【作者】 陶明慧；

【导师】 任建岳；

【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） ， 光学工程， 2012， 博士

【摘要】 随着遥感相机技术对地面分辨率和视场覆盖范围的要求越来越高，单图像传感器能够提供的图像信息已经不能满足人们的需求，基于大面阵CCD的制造工艺越来越复杂、成本越来越高的情况，不断有人提出CCD拼接的技术满足目前对大视场的要求。对目前的CCD拼接技术进行总结后发现反射式分光法具有能量利用率高、拼接长度长、无像差等优点而成为具有广泛应用前景的拼接技术。本文的反射镜光学拼接方法使用普通反射镜为分光元件，比其它的反射式分光法使用的反射镜加工工艺简单、扩展性更好，具有很高的实际应用价值。但是反射式分光法不可避免都存在渐晕问题，本文针对反射镜光学拼接系统中的渐晕现象进行了深入的研究，并提出了有效的校正方法实现了渐晕的处理。1．从几何光学与物理光学两个角度分别对反射镜光学拼接系统的渐晕现象进行了定量分析：从几何光学角度分析了由于反射镜遮挡导致的部分光线无法在像面上成像出现的能量损失问题，推导出这种情况下渐晕系数的表达式；从物理光学角度分析了由于反射镜边缘直边衍射导致的像面光强分布不均的问题，推导出了这种情况下的渐晕衍射能量分布。为了对反射镜拼接系统的成像质量进行具体分析，在ZEMAX中采用多重结构的方法构建了系统的成像模型，使用动态数据交换机制和光学追迹的方法求解系统的PSF和MTF，以此来分析系统的成像特性和渐晕失真的原因。2．为了对反射镜拼接渐晕处理提供一个良好的实验平台，设计了一套基于反射镜光学拼接的双面阵CCD成像系统。使用KAI-2093为图像传感器，详细分析了它的工作原理，并设计了一套合适的驱动电路和高速图像采集电路，采用FPGA片内缓存和RAM块乒乓工作相结合的方式提高了系统信号的完整性和实时性。设计了一种自动曝光控制算法，采用二维搜索的方法判断环境的光照条件，结合KAI-2093的曝光灰度曲线得到曝光补偿曲线，为后续的图像渐晕处理做好了准备。3．提出基于理论、辐射定标和退化模型的三种反射镜拼接渐晕处理方法，详细阐述了三种方法的原理和实现过程。基于理论模型的方法以分析得到的光学模型为基础，采用按照像元将像面分割的方法得到补偿因子，由于该方法没有考虑电子学的影响，所以它的意义目前还只限于理论分析；基于辐射定标的方法使用的是实验室辐射定标的数据，所以由此得到的补偿因子具有很好的针对性；为了弥补辐射定标法灵活性方面的缺陷，根据实际成像图片的灰度分布提出了系统渐晕的退化模型，采用最速下降法求解退化模型的几个参数，最后使用退化模型对渐晕图像进行补偿。4．文中最后采用图像配准和拼接的方法对反射镜拼接系统的渐晕图片进行处理，目的是为了减少系统测量上的复杂度、降低系统对测试设备的依赖性。输出调整为单片独立输出方式，采用Harris算子提取图像特征点，使用归一化相关法进行粗匹配，采用双向匹配法进一步提纯，最后采用RANSAC算法剔除所有的误匹配特征点，求解出变换模型的参数，采用分段加权平均法和基于梯度法结合的融合方法对图像重叠区域进行处理，最终得到整幅亮度均匀的图像。更多还原

【Abstract】 With the need of high ground resolution and large ground cover in remotecarema system, monolithic image sensors are not large enough to achieve therequirement. Also because the manufacture technics of large area Charge CoupledDevice(CCD) are more complex and more cost, CCD mosaic technics have beenproposed to meet the requirement in the large field of view(FOV). In this paperseveral mosaic methods are analyzed and compared. To avoid their drawbacks themethod based on reflectors is proposed because of its advantages in higher energyefficiency, longer splice length, no-abberation, et al.In the paper, the reflector to be used is commonly and suits to extend, so it has abroad application prospect. However there is vignetting phenomenon in the mosaicsystem based on reflectors which will be discussed and disposed. The primarycoverage of this paper is as follows:First, from geometrical optics and physical optics, we analysed the generationmechanism of the vignetting phenomenon, and proposed the vignetting distributionmodel in the middle of the optical field separately. An equivalent optical model basedon muti-configuration technique is set up in ZEMAX for the analysis in the system’soptical characteristics. The system’s Point Spread Function(PSF) and ModulationTransfer Function(MTF) are calculated through dynamic data exchange and opticaltrail methods to analyze the causation of the vignetting phenomenon. Second, in order to prove the feasibility of the vignetting correction methods andunderstand the technology difficulties in the design, a prototype camera using thereflector mosaic structure was built which has two area CCD. We analyzed theinterline transfer area CCD KAI-2093’s working principle in detail to put forward anappropriate driving and high speed image collection scheme. we use cache in FPGAand RAM working in ping-pang mode in order to realize signal’s integrality and realtime working mode,. An auto-exposure algorithm was designed to compensate thecamera’s exposure parameters which takes advantage of the large brightness contrastamong different elements in the image, dynamically distinguishes the main object andthe background. All of the above are prepared for the vignetting corrections.Third，the methods based on the theory, the radiate calibration and the degeneratemodel are proposed in this paper, separately. The theory correction method is based onthe optical model gained in Chapter Three, which compensates the pixel’s grey valueaccording to the compensation factors achieved by dividing the image plane. But thismethod does not consider the impact of electronics, so it only has sense in thetheoretical analysis. The second method uses the data from the radiate calibration, sothe compensation factors are more accurate but not flexible. So in order to make upthe limitation, after the analysis of the actual image’s grey distribution, the degeneratemodel method is proposed to compensate the vignetting image, which uses thesteepest descent method to get the parameters.Fourth, in the end we use the image registration and mosaic algorithm to correctthe vignetting image. The output mode is turned to single, and the scheme uses theHarris operator to pick up the feature points, the NCC function to rough match, thetwo-way matching method for further purification, and finally the RANSACalgorithm to remove all the wrong matching points to obtain the transformation model.The segment weighted average method and the method based on gradient arecombined to process the overlap area to ultimately get the uniform brightness image.更多还原