Skip to main content
ARS Home » Research » Publications at this Location » Publication #58957
Title: NEURAL NETWORK PATTERN RECOGNITION OF PHOTOACOUSTIC FTIR SPECTRA AND KNOWLEDGE-BASED TECHNIQUES FOR DETECTION OF TOXIGENIC FUNGI IN CORN

Author
item Gordon, Sherald
item Wicklow, Donald
item WHEELER, BRUCE - BECKMAN INSTITUTE
item SCHUDY, ROBERT - SYMBOLICS, INC
item Greene, Richard

Submitted to: Rocky Mountain Analytical Conference Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: 6/12/1996
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Fourier transform infrared photoacoustic spectroscopy (FTIR-PAS), a highly sensitive probe of the surfaces of solid substrates, is used to detect pathogenic fungal contamination in corn. Kernels of corn infected with toxigenic fungi, such as Aspergillus flavus, display FTIR-PAS spectra that differ significantly from spectra of uninfected kernels. Photoacoustic infrared spectral features were identified, and an artificial neural network was trained to distinguish contaminated from uncontaminated corn by pattern recognition. Software was written for computer extraction of the infrared spectral features and neural network classification. A friendly graphical user interface allows a non-chemist to easily discover spectral features useful in the analyses. Work is in progress to integrate epidemiological information about cereal crop fungal disease into the spectral pattern recognition program to produce a more knowledge-based, and hence, more reliable technique. A model of a hierarchically organized expert system is proposed, using epidemiological factors such as plant stress and susceptibility to infection, weather, insect vectors, handling and storage conditions, in addition to the analytical data to predict A. flavus and other kinds of toxigenic fungal contamination that might be present in food grains.