Building damage detection from post-quake remote sensing image based on fuzzy reasoning

The paper presents an approach for building damage detection from high resolution remote sensing image using multi-feature analysis and the fuzzy reasoning procedure. The selected area of our study is in Yushu, which was strongly hit by 7.1-magnitude earthquake. The study area contains 101 buildings, of which 46 are collapsed and 55 are un-collapsed. First, the buildings were selected one-by-one from the GIS data and remote sensing image. Second, three categories of features were analyzed to describe the differences between the collapsed buildings and un-collapsed ones, including spectral feature, texture feature and gradient feature. Last, a final decision was made through considering the variety of feature parameters utilizing fuzzy reasoning. The overall accuracy of building damage detection was 91.09%, of the total 46 collapsed buildings, 42 were detected correctly by the proposed approach, giving 91.30% producer´s accuracy.