An artificial neural network based encoding of an invariant sammon map for real-time projection of patterns from odour sensor arrays

K C Persaud, H G Byun

    Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

    Abstract

    Odour sensor arrays produce multidimensional patterns of data that may be used as 'fingerprints' to differentiate between different volatile chemicals or mixtures of chemicals. A problem that arises for the human observer, is how to visualise multidimensional data in order to determine whether a certain pattern is similar or different from known patterns. A requirement is to reduce multidimensional data to two or three dimensions, and project unknown patterns on to the resulting map in real time, allowing the observer to make judgements on the incoming data. The SAMMAN network was adopted, but modified by the use of a previously trained neural network encoding principal component analysis to provide weights used to initialise the SAMMAN network. This provided a robust system that was invariant to changes in the order of the data set, and allowed rapid projection of unknown data on to a Sammon map.
    Original languageEnglish
    Title of host publicationInternational Conference on Advances in Pattern Recognition
    Subtitle of host publicationProceedings of ICAPR ’98, 23–25 November 1998, Plymouth, UK
    EditorsSameer Singh
    Place of PublicationLondon
    PublisherSpringer Nature
    Pages187-194
    ISBN (Print)978-1-4471-1214-3
    Publication statusPublished - 1999

    Fingerprint

    Dive into the research topics of 'An artificial neural network based encoding of an invariant sammon map for real-time projection of patterns from odour sensor arrays'. Together they form a unique fingerprint.

    Cite this