基于聚类思想的红外弱小目标检测  

Infrared small target detection based on clustering idea

作　　者：饶俊民 穆靖 刘士建[1,3] 公劲夫 李范鸣[1,3] RAO Jun-Min;MU Jing;LIU Shi-Jian;GONG Jin-Fu;LI Fan-Ming(Key Laboratory of Infrared System Detection and Imaging Technology,Chinese Academy of Sciences,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China;Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China)

机构地区：[1]中国科学院红外探测与成像技术重点实验室,上海200083 [2]中国科学院大学,北京100049 [3]中国科学院上海技术物理研究所,上海200083

出　　处：《红外与毫米波学报》2023年第4期527-537,共11页Journal of Infrared and Millimeter Waves

基　　金：国家十四五预研课题(项目编号514010405-207)。

摘　　要：针对复杂背景下尺寸未知的红外弱小目标检测难题,一种基于聚类思想的红外弱小目标检测方法被提出。首先,利用小目标形态学特征对原始红外图像进行预处理,生成新的密度特征图。其次,使用改进的密度峰聚类算法对潜在候选目标进行粗定位。然后,针对潜在目标的局部候选集,采用加权模糊集聚类算法对局部候选集进行目标与背景区域的精细分割,利用目标与背景之间的差异性在增强目标的同时抑制虚警。最后,对处理后的局部候选集进行自适应阈值提取真实目标。实验结果表明,与7种对比算法相比,该算法对尺寸未知的小目标具有良好的鲁棒性和检测性能。In order to solve the problem of detecting infrared small targets of unknown size in complex background,an infrared small target detection algorithm based on the clustering idea is proposed.First,the original infrared image is preprocessed by using small target morphological features to generate a new density feature map.Secondly,the potential candidate targets are coarsely localized with an improved density-peak clustering algorithm.Then,the local candidate sets of potential targets are constructed.A weighted fuzzy set clustering algorithm is used to finely segment the target and background regions of the image block,and then the difference between the target and background is adopted to suppress false alarms while enhancing the target.Finally,an adaptive threshold is applied to the processed local candidate set to extract the real target.Experimental results show that the proposed algorithm has good robustness and detection performance for small targets of unknown size in comparison with the other seven methods.

关 键 词：红外弱小目标检测 聚类 密度峰聚类 模糊集 分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]