基于形态学和改进管道滤波的红外小目标检测  

Infrared Small Target Detection Based on Morphology and Improved Pipeline Filter

作　　者：赵明晶 李伟 陶然[1] 马鹏阁[3] 揭斐然[4] ZHAO Mingjing;LI Wei;TAO Ran;MA Pengge;JIE Feiran(School of Information and Electronics,Beijing Institute of Technology,Beijing 100081,China;Science and Technology on Electro-optic Control Laboratory,Beijing 100081,China;School of Intelligent Engineering Zhengzhou,University of Aeronautics,Zhengzhou,Henan 450015,China;Luoyang Electro-optical Equipment Research Institute of Aviation Industry Corporation of China,Luoyang,Henan 100007,China)

机构地区：[1]北京理工大学信息与电子学院,北京100081 [2]光电控制技术重点实验室,北京100081 [3]郑州航空工业管理学院智能工程学院,河南郑州450015 [4]中国航空工业集团公司洛阳电光设备研究所,河南洛阳100007

出　　处：《北京理工大学学报》2023年第7期755-763,共9页Transactions of Beijing Institute of Technology

基　　金：航空科学基金资助项目(20200051072001)。

摘　　要：红外小目标的正确检测是图像处理领域中的一个重要课题,现有方法没有充分挖掘小目标的特征导致目标容易淹没在复杂的背景中.此外,目标的运动轨迹信息没有完全利用也使得目标和孤立噪声不易区分.为了解决该问题,设计了一种基于多属性形态学和改进管道滤波相结合的算法.对红外图像构建最大树,并通过面积、高度和对角线属性综合提取小目标的多种特征达到目标增强背景抑制的目的,融合多属性形态学得到的结果确定候选目标,通过改进的管道滤波利用目标运动轨迹的规律性进一步去除类目标噪声.将所提出的方法在4个数据集与4种方法进行比较,实验结果证明所提出的方法能在最大程度上抑制背景杂波的同时增强目标,对多种复杂场景下的不同类型、尺寸不一和亮度差异大的目标检测效果鲁棒.Detecting small target correctly in infrared images is an important topic in the field of image processing.Existing methods do not fully explore the characteristics of small targets,therefore,targets are easily submerged in the complex background.Furthermore,target trajectory information can not be fully utilized,making it difficult to distinguish the target from the isolated noise.To solve above problems,a novel infrared small target detection method was proposed based on multi-attribute morphology and improved pipeline filter.Firstly,to achieve target enhancement and background suppression,a max-tree was constructed for infrared image,and some features,such as area,height and diagonal attributes,were extracted synthetically from small target.Then,fusing the results of multi-attribute morphology,the candidate targets were determined.And utilizing an improved pipeline filter,considering the regularity of target motion,the target-like noise was removed from the candidate targets.Finally,some experiments were carried out to compare the proposed method with other four methods on four image data sets.The experiment results show that the proposed method can suppress the background clutter to the maximum extent and enhance the target simultaneously.Furthermore,the proposed method can keep its robust to the targets with different types,diverse sizes,and large brightness differences in a variety of complex scenes.

关 键 词：图像处理 红外图像 小目标检测 数学形态学 改进管道滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]