基于图像块SIFT特征学习的多源遥感图像匹配方法  

Multi-source Remote Sensing Image Matching Method Combined with SIFT and Image Patch Feature Learning

作　　者：李梓谦 付志涛 聂韩 陈思静 张菲菲 LI Ziqian;FU Zhitao;NIE Han;CHEN Sijing;ZHANG Feifei(Faculty of Land Resources Engineering,Kunming University of Science and Technology,Kunming 650000,China;School of Computer Science,Guangdong University of Education,Guangzhou 510303,China)

机构地区：[1]昆明理工大学国土资源工程学院,昆明650000 [2]广东第二师范学院计算机学院,广州510303

出　　处：《遥感信息》2023年第1期155-162,共8页Remote Sensing Information

基　　金：云南省科技厅基础研究计划面上项目(202101AT070102、202101BE070001-037);广州市科技计划项目(201804010280)。

摘　　要：对于多源遥感图像因成像原理、时相差异以及分辨率等因素导致的匹配困难问题,提出了一种基于图像块SIFT特征学习的多源遥感图像匹配方法。首先,通过尺度不变特征变化(scale-invariant feature transform,SIFT)提取图像特征点并截取对应的图像块对;其次,利用多源遥感图像匹配网络(matching nerul network,MNN)学习图像块特征并输出匹配点对,结合快速样本一致性(fast sample consensus,FSC)方法优化匹配结果;最后,计算图像变换矩阵实现多源遥感图像配准。为验证本文方法的有效性,制作了8000对多源遥感图像数据集对MNN进行网络训练,并与FSC-SIFT(fast sample consensus-scale-invariant feature transform)、PSO-SIFT(position scale orientation-scale-invariant feature transform)以及跨模态图像匹配网络Contextdesc进行对比实验,结果表明本文方法在正确匹配点数量、匹配精度等方面具有一定优越性。For the difficulties of multi-source remote sensing image matching caused by imaging principle,phase difference and resolution,a multi-source remote sensing image matching method combined with SIFT and image patch feature learning is proposed in this paper.Firstly,the feature points are extracted by the scale invariant feature transform method and the corresponding image patch pairs are intercepted.Secondly,the matching point pairs are output by using the matching nerul network(MNN)and the fast sample consensus(FSC)method.Finally,the image transformation matrix is calculated to realize multi-source remote sensing image registration.In order to verify the effectiveness of the proposed method,this paper makes 8000 pairs of multi-source remote sensing image data sets for network training of MNN.Then it is compared with fast sample consumption scale invariant feature transform(FSC-SIFT),position scale orientation scale invariant feature transform(PSO-SIFT)and cross modal image registration network(Contextdesc).The results show that the proposed method has advantages in the number of correct matching points and matching accuracy.

关 键 词：多源遥感图像 图像块匹配 图像配准 特征学习 神经网络 

分 类 号：P237[天文地球—摄影测量与遥感]