光学精密工程2024，Vol.32Issue(19)：2971-2985,15.DOI:10.37188/OPE.20243219.2971
基于频域-空域混合注意力的降质遥感图像质量增强
Degradation remote sensing image quality enhancement based on frequency-domain-spatial-domain hybrid attention
摘要
Abstract
Traditional image processing algorithms lack stability,and deep learning algorithms fall short of engineering requirements for remote sensing due to insufficient training datasets and high computational de-mands.To tackle this,the paper integrates degradation modeling and image processing.It introduces a method for creating a remote sensing image enhancement dataset using a Zernike polynomial degradation model.Additionally,it designs an algorithm for enhancing degraded remote sensing images using a hybrid frequency-domain and spatial-domain attention mechanism.This algorithm employs a dual-domain selec-tion module and a frequency feature residual module to improve the learning of high-frequency image tex-tures and details in both domains.The hybrid attention mechanism further boosts feature extraction capa-bilities.The algorithm's performance was validated against five common methods using NIQE values,vi-sualization effects,MTF curves,and inference efficiency.Results show that the proposed approach signifi-cantly reduces NIQE values and improves MTF curves,leading to clearer images and substantially en-hancing degraded image quality.For images with a specific pixel size of 27 620×29 200,the algorithm processes them in just 27 s,compared to hours for traditional methods,thus meeting engineering timeliness re-quirements.This research offers a rapid and effective solution for addressing satellite imaging degradation.关键词
光学遥感图像/图像增强/深度学习/Zernike多项式/图像退化模型Key words
optical remote sensing images/image enhancement/deep learning/Zernike polynomial/im-age degradation model分类
计算机与自动化引用本文复制引用
魏花,唐熊忻,聂海涛,王静,杨瀚翔,夏媛媛,徐帆江..基于频域-空域混合注意力的降质遥感图像质量增强[J].光学精密工程,2024,32(19):2971-2985,15.基金项目
国家重点研发计划资助项目(No.2021YFB3601404) （No.2021YFB3601404）
