摘要
针对传统降维算法处理后分类识别精度不高的问题,提出一种双重L2,1稀疏表示降维算法。将分组稀疏表示的思想引入高光谱降维算法中,利用基于L2,1范数约束的稀疏表示方法重构高光谱图像,获得特征强化的高光谱数据,对强化后的数据进行基于L2,1范数约束的稀疏表示,将稀疏系数矩阵作为高维数据的低维表示。在Indian Pines数据集上的分类结果表明,总体分类精度能够达到94.33%。该算法利用L2,1稀疏表示的结果组内稠密组间稀疏的特点,强化特征的同时提高了低维数据的分类识别效果。
The idea of grouped sparse representation was introduced into the hyperspectral dimensionality reduction algorithm to solve the problem of low accuracy when using traditional dimensionality reduction algorithm.The hyperspectral image was reconstructed using the sparse representation based on L2,1 parametric constraints to obtain the feature-enhanced hyperspectral data,and the enhanced data were subjected to the sparse representation based on L2,1 parametric constraints,and the sparse coe-fficient matrix was used as the low-dimensional representation of the high-dimensional data.The classification results on the Indian Pines dataset show that the overall classification accuracy can reach 94.33%.The algorithm takes advantage of the intra-group dense inter-group sparsity of the resultant L2,1 sparse representation to enhance the features while improving the classi-fication recognition of low-dimensional data.
作者
董隽硕
吴玲达
DONG Jun-shuo;WU Ling-da(Complex Electronic System Simulation Laboratory,Space Engineering University,Beijing 101416,China)
出处
《计算机工程与设计》
北大核心
2023年第4期1122-1128,共7页
Computer Engineering and Design
基金
国家自然科学基金项目(61801513)。
关键词
分组稀疏
高光谱
L2
1范数
稀疏表示
稀疏重构
group sparsity
hyperspectral
L2,1 norm
sparse representation
sparse reconstruction
