Estimating urban impervious surfaces using LS-SVM with multi-scale texture

Abstract

Various methodologies have been used to estimate and map percent impervious surface using medium resolution remote sensing imagery. However, there appears to be few study conducted on the use of SVR for estimating ratio of impervious surfaces. The aim of this paper is to compare the effectiveness both of two advanced algorithms and three feature set for estimating and describing impervious surface. Landsat imagery (acquired on Sep. 16, 2000 and Apr. 2, 2006) in Nanjing, China, were used for the analysis. The linear spectral mixture analysis (LSMA) and least-squares support vector machine (LS-SVM) were employed to extract impervious surface. Accurate assessment was performed against a high-resolution IKONOS image. The results show that LS-SVM was more effective than LSMA in extracting impervious surfaces with high statistical accuracy. The root-mean-square error (RMSE) of the impervious surface map using LS-SVM model was 0.106 compared with 0.246 using LSMA. Also, the LS-SVM with multi-scale texture was obtained the lowest error than the spectrum and single scale texture. It is demonstrated that the LS-SVM with multi-scale texture is of capability of handling the nonlinear mixing of the image spectrum and the complex distribution of urban objects.