红外与可见光图像融合算法研究  

Research on Infrared and Visible Image Fusion Algorithm

作　　者：李秒 郝元宏 许树园 LI Miao;HAO Yuanhong;XU Shuyuan(North Automatic Control Technology Institute,Taiyuan 030006,China;China North Industries Corporation,Beijing 100053,China)

机构地区：[1]北方自动控制技术研究所,太原030006 [2]中国北方工业有限公司,北京100053

出　　处：《火力与指挥控制》2025年第3期165-177,共13页Fire Control & Command Control

基　　金：国家级预研课题基金资助项目(XXX构建技术)。

摘　　要：红外和可见光图像具有互补性,通过图像融合技术可以获得更加丰富的信息,提高图像质量,满足多元的应用需求。针对红外和可见光图像融合方法,分别从像素级、特征级以及决策级3个层面对现有方法进行了归纳分析。在各融合层面,将图像融合技术归纳为传统方法和深度学习方法两大类,着重梳理了近两年出现的轻量化深度学习方法,重点关注了近年涌现的Transformer方法,并给出了各类方法的优缺点。针对IVIF实际应用问题,提出了一种工程化框架,为IVIF技术落地提供参考思路。Infrared and visible images are complementary,and more abundant information can be obtained through image fusion technology,which can improve image quality and meet multiple application requirements.Aiming at the infrared and visible image fusion(IVIF)method,the existing methods from three levels:pixel level,feature level and decision-making level are summarized and analyzed.At each fusion level,image fusion technology is summarized into two categories:traditional methods and deep learning methods.The lightweight deep learning methods that have appeared in the past two years are specifically sorted out,and the Transformer method that has emerged in recent years is focused on,and the advantages and the disadvantages of various methods are presented.Finally,aiming at the practical application problems of IVIF,an engineering framework is proposed to provide reference ideas for the implementation of IVIF technology.

关 键 词：图像融合 红外图像 可见光图像 深度学习 无监督学习 轻量化模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]