Superresolution reconstruction of optical remote sensing images based on a multiscale attention adversarial network

Qi Zhang,Rui-Sheng Jia,Zeng-Hu Li,Yong-Chao Li,Hong-Mei Sun

DOI: https://doi.org/10.1007/s10489-022-03548-7

IF: 5.3

2022-04-06

Applied Intelligence

Abstract:Due to the influence of imaging equipment and environmental conditions on optical remote sensing image acquisition, image resolution is generally low. Superresolution reconstruction technology is an important way to improve image quality. However, the existing optical remote sensing image superresolution reconstruction methods have some problems, such as insufficient feature extraction, blurred texture details of reconstructed images, and excessive network accumulation. To solve the above problems, a superresolution reconstruction method for optical remote sensing images based on a multiscale attention adversarial network is proposed in this paper. The method takes a generative adversarial network (GAN) as the basic framework. The generator uses four multiscale attention residual blocks (MSARBs) to extract image multiscale feature information and carries out feature fusion through a binary feature fusion structure (BFFS) to generate more realistic images. The discriminator uses a depth convolution network to distinguish the differences between real images and superresolution images. In the aspect of loss function construction, the perceptual loss and adversarial loss are combined to improve the perceptual quality of the images. Experimental results show that this method is superior to the compared algorithm in regard to the objective evaluation metrics of peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), and its reconstructed images have better visual effect.

computer science, artificial intelligence