主办单位:中国气象局沈阳大气环境研究所
国际刊号:ISSN 1673-503X
国内刊号:CN 21-1531/P

气象与环境学报 ›› 2019, Vol. 35 ›› Issue (3): 87-93.doi: 10.3969/j.issn.1673-503X.2019.03.011

• 简报 • 上一篇    下一篇

基于深度学习的FY3D/MERSI和EOS/MODIS云检测模型研究

瞿建华, 鄢俊洁, 薛娟, 郭雪星   

  1. 北京华云星地通科技有限公司, 北京 100081
  • 收稿日期:2019-01-14 修回日期:2019-02-22 出版日期:2019-06-30 发布日期:2019-06-28
  • 作者简介:瞿建华,男,1976年生,高级工程师,主要从事气象卫星遥感应用方面的研究,E-mail:qujh@cma.gov.cn。
  • 基金资助:
    风云三号(02)批地面应用系统"FY-3湿地遥感监测评价应用示范(FY-3(02)-UDS-1.7.1)"项目资助。

Research on the cloud detection model of FY3D/MERSI and EOS/MODIS based on deep learning

QU Jian-hua, YAN Jun-jie, XUE Juan, GUO Xue-xing   

  1. Beijing Huayun Shinetek Science and Technology Co., Ltd., Beijing 100081, China
  • Received:2019-01-14 Revised:2019-02-22 Online:2019-06-30 Published:2019-06-28

摘要: 针对FY3D/MERSI和EOS/MODIS的云检测问题,提出了一种基于深度学习技术的全自动云检测算法,首次将深度学习引入到卫星影像云检测领域。本算法使用深度全卷积神经网络(Deep Convolutional Neural Networks)作为核心结构,基于EOS/MODIS基本云检测原理选择合适的通道作为特性向量参数,针对不同的场景进行分类和网络模型的训练,最终得到基于深度学习的云检测模型。经过EOS/MODIS数据和FY3D/MERSI数据的测试,云检测的精度达到98%以上,可以看出基于深度学习的云检测算法能够用于云检测,该算法具有效率高、精度高等特点,云检测效果理想。

关键词: FY3D/MERSI, EOS/MODIS, 云检测, 卷积神经网络, 深度学习

Abstract: Based on deep learning technology,a fully automatic cloud detection algorithm for FY3D/MERSI and EOS/MODIS image is proposed and is firstly introduced in cloud detection study.Using Deep Convolutional Neural Networks as core structure and choosing some appropriate channels as characteristic parameters based on MODIS basic cloud detection theory,classifying and network model training are conducted according to the different scenarios.As a result,cloud detection model based on deep learning is setup and its cloud detection accuracy is over 98% by testing EOS/MODIS and FY3D/MERSI data,indicating that the algorithm can be applied in cloud detection with the characteristics of high efficiency and high accuracy as well as ideal cloud detection effect.

Key words: FY3D/MERSI, EOS/MODIS, Cloud detection, Convolutional neural network, Deep learning

中图分类号: