提供了借鉴。

Alongside with the rapid increase of spatial data, especially remote sensing data and the growing application de- mands, spatial data organizational efficiency and processing speed have already become the bottleneck hindering its applica- tion. At present, the earth partition theory and high performance computing technology provide a possible approach for solving the above problems. In response to these problems, this paper put forward a new parallel computing pattern which was based on the remote sensing image data model of partition facets template. The theoretical basis of the new pattern included the earth partition model and the parallel computing theory. Guided by the hybrid parallel computing framework of MPI and 0penMP, the researchers discovered the parallel algorithm library and adjusted the internal application method so as to realize parallel implementation of computing tasks. Meanwhile, the new computing method was based on the parallel computing model of re- mote sensing images. It described the details as follows. Firstly, in accord with the specific application needs, this paper chose the appropriate remote sensing image and subdivide the image based on the EMD model, thus getting different levels par- tition facets. Secondly, it extracted the features of partition facets, generated the partition facets template, and built the data- base of partition templates. Thirdly, it analyzed computing hotspot of the concrete processing algorithm and implemented the corresponding parallel processing model. At last, it designed the parallel class and form the class library of parallel processing algorithm of remote sensing image. On the whole, the new method could be easily applied, while users need not master the relevant parallel developing experience or knowledge about the subdivision organization mechanism of spatial data. By instanti- ating the specific parallel processing class, initializing the parameters, and calling the method interface of the specific parallel class, they were able to implement the paralleling process of remote sensing image. Besides, it verified the effectiveness of the proposed method by a remote sensing image segmentation instance. The experiments demonstrate the segmentation result, while the running speed of the segmentation algorithm can be improved. The proposed method is of considerable practical sig- nificance and reference value for improving the processing ability of remote sensing image.