On semi-supervised modified Fuzzy C-Means algorithm for Remote Sensing Clustering

Focusing on the problem that prior knowledge is always ignored in the Remote Sensing Classification by the unsupervised Fuzzy C-Means, a semi-supervised modified Fuzzy C-Means model for Remote Sensing image processing is proposed. The proper cluster centrals are obtained after a fast iteration going through the whole prior knowledge, which overcomes the affectation by the stochastic initializing the central of cluster. Whatpsilas more, an impact factor of labeled samples is added in the process of cyclic iteration, which efficiently deals with the problem of different spectrum characteristics with the same object, and guides the cluster direction to the correct direction to improve the convergent speed and the image segmentation precision. In addition, fundamental framework of the Fuzzy C-Means is updated for the remote sensing image segmentation, and the output of the fuzzy cluster iteration is fuzzed in reverse and automatically matches the attribute of the cluster results. In the end, error matrix and the consistence factor are introduced to verify the algorithm true effectiveness.