Relevant aspects of quantification and sample heterogeneity in hyperspectral image resolution
| Autor: | Anna de Juan, Romà Tauler, James Burger, Sara Piqueras |
|---|---|
| Rok vydání: | 2012 |
| Předmět: |
Chemical imaging
Multivariate statistics Pixel Computer science Calibration (statistics) business.industry Process Chemistry and Technology Image map Hyperspectral imaging Pattern recognition Context (language use) Computer Science Applications Analytical Chemistry Computer vision Artificial intelligence business Image resolution Spectroscopy Software |
| Zdroj: | Chemometrics and Intelligent Laboratory Systems. 117:169-182 |
| ISSN: | 0169-7439 |
| DOI: | 10.1016/j.chemolab.2011.12.004 |
| Popis: | NIR-infrared chemical imaging (NIR-CI) is a powerful technique to provide spatial and spectral information about the samples analyzed. NIR-CI presents a big potential to obtain accurate and reliable information about the quality of end products and it is also an excellent tool for process control. Quantitative analysis and heterogeneity studies on NIR images are often carried out by multivariate calibration techniques, designed in this context as Multivariate Image Regression (MIR). Multivariate Curve Resolution—Alternating Least Squares (MCR-ALS) is another data analysis method, aimed mainly at recovering the pure spectra and distribution maps of images. This straightforward information can be potentially applied for quantitative purposes and heterogeneity description when image multiset structures are analyzed. In this work, the potential of MCR-ALS to provide quantitative and heterogeneity information has been explored paying special attention to particular aspects, such as the preprocessing used for resolution and the effect of several factors on the global quantitative results, namely: a) the design of the regions of interest (ROIs) included in the multiset image structure, b) the constraints applied in the resolution step and c) the calibration/validation strategy applied. Quantitative information at a pixel level has also been carried out to study the heterogeneity of the samples analyzed, stressing the difference between the constitutional heterogeneity (population statistics describing the quantitative distribution or pixel-to-pixel variability in concentration) and distributional heterogeneity (image maps expressing the variability in the spatial distribution of compounds in the image). To perform the work described above, real NIR images of non-homogeneous mixtures of acetylsalicilyc, caffeine and starch of different compositions have been used. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect