融合光谱度量标记迁移和Tri-training的高光谱遥感图像半监督分类算法  

Semi-supervised classification algorithm for hyperspectral remote sensing images fusing spectral measure-based label transfer and tri-training

作　　者：曹峰[1] 李文涛 骆剑承 李德玉[1,3] 钱宇华[1,3,4] 白鹤翔[1] 张超[1] CAO Feng;LI Wentao;LUO Jiancheng;LI Deyu;QIAN Yuhua;BAI Hexiang;ZHANG Chao(School of Computer and Information Technology(School of Big Data),Shanxi University,Taiyuan 030006,China;State Key Laboratory of Remote Sensing Science,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China;Key Laboratory of Computational Intelligence and Chinese Information Processing,Ministry of Education,Shanxi University,Taiyuan 030006,China;Institute of Big Data and Industry,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西大学计算机与信息技术学院(大数据学院),山西太原030006 [2]中国科学院空天信息创新研究院遥感科学国家重点实验室,北京100101 [3]山西大学计算智能与中文信息处理教育部重点实验室,山西太原030006 [4]山西大学大数据与产业研究院,山西太原030006

出　　处：《大数据》2023年第6期72-89,共18页Big Data Research

基　　金：国家自然科学基金资助项目(No.62072291,No.42071316,No.62072294,No.61672332,No.41871286,No.61806116);山西省重点研发计划项目(No.201903D421003,No.201903D421041);山西省教育厅科技成果转化培育项目(No.2020CG001)。

摘　　要：针对海量的高光谱遥感图像光谱和丰富的空间信息中可用于分类的有标记样本远少于无标记样本的数据特性,提出了一种融合光谱度量标记迁移和Tri-training的高光谱遥感图像半监督光谱-空间分类算法。该算法提出了一种基于光谱度量的标记迁移方法,通过结合迁移标记和Tri-training预测标记进行扩充样本标记预测,提高了扩充样本标记的准确性。同时,该算法基于空间相关性选择扩充样本,综合运用光谱和空间特征提升图像分类的精度。在两个公开的高光谱遥感图像数据集上进行了实验,结果表明该算法优于基于Tri-training算法的高光谱遥感图像的分类性能。Aimed at the problem that a large number of hyperspectral remote sensing images were rich in spectral and spatial information,and the labeled samples available for image classification were far less than unlabeled samples,a semisupervised spectral-spatial classification algorithm was proposed by fusing spectral measure-based label transfer and Tri-training.A spectral measure-based label transfer method was proposed for our algorithm.The transferred labels and predicted labels for Tri-training algorithm were used to predict the labels of expanded unlabeled samples,which can promoted the prediction accuracies of labels for expanded unlabeled samples.Meanwhile,our algorithm selectel expanded samples based on spatial correlation,and used spectral and spatial features to improve the accuracy of image classification.Experimental study was executed on two public hyperspectral remote sensing image datasets,and the results showed that the proposed algorithm outperform tri-training algorithm.

关 键 词：高光谱图像分类 半监督分类 纹理特征 光谱度量 Tri-training算法 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]