Using neural networks and genetic algorithms to enhance performance in an electronic nose

Area of Science:

• Chemometrics
• Sensor Technology
• Machine Learning

Background:

• Accurate classification of odorous substances is crucial for various applications.
• Traditional methods face challenges in sensitivity, repeatability, and discernment.
• Electronic noses offer a promising approach for odor analysis.

Purpose of the Study:

• To develop an improved electronic nose system for odor classification.
• To enhance classification performance by integrating advanced machine learning techniques.
• To evaluate the system's effectiveness on diverse odorant classes.

Main Methods:

• Utilized a 32-sensor electronic nose array for data acquisition.
• Applied time-windowing functions for response partitioning and Karhunen-Loéve expansion (KLE) for dimensionality reduction.
• Employed a genetic algorithm (GA) for optimizing neural network (NN) training parameters and feature selection, creating a GA-supervised NN (GANN).

Main Results:

• The GANN system demonstrated superior performance compared to a standard NN classifier.
• Effective feature extraction and parameter optimization were achieved using the GA.
• The GANN successfully classified fragrances, hog farm air, and soft beverages.

Conclusions:

• The proposed GA-supervised NN system significantly improves odor classification accuracy.
• This approach addresses key challenges in sensitivity, repeatability, and discernment.
• The GANN is a robust and effective tool for electronic nose applications in odor analysis.