【作者】 么彬；

【导师】 李海森；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2009， 博士

【摘要】 有效地开展海洋资源开发、海洋工程以及相关领域研究的前提是获取可靠的水下场景信息,而反应这些信息最直观的手段就是水声图像。因此,具有三维地形测量能力的多波束测深仪和二维地貌成像能力的侧扫声纳成为上述领域使用最为广泛的海洋勘查仪器,而国际上这两类仪器又都朝着相互融合能够同时提供地形和地貌信息的方向发展。本文以国家863计划海洋技术领域目标导引类课题“浅水宽覆盖多波束测深系统”和黑龙江省中小企业创新基金资助项目“高分辨浅水多波束测深系统”等科研项目为背景,在多波束测深仪的理论基础上研制具有水下场景二维/三维成像能力的高分辨海底地形地貌探测系统,并对其相关技术展开研究,因此论文的研究内容可以分为理论研究和工程实践两大部分。理论研究部分主要包括多子阵波束域海底探测算法实现对海底散射回波信号的方位-强度联合估计以及在海底地形地貌探测中的适用性研究和基于MRF模型的声纳图像分割方法的研究。工程实践部分主要包括：多子阵波束域海底探测算法实时实现以及高分辨海底地形地貌探测系统的设计、实现和试验验证。首先,在对水下声学图像处理相关领域的研究进展分析的基础上,探讨了实现高分辨地形地貌探测所需解决的关键问题。随后针对海底散射回波信号的方位-强度联合估计问题,引入两种新的信号处理方法,多子阵波束域CAATI(Multiple Sub-array Beamspace-CAATI,简称MSB-CAATI)算法和多子阵波束域求根MUSIC (Multiple Sub-array Beamspace-Root-MUSIC,简称MSB-RMU)。通过计算机仿真研究了两种算法在低信噪比和小快拍数情况下的方位和强度估计性能,并利用根据IHO测量规范建立水下场景和实际海试数据验证算法的实用性。在声纳图像分割方面,针对传统的基于Markov随机场模型的图像分割方法存在模型参数估计不准确、求解速度慢、且容易陷入局部最优解等缺陷,提出一种固定参数的简化模型,通过对合成的纹理图像和实际声纳图像分割对算法的可行性和分割效果进行了理论分析。在工程实践方面,针对直接影响算法并行实现效率的两个因素,即并行处理平台的结构和算法的可并行性,展开研究。在并行处理平台结构方面,在对k元n立方网络研究的基础上提出了一种适用性更广的广义K元n立方网络,并对新网络的拓扑性质和路由算法进行研究。在算法可并行性方面,通过对MSB-RMU算法运算量的分析确定平均协方差矩阵的特征分解是运算量相对集中的部分,为此研究了基于并行Jacobi方法的特征值分解方法以及在FPGA上的实现方案,设计并实现了一套基于高速DSP和FPGA的并行信号处理系统。最后完成了高分辨海底地形地貌探测系统的工程实现,并对系统性能和各项技术指标进行了全面验证试验和考核。更多还原

【Abstract】 The precondition of the efficient research in ocean exploring resource, ocean engineering and other correlated fields is acquiring the credible information of underwater scene. As the most obvious measure to reflecting the information is underwater acoustic image, the multibeam bathymetry system with the capability of 3-D imaging and the side scan sonar with the capability of 2-D imaging become the ocean survey instruments used in above fields most broadly. And internationally the two kinds of instruments are all developing toward the direction that they can combine and provide information of terrain and physiognomy at the same time. The background of this paper is the project supported by National High Technology Research and Development Program of China, "Shallow Water Wide-coverage Multibeam Bathymetry System" and the project supported by Small and Medium-sized Enterprise Innovation Funds of Heilongjiang province, "High Resolution Shallow Water Multibeam Bathymetry System", etc. The purpose of this paper is to develop a set of high resolution seafloor terrain and physiognomy detection system that has the ability of underwater scene 2-D/3-D imaging on the basis of the theory of multibeam bathymetry system, and to research the correlated techniques. So the paper can be divided into two parts that theory research and engineering realization. The theory research includes:estimation of direction of arrival (DOA) and intensity of arrival (IOA) of seabed backscattering using multiple sub-array beam-space seafloor detection algorithm and the usability of the algorithm in the seafloor terrain and physiognomy detection system and sonar image segmentation algorithm based on Markov random field model. The engineering realization includes:the real-time realization of multiple sub-array beam-space seafloor detection algorithm and the design, realization and test proof of high resolution seafloor bathymetry and physiognomy detection system. Firstly, based on analysis of researching course of the correlated fields of underwater acoustic image processing, the key problems, which need to be resolved when realizing the high resolution seafloor terrain and physiognomy detection system, are discussed. And then to solve the problem of the estimation of DOA and IOA of the seabed backscattering signal, two kinds of new algorithm of signal processing, Multiple Sub-array Beam-space CAATI and Multiple Sub-array Beam-space Root-MUSIC, have been proposed. The estimation performance of two kinds of algorithm in the case of low SNR and few snaps has also been studied by computer simulation. And the processing of the sea test data and the simulation underwater scene data according to the IHO standard confirms the usability of the algorithm further more.In the part of sonar image processing, to the problem of traditional image segmentation algorithm based on Markov random field model, such as inaccurate model parameter, long resolving time and easy to trap in local optimum solution etc, a simplified model with fixed parameters is proposed. By applying the algorithm to the synthesized texture image and actual side scan sonar image, the practicability and segmentation performance of the algorithm have been analyzed theoretically.At the aspect of the architecture of the parallel processing platform, a new direct interconnect networks topology, generalized K-ary n-cubes networks based on k-ary n-cubes networks is proposed and the topological properties and routing algorithms have been researched too. After analyzing the algorithm computation load, a theory has been known that the Eigenvalue decomposition (EVD) of average covariance matrix is the part in which most computation load concentrates. So the method of EVD based on parallel Jacobi and the realization on FPGA have been studied. A parallel signal process system based on FPGA and DSP has also been designed and actualized.At last, the engineering realization and performance testing of high resolution seafloor terrain and physiognomy detection system have been completed.更多还原