Fuzzy image regions for estimation of impervious surface areas

Abstract

A fuzzy image segmentation approach for qualitative classification of land cover was proposed recently. In this letter, such an approach is applied for estimation of impervious surface areas from Landsat-TM images. The method involves four main stages: (i) pre-processing for radiometric normalization and independent component transformation, (ii) fuzzy segmentation to create fuzzy image regions representing membership values to land cover classes, (iii) feature analysis to evaluate contextual properties of fuzzy image regions, and (iv) regression to estimate impervious surface area. In this letter, a support vector machine technique was applied to conduct supervised learning tasks. Experimental results suggest that the method provides an accurate and simple alternative for quantitative analysis of urban land cover.

Quantitative land cover change analysis using fuzzy segmentation

“…Area error was calculated as the ratio between predicted area and reference area. ISA estimation, an earlier land cover change analysis using a simpler fuzzy segmentation approach (Lizarazo, 2010) was used as benchmark. In that method, fuzzy segmentation, the first stage, outputs fuzzy-fuzzy image regions.…”

“…Although the FIRME implementation reported in this paper used a GAM & RF combination, the method can be modified for any machine learning technique. Lizarazo (2010), for example, applied the simple variant of fuzzy segmentation above mentioned using a Support Vector Machine (SVM) technique. The main advantages of the fuzzy segmentation approach are its simplicity and transferability.…”

“…This work aims to determine whether the inclusion of several spatial and spectral properties of fuzzy image regions contributes to improve the estimation of ISA values. An earlier ISA estimation using a fuzzy image partition with no inclusion of spatial properties (Lizarazo, 2010) is used as benchmark.…”

“…Current approaches for impervious surfaces mapping make use of automated masking, visual interpretation and pixel-based regression algorithms (Clapham, 2005;Jantz et al, 2005). Such approaches produce very accurate results but they are labour intensive and time consuming (Lizarazo, 2010). This paper explores the potential of the Fuzzy Image Regions Method (FIRME) for continuous classification of land cover.…”

Segmentation of Remotely Sensed Imagery: Moving from Sharp Objects to Fuzzy Regions

“…the fraction of impervious surface in each pixel, or the percentage of tree cover at each location. Lizarazo (2010) applied the fuzzy segmentation method for estimation of impervious surface area (ISA) values in Montgomery County, Maryland, USA, from Landsat-TM ortorectified images. The Landsat images, collected in 1990 and 2000, comprised of seven spectral channels with pixel size of 28.5 m at 50 m root mean square error (RMSE) positional accuracy.…”