改进的U-Net网络在遥感影像道路提取中的应用  

Application of Improved U-Net Network in Road Extraction from Remote Sensing Images

作　　者：孔祥稳 王常颖 张世超 李劲华 隋毅 KONG Xiangwen;WANG Changying;ZHANG Shichao;LI Jinhua;SUI Yi(School of Computer Science and Technology,Qingdao University,Qingdao,Shandong 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,山东青岛266071

出　　处：《遥感信息》2022年第2期97-104,共8页Remote Sensing Information

基　　金：山东省重点研发计划重大科技创新工程项目(2019JZZY020101);全国统计科学研究项目(2020335)。

摘　　要：针对高分辨率遥感影像中道路目标结构复杂且背景地物多样的问题,设计了一种适用于高分辨率遥感影像道路提取的SM-Unet网络。为捕获孤立道路区域的长距离关系的同时也能关注局部信息,网络编码器下采样前加入条纹池化模块;为增强网络对复杂场景中道路区域上下文信息的获取能力,使道路特征表示更有辨别力,编码器最后卷积层后加入混合池化模块。为验证SM-Unet模型提取道路的能力,选择我国高分二号遥感影像为数据集开展道路提取实验。结果表明,SM-Unet网络训练的道路提取模型在精确率、召回率、F 1分值、平均交并比等评价指标上,均优于U-Net、FCN、DeepLabV3+等网络模型。同时,在道路提取的完整性方面,提取效果最优。Aiming at the problems of complex structure of road targets and diverse background objects in high resolution remote sensing images,a SM-Unet network for road extraction from high resolution remote sensing images is proposed.In order to capture the long-distance relationship of isolated road area and pay attention to local information,the network encoder adds strip pooling module before subsampling.In order to enhance the network’s ability to obtain the context information of the road area in the complex scene and make the road feature representation more discriminative,the encoder finally adds the mixed pooling module after the convolution layer.In order to verify the ability of the SM-Unet model to extract roads,China’s GF-2 remote sensing image is selected as the data set to carry out the road extraction experiment.The results show that the road extraction model trained by the SM-Unet network has high precision,recall,F 1 and IoU.All of them are better than that of U-Net,FCN,DeepLabV3+and other network models.At the same time,in the integrity of road extraction,the extraction effect of the proposed method is the best.

关 键 词：道路提取 高分二号 U-Net 条纹池化模块 混合池化模块 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]