基于改进加权核范数的红外弱小目标检测  

Infrared dim target detection based on improved weighted kernel norm

作　　者：翟昊 罗晓琳 吴令夏 王荣昌 ZHAI Hao;LUO Xiao-Lin;WU Lin-Xia;WANG Rong-Chang(Army Academy of Artillery and Air Defense,Hefei 230031,China;Anhui Key Laboratory of Polarized Imaging Detection Technology,Hefei 230031,China)

机构地区：[1]中国人民解放军陆军炮兵防空兵学院,安徽合肥230031 [2]偏振光成像探测技术安徽省重点实验室,安徽合肥230031

出　　处：《激光与红外》2021年第6期776-781,共6页Laser & Infrared

摘　　要：针对传统基于鲁棒主成分分析(RPCA)的红外弱小目标检测算法对噪声不敏感,算法运行时间长,鲁棒性不强的问题,提出一种重加权红外小目标图像模型,并用非精确增广拉格朗日乘子法(AIALM)求解。该方法首先将原始红外图像转化为红外块图像模型,然后采用重加权核范数对背景块图像进行约束,较好地保留了背景边缘。针对单纯使用l1范数不能抑制某些噪声或杂波的问题,引入了加权l1范数,进一步增强了目标图像的稀疏性。最后,将红外块图像模型转化为重加权RPCA问题,并用AIALM求解。通过大量实验表明:该算法在抑制背景杂波以及目标检测性能方面要优于其他传统算法。Aiming at the problems of the traditional Robust Principal Component Analysis(RPCA)-based infrared small target detection algorithm that is insensitive to noise,the algorithm runs for a long time,and the robustness is not strong,a re-weighted infrared small target image model is proposed,and non-precision enhancement is used,which is Wide Lagrangian multiplier method(AIALM)solution.This method first converts the original infrared image into an infrared block image model,and then uses a re-weighted kernel norm to constrain the background block image,which better preserves the background edge.Aiming at the problem that using norm alone cannot suppress some noise or clutter,a weighted norm is introduced to further enhance the sparsity of the target image.Finally,the infrared block image model is transformed into a weighted RPCA problem and solved by AIALM.A large number of experiments show that this algorithm is better than other traditional algorithms in suppressing background clutter and target detection performance.

关 键 词：小目标检测 红外块图 低秩稀疏矩阵 鲁棒主成分分析 重加权 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]