基于多特征融合的红外弱小目标检测  

Infrared small target detection based on multi-feature fusion

作　　者：武杰[1] 蔺素珍[1] 禄晓飞[2] 李大威[1] 张海松 Wu Jie;Lin Suzhen;Lu Xiaofei;Li Dawei;Zhang Haisong(College of Data Science and Technology,North University of China,Taiyuan 030051,China;Jiuquan Satellite Launch Center,Jiuquan 735000,China)

机构地区：[1]中北大学大数据学院,太原030051 [2]酒泉卫星发射中心,酒泉735000

出　　处：《电子测量技术》2022年第8期108-115,共8页Electronic Measurement Technology

基　　金：国家自然科学基金(61702465);国家自然科学基金(61774138);山西省自然科学基金(201901D111151);山西省研究生创新项目(2021Y622)资助。

摘　　要：为解决红外弱小目标检测领域中基于单类先验知识的人类视觉系统检测方法检测准确率低、虚警率高以及显著图计算复杂等问题,提出一种在复杂背景条件下对红外弱小目标多种特性进行融合处理的检测方法。通过融合红外弱小目标的局部灰度值大、自身灰度信息符合二维高斯分布以及与邻域相似度低的三大特性,利用协方差检测和相似度对比计算得到显著图,对显著图进行简单阈值分割得到真实目标。对不同复杂背景和不同数据类型的红外源图像进行弱小目标检测实验,结果表明:与基线算法相比本文所提算法检测结果背景抑制因子和信杂比增益均提高2~3倍,交并比为HVS类方法最优,ROC曲线在较低虚警率时获得最高检测准确率。本文方法将红外源图像中弱小目标多个特性进行有效融合,提高检测精度的同时降低了显著图计算复杂度,在不同复杂背景和杂波干扰的情况下仍能取得较好的目标定位和背景抑制效果。In order to solve the problems of low detection accuracy, high false alarm rate and complex calculation of saliency map based on single-class prior knowledge of human visual system detection method in the field of infrared small target detection, a detection method that fuses various characteristics of infrared small targets under complex background conditions is proposed. By fusing the three characteristics of infrared small targets that the local gray value is large, its own gray information conforms to the two-dimensional Gaussian distribution, and the similarity with the neighborhood is low, the saliency map is calculated by covariance detection and similarity comparison. And then threshold segmentation of the saliency map to get the real target. The small target detection experiments are carried out on infrared source images with different complex backgrounds and different data types. The results show that: compared with the baseline algorithm, the detection results of the proposed algorithm in this paper increase the background suppression factor and the signal-clutter ratio gain by 2~3 times, the intersection of union is the best in the HVS method, and the ROC curve obtains the highest detection accuracy at a lower false alarm rate. The method in this paper effectively fuses multiple characteristics of small targets in the infrared source image, improves the detection accuracy and reduces the complexity of the algorithm, and can still achieve good target positioning and background suppression in the case of different complex backgrounds and clutter interference.

关 键 词：弱小目标检测 红外图象 目标背景灰度对比度 协方差检测 目标背景相似性 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]