Infrared mapping of inorganic materials: a supervised method to select relevant spectra

Abstract

Spectroscopic imaging is expanding thanks to new instrumental concepts, technological developments, easier use and lower costs. This leads to collect and to handle huge hyperspectral databases. Among data providing useful information, there may also be unwanted information (i.e., either biased or irrelevant with respect to the information one wants to retrieve from the analysis). The classical approach aims at building a spectra library, which contains both informative spectra and outliers, so as to train learning algorithms. However, reference spectra cannot always be acquired, especially for samples prone to ageing or changes when exposed to humidity, temperature… Thus, building a library with reference spectra is not always possible. To handle this issue, a new supervised method (SSMS for Supervised Selective Method based on SIMPLISMA) has been designed and is described in this article, to identify and exclude unwanted spectra from the resolution process when no library is available. SSMS relies on a supervised exclusion of the unwanted spectra. It ensures both a quick treatment and an accurate analysis of the data (reduced number of representative spectra to supervise). This new method is applicable to any type of hyperspectral database. In this work, its efficiency is demonstrated on a database acquired using a FT-IR microscope. To avoid issues arising from the acquisition of maps with classic ATR crystals (cross-contamination and successive residual imprints on the material), the use of a new set-up called static-ATR is explored. In addition, the combined use of SSMS and of a physical model permits to identify the origins of the outlier spectra. Thus, it becomes possible to improve the experimental method (sample preparation, acquisition parameters…). Finally, with a constrained Alternating Least Squares method (ALS), relevant chemical information is obtained. The robust method developed here permits to achieve chemical maps at the micron scale for inorganic materials.