Remote-Sensing Image Superresolution Based on Visual Saliency Analysis and Unequal Reconstruction Networks
Libao Zhang,Donghui Chen,Jie Ma,Jue Zhang
DOI: https://doi.org/10.1109/tgrs.2019.2960781
IF: 8.2
2020-06-01
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Remote-sensing images (RSIs) generally have strong spatial characteristics for surface features. Various ground objects, such as residential areas, roads, forests, and rivers, differ substantially. According to this visual attention characteristic, regions with complicated texture features require more realistic details to reflect a better description of the topography, while regions such as farmlands should be smooth and have less noise. However, most existing single-image superresolution (SISR) methods fail to fully utilize these properties and therefore apply a uniform reconstruction strategy to the whole image. In this article, we propose a novel saliency-driven unequal single-image reconstruction network in which the demands of various regions in the superresolution (SR) process are distinguished by saliency maps. First, we design a new gradient-based saliency analysis method to produce more accurate saliency maps with imagewise annotations. The method utilizes the superiority of a multireception field to extract both high-level features and low-level features. Second, we propose a novel saliency-driven gate conditional generative adversarial network, where the saliency map is regarded as a medium during the training procedure of the whole network. The saliency map is regarded as a pixelwise condition in a generator to enhance the training capability of the network. Additionally, we design a new loss function that combines normalized content loss, saliency-driven perceptual loss, and gate-control adversarial loss to further refine details of texture-complex areas for RSIs. We evaluate the performance of our algorithm and compare it with many other state-of-the-art SR methods using a remote-sensing data set. The experimental results show that our approach achieves the optimal outcome in salient areas. Our method attains the best effect on global quality and visual performance.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics