The accuracy of semi-supervised hyperspectral image classification generally improves with an increase of the number of labeled pixels.However,acquiring of labeled samples is time-consuming and laborious,and depends on expert knowledge.To address this issue,this paper proposes a new method to generate virtual samples with pseudo-labels.Based on the convex set theory in mathemat-ics,the proposed method can generate a large number of virtual samples with pseudo-labels by using a small number of labeled samples.An effectively expands the training sample set and significantly improves the classification results of semi-supervised classifiers.To verify the validity of the proposed method,extensive tests were performed on two common real hyperspectral data sets,Indi-an Pines and Pavia University.Experimental results show that when the proposed method is used to classify hyperspectral images with a small number of labeled samples,the index values of the three evaluation classification results are significantly improved.
关键词
高光谱图像/虚拟样本/伪标签/半监督分类/凸集
Key words
hyperspectral image/virtual sample/pseudo-label/semi supervised classification/convex set
维普
万方数据