遥感影像阴影检测与去除算法研究

发布时间：2019-02-25 07:35

【摘要】：对地观测技术的迅猛发展,迎来了卫星遥感前所未有的发展阶段。高分辨率遥感图像的出现使得人们探索和认识自然界步入了新的里程碑,高大建筑物、树木等遮挡太阳光线难以避免在遥感图像中形成大片阴影区。阴影的存在影响图像信息的判读与解译,给后续遥感图像处理带来诸多困难,如目标分类识别,图像配准等任务；为有效利用阴影数据,遥感影像阴影处理成为研究热点,而阴影检测与阴影去除是阴影处理中相互依存的两个方面。本文对现有阴影检测算法-基于黑体辐射模型与自适应特征选择的阴影检测算法进行了学习与仿真。分析研究发现,现有阴影检测算法对亮度均匀且亮度低的阴影区域有较好的检测性能。不过,由于地物复杂性,遥感图像中存在许多非匀质阴影和亮阴影,现有阴影检测算法对非匀质阴影和亮阴影存在漏检问题。为提高对非匀质阴影与亮阴影的检测效果,本文提出了一种结合局部分类水平集与颜色特征的遥感影像阴影检测方法,该方法首先结合阴影区域的亮度非均匀性,采用局部分类水平集分割遥感图像的阴影区域,然后通过分析绿地与阴影颜色特征分量的差别以去除候选阴影区中被误检的绿地。实验结果表明所提出的方法优于现有的黑体辐射模型与自适应特征选择的方法,有效克服了传统方法对非匀质阴影与亮阴影的漏检问题,且整个检测过程无需人工干预。在阴影检测算法能较好检测定位出阴影的基础上,论文进行阴影去除算法的研究。仿真分析了现有的基于颜色恒常、基于样本学习、自适应非局部正则的阴影去除算法。分析研究发现,现有阴影去除算法存在颜色失真、信息失真,或是需过多人工参与操作、阴影去除后纹理信息保持不够好等问题。为此,论文设计了一种基于Curvelet纹理方向非局部正则的阴影去除算法。该算法对自适应非局部正则的阴影去除算法从三个方面进行了改进。(1)采用论文提出的“结合局部分类水平集与颜色特征的遥感影像阴影检测方法”提取阴影区；(2)为消除半影的影响,对阴影边缘进行弱化；(3)采用Curvelet小波提取阴影区方向因子用于正则化处理,增强阴影去除区域的纹理细节信息。实验结果表明所设计的阴影去除算法提高了算法可操作性、对阴影去除区域的纹理细节信息的保持优于自适应非局部正则的阴影去除算法。最后,本文基于MATLAB平台设计了一款遥感影像阴影检测与去除算法可视化演示系统,并给出了该系统的使用说明和演示结果。
[Abstract]:The rapid development of Earth observation technology ushered in an unprecedented development stage of satellite remote sensing. The appearance of high-resolution remote sensing images makes people explore and understand the nature into a new milestone. It is difficult to avoid the formation of large shadow areas in remote sensing images by blocking solar rays such as tall buildings and trees. The existence of shadow affects the interpretation and interpretation of image information, and brings many difficulties to the subsequent remote sensing image processing, such as target classification and recognition, image registration and other tasks. In order to utilize shadow data effectively, shadow processing of remote sensing images has become a hot topic, and shadow detection and shadow removal are two interdependent aspects in shadow processing. In this paper, shadow detection algorithms based on blackbody radiation model and adaptive feature selection are studied and simulated. It is found that the existing shadow detection algorithms have better detection performance for shadow regions with uniform brightness and low brightness. However, due to the complexity of ground objects, there are many non-homogeneous shadows and bright shadows in remote sensing images, and the existing shadow detection algorithms have the problem of missing detection of non-homogeneous shadows and bright shadows. In order to improve the detection effect of non-homogeneous shadow and bright shadow, this paper presents a method of shadow detection in remote sensing image combining local classification level set and color feature. This method first combines the brightness inhomogeneity of shadow region. The shadow region of remote sensing image is segmented by using local classification level set, and then the difference between green space and shadow color characteristic component is analyzed to remove the false detected green space in candidate shadow area. The experimental results show that the proposed method is superior to the existing blackbody radiation model and adaptive feature selection method, and effectively overcomes the problem of missing detection of non-homogeneous and bright shadows by traditional methods, and the whole detection process does not require manual intervention. On the basis of shadow detection algorithm which can detect and locate shadow, shadow removal algorithm is studied in this paper. The existing shadow removal algorithms based on color constant, sample learning and adaptive nonlocal regularization are simulated and analyzed. It is found that the existing shadow removal algorithms have some problems such as color distortion, information distortion, or too much manual participation, and the texture information is not good enough after shadow removal. Therefore, a non-local regular shadow removal algorithm based on Curvelet texture direction is designed. The algorithm improves the adaptive non-local regularization shadow removal algorithm from three aspects. (1) the shadow region is extracted by the "shadow detection method combining local classification level set and color features" proposed in this paper; (2) in order to eliminate the influence of penumbra, the shadow edge is weakened; (3) the direction factor of shadow region is extracted by Curvelet wavelet to enhance the texture details of shadow removal region. The experimental results show that the proposed shadow removal algorithm improves the maneuverability of the algorithm and maintains the texture details of the shadow removal region better than the adaptive non-local regular shadow removal algorithm. Finally, based on the MATLAB platform, a visual demonstration system of shadow detection and removal algorithm for remote sensing images is designed, and the application and demonstration results of the system are given.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】