Chemometrics, imaging and spectroscopy laboratory – Department of Life Sciences, University of Modena and Reggio Emilia

Abstract

Following the previous papers of our colleagues from the University of Genova and from the University of Rome “La Sapienza” in the series of articles presenting the Italian research groups active in the field of NIR spectroscopy, this paper aims at introducing the main activities of the Chemometrics, Imaging and Spectroscopy Laboratory (CHIMSLAB) of the University of Modena and Reggio Emilia. The group is headed by Prof. Alessandro Ulrici, associate professor in Analytical Chemistry and Coordinator of the Research Doctorate School in Food and Agricultural Science, Technology and Biotechnology (STEBA) of the University of Modena and Reggio Emilia. CHIMSLAB team also includes Dr. Giorgia Foca, as assistant professor, and Dr. Rosalba Calvini, as post-doc researcher. We would also like to mention our former PhD students, Dr. Carlotta Ferrari and Dr. Giorgia Orlandi, who gave a fundamental contribution to our recent activities. In addition, in 2017, we had the pleasure to host Prof. Sylvio Barbon Junior (Computer Science Department, Londrina State University) and Dr. Ana Paula A. C. Barbon (Animal Science Department, Londrina State University) as visiting researchers. CHIMSLAB research group is also affiliated to BIOGEST-SITEIA, the interdepartmental research centre of the University of Reggio Emilia working on the improvement and valorisation of agri-food biological resources. The keywords in the group name recall our main research activities: the application and development of chemometric methods for data modelling with a specific interest in spectroscopic and imaging data. In particular, the application of near infrared (NIR) spectroscopy, computer vision and NIR hyperspectral imaging (NIR-HSI) in the agri-food sector represents a considerable part of our expertise. However, thanks to collaborations with other research groups, we had the possibility of applying chemometric modelling to a wide range of research fields, including cultural heritage, electrochemical sensing, microbiology and entomology, among others. Concerning our research topics of main interest for the readers of NIR News, in the past years, we focused on two key aspects of spectroscopic and imaging data analysis: variable selection and data dimensionality reduction. Feature selection is a crucial aspect in the analysis of spectroscopic signals, since the selection of the spectral regions of interest for the problem at hand usually allows to discard noise and to obtain calibration or classification models with higher performances. For these reasons, starting from the beginning of our research activities, the application of state-of-art variable selection methods and the development of new selection strategies have represented key topic of our work. The importance of variable selection methods is even more relevant when dealing with NIR hyperspectral images. Indeed, usually, industrial applications require sorting technologies meeting the requirements of fast time of analysis and low costs. Therefore, variable selection is generally applied to hyperspectral data acquired at the laboratory scale in order to find few wavelengths relevant for the problem at hand to be implemented in a faster and cheaper multispectral imaging system. In this context, a recent collaboration with Prof. Jose Amigo (University of Basque Country) and Caff e Molinari S.p.A. aimed at stimulating the implementation of a multispectral imaging system based on only four wavelengths for the classification of Arabica and Robusta green coffee beans. The four wavelengths were selected through the application of sparse-based variable selection methods to hyperspectral data, and the key aspect of our simulation consisted in the identification of relevant descriptors derived from the reflectance values registered at the four wavelengths in order to obtain classification performances similar to those obtained with the hyperspectral imaging system. The second key topic of our research activity concerns the development of data dimensionality methods