融合主动学习的高光谱图像半监督分类

发布时间：2018-01-20 06:13

本文关键词： 高光谱图像半监督分类支持向量机主动学习差分进化　出处：《哈尔滨工程大学学报》2017年08期 　论文类型：期刊论文

【摘要】：针对高光谱数据维数高、有标签样本少等特点,采用半监督分类利用未标记样本信息提高高光谱图像分类精度。主动学习研究训练样本的选择方法,以少量的标记样本得到尽可能好的泛化能力。本文提出了一种结合主动学习算法的半监督分类算法。该方法使用支持向量机作为基本的学习模型,通过主动学习方法选取训练样本,以伪标记的形式加入到分类器的训练中,结合验证分类器迭代选出置信度较高的伪标记样本,通过差分进化算法交叉变异伪标记样本扩充标记样本群。在两个数据集上进行仿真实验,与传统分类算法相比,所提算法的总体分类精度分别提高了1.97%、0.49%,表明该算法能够有效地提升主动学习样本选择的效率,在有限带标记样本情况下提高了分类器精度。
[Abstract]:In view of the high dimension of hyperspectral data and the small number of labeled samples, semi-supervised classification is used to improve the accuracy of hyperspectral image classification by using unlabeled sample information. In this paper, a semi-supervised classification algorithm combining active learning algorithm is proposed, which uses support vector machine as the basic learning model. The training sample is selected by active learning method and added to the training of classifier in the form of pseudo-label, and the pseudo-label sample with high confidence is selected by iterative verification classifier. Through the differential evolution algorithm cross-mutation pseudo-label samples to expand the tagged sample group. Two data sets on the simulation, compared with the traditional classification algorithm. The overall classification accuracy of the proposed algorithm is improved by 1.97 and 0.49 respectively, which shows that the algorithm can effectively improve the efficiency of active learning sample selection. The accuracy of the classifier is improved in the case of finite labeled samples.
【作者单位】：哈尔滨工程大学信息与通信工程学院;
【基金】：国家自然科学基金项目(61675051) 黑龙江省自然科学基金项目(F201409)
【分类号】：TP751
【正文快照】： 网络出版地址:http://www.cnki.net/kcms/detail/23.1390.u.20170426.1801.068.html随着高光谱成像技术及相关数据的不断发展,高光谱图像分类技术已成为遥感领域的研究热点问题[1-2]。相对于其他遥感图像,高光谱图像高维非线性、波段间相关性高以及训练样本标记难以获得等特点

【相似文献】