Machine learning methods for wine IR spectra analysis
At a glance
- Project leader : Prof. Dr. Thomas Ott
- Project team : Prof. Dr. Urs Mürset, Dr. Robert Rohrkemper
- Project status : completed
- Funding partner : Internal
- Contact person : Thomas Ott
Description
Infrared (IR) spectra of wine from two datasets have been
analyzed. Categories were created
automatically via machine learning methods. These categories group
the wine by specific
type as well as color. The classification methods successfully
achieved less than 5% error.
Specific parameters were also quantified via regression methods,
also with less than 5% error.
Some parameters were not previously documented via IR spectroscopy
for wine and include
tannins, alcohol, pH, AcOH, and density. The project report also
includes discussions about the
overall context of wine IR spectroscopy and its applications. A
full evaluation was performed
of the OPUS software offered by Bruker. A detailed list of possible
improvements to the
software is provided.