Robust non-negative supervised low-rank discriminant embedding (NSLRDE) for feature extraction
Robust non-negative supervised low-rank discriminant embedding (NSLRDE) for feature extraction
复制标题
DOI：
10.1007/s13042-022-01752-y
复制
发表时间：
2023-01
影响因子：
5.6
通讯作者：
M. Wan;Chengxu Yan;Tianming Zhan;Hai Tan;Guowei Yang
中科院分区：
文献类型：
--
作者：
M. Wan;Chengxu Yan;Tianming Zhan;Hai Tan;Guowei Yang
在众多的特征提取技术中，非负矩阵分解(NMF)技术忽略了数据的全局表示，而侧重于数据的局部结构信息。然而，全局表示通常比数据噪声更健壮。因此，针对上述问题，结合局部信息数据、全局表示和低阶表示的特点，提出了一种非负监督低阶判别嵌入模型(NSLRDE)，以提高算法的稳健性。该算法将数据分解为清洁数据和噪声数据，并对直通范数进行稀疏约束，增强了对噪声的鲁棒性。此外，该算法利用低阶表示学习和非负分解进一步增强了算法的健壮性。最后，结合图嵌入算法，保留局部和全局数据。我们还将该方法应用于各种噪声数据库，以测试其有效性。
Among many feature extraction technologies, non-negative matrix factorization (NMF) technology ignores the global representation of data and focuses on the local structure information of data. However, the global representation is often more robust than data noise. Therefore, aiming at solving the above problems, combined with the characteristics of local information data, global representation and low-rank representation, a non-negative supervised low-rank discriminant embedding model (NSLRDE) is proposed to improve the robustness of the algorithm. The algorithm decomposes the datainto clean dataand noise data, and sparsely constrainsthrough-norm to enhance the robustness to noise. In addition, the algorithm uses low-rank representation learning and non-negative decomposition to further enhance the robustness of the algorithm. Finally, combined with graph embedding algorithm, local and global data are retained. We also apply the method to various noise databases to test the effectiveness.