文章摘要

文章信息/Info

Title:: Seismic Debris Flow Information Detection for High Resolution Images of UAV Based on Transfer Learning

作者:: 郭加伟 ¹; 2 ; 李永树 ¹; 王洪蜀 ³; 鲁恒 ⁴; 5; (1.西南交通大学地球科学与环境工程学院,四川成都 611756; 2.成都市规划信息技术中心,四川成都 610094; 3.四川水利职业技术学院测绘工程系,四川成都 611231; 4.四川大学水力学与山区河流开发保护国家重点实验室,四川成都 610065; 5.四川大学水利水电学院,四川成都 610065)

Author(s):: GUO Jiawei¹; 2; LI Yongshu¹; WANG Hongshu³; LU Heng⁴; 5
1 Faculty of Geosciences and Environmental Engineering; Southwest Jiaotong University; Chengdu 611756; Sichuan; China 2 Information Technology Center of; Chengdu Planning and Management Bureau,Chengdu 610094,Sichuan,China 3 Department of Surveying and Mapping Engineering,Sichuan Water Conservancy Vocational College,Chengdu 611231,Sichuan,China 4 State Key Laboratory of Hydraulics and Mountain River Engineering,Sichuan University,Chengdu 610065,Sichuan,China 5 College of Hydraulic and Hydroelectric Engineering,Sichuan University,Chengdu 610065,Sichuan,China

Keywords:: earthquake; debris flow; UAV high-spatial resolution images; transfer learning; information detection

摘要:: 地震后经常会引发大量的泥石流灾害(称作地震泥石流),容易造成极大的破坏,无人机低空遥感技术以其便捷、时效性强等特点成为一种快速获取灾害信息的手段,但其影像的光谱信息较为缺乏,较难准确地检测地震泥石流灾害信息。针对以上问题,提出了一种基于迁移学习机制地震泥石流检测方法,该方法在已构建地震泥石流灾害样本库的基础上,将卷积神经网络训练得到的特征迁移到地震泥石流灾害信息检测中,完成地震泥石流灾害信息的自动检测,并将面向对象的地震泥石流灾害信息检测结果与迁移学习支持下的检测结果进行了对比与分析。结果表明:基于迁移学习的地震泥石流灾害信息检测结果在精度上稍优于面向对象的地震泥石流灾害信息检测结果,且前者在保持地震泥石流的平滑性和完整性上要优于后者。

Abstract:: A lot of debris flow disasters(known as the earthquake debris-flow)occurred in post-earthquake,which cause great damage. UAV low-altitude remote sensing technology has the characteristics of convenience,timeliness and so on,and becomes a means of rapid access to disaster information. However,the spectral information of the UAV images is not enough,and it is difficult to detect the information of the earthquake debris flow disaster accurately. Taking into account of the above problems,a method based on transfer learning mechanism for earthquake debris flow detection was proposed. Based on the established earthquake debris flow disaster sample database,features trained by convolution neural network were transferred to earthquake debris flow information detection and the information was detected automatically. The earthquake debris flow information detected based on object-oriented and transfer learning were compared and analyzed. The experimental results showed that the information detection result of earthquake debris flow disaster based on transfer learning was slightly better than that of object-oriented and the former was better than the latter in maintaining smoothness and integrity of earthquake debris flow.