福建电脑2024，Vol.40Issue(1)：33-38,6.DOI:10.16707/j.cnki.fjpc.2024.01.006
多尺度自相似遥感图像超分辨率重建网络设计
Hybrid-scale Self-similarity Remote Sensing Image Super-resolution Reconstruction Network Design
摘要
Abstract
The area captured in remote sensing images is generally large,so targets with similar features have a higher probability of repeating themselves in the image.In response to this characteristic,this paper proposes a multi-scale self-similar remote sensing image super-resolution reconstruction network.By introducing a global context module in the SSEM network structure to obtain the internal recursion of single scale and cross scale information within the image,and introducing a pixel attention module in the upsampling module to enhance its feature detail extraction ability.Tests on the UC Merced dataset show that the PSNR of our algorithm is 0.11dB,0.15dB,and 0.05dB higher than that of the HSENet algorithm at 2x,3x,and 4x scales,respectively;In terms of SSIM metrics,our algorithm outperforms the HSENet algorithm by 0.0058 and 0.0013 at 3x and 4x scales,respectively.关键词
遥感图像/自相似/超分辨率/卷积神经网络Key words
Remote Sensing Images/Self-Similarity/Super-Resolution/Convolutional Neural Networks分类
信息技术与安全科学引用本文复制引用
何松,唐程华,陈俊宽,谢唯嘉..多尺度自相似遥感图像超分辨率重建网络设计[J].福建电脑,2024,40(1):33-38,6.基金项目
本文得到江西省研究生创新专项(No.YC2022-S640)资助. （No.YC2022-S640）
