利用Deeplab v3提取高分辨率遥感影像道路  

Road Extraction of High Resolution Remote Sensing Imagery Based on Deeplab v3

作　　者：韩玲[1,2] 杨朝辉 李良志 刘志恒 黄勃学[1] HAN Ling;YANG Zhaohui;LI Liangzhi;LIU Zhiheng;HUANG Boxue(School of Geology Engineering and Geomatics,Chang’an University,Xi’an 710054,China;Shaanxi Key Laboratory of Land Consolidation,Xi’an 710054,China)

机构地区：[1]长安大学地质工程与测绘学院,西安710054 [2]陕西省土地整治重点实验室,西安710054

出　　处：《遥感信息》2021年第1期22-28,共7页Remote Sensing Information

基　　金：装备预研教育部联合基金项目(6141A02022376);陕西省土地整治重点实验室基金项目(2018-ZZ04)。

摘　　要：针对传统道路提取方法存在的道路边缘粗糙、抗干扰性弱、提取精度低等问题,提出了一种基于编码解码器的空洞卷积模型(Deeplab v3)的道路提取方法。首先,对原始高分辨率遥感影像进行标注;其次,利用标注数据集对Deeplab v3模型进行训练、测试;最后,得到高分辨率遥感影像道路提取结果。分析结果可知,该模型能够较好地提取高分辨率遥感影像中的道路边缘特征,相比其他道路提取方法具有更高的提取精度和更加完整的道路信息,正确率可达到93%以上。A new road extraction method based on the Deeplab v3 model is proposed to solve the problems of traditional road extraction methods such as rough road edge,weak anti-interference and low extraction accuracy existing.A three-step procedure is developed in this study for extracting roads based on high-resolution remote sensing image.Firstly,label the high-resolution remote sensing image.Secondly,the Deeplab v3 model is trained and tested by using the label data set.Finally,get the road extraction results of the high-resolution remote sensing image.The results indicate that the Deeplab v3 model can excellently extract the road edge features combined with the high-resolution remote sensing image.Compared with other road extraction methods,this proposed method displays more complete extracted road information and higher extraction accuracy,which has the accuracy over 93%.

关 键 词：道路提取 高分辨率遥感影像 深度学习 Deeplab v3 空洞卷积 空洞空间金字塔池化(ASPP) 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]