Odor recognition VS. classification in artificial olfaction

Baranidharan Raman, Joshua Hertz, Kurt Benkstein, Steve Semancik

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


Most studies in chemical sensing have focused on the problem of precise identification of chemical species that were exposed during the training phase (the recognition problem). However, generalization of training to predict the chemical composition of untrained gases based on their similarity with analytes in the training set (the classification problem) has received very limited attention. These two analytical tasks pose conflicting constraints on the system. While correct recognition requires detection of molecular features that are unique to an analyte, generalization to untrained chemicals requires detection of features that are common across a desired class of analytes. A simple solution that addresses both issues simultaneously can be obtained from biological olfaction, where the odor class and identity information are decoupled and extracted individually over time. Mimicking this approach, we proposed a hierarchical scheme that allowed initial discrimination between broad chemical classes (e.g. contains oxygen) followed by finer refinements using additional data into sub-classes (e.g. ketones vs. alcohols) and, eventually, specific compositions (e.g. ethanol vs. methanol) [1]. We validated this approach using an array of temperature-controlled chemiresistors. We demonstrated that a small set of training analytes is sufficient to allow generalization to novel chemicals and that the scheme provides robust categorization despite aging. Here, we provide further characterization of this approach.

Original languageEnglish
Title of host publicationOlfaction and Electronic Nose
Subtitle of host publicationProceedings of the 14th International Symposium on Olfaction and Electronic Nose, ISOEN 2011
Number of pages4
StatePublished - 2011
Event14th International Symposium on Olfaction and Electronic Nose, ISOEN 2011 - New York City, NY, United States
Duration: May 2 2011May 5 2011

Publication series

NameAIP Conference Proceedings
ISSN (Print)0094-243X
ISSN (Electronic)1551-7616


Conference14th International Symposium on Olfaction and Electronic Nose, ISOEN 2011
Country/TerritoryUnited States
CityNew York City, NY


  • Bio-inspired processing
  • Chemical sensor array
  • Pattern recognition
  • Sensor drift


Dive into the research topics of 'Odor recognition VS. classification in artificial olfaction'. Together they form a unique fingerprint.

Cite this