Fig 1.
Architecture of the proposed MCSCN model.
Our network consists of three main parts: BFE, MCSCM, and reconstruction module. We introduce a MCSCM to complement the information and make full use of the different scales of feature information.
Fig 2.
It mainly contains encoder and decoder layers, and we can use different kernel sizes to achieve different scales of information.
Table 1.
PSNR comparison of different kernel sizes for the MCSCM.
Fig 3.
The PSNR and SSIM of the different number of stacked DCUs.
Fig 4.
The visual comparison results magnified by an upscaling factor 4.
Table 2.
PSNR and SSIM comparison results among different methods.
Boldface indicates the best performance and italics indicate the second-best performance.
Table 3.
PSNR/SSIM comparison on remote sensing test datasets among different methods with up-scaling factor ×4.
Boldface indicates the best performance and italics indicate the second-best performance.
Fig 5.
The comparative results of Test20 dataset magnified by an up-scaling factor 3.
Fig 6.
The visual comparison of Test20 dataset SR obtained using different methods with an up-scaling factor 4.
Fig 7.
Comparison between model complexity and image quality.
The left vertical axis is the number of parameters, and the right vertical axis is the size of the model file.