结合局部全局一致性和支持向量机的半监督分类方法  

Semi-supervised classification method combining local and global consistency and support vector machine

作　　者：池辛格 王立国[1] CHI Xinge;WANG Liguo(College of Information and Communications Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2021年第1期48-54,共7页Applied Science and Technology

基　　金：国家自然科学基金项目(62071084)。

摘　　要：半监督分类方法通过提取无标签样本的信息,结合有限的有标签样本,克服了高光谱图像准确地物标签样本不足的问题,有效提高了图像的分类精度。局部全局一致性算法是一种基于图的标签传递半监督分类方法,具有性能良好,易于求解以及能够有效反映所有样本间关系的优点,但是其分类结果极不稳定,不利于实际应用。支持向量机是高光谱图像的分类领域应用最广的监督分类方法,拥有较强的鲁棒性,但是由于高光谱图像中有标签的样本不一定能够代表该类地物的全部特征,其分类结果也有可能出现波动性较大的问题,分类精度不理想。因此本文提出了一种结合局部全局一致性和支持向量机的半监督分类算法,通过迭代不断提取两种算法中分类结果相同的部分扩充有标签样本集,然后通过支持向量机进行分类,大幅度提高了分类精度和稳定度。The semi-supervised classification method overcomes the problem of insufficient accurate object label samples for hyperspectral images by extracting information from unlabeled samples and combining with limited labeled samples,and effectively improves the classification accuracy of images.The local global consensus algorithm is a semisupervised classification method based on the label transfer.It has the advantages of good performance,easy to solve and can effectively reflect the relationship among all samples,but the randomness of the label transfer process makes its classification results extremely unstable,and not conducive to practical applications.Support vector machine is the most widely used supervised classification method in the field of hyperspectral image classification.It has strong robustness.However,because the labeled samples in hyperspectral images may be unable to represent all the features of such objects.As a result,there may be problems with greater volatility,and the classification accuracy is not ideal.Therefore,this paper proposes a semi-supervised classification algorithm that combines local global consistency and support vector machines.It iteratively extracts the partially expanded labeled sample set with the same classification result in the two algorithms,and then classifies them through the support vector machine.This method greatly improves the classification accuracy and stability.

关 键 词：高光谱 高光谱图像 高光谱图像处理 高光谱图像分类 半监督分类 机器学习 局部全局一致性 支持向量机 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置] TP751[自动化与计算机技术—控制科学与工程]