用于遥感图像超分辨率重建的残差对偶回归网络

Residual dual regression network for super-resolution reconstruction of remote sensing images

使用人工模拟的高-低分辨率图像对易导致在对真实遥感图像超分辨率重建时模型泛化能力差,针对此问题,结合残差通道注意力网络(residual channel attention network,RCAN)的二次残差(residual in residual,RIR)模块,改进对偶回归网络(dual regression networks,DRN),提出了残差对偶回归网络(residual dual regression network,RDRN)。选取LandCover.ai和DIOR航空图像数据集的10000张512像素×512像素图像构成样本数据集,用于训练和测试网络,并将重建结果与现有其他超分辨率网络模型的重建结果对比评价。实验结果表明,RDRN在重建质量和模型参数量方面均表现优异,能够在较低模型复杂度的情况下实现较好的超分重建效果,且对不同低分辨率遥感图像具有较好的泛化能力。

In order to solve the problem of poor model generalizing ability in real super-resolution reconstruction of remote sensing images,which is easily caused by the use of artificial high-low resolution image pairs,combined with the residual in residual(RIR)module of residual channel attention network(RCAN),dual regression network(DRN)is improved,and residual dual regression network(RDRN)is proposed.Ten thousand 512×512 pixel images from LandCover.ai and DIOR aerial image data sets were selected to form the sample data set for training and testing the network,and the reconstruction results were compared with those of other super-resolution network models.The experimental results show that RDRN has an excellent performance in both reconstruction quality and model parameters.It can achieve a better super segmentation reconstruction effect with lower model complexity and has good generalization ability for different low-resolution remote sensing images.