A New Super Resolution Framework Based on Multi-Task Learning for Remote Sensing Images.

Li Yan,Kun Chang

doi:10.3390/s21051743

Abstract

Super-resolution (SR) algorithms based on deep learning have dominated in various tasks, including medical imaging, street view surveillance and face recognition. In the remote sensing field, most of the current SR methods utilize the low-resolution (LR) images that directly bicubic downsampled the high-resolution (HR) images as not only train set but also test set, thus achieving high PSNR/SSIM scores but showing performance drop in application because the degradation model in remote sensing images is subjected to Gaussian blur with unknown parameters. Inspired by multi-task learning strategy, we propose a multiple-blur-kernel super-resolution framework (MSF), in which a multiple-blur-kernel learning module (MLM) optimizes the parameters of the network transferable and sensitive for SR procedures with different blur kernels. Besides, to simultaneously exploit the prior of the large-scale remote sensing images and recurrent information in a single test image, a class-feature capture module (CCM) and an unsupervised learning module (ULM) are leveraged in our framework. Extensive experiments show that our framework outperforms the current state-of-the-art SR algorithms in remotely sensed imagery SR with unknown Gaussian blur kernel.

Highlights

In digital image processing, low-resolution (LR) images are generally viewed as a result of a degradation function of high-resolution (HR) images
The distinctive performance of our multiple-blur-kernel super-resolution framework (MSF) springs from its ability to tackle various blur kernels and to utilize the Through the comparison experiment, our MSF outperform other state-of-the-art SR algorithms with a large gap on the vehicle, ship, and plane classes
We have proposed a new framework to tackle the super-resolution problem in remote sensing images by exploiting a multi-task learning strategy and an unsupervised learning strategy

Summary

Introduction

Low-resolution (LR) images are generally viewed as a result of a degradation function of high-resolution (HR) images. There are two widely explored SR methods: single-image SR (SISR) and multi-image SR (MISR). MISR should have played a significant role in SR for remotely sensed imagery because cameras in satellites, airplanes and drones periodically generate images of a same scene. The problems, such as image alignment, climate variation, and the change of scene content, inhibit the MISR application in remote sensing images. The SISR algorithms that achieve state-of-the-art performance often involve advancement in deep learning

Methods

Results

Discussion

Conclusion