Interpretability of Machine Learning methods andmethodology for analyzing hyperspectral images, discover the 2 Ondalys’ scientific communications at the NIR2025 conference in June 2025, in Roma.
Organized by SISNIR – Società Italiana di Spettrosocopia NIR, the 22nd International Conference on Near Infrared Spectroscopy – NIR2025, will be held from June 8th toJune 12th 2025 in Roma, Italy.
Expert in Chemometrics and Machine Learning methods applied to spectral data, the ondalys team will attend this conference to discuss about spectroscopic data analysis and will present 2 communications on the following themes :
Machine Learning interpretability methods applied to calibration models developed on NIR spectroscopic data
In the past decades, Machine Learning (ML) models have become more and more complex, leading to improvements in their predictive performance. However, these models can often be described as “black boxes”, in the sense that it is very difficult to explain how results are obtained by a model from the input data. As complex Machine Learning models are increasingly used to make decisions, for instance in industrial or medical applications, there is a growing need to improve their interpretability in order to have greater confidence in their results, providing the so-called Explainable AI (Artificial Intelligence).
This study is focused on the interpretability of Machine Learning models after model calibration on near-infrared spectroscopic data, a.k.a. the “post-hoc explanation” of ML models.
Learn more about the advanced Machine Learning methods
Machine Learning methods for sugar quantification in grapes based on NIR Hyperspectral Imaging
Hyperspectral imaging [1] has applications in many fields, including agriculture, environment, medicine and industry. It can be used to classify objects according to their composition, or to quantify compounds present on the surface and show their spatial distribution. This last case is often challenging as the reference is not known for each pixel. Image processing is even more important in this case to extract the Region of Interest (ROI) to build the model but also to be able to apply it on new images.
This scientific study presents the methodology for analyzing hyperspectral images for grape bunch maturity prediction directly in vineyards. In this example, the parameter to be predicted is the sugar content (in °Brix) on each whole bunch of grapes, providing an average measurement of sugar content per image
Learn more about how and why analysizing hyperspectral images – HSI
NIR2025 will highlight all aspects of near-infrared spectroscopy, spanning from methodological and technological advances in NIR spectroscopy (near-infrared imaging and miniaturized spectrometers) to applications in various fields (agri-food, pharmaceutical, chemical, biotech, etc.) through the analysis of spectroscopic data and hyperspectral images.
Every two years, in fact, the International Council for Near Infrared Spectroscopy (ICNIRS – https://icnirs.org/) organizes the main international conference dedicated to NIR spectroscopy.
Each edition permits to to catch up on all the NIR news (theory, applications, innovative instrumental solutions).
