【作者】 杨利红；

【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】 随着对宽覆盖地面卫星图像需求的不断增加，大视场航天相机成为航天技术领域研究的重点和热点。大视场航天相机的研制要耗费巨大的人力物力资源，其主要目的是获取高质量宽覆盖的遥感图像，因此图像质量是大视场航天相机成败的关键。大视场航天相机通过多片CCD拼接和多通道成像的设计来获取宽覆盖的地面图像，每一片CCD作为一个成像通道单独生成一幅数字图像。由于各片CCD之间存在拼接重叠像元，因此地面获取的各通道图像之间不连续，相邻通道图像之间有一定的重叠区域。在大视场航天相机的成像过程中，由于CCD传感器性能、卫星平台振动、成像系统离焦、大气湍流作用等因素的影响，会造成遥感图像退化，在现有的技术条件下，无法在相机上附加完善的星上图像处理系统，因此地面获取的各通道图像均存在退化变模糊现象。为了获得覆盖整个大视场的连续无缝宽覆盖图像，并尽可能提高宽覆盖图像的质量，必须对各通道图像进行质量退化复原和连续无缝拼接复原。结合大视场航天相机各通道图像不连续且存在质量退化的特点，本文从两个方面进行图像复原研究，首先针对各通道图像的退化问题进行图像质量退化复原，以得到细节更清晰的图像，使感兴趣的目标更易识别；其次对提升清晰度后的各通道图像进行连续无缝的自动拼接以得到最终的高质量宽覆盖图像。本文所处理的遥感图像数据量巨大，在保证图像质量提升效果的前提下，时间开销是设计图像复原算法过程中需考虑的一个关键问题。针对各通道图像退化问题的质量退化复原是本文的研究重点。详细分析了大视场航天相机遥感图像质量退化的原因，建立了图像退化和复原的模型，将其复原过程分为去噪处理和反卷积处理两个步骤实现。去噪处理过程中，提出结合奇异点检测的改进陷波滤波器法和自适应小波软阈值去噪法实现了遥感图像中条带噪声的消除，在去除噪声的同时尽可能地保留了图像的边缘特征信息。反卷积处理过程的关键是在轨点扩散函数(Point Spread Function, PSF)的估计。本文采用刀刃法利用遥感图像中具有刀刃特征的子图像来估计成像系统的在轨PSF，提出对边缘扩散函数(Edge Spread Function, ESF)进行Fermi拟合来保证PSF估计的精度。得到成像系统的在轨PSF后，以它为参数，采用性能良好且时间开销小的维纳滤波算法进行反卷积处理，提出基于边缘检测的最优窗维纳滤波算法克服了直接维纳滤波导致的寄生波纹和振铃波纹现象，质量退化的各通道图像经该算法处理后，清晰度明显提升，视觉效果良好。实验表明，复杂地形遥感图像被复原后，其清晰度评价指标灰度平均梯度(Gray Mean Grads,GMG)从4.755提高到11.333，拉普拉斯和(Laplacian Sum, LS)从18.676提高到58.493，图像质量得到大幅提升。经过去噪处理和反卷积处理后，各通道图像的质量退化现象被复原，图像清晰度提高，特征更易提取和识别，提高了各通道图像连续无缝自动拼接的成功率和效率。本文将各通道图像连续无缝的自动拼接过程分为图像配准和图像融合两个步骤实现。提出基于轮廓的最优步进模板匹配算法实现配准，该算法执行时间约为传统模板匹配算法的0.85%和SSDA算法的1.3%，是一种适合大视场航天相机遥感图像的快速配准算法。融合算法采用渐入渐出法，该方法实现简单，计算速度快，消除拼接接缝的效果好，适合数据量巨大的遥感图像融合处理。更多还原

【Abstract】 With the increasing demand for ground images with wide coverage, the spacecamera with wide field of view becomes the focus and the hot spot in the field ofspace technology. Its development costs enormous human and material resources. Itsmain products are high-quality remote sensing images with wide coverage, so theimage quality is the key to the success of the space camera with wide field of view.The space camera with wide field of view obtains ground images by multi-chipCCD mosaic and multiple imaging channels. Because of the overlapped pixelsbetween the stitching CCDs, the multi-channel images are discontinuous. There isan overlapped region between the images of adjacent channels. In the imagingprocess, many factors will result in image degradation, such as the CCDperformance, the vibration of the satellite, the defocus of the imaging system andatmospheric turbulence. Because the space camera can’t attach a perfect imageprocessing system in the existing condition, the acquired multi-channel images areblurry due to degradation. In order to obtain a continuous and seamless widecoverage image which covers the entire field of view and maximize the quality ofthe wide coverage image, the image degradation of each channel needs to berecovered. At the same time, the multi-channel images need to be stitched toacquire a continuous and seamless wide coverage image.Due to the discontinuity of multi-channel images and image degradation of each channel, this paper deals with image restoration of the space camera with wide fieldof view in two aspects. First of all, the image degradation of each channel needs tobe recovered to get clearer images, making the interested targets easier to identify.Secondly, the multi-channel images whose clarity is enhanced need to be stitchedautomatically to get a high quality continuous and seamless wide coverage image.The data amount of the remote sensing images dealt in this paper is huge. Therefore,the time overhead of the restoration algorithms is a key issue to consider in thecondition that the enhancement of image quality is ensured.Restoration on image degradation is the focus of this paper. The degradationreasons of the remote sensing images are analyzed detailedly. Image degradation andrestoration models are built. The recovered process is divided into denoising anddeconvolution. In the denoising step, the improved notch filtering method withsingular point detection and wavelet method with adaptive soft threshold areproposed. The two methods not only remove almost all the strip noise in the remotesensing images but also retain the edge information well. The core of thedeconvolution step is estimation of the on-orbit point spread function. Theknife-edge method is adopted to estimate the on-orbit point spread function of theimaging system. In order to ensure the accuracy of the point spread functionestimation, the Fermi fitting is done to the edge spread function. Due to the wellperformance and small time overhead, the Wiener filtering algorithm is used torestore the degraded images with the estimated point spread function as a parameter.In order to overcome the parasitic ripple and ringing phenomenon caused by thetraditional Wiener filtering, the optimal window and edge detection are combinedwith the Wiener filtering to restore images. The clarity of the degraded imagesprocessed by this method is significantly improved. After the remote sensing imagewith complex terrain is recovered, its gray mean grads (GMG) increases form4.775to11.333, and its Laplacian Sum (LS) increases form18.676to58.493.After image degradation of each channel is recovered, the clarity is improvedand the features are easier to extract. All this are benefit to the continuous and seamless automatic mosaic of the multi-channel images. The automatic mosaic isdivided into image registration and image fusion. The template matching algorithmwith optimal steps is proposed as image registration algorithm. The execution timeof this algorithm is approximately0.85%of the traditional template matchingalgorithm and1.3%of SSDA algorithm. It is a fast image registration algorithmwhich is suitable for the space camera with wide field of view. The fade-in andfade-out method is used to complete image fusion. It is a simple and fast algorithmwith wonderful effect in eliminating the stitching seams. It is suitable for the remotesensing images with a huge amount of data.更多还原