基于改进SRGAN的无人机航拍图像去雾算法
Improved SRGAN-based algorithm for defogging UAV aerial images
王朝辉 1严一鸣 2韩晓微 3梁天一 1万子慷 1王起钢1
作者信息
- 1. 沈阳大学,视觉感知与智能计算沈阳市重点实验室,辽宁沈阳 110044
- 2. 中国电子科技集团公司第十一研究所,北京 100015
- 3. 沈阳大学,科技创新研究院,辽宁沈阳 110044
- 折叠
摘要
针对航拍图像往往受雾霾天气影响出现图像模糊、细节丢失等问题,本研究提出了一种基于改进SRGAN的无人机航拍图像去雾算法,旨在快速高效地去除航拍图像中的雾霾并恢复图像细节和纹理信息.本文重新设计判别器核心结构SResblock并引入CBAM注意力机制,完成了对原始SRGAN的改进,提出DH-SRGAN算法.在VISDRONE户外航拍合成雾数据集上测试结果显示,本算法在单幅图像去雾方面取得了显著提升,去雾后的图像与原始图像PSNR达24.48 dB、SSIM达95.29％,两项指标均优于传统算法.相比原始SRGAN,DH-SR-GAN算法更加轻量化,适合嵌入到无人机侦察任务中的图像预处理流程.
Abstract
Aiming at the problem that aerial images are often affected by hazy weather with image blurring and loss of details,an improved SRGAN algorithm is proposed to remove haze in aerial images quickly and efficiently and restore image details and texture information.In this paper,the core structure of discriminator SResblock is redesigned and CBAM attention mechanism is introduced to improve the original SRGAN,and DH-SRGAN algorithm is proposed.The test results on the VISDRONE outdoor aerial synthetic fog dataset show that the proposed algorithm achieves signifi-cant improvement in the fog removal of a single image,with the defogged image reaching 24.48 dB PSNR and 95.29％SSIM compared to the original image,which are better than the traditional algorithms in both metrics.Compared with original SRGAN,the DH-SRGAN algorithm is more lightweight and suitable for embedding into the image prepro-cessing process of UAV reconnaissance missions.
关键词
图像去雾/DH-SRGAN/深度学习/残差结构/注意力机制Key words
image defogging/DH-SRGAN/deep learning/residual structure/attention mechanism引用本文复制引用
基金项目
辽宁省应用基础研究计划项目(2023JH2)
辽宁省应用基础研究计划项目(101300205)
沈阳市科技计划项目(23-407-3-33)
出版年
2024
NETL
NSTL
万方数据