摘要
利用Landsat8遥感卫星影像数据制作影像融合数据集,提出了一种双通道融合网络,并利用结果影像的质量指数对网络融合性能进行评估,分析与双三次卷积插值和GS影像融合方法的差异。结果表明,该网络加强了对高频空间信息的提取,在更高效提取空间特征的同时,减弱了融合过程中对多光谱影像光谱特征的影响,从而提高了融合影像的综合影像质量(QNR=0.8852)。
We used Landsat 8 remote sensing satellite image data to make the image fusion dataset,put forward a dual-channel fusion network,evaluated the performance of this network by the quality index of remote sensing images,and analyzed the differences between the network and bicubic interpolation and GS image fusion methods.The results show that the network proposed in this paper can strengthen the extraction of high-frequency spatial information.While extracting spatial features more efficiently,it also weakens the influence on the spectral features of multi-spectral images during the fusion process,thus improving the comprehensive image quality of fusion images(QNR=0.8852).
作者
靳道明
李路沙
JIN Daoming;LI Lusha(China Railway Design Corporation,Tianjin 300142,China)
出处
《地理空间信息》
2023年第11期1-4,共4页
Geospatial Information
关键词
深度学习
遥感影像融合
双通道卷积神经网络
多尺度特征
deep learning
remote sensing image fusion
dual-channel convolutional neural network
multi-scale feature