一种针对图像融合的Dense-FCNN算法研究  

Research on Dense-FCNN Algorithm for Image Fusion

作　　者：郑迦馨 罗银辉[1] 吴岳洲 王宇[2] ZHENG Jiaxin;LUO Yinhui;WU Yuezhou;WANG Yu(School of Computer Science,Civil Aviation Flight University of China,Guanghan 618307;School of Computer Science,Nanjing Normal University,Nanjing 210023)

机构地区：[1]中国民用航空飞行学院计算机学院,广汉618307 [2]南京师范大学计算机学院,南京210023

出　　处：《计算机与数字工程》2025年第1期209-213,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目“大数据驱动的飞行训练智能评估理论与方法”(编号:U2033213);国家重点研发计划项目(编号:2021YFF0603904);中央高校基本科研业务费基金项目(编号:ZJ2022-004)资助。

摘　　要：高空间和高光谱分辦率的图像能够给计算机视觉处理带来更全面的信息,特别是在遥感图像融合方面的研究一直备受关注。近年来,基于卷积网络的遥感图像融合方法在遥感图像融合领域带来了突破性的进展。针对遥感图像融合质量问题,提出了一种Dense-FCNN的网络,该网络融合了DenseNet网络和FusionCNN的特点,基于充分利用特征的思想,将DenseBlock模块加入到FusionCNN的不同结构中,并做出对比分析。结果表明,改进后的Dense-FCNN网络在融合质量上均有所提升,最好的模型效果提升了3.4%,并且,融合图像能更好地保留Pan图像的空间信息和MS图像的光谱信息,对于获取更全面的遥感图像信息具有实际应用价值。High spatial and hyperspectral images can bring more comprehensive information to computer vision processing,es⁃pecially in remote sensing image fusion.In recent years,the remote sensing image fusion method based on convolution network has brought a breakthrough in the field of remote sensing image fusion.Aiming at the quality problem of remote sensing image fusion,a kind of Dense-FCNN network is proposed.The network combines the characteristics of DenseNet network and FusionCNN.Based on the idea of making full use of the characteristics,the DenseBlock module is added to the different structures of FusionCNN,and a comparative analysis is made.The results show that the improved dense FCNN network has improved the fusion quality,and the best model effect has increased by 3.4%.Moreover,the fusion image can better retain the spatial information of Pan image and the spectral information of MS image,which has practical application value for obtaining more comprehensive remote sensing image in⁃formation.

关 键 词：融合质量 卷积网络 图像融合 空间信息 光谱信息 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]