基于多分支CNN的高光谱与全色影像融合处理

Pansharpening Based on Multi-Branch CNN

高光谱与全色影像融合旨在通过融合高空间分辨率的全色影像与低空间分辨率的高光谱影像来获得高空间分辨率的高光谱影像。基于深度卷积神经网络(CNN),提出了一种遥感影像融合方法,利用两个独立的分支网络逐级从高光谱和全色影像中提取光谱和空间特征。该融合网络由两个分支网络和一个主线网络组成,利用两个分支网络分别从高光谱与全色影像中提取空谱特征,主线网络基于分支网络提取的特征,重建得到最终融合的高空间分辨率的高光谱影像。在CAVE和Pavia Center数据集上分别进行了实验验证,通过对比可以发现,所提出的融合算法在空间细节和光谱保真度上较当前主流算法均表现出更优异的性能。

Pansharpening aims to obtain hyperspectral images with high spatial resolutions by fusing hyperspectral images with low spatial resolutions and panchromatic images with high spatial resolutions together.This paper introduces a remote sensing image fusion method based on a deep convolutional neural network(CNN),which extracts spectral and spatial features step by step from hyperspectral and panchromatic images using two independent branch networks.The proposed fusion network is composed of two branches and a main network.The two independent branch networks are used for extracting the spatial-spectral features from hyperspectral and panchromatic images,while based on the features extracted from the branch network,the main network is used to reconstruct and the final fused hyperspectral images with high spatial resolutions are obtained.The experimental verifications were conducted on both CAVE and Pavia Center datasets.Through comparison,one can see that the proposed fusion algorithm outperforms the prevailing algorithms in terms of spatial detail and spectral fidelity.