计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (29): 160-163.DOI: 10.3778/j.issn.1002-8331.2009.29.048

• 图形、图像、模式识别 • 上一篇    下一篇

协同训练支持向量机对遥感影像的分类研究

李广水1,2,宋丁全1,郑 滔3,李 杨2,苏继申4   

  1. 1.金陵科技学院,南京 211169
    2.南京林业大学 森林资源与环境学院,南京 210037
    3.南京大学 软件学院,南京 210093
    4.南京市园林科学研究所,南京 210037
  • 收稿日期:2009-05-26 修回日期:2009-07-15 出版日期:2009-10-11 发布日期:2009-10-11
  • 通讯作者: 李广水

Research Tri-training SVMS for remote sensing image classification

LI Guang-shui1,2,SONG Ding-quan1,ZHENG Tao3,LI Yang2,SU Ji-shen4   

  1. 1.Jinling Institute of Technology,Nanjing 211169,China
    2.College of Forest Resource and Environment,Nanjing Forestry University,Nanjing 210037,China
    3.Software Institute,Nanjing University,Nanjing 210093,China
    4.Nanjing Institute of Landscape Science,Nanjing 210037,China
  • Received:2009-05-26 Revised:2009-07-15 Online:2009-10-11 Published:2009-10-11
  • Contact: LI Guang-shui

摘要: 协同训练可以提高半监督分类器的分类精度,而如何构建具有冗余特性的训练集是其关键所在。依据遥感影像的纹理特征,提出了基于纹理特征值及像素灰度值构建的两个训练集上协同训练支持向量机的算法CTSVMTRS。仿真实验比较了在不同训练集上CTSVMTRS的分类效果,在叠代训练过程中,两类数据集的所有过程的测试结果都存在的明显差异验证了提出的观念。

关键词: 协同训练, 支持向量机, 遥感图像, 纹理分析, 机器学习

Abstract: Tri-training applied in semi-supervised learning can improve the classification precision,but,how to construct two redundance data sets is the key for Tri-training.With the analysing on texture property of remote sensor image,the algorithm CTSVMTRS for Tri-training SVMS in remote sensing image based on two data sets that one is from pixel value and another is from calculating texture property is presented.In the experiment,the keeping distinction of tested result from different Tri-training SVMS generated from two kinds of data sets in each cycle proves the algorithm is effective.

Key words: Tri-training, Support Vector Machines(SVM), remote sensing image, texture analysis, machine learning

中图分类号: