【摘要】 2010年1月12日海地发生7.3级大地震,造成首都太子港大量建筑物损毁。从震后甚高分辨率遥感影像中可以发现在倒塌和部分倒塌建筑物的周围存在很多瓦砾。因此,可以将瓦砾作为建筑物倒损的震害特征。分别采用基于像元的方法、面向对象的方法、综合地统计学纹理的面向对象方法自动提取建筑物瓦砾,并对3种方法的分类精度进行了评价,研究结果表明:单纯面向对象分类方法的总体分类精度低于基于像元的分类方法;当加入纹理特征后,总体Kappa系数和瓦砾的用户精度明显提高,然而瓦砾的生产者精度的改善并不明显。更多还原

【Abstract】 An M 7.3 strong earthquake occurred in Port-au-Prince,Haiti on Jan.12,2010 has destroyed a large number of buildings.From the post-earthquake very high resolution remote sensing image,we can find there are a lot of rubbles around collapsed and partly damaged buildings.Therefore,we take these rubbles as seismic-damage characteristics of the collapsed buildings and use the following three methods—pixel-based method,object-oriented method and object-oriented method combined with geo-statistical texture to extract rubbles automatically.Finally,we assess the accuracy of the three classifications of the remotely-sensed data through the three methods.It turns out that the accuracy of pixel-based method is less than that of object-oriented method.The overall Kappa coefficient and user’s precision of rubbles have improved remarkably when adding texture feature.However,there is no obvious improvement of the producer precision of the rubbles.更多还原