RESEARCH : Machine Learning for Scent Verbal Description

Who is working on it:

Read a longer description of the project here or watch short Alex's presentation at the WKU Physics and Astronomy Colloquium: Project Overview:

The main goal of the project is to develop Multidimensional Gas Chromatography based artificial olfactory system which will provide a verbal description of various scents. Gas Column Chromatography (GCC) is widely used in analytical chemistry to separate and analyze chemicals from a mixture. The Applied Physics Institute (API) at Western Kentucky University developed a compact gas chromatograph in combination with an array of highly integrated and selective metal oxide (MOX) sensors. Due to the multisensory detector used in the experimental setup, the device collects multiple chromatograms in a single run.

Quality of odor is an important property which is used to characterize food and beverage, air and water quality, cosmetics, medicine and other areas. The GCC experimental setup provides valuable data, but there is a need for this data to be interpreted further to provide verbal description of an odor.

In this project we will develop a Deep Learning based algorithm to provide a verbal description of an odor sample based on an experimentally obtained chromatogram. The odor will be analyzed using multisensory gas chromatography experimental setup. Each chromatogram will be 1) analyzed to obtain a set of peak positions as well as areas under each peak, and then 2) matched with a 146-dimensional vector containing verbal descriptors. Verbal descriptors are provided in "The Atlas of Odor Character Profiles" by A. Dravnieks. In order to map experimental data (chromatograms) to a verbal description in the 146-dimensional space, we propose to use a Convolutional Neural Network.