Skip to main content
ARS Home » Research » Publications at this Location » Publication #142215
Title: SIMPLE ALGORITHMS FOR THE CLASSIFICATION OF VISIBLE/NIR AND HYPERSPECTRAL IMAGING SPECTRA OF CHICKEN SKINS, FECES, AND FECAL CONTAMINATED SKINS

Author
item LIU, Y - UGA
item Windham, William
item Lawrence, Kurt
item Park, Bosoon

Submitted to: Journal of Applied Spectroscopy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/17/2003
Publication Date: 12/1/2003
Citation: Liu, Y., Windham, W.R., Lawrence, K.C., Park, B. 2003. SIMPLE ALGORITHMS FOR THE CLASSIFICATION OF VISIBLE/NIR AND HYPERSPECTRAL IMAGING SPECTRA OF CHICKEN SKINS, FECES, AND FECAL CONTAMINATED SKINS. Journal of Applied Spectroscopy. 57(12):1609-1612.

Interpretive Summary: Legislation requires that each poultry carcass at processing plants be inspected by FSIS inspectors. On-line visual and manual inspection of fecal contaminated chicken carcasses is both labor intensive, prone to human error and day-to-day and inspector-to-inspector variations. Hence, ARS has developed multispectral and hyperspectral imaging systems for use in real-time on-line detection of fecal contaminated carcasses. To further improve the mathematical model and select better key wavelengths it is necessary to explore the fundamental visible light spectral features of chicken skins and feces. Moreover, it is important to compare the detection results by using the key wavelengths on both visible light data and imagings. We determined additional key wavelengths and new algorithm methodology from visible light data and successfully applied the models to images of contaminated carcasses. The results strengthen the scientific understanding of skin / feces color on key wavelength selection. The outcome provides guidance to further design the real-time on-line imaging system.

Technical Abstract: This study presented a novel methodology to analyze and classify visible/NIR spectra of uncontaminated chicken skins and pure chicken feces as well as hyperspectral imaging spectra of fecal contaminated chicken skins. The results revealed that key wavelengths in both subtraction and ratio algorithm could be used to perform the classification analysis between skin and feces with a great success. The results were in agreement with classical chemometric modeling methodology. Obviously, the algorithm approach is most attractive and interesting since, in its simplest form, there is no calibration model, which is commonly built from a larger set of spectral data. In addition, there are less data points, which could benefit the rapid processing of spectra from an on-line imaging system.