Remote Sensing Image Super-Resolution Via Attentional Feature Aggregation Generative Adversarial Network
Feng Cai, Keyu Wu, Feng Wang · IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium · 2022
The extraction of high-frequency details is generally neglected in single image super-resolution (SISR) for remote sensing images. In this paper, we propose an attentional feature aggregation generative adversarial network (AFA-GAN) with the capability of strong feature extraction and attentional feature fusion to generate high-resolution remote sensing images. We adopt the residual feature aggregation framework for the feature extraction to make full use of the hierarchical features on the residual branches. To better fuse global and local features with inconsistent scales, an attentional feature fusion mechanism is utilized in residual feature aggregation modules. The comprehensive experiments with state-of-the-art SISR methods on the UC Merced dataset demonstrate the effectiveness and superiority of our AFA-GAN.