Due to the scale diversity of geographical objects, hierarchical remote sensing image segmentation plays an important role in object-oriented image analysis. In this paper, a hierarchical remote sensing images segment...
详细信息
ISBN:
(纸本)9781538636756
Due to the scale diversity of geographical objects, hierarchical remote sensing image segmentation plays an important role in object-oriented image analysis. In this paper, a hierarchical remote sensing images segmentation method with unsupervised deep learning features is proposed. The unsupervised deep learning features of an image are extracted with a sparse convolutional auto-encoder. Both deep features and the clustering information of deep features are utilized in hierarchical image segmentation which is a bottom-up region merging process. The iterative region merging process starts from an initial partition of an image under a merging criterion. Finally, a tree-like image segmentations hierarchy which contains ground objects of different scales is obtained. The proposed method integrates the advantages of unsupervised deep learning features with region merging-based hierarchical image segmentation. The experiment results have shown the proposed method is superior to the methods using only spectral or spatial features.
暂无评论