William R. Windham

HYPERSPECTRAL IMAGING FOR DETECTING FECAL AND INGESTA CONTAMINANTS ON POULTRY CARCASSES

A hyperspectral imaging system including a camera with prism-grating-prism spectrograph, fiber optic line lighting, motorized lens control, and hyperspectral image-processing software was developed for poultry safety inspection, particularly identification of fecal and ingesta contamination on poultry carcasses. Both spectral and spatial image data between 400 and 900 nm with 512 spectral bands were acquired from fecal and ingesta contaminated poultry carcasses. Four dominant wavelengths (434, 517, 565, and 628 nm) were selected by principal component analysis from visible/near-infrared spectroscopy for wavelength selection of hyperspectral images. A calibration model for the hyperspectral imaging system was developed from calibration lighting sources (HgAr, Kr, and lasers) for accurate band selection from the hyperspectral images to identify spatial and spectral characterization of fecal and ingesta contaminants. Hyperspectral image processing algorithms, specifically band ratio of dual-wavelength (565/517) images and histogram stretching, were effective in the identification of fecal and ingesta contamination of poultry carcasses. Test results indicated that the detection accuracy was 97.3% for linear and 100% for non-linear histogram stretching. This article presents the research results that hyperspectral imaging can be used effectively for detecting feces (from the duodenum, ceca, and colon) and ingesta on poultry carcasses and demonstrates the potential application for on-line safety inspection of poultry.

CALIBRATION OF A PUSHBROOM HYPERSPECTRAL IMAGING SYSTEM FOR AGRICULTURAL INSPECTION

A method to calibrate a pushbroom hyperspectral imaging system has been demonstrated for use in agricultural ninspection where the imaged object is close to the imaging system. The method consists of a modified geometric control point ncorrection to remove smile and keystone effect from the system, and both wavelength and distance calibrations to reduce the nwavelength and distance errors to less than 0.5 nm and 0.01 mm (across entrance slit–width), respectively. Next, a npixel–by–pixel percent reflectance calibration was performed at all wavelengths with dark current and 99% reflectance ncalibration–panel measurements, and results were verified with measurements on a certified gradient Spectralon panel with nnominal values ranging from 12% to 99%. Results indicate that the method is capable of calibrating the hyperspectral system nacross the entire spectral range of the detector, but errors increase below 420 nm and above 840 nm. Further research should nbe performed to evaluate the stability of the calibration over time, and techniques must be developed to implement the ncalibration for real–time analysis.

Hyperspectral Imaging for Detecting Fecal and Ingesta Contamination on Poultry Carcasses

A hyperspectral imaging system including camera with prism-grating-prism nspectrograph, fiber optic line lighting, motorized lens control, and hyperspectral image nprocessing software was developed for poultry safety inspection, particularly the identification of nfecal and ingesta contamination on poultry carcasses. Both spectral and spatial image data nbetween 400 and 900 nm with 512 spectral bands were acquired from fecal and ingesta ncontaminated poultry carcasses. Four dominant wavelengths (434, 517, 565, and 628 nm) were nselected by principal component analysis from visible/near-infrared spectroscopy to apply for nwavelength selection of hyperspectral images. A calibration model for the hyperspectral nimaging system was developed from calibration lighting sources (HgAr, Kr, and Lasers) for naccurate band selection from hyperspectral images to identify spatial and spectral ncharacterization of fecal and ingesta contaminants. Hyperspectral image processing algorithms, nspecifically band ratio of dual-wavelength (565/517) images and histogram stretching, were neffective on the identification of fecal and ingesta contamination of poultry carcasses. This algorithm can be further applied for real-time identification of fecal contamination on poultry ncarcasses in the processing line. This paper presents the research results that hyperspectral nimaging can be used effectively for detecting feces (from duodenum, ceca, and colon) and ningesta on poultry carcasses and demonstrates potential application for on-line processing of npoultry for safety inspection.

Effect of Intestinal Content Contamination on Broiler Carcass Campylobacter Counts

Intestinal contents may contaminate broiler carcasses during processing. The objective of this study was to determine what effect various levels of intestinal contents had on the numbers of Campylobacter detected in broiler carcass rinse samples. Eviscerated broiler carcasses were collected from the shackle line in a commercial processing plant immediately after passing through an inside/outside washer. Broiler carcasses were cut longitudinally into contralateral halves using a sanitized saw. Cecal contents from the same flock were collected, pooled, homogenized, and used to contaminate carcass halves. Paired carcass halves were divided into groups of eight each, and then cecal contents (2, 5, 10, 50, or 100 mg) were placed onto one randomly selected half of each carcass, while the corresponding half of the same broiler carcass received no cecal contents. Campylobacter counts from carcass halves with cecal contamination were compared to the uncontaminated halves of the same carcasses using a paired t test. Carcass halves with 5 mg or more of surface cecal contamination had significantly higher numbers of Campylobacter than those without (P < 0.01). Carcass halves contaminated with only 5 mg of cecal contents had an average of 3.3 log CFU Campylobacter per ml of rinse, while corresponding uncontaminated carcass halves had 2.6 log CFU Campylobacter per ml of rinse. These data indicate that even small (5 mg) amounts of cecal contents can cause a significant increase in the numbers of Campylobacter on eviscerated broiler carcasses. Therefore, it is important to keep such contamination to a minimum during processing.

ALGORITHM DEVELOPMENT WITH VISIBLE/NEAR-INFRARED SPECTRA FOR DETECTION OF POULTRY FECES AND INGESTA

The USDA Agricultural Research Service has developed a method and a hyperspectral imaging system to detect nfeces (from duodenum, ceca, and colon) and ingesta on poultry carcasses. The method first involves the use of multivariate ndata analysis on visible/near-infrared (Vis/NIR) reflectance spectra of fecal and uncontaminated skin samples for nclassification of contaminates and selection of key wavelengths. Four dominant wavelengths (434, 517, 565, and 628 nm) nwere identified by intensity of principal component (PC) loading weights. Key wavelengths were validated on hyperspectral nimages of contaminated broiler carcasses. Specifically, with a quotient of 565 nm/517 nm, 100% of the fecal contaminates nwere detected in a limited population of broilers fed a corn/soybean meal diet. The objectives of this research was to validate nthe 565 nm/517 nm quotient to classify uncontaminated skin from feces/ingesta with broilers fed corn, milo, or wheat diets nand to investigate the use of single-term linear regression (STLR) to select key wavelengths for classification. Feces (N = n369) and uncontaminated broiler breast skin (N = 96) were analyzed from 440 to 880 nm. The overall accuracy of detecting ncontamination for any type of feed with the 565 nm/517 nm quotient was 99% with 16 uncontaminated skin samples classified nas contaminates (false positive). STLR optimized a new quotient of 574 nm/588 nm, which classified 100% of contaminates ncorrectly with no false positives. The shift in the denominator from 517 to 588 nm is possibly due to greater fecal color nvariation from broilers fed wheat or milo. In addition, dividing by 588 nm minimized the effect of lightness (L*) on nclassification. The use of the STLR to scan the spectral data to find wavelengths correlated with the dependent variable is nan alternative to selecting key wavelengths based on the intensity of PC loading weights. Although models from Vis/NIR nspectroscopy and STLR performed well, they need to be validated on hyperspectral images of uncontaminated and ncontaminated carcasses.

Improved Hyperspectral Imaging System for Fecal Detection on Poultry Carcasses

The USDA Agricultural Research Service (ARS) has developed a hyperspectral imaging system to detect fecal contaminants on poultry carcasses. The system measures the intensity of reflected light energy from about 400 nm to 1000 nm and has been used as a research tool to identify key wavelengths for detecting contaminants. Selected wavelengths are to be used in a real-time multispectral system for contaminant detection. The ARS has reported that the ratio of reflectance images at 565 nm and 517 nm was able to identify fecal contaminants. However, this ratio alone also misclassified numerous non-fecal carcass features (false positives). Recent modifications to the system, including improved lighting, a new camera, a new spectrograph, and a new algorithm with an additional wavelength, have increased fecal detection accuracy while reducing the number of false positives. The new system was used to collect hyperspectral data on 56 stationary poultry carcasses. Carcasses were contaminated with both large and small spots of feces from the duodenum, ceca, and colon, and ingesta from the crop. A total of 1030 contaminants were applied to the carcasses. The new system and algorithm correctly identified over 99% of the contaminants with only 25 false positives. About a quarter of the carcasses had at least one false positive pixel.

Visible/NIR Spectroscopy for Characterizing Fecal Contamination of Chicken Carcasses

Zero tolerance of feces on the surfaces of meat and poultry carcasses during nslaughter was established as a standard to minimize the likelihood of microbial pathogens. nMicrobial pathogens can be transmitted to humans by consumption of contaminated meat and npoultry. Compliance with zero tolerance in meat processing establishments is currently verified nby visual observation. The objective of this study was to investigate the use of visible, near-infrared nreflectance spectroscopy as a method to discriminate between uncontaminated poultry nbreast skin and feces, and to select key wavelengths for use in a hyperpspectral system. Feces n(n = 102), uncontaminated poultry breast skin, and skin contaminated with fecal spots were nscanned from 400 to 2498 nm. The spectra were reduced by principal component (PC) nanalysis. The first four PCs explained 99.8% of the spectral variation. PC 1 was primarily nresponsible for the separation of uncontaminated skin from feces and for the separation of nuncontaminated skin from contaminated skin. . A Classification model was able to classify fecal ncontaminated skin from the spectral data with a success rate of 95% Key wavelengths were nidentified by intensity of loading weights at 628 nm for PC 1, 565 nm for PC 2 and 434 and 517 nm for PC 4. Visual assessment of loading weights suggests that discrimination was dependent non the spectral variation related to fecal color and myoglobin and/or hemoglobin content of the nuncontaminated breast skin.

Detection of Citrus Huanglongbing by Fourier Transform Infrared–Attenuated Total Reflection Spectroscopy

Citrus Huanglongbing (HLB, also known as citrus greening disease) was discovered in Florida in 2005 and is spreading rapidly amongst the citrus growing regions of the state. Detection via visual symptoms of the disease is not a long-term viable option. New techniques are being developed to test for the disease in its earlier presymptomatic stages. Fourier transform infrared-attenuated total reflection (FT-IR-ATR) spectroscopy is a candidate for rapid, inexpensive, early detection of the disease. The mid-infrared region of the spectrum reveals dramatic changes that take place in the infected leaves when compared to healthy non-infected leaves. The carbohydrates that give rise to peaks in the 900–1180 cm−1 range are reliable in distinguishing leaves from infected plants versus non-infected plants. A model based on chemometrics was developed using the spectra from 179 plants of known disease status. This model then correctly predicted the status of >95% of the plants tested.

VISIBLE/NIR SPECTROSCOPY FOR CHARACTERIZING FECAL CONTAMINATION OF CHICKEN CARCASSES

Zero tolerance of feces on the surfaces of meat and poultry carcasses during slaughter was established as a nstandard to minimize the likelihood of microbial pathogens. Microbial pathogens can be transmitted to humans by nconsumption of contaminated meat and poultry. Compliance with zero tolerance of feces in meat processing establishments nis currently verified by visual observation. The objective of this study was to investigate the use of visible, near–infrared nreflectance spectroscopy as a method to discriminate between uncontaminated poultry breast skin and feces, and to select nkey wavelengths for use in a hyperpspectral system. Feces (n = 102), uncontaminated poultry breast skin, and skin ncontaminated with fecal spots were analyzed from 400 to 950 nm. The spectra were reduced by principal component (PC) nanalysis. The first four PCs explained 99.8% of the spectral variation. PC 1 was primarily responsible for the separation of nuncontaminated skin from feces and for the separation of uncontaminated skin from contaminated skin. A classification model nwas developed and evaluated to classify fecal–contaminated skin from the spectral data with a success rate of 95%. Key nwavelengths were identified by intensity of loading weights at 628 nm for PC 1, 565 nm for PC 2, and 434 and 517 nm for nPC 4. Discrimination was dependent on the spectral variation related to fecal color and myoglobin and/or hemoglobin ncontent of the uncontaminated breast skin.

Detection by Hyperspectral Imaging of Shiga Toxin–Producing Escherichia coli Serogroups O26, O45, O103, O111, O121, and O145 on Rainbow Agar

The U.S. Department of Agriculture, Food Safety Inspection Service has determined that six non-O157 Shiga toxin-producing Escherichia coli (STEC) serogroups (O26, O45, O103, O111, O121, and O145) are adulterants in raw beef. Isolate and phenotypic discrimination of non-O157 STEC is problematic due to the lack of suitable agar media. The lack of distinct phenotypic color variation among non-O157serogroups cultured on chromogenic agar poses a challenge in selecting colonies for confirmation. In this study, visible and near-infrared hyperspectral imaging and chemometrics were used to detect and classify non-O157 STEC serogroups grown on Rainbow agar O157. The method was first developed by building spectral libraries for each serogroup obtained from ground-truth regions of interest representing the true identity of each pixel and thus each pure culture colony in the hyperspectral agar-plate image. The spectral library for the pure-culture non-O157 STEC consisted of 2,171 colonies, with spectra derived from 124,347 of pixels. The classification models for each serogroup were developed with a k nearest-neighbor classifier. The overall classification training accuracy at the colony level was 99%. The classifier was validated with ground beef enrichments artificially inoculated with 10, 50, and 100 CFU/ml STEC. The validation ground-truth regions of interest of the STEC target colonies consisted of 606 colonies, with 3,030 pixels of spectra. The overall classification accuracy was 98%. The average specificity of the method was 98% due to the low false-positive rate of 1.2%. The sensitivity ranged from 78 to 100% due to the false-negative rates of 22, 7, and 8% for O145, O45, and O26, respectively. This study showed the potential of visible and near-infrared hyperspectral imaging for detecting and classifying colonies of the six non-O157 STEC serogroups. The technique needs to be validated with bacterial cultures directly extracted from meat products and positive identification of colonies by using confirmatory tests such as latex agglutination tests or PCR.