Abstract:

This paper proposes a semi-supervised collaborative classification for hyperspectral remote sensing image with combination of cluster feature and SVM. The frame of our method combines kernel-spectral fuzzy C-means and semi-supervised SVM to improve the classification accuracy, through making full use of the advantages of classification and clustering. In details, ClusterLoss, ClassConsistent, classification difference and sample difference are created to build the collaborative classification frame, which can make the best of limited labeled samples and lot unlabeled data. This approach can minimize the cost of acquisition of labeled samples and in some degree solve the problem that support vector increases linearly with the number of training samples. Experimental results show that classification accuracy of the proposed method is more effective than that of semi-supervised SVM.

Key words: SVM, semi-supervised classification, hyperspectral remote sensing image, cluster feature