Algorithm development with visible-near-infrared spectral for detection of poultry feces and injesta

Authors: Windham, William; Smith, Douglas; Park, Bosoon; Lawrence, Kurt; Feldner, Peggy

Submitted to: Transactions of the ASAE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/17/2003
Publication Date: 11/18/2003
Citation: Windham, W.R., Smith, D.P., Park, B., Lawrence, K.C., Feldner, P.W. 2003. Algorithm development with visible-near-infrared spectral for detection of poultry feces and injesta. Transactions of the ASAE. 46(6):1733-1738.

Interpretive Summary: On-line visual and manual inspection of fecal contaminated chicken carcasses is conducted by the Food Safety and Inspection Service (FSIS) to ensure a safe meat supply to consumers. The inspection process is both labor intensive and prone to human error. The USDA Agricultural Research Service has developed a method and a hyperspectral imaging system to detect feces (from duodenum, ceca and colon) and ingesta on poultry carcasses. To further improve the method it is necessary to test the method using feces and ingesta from poultry fed different diets. Mathematical equations developed from poultry fed a corn diet successfully detected 100% of the feces from poultry fed milo and wheat diets. The models also detected some uncontaminated skin as feces (false positives). In addition, a new computer program was tested to simplify the search for key fecal detection wavelengths and aided in 'fine-tuning' the wavelengths. Fecal detection models, specifically a division of 2 and/or 3 key wavelengths was 100% successful in detecting contamination with no false positives. These models and wavelengths are the basis for the on-line multispectral imaging system to be used in processing plants. The system will aid FSIS in inspectors to identify carcasses contaminated with feces.

Technical Abstract: The USDA Agricultural Research Service has developed a method and a hyperspectral imaging system to detect feces (from duodenum, ceca and colon) and ingesta on poultry carcasses. The method first involves the use of multivariate data analysis on visible/near-infrared (Vis/NIR) reflectance spectra of fecal and uncontaminated skin samples for classification and selection of key wavelengths. Four dominant wavelengths (434, 517, 565, and 628 nm) were identified by intensity of principal component (PC) loading weights. Specifically, with a quotient of 565-nm/517-nm, 100% of the fecal contaminates were detected. The objectives of this research to validate the 565-nm/517-nm quotient to classify uncontaminated skin from feces/ingesta with broilers fed corn as well as milo and wheat diets and to investigate the use of single-term linear regression (STLR) to select key wavelengths to detect feces. Feces (N=369) and uncontaminated broiler breast skin (N=96) were analyzed from 440 to 880 nm. The overall accuracy of detecting contamination with the 565-nm/517-nm quotient was 99% with 16 false positive. STLR optimized a new quotient of 574-nm/588-nm which classified 100% of contaminates correctly. The shift in the denominator from 516 to 588 nm is possible due to greater fecal color variation from broilers fed wheat and milo. The use of the STLR to scan the spectral data to find key wavelengths is a good alternative to selecting key wavelengths based of the intensity of PC loading weights. Although models from Vis/NIR spectroscopy and STLR performed well, they need to be validated on hyperspectral images of uncontaminated and contaminated carcasses.