将两种基于地统计学的纹理特征加入到高分辨率遥感影像的城市建筑物倒塌探测中,考察了多尺度纹理对探测结果的影响.采用基于单类支持向量机的多时相直接分类方法提取建筑物倒塌信息.以伊朗巴姆地区2003 年12 月地震前后的Quickbird 遥感影像为数据源,评价和验证了本文方法的有效性.研究表明,将多尺度的空间和时相纹理信息加入到高分辨率遥感影像的倒塌建筑物探测中,可以有效提高分类精度,该方法得到的结果可应用于灾害救援及评估.

The availability of very high resolution (VHR) satellite imagery has made analysis of remotely sensed data an increasingly effective tool for detection of post-earthquake urban building damage. While methods that use spectral information alone are often ineffective owing to spectral similarity between undamaged and damaged buildings, the detailed structure information discernable in VHR images makes it possible to add spatial and temporal features in the detection. This paper proposed a method to extract multi-temporal texture by the Pseudo Cross Variogram (PCV) and multiband texture by the Multivariate Variogram (MV). The derived texture features were combined with spectral information for building damage detection in 2003 Bam earthquake using multi-temporal Quickbird images. The performance of the two texture features was evaluated at both optimalscale level and multi-scale level. The results showed that incorporating multi-temporal and multiband textures could signif icantly increase detection accuracy, and that multi-scale textures performed better than uni-scale textures.