摘要
遥感影像中含有较为丰富的高频信息,但受硬件设备限制和外界环境干扰,高频信息边缘部分较为模糊,影响遥感影像的空间分辨率质量。针对这一问题提出一种基于边缘检测和密集残差的遥感影像超分辨率重建方法。该方法利用边缘检测算子提取影像边缘细节特征,采用密集残差模块代残差网络复用底层特征,并将每一层的特征图进行加权融合,最后通过生成器完成遥感影像的超分辨率重构,实现遥感影像分辨率质量的提高。实验结果表明,该算法指标优于传统的双三次插值、SRCNN、VDSR和SRGAN,为遥感影像重建提供新的解决方法与技术思路。
Remote sensing images contain rich high-frequency information,but due to hardware limitations and external environmental interference,the edge of high-frequency information is blurred,which affects the spatial resolution quality of remote sensing images.Aiming at this problem,this paper proposes a remote sensing images super-resolution reconstruction method based on edge detection and dense residuals.This method used edge detection operators to extract image edge detail features,adopted dense residual module to replace the residual network to multiplex the underlying features,and weighted and fused the feature maps of each layer.The super-resolution of the remote sensing image was completed by the generator.It realized the improvement of the resolution quality of remote sensing images.The experimental results show that the index of the proposed method is superior to the traditional bicubic interpolation,SRCNN,VDSR and SRGAN.It provides a new solution and technical ideas for the reconstruction of remote sensing images.
作者
李景文
陈文达
姜建武
Li Jingwen;Chen Wenda;Jiang Jianwu(Guangxi Key Laboratory of Spatial Information and Surveying,Guilin 541004,Guangxi,China;College of Geomatics,Guilin University of Technology,Guilin 541004,Guangxi,China)
出处
《计算机应用与软件》
北大核心
2023年第2期240-245,共6页
Computer Applications and Software
基金
国家自然科学基金项目(41961063)
国家文化和旅游科技创新工程项目(2019-011)。
关键词
遥感影像
超分辨重建
生成对抗网络
边缘特征
Remote sensing image
Super-resolution reconstruction
Generative adversarial network
Edge feature
