Super-resolution of Aerial Images Based on Global Dependency Transformer
Xue Qin Tang, Zhongjun Mao, Junqin Zhao, Jin Liu, Haihang Xu, Hongchuan Song, Shuai Guo · 2025
For the problem of global feature extraction difficulty in the super-resolution task of aerial images, a super-resolution reconstruction model for aerial images based on global dependence transformer is proposed. The proposed model innovatively introduces a Re-parameterized Spatial Frequency Block (RSFB) with global feature extraction capability, which compensates for the defect of the original model's weak global information perception capability. In addition, the proposed model also constructs a Dual-channel Residual Group (DARG) with the ability of spatial and channel dual-channel feature aggregation, which substantially improves the model's reliance on global information. Experimental results on the aerial image dataset NWPU-RESISC45 show that the proposed model has obvious advantages in PSNR and SSIM metrics.