计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (18): 166-172.DOI: 10.3778/j.issn.1002-8331.1805-0427

• 图形图像处理 • 上一篇    下一篇

基于深度贝叶斯主动学习的高光谱图像分类

杨承文,李吉明,杨东勇   

  1. 1.浙江工业大学 信息工程学院,杭州 310023
    2.浙江警察学院,杭州 310053
  • 出版日期:2019-09-15 发布日期:2019-09-11

Active Learning for Hyperspectral Image Classification with Deep Bayesian

YANG Chengwen, LI Jiming, YANG Dongyong   

  1. 1.College of Information Engineering, Zhejiang University of Technology, Hangzhou 310023, China
    2.Zhejiang Police College, Hangzhou 310053, China
  • Online:2019-09-15 Published:2019-09-11

摘要: 针对高光谱图像分类中标记样本获取费时费力,无标记数据难以得到有效利用以及主动学习与深度学习结合难等问题,结合贝叶斯深度学习与主动学习的最新进展,提出一种基于深度贝叶斯的主动学习高光谱图像分类算法。利用少量标记样本训练一个卷积神经网络模型,根据与贝叶斯方法结合的主动学习采样策略从无标记样本中选择模型分类最不确定性的样本,选取的样本经人工标记后加入到训练集重新训练模型,减小模型不确定性,提高模型分类精度。通过PaviaU高光谱图像分类的实验结果表明,在少量的标记样本下,提出的方法比传统的方法分类效果更好。

关键词: 高光谱遥感图像, 贝叶斯深度学习, 主动学习, 分类

Abstract: In order to solve the problems that the acquisition of labeled samples are time-consuming and difficult for hyperspectral remote sensing images classification, unlabeled samples have not been used efficiently, and active learning cannot combine well with deep learning. Bayesian active learning hyperspectral image classification algorithm is proposed, arrcoding to the latest advances in Bayesian deep learning and active learning. A convolutional neural network model is trained with a small number of labeled samples, and the most uncertain samples are selected from the unlabeled samples according to the active learning sampling strategy combined with the Bayesian method. The selected samples are added to the training set to update the model. Then model uncertainty will be reduced and the model classification effect will be improved. The experimental results of PaviaU hyperspectral image classification show that using a small number of labeled samples, the proposed method performs better than the traditional method.

Key words: hyperspectral remote sensing image, Bayesian deep learning, active learning, classification