Multiattention Generative Adversarial Network for Remote Sensing Image Super-Resolution

Image super-resolution (SR) methods can generate remote sensing images with high spatial resolution without increasing the cost of acquisition equipment, thereby providing a feasible way to improve quality images. Clearly, image SR is severe ill-posed problem. With development deep learning, powerful fitting ability neural networks has solved this problem some extent. Since texture information various are totally different from each other, in paper, we proposed network based on generative adversarial (GAN) achieve images, named multi-attention (MA-GAN). The main body generator MA-GAN contains three blocks: pyramid-convolutional residualdense (PCRD) block, attention-based upsampling (AUP) block and fusion (AF) block. Specifically, developed attention-pyramid convolutional (AttPConv) operator PCRD combines multi-scale convolution channel attention (CA) automatically learn adjust scale residuals for better representation. established AUP utilizes pixel (PA) perform arbitrary scales upsampling. And AF employs branch (BA) integrate upsampled low-resolution high-level features. Besides, loss function takes both feature into consideration guide learning procedure generator. We have compared our approach several state-of-the-art number scenes, experimental results consistently demonstrate effectiveness MA-GAN. For study replication, source code will be released at: https://github.com/ZhihaoWang1997/MA-GAN.