Electronic nose with a new feature reduction method and a multi-linear classifier for Chinese liquor classification

Area of Science:

• * Analytical Chemistry
• * Chemometrics
• * Artificial Intelligence

Background:

• * Accurate classification of Chinese liquors is crucial for quality control and authentication.
• * Existing methods often lack efficiency in handling complex aroma profiles.
• * Electronic nose (e-nose) technology offers a promising approach for objective aroma analysis.

Purpose of the Study:

• * To develop and evaluate a new feature reduction technique for e-nose data.
• * To classify Chinese liquors based on aroma style, raw materials, and geographical origin.
• * To compare the performance of different classification algorithms.

Main Methods:

• * An electronic nose system was employed for data acquisition.
• * A hybrid feature reduction method combining feature selection (information theory-based) and feature extraction (Kernel Entropy Component Analysis) was implemented.
• * Classification was performed using Back Propagation Artificial Neural Network (BP-ANN), Linear Discriminant Analysis (LDA), and a multi-linear classifier.

Main Results:

• * The combined feature reduction technique successfully reduced feature space dimensions to 41 and then to 12.
• * The multi-linear classifier achieved the highest classification rate of 97.22% for aroma style.
• * Classification by raw materials and geographical origin yielded rates of 98.75% and 100%, respectively.

Conclusions:

• * The proposed feature reduction method significantly enhances the efficiency and accuracy of e-nose systems.
• * The multi-linear classifier demonstrates superior performance in classifying Chinese liquors.
• * The e-nose system, coupled with advanced data analysis, provides a reliable tool for liquor authentication and categorization.