少样本条件下高光谱图像空-谱特征分析

A Spatial-spectral Feature Analysis of Hyperspectral Images with Limited Samples

在高光谱图像(HSI)分类中,由于标记样本的获取十分耗时耗力,少样本问题一直是该领域的重要研究问题之一.本文先对HSI进行多种空间特征提取,并将这些特征与谱特征融合,以形成多种空-谱特征.然后对多种空-谱特征及其融合进行了实验对比分析.在3个基准HSI数据集上的实验结果表明,在少样本条件下,空-谱特征融合下的HSI分类精度显著高于仅用谱特征的分类精度;多空-谱特征融合方法的分类精度显著优于单一空-谱特征方法的分类精度.

In hyperspectral image(HSI) classification, the problem of limited number of samples has always been one of the important research problems due to the time-consuming and labor-intensive acquisition of labeled samples. In this paper, a variety of spatial features were extracted from HSI, and these features were fused with spectral features to form a variety of space-spectral features. Then, these spatial-spectral features and their fusions were experimentally compared and analyzed. The experimental results on three benchmark HSI datasets show that with limited samples, the classification accuracy of spatial-spectral feature fusion is significantly higher than that of the method with only spectral features, and the classification accuracy of the multi-spatial-spectral feature fusion method is significantly better than that of the single spatial-spectral feature method.