36B-24 |
|
A. VARGAS, Biological Resources Engineering, Univ. of Maryland, 1426 Ansc/AgEn Building, College Park, MD 20742, M. Kim, Instrumentation and Sensing Laboratory, United States Department of Agriculture - ARS, Bldg. 303, BARC-East, 10300 Baltimore Avenue, Beltsville, MD 20705-2350, Y. Tao, Dept. of Biological Resource Engineering, Univ. of Maryland, College Park, MD 20742, A. Lefcourt, USDA - ARS, Instrumentation and Sensing Laboratory, Bldg. 303 BARC-East, 10300 Baltimore Ave., Beltsville, MD 20705-2350, Y.-R. Chen, Y. Luo, Produce Quality & Safety Lab., USDA-ARS-Beltsville Agricultural Research Center, 10300 Baltimore Ave., Bldg. 002, Rm. 117, Beltsville, MD 20705, and Y. Song, FDA, Division of Food Processing and Packaging, 6502 S. Archer Rd., Summit-Argo, IL 60501. Fecal contamination detection on produce has become increasingly important since the Food and Drug Administration (FDA) averred fecal contamination as a major source of human pathogens. The Instrumentation and Sensing Laboratory (ISL), United States Department of Agriculture (USDA) developed a laboratory based Hyperspectral Imaging System (HIS) for anomaly detection on produce and meat. In this paper hyperspectral fluorescence images acquired using the HIS were used to examine if detection of fecal contamination on cantaloupes is possible. The system uses a UVA light source for fluorescence imaging and scans images in the 425 to 770 nm range. Fluorescent images were taken after samples were treated with diluted cow feces at different concentrations and volumes. ubsequently, the images were subjected to several mathematical algorithms such as permutation of band ratios, unsupervised classification, and principal component analysis (PCA). Results indicate that fluorescence images at 675 nm exhibited the greatest contrast between treated and untreated surfaces. Detection rates were improved using ratio images; in particular, higher detection rates were obtained for all volumes and concentrations of feces treatments using the 695/595, 675/555 and 555/665 nm image. Unsupervised classification images were more effective in allowing removal of unwanted areas, and isolating treated areas. PCA showed that the first 6 principal component (PC) images exhibited useful results for contamination detection. PC-2 and PC-5 displayed best contrast for contamination detection. False alarms presented a persistent problem when trying to identify fecal contamination, however, PC-5 provided contrast between them, creating ideal conditions for masking out the false alarms.
Session 36B, Biotechnology: General
2005 IFT Annual Meeting, July 15-20 - New Orleans, Louisiana |