Nonlinear Mixing Characteristics of Reflectance Spectra of Typical Mineral Pigments

Hyperspectral technology has been used to identify pigments that adhere to the surfaces of polychrome artifacts. However, the colors are often produced by the mixing of pigments, which requires that the spectral characteristics of the pigment mixtures be considered before pigment unmixing is conducted. Therefore, we proposed an experimental approach to investigate the nonlinear degree of spectral reflectance, using several mixing models, and to evaluate their performances in the study of typical mineral pigments. First, five mineral pigments of azurite, malachite, cinnabar, orpiment, and calcite were selected to form five groups of samples, according to their different mass ratios. Second, a fully constrained least squares algorithm based on the linear model and three algorithms based on the nonlinear model were employed to calculate the proportion of each pigment in the mixtures. We evaluated the abundance accuracy as well as the similarity between the measured and reconstructed spectra produced by those mixing models. Third, we conducted pigment unmixing on a Chinese painting to verify the applicability of the nonlinear model. Fourth, continuum removal was also introduced to test the nonlinearity of mineral pigment mixing. Finally, the results indicated that the spectral mixing of different mineral pigments was more in line with the nonlinear mixing model. The spectral nonlinearity of mixed pigments was higher near to the wavelength corresponding to their colors. Meanwhile, the nonlinearity increased with the wavelength increases in the shortwave infrared bands.

Spatial Low-Rank Tensor Factorization and Unmixing of Hyperspectral Images

This work presents a method for hyperspectral image unmixing based on non-negative tensor factorization. While traditional approaches may process spectral information without regard for spatial structures in the dataset, tensor factorization preserves the spectral-spatial relationship which we intend to exploit. We used a rank-(L, L, 1) decomposition, which approximates the original tensor as a sum of R components. Each component is a tensor resulting from the multiplication of a low-rank spatial representation and a spectral vector. Our approach uses spatial factors to identify high abundance areas where pure pixels (endmembers) may lie. Unmixing is done by applying Fully Constrained Least Squares such that abundance maps are produced for each inferred endmember. The results of this method are compared against other approaches based on non-negative matrix and tensor factorization. We observed a significant reduction of spectral angle distance for extracted endmembers and equal or better RMSE for abundance maps as compared with existing benchmarks.

Application of spatially related MRF model in NMF hyperspectral unmixing

Aiming at Non-negative Matrix Factorization (NMF)’s problem of initialization and "local minima" in hyperspectral unmixing, a NMF linear unmixing algorithm with spatial correlation constrains (SCNMF) based on Markov Random Field (MRF) was proposed. Firstly, Hyperspectral Signal identification by minimum error (HySime) method was adopted to estimate the number of endmembers, initialized endmember matrix and abundance matrix by Vertex Component Analysis (VCA) and Fully Constrained Least Squares (FCLS) respectively. then established energy function to depict the spatial distribution characteristics of ground objects by MRF model. Finally, spatial correlation constraint based on MRF model and NMF standard objective function were combined in the form of altemating iteration to estimate endmember spectrum and abundance of hyperspectral image. Theoretical analysis and experimental results indicated that, the endmember decomposition precision of SCNMF is 10.6% higher than that of Minimum Volume Constrained NMF (MVC-NMF), 12.3% higher than that of Piecewise Smoothness NMF with Sparseness Constraints(PSNMFSC), 14.1% higher than that of NMF with Alternating Projected Subgradients(APS-NMF); the abundance decomposition precision of SCNMF is 14.4% higher than that of MVC-NMF, 15.9% higher than that of PSNMFSC, 15.3% higher than that of APS-NMF.The proposed SCNMF can remedy NMF's deficiency in describing spatial correlation characteristics, and decrease spatial energy distribution error.