首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 216 全文下载次数: 161
引用本文:

DOI:

10.11834/jrs.20219209

收稿日期:

2019-06-16

修改日期:

2021-04-30

PDF Free   EndNote   BibTeX
基于改进全卷积网络的高分辨率遥感影像农村道路提取方法
摘要:

目前基于高分辨率遥感影像提取农村道路的研究较少,且既有方法提取的效果欠佳。针对上述问题,提出了一种适用于高分辨率遥感影像的农村道路提取的全卷积网络模型DC-Net(Dilated Convolution Network)。该模型基于全卷积的编解码结构来提取道路深度特征信息,同时针对农村道路细长的特点,在解编码层之间加入了以空洞卷积为基础的ASPP(Atrous Spatial Pyramid Pooling)结构来提取道路的多尺度特征信息,在不牺牲特征空间分辨率的同时扩大了特征感受野(Field-of-View, FOV),从而提高细长农村道路的识别率。以长株潭城市群郊区部分区域为试验区,采用高分二号国产影像为实验数据,并与经典的几种全卷积网络进行对比分析。实验结果表明:本文提出的改进全卷积网络道路提取模型能够有效地提取高分辨率遥感影像中农村道路的细节特征,总体提取效果较好,为农村地区道路自动提取提供了一种新的原创性方法。

Road Extraction in Rural Areas from High Resolution Remote Sensing Image Using a improved Full Convolution Network
Abstract:

At present, there are few studies on rural road extraction based on high-resolution remote sensing images, and the extraction effect of existing methods is not good enough. Aiming at the above problems, a full convolution network model dc-net suitable for rural road extraction with high-resolution remote sensing images is proposed. Decoding structure of the model is based on the convolution depth way to extract characteristic information, at the same time according to the characteristics of rural roads elongated in solving coding layer between joined the empty convolution based ASPP (Atrous Spatial Pyramid Pooling) structure to extract road multi-scale characteristic information, without sacrificing features of Spatial resolution at the same time expand receptive Field (Field-of-View, FOV), improved the recognition rate of slender rural roads. Part of the suburban area of Changsha-zhuzhou-xiangtan Urban Agglomeration was taken as the experimental area, and the GF-2 domestic image was used as the experimental data, which was compared and analyzed with several classical full-convolution networks. The experimental results show that the improved full convolution network road extraction model proposed in this paper can effectively extract the characteristic information of rural roads in high-resolution remote sensing images with good overall effect. It provides a new method for automatic road extraction in rural areas.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮