Remote Sensing Image Classification Method Based on Preferential Adaptive Interval-Value Fuzzy C-Means

Abstract

The heterogeneity of objects within the class and the ambiguity of objects between the classes in remote sensing images cause the uncertainty in ground objects classification. Fuzzy set theory could express the fuzziness effectively, while interval-value data model can reflect the uncertainty of the data. Therefore, combining the interval-value data model and fuzzy c-means algorithm, a preferential adaptive interval-value fuzzy c-means (PA-IVFCM) algorithm is proposed in this paper. The overall interval width of the category is adjusted by normalizing mean square error in the class, the interval modeling of the data is selected by using the preferential factor dynamically, thereby increasing the intra-class compactness and the boundary separability. The experimental results show that PA-IVFCM method can be effectively applied in the SPOT5 remote sensing data classification, and the overall classification accuracy and Kappa coefficients are greatly improved compared with the existing popular fuzzy classification methods.