基于半监督子空间迁移的稀疏表示遥感图像场景分类方法

利用已有的标记数据对新领域图像进行分类是遥感图像场景分类的重要研究方向。提出了一种基于半监督子空间迁移的稀疏表示（sparse representation method based on semi-supervised transfer learning subspace，SR-SSTLS）遥感图像场景分类方法。为减少源域和目标域数据分布变化，将不同数据域的遥感图像投影至共享子空间。源域和目标域数据在投影子空间协同学习共享字典，使得带标记的源域数据辅助目标域模型的建立。同时，建立了基于源域、目标域、源域-目标域标记数据的拉普拉斯图矩阵和目标域未标记数据的拉普拉斯正则化项，使得目标域中的数据均得到很好编码。在多个遥感图像数据集上的实验结果均证明了SR-SSTLS方法的有效性。

A sparse representation method based on semi-supervised transfer learning subspace for remote sensing scene classification

How to use the existing labeled datasets to classify new images is an important research direction in remote sensing scene classification.In this paper,a sparse representation method based on semi-supervised transfer learning subspace (SR-SSTLS) is proposed for remote sensing scene image classification.In order to reduce the influence of the data distribution of source domain and target domain,remote sensing images of different domains are projected into a shared subspace,and in such shared subspace,a shared dictionary is learned from the data of both source domain and target domain,so that a large number of labeled source domain data can assist the model establishment in target domain.In order to make the data in the target domain be well encoded,the Laplacian matrixes of labeled data are built in source domain,target domain,and source-target domain. Meantime,the Laplacian regularization term of unlabeled data is built in target domain. Experimental results on several remote sensing image datasets show the effectiveness of SR-SSTLS.