Single Remote Sensing Image Super-Resolution via a Generative Adversarial Network With Stratified Dense Sampling and Chain Training

Super-resolution methods have significantly contributed to the improvement of the spatial resolution of remote sensing images. The development of deep learning empowers novel methods to learn informative feature representation from massive low-resolution (LR) and high-resolution (HR) image pairs. Conventional remote sensing image super-resolution methods, however, may fail in large-scale (×8,×9) super-resolution tasks. Specifically, a larger scale factor corresponds to the less information in LR images, which is a considerable challenge to super-resolution. To address the issue, we propose a novel method for single remote sensing image super-resolution (SRSISR) based on stratified dense-sampling to effectively extract image features. Specifically, the proposed super-resolution dense-sampling residual attention network (SRDSRAN) combines dense-sampling and residual learning to improve multi-level feature fusion and gradient propagation, and employs local and global attentions to learn important features and long-range interdependence in the channel and spatial dimensions. Meanwhile, we also devise a discriminator model using local and global attentions, and with the loss function integrating L ₁ pixel loss, L ₁ perceptual loss, and relativistic adversarial loss to obtain the perceptually realistic images. Besides, we introduce a chain training to promote the performance and expedite the training process for large-scale super-resolution. Experimental results on UC Merced image and other multispectral data demonstrated that our SRDSRAN outperformed the current state-of-the-art methods quantitatively and in visual quality, and obtained a higher classification accuracy in scene classification, proving its potential for applications with other downstream tasks. We will open source our code at https://github.com/Meng-333/SRADSGAN.