|Chao, Kuanglin - Kevin Chao|
Submitted to: Journal of Food Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/5/2011
Publication Date: 10/2/2011
Publication URL: http://handle.nal.usda.gov/10113/55076
Citation: Yang, C., Kim, M.S., Kang, S., Cho, B., Chao, K., Lefcourt, A.M., Chan, D.E. 2011. Red to far-red multispectral fluorescence image fusion for detection of fecal contamination on apples. Journal of Food Engineering. 108:312-319. Interpretive Summary: Because outbreaks of foodborne illnesses related to fruits and vegetables have been reported in increasing numbers during the past decade, the need to develop methods to detect contamination of fruits and vegetables caused by bacteria spread through the feces of livestock or wildlife is growing more urgent. The precise and fast detection of feces on the surfaces of fresh produce will be essential to protecting public health from potentially contaminated fruits and vegetables. The EMFSL research group has developed a rapid fluorescence detection algorithm to inspect fresh produce for fecal contamination, based on the use of a non-destructive hyperspectral line-scan machine vision system. Hyperspectral fluorescence images of apples contaminated with bovine fecal matter were acquired by a line-scan imaging system utilizing newly available violet LEDs as the excitation light source. A multispectral algorithm was derived from hyperspectral image analysis of fluorescence emissions produced by the violet LED excitation. The developed algorithm detected contamination spots without misidentifying any normal apple surfaces as fecal contamination. The method and results described in this investigation can be used as a food safety inspection tool by the food processing industry and regulatory agencies to minimize food safety risks for the general public, and to provide information helpful to food process scientists and engineers.
Technical Abstract: This research developed a multispectral algorithm derived from hyperspectral line-scan fluorescence imaging under violet/blue LED excitation for detection of fecal contamination on Golden Delicious apples. Using a hyperspectral line-scan imaging system consisting of an EMCCD camera, spectrograph, and lenses equipped with a pair of violet LED line lights, fluorescence images of 59 Golden Delicious apples were acquired. The developed algorithm required the intensities from only four wavebands, 680 nm, 684 nm, 720 nm, and 780 nm, for computation of simple functions for precise detection of feces. The spectral analysis showed two spectral characteristics: the shift from the peak at 684 nm for apples to the peak at 680 nm for feces, and the shoulder curve was formed by 684 nm, 720 nm, and 780 nm for feces. Based on linear regression and second-order polynomial regression on1200 sample pixels from 8 of 59 apples, only six sample pixels of feces were required to form the boundary to separate apples and feces. The algorithm could detect more than 99% of different dilutions of feces on apples and obtained no false-positive detection on more than 90% of normal apple surfaces. The highly precise detection of feces showed that a simple multispectral fluorescence imaging algorithm based on violet/blue LED excitation can be used to detect fecal contamination on fast-speed apple processing lines.