测绘通报 ›› 2018, Vol. 0 ›› Issue (4): 57-62.doi: 10.13474/j.cnki.11-2246.2018.0110

• 行业观察 • 上一篇    下一篇

样本迁移支持下的遥感影像自动分类方法

林聪1,2, 李二珠1,2, 杜培军1,2   

  1. 1. 南京大学卫星测绘技术与应用国家测绘地理信息局重点实验室, 江苏 南京 210023;
    2. 南京大学江苏省地理信息技术重点实验室, 江苏 南京 210023
  • 收稿日期:2017-07-17 出版日期:2018-04-25 发布日期:2018-05-03
  • 通讯作者: 杜培军。E-mail:dupjrs@126.com E-mail:dupjrs@126.com
  • 作者简介:林聪(1994-),男,硕士生,主要研究方向为遥感图像处理、长时间序列遥感应用。E-mail:lcnjucumt@126.com
  • 基金资助:

    国家自然科学基金重点基金(41631176);中国地质调查局三峡后续工作科研项目(0001792015CB50002)

An Automatic Approach for Remote Sensing Classification Supported by Sample Transfer

LIN Cong1,2, LI Erzhu1,2, DU Peijun1,2   

  1. 1. Key Laboratory for Satellite Mapping Technology and Applications of State Administration of Surveying, Mapping and Geoinformation of China, Nanjing University, Nanjing 210023, China;
    2. Jiangsu Provincial Key Laboratory of Geographic Information Science and Technology, Nanjing University, Nanjing 210023, China
  • Received:2017-07-17 Online:2018-04-25 Published:2018-05-03

摘要:

迁移学习是运用已有知识对相关的不同领域的问题进行求解的一种机器学习方法,本文结合这一方法,提出了一种基于先验知识的样本自动选取方法,并构建了一套土地覆盖自动分类的算法框架。该方法主要面向Landsat数据,通过图像变化检测技术与光谱形状编码的方法,从源领域中迁移适用的地物类别知识并标记在目标影像中,使用SVM完成基于样本迁移的自动分类流程。结果表明,该方法可以获得可靠的自动分类结果,一定程度上满足遥感信息的大范围提取与长时间序列处理分析的发展需求。

关键词: 迁移学习, 土地覆盖, 变化检测, 光谱形状编码, 自动分类

Abstract:

This paper proposes a novel automatic classification approach to acquire land cover maps by classifying remote sensing images in the time series.Different from other sample collection methods,the proposed method tries to define a precise training set for target domain automatically by transfer learning.This method is proposed for Landsat TM images.This is done by change detection method and spectral curve shape vector.Firstly,the unchanged labels are located by change vector analysis.Then the prior class knowledge from source domain is transferred to the target images,taking advantage of the already available knowledge on the land cover products related to source images.Finally,the target image is classified by support vector machine.The result shows that the approach is effective in automatically obtaining land-cover classification maps.

Key words: transfer learning, land cover, change detection, spectral curve shape vector, automatic classification

中图分类号: