小样本高光谱图像分类的基于SVM的标记选择方法
In this paper, a new marker selection technique is proposed using SVM over-fitting. The technique is implemented when both spatial and spectral information are extracted for hyperspectral image classification. Spatial information is extracted using Extended Morphological Profile with duality. Nonparametric feature extraction techniques are used to reduce the redundant and irrelevant information from the spatial and spectral information. VLTSA is performed on the hyperspectral image. One to ten training samples per class are examined and it is concluded that better classification accuracy is obtained when NWFE is used as FE. It is also investigated that even using very low training samples still a reasonably fine classifications can be obtained by using the marker selection method of SVM over-fitting.
高光谱遥感影像 图像分类 标记选择 支持向量机
Farid Muhammad Imran 何明一
国内会议
西安
中文
126-131
2015-09-01（万方平台首次上网日期，不代表论文的发表时间）