结合风格迁移与边缘特征的SAR与光学影像匹配  

SAR and optical image matching based on style migration and edge features

作　　者：潘明 谭海 钟旭辉 PAN Ming;TAN Hai;ZHONG Xuhui(School of Geomatics,Liaoning Technical University,Fuxin Liaoning 123000,China;Land Satellite Remote Sensing Application Center,Beijing 100048,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]自然资源部国土卫星遥感应用中心,北京100048

出　　处：《北京测绘》2023年第7期938-943,共6页Beijing Surveying and Mapping

基　　金：国家自然科学基金(42274029)。

摘　　要：针对合成孔径雷达(SAR)影像与光学影像匹配中传统算法匹配效果不佳的问题,提出一种基于风格迁移与边缘特征的归一化互相关匹配算法。首先使用循环一致对抗生成网络(CycleGAN)训练从SAR影像到光学影像的风格迁移网络,将SAR影像风格迁移生成人工光学影像,其次将SAR影像及人工光学影像进行边缘提取并融合边缘得到SAR影像边缘特征,同时提取光学影像边缘特征,随后基于二值边缘特征图进行双向一致性约束下的归一化互相关匹配,最后使用快速抽样一致性算法进行误匹配剔除。实验结果表明,本算法比合成孔径雷达尺度不变特征变换(SAR-SIFT)算法与位置尺度定向不变特征变换(PSO-SIFT)算法更适应于SAR与光学影像之间的匹配,而相对于绝对相位方向梯度直方图(HAPCG)算法与辐射不变特征变换(RIFT)算法,本文算法具有更高的匹配成功率以及更低的匹配用时。Aiming at the poor matching effect of traditional algorithms in the matching of synthetic aperture radar(SAR)images and optical images,a normalized cross-correlation matching algorithm based on style migration and edge features was proposed.First,the style migration network from SAR image to optical image was trained by using the cycle consistent antagonism generation network(CycleGAN),and the SAR image style was transferred to generate the artificial optical image.Secondly,the SAR image and the artificial optical image were edge extracted and fused to obtain the edge features of the SAR image.At the same time,the edge features of the optical image were extracted,and then the normalized cross-correlation matching under the bidirectional consistency constraint was performed based on the binary edge feature map,Finally,the fast sampling consistency algorithm was applied to eliminate the mismatch.The experimental results showed that this algorithm was more suitable for the matching between SAR and optical images than SAR scale-invariant feature transform(SAR-SIFT)algorithm and position scale orientation-scale invariant feature transform(PSO-SIFT)algorithm.Compared with histogram of absolute phase consistency gradients(HAPCG)algorithm and radiation-variation insensitive feature transform(RIFT)algorithm,this algorithm had higher matching success rate and lower matching time.

关 键 词：影像匹配 风格迁移 边缘检测 合成孔径雷达影像 光学影像 归一化互相关 

分 类 号：P237[天文地球—摄影测量与遥感] TP420.20[天文地球—测绘科学与技术]