基于Fisher字典学习稀疏表示的高光谱图像分类

近年基于稀疏表示的分类框架(Sparse Representation based Classification,SRC)在计算机视觉和模式识别领域取得了巨大成功,高光谱图像解译也逐渐引入稀疏表示方法。针对基于SRC的高光谱图像分类算法随机抽取训练样本构成字典较难捕获相似类别的相对差异性信息问题,提出采用Fisher字典学习方法增强相似类训练样本的可区分性。此外,考虑到高光谱图像具有较强空间相关性的特点,设计一种简单有效的投票策略进行类别判决。大量实验表明:基于Fisher字典学习的联合投票分类方法能够较好地改善高光谱分类精度。

Hyperspectral Image Classification Using Fisher Dictionary Learning based Sparse Representation

Inspired by the recent success of Sparse Representation based Classification (SRC) framework in computer vision and pattern recognition areas,the HyperSpectral Image (HSI) processing community has witnessed a surge of papers focusing on the utilization of sparse prior for effective classification.In sparse representation based on HSI classification,relative difference information between similar classes can hardly be captured for the reason of randomly sampling.We first apply a novel fisher discriminative dictionary learning method,which improves the discriminative and the reconstruction capability of the dictionary.Secondly,motivated by the assumption that spatially adjacent samples are statistically related,a majority voting scheme incorporating contextual information is proposed to predict category labels.Experimental results demonstrated that this method can significantly improve HSI classification accuracy.