基于核字典学习的图像分类

摘要

Automatic classification of aerial images is one of the most challengeable tasks due to its high-dimension data and complex context.In order to tackle the problems of high feature dimension and linearly inseparable in original data,this paper proposed a recognition algorithm combining kernel dictionary learning and discriminant analysis based on dictionary learning and sparse representation.First of all,it learned a kernel dictionary that explored the underlying structure of data,then obtained the sparse representations of samples by the kernel dictionary.Secondly,it employed the linear discriminant analysis to make these sparse representations more separable.Finally,it used classical support vector machine for classification.Experimental results show that this method based on kernel dictionary learning and discriminant analysis has superior recognition performance in comparison with the methods based on traditional feature extraction in subspace and dictionary learning.%航拍图像往往具有场景复杂、数据维度大的特点,对于该类图像的自动分类一直是研究的热点.针对航拍原始数据特征维度过高和数据线性不可分的问题,在字典学习和稀疏表示的基础上提出了一种结合核字典学习和线性鉴别分析的目标识别方法.首先学习核字典并通过核字典获取目标样本的稀疏表示,挖掘数据的内部结构;其次采用线性鉴别分析,加强稀疏表示的可分性;最后利用支持向量机对目标进行分类.实验结果表明,与传统基于子空间特征提取的算法和基于字典学习的算法相比,基于核字典学习与鉴别分析的算法分类性能优越.