基于多层卷积的红外与可见光图像融合算法  

Infrared and Visible Image Fusion Based on Multi-layer Convolution

作　　者：陈海秀[1,2] 房威志 陆康 陆成 黄仔洁 陈子昂 CHEN Haixiu;FANG Weizhi;LU Kang;LU Cheng;HUANG Zijie;CHEN Ziang(School of Automation,Nanjing University of Information Science&Technology,Nanjing 210000,China;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing University of Information Science&Technology,Nanjing 210000,China)

机构地区：[1]南京信息工程大学自动化学院,南京210000 [2]南京信息工程大学江苏省大气环境与装备技术协同创新中心,南京210000

出　　处：《电光与控制》2024年第9期12-17,44,共7页Electronics Optics & Control

基　　金：国家自然科学基金(61302189);江苏省研究生科研与实践创新计划项目(SJCX23_0383)。

摘　　要：针对复杂背景下纹理细节信息丢失、融合图像视觉感受较差等问题,提出了一种基于多层卷积的红外与可见光图像融合算法。该算法的网络框架分为编码器、解码器和融合网络3个部分。在编码器中引入高效通道注意力机制对源图像进行编码处理,融合多层卷积块、梯度卷积块、下采样卷积块以及卷积空间通道注意力机制等形成多层卷积融合网络(MCFN),通过该融合网络进行特征融合,利用解码器重建输出融合图像。选取了5种现有算法与所提算法用8种客观评价指标在两种数据集上进行比较,结果表明,所提算法融合后的图像目标突出、细节清晰、轮廓明显、指标提升显著,符合人体视觉感受。To solve the problems of texture detail loss and poor visual perception of fused images in complex background,an infrared/visible image fusion algorithm based on multi-layer convolution is proposed.The network framework of the algorithm is divided into encoder,decoder and fusion network.An Efficient Channel Attention(ECA)mechanism is introduced into the encoder to encode the source image.Multi-layer Convolutional Fusion Network(MCFN)is formed by fusion of multi-layer convolution blocks,gradient convolution blocks,down sampled convolution block and the attention mechanism of convolution space channels.Feature fusion is performed through MCFN.Then the output fusion image is reconstructed through the decoder.Eight objective evaluation indicators are selected for comparison with five existing algorithms on two datasets.The results show that the fusion image of the proposed algorithm has prominent object,clear details and obvious contour,and the index is improved significantly,which accords with human visual perception.

关 键 词：图像融合 红外图像 可见光图像 多层卷积 融合网络 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]