中文摘要：

针对基于图的半监督图像分类方法扩展性差的问题,结合了mean shift图像聚类算法和基于锚点建图的方法并将其应用于遥感图像的分类中。首先采用mean shift聚类算法对遥感图像聚类;其次根据基于锚点建图的半监督分类方法,选取聚类中心作为锚点,利用锚点集和标记样本集建图,达到缩小图规模的目的,并建立锚点与样本间的关联矩阵;然后通过分类器得到锚点的标记信息;最后由样本与锚点间的关联矩阵还原得到遥感图像的分类结果。实验结果表明,该方法对遥感图像分类时,能够有效地降低计算复杂度,同时获取较好的分类结果。

英文摘要：

To overcome the poor scalability of the graph-based semi-supervised image classification method, the method combining the mean shift image clustering algorithm with anchor graph construction is applied to the classification of remote sens- ing image. The mean shift clustering algorithm is used to cluster the remote sensing image. According to the semi-supervised image classification method based on anchor graph regularization （AGR）, the cluster center is selected as an anchor point, and the anchor point set and marking sample set are used to construct the graph to reduce the graph size and establish the correla- tion matrix between the anchor point and sample. The marking information of the anchor point is obtained with the classifier. And then the classification result of the remote sensing image is obtained according to the correlation matrix between the sample and anchor point. The experimental results show that the presented method can reduce the computational complexity effectively and obtain the satisfied classification results while classifying the remote sensing image.