基于CNN的高光谱和多光谱图像融合方法研究  

Research on hyperspectral and multispectral image fusion methods based on improved convolutional neural networks

作　　者：齐济 杨海涛 孔卓 QI Ji;YANG Haitao;KONG Zhuo(Graduate School of Aerospace Engineering University,Huairou 101400,China;School of Aerospace Information,University of Aerospace Engineering,Huairou 101400,China)

机构地区：[1]航天工程大学研究生院,北京101400 [2]航天工程大学航天信息学院,北京101400

出　　处：《兵器装备工程学报》2022年第11期81-87,129,共8页Journal of Ordnance Equipment Engineering

摘　　要：为了解决高光谱图像空间分辨率较低的问题,提出一种基于卷积神经网络(CNN)的图像融合方法。首先,将输入的高光谱图像和多光谱图像输入至一个双分支网络进行特征提取,以获得高光谱图像中的高频光谱信息和多光谱图像中的高频空间特征;其次,通过特征融合网络将提取到的特征进行初步融合,形成紧凑的图像特征;再通过光谱重建网络,从融合后的特征中恢复重建高光谱图像,以初步提高其空间分辨率;最后,对上一步重建的高光谱图像进行空间边缘和光谱边缘信息的重构,进一步提升高光谱图像的空间分辨率。经验证,所提出的方法能将高光谱图像的更详细的细节信息保留,在主观视觉和客观评价指标PSNR、RMSE、ERGAS以及SAM上具有一定的优势。In order to solve the problem of low spatial resolution of hyperspectral images,an image fusion method based on convolutional Neural Network(CNN)was proposed.The input hyperspectral image and multispectral image were input to a dual-branch network for feature extraction to obtain high-frequency spectral information in hyperspectral images and high-frequency spatial features in multispectral images;The extracted features were initially fused through the feature fusion network,with the aim of forming compact image features;Through the spectral reconstruction network,the hyperspectral image was restored and reconstructed from the fused features to initially improve its spatial resolution;The spatial edge and spectral edge information of the hyperspectral image reconstructed in the previous step further improves the spatial resolution of the hyperspectral image.After experimental verification,the proposed method can retain the more detailed detailed information of hyperspectral images,which has certain advantages in subjective visual and objective evaluation indicators PSNR,RMSE,ERGAS and SAM.

关 键 词：图像融合 卷积神经网络 高光谱图像 多光谱图像 方法研究 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]