面向灾害信息提取的SAR图像并行算法设计与实现
发布时间：2018-11-20 09:01
【摘要】：随着遥感技术的发展,遥感在各个领域的应用越来越普遍,当前我国的遥感应用正处于发展阶段,有必要对人类赖以生存的自然环境进行长期的动态监测,对灾害信息进行有效的提取。合成孔径雷达(Synthetic Aperture Radar——SAR)是一种高分辨率成像雷达,具有全天时、全天候的特点,在灾害检测方面具有独特的优势,但由于SAR图像一般含有海量数据信息,而且SAR图像灾害算法计算模型比较复杂,所以针对SAR图像的灾害算法处理会耗费大量的时间。通常在灾害处理时必须依据图像快速实时地得出计算结果从而制定出可行的保护措施,因此有必要研究和设计SAR图像灾害检测算法的快速处理方法,并将其集成于SAR图像处理系统中。本文研究了面向灾害信息提取的快速处理算法,并分别给出了基于MPI+OpenMp和OpenCL的两种不同形式的并行处理方案,同时将设计好的算法模块集成于SAR图像处理软件中,主要研究工作如下:(1)设计了基于MPI集群加OpenMp共享存储的混合并行算法。主要针对道路损毁算法设计了多节点和节点内部的两级并行方案。在道路损毁提取算法的MPI+OpenMp混合并行实现中,对SAR图像数据进行了详细的数据分块。通过实验分析了节点数对混合模型算法的影响,证明了计算机集群与OpenMp的混合并行方案在小型试验室相对于串行算法的优越性。(2)由于GPU在处理SAR图像海量数据方面的优势,本文主要设计了基于OpenCL的道路损毁并行算法。在具体设计过程中,首先对算法可并行部分进行了性能优化,进一步提升了算法性能。基于优化后的算法,详细设计了存储以及线程划分方案,通过实验测试得到了最大13倍的并行加速比。(3)开发了SAR图像面向灾害信息提取软件,采用面向对象的构建方法,对各个功能模块进行了详细的设计。将算法模块和系统界面分离,实现了整个系统的低耦合性。对于每个灾害处理算法,针对不同的硬件环境,使用OpenMp+MPI和OpenCL进行并行加速,并将加速算法集成到软件系统中。
[Abstract]:With the development of remote sensing technology, the application of remote sensing in various fields is becoming more and more common. At present, the application of remote sensing in our country is in the developing stage. It is necessary to carry out long-term dynamic monitoring of the natural environment on which human beings depend for survival. The disaster information is extracted effectively. Synthetic Aperture Radar (Synthetic Aperture Radar--SAR) is a kind of high-resolution imaging radar, which has the characteristics of all-day, all-weather, and has unique advantages in disaster detection. However, SAR images generally contain massive data information. Moreover, the computational model of SAR image disaster algorithm is complex, so the disaster algorithm processing for SAR image will take a lot of time. Usually, in the disaster processing, we must get the calculation results quickly and in real time according to the image, so it is necessary to study and design the fast processing method of the SAR image disaster detection algorithm, so it is necessary to study and design the fast processing method of the disaster detection algorithm of the SAR image. It is integrated into SAR image processing system. In this paper, the fast processing algorithms for disaster information extraction are studied, and two different parallel processing schemes based on MPI OpenMp and OpenCL are presented, and the designed algorithm modules are integrated into the SAR image processing software. The main research work is as follows: (1) A hybrid parallel algorithm based on MPI cluster and OpenMp shared storage is designed. A two-level parallel scheme of multi-node and inside-node is designed for road damage algorithm. In the MPI OpenMp hybrid parallel implementation of road damage extraction algorithm, the SAR image data is divided into blocks in detail. The effect of the number of nodes on the hybrid model algorithm is analyzed through experiments. It is proved that the hybrid parallel scheme of computer cluster and OpenMp is superior to serial algorithm in small laboratory. (2) because of the advantage of GPU in processing massive data of SAR image, this paper mainly designs a road damage parallel algorithm based on OpenCL. In the specific design process, the performance of the parallelism part of the algorithm is optimized, which further improves the performance of the algorithm. Based on the optimized algorithm, the storage and thread partition schemes are designed in detail, and the maximum parallel speedup ratio of 13 times is obtained through experimental tests. (3) the SAR image disaster information extraction software is developed, and the object-oriented construction method is adopted. Each function module is designed in detail. The algorithm module is separated from the system interface to realize the low coupling of the whole system. For each disaster processing algorithm, OpenMp MPI and OpenCL are used for parallel acceleration for different hardware environments, and the acceleration algorithm is integrated into the software system.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
本文编号：2344481
[Abstract]:With the development of remote sensing technology, the application of remote sensing in various fields is becoming more and more common. At present, the application of remote sensing in our country is in the developing stage. It is necessary to carry out long-term dynamic monitoring of the natural environment on which human beings depend for survival. The disaster information is extracted effectively. Synthetic Aperture Radar (Synthetic Aperture Radar--SAR) is a kind of high-resolution imaging radar, which has the characteristics of all-day, all-weather, and has unique advantages in disaster detection. However, SAR images generally contain massive data information. Moreover, the computational model of SAR image disaster algorithm is complex, so the disaster algorithm processing for SAR image will take a lot of time. Usually, in the disaster processing, we must get the calculation results quickly and in real time according to the image, so it is necessary to study and design the fast processing method of the SAR image disaster detection algorithm, so it is necessary to study and design the fast processing method of the disaster detection algorithm of the SAR image. It is integrated into SAR image processing system. In this paper, the fast processing algorithms for disaster information extraction are studied, and two different parallel processing schemes based on MPI OpenMp and OpenCL are presented, and the designed algorithm modules are integrated into the SAR image processing software. The main research work is as follows: (1) A hybrid parallel algorithm based on MPI cluster and OpenMp shared storage is designed. A two-level parallel scheme of multi-node and inside-node is designed for road damage algorithm. In the MPI OpenMp hybrid parallel implementation of road damage extraction algorithm, the SAR image data is divided into blocks in detail. The effect of the number of nodes on the hybrid model algorithm is analyzed through experiments. It is proved that the hybrid parallel scheme of computer cluster and OpenMp is superior to serial algorithm in small laboratory. (2) because of the advantage of GPU in processing massive data of SAR image, this paper mainly designs a road damage parallel algorithm based on OpenCL. In the specific design process, the performance of the parallelism part of the algorithm is optimized, which further improves the performance of the algorithm. Based on the optimized algorithm, the storage and thread partition schemes are designed in detail, and the maximum parallel speedup ratio of 13 times is obtained through experimental tests. (3) the SAR image disaster information extraction software is developed, and the object-oriented construction method is adopted. Each function module is designed in detail. The algorithm module is separated from the system interface to realize the low coupling of the whole system. For each disaster processing algorithm, OpenMp MPI and OpenCL are used for parallel acceleration for different hardware environments, and the acceleration algorithm is integrated into the software system.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
【参考文献】
相关期刊论文 前1条
1 滕人达;刘青昆;;CUDA、MPI和OpenMP三级混合并行模型的研究[J];微计算机应用;2010年09期
相关硕士学位论文 前1条
1 程香丽;地震灾区道路崩塌体岩块施工难度系数专题图研究[D];西南交通大学;2011年
，本文编号：2344481
本文链接：https://www.wllwen.com/kejilunwen/wltx/2344481.html
