| 基于双分支卷积神经网络的SAR与多光谱图像融合实验 |
| |
| 引用本文： | 吴佼华,杨学志,方帅,董张玉.基于双分支卷积神经网络的SAR与多光谱图像融合实验[J].地理与地理信息科学,2021,37(2):22-30. |
| |
| 作者姓名： | 吴佼华 杨学志 方帅 董张玉 |
| |
| 作者单位： | 合肥工业大学计算机与信息学院,安徽 合肥230009;工业安全与应急技术安徽省重点实验室,安徽 合肥230009;工业安全与应急技术安徽省重点实验室,安徽 合肥230009;合肥工业大学软件学院,安徽 合肥230009;智能互联系统安徽省实验室,安徽 合肥230009;合肥工业大学计算机与信息学院,安徽 合肥230009;工业安全与应急技术安徽省重点实验室,安徽 合肥230009;智能互联系统安徽省实验室,安徽 合肥230009 |
| |
| 基金项目： | 中央高校基本科研业务费专项资金项目;安徽省重点研究;开发计划项目 |
| |
| 摘 要： | 针对合成孔径雷达(Synthetic Aperture Radar,SAR)和多光谱(Multi-Spectral,MS)融合图像中存在的空间细节模糊和颜色失真问题,该文兼顾光谱监督和空间细节监督,设计光谱损失函数和空间细节损失函数,提出一种基于双分支卷积神经网络(Convolution Neural Network,CNN)的SAR和MS图像融合算法。该算法网络框架包含光谱保持和细节提升两个分支:光谱保持分支通过上采样MS图像连接到网络的输出,直接将光谱信息传递到融合图像中;细节提升分支对SAR和MS图像通过高通滤波提取高频细节信息,然后应用CNN对细节信息进行特征提取、特征融合及重建,最后将重建的细节信息叠加到上采样的MS图像,得到融合结果。以哨兵-1B GRD级别的SAR图像和Landsat8卫星多光谱图像为实验数据,通过与传统融合算法和深度学习算法RSIFNN进行对比,结果表明,该文算法在定性和定量评价方面效果更好,能够在保持光谱信息的基础上增强多光谱图像的空间细节信息,有利于后续地物分类和目标识别等工作的开展。
|
| 关 键 词： | 合成孔径雷达图像 多光谱图像 图像融合 空间细节信息 卷积神经网络 |
SAR and Multispectral Image Fusion Experiment Based on Dual Branch Convolutional Neural Network |
| |
| WU Jiao-hua,YANG Xue-zhi,FANG Shuai,DONG Zhang-yu.SAR and Multispectral Image Fusion Experiment Based on Dual Branch Convolutional Neural Network[J].Geography and Geo-Information Science,2021,37(2):22-30. |
| |
| Authors: | WU Jiao-hua YANG Xue-zhi FANG Shuai DONG Zhang-yu |
| |
| Institution: | (School of Computer and Information,Hefei University of Technology,Hefei 230009;Anhui Province Key Laboratory of Industrial Safety and Emergency Technology,Hefei 230009;School of Software,Hefei University of Technology,Hefei 230009;Intelligent Interconnected System Laboratory of Anhui Province,Hefei 230009,China) |
| |
| Abstract: | Aiming at the problems of spatial detail blur and color distortion in synthetic aperture radar(SAR)and multi-spectral(MS)fusion images,this paper designs spectral loss function and spatial detail loss function considering both spectral supervision and spatial detail supervision,and proposes a SAR and MS image fusion algorithm based on dual branch convolution neural network(CNN).The network framework of the algorithm consists of two branches:spectrum preserving and detail enhancing.Spectrum preserving branch uses up-sampled MS images to connect to the output of the network,and directly transfers the spectral information to the fusion image;detail enhancing branch first extracts high-frequency detail information from SAR and MS images by high pass filtering,then uses CNN to carry out feature extraction,feature fusion and reconstruction of the detail information.Finally,the reconstructed detail information is superimposed on the up-sampled MS images to obtain the fusion result.Compared with the traditional fusion algorithms and deep learning fusion algorithms,the proposed algorithm achieves better results in visual evaluation and quantitative evaluation,and can enhance the spatial details of MS images on the basis of maintaining spectral information,which is conducive to the following work such as terrain classification and target recognition. |
| |
| Keywords: | synthetic aperture radar image multi-spectral image image fusion spatial detail information convolution neural network |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
