四川轻化工大学学报（自然科学版）2025，Vol.38Issue(6)：51-59,9.DOI:10.11863/j.suse.2025.06.06
多尺度自适应RCAN的遥感图像超分辨率重建
Super-resolution Reconstruction of Remote Sensing Images Based on Multi-scale Adaptive RCAN
摘要
Abstract
In order to solve the problems of blurred edges,inconspicuous details and easy loss of feature information in super-resolution remote sensing images,a multi-scale adaptive Residual Channel Attention Network(RCAN)remote sensing image reconstruction network based on the RCAN architecture has been proposed,using convolutional kernels of different sizes to extract features,combined with adaptive residual connections,which effectively enhances the feature extraction ability and high-resolution image reconstruction ability of the model.The experimental results conducted on the two test sets indicate that the peak signal-to-noise ratio of the proposed algorithm reaches 37.711 dB and 37.526 dB at×2 magnification,respectively,the structural similarity reaches 0.960 and 0.962 at×2 magnification,and the learned perceptual image patch similarity scores of×2 magnification scale reach 0.016 and 0.025,which are better than the remote sensing images of super-resolution reconstruction models such as RCAN,Mixed High-order Attention Nework(MHAN)and Residual Dense Network(RDN).Moreover,the reconstructed remote sensing images are clearer in terms of details and contours,indicating the effectiveness and robustness of the proposed model.关键词
超分辨率重建/多尺度特征/注意力机制/残差网络Key words
super-resolution reconstruction/multi-scale features/attention mechanisms/residual network分类
信息技术与安全科学引用本文复制引用
雍婷婷,刘小芳..多尺度自适应RCAN的遥感图像超分辨率重建[J].四川轻化工大学学报（自然科学版）,2025,38(6):51-59,9.基金项目
国家自然科学基金面上项目(42471437) （42471437）
教育部高等教育司产学合作协同育人项目(202101038016) （202101038016）
高层次创新人才培养专项项目(B12402005) （B12402005）
四川轻化工大学人才引进项目(2021RC16) （2021RC16）
四川轻化工大学研究生创新基金资助项目(Y2024124) （Y2024124）
