Abstract:
In the absence of high-resolution (HR) sensors, super-resolution (SR) algorithms for remote sensing imagery improve the spatial resolution of the images. Currently, most ...Show MoreMetadata
Abstract:
In the absence of high-resolution (HR) sensors, super-resolution (SR) algorithms for remote sensing imagery improve the spatial resolution of the images. Currently, most of the SR algorithms are based on deep learning methods, e.g., convolutional neural networks (CNNs). In particular, the generative adversarial networks (GANs) have demonstrated accepted performances in image SR owing to their powerful generative capabilities. However, remote sensing images have complex feature types, which largely limits the performances of GAN-based SR methods for real satellite images. To address this issue, an attention mechanism and a multiscale structure are introduced into the generator of the GAN network, and a multiscale attention GAN (MSAGAN) is constructed in this study. We sequentially arrange the channel attention module and the spatial attention module after the multiscale structure to emphasize important information, suppress unimportant information details, and improve the model’s performance. Furthermore, we add residual connections and dense blocks to further enhance the performance of the generative network by increasing its depth. We compared with other existing deep learning-based SR methods, and our proposed MSAGAN algorithm performed better in generating high spatial satellite images.
Published in: IEEE Geoscience and Remote Sensing Letters ( Volume: 20)