基于残差网络结构的道路语义分割  

Road Semantic Extraction Based on Residual Network Structure

作　　者：魏永林 陈宜金[1] 陈俊美 WEI Yonglin;CHEN Yijin;CHEN Junmei(School of Earth Sciences and Surveying Engineering,China University of Mining and Technology,Beijing 100083,China)

机构地区：[1]中国矿业大学(北京)地球科学与测绘工程学院,北京100083

出　　处：《北京测绘》2021年第4期445-448,共4页Beijing Surveying and Mapping

基　　金：中央高校基本科研业务费专项(2010YD06)。

摘　　要：从高分辨率光学遥感影像中提取地物信息受到广泛关注,道路提取是其中一项重要任务。设计了一种基于残差网络结构的深度学习神经网络模型,并将该模型与FCN网络模型和U-Net网络模型输出结果进行了对比。结果表明,本文方法不仅在F 1得分和IOU上均优于其他两个模型,并且在提取的道路平滑度和图像噪声处理方面也具有一定的优势。Extracting mapinformation from high-resolution optical remote sensing images has received widespread attention,and road extraction is one of the important tasks.This paper designed a deep learning neural network model based on residual network,and compared it with FCN network model and U-Net network model.Experimental analysis showed that the method in this paper was superior to the other two models in comparison of extraction results,both in F 1 score and IOU(Intersection Over Union),and has certain advantages in road extraction smoothness and noise processing.

关 键 词：高分辨率遥感影像 道路提取 卷积神经网络 残差网络 

分 类 号：P237[天文地球—摄影测量与遥感]