26 July 2007 Support vector regression with genetic algorithms for estimating impervious surface and vegetation distributions using ETM+ data
Author Affiliations +
Abstract
Accurate estimation of impervious surface and vegetation is a key issue in monitoring urban area and assessing urban environments. It has been proved that the nonlinear models for spectral mixture analysis outperform the linear models in the literature. However, the mapping functions of nonlinear models require to be predefined which are difficult to be determined. Support vector regression (SVR) has shown success in dealing with nonlinear problem, such as estimation and prediction. In this paper, genetic algorithm (GA) was employed to determine the optimal parameters of SVR automatically, which were applied to SVR model. Further, a GA-SVR model with multi sets of parameters (Multi-GA-SVR) was applied to estimate the distributions of impervious surface and vegetation. The results showed that Multi-GA-SVR achieved a higher accuracy than GA-SVR with single set of parameters (Single-GA-SVR) and the traditional linear mixture model (LMM), with an overall root mean square error measure (RMSE) of 0.15 for three distributions. It is demonstrated that the proposed approach is a promising approach for estimation of impervious surface and vegetation.
© (2007) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Liang Chen, Youjing Zhang, Bo Chen, "Support vector regression with genetic algorithms for estimating impervious surface and vegetation distributions using ETM+ data", Proc. SPIE 6752, Geoinformatics 2007: Remotely Sensed Data and Information, 67523L (26 July 2007); doi: 10.1117/12.761250; https://doi.org/10.1117/12.761250
PROCEEDINGS
10 PAGES


SHARE
Back to Top