摘要
针对遥感图像数据具有的高维数、非线性以及海量无标记样本的特性,提出了一种基于混合熵和L1范数的概率型最小二乘支持向量机分类方法.将准熵和熵差分融合,构造一种混合熵用以从海量无标记样本集中选出最有"价值"的待标记样本;基于L1范数距离度量,进一步从待标记样本集中筛选出孤立点和冗余点加以剔除;基于初始已标记样本以及筛选得到的样本,训练得到概率型最小二乘支持向量机.对反射光学系统的成像光谱仪(ROSIS)高光谱遥感图像进行了分类实验.结果表明:所提分类方法的总精度和Kappa系数分别达到了89.90%和0.868 5,能够以较少的训练样本得到较高的分类精度,其更适于处理遥感图像分类问题.
Aiming at remote sensing image data having properties of high-dimension, nonlinearity, and massive unlabeled samples, a kind of probability least squares support vector machine （PLSSVM） classification method based on hybrid entroy and Ll-norm was proposed. At first, a hybrid entroy was designed by combining quasi-entropy with entropy difference, which was used to select the most ＇valuable＇ unlabeled samples from the massive unlabeled sample set. In the second step, a L^-norm distance mectric was used to further select and to remove outliers and redundant data from the most ＇valuable＇ unlabeled samples. At last, the original labeled and the selected unlabeled samples were adopted to train the PLSSVM. Experimental results on ROSIS hyperspectral remote sensing image show that the overall accuracy and Kappa coeffi- cient of the proposed classification method reach 89.90% and 0. 868 5 respectively. The pro- posed method can obtain higher classification accuracy with few training samples, which is much applicable for classification problem of remote sensing image.
出处
《中国矿业大学学报》
EI
CAS
CSCD
北大核心
2012年第6期971-977,共7页
Journal of China University of Mining & Technology
基金
国家自然科学基金项目(60804022
60974050
61072094)
教育部新世纪优秀人才支持计划项目(NCET-08-0836
NCET-10-0765)
霍英东教育基金会青年教师基金项目(121066)
关键词
遥感图像
混合熵
L1范数
主动学习
概率型最小二乘支持向量机
remote sensing image
hybrid entropy
L1 norm
active learning
probability leastsquares support vector machine
