Quantitative land cover change analysis using fuzzy segmentation

Fuzzy image segmentation was proposed recently as an alternative GEOBIA method for conducting discrete land cover classification. In this paper, a variant of fuzzy segmentation is applied for continuous land cover change analysis. The method comprises two main stages: (i) estimation of compositional land cover for each data by fuzzy segmentation; and (ii) change analysis using a fuzzy change matrix. The fuzzy segmentation stage outputs fuzzy-crisp and crisp-fuzzy image regions whose spectral and geometric properties are measured to populate the set of predictors used to estimate land cover at single dates. The variant of fuzzy image segmentation is implemented using advanced machine learning techniques and tested in a rapidly urbanizing area using Landsat multi-spectral imagery. Experimental results suggest that the method produces accurate characterization of continuous land cover classes. Thus, the proposed method is potentially useful for enhancing the current GEOBIA perspective which focuses mainly on discrete land cover classifications.

► A method for compositional land cover mapping using fuzzy segmentation is tested in a rapidly urbanizing region.
► Advanced regression techniques are applied to conduct the supervised regression task surface areas.
► The novel method enables geographic object-based image analysis for both continuous and discrete land cover classification.
► Spatial and spectral properties of fuzzy image regions are used as predictors for estimation of impervious.
► Results show that a rich set of image regions properties may increase the accuracy of the continuous classification.