基于双树复小波的单幅遥感图像自适应去除云雾原理与方法的研究

Cloud and fog are the important factors disturbing remote sensing imaging for land which influence the real-time image interpretation and decisions. Removing cloud and fog from a single remote sensing image is benefit to improve remote sensing data validity, and to save data collecting time and cost. Firstly, based on Streamer radiation and transfer mode, radiation characteristics of cloud and fog in the case of various situations and imaging conditions are studied. Simulation methods of remote sensing imaging of cloud and fog by deducing their distribution data in two dimensions and applying optical imaging theory are discussed. On the basis of atmospheric radiation theory, methods of obtaining simulated remote sensing images are studied when the ground scene, cloud and fog are imaged together. Secondly, characteristics of coefficients in sub-belts obtained by decomposing the data of scene, cloud and fog in the simulated remote sensing images by dual-tree complex wavelet respectively are analyzed. The threshold algorithm operated in dual-tree complex wavelet levels is investigated. According to the generalized Gauss distribution model and by choosing KLD measurement distance as similarity criterion, the method of determining parameters of optimum filters using the design method of q-shift dual-tree complex wavelet filters is researched in the case of various features of cloud and fog and imaging conditions. In addition, algorithm to removing cloud and fog from remote sensing images which is capable of selecting parameters of optimum filters adaptively is investigated using the maximum entropy as the criterion. Finally, an experiment system which has the ability to process and remove cloud and fog from remote sensing images real-timely is developed by designing software and hardware. Real-time processing experiments in the platform of PC and the experiment system are performed. Research results are examined and improved.

云雾是对地遥感成像中重要的干扰因素，影响图像的实时判读和决策。单幅遥感图像去云雾有助于提高数据有效性，节约采集时间和成本。首先基于Streamer辐射传输模式，研究云雾不同状态和成像条件下的辐射特性，探讨通过求解云雾二维分布数据和运用光学成像理论模拟云雾遥感成像的方法。基于大气辐射理论，研究地面景物和云雾叠加成像获取模拟遥感图像的方法。然后，分析模拟遥感图像中云雾数据和景物数据分别经双树复小波变换各子带系数的特性，研究双树复小波变换分层阈值算法。根据广义高斯分布统计模型和以KLD测度距离为相似度准则，研究不同云雾特征和成像条件下，以q-移双树滤波器设计方法确定最优变换滤波器组参数的方法。在此基础上，研究以极大值熵为准则具有自适应选择优化滤波器参数性能的去除遥感图像云雾的方法。最后，通过软、硬件设计，开发遥感图像去云雾实时处理实验系统。开展PC机和航拍实时处理实验，检验和改进研究成果。

云雾是对地遥感成像中重要的干扰因素，影响图像的实时判读和决策。单幅遥感图像去云雾有助于提高数据有效性，节约采集时间和成本。首先基于大气辐射传输模式，研究了云雾不同状态和成像条件下的辐射特性，建立了受云雾干扰的遥感成像退化模型，提出了地面景物和云雾叠加成像获取模拟遥感图像的方法，研究表明大气透过率函数决定了云雾的浓度、形状和分布。然后，分析了遥感图像中云雾数据和景物数据分别经双树复小波变换后各子带系数的频率特性，确定了以双树复小波变换分层法为基础的去云雾处理思路。研究q-移双树滤波器设计方法和由滤波器获得双树复小波基函数的生成步骤，探讨了滤波器参数的确定方法。接着，研究了单幅图像质量评价方法。分析了图像的视觉显著性和感知锐度特征，提出了基于视觉显著性的局部感知锐度的图像质量评价算法。分析了图像在不同层次模糊中局部锐度特征差异，结合双树复小波变换的多分辨率特性和方向选择性，提出了基于局部锐度特征和双树复小波相结合的无参图像质量评估算法，为提高算法自适应性奠定基础。结合大气辐射传输模式、暗通道先验、导向滤波，提出了基于双树复小波变换的单幅遥感图像自适应去除云雾方法，给出详细的实现步骤。最后，通过软、硬件设计，开发遥感图像去云雾实时处理实验系统，开展图像处理实验。研究表明，本项目提出的去除云雾方法处理效果好，自适应强。