一种基于改进U-Net的高分影像公路线路提取方法  

A highway extraction method from high-resolution images based on improved U-Net

作　　者：廉慧洁 王维高 朱军 唐苒然 谢亚坤 LIAN Huijie;WANG Weigao;ZHU Jun;TANG Ranran;XIE Yakun(Faculty of Geosciences and Environmental Engineering,Southwest Jiaotong University,Chengdu 610097,China;Sichuan Communication Surveying&Design Institue Co.,Ltd.,Chengdu 610031,China)

机构地区：[1]西南交通大学地球科学与环境工程学院,成都610097 [2]四川省交通勘察设计研究院有限公司,成都610031

出　　处：《时空信息学报》2023年第3期335-344,共10页JOURNAL OF SPATIO-TEMPORAL INFORMATION

基　　金：四川省科技计划项目(2022YFS0533);四川省交通运输科技项目(2020-B-05);重庆公路科技项目(R113520H01095)。

摘　　要：高质量的公路网信息在区域经济发展、灾害应急管理和土地规划中发挥着重要作用。本文提出一种基于改进U-Net(u-shapednetwork)模型的高分影像公路线路提取方法。首先,从功能、路线设计、分级标准和构造及横断面四个方面剖析公路线路特征,厘清公路与道路的区别;然后,联合通道与空间注意力模块,提出了一种改进的U-Net网络模型;最后,以重庆市南岸区和巴南区为研究区,建立数据集并进行公路线路信息提取实验。结果表明,与已有方法相比,精确度、F1分数、IoU指标有明显提升。本文方法在高分辨率影像提取公路信息方面具有可行性与有效性。High-quality highway network information plays an important role in regional economic development,disaster emergency management,and land planning.This research proposed a highway extraction method based on an improved U-Net model from high-resolution images.Firstly,the characteristics of highway are analyzed from four aspects:function,route design,grading standards and structure,and cross-section,and the difference between highway and road is clarified.Then,an improved U-Net network model is proposed by combining channel and spatial attention modules.Finally,the Nanan District and Banan District of Chongqing City were used as the research areas to establish a data set and the highway information extraction experiment was carried out.The results showed that compared with the existing methods,the accuracy,F1 score,and IoU index of the proposed method are significantly improved.This proves the feasibility and effectiveness of the proposed method in highway information extraction from high-resolution images.

关 键 词：高分辨率影像 公路线路提取 改进U-Net模型 通道注意力机制 空间注意力机制 

分 类 号：U412.3[交通运输工程—道路与铁道工程]