改进的LLGC高光谱图像半监督分类

发布时间：2018-03-21 11:19

本文选题：半监督分类　切入点：局部全局一致性　出处：《哈尔滨工程大学学报》2017年07期 　论文类型：期刊论文

【摘要】：针对高光谱数据波段多,地物标签获取代价高,带标记的样本数量少,分类过程中容易引起Hudges现象。本文提出一种基于改进的局部全局一致性(learning with local and global consistency,LLGC)算法的半监督分类方法。通过边缘采样法(margin sampling,MS)选取最富含信息量的无标签样本,加入到训练集来扩充训练样本;用KNN算法计算相似度进一步优选无标签样本,去除噪声点和存在的野值点;使用改进的局部全局一致性算法对无标签样本集进行分类标记,得到各类别的分类结果。实验结果表明,本文方法在充分利用无标签样本的情况下,有效地提高了带有少量标签样本的高光谱图像的分类精度。
[Abstract]:In view of the high frequency bands of hyperspectral data, the high cost of acquiring feature labels and the small number of labeled samples, This paper presents a semi-supervised classification method based on the improved local global consistency learning with local and global consistency algorithm. The margin sampling method is used to select the most informative unlabeled samples. Add the training set to expand the training sample; use KNN algorithm to calculate the similarity to select the untagged sample further, remove the noise points and the existing outliers; use the improved local global consistency algorithm to classify the untagged sample set, and label the untagged sample set by using the improved local global consistency algorithm. The experimental results show that the proposed method can effectively improve the classification accuracy of hyperspectral images with a small number of labeled samples under the condition of making full use of unlabeled samples.
【作者单位】：哈尔滨工程大学信息与通信工程学院;江南机电设计研究所;
【基金】：国家自然科学基金项目(60802059) 教育部博士点新教师基金项目(200802171003)
【分类号】：TP751

【相似文献】