|Heitschmidt, Gerald - Jerry|
Submitted to: Near Infrared Spectroscopy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/12/2006
Publication Date: 8/18/2006
Citation: Lawrence, K.C., Windham, W.R., Park, B., Heitschmidt, G.W., Smith, D.P. 2006. Partial Least Squares Regression of Hyperspectral Images for Contaminant Detection on Poultry Carcasses. Near Infrared Spectroscopy Journal. 14(4):223-230.
Interpretive Summary: According the Food Safety & Inspection Service, there can be no fecal contamination on processed poultry carcasses when they enter the chiller tanks. ARS has developed an advanced imaging system based on hyperspectral imaging to detect the fecal contaminants. Although current hyperspectral systems are too slow for continuous on-line inspection, new system coming on the market should be able to keep up with the 140 bird per minute line speeds. This paper reports on a method to utilize the broad spectral response from a hyperspectral image system to detect fecal contaminants. A statistical analysis technique, known as partial least squares regression, was used to predict the contaminants with about 95% accuracy for over 400 contaminant spots while only 26 false responses were found. About 1/3 of the false positives were from bloody wingtips which would not be seen in a commercial processing line. The new algorithm now needs to be evaluated with one of the new high speed hyperspectral imaging systems.
Technical Abstract: The U.S. Department of Agriculture has developed multispectral and hyperspectral imaging systems to detect faecal contaminants. Until recently, the hyperspectral imaging system has been used as a research tool to detect a few optimum wavelengths for use in a multispectral imaging system. However, with the development of CMOS cameras, discrete wavelengths or subsets of the full-detector range can be used to greatly increase the speed of hyperspectral imaging systems. This paper reports on the use of a broad-spectrum multivariate statistical analysis technique for detecting contaminants with hyperspectral imaging. Partial least squares regression (PLSR) was used for model development. Calibration models from spatially averaged region of interest data were developed with and without smoothing, with and without scatter correction, and with and without first derivative (difference) preprocessing. Results indicate that using the full spectral range with scatter correction was needed for good model development. Furthermore, validation of the various calibration models indicated that preprocessing with scatter correction, nine-point boxcar smoothing, and first derivative preprocessing resulted in the best validation with about 95% of the over 400 contaminants detected with only 26 false positives (errors of commission). About one-third of the false positives were from bruised wingtips which would not be visible during in-plant commercial processing.