A NEW FRAMEWORK OF BUILDING COLLAPSE EXTRACTION BASED ON AZIMUTH TIME-FREQUENCY ANALYSIS

Rapidly obtaining building damage information after the earthquake is vitally important to earthquake rescue and decision making. In this paper, we propose a new framework of building collapse extraction by combining polarimetric information and azimuth time-frequency analysis. The proposed framework consists of two main steps, i.e. open area removal and intact building identification. In the first step, we introduce the Shannon entropy to polarimetric parameters derived from H/Lambda/alpha decomposition to remove open areas. Since intact buildings with orientation angles present a similar scattering mechanism to collapsed buildings, it seriously hampers the building collapse extraction accuracy. To solve this problem, in the second step we try to use azimuth time-frequency analysis to eliminate this deficiency. In the azimuth direction, sub-aperture images are obtained by spectrum splitting, which corresponds to observe the targets from different azimuth look angles. The nonstationary discrimination method is used combined with the SPAN parameter to identify intact buildings including intact buildings with orientation angles. The feasibility of the second step has been demonstrated by using E-SAR PolSAR data. The potential advantage of the proposed framework when we apply it to the earthquake-induced building collapse extraction includes that only one post-earthquake polarimetric SAR (PolSAR) image is needed. Moreover, orientated intact buildings can be identified and the final accuracy of building collapse extraction can be probably improved as a result.National Natural Science Foundation of China [41571337]CPCI-S(ISTP)4259-426