基于聚类字典学习和稀疏表示的SAR图像抑斑方法

发布时间：2018-06-10 21:12

本文选题：合成孔径雷达 + 相干斑抑制　；参考：《系统工程与电子技术》2017年08期

【摘要】：针对合成孔径雷达(synthetic aperture radar,SAR)图像相干斑抑制问题,提出一种基于聚类字典学习和稀疏表示的SAR图像抑斑方法。本方法以相干斑噪声的非对数加性模型为基础,通过改进相似度测度的K-means聚类和主成分分析方法进行字典学习,克服了相干斑噪声非高斯性带来的影响,形成具有结构性聚类的字典原子;在稀疏分解方面,通过引入方差稳定因子,建立了适用于抑制SAR相干斑噪声的稀疏表示模型,并通过交替迭代算法进行代价方程求解;同时算法还增加了点目标保护措施,避免了对图像点目标"过滤波"。通过卫星、无人机SAR图像的抑斑实验证明,相比经典的SAR图像抑斑方法,所提的方法在抑斑的视觉效果上和客观评价指标上都有较大的提升。
[Abstract]:Aiming at the problem of speckle suppression in synthetic aperture radar (SAR) images, a speckle suppression method based on clustering dictionary learning and sparse representation is proposed. This method is based on the non-logarithmic additive model of speckle noise. By using K-means clustering and principal component analysis (PCA), the influence of speckle noise non-Gao Si is overcome. In the aspect of sparse decomposition, a sparse representation model suitable for suppressing SAR speckle noise is established by introducing variance stability factor, and the cost equation is solved by alternating iterative algorithm. At the same time, the algorithm also adds the point target protection measures to avoid the image point target "filter wave". The experiments on SAR images of satellite and UAV show that compared with the classical SAR image suppression method, the proposed method can improve the visual effect of speckle suppression and the objective evaluation index.
【作者单位】：北京航空航天大学无人系统研究院;北京航空航天大学虚拟现实技术与系统国家重点实验室;
【基金】：国家自然科学基金创新研究群体科学基金(61521091);国家自然科学基金(61272348,61572054)资助课题
【分类号】：TN957.52

【参考文献】