联合辐射指数的多流融合网络空谱融合方法  

Spatial-spectral fusion method of multi-stream fusion network based on joint radiometric indices

作　　者：雷大江[1,2] 王纪 张策 张莉萍[2] 李伟生[2] 于洪[1] LEI Dajiang;WANG Ji;ZHANG Ce;ZHANG Liping;LI Weisheng;YU Hong(College of Computer,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Key Laboratory of Image Cognition,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065 [2]重庆邮电大学重庆市图像认知重点实验室,重庆400065

出　　处：《太赫兹科学与电子信息学报》2023年第1期88-94,共7页Journal of Terahertz Science and Electronic Information Technology

基　　金：国家自然科学基金资助项目(61972060,61802148,U1401252);重庆市杰出青年基金资助项目(cstc2014jcyjjq40001);重庆市海外留学人员创新创业基金资助项目(cx2018120)。

摘　　要：结合遥感图像融合的特点,提出一种联合辐射指数的多流融合生成对抗网络的遥感图像融合方法。该方法从初始的多光谱图像和全色图像中按照特定的规则提取特征图,利用生成器子网络分别提炼输入的多源图像,在特征域上叠加特征后通过生成器的主网络进行融合。根据遥感领域中多光谱图像和全色图像在波段上的特性,在判别器中引入调制传递函数(MTF)来判别融合图像的光谱信息和空间结构信息。为评估所提方法的有效性,进行视觉分析,并与其他算法进行客观评价的比较。实验结果表明,该方法在视觉效果和客观评价上优于其他算法。A remote sensing image fusion method is proposed based on multi-stream fusion generative adversarial network with radiation index.This method extracts feature maps from the MultiSpectral(MS)image and Panchromatic(PAN)image according to specific rules,and uses the generator sub-network to refine the input multi-source image,then fuses images through the main network of the generator after superimposing the features.According to the characteristics of multi-spectral and panchromatic images in the remote sensing field,the Modulation Transfer Function(MTF)is introduced into the discriminator to simultaneously discriminate the spectral information and spatial structure information of the fused images.In order to evaluate the effectiveness of the proposed method,its visual analysis is conducted and it is compared with other algorithms for objective evaluation.Experimental results show that this method is superior to other algorithms in visual effects and objective evaluation.

关 键 词：遥感图像融合 生成对抗网络 辐射指数 多流融合 

分 类 号：TN911.73[电子电信—通信与信息系统]