Classification of whiskies and drinking glasses based on ion mobility spectrometry measurements
Närhi, Juha (2025)
2025
Tieto- ja sähkötekniikan kandidaattiohjelma - Bachelor's Programme in Computing and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2025-05-15
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202505145451
https://urn.fi/URN:NBN:fi:tuni-202505145451
Tiivistelmä
Alongside other alcoholic beverages, whiskies require reliable authentication methods to prevent fraud, such as mislabeling and adulteration. To support such authentication efforts, this thesis explored whether data collected from an ion mobility spectrometry (IMS)-based electronic nose (e-nose) could be used to identify different whiskies and determine from which glass the measurement had been taken.
In the measurement setup, the IMS device was placed at the rim of the glass to mimic a realworld sniffing situation, where volatile compounds from the whisky evaporate and are typically perceived by the human nose. Three different whiskies and three different types of glasses were used, resulting in a total of nine whisky-glass combinations. The use of different glasses was motivated by the aim of investigating whether the shape of the glass affects the IMS response and thereby the classification accuracy. Each combination was measured for ten minutes under controlled conditions, and the measurements were repeated five times.
After data collection, the data was preprocessed and then multiple machine learning algorithms, including a Long Short-Term Memory (LSTM) neural network, Support Vector Classification (SVC), and Linear Discriminant Analysis (LDA), were applied to perform the classification. The validation was performed using session-based cross-validation.
The highest classification accuracies were obtained using the Support Vector Classifier (SVC): 87% for whisky identification, 69% for glass identification, and 67% for classifying whisky-glass combinations. While the glasses generated somewhat distinctive IMS responses, the classification of whiskies proved to be consistently more reliable and accurate than the identification of the glasses. The results were encouraging and provide a foundation for future research and potential real-world applications, if classification accuracy can be further improved.
In the measurement setup, the IMS device was placed at the rim of the glass to mimic a realworld sniffing situation, where volatile compounds from the whisky evaporate and are typically perceived by the human nose. Three different whiskies and three different types of glasses were used, resulting in a total of nine whisky-glass combinations. The use of different glasses was motivated by the aim of investigating whether the shape of the glass affects the IMS response and thereby the classification accuracy. Each combination was measured for ten minutes under controlled conditions, and the measurements were repeated five times.
After data collection, the data was preprocessed and then multiple machine learning algorithms, including a Long Short-Term Memory (LSTM) neural network, Support Vector Classification (SVC), and Linear Discriminant Analysis (LDA), were applied to perform the classification. The validation was performed using session-based cross-validation.
The highest classification accuracies were obtained using the Support Vector Classifier (SVC): 87% for whisky identification, 69% for glass identification, and 67% for classifying whisky-glass combinations. While the glasses generated somewhat distinctive IMS responses, the classification of whiskies proved to be consistently more reliable and accurate than the identification of the glasses. The results were encouraging and provide a foundation for future research and potential real-world applications, if classification accuracy can be further improved.
Kokoelmat
- Kandidaatintutkielmat [10645]