Applying convolution based processing methods to a dual channel, large array artificial olfactory mucosa

Our understanding of the human olfactory system, particularly with respect to the phenomenon of nasal chromatography, has led us to develop a new generation of novel odour-sensitive instruments (or electronic noses). This novel instrument is in need of new approaches to data processing so that the i...

全面介紹

Saved in:
書目詳細資料
Main Authors: Taylor, James, Che Harun, F. K., Covington, J. A., Gardner, J. W.
格式: Conference or Workshop Item
出版: 2009
主題:
在線閱讀:http://eprints.utm.my/id/eprint/14842/
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
實物特徵
總結:Our understanding of the human olfactory system, particularly with respect to the phenomenon of nasal chromatography, has led us to develop a new generation of novel odour-sensitive instruments (or electronic noses). This novel instrument is in need of new approaches to data processing so that the information rich signals can be fully exploited; here, we apply a novel time-series based technique for processing such data. The dual-channel, large array artificial olfactory mucosa consists of 3 arrays of 300 sensors each. The sensors are divided into 24 groups, with each group made from a particular type of polymer. The first array is connected to the other two arrays by a pair of retentive columns. One channel is coated with Carbowax 20M, and the other with OV-1. This configuration partly mimics the nasal chromatography effect, and partly augments it by utilizing not only polar (mucus layer) but also non-polar (artificial) coatings. Such a device presents several challenges to multi-variate data processing: a large, redundant dataset, spatio-temporal output, and small sample space. By applying a novel convolution approach to this problem, it has been demonstrated that these problems can be overcome. The artificial mucosa signals have been classified using a probabilistic neural network and gave an accuracy of 85%, Even better results should be possible through the selection of other sensors with lower correlation.