Typical disaster damage target extraction method based on object oriented classification

Abstract

The extraction of typical targets (such as water, road and damaged buildings) of nature disaster is important to the emergency management, post-disaster damage assessment and disaster monitoring. This paper shows an objected oriented method to extract water, road and building targets in the earthquake of Wenchuan(2008), Yushu(2010) and Haiti(2010). We built the optimal feature sets with spectral features, shape features, texture features and context features. Experiment result shows that this method can extract flood area, road and damaged buildings effectively and achieve a relatively high accuracy. These experimental studies are leading to the opportunity to integrate classical damage survey and image oriented semi-automatic interpretation.