基于残差网络约束的道路提取

Road Extraction Based on Residual Network Constraints

遥感图像的道路识别是一个重要研究课题,对GIS平台构建和数据更新有着重大贡献。然而,遥感图像中地表信息丰富,道路建设复杂,这给提取道路造成了一定的困难。本文基于残差网络模块,从以下两方面进行研究:1)构建了数据集,其包含城市、乡村道路,并运用镜像、旋转等手段对数据集进行了扩充;2)针对ResNet提出了一种改进方法,并利用数据集对网络进行了样本训练和特征学习,进行道路提取。

Road recognition in remote sensing images is an important research topic,which makes a great contribution to the construction of GIS platform and data update.However,the remote sensing image is rich in surface information and complex road construction,which causes some trouble for road extraction.Based on the residual network module,this paper studies from the following two aspects:1) construct the data set,which includes urban and rural roads,and expand the data set by means of mirror image and rotation;2) propose an improved method for ResNet,and use the data set to train samples and learn features of the network for road extraction.