Semi-supervised Method with Spatial Weights based Possibilistic Fuzzy c-Means Clustering for Land-cover Classification

In remote sensing image analysis, the accuracy of the results depends not only on the accuracy of the image acquisition process but also on the segmentation and classification accuracy of the image. The fuzzy classification technique works by dividing the pixels of the image into sets of fuzzy clusters by iteratively optimizing the objective function to update the cluster membership and center centroid. This technique overcomes the disadvantages of hard clustering; However, this method is quite sensitive to interference and extraneous elements. In this paper, we propose a novel semi-supervised clustering method with spatial weights (SPFCM-W) for multi-spectral remote sensing image land-cover classification by the extension of the possibilistic fuzzy c-means (PFCM) algorithm, in which spatial weights of the pixels and labeled data are used to increase the accuracy of clustering results when the data structure of input patterns is non-spherical and complex. Results obtained on two kinds of multi-spectral remote sensing images (Landsat-7 ETM+, Sentinel-2A) by comparing the proposed technique with some variations of the fuzzy clustering algorithm demonstrate the good efficiency and high accuracy of the proposed method.