Automatic Recognition of Built-up Areas in China Using CBERS-2B HR Data

Abstract

The study presented here evaluates the possibility to map the human settlement patterns in China using CBERS-2B 2.36m-resolution single-band panchromatic images as input. The automatic image information extraction workflow designed by the Joint Research Center, European Commission (JRC) is used for the global human settlement layer (GHSL) production. The classification methodology relies on textural and multi-scale morphological image feature retrieval and automatic learning and classification from low resolution globally-available reference layers. According to the experimental results, the most promising learning and classification option is the one using perscene area-matching techniques with MODIS 500 “urban areas” as training layer. This option provides an average accuracy of 98.13% with 5.66% of standard deviation. More refined analysis is necessary using detailed reference datasets in order to understand the value added to the propose methodology.