|Kim, Taemin -|
|Cho, Byoung-Kwan -|
Submitted to: Colombian Journal of Animal Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: February 23, 2010
Publication Date: June 1, 2010
Repository URL: http://HDL.HANDLE.NET/10113/47717
Citation: Kim, T., Cho, B., Kim, M.S. 2010. Emission filter design to detect poultry skin tumors using Fluorescene hyperspectral imaging. Colombian Journal of Animal Science. 23:9-16. Interpretive Summary: Development of rapid and reliable machine vision-based inspection systems to ensure safe production of poultry during post-harvest processing is needed. To develop machine vision-based food safety and quality inspection technologies a thorough understanding of the spectral and spatial attributes of samples is essential. We developed a hyperspectral imaging platform capable of fluorescence imaging that allows for relatively large sample evaluations for sequences of individual wavelengths across broad spectral regions. Using hyperspectral fluorescence images of whole chicken carcasses with skin tumors, we developed a numerical method to determine optimal fluorescence emission bands to detect the tumors. These spectral bands are important in designing a rapid on-line inspection system for use in the processing plants. This information is useful to the FSIS and poultry processing plants.
Technical Abstract: The secure production of disease-free meat is crucial in the mass production environment. The fluorescence spectra of poultry have been gaining the practical use because the fluorescence response is very sensitive in detecting a particular biological element. A hyperspectral image contains spectral information measured as a sequence of individual wavelengths across broad spectral regions. This paper presents an optimal design method of emission filter in a hyperspectral fluorescence imaging system to detect skin tumors on poultry carcasses. The proposed method designs the optimal emission filter using the linear discriminant analysis. It provides the optimal weighting of emission wavelengths in terms of discriminant power. The weights prioritize wavelengths to select significant spectral bands. Skin tumor parts in resultant fluorescence images of chicken carcasses are enhanced drastically so that they are distinguishable by the pixel-wise intensity. The method can be extended to detect other biomedical abnormalities as well.