Multiple kernel collaborative fuzzy clustering algorithm with weighted super-pixels for satellite image land-cover classification

Applications are dealing with processing the large data size, especially in satellite multi-spectral imagery analysis. In fact, many reasons related to security respect, big data or limited bandwidth, the datasets cann't be centrally clustered by the FCM algorithm or its variants, so the collaborative clustering algorithms are efficiently used. The clustering algorithms have been applied to analyze the surface of the earth, specifically a variety of multispectral satellite image (MSI) classification. In this paper, a novel collaborative fuzzy clustering based framework is proposed, in particular, which is Multiple kernel Collaborative Fuzzy C-means Clustering with weighted super-pixel granulation technique (SMKCFCM algorithms) for satellite image classification. There are two phases consisting of (1) MSI are preprocessed by grouping of the similar pixels into super-pixels granules and then (2) the super-pixels granules are clustered using the collaborative fuzzy clustering with multiple kernels and the their weights. The granule's weight is determined based on its size. It means that amount of the considered objects reduces from a large amount of pixels to only a few hundred super-pixels. In the final step, the performance is improved by the collaborative clustering algorithm on multiple sites of image data. This method is combined with multiple kernels that implicitly convert the original feature space into a higher dimensional space via a non-linear map. This transformation leads to greatly increases the linear separability of the non-spherical and complex input patterns. Experiments were performed on the multi-spectral satellite image datasets and the validity indices were summarized with comparison between the SMKCFCM algorithm and some algorithms in the family of collaborative fuzzy clustering. © 2019 Elsevier Ltd