Author
![]() |
Windham, William |
![]() |
Heitschmidt, Gerald - Jerry |
![]() |
Lawrence, Kurt |
![]() |
Park, Bosoon |
![]() |
Smith, Douglas |
Submitted to: International Journal of Poultry Science
Publication Type: Peer Reviewed Journal Publication Acceptance Date: 12/29/2007 Publication Date: 1/29/2008 Citation: Windham, W.R., Heitschmidt, G.W., Lawrence, K.C., Park, B., Smith, D.P. 2008. Effect of spectrally mixed pixels on detection of cecal contaminated broiler carcasses. International Journal of Poultry Science. 6(12):955-959. 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 processes is both labor intensive and prone to human error. The USDA Agricultural Research Service has developed a method in conjunction with a technology called hyperspectral imaging to detect feces on poultry carcasses. To further improve the method it is necessary to test its ability to detect very small (10 mg and less) fecal contaminants on chicken carcasses. Mathematical models developed for small contaminants were able to detect 100% of the chicken carcasses contaminated with 5 and 10 mg of fecal material. The models only detected 75% of the 2 mg fecal spots applied to the carcasses. The models also detected some very small spots on uncontaminated skin as feces (false positives). However, contaminants of this size do not increase bacterial counts of broiler carcasses. The very small false positives detected are in the same size as the 2 mg fecal spots and thus can be ignored. The system will aid FSIS in inspection of fecal contaminated carcasses. Technical Abstract: Detection of small masses (i.e. 10 mg and less) of fecal contaminants on broiler carcasses presents a significant challenge when using a multispectral imaging system. In contrast to the spectrally noncontiguous multispectral imagery, hyperspectral imagery can be seen as a single image with a contiguous spectrum of reflectance values associated with each image pixel. On a broiler carcass, the spectra may be recognizable as feces provided the contaminant fills or almost fills the pixel in the corresponding scene. Pixels partially filled (i.e. mixed pixels) by a contaminant result in a spectral signature that is a mixture of feces and carcass skin. Mixed pixels with small fecal masses on broiler carcasses can be problematic to accurately detect. The objective of this study was to determine whether hyperspectral imagery offered an improved detection rate of cecal contamination of known mass (2 to 10mg) relative to multispectral imagery. On each of three replicate sample days, twenty-four eviscerated, pre-chilled broiler carcasses were collected from a commercial processing plant. Cecal contents from the same flock were also collected and used to contaminate the carcasses. Carcass halves were first imaged uncontaminated and then imaged again after cecal contents (2, 5, or 10 mg) had been applied to the carcasses. Contaminants were predicted by decision tree (DT) and mixture tuned matched filter (MTMF) classifiers, and results compared. The DT classifier, applied to the multispectral imagery, detected 63, 80, and 100% of the cecal mass applied at about 2, 5, and 10 mg, respectively. The low detection accuracy of the 2 and 5 mg mass was due to some contaminated mixed pixels that either went under-detected or in some cases undetected altogether (false negatives). The MTMF classifier, applied to the hyperspectral imagery, detected 88% of 2 mg ceca and 100% of the 5 and 10 mg contaminants. At an applied mass of about 2, 5, and 10 mg, the MTMF classifier detected 55, 52, and 53%, respectively more cecal contaminated pixels than the DT classifier. The DT classifier incorrectly identified 104, 59, and 56 false positives on carcasses contaminated with about 2, 5 and 10 mg of ceca. On average, these false positives occurred on 36% of the carcasses. The MTMF classifier detected far fewer false positives on 15% of the carcasses. |