计算机工程

• 图形图像处理 • 上一篇    下一篇

基于块近邻的边界约束标签平滑高光谱图像分类方案

陈善学,桂成名,王一宁   

  1. (重庆邮电大学 移动通信技术重庆市重点实验室,重庆 400065)
  • 收稿日期:2016-10-25 出版日期:2017-11-15 发布日期:2017-11-15
  • 作者简介:陈善学(1966—),男,教授、博士,主研方向为图像处理、数据压缩;桂成名、王一宁,硕士研究生。
  • 基金项目:
    重庆市教委科学技术研究项目(KJ1400416)。

Hyperspectral Image Classification Scheme with Boundary Constrain Label Smoothing Based on Block Neighbor

CHEN Shanxue,GUI Chengming,WANG Yining   

  1. (Chongqing Key Lab of Mobile Communications Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
  • Received:2016-10-25 Online:2017-11-15 Published:2017-11-15

摘要: 为提高高光谱图像分类精度,结合光谱信息、邻域信息和边界信息提出一种高光谱图像分类方案。利用局部费希尔判别分析算法进行降维操作并获取边界信息。根据块近邻分类器算法结合光谱和邻域2个维度获得判决信息。采用边界信息对块近邻分类器算法获得的分类标签进行标签平滑操作。在3个真实地物高光谱数据集上进行实验,结果表明该方案稳定有效地提高了高光谱图像的分类精度。

关键词: 高光谱图像分类, 块近邻分类器, 标签平滑, 边界约束, 局部费希尔判别分析

Abstract: In order to improve the accuracy of hyperspectral image classification,combined with spectral information,neighborhood information and boundary information,this paper proposes a hyperspectral image classification scheme.The method takes the Local Fisher Discriminant Analysis(LFDA) algorithm to reduce the dimension and get the boundary information.The proposed Block Nearest Classifier(BNC) algorithm is used to get the discriminant information with the spectral feature and neighbor feature.The boundary information is used to smooth the classification label obtained from BNC algorithm.Experiment is carried out on hyperspectral dataset of 3 real ground objects.Results show that the proposed scheme improves the classification accuracy of hyperspectral imag effectively and robust.

Key words: hyperspectral image classification, Block Neighbor Classifier(BNC), label smoothing, boundary constrain, local Fisher discriminant analysis

中图分类号: