2-d image localization in hyperspectral image analysis of pharmaceutical materials

 

introduction discriminant analysis (da) is often used in hyperspectral image processing to classify multiple components on a sample matrix. methods in da, a customized threshold is typically applied in a univariate domain, such as a projected score or predicted concentration value, in order to differentiate multiple components. however, determination of an appropriate threshold is subjective and influenced by both material- and user-dependent factors. moreover, the potential of different thresholds to yield conflicting results poses a significant concern and challenge for pharmaceutical scientists. results the photon radial diffusion among adjacent pixels in a near-infrared chemical image has been documented in literature to be a significant contributor to the challenges associated with the threshold-dependent image analysis. therefore, in this study, a 2-d image of a score scatter plot was developed to visualize and identify the localization of individual components in a sample matrix, instead of a univariate projected score. conclusions an nir chemical imaging dataset of cylindrical compacts of binary mixtures, including microcrystalline cellulose and different sized polystyrene microspheres, was used to demonstrate the efficiency of the method. the application of the method on hyperspectral image analysis on pharmaceutical multi-component systems was also presented.