一种新的两分类器融合的空谱联合高光谱分类方法

本文提出一种两分类器融合的高光谱空谱联合分类方法,首先利用子空间多项式逻辑回归在图像的特征子空间中分类,得到满概率图;根据满概率将每个像元分至概率最大的两个最可信类别,并在原始空间中构建最可信类别字典,利用稀疏解混对每个像元在最可信类别字典下进行稀疏表示,得到稀疏概率图;最后将满概率图和稀疏概率图线性融合,并利用边缘保持的马尔可夫正则项挖掘图像空间信息,得到具有边缘保持的空谱分类模型.实验表明,提出的两分类器融合方法即使在训练样本较少时也比现有方法得到更好的分类结果.

A Novel Two-Classifier Fusion Method for Spectral-Spatial Hyperspectral Classification

This paper presents a new multiple-classifier approach for spectral-spatial classification of hyperspectral images(HSI).Firstly,subspace based multinomial logistic regression(MLRsub) method is used to calculate the full probability of each pixel in the feature space;Secondly,the sub-dictionary is constructed by the training samples of the most two reliable classes,which is determined by the full probability for each pixel.Then,sparse unmixing(SU) is used to calculate the sparse probability in the original HSI.Finally,the full probability and sparse probability are fused linearly and the spatial information is exploit by an edge preserving Markov random field(MRF) regularizer.Experimental results indicate that our proposed multiple-classifier leads to better classification performance than the state-of-the-art methods,even with small training samples.