Spectral peak verification and recognition using a multilayered neural network

Details

• Personal Author:
  Levine SP ; Tomellini SA ; Wythoff BJ
• Description:
  The development of a peak validation system and stand alone peak recognition system, based on a neural network model termed the multilayered perceptron, was described for vapor phase infrared spectra. The system developed consisted of three programs: a visual peak evaluation program; a network training, diagnostic, and peak verification program; and a stand alone peak picking program. The peak evaluation program provided correct output values used in network training. Each input pattern was graphically displayed by this program and scaled analogously to the network input normalization algorithm, and the correct output was specified for that input pattern. The infrared spectral data were 2 centimeter resolution spectra of vapor phase species. A portion of noise data from the spectrum was added to each input vector. Thirteen data points described most of the bands and 25 noise points were sufficient to adequately describe the noise. The network was trained twice by using these same parameters. The error in mapping the training values decreased as the number of hidden nodes increased. A standalone peak recognition program was written to perform autonomous evaluation of a test spectrum. This program can use any network architecture and connection weights created with the training/evaluation program. A data window equal in width to the number of inputs for the network was incrementally moved down the spectrum data points, one at a time. A simple prefilter reduced the volume of data. The authors conclude that this system for interpreting infrared spectral data has implications in every chemical application where the recognition of peak shape signals in analytical data is important. [Description provided by NIOSH]
• Subjects:
  Computers Infrared Rays Occupational Health Safety Software
• Keywords:
  Computer-models Grants-other Infrared-spectrophotometry Mathematical-models NIOSH-Grant NIOSH-Publication Quantitative-analysis
• ISSN:
  0003-2700
• Document Type:
  Text
• Funding:
  Grant
• Genre:
  Journal Article
• Place as Subject:
  Michigan ; OSHA Region 5
• CIO:
  National Institute for Occupational Safety and Health (NIOSH)
• Topic:
  Workplace Safety & Health
• Location:
  North America
• Volume:
  62
• Issue:
  24
• NIOSHTIC Number:
  nn:00229145
• Citation:
  Anal Chem 1990 Dec; 62(24):2702-2709
• Contact Point Address:
  Environmental & Indust Health School of Public Health II 1420 Washington Heights Ann Arbor, MI 48109-2029
• Federal Fiscal Year:
  1991
• NORA Priority Area:
  Other Occupational Concerns ; Grants Other
• Performing Organization:
  University of Michigan at Ann Arbor, Ann Arbor, Michigan
• Peer Reviewed:
  True
• Start Date:
  19870929
• Source Full Name:
  Analytical Chemistry
• End Date:
  19911231
• Collection(s):
  National Institute for Occupational Safety and Health
• Main Document Checksum:
  urn:sha-512:e6d356e79bf903380e1e83c5bbe7459c49b67d321b0414413523825885df2a5096cf7c39b159547d84040d98e22eb46b98d0c8a836998e9754b4af2d3a30b589
• Download URL:
  https://stacks.cdc.gov/view/cdc/205622/cdc_205622_DS1.pdf
• File Type:
  [PDF - 2.06 MB ]