• Current Optics and Photonics
• /
• v.5 no.4
• /
• pp.431-443
• /
• 2021

Sparse unmixing has been proven to be an effective method for hyperspectral unmixing. Hyperspectral images contain rich spectral and spatial information. The means to make full use of spectral information, spatial information, and enhanced sparsity constraints are the main research directions to improve the accuracy of sparse unmixing. However, many algorithms only focus on one or two of these factors, because it is difficult to construct an unmixing model that considers all three factors. To address this issue, a novel algorithm called multiview-based spectral weighted and low-rank row-sparsity unmixing is proposed. A multiview data set is generated through spectral partitioning, and then spectral weighting is imposed on it to exploit the abundant spectral information. The row-sparsity approach, which controls the sparsity by the l_2,0 norm, outperforms the single-sparsity approach in many scenarios. Many algorithms use convex relaxation methods to solve the l_2,0 norm to avoid the NP-hard problem, but this will reduce sparsity and unmixing accuracy. In this paper, a row-hard-threshold function is introduced to solve the l_2,0 norm directly, which guarantees the sparsity of the results. The high spatial correlation of hyperspectral images is associated with low column rank; therefore, the low-rank constraint is adopted to utilize spatial information. Experiments with simulated and real data prove that the proposed algorithm can obtain better unmixing results.

• Journal of the Korean Mathematical Society
• /
• v.36 no.2
• /
• pp.333-344
• /
• 1999

In [4], it was shown that an n by n orthogonal matrix which has a row of nonzeros has at least ( log2n + 3)n - log2n +1 nonzero entries. In this paper, the matrices achieving these bounds are constructed. The analogous sparsity problem for m by n row-orthogonal matrices which have a row of nonzeros in conjectured.

• Communications of the Korean Mathematical Society
• /
• v.15 no.4
• /
• pp.587-595
• /
• 2000

We contain a simple construction for the sparse n x n connected orthogonal matrices which have a row of p nonzero entries with 2$\leq$p$\leq$n. Moreover, we study the analogous sparsity problem for an m x n connected row-orthogonal matrices.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.7
• /
• pp.3194-3216
• /
• 2018

Slow Feature Discriminant Analysis (SFDA) is a supervised feature extraction method inspired by biological mechanism. In this paper, a novel method called Two Dimensional Slow Feature Discriminant Analysis via $L_{2,1}$ norm minimization ($2DSFDA-L_{2,1}$) is proposed. $2DSFDA-L_{2,1}$ integrates $L_{2,1}$ norm regularization and 2D statically uncorrelated constraint to extract discriminant feature. First, $L_{2,1}$ norm regularization can promote the projection matrix row-sparsity, which makes the feature selection and subspace learning simultaneously. Second, uncorrelated features of minimum redundancy are effective for classification. We define 2D statistically uncorrelated model that each row (or column) are independent. Third, we provide a feasible solution by transforming the proposed $L_{2,1}$ nonlinear model into a linear regression type. Additionally, $2DSFDA-L_{2,1}$ is extended to a bilateral projection version called $BSFDA-L_{2,1}$. The advantage of $BSFDA-L_{2,1}$ is that an image can be represented with much less coefficients. Experimental results on three face databases demonstrate that the proposed $2DSFDA-L_{2,1}/BSFDA-L_{2,1}$ can obtain competitive performance.

• Korean Management Science Review
• /
• v.15 no.1
• /
• pp.49-62
• /
• 1998

The purpose of this paper is to develope a large-scale simplex method program LPAKO. Various up-to-date techniques are argued and implemented. In LPAKO, basis matrices are stored in a LU factorized form, and Reid's method is used to update LU maintaining high sparsity and numerical stability, and further Markowitz's ordering is used in factorizing a basis matrix into a sparse LU form. As the data structures of basis matrix, Gustavson's data structure and row-column linked list structure are considered. The various criteria for reinversion are also discussed. The dynamic steepest-edge simplex algorithm is used for selection of an entering variable, and a new variation of the MINOS' perturbation technique is suggested for the resolution of degeneracy. Many preprocessing and scaling techniques are implemented. In addition, a new, effective initial basis construction method are suggested, and the criteria for optimality and infeasibility are suggested respectively. Finally, LPAKO is compared with MINOS by test results.