基于并行网络的低照度可见光和红外图像融合方法  

The fusion method of low-light visible light and infrared images based on parallel networks

作　　者：周晔 杜晓雨 谭亚军 张静[1] ZHOU Ye;DU Xiaoyu;TAN Yajun;ZHANG Jing(North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学,太原030051

出　　处：《激光杂志》2025年第3期133-141,共9页Laser Journal

基　　金：国防基金项目(No.2021-JCJQ-JJ-0726)。

摘　　要：针对现有的融合算法在低光环境下细节损失严重、亮度和对比度较低的问题,提出了基于并行网络的低照度可见光和红外图像融合方法(PNLLFusion),最大程度保留了源图像细节信息并提高了亮度与对比度。该方法实现了融合与亮度提升的并行进行,减少了由于增强算法和融合算法不兼容带来的信息损失,此外在Squeeze-and-excitation(SE)模块上添加残差结构和在自注意力网络中增加梯度计算以保留更多的纹理和边缘信息。在LLVIP数据集和TNO数据集上验证了方法的有效性,结果证明:与经典融合算法相比,该方法在低光照环境下能保留更多的源图像细节信息,并且能提高图像的对比度和亮度,在主观和客观评价上均能取得较好或相近的结果。To address the loss of details,low brightness,and contrast in existing fusion algorithms,the article proposes the Fusion of Low-Light visible light and infrared images based on Parallel Networks(PNLLFusion).PNLLFusion aims to maximize the preservation of detail information from the source images and enhance brightness and contrast.This method implements parallel fusion and brightness enhancement,reducing information loss caused by incompatibility between enhancement and fusion algorithms.Additionally,residual structures are added on the Squeeze-and-Excitation(SE) module,and gradient computation is incorporated into the self-attention network to preserve more texture and edge information.The effectiveness of the method is validated on the LLVIP dataset and TNO dataset.Experimental results demonstrate that compared to classical fusion algorithms,this method can preserve more detail information from the source images in low-light environments,while also improving image contrast and brightness.It achieves good or comparable results in both subjective and objective evaluations.

关 键 词：图像融合 RETINEX理论 图像增强 Squeeze-and-excitation 

分 类 号：TN911.73[电子电信—通信与信息系统]