基于二级字典的联合稀疏表示高光谱图像分类

针对多数传统分类算法应用于高光谱分类存在的分类精度较低、光谱信息利用不充分的问题,在基于核函数的联合稀疏表示分类方法的基础上提出了一种基于二级字典的联合稀疏表示的高光谱分类算法。在字典原子前加入待测像元与该原子的引力,以达到更快捷地找到与待测像元相匹配的原子的目的。加入的引力值由万有引力公式改进的适应于高光谱图像的公式计算而来。为了使得稀疏重构后的残差波段中包含的具有一定意义的分类鉴别信息被充分挖掘,本文采用指数平滑公式对残差信息进行再利用。通过在Indian Pine数据集和Salina-A数据集上进行实验,验证了所提算法可以提升分类精度。

Joint sparse representation of hyperspectral image classification based on secondary dictionary

In view of the problems of low classification accuracy and insufficient utilization of spectral information in most traditional classification algorithms when applied in hyperspectral classification, based on the joint sparse representation of hyperspectral image classification based on kernel function, the joint sparse representation of hyperspectral image classification based on secondary dictionary is proposed. The gravitation between the pixel to be measured and the atom is added in front of the dictionary atom, so as to find the atom matching with the pixel to be measured more quickly. The added gravitational value is calculated by the formula adapted to the hyperspectral image modified by the gravitational formula. In order to explore the meaningful classification and identification information contained in the residual band after sparse reconstruction fully, this paper reuses the residual information is reused by using exponential smoothing formula. Experimental results in Indian Pines and Salina-A data sets show that the proposed algorithm achieves the purpose of improving the classification accuracy of hyperspectral images.