基于任务树调度的大尺度的遥感影像并行镶嵌技术的研究

发布时间：2018-08-14 13:36

【摘要】：在遥感技术与相关领域飞速发展的今天，遥感影像的内容信息分析已逐渐成为行业内部研究主流，单一尺度的遥感影像分析已不能满足对大区域预警与环境分析的要求，遥感影像镶嵌已成为遥感影像处理中的一个热门话题，然而遥感影像的特性与行业要求使得遥感影像镶嵌技术复杂冗长，处理流程繁杂不易简化，选择正确的方式方法处理大规模大尺度遥感图像的镶嵌问题变得十分重要，有效利用软硬件资源，充分分配计算资源与计算空间需要算法的支持与匹配。针对上述情况，本文研究了图像镶嵌技术，并行编程技术，任务调度方式方法，并分析了解了现有的大尺度遥感镶嵌算法中存在的问题，针对运行速率低与可扩展性差等普遍问题，提出了基于任务树调度的大尺度遥感影像并行镶嵌技术，，通过对大尺度遥感影像集进行解耦，解析影像间关系，构建具有前驱后续逻辑关系的任务树，利用高性能集群本地资源管理器分配集群高效的运算能力，智能调度监控，将已触发任务发送至就绪状态队列，等待空闲节点运行，节点中的子镶嵌任务采用MPI(Message Passing Interface)并行运算编程方法实现，达到多层次并行，充分利用现有硬件设施的目的。本文提出的调度方案为基于关键路径和状态队列（CPDS-SQ）的动态DAG(Directed Acyclic Graph)调度策略，并根据此调度策略完成了并行调度TTM（Task-tree Mosaic）系统，动态生成任务调度序列，有效减少任务树长度，增强任务并行度，大幅度缩减了并行镶嵌的运算时间，同时，动态调度过程中的纠错重发机制提高了程序的鲁棒性，对大尺度遥感图像的镶嵌实现提供了良好基础。本文针对上述调度方法设计实现了大尺度遥感镶嵌系统，详细阐述了系统软硬件的实现过程与结果。
[Abstract]:With the rapid development of remote sensing technology and related fields, the content information analysis of remote sensing image has gradually become the mainstream of research in the industry. The single scale remote sensing image analysis can no longer meet the requirements of regional early warning and environmental analysis. Remote sensing image mosaic has become a hot topic in remote sensing image processing. However, the characteristics and industry requirements of remote sensing image make remote sensing image mosaic technology complicated and lengthy, and the processing process is not easy to simplify. It is very important to choose the correct method to deal with the mosaic problem of large-scale and large-scale remote sensing images. The efficient use of software and hardware resources and the adequate allocation of computing resources and computing space need the support and matching of algorithms. In view of the above situation, this paper studies image mosaic technology, parallel programming technology, task scheduling method, and analyzes the existing problems in large-scale remote sensing mosaic algorithm. Aiming at the common problems such as low running rate and poor scalability, the parallel mosaic technology of large-scale remote sensing image based on task tree scheduling is proposed. By decoupling the large-scale remote sensing image set, the relationship between the images is analyzed. The task tree with the preprocessing and subsequent logic relationship is constructed, and the high performance cluster local resource manager is used to allocate the cluster's efficient computing power. The task is sent to the ready state queue to wait for the idle node to run, and the triggered task is sent to the ready state queue. The sub-mosaic task in the node is implemented by MPI (Message Passing Interface) parallel operation programming method to achieve multi-level parallelism and make full use of the existing hardware facilities. The scheduling scheme proposed in this paper is a dynamic DAG (Directed Acyclic Graph) scheduling strategy based on critical path and state queue (CPDS-SQ). According to this scheduling strategy, a parallel scheduling TTM (Task-tree Mosaic) system is implemented, which dynamically generates task scheduling sequences and effectively reduces the length of task tree. The task parallelism is enhanced and the computation time of parallel mosaic is greatly reduced. At the same time, the error correction and retransmission mechanism in dynamic scheduling improves the robustness of the program and provides a good foundation for the realization of large-scale remote sensing image mosaic. In this paper, a large scale remote sensing mosaic system is designed and implemented in accordance with the above scheduling method, and the realization process and results of the software and hardware of the system are described in detail.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【共引文献】