基于迁移学习的卫星云图云分类
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金资助项目(61503192);江苏省自然科学基金资助项目(BK20161533;BK20131002);江苏省六大人才高峰(2014-XXRJ-007);大学生实践创新训练计划项目(201610300088)


Satellite imagery cloud classification based on transfer learning
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 文章评论
  • |
  • 资源附件
    摘要:

    针对在基于机器学习的云图识别中,由于不存在公认的云分类样本库的现实条件下,带来的训练样本数量不足和不平衡,从而难以获得可靠的分类模型的问题,利用迁移学习中的多源加权Tradaboost算法(内部采用极限学习机作为分类器)来进行卫星云图云的检测。利用多人(多源)标注的大量厚云的样本,构成多源辅助样本集;利用少量标注的薄云样板构成目标样本集。使用迁移学习和辅助样本集,对仅在薄云样本集下的训练获得的极限学习机分类器进行辅助训练,提高其薄云识别率。基于国家卫星气象中心的HJ-1A/B的卫星数据实验结果表明,迁移学习可以充分利用容易获得的大样本厚云辅助样本知识,对同类型有关联的小样本薄云分类器进行识别提高。实验表明,迁移学习算法可以进一步用于更多多源样本和其他云分类的任务。

    Abstract:

    Cloud fraction is the basis for the application of meteorological satellites.Compared with traditional methods,the existing methods based on Machine Learning(ML) showing better performance in the utilization of all of the characteristics and optical parameters of the satellite cloud.However,at present there has been no cloud classification standard large sample set,which is a critical problem in this area.This forces researchers to build their own small databases,but in the process of building it,samples of some categories are difficult to label,while those of some relative categories are easy to label,thus the distribution of the sample set is complex and uneven,causing the ML to inhibit cloud detection and cloud fraction performance.In order to resolve this problem,in this paper,based on NSMC's HJ-1A/B satellite imagery data,the Transfer Learning (TL) approach is used for cloud detection.More specifically,the researchers labeled a large number of thick cloud samples,built six sets,and used these as auxiliary sample sets;they then labeled a small number of thin cloud samples,built one set,and used this as a task sample set.The results of the study show that TL,when combined with Extreme Learning Machine (ELM),can utilize the knowledge of the sample sets' auxiliary thick cloud to improve the ELM's thin cloud identification accuracy,which was only trained by the thin cloud sample set.This shows that the TL algorithm would be a good choice in a greater number of categories of cloud classification work when the sample set is complex and uneven.

    参考文献
    相似文献
    引证文献
引用本文

胡凯,严昊,夏旻,徐同,胡伟,徐春燕,2017.基于迁移学习的卫星云图云分类[J].大气科学学报,40(6):856-863. HU Kai, YAN Hao, XIA Min, XU Tong, HU Wei, XU Chunyan,2017. Satellite imagery cloud classification based on transfer learning[J]. Trans Atmos Sci,40(6):856-863. DOI:10.13878/j. cnki. dqkxxb.20170106002

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2017-01-06
  • 最后修改日期:2017-04-08
  • 录用日期:
  • 在线发布日期: 2017-12-27
  • 出版日期:

地址:江苏南京宁六路219号南京信息工程大学    邮编:210044

联系电话:025-58731158    E-mail:xbbjb@nuist.edu.cn    QQ交流群号:344646895

大气科学学报 ® 2024 版权所有  技术支持:北京勤云科技发展有限公司