【作者】 祝永坚；

【导师】 陈钱；

【作者基本信息】 南京理工大学 ， 光学工程， 2007， 博士

【摘要】 巡航导弹的飞速发展决定了其在未来高科技战争中占有举足轻重的地位，而航空成像和景象匹配制导技术的结合给了巡航导弹明亮的眼睛。由于景象匹配的基准图和实时图是在不同时间、不同成像条件下、用不同探测器、从不同视角获得的，即使是同一个物体，在图像中物体所表现出来的光学特性(灰度值、颜色值等)、几何特性(外形、大小等)及空间位置(图像中位置、方向等)都有很大的不同，再加上噪声、干扰物体等因素的存在使得图像有很大的不同。当巡航导弹在夜间飞行时，所采集的实时图为微光夜视图像，而基准图一般都是白天气象条件较好的时候采集的卫星图片或者航拍图片，而日间与夜间的景物在图像中呈现明显的不同，因此匹配时就容易产生偏差，甚至误匹配，极大地影响了制导的精度。为此需要对微光成像进行仿真，通过完整地重现微光成像系统的各个工作环节，把正常日照条件下的图像(基准图)精确地转换为微光夜视图像。微光景象匹配基准图模拟生成技术是提高巡航导弹夜间景象匹配制导精度的有效途径之一。微光图像相比于日光图像最主要的特征就是信噪比低。本文首先从直方图、时空域噪声、图像相关性、灰度层次丰富性、图像信息量等角度考察了微光图像的特征，在分析二维直方图的基础上提出用信息容量这一示性参数来表征夜视图像信息的丰富程度和局部相关程度，这些特征对评价微光图像仿真效果具有重要的指导意义。大气传输是光电成像的必经过程。图像天候模型的建立向来是个难点，大气是个随机变化的复杂体，受到各种自然因素的影响。本文分析了光电成像过程中大气这一传输介质对成像质量的影响，采用辐射传输理论描述了大气介质的吸收和散射特性，建立了大气成像系统的传递函数模型。根据数学模型，编写了相应的计算机仿真软件，并且将LOWTRAN7作为一个独立模块调入以模拟大气造成的图像能量衰减和能量分布变化，为微光成像系统仿真奠定了基础。光谱反射特性是景物成像后显示出区别于其他景物的特征的根本原因，也是白天与夜间图像对比度发生变化的根本原因。本文以地面景物的光谱反射特性为突破口，从ICCD相机的工作原理出发，分析了日光与夜天光的光谱分行，深入讨论了地面光谱反射特性对成像的影响，对比研究了白天与夜间地面反射特性造成的日光图像与微光图像的对比度差异，分析了ICCD相机的光谱辐射响应和空间响应特性，建立了不同夜间光照情况下微光成像系统的统一模型，包括：ICCD相机积分成像模型、ICCD相机MTF链模型、ICCD相机噪声模型。针对微光成像仿真对图像分割的实际需求，本文提出了一种基于微光图像特征的图像分割新方法。讨论了K-均值聚类在图像分割中的作用，在详细分析了Canny算法整个过程的基础上，采用二维直方图最大熵法计算得到对Canny算法性能有决定意义的高门限。这种自适应方法不仅改善了Canny算法的分割效果，同时也增强了Canny算法的适用性和实用性。最后，运用数学形态学对边缘图像进行二次处理，使得图像边缘更清晰、特征更明显，得到较好的图像分割结果。最后，基于本文所分析和建立的模型，作者研制了“微光视景生成系统”，一个完整的蕴含LOWTRAN7计算内核、图像分割和分类以及ICCD后段仿真的微光景象匹配基准图模拟生成系统，并给出了实验结果。更多还原

【Abstract】 The rapid development of cruise missiles makes them play an important role in futurehigh-tech wars. The combination of aerial imaging and scene matching brings cruisemissiles bright eyes. Since the scene matching base map and the real-time map arecaptured at different time, in different imaging conditions, using different detectors, fromdifferent visual angles, different maps appear entirely different in optical characteristics(gray, color, etc.), geometric characteristics (shape, size, etc.) and spatial characteristics(location, direction, etc.), even if the same object. Furthermore, noise and interferencebring differences. When cruise missiles fly at night, the real-time map is night visionimages, while the base map is composed of satellite images or aerial photographs acquiredduring daytime in better weather conditions. The difference between daytime images andnight vision images greatly impact on the accuracy of scene matching and guidance.Therefore, the low-light-level imaging simulation is of necessariness. By reconstructing theLLL imaging system, daytime images (base map) are accurately converted to night visionimages. LLL Scene Matching base map simulation technology is one of the effective waysto improve the scene matching guidance accuracy of cruise missiles at night.The most important characteristic of LLL images is poor signal-to-noise ratiocompared to daytime images. Characteristics of LLL images are analyzed at the beginningof this paper, such as histograms, time-and-space-domain noise, image correlation, richnessof gray level and image information. Based on the analysis of two-dimensional histograms,information capacity is proposed as an indicator parameter to describe the richness andlocal correlation of night Vision image information. These characteristics are of vitaldirective significance to the evaluation of LLL image simulation.Atmospheric propagation is necessary in optoelectronic imaging process. Buildingimage weather model is always difficult, since the atmosphere is a complex systemchangeing randomly and effected by natural factors. In this paper, the influence ofatmospheric propagation on image quality in optoelectronic imaging process is analyzed.Radiative transfer theory is used to describe the atmospheric absorption and scatteringproperties and the atmospheric imaging system transfer function model is advanced. Basedon the mathematical model, the corresponding computer simulation software is developed.And LOWTRAN7 is embeded as a separate module to simulate the energy attenuation ofimages and energy distribution change due to atmospheric propagation, which laid the foundation for LLL imaging simulation.Spectral reflectance properties are the fundamental reasons for distinguishing onescene from another, as well as the contrast changes between daytime images and nightvision images. Based on the ground spectral reflectance properties and operationgprinciples of ICCD, spectral distributions of daylight and night sky light are analyzed. Theinfluence of ground spectral reflectance properties on imaging process is deeply discussed.The contrast difference between daytime images and LLL images is studied. Spetralradiance response and spatial response of ICCD are analyzed. A series of unifying modelsof LLL scene simulation is builded under different night illumination conditions, includingthe integral imaging model, MTF chain model and noise model of ICCD.,In order to meet the practical requirement of LLL imaging simulation, a novel imagesegmentation method based on low-light-level image characteristics is presented in thispaper. The contribution of K-means clustering in image segmentation is discussed. Thewhole process of Canny algorithm is analyzed in details. The maximum entropy based ontwo-dimensional histogram is proposed to determine the high threshold of Cannyalgorithm. This adaptive approach improves the effect of Canny algorithm and alsoenhances the applicability and practicability. Mathematical morphology is used in the edgeimage processingto get better image segmentation results.Finally, based on the models analyzed and builded in the paper, a real-time LLL scenesynthesis system is developed, including computing kernel of LOWTRAN7, imagesegmentation and category, ICCD simulation. And the experimental results are given atlast.更多还原