基于结构低秩编码的复杂环境红外弱小目标检测算法  

Infrared dim small target detection algorithm based on structural low-rank coding under complex environment

作　　者：赵爱罡[1] 王宏力[1] 杨小冈[1] 陆敬辉[1] 黄鹏杰 

机构地区：[1]第二炮兵工程大学,西安710025

出　　处：《中国惯性技术学报》2015年第5期662-669,共8页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金(61203189;61374054)

摘　　要：针对复杂环境红外弱小目标检测难的问题,依据背景慢变特性,提出了一种将背景优化和低秩表达相结合的结构低秩编码小目标检测算法。首先,利用梯度0l范数约束提取背景中梯度较大的成分,保留灰度快变结构,同时平滑慢变结构,对背景进行优化;其次,使用核函数刻画背景图像块之间的低秩特性,用秩描述背景的主要结构并进行建模;最后,分解得到的误差矩阵具有稀疏性,主要包含快变的小目标结构,通过稀疏矩阵1,2l范数定位红外弱小目标。实验结果表明,结构低秩编码检测算法能够有效发掘复杂背景图像块之间的关系,抑制杂波干扰,在虚警为2时,最低检测率为92%。提高了复杂环境下红外弱小目标的检测性能,基本能满足实际应用要求。Aiming at the problem of dim small target detection under complex environment,a small-target detection algorithm with structural low-rank coding（SLRC） is put forward based on background＇s slow varying,which combines background optimization with low-rank representation. Firstly,the background components with larger gradient are extracted using 0l norm restrict of gradient. The grayscale rapid-varying structure is retained,and the slow-varying structure is smoothed. The background is optimized by this way. Secondly,the low-rank between pieces of background is modeled by the nuclear norm. And the model is built based on the background＇s main structure,which is described by rank. At last,the error matrix by decomposition is sparse,which contains small-target rapid-varying structure. The infrared dim small target is located by 1,2l norm of error matrix. Experiment results show that the SLRC detection algorithm can effectively explore the relationships between complex backgrounds and depress the jam of clutter. The minimum detection rate can be up to 92% when false-alarm is 2. These improve the detection performance of infrared dim small target under complex environment,basically satisfying the actual application requirements.

关 键 词：复杂环境 红外小目标 低秩表达 0l范数约束 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]