不同特征融合的震后损毁建筑物识别研究*

(1.中国地震局工程力学研究所,黑龙江 哈尔滨 150080; 2.中国地震局地壳应力研究所,北京 100085)

遥感影像; 多特征融合; 纹理特征; 损毁建筑物

Research on Earthquake Damaged Building Extraction by Different Features Fusion
LI Qiang1,2, ZHANG Jingfa2

(1. Institute of Engineering Mechanics, CEA, Harbin 1500801, Heilongjiang, China)(2.Institute of Crustal Dynamics, CEA, Beijing 100085, China)

remote sensing image; multi-feature fusion; texture feature; damaged buildings

备注

采用玉树地震QuickBird影像,对不同损毁级别的建筑物进行采样分析,提取不同损毁情况表征纹理特征的灰度共生矩阵中区分度较好的特征参数。比较不同特征融合之后损毁建筑物提取精度。结果 表明,灰度共生矩阵中纹理特征参数对不同损毁程度的建筑物具有不同的区分度,光谱特征与纹理特征结合,损毁建筑物识别精度最高,不同的特征参数混淆使用会造成信息冗余,从而降低信息提取精度。在实际工作中,要根据遥感震害机理选取合适的特征组合,提高损毁建筑物的提取精度。

Firstly, using QuickBird image of Yushu Earthquake, we extracted characteristic parameters which have better discrimination in the gray level co-occurrence matrix(GLCM)characterized by texture features in different damage situation through sampling and analysis of buildings in different damage level. And then on the basis of it, we compared the extraction accuracy of damaged buildings by using the different features merged. The results show that the texture features parameters in GLCM have different distinguish on the buildings in different damage degree. Combined the spectral feature with texture feature, the identification precision of the damaged building is highest, however, the confusion of different characteristic parameters may lead to redundant information, resulting in reducing the accuracy of the extracted information. In practice working, based on remote sensing mechanism of damage, we should select the appropriate feature combination to improve the extraction accuracy of damaged buildings.