Pre-processing methods for automatic pigment recognition through Vis–NIR reflectance spectra

The present work introduces the preliminary outcome of one of the tasks inside the MOBARTECH project, that is the creation of the prototype for a compact FORS spectrophotometer capable of automatically identifying pigments and dyes. On the road toward this goal, we focused on the automation of pigment identification, as aspect of great interest which would allow real time analysis with benefit especially for unskilled operators. In this paper, we evaluate high-order derivative approximation pre-processing and features extraction methods, strictly related to the way the database is built; we have chosen to use a dynamic database to be able to elaborate data and give an answer in real time, independently from the experimental set up, and we are planning an interface suitable for both non-skilled and expert users. The aim of the present research is thus to evaluate how spectra pre-processing affects the response and the ability of the search algorithm to correctly identify an unknown spectrum, in order to develop a totally automated pipeline for spectra identification or to prefer an interactive framework.