基于半监督集成的遥感图像的分割和分类-模式识别与智能系统专业论文.docxVIP

ABSTRACT ABSTRACT III AB STRACT The earth observation technology of remote sensing has been widely used in battlefield intelligence and reconnaissance，target recognition，climate change monitoring，vegetation monitoring，disaster monitoring，digital earth etc．Segmentation， classification and recognition are the key technologies in remote sensing image processing and interpretation．In recent years，semi—supervised learning has attracted increasing attention，SO this paper have a research on remote sensing image processing using the semi—supervised learning theory and methods．The main contributions of this paper call be summarized as follows： 1．A semi—supervised spectral clustering algorithm for large-scale data clustering is proposed．The uniform sampling technique combined with nearest neighbor rule is used for the large-scale data clustering．Pairwise constraints are incorporated into spectral cluster to construct semi—supervised spectral clustering in which the self-tuning parameter is applied to avoid the selection of the scaling parameter． The experimental results on the segmentation of multi-spectral remote sensing image show the effectiveness and potential of the proposed method． 2．Based on the Semi—Boost ensemble method with semi-supervised path-based similarity,a target recognition system for the SAR images is put forward．Add the pairwise constraints information generated by the labeled samples into the similarity matrix．And guided by the similarity matrix．the unlabeled samples、析tll lligh confidence are added into the training samples．and finally the performance of the algorithm is improved by ensembling multi—SVMs with weighted component．The experimental results show that the target recognition accuracy of the proposed algorithm is higher than that of other traditional methods． 3．An spectral clustering ensemble method based on boosting is proposed and applied to the segmentation of SAR images．The individual spectral clustering algorithm is constructed by adaptive