摘要： 为提高高光谱遥感影像在训练样本不足时的分类精度，提出一种基于线性邻域传播的改进加权K近邻算法。采用线性邻域传播(LNP)算法获取无标签数据属于各类别的概率，将其作为类别信息，以增加训练样本数量，提高K近邻算法的分类效果，并降低错误分类带来的风险。实验结果表明，对于高光谱遥感影像，该算法具有较好的分类效果，优于传统的KNN算法、距离加权KNN算法以及LNP等半监督分类算法。

关键词: 半监督学习, 高光谱遥感, 分类, 线性邻域传播, 加权K近邻, 流形学习

Abstract: To improve the classification accuracy of hyperspectral remote sensing image when lack of training data, this paper proposes a Weighted K-nearest Neighbor(WKNN) algorithm based on Linear Neighborhood Propagation(LNP). In order to increase the number of training data and improve the classification accuracy, it obtains the unlabeled datas’ probability for each class by LNP algorithm. By this, it can drop the misclassification risk of LNP. Experimental results show that this algorithm has a better performance than other supervised classification algorithms like K-nearest Neighbor(KNN) algorithm, distance WKNN algorithm, and LNP semi-supervised classification algorithm.

Key words: semi-supervised learning, hyperspectral remote sensing, classification, Linear Neighborhood Propagation(LNP), Weighted K-nearest Neighbor(WKNN), manifold learning

中图分类号: 

• TP301.6