图割能量驱动的高分辨率遥感影像震害损毁建筑物检测

发布时间：2019-02-26 11:44

【摘要】：为充分利用高分辨率遥感影像提供的细节信息,提高震害损毁建筑物检测精度,提出了一种图割框架下融合形状、边缘、角点等多种类型特征的损毁建筑物检测方法。该方法首先利用震前建筑物线划图提取包含单个建筑物的局部影像,用于图割能量函数建模,并分别根据建筑物的位置、形状、边缘以及角点构造能量函数的各约束项。在此基础上,通过最大流/最小割算法求解能量函数最小值,依据最小割能量进行地震前后局部影像中建筑物的相似性度量。最后利用最大期望算法(expectation maximization,EM)求得最小割能量的分类阈值,并根据错分率估值执行后处理以获得最终变化检测结果。采用2011年3月11日东日本大地震前后石卷港的高分辨率遥感影像进行试验,结果表明本文提出的方法能有效检测出损毁建筑物。
[Abstract]:In order to make full use of the detailed information provided by high-resolution remote sensing images and to improve the detection accuracy of earthquake damage damaged buildings, a new method for detecting damaged buildings is proposed, which combines shape, edge, corner and other features under the frame of graph cutting. In this method, the local image containing a single building is extracted from the building line map before the earthquake, which is used to model the energy function of a single building, and the constraints of the energy function are constructed according to the position, shape, edge and corner of the building. On this basis, the maximum flow / minimum cut algorithm is used to solve the minimum value of energy function, and the similarity of buildings in local images before and after earthquake is measured according to the minimum cut energy. Finally, the maximum expectation algorithm (expectation maximization,EM) is used to obtain the classification threshold of the minimum cut energy, and the post-processing is performed according to the estimation of the error rate to obtain the final change detection results. The high-resolution remote sensing images of Shihuo Harbor before and after the East Japan earthquake on March 11, 2011 are tested. The results show that the method proposed in this paper can effectively detect damaged buildings.
【作者单位】：中南大学有色金属成矿预测与地质环境监测教育部重点实验室地球科学与信息物理学院;
【基金】：国家自然科学基金(41301453) 湖南省自然科学基金(2017JJ3378) 国家973计划(2012CB719903) 教育部博士点基金(20130162120027)~~
【分类号】：P237

【参考文献】