Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (22): 197-204.DOI: 10.3778/j.issn.1002-8331.1708-0045

Previous Articles     Next Articles

3D Gabor based multi-view active learning for hyperspectral image classification

YAO Qiong1, XU Xiang1,2, ZOU Kun1   

  1. 1.College of Computer Engineering, University of Electronic Science and Technology of China, Zhongshan Institute, Zhongshan, Guangdong 528400, China
    2.School of Geography and Planning, Sun Yat-sen University, Guangzhou 510275, China
  • Online:2018-11-15 Published:2018-11-13

基于3D Gabor多视图主动学习的高光谱图像分类

姚  琼1,徐  翔1,2,邹  昆1   

  1. 1.电子科技大学中山学院 计算机学院,广东 中山 528400
    2.中山大学 地理科学与规划学院,广州 510275

Abstract: The acquisition of labeled samples is a difficult issue on hyperspectral remote sensing images classification. The multi-view active learning scheme how to query the most informative samples is carried out. Firstly the 3D Gabor filters with different scales and directions are adopted to extract the spectral and spatial features simultaneously. Then, the obtained 3D Gabor features with stronger discriminant ability are selected to construct multiple views. Finally, a new query strategy based on the Minimum Posteriori Probability Difference(MPPD) for multi-view active learning is proposed. Experiments are conducted on two benchmark hyperspectral image datasets, including ROSIS Pavia University and AVIRIS Indiana Pines, by using the proposed multi-view active learning framework and the MPPD query strategy. In the initial step, 30 labeled samples are used to learn the classifiers, and after 100 iterations, the overall accuracies of Pavia and Indiana datasets reach 94.16% and 91.30% respectively, which shows that the 3D Gabor filters can effectively extract the spectral and spatial features, and generate multiple views with diversity and complementarity. At the same time, the MPPD strategy can query the most valuable samples.

Key words: hyperspectral image classification, multi-view learning, active learning, query strategy, 3D Gabor

摘要: 针对高光谱遥感图像中标记样本获取困难的问题,研究如何选择少量高质量的查询样本进行交互标记的多视图主动学习算法。首先采用不同尺度和方向的三维Gabor滤波器组提取高光谱图像空谱特征;然后挑选出类别判别能力较强的三维Gabor特征来构建多视图;最后提出一种基于多视图后验概率差异最小(MPPD)的样本查询策略。实验初选30个标记样本,经过100次迭代后,三维Gabor特征多视图结合MPPD查询策略在ROSIS Pavia University和AVIRIS Indiana Pines两个数据集上的总体分类精度分别达到94.16%和91.30%,表明通过三维Gabor可以有效提取高光谱遥感图像空谱特征,提供具有多样性和互补性的特征视图。结合MPPD查询策略能挑选出最有价值的查询样本。

关键词: 高光谱图像分类, 多视图学习, 主动学习, 查询策略, 三维Gabor