测绘学报 ›› 2019, Vol. 48 ›› Issue (3): 330-338.doi: 10.11947/j.AGCS.2019.20180005

• 摄影测量学与遥感 • 上一篇    下一篇

基于Encoder-Decoder网络的遥感影像道路提取方法

贺浩1, 王仕成1, 杨东方1, 王舒洋2, 刘星1   

  1. 1. 火箭军工程大学控制工程系, 陕西 西安 710025;
    2. 火箭军工程大学信息工程系, 陕西 西安 710025
  • 收稿日期:2018-01-08 修回日期:2018-11-15 出版日期:2019-03-20 发布日期:2019-04-10
  • 通讯作者: 杨东方 E-mail:yangdf301@163.com
  • 作者简介:贺浩(1991-),男,博士生,研究方向为深度学习和计算机视觉。E-mail:hehao209@126.com
  • 基金资助:
    国家自然科学基金(61403398;61673017);陕西省自然科学基金面上项目(2017JM6077)

An road extraction method for remote sensing image based on Encoder-Decoder network

HE Hao1, WANG Shicheng1, YANG Dongfang1, WANG Shuyang2, LIU Xing1   

  1. 1. The Rocket Force University of Engineering, The Department of Control Engineering, Xi'an 710025, China;
    2. The Rocket Force University of Engineering, The Department of Information Engineering, Xi'an 710025, China
  • Received:2018-01-08 Revised:2018-11-15 Online:2019-03-20 Published:2019-04-10
  • Supported by:
    The National Nature Science Foundation of China (Nos. 61403398;61673017);The General Project of Shaanxi Nature Science Foundation (No. 2017JM6077)

摘要: 针对道路目标特点,设计实现了用于遥感影像道路提取的Encoder-Decoder深度语义分割网络。首先,针对道路目标局部特征丰富、语义特征较为简单的特点,设计了较浅深度、分辨率较高的Encoder-Decoder网络结构,提高了分割网络的细节表示能力。其次,针对遥感影像中道路目标所占像素比例较小的特点,改进了二分类交叉熵损失函数,解决了网络训练中正负样本严重失衡的问题。在大型道路提取数据集上的试验表明,所提方法召回率、精度和F1-score指标分别达到了83.9%、82.5%及82.9%,能够完整准确地提取遥感影像中的道路目标。所设计的Encoder-Decoder网络性能优良,且不需人工设计提取特征,因而具有良好的应用前景。

关键词: 遥感, 道路提取, 深度学习, 语义分割, 编解码网路

Abstract: According to the characteristics of the road features, an Encoder-Decoder deep semantic segmentation network is designed for road extraction of remote sensing images. Firstly, as the features of the road target are rich in local details and simple in semantic features, an Encoder-Decoder network with shallow layers and high resolution is designed to improve the ability of representing detail information. Secondly, as the road area is small proportion in remote sensing images, the cross-entropy loss function is improved, which solves the imbalance between positive and negative samples in training process. Experiments on large road extraction dataset show that, the proposed method gets the recall rate 83.9%, precision 82.5% and F1-score 82.9%, which can extract the road targets in remote sensing images completely and accurately. The Encoder-Decoder network designed in this paper performs well in road extraction task and needs less artificial participation, so it has a good application prospect.

Key words: remote sensing, road extraction, deep learning, semantic segmentation, Encoder-Decoder network

中图分类号: