【作者】 李一芒；

【导师】 何昕；

【作者基本信息】 中国科学院研究生院(长春光学精密机械与物理研究所) ， 机械电子工程， 2014， 博士

【摘要】 在东海海域进行防空识别区的划立，标志着我国对外部威胁的预警工作迈入了新阶段，同时说明了预警工作在国防领域的重要地位。雷达作为一种有源探测设备，在预警工作中一般担负主要任务，但其缺陷难以克服。作为雷达探测的补充手段，光学无源预警系统受到了广泛的关注。实时图像处理系统是光学无源预警系统的核心部分，其相关研究主要分为硬件结构和软件算法两个方面。近30年，数字图像处理技术发展迅速，硬件结构方面朝着速度更快、集成化更好的方向发展，软件算法方面则涌现出了很多高效、易于工程实现的方法。这些都为光学无源预警系统的发展打下了坚实的基础。本文以周扫式多谱段无源预警实时图像处理系统为研究对象，开展的内容主要有：一、根据预警系统采用多谱段探测器进行“线面结合”的成像方式，且线阵探测器圆周扫描成像为主的特点，提出了新的预警策略和图像处理机制；二、对基于数字图像处理技术的目标检测与目标识别算法进行了讨论研究，为实现系统功能提供理论基础；三、对实现目标检测与识别的特定硬件平台加以研究和设计，以实时地完成系统对外部威胁的预警功能。周扫式多谱段无源预警系统能在一定策略下通过多个谱段的光学探测器进行目标探测和识别，本文以红外线阵探测器和可见光面阵探测器所组成的多谱段探测系统为例，开展对周扫式多谱段无源预警系统的研究。针对不同谱段探测器的应用，提出线面结合的探测方式。提出采用多级预警方法针对线阵探测器圆周扫描过程中发现的目标和面阵探测器进行确认时所识别的目标给出不同级别的预警信号；结合线阵探测器捕获图像的特点和图像处理平台高速率、大数据量的特点，定义图像子帧，对大视场图像进行分割处理；采用处理器内部多路处理的方法取代原有的等待方法，解决因对大视场图像进行分割而可能引起目标损坏的问题。提出了一种基于多方位一维滤波器的目标检测算法对红外线阵探测器捕获图像进行处理，通过改变各方位滤波器的参数检测不同形状的目标，使滤波器在二维空间内尺度更灵活，解决了传统二维滤波器各项同性的问题。提出了一种基于充分降维算法和支持向量机的目标分类方法，通过充分降维理论，降低支持向量的个数和向量的特征维数，充分降维方法能够在不影响分类结果的同时将维数降到最低，较其它特征简化算法更科学。根据多谱段预警系统的技术指标要求，设计一种多处理器并行架构的数字图像处理系统，通过多核间任务分配与信息交互提升系统对信息的处理速度。在数字图像处理系统中集成了PCI-e通信单元，使系统能与控制计算机间实现高速数据交互，同时优化了PCI-e通信单元，通过FPGA内核实现通信协议，取代了采用控制芯片实现通信协议的方法。在对算法的推导和对嵌入式实时图像处理系统的硬件设计过程中，进行了多项实验测试以验证所提算法的正确性和可行性。最后通过采集空中飞机目标的实验对系统的性能加以评估和分析。实验结果证明，系统的算法及硬件设计能够实现对空探测和预警功能，达到了预期目的。更多还原

【Abstract】 Setting air defense identification zone in the East China Sea, marks China’searly warning to the external threat has entered a new stage and shows theimportance of early warning in the national defense field. Radar is used as an activedetection equipment, taking the main task in the early warning work. But it’s difficultto overcome the shortcomings. As a supplementary means of radar, optical passiveearly warning system has received extensive concern and attention. Real-time imageprocessing system is the core of the optical passive warning systems and the relatedresearch is divided into hardware and software algorithms. In recent30years, digital image data processing technology has been developed rapidly. Thehardware is faster and better integrated in the direction of development. For thesoftware algorithm, there has been a lot of high efficiency method forengineering applications. These has established solid foundation for the developmentof optical passive warning system.Researching on real-time image processing system of multi-spectral scanningpassive early warning system was the main task in this paper. The main contentsincluded the following. First, according to the characteristics of the “combining lineand surface” imaging mode that line array detector circle scanning is the mainmode, it proposed new early warning strategy and image processing mechanism.Second, the target detection and recognition algorithm was discussed based on digital image processing technology and provided a theoretical basis in order torealize the system function. Finally, the specific hardware platform was designed toachieve the target detection and recognition and complete the system function inreal-time.Passive multi-spectral scanning early warning system was able to finish targetdetection and recognition by multiple spectrum optical detectors in a certain strategy.An infrared linear array detector and a visible light surface array detector wascomposed as an example for the research on passive multi-spectral scanning earlywarning system in this paper. For different spectral detector applications, it proposed“combining line and surface” detection methods. Different early warning levelsignals of was given based on multilevel early warning method for the identificationof the “combining line and surface” imaging mode. Combined with thecharacteristics of line array infrared detector and high rate and huge data of imageprocessing hardware platform, image sub frame was defined to process thesegmentation of wide field image. Using multi-processor internal processing methodto replace the original method of waiting, it solved the problem by targetsegmentation. It proposed a novel target detection algorithm based onone-dimensional local filter for the image captured by infrared array detector.Through changing the parameters, it was able to detect different shapes. Theproposed filter solving the traditional2D filter isotropic problems was more flexiblein two-dimensional space scale. A classification method based on support vectormachine using sufficient dimension reduction algorithm was proposed to identify thetargets. The dimension of feature and the number of vectors was reducedthrough sufficient dimension reduction theory. Sufficient dimensionreduction method didn’t affect the effective features at the sametime the dimension was reduced to the lowest and the method was more scientificcompared with other features simplified method. According to technical indicators ofmultispectral warning system, a digital image processing system based onmulti-processor parallel architecture was designed. Task allocation and information exchange between multi-cores was used to improve the processing speed of thesystem. The PCI-e communication unit was integrated in the digital imageprocessing system instead of using the control chip to realize the communicationprotocol to realize the implementation of high speed data communication.The feasibility and correctness of the algorithm and the real-time imageprocessing system hardware platform was verified by several experimental.Finally, the evaluation on the performance of the system was analyzed throughthe experiment on some images including aerial target. The experimental resultsshow that, the algorithm and the hardware design of the system to meet therequirement of the early warning system and achieve the expected goal.更多还原