一种改进U-Net的高分辨率遥感影像道路提取方法  

High-resolution Remote Sensing Image Road Extraction Method for Improving U-Net

作　　者：王卓 闫浩文 禄小敏 冯天文 李亚珍 Wang Zhuo;Yan Haowen;Lu Xiaomin;Feng Tianwen;Li Yazhen(Faculty of Geomatics,Lanzhou Jiaotong University,Lanzhou 730070,China;National-Local Joint Engineering Research Center of Technologies and Application for National Geographic State Monitoring,Lanzhou 730070,China;Gansu Provincial Engineering Laboratory for National Geographic State Monitoring,Lanzhou 730070,China;Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]兰州交通大学测绘与地理信息学院,甘肃兰州730070 [2]地理国情监测技术应用国家地方联合工程研究中心,甘肃兰州730070 [3]甘肃省地理国情监测工程实验室,甘肃兰州730070 [4]中国科学院西北生态环境资源研究院,甘肃兰州730000 [5]中国科学院大学,北京100049

出　　处：《遥感技术与应用》2020年第4期741-748,共8页Remote Sensing Technology and Application

基　　金：国家重点研发计划项目(2017YFB0504203);国家自然科学基金项目(41671447、41861055);国家青年基金项目(41801395);中国博士后科学基金(2019M653795);兰州交通大学优秀平台(201806)。

摘　　要：从遥感影像中准确高效地提取道路信息,对基础地理数据库的建立与维护具有重大意义。高分辨率遥感影像背景信息复杂,导致现有算法无法较好地从中提取道路信息。U-Net网络在图像分割方面有较好的实验效果,但道路分割结果准确性不佳,因此,提出了一种改进U-Net网络的高分辨率影像道路提取方法。首先,设计基于U-Net的网络结构,将VGG16作为网络编码结构,可更好地提取特征语义信息;其次,利用Batch Normalization与Dropout解决网络训练过程中出现的过拟合;最后,对训练数据利用旋转与镜像变换进行扩充,采用ELU激活函数,提升了网络训练速度。实验结果表明:该方法可以较为准确高效地提取道路信息。Accurate and efficient extraction of road information based on remote sensing image is a great significance for the establishment and maintenance of basic geographic databases.Due to the complex background information of high-resolution remote sensing images,existing algorithms cannot extract road information very well.U-Net network has good experimental results in image segmentation,but the accuracy of road segmentation results is not good.For this reason,this paper proposes a high-resolution image road extraction method based on improved U-Net network.Firstly,the U-Net-based network structure is designed and implemented.The network uses VGG16 as the network coding structure,which can extract feature semantic information better.Secondly,the use of Batch Normalization and Dropout solves the phenomenon of over-fitting that occurs during the network training process.Finally,the training data is expanded by rotation and mirror transformation,and the ELU activation function is used to improve the network training speed.The experimental results show that the method can extract road information more accurately and efficiently.

关 键 词：高分辨率遥感影像 道路提取 U-Net网络 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]