基于深度学习的多帧遥感降质图像三维重建算法  

A deep learning based 3D reconstruction algorithm for multiframe degraded remote sensing images

作　　者：石力源 SHI Liyuan(Hangzhou Dianzi University,Hangzhou 310018,China)

机构地区：[1]杭州电子科技大学,浙江杭州310018

出　　处：《现代电子技术》2024年第6期161-164,共4页Modern Electronics Technique

摘　　要：为提升多帧遥感降质图像对比度以及图像质量,提出一种基于深度学习的多帧遥感降质图像三维重建算法。采用三角函数变换方法并结合高通滤波器,增强多帧遥感降质图像对比度;再以包含生成器和判别器的生成对抗网络为基础,在判别器中引入自注意力层,设计自注意力机制残差模块,生成自注意力生成对抗网络模型;最后将增强后的图像输入模型进行学习和训练,获取多帧遥感降质图像的全局特征后,实现多帧遥感降质图像三维重建。测试结果表明,所提算法具有较好的多帧遥感降质图像增强能力,能够提升图像对比度,并且渗透指数(PI)均在0.92以上,重构效果良好。In order to improve the contrast and image quality of multiframe degraded remote sensing images,a deep learning based 3D reconstruction algorithm for multiframe degraded remote sensing images is proposed.The trigonometric function transformation method combined with high pass filter is used to enhance the contrast of multiframe degraded remote sensing images.Based on a generative adversarial network that includes generators and discriminators,a self attention layer is introduced into the discriminator,and a residual module of the self attention mechanism is designed to generate a self attention generative adversarial network model.The enhanced image input model is learned and trained to obtain global features of multiple degraded remote sensing images,and then three-dimensional reconstruction of multiple degraded remote sensing images is achieved.The testing results show that the algorithm has good ability to enhance multi frame degraded remote sensing images and improve image contrast.The permeability indexs(PI) are all above 0.92,and the reconstruction effect is good.

关 键 词：多帧遥感图像 降质图像 深度学习 三维重建 图像增强 生成对抗网络 自注意力层 全局特征 

分 类 号：TN957.52-34[电子电信—信号与信息处理] TP391[电子电信—信息与通信工程]