基于多视角学习的多光谱和PolSAR影像特征级协同分类  

Feature-level Collaborative Classification of Multi-spectral and PolSAR Images Based on Multi-view Learning

作　　者：王斌 肖艳 WANG Bin;XIAO Yan(Geographic Information Branch,Changchun Institute of Surveying and Mapping,Changchun 130021,China;College of Exploration and Surveying Engineering,Changchun Insititute of Technology,Changchun 130012,China)

机构地区：[1]长春市测绘院地理信息分院,吉林长春130021 [2]长春工程学院勘查与测绘工程学院,吉林长春130012

出　　处：《地理空间信息》2024年第7期50-53,共4页Geospatial Information

基　　金：吉林省科技厅资助项目(YDZJ202201ZYTS499)。

摘　　要：现有多光谱和PolSAR影像特征级协同分类研究大多忽视了不同数据源特征间的互补性关系,因此通过引入多视角学习技术,提出了一种新的多光谱和PolSAR影像特征级协同分类方法。首先将多光谱影像特征和PolSAR影像特征看作两种不同视角,采用典型相关分析算法进行特征融合;然后将融合特征、多光谱和PolSAR影像特征组合为一个特征集;最后进行特征选择和分类。以吉林省长春市部分区域为研究区,以Landsat8和RadarSat-2影像为数据源,利用该方法进行土地覆被分类,取得了较好的效果,总体精度和Kappa系数分别为91.80%和0.89;并通过对比方法进一步证明了该方法的有效性。Most of the existing researches on feature-level collaborative classification of multi-spectral and PolSAR images ignore the complementarity between features of different data sources.In order to solve this problem,we proposed a new feature-level collaborative classification method of multi-spectral and PolSAR images by introducing multi-view learning technology.In this method,multi-spectral image features and PolSAR image features were regarded as two different views,and the features of two views were fused by canonical correlation analysis algorithm at first.Then,the fusion features,multi-spectral image features and PolSAR image features were combined into a feature set.Finally,feature selection and classification were performed.Taking the southeastern part of Changchun City,Jilin Province as the research area,Landsat8 and RadarSat-2 images as the data source,we used the proposed method to conduct land-use classification,and obtained good classification results.The overall accuracy was 91.80%and the Kappa value was 0.89.In addition,we further proved the effectiveness of this method by comparison.

关 键 词：遥感 协同分类 多视角学习 典型相关分析 多光谱 POLSAR 

分 类 号：P237[天文地球—摄影测量与遥感]