测绘通报 ›› 2020, Vol. 0 ›› Issue (9): 110-113.doi: 10.13474/j.cnki.11-2246.2020.0294

• 技术交流 • 上一篇    下一篇

深度学习在高分辨率遥感影像冬油菜提取中的应用

杨泽宇1, 张洪艳2, 明金1, 冷伟1, 刘海启3, 游炯3   

  1. 1. 武汉珈和科技有限公司, 湖北 武汉 430000;
    2. 武汉大学测绘遥感信息工程国家重点实验室, 湖北 武汉 430000;
    3. 农业农村部耕地利用遥感重点实验室/农业农村部规划设计研究院, 北京 100000
  • 收稿日期:2019-12-27 修回日期:2020-03-16 出版日期:2020-09-25 发布日期:2020-09-28
  • 通讯作者: 张洪艳。E-mail:zhanghongyan@whu.edu.cn E-mail:zhanghongyan@whu.edu.cn
  • 作者简介:杨泽宇(1987-),女,硕士,工程师,从事农业遥感研究工作。E-mail:yangzywhu@sina.cn
  • 基金资助:
    农业农村部耕地利用遥感重点实验室开放课题(2019LCLU002)

Extraction of winter rapeseed from high-resolution remote sensing imagery via deep learning

YANG Zeyu1, ZHANG Hongyan2, MING Jin1, LENG Wei1, LIU Haiqi3, YOU Jiong3   

  1. 1. Wuhan Jiahe Technology Co., Ltd., Wuhan 430000, China;
    2. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430000, China;
    3. Key Laboratory of Cultivated Land Use, MARA/Academy of Agricultural Planning and Engineering, MARA, Beijing 100000, China
  • Received:2019-12-27 Revised:2020-03-16 Online:2020-09-25 Published:2020-09-28

摘要: 近年来,深度学习在基于高分辨率遥感影像的农作物种植信息提取领域应用广泛。本文充分利用油菜在盛花期的光谱特征,提出了基于深度学习理论的单时相高分辨率遥感影像油菜分布提取方法。以2016年湖北省沙洋县作为研究区域,获取油菜盛花时期高分一号(GF-1)影像,并以沙洋县为基础影像,通过手工标记制作油菜训练样本。设计两种深度学习框架模型,一种以卷积神经网络(CNN)为框架,构建一维卷积神经网络(1D-CNN)模型,第二种以循环神经网络(RNN)为框架,组合门控循环单元(GRU)模型,训练标准样本模型,完成油菜分类提取。最后,与传统支持向量机(SVM)、随机森林(RF)方法进行了结果对比。试验结果表明,本文设计的基于深度学习CNN和RNN模型提取的冬油菜空间分布精度和面积精度皆优于其他两种方法,为进一步实现冬油菜提取自动化提供试验基础。

关键词: 油菜提取, 深度学习, 卷积神经网络(CNN), 循环神经网络(RNN), 高分一号

Abstract: Recently, deep learning technology has been widely used in the crop extraction from high-resolution remote sensing data. This paper exploits the use of the spectral characteristics of rapeseed in the flowering period, and proposes a rapeseed extraction method from single-phase high-resolution remote sensing image based on deep learning theory. The paper employs GF-1 satellite data during the rapeseed flowering period, Shayang City, Hubei Province, as the research data. Firstly, rapeseed training samples are annotated on the image by manual labeling. Then, two deep learning framework models are built, including a one-dimensional convolutional neural network (1D-CNN) and a recurrent neural network (RNN), to conduct the rapeseed extraction with the help of annotated training samples. Finally, the rapeseed extraction results are evaluated by comparing with traditional support vector machine (SVM) and random forest (RF) methods. The experimental results show that the spatial distribution accuracy and area accuracy of winter rapeseed based on deep learning CNN and RNN models designed in this paper are better than the other two methods, which provides guidance for automation of large area winter rapeseed extraction.

Key words: extraction of winter rapeseed, deep learning, CNN, RNN, GaoFen-1

中图分类号: