基于C6678平台的光学遥感图像典型目标检测

CASIA OpenIR > 毕业生 > 硕士学位论文

	基于C6678平台的光学遥感图像典型目标检测
	冯云海
	2016-05-27
学位类型	工程硕士
中文摘要	开展光学遥感图像典型目标检测特别是对港口及海面进行舰船检测等研究对于濒海国家的安全与发展具有重要意义。舰船检测系统作为其中一项重要应用，在特殊环境下需要配置高性能、低功耗、轻量小型的处理设备，因而数字信号器(DSP)得到大量应用。作为多核高性能DSP的代表，德州仪器(TI)公司的C6678尤其适合进行图像处理任务, 但是基于C6678进行此项专门的研究鲜有公开成果发表。卫星光学遥感图像一般尺寸较大，针对大尺寸图像的处理，在PC端有成熟的解决方案，但在内存受限的条件下没有一种较好的应对方案。基于以上背景，本论文关注C6678平台上超大尺寸光学遥感图像的并行处理，并以舰船检测作为典型展开研究。为了充分发挥C6678的多核优势，从多方面进行分析进行设计与优化：外部数据传输方面，使用高速以太网接口，实现了基于TCP连接的高速Socket传输方案，从数据输入阶段就大幅度提升了传输速度；数据存储方面，结合C6678的多级存储结构，针对大尺寸图像处理情境下内存受限的问题，提出了数据分片与数据共享的应对策略；数据处理方面，分为预处理、目标检测、后处理三个步骤，预处理能够大幅减少计算量，提高整体效率；并行模式方面，详细分析了两种不同的模式并确定了Master/Slave并行模式，使用MessageQ实现核间通信与同步。本论文的主要贡献与创新点包括：（1）针对特大尺寸光学遥感图像以及处理平台内存限制的情况，提出了“外部、内部数据分块”的策略。（2）在C6678上设计并实现了基于Master/Slave方式的并行处理方案，使用MessageQ实现了C6678多核之间的通信与任务同步。（3）实现了并行处理方式的光学遥感图像舰船目标检测算法并针对C6678进行了优化。（4）使用NDK实现了TCP连接的Socket以太网高速数据传输。综上所述，在研究了C6678的外围高速数据传输接口、多级存储结构、核间通信与同步机制的基础上，建立了C6678平台上光学遥感图像并行处理框架。该框架充分发挥了C6678的多核优势，提高了光学遥感图像处理的效率。
英文摘要	It is of great significance to carry out the research on typical target detection, especially on ship detection in the port or on the sea in optical remote sensing image, for the sake of safety and the development of the coastal countries. As an important application of ship detection system, the processing platform need to be equipped with devices with high performance, low power consumption, light weight and compact under special circumstances, and this is why the digital signal processors (DSP) are widely used. Texas Instruments’ multi-core, high performance DSP, namely, C6678 is very suitable for image processing tasks. However, seldom published research materials for this special topic on C6678 can be found. The major feature of satellite optical remote sensing image is the large size both in pixel and volume. To processing large sized image, there are good solutions on PC, but there is no good deal on DSPs in condition that the memory is limited. Provided all the above, this paper focuses on the parallel processing of super large sized optical remote sensing image on the C6678 platform, taking ship detection as a typical study case. In order to make full use of the multi-core C6678, design and optimization are analyzed from several aspects: as for the external data transmission, high-speed Ethernet interface is used, high speed Socket transmission scheme based on the TCP connection is established, which will greatly enhance the transmission speed during the data from the input stage; as for data storage, multi-level storage structure with C6678 and limited memory problems are considered, strategies are made for data slicing and data sharing; as for data processing, which is divided into preprocessing, target detection and post processing, can significantly reduce the amount of calculation and improve the overall efficiency; as for parallel model, a detailed analysis of two different kinds of parallelism are made and the Master/Slave is selected as the parallel mode, with the use of MessageQ to achieve inter core communication and synchronization. Creative work with innovative ideas can be described as follows: (1) Data slicing is performed outside and inside the DSP to deal with large-size images. (2)Master/Slave parallelism with MessageQ inter-core communication and task synchronization is implemented on C6678 DSP. (3) Refined ship detection algorithm which is suitable for C6678 is implemented in a parallel way. (4)A TCP connection is established between C6678 and the host using Socket model. In summary, based on the study of the high speed data transmission peripherals, the multi-level storage structure, inter core communication and synchronization mechanism of C6678, we build an optical remote sensing image parallel processing framework on the C6678 DSP, which makes full use of the C6678 multi-core superiority and enhances the processing efficiency of the optical remote sensing image.
关键词	光学遥感图像舰船检测 C6678 并行处理
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/11566
专题	毕业生_硕士学位论文
作者单位	中国科学院自动化研究所
第一作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	冯云海. 基于C6678平台的光学遥感图像典型目标检测[D]. 北京. 中国科学院大学,2016.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
冯云海_基于C6678平台的光学遥感图像（1700KB）	学位论文		限制开放	CC BY-NC-SA