迁移学习支持下的高分影像山地滑坡灾害解译模型  

An Automatic Interpretation Model for Mountains Landslide Disaster of High-Resolution Remote Sensing Images Based on Transfer Learning

作　　者：郭加伟[1,2] 李永树[1] 李政[1,2] 刘锟铭 张帅毅[1,2] GUO Jiawei LI Yongshu LI Zheng LIU Kunming ZHANG Shuaiyi(Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu 611756, China Key Laboratory of Geo-special Information Technology, Ministry of Land and Resources, Chengdu University of Technology, Chengdu 610059, China)

机构地区：[1]西南交通大学地球科学与环境工程学院,四川成都611756 [2]成都理工大学国土资源部地学空间信息技术重点实验室,四川成都610059

出　　处：《测绘科学技术学报》2016年第5期496-501,共6页Journal of Geomatics Science and Technology

基　　金：国家"十二五"科技支撑计划项目(2014BAL01B04);中央高校基本科研业务费专项资金项目(2682013ZT26);2014基础测绘科技计划项目;国土资源部地学空间信息技术重点实验室开放基金项目(KLGSIT2015-03);四川省应急测绘与防灾减灾工程技术研究中心开放基金

摘　　要：无人机低空遥感是近年来新兴的一种快速获取灾情信息的手段,如何利用无人机高分影像构建滑坡灾害解译模型是实现快速自动解译滑坡的关键。针对该问题,对比了多种影像特征提取方法,将迁移学习(TL)特征和支持向量机(SVM)引入到构建滑坡灾害自动解译模型中,提出了一种TL支持下的高分影像滑坡灾害解译模型。选取5·12汶川地震及4·20芦山地震系列无人机影像构建了滑坡灾害样本库并进行了实验,TL特征方法整体分类准确度ACC为95%,ROC达到0.98,识别准确率达到97%。结果表明,所提方法可用于高分影像滑坡自动解译,同时可用于大面积高分影像中快速山地滑坡灾害定位及检测。In recent years, low-altitude UAV remote sensing is emerging as a means to rapid access to disaster information. How to quickly and accurately interpret the landslide after a landslide disaster can provide reliable data sources for subsequent monitor and forecast. Then how to use UAV high spatial resolution images to construct an interpretation model for mountain landslide disaster is the key to achieve fast and automatic interpretation of landslide. To solve this problem, multiple image feature extracting methods were compared firstly. Then, the transfer learning （TL） and support vector machine（SVM） were introduced to construct the automatic interpretation model, and an automatic interpretation model for mountain landslide disaster with high spatial resolution images based on transfer learning method was proposed. Finally, to construct landslide disaster sample database, UAV series images of 5.12 and 4-20 earthquake were selected, and the experiments based on the database were carried out. The comprehensive classification accuracy ACC of transfer learning feature method was 95%, ROC was 0.98, and the recognition accuracy was 97%. Experimental results showed that the proposed method could be used for landslide automation interpretation with high spatial resolution remote sensing images, and quick location and detection of landslide disaster with high spatial resolution images for large area.

关 键 词：山地滑坡 迁移学习 无人机影像 自动解译 灾害检测 

分 类 号：P237[天文地球—摄影测量与遥感]