Impacts of feature selection for urban impervious surface extraction using optical image and SAR data

Abstract

Multisource remote sensing data can provide complementary information for object information extraction. However, the increasing of feature types and the dimensions, it is critical for OBIA (Object- Based Image Analysis, OBIA) that the identification of features and the separability between classes. Automatic the features selection and thresholds calculation can avoid time-consuming trial-and-error practice and efficient obtain important features. This article compared two features selection methods with SEaTH (Seperability and Thresholds) algorithm and CART (Classification And Regression Tree, CART), then the selected features was used to extract urban impervious surface by using GF-2 image data and Sentinel 1-A data in wuhan city of Hubei province, the experimental results shown that the impervious surface extraction accuracy with the features and thresholds by CART was higher than SEaTH algorithm. Thus, CART is much more efficient object-oriented feature selection algorithm for impervious surface extraction using optical image and sentinel 1-A data.