高分辨率遥感影像滑坡纹理特征提取方法  

A Method for Extracting Texture Features of Landslide in High Resolution Remote Sensing Images

作　　者：王庆雅 张锦[1] WANG Qingya;ZHANG Jin(College of Mining Engineering, Taiyuan University of Technology, Taiyuan 030006, China)

机构地区：[1]太原理工大学矿业工程学院,太原030006

出　　处：《太原理工大学学报》2021年第4期547-556,共10页Journal of Taiyuan University of Technology

基　　金：国家重点研发计划地球观测与导航专项子课题(2018YFB0505402);国家自然科学基金面上项目(41771443)。

摘　　要：针对目前滑坡提取所需数据多样、算法复杂、特征冗余且对于单体滑坡的纹理特征研究还不够充分等问题,根据Gabor滤波进行滑坡纹理特征分析,并融合光谱特征利用支持向量机实现滑坡提取。首先对滑坡影像进行多尺度多方向Gabor变换,其次对同一方向不同尺度Gabor特征进行融合得到多方向特征,最后对多方向Gabor纹理特征进行最大值过滤,构成多尺度多方向Gabor纹理特征。结果表明,对比局部二值模式、局部空间自相关,得出纹理分析方法对滑坡提取效果最好,具有较高的提取精度。High resolution remote sensing images play an important role in landslide recognition.In order to solve the problems of diverse data,complex algorithms,redundant features,and insufficient research on the specific texture characteristics of a single landslide,Gabor filtering was used for texture feature extraction and fusion of spectral features using Support Vector Machine to achieve landslide extraction.First,multi-scale and multi-direction Gabor transformation is carried out on the landslide image.Then,the Gabor features of different scales in the same direction are fused to obtain the multi-directional features.Finally,the multi-directional Gabor features are filtered at the maximum to form the multi-scale and multi-directional Gabor texture features.The results show that,compared with the local binary model and local spatial autocorrelation,the texture analysis method in this paper has the best effect on landslide extraction and a high extraction accuracy.

关 键 词：高分辨率遥感影像 多尺度多方向Gabor纹理特征 滑坡提取 

分 类 号：P642[天文地球—工程地质学] P237[天文地球—地质矿产勘探]