SAR图像稀疏表示模型的实证研究  

Empirical study on sparse representation model of SAR images

作　　者：黄柯蒙 姜娜娜 赵文博 郑妍昕 刘文平 朱炬波 HUANG Kemeng;JIANG Nana;ZHAO Wenbo;ZHENG Yanxin;LIU Wenping;ZHU Jubo(School of Artificial Intelligence,Sun Yat-sen University,Zhuhai 519082,China)

机构地区：[1]中山大学人工智能学院,广东珠海519082

出　　处：《中山大学学报（自然科学版）（中英文）》2024年第4期107-114,共8页Acta Scientiarum Naturalium Universitatis Sunyatseni

基　　金：国家自然科学基金(U21B2039)。

摘　　要：对合成孔径雷达(SAR)场景使用稀疏表示算法得到基函数字典示例,滤波前后图像稀疏表示的对比研究表明:相干斑噪声对SAR场景稀疏表示的字典结果有影响。选取浦江二号、ALOS2和SIR-C的特定SAR图像数据,通过设置单因素条件探讨优化算法、样本内容、数据集大小、雷达分辨率、极化方式、波段对字典结果的影响。结果表明:(1)SAR场景稀疏表示学习出的字典和雷达波段、分辨率、极化方式有关,和所选取的不同样本内容、数据集的大小以及所使用的优化算法无关。(2)C波段比L波段更能反应SAR场景的稀疏性。(3)降采样数据集更能反应SAR场景的稀疏性。(4)HH、VV极化图像学习出的字典更具有本质特征。An example of using the sparse representation algorithm to obtain a basis function dictionary for synthetic-aperture radar(SAR)scenes.The comparative of sparse representation of images before and after filtering shows that speckle noise has an impact on the dictionary results of sparse representation of SAR scenes.Select specific SAR image data from Pujiang No.2,ALOS2,and SIR-C,it is discussed that the effects of optimization algorithm,sample content,dataset size,radar resolution,polarization method,and band on dictionary results by setting single factor conditions.The results show that:(1)The dictionary learned from sparse representation of SAR scenes is related to radar band,resolution and polarization mode,and is independent of sample contents,datasets size,and optimization algorithms.(2)C-band can better reflect the sparsity of SAR scenes than L band.(3)The downsampling dataset can better reflect the sparsity of SAR scenes.(4)The dictionaries learned from HH and VV polarized images have more essential features.

关 键 词：SAR 稀疏表示 基函数字典 相干斑噪声 极化 

分 类 号：TN95[电子电信—信号与信息处理]