基于注意力U-Net网络的红外夜视质量增强研究  

Study on quality enhancement for infrared night vision based on attention U-Net network

作　　者：祁宏宇[1] QI Hongyu(School of Electronics and Information,Jiangsu Vocational College of Business,Nantong 226011,China)

机构地区：[1]江苏商贸职业学院电子与信息学院,江苏南通226011

出　　处：《光学技术》2023年第5期623-630,共8页Optical Technique

基　　金：南通市科技局社会民生计划项目(MS12021060)。

摘　　要：针对红外夜视遥感系统成像质量差的问题,提出一种目标与背景分离的多模态图像融合方法来改善红外夜视的成像质量。一方面,采用注意力U-Net对红外与可见光图像的目标区域进行分割与融合处理,利用U-Net强大的学习能力充分保留原图像中的目标信息;另一方面,通过引导滤波器对红外与可见光图像的背景区域进行分解,采取不同的融合策略处理基层信息与细节层信息,增强背景中的显著区域。在TNO数据集上的对比实验结果表明,该方法在主观视觉评价与客观量化评价两方面均优于其它对比方法。Aiming at the problem of low imaging quality of infrared night vision remote sensing system,a multi-modal image fusion method with separation of the target and background,to improve the imaging quality of infrared night vision.On the one hand,the attention U-Net is adopted to segment and fuse the target region of the infrared image and the visible light image,the powerful leaning ability of U-Net is taken advantage to preserve the target information of the source images;on the other hand,the background regions of the infrared image and the visible light image are decomposed through the guided filter,different fusion strategies are used to fuse base layer information and detail layer information,so that the salient regions of the background are enhanced.Compared experiments are carried on the TNO dataset,the results show that the proposed method outperforms the other compared methods on both subjective visual evaluation and objective quantitative evaluation.

关 键 词：红外夜视系统 图像融合 图像质量增强 深度神经网络 引导滤波器 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]