摘要
特征提取和特征选择是模式识别的关键问题之一,它影响到分类器的设计及其性能.高光谱图像数据是超高维多特征数据集,如何实现高维特征空间的特征压缩和特征提取是一个重要课题.基于高光谱图像谱图合一、数据维度高的数据结构特点,该文从光谱和图像两个层面分别综述了主成分分析、最小噪声分离、独立成分分析等光谱特征提取方法以及基于颜色、纹理、形状等图像特征提取方法.还详细介绍了核主成分分析和投影寻踪方法这两种高光谱特征提取新方法,并给出了以上方法的应用实例.特征提取和特征选择的研究将为后续的高光谱图像分类奠定良好的基础.
Feature extraction and feature selection is an important subject in pattern recognition, which affects the performance of the classifier. Based on the character of high dimensional hyperspectral image, a key factor is how to reduce the dimension. Usually dimension reduction can be processed on two aspects, spectral and image. In this paper, different methods are summarized which in- clude the principal component analysis, the minimum noise separation, independent component analysis, texture, shape, etc. The article also introduced some new hyper-spectral feature extraction method for the kernel principal component analysis and projection pursuit method. Meanwhile the application of the methods mentioned above is instantiated. The research on feature extraction and feature selection will be the most essential premise for the hyper-spectral image classification.
出处
《广西师范学院学报（自然科学版）》
2015年第2期39-43,共5页
Journal of Guangxi Teachers Education University(Natural Science Edition)
基金
广西教育厅项目(201203YB103)
关键词
高光谱图像
特征提取
特征选择
主成分分析
最小噪声分离
独立成分分析
核主成分分析
投影寻踪
hyper-spectral image
feature extraction
feature selection
principal component analysis
the minimum noise separation
independent component analysis
kernel principal component analysis
projection pursuit
