Content uploaded by Chun-Chieh Yang
Author content
All content in this area was uploaded by Chun-Chieh Yang on May 20, 2015
Content may be subject to copyright.
Applied Engineering in Agriculture
Vol. 24(4): 475‐485 2008 American Society of Agricultural and Biological Engineers ISSN 0883-8542 475
HIGH THROUGHPUT SPECTRAL IMAGING SYSTEM FOR
WHOLESOMENESS INSPECTION OF CHICKEN
K. Chao, C.‐C. Yang, M. S. Kim, D. E. Chan
ABSTRACT. An online line‐scan imaging system containing an electron‐multiplying charge‐coupled device detector and line‐
scan spectrograph was used for identifying wholesome and unwholesome freshly slaughtered chicken carcasses on high‐speed
commercial chicken processing lines. Hyperspectral images were acquired using the line‐scan imaging system for
5549 wholesome chicken carcasses and 93 unwholesome chicken carcasses on a commercial processing line, for analysis to
optimize ROI size and location and to determine the key intensity waveband and ratio wavebands to be used for online
inspection. Multispectral imaging algorithms were developed for real‐time online identification of wholesome and
unwholesome chicken carcasses. The imaging system inspected over 100,000 chickens on a commercial 140‐bpm kill line
during continuous operation and achieved over 99% accuracy in identifying wholesome chickens and over 96% accuracy
in identifying unwholesome chickens. A system of this type can perform food safety inspection tasks accurately and with less
variation in performance at high speeds (e.g., at least 140 bpm), and help poultry plants to improve production efficiency and
satisfy increasing consumer demand for poultry products.
Keywords. Food safety, HACCP, Poultry, Spectral imaging.
he 1957 Poultry Product Inspection Act mandated
postmortem inspection of every bird carcass pro‐
cessed by a commercial facility. Since then, USDA
inspectors have conducted on‐site organoleptic in‐
spection of all chickens processed at U.S. poultry plants for
indications of diseases or defects. Inspectors of the USDA
Food Safety and Inspection Service (FSIS) examine by sight
and by touch the body, the inner body cavity surfaces, and the
internal organs of every chicken carcass during processing
operations.
With the 1996 final rule on Pathogen Reduction and
Hazard Analysis and Critical Control Point (HACCP)
systems (USDA, 1996), FSIS implemented the HACCP and
Pathogen Reduction programs in meat and poultry process‐
ing plants throughout the country to prevent food safety
hazards. More recently, FSIS has also been testing the
HACCP‐Based Inspection Models Project (HIMP) in a small
number of volunteer plants (USDA, 1997). HIMP require‐
ments include zero tolerance for unwholesome chickens
exhibiting symptoms of “septox” – a condition of either
septicemia or toxemia. Wholesome chickens do not exhibit
Submitted for review in November 2007 as manuscript number FPE
7282; approved for publication by the Food & Process Engineering
Institute of ASABE in April 2008.
Mention of a product or specific equipment does not constitute a
guarantee or warranty by the U.S. Department of Agriculture and does not
imply its approval to the exclusion of other products that may also be
suitable.
The authors are Kuanglin Chao, Research Scientist, USDA‐ARS Food
Safety Laboratory, Beltsville, Maryland; Chun‐Chieh Yang, ASAE
Member, Research Associate, Department of Biosystems and Agricultural
Engineering, University of Kentucky, Lexington; Moon S. Kim, Research
Physical Scientist, and Diane E. Chan, Agricultural Engineer, USDA‐ARS
Food Safety Laboratory, Beltsville, Maryland. Corresponding author:
Kuanglin Chao, Food Safety Laboratory, USDA‐ARS, Building 303,
BARC‐East, 10300 Baltimore Ave., Beltsville, MD 20705; phone:
301‐504‐8450; fax: 301‐504‐9466; e‐mail: kevin.chao@ars.usda.gov.
symptoms of “septox” and the unwholesome birds must be
removed from the processing line.
Septicemia is caused by the presence of pathogenic
microorganisms or their toxins in the bloodstream, and
toxemia results from toxins produced by cells at a localized
infection or from the growth of microorganisms. Septox birds
are considered to be unwholesome and USDA inspectors
remove these unwholesome birds from the processing lines
during their bird‐by‐bird inspections, which can, by law, be
conducted at a maximum speed of 35 birds per minute (bpm)
for an individual inspector. The inspection process is subject
to human variability, and the inspection speed restricts the
maximum possible output for the processing plants while
also making inspectors prone to fatigue and repetitive injury
problems. This limit on production throughput, combined
with increases in chicken consumption and demand over the
past two decades, places additional pressure on both chicken
production and the food safety inspection system. U.S.
poultry plants now process over 8.8 billion broilers annually
(USDA, 2007). During processing at a typical U.S. poultry
plant, birds are first slaughtered on kill lines and then
transferred to evisceration lines on which inspection stations
are located. Commercial evisceration lines in the United
States currently may be operated at speeds up to 140 bpm;
however, such processing lines require up to four inspection
stations, each with an FSIS inspector to conduct bird‐by‐bird
inspection at the 35‐bpm speed limit.
Machine vision technologies have been developed to
address a variety of food and agricultural processing
applications. Various sensing techniques such as RGB
(red/green/blue) color imaging, visible and near‐infrared
(Vis/NIR) spectroscopy and imaging, fluorescence spectros‐
copy and imaging, and X‐ray imaging, have been investi‐
gated for potential use in food processing and online
inspection applications (Daley et al., 1994; Delwiche, 2003;
Jing et al., 2003; Kim et al., 2003; Lu, 2007; Windham et al.,
2003).
T
476 APPLIED ENGINEERING IN AGRICULTURE
A variety of methods for imaging whole chicken carcasses
and chicken viscera/organs have been investigated for use in
food safety inspection of poultry. RGB color imaging of
chicken spleens, hearts, and livers was found capable of
identifying poultry disease conditions including leucosis,
septicemia, airsacculitis, and ascites in the laboratory (Tao
et al., 1998; Chao et al., 1999), but these methods required
precise presentation of the visceral organs and thus were
unsuitable for conventional poultry processing lines. A
two‐camera system using two wavebands in the visible
spectrum for whole‐carcass imaging was able to separate
90% of wholesome and unwholesome chickens at processing
line speeds up to 70 bpm, but was not feasible for higher
speed processing (Park and Chen, 2000; Chao et al., 2002).
Thus there remains a need to develop imaging systems
that can inspect chickens for wholesomeness in commercial
processing lines which operate at speeds of at least 140 bpm.
The objectives of this study were to provide an online
line‐scan imaging system capable of both hyperspectral and
multispectral visible/near‐infrared reflectance, and a method
of using the system to inspect freshly slaughtered chickens on
a processing line for wholesomeness and unwholesomeness.
MATERIALS AND METHODS
HIGH THROUGHPUT SPECTRAL IMAGING SYSTEM
The spectral imaging system (fig. 1) consisted of an
Electron‐Multiplying Charge‐Coupled‐Device (EMCCD)
camera, an imaging spectrograph, a C‐mount lens, and a pair
of high power, broad‐spectrum white light‐emitting‐diode
(LED) line lights. The EMCCD camera (PhotonMAX 512b,
Roper Scientific, Inc., Trenton, N.J.) has approximately 512
× 512 pixels and is thermoelectrically cooled to approxi‐
mately ‐70°C (via a three‐stage Peltier device). An imaging
spectrograph (ImSpector V10 OEM, Specim/Spectral Imag‐
ing Ltd., Oulu, Finland), and a C‐mount lens (Rainbow
A
BDC
E
Figure 1. A photograph of the hyperspectral/multispectral imaging
inspection system on a commercial chicken processing line. A:
Electron‐Multiplying Charge‐Coupled‐Device (EMCCD) camera; B:
line‐scan spectrograph; C: lens assembly; D: LED lighting system; E:
data processing unit.
S6x11, International Space Optics, S.A., Irvine, Calif.) are
attached to the EMCCD imaging device. The spectrograph
aperture slit of approximately 50 microns limits the
instantaneous field of view (IFOV) of the imaging system to
a thin line. Light from the linear IFOV is dispersed by a
prism‐grating‐prism line‐scan spectrograph and projected
onto the EMCCD imaging device. The spectrograph creates
a two‐dimensional (spatial and spectral) image for each
line‐scan, with the spatial dimension along the horizontal
axis and the spectral dimension along the vertical axis of the
EMCCD imaging device. The imaging device is coupled
with a 16‐bit digitizer (CCI‐23, Andor Technology Limited,
South Windsor, Conn.) with a pixel‐readout rate of
approximately 10 MHz. The digitizer performs rapid
analog‐to‐digital conversion of the image data for each
hyperspectral or multispectral line‐scan image. These data
are then processed by the computer for image analysis and
classification of wholesome and unwholesome pixels in the
line‐scan images.
The spectral imaging system requires calibration before
line‐scan images can be acquired. Re‐calibration is generally
not required unless the physical arrangement of the
components of the imaging system is disturbed. The first step
in the calibration process was to cool the imaging system to
its operating temperature of ‐70°C. The next step was to set
image binning, which is determined by the spectral distribu‐
tion of useful wavelengths and the size of spatial image
features to be processed for the application. The original
image size, 512 × 512 pixels, was reduced by 1 × 4 binning
to result in line‐scan images with a spatial resolution of
512 pixels (512 divided by 1) and a spectral resolution of
128 pixels (512 divided by 4) in the spectral dimension. The
binning process adds together photons from adjacent pixels
in the detector array and was performed by the shift register
of the EMCCD imaging device. This produced a reduced
number of pixels to be digitized by the 16‐bit A/D PCI board
for the computer to process. Reducing total pixel readout
time decreased the acquisition time of each line‐scan image,
which allowed higher image acquisition speed for the
EMCCD imaging device. Because the useful spectrum of
light did not span the entire width of the EMCCD detector,
the first 20 and last 53 spectral bands were discarded,
resulting in a line‐scan image size of 512 × 55 pixels.
The next step in the calibration process was spectral
waveband calibration that identified each spectral channel
with a specific wavelength. A neon‐mercury calibration lamp
(Oriel Instruments, Stratford, Conn.) was utilized for spectral
calibration; the mercury peaks at 435.84 and 546.07 nm were
found to correspond to the 8th and 25th bands, respectively,
and neon peaks at 614.31, 640.23, 703.24, and 724.52 nm
corresponded to the 35th, 39th, 49th, and 52nd bands,
respectively. The following second‐order polynomial regres‐
sion was calculated from the reference wavelength peaks of
the mercury and neon spectra to calibrate the spectral axis:
70.39303.601.0 2+×+×=λ cc nn (r2 = 0.9999) (1)
where l is the wavelength (nm), and nc is the spectral channel
number. The hyperspectral imaging data ranged from
399.94 nm (the first band) to 750.42 nm (the 55th band) with
an average bandwidth of 6.02 nm. The distance between the
lens and IFOV target area was 914 mm, with the LED line
lights illuminating the IFOV target area from a distance of
477Vol. 24(4): 475‐485
214 mm. The IFOV spanned 177.8 mm, which translated into
512 spatial pixels, with each pixel representing an area of
0.12 mm2.
Following system calibration, the spectral imaging sys‐
tem was ready to use for the acquisition of reference line‐scan
images. Prior to acquiring hyperspectral chicken images,
acquisition of a white reference image was performed using
a 99% diffuse reflectance standard (Spectralon, LabSphere,
Inc., North Sutton, N.H.) illuminated by the lighting system;
acquisition of a dark reference image was performed by
acquiring an image with the lens covered by a non‐reflective
opaque black fabric. These reference line‐scan images were
used to calculate the pixel‐based relative reflectance for raw
line‐scan images as follows:
DR
DI
I−
−
=0 (2)
where I is the relative reflectance, I0 is the raw reflectance,
D is the dark reference, and R is the white reference.
PROCEDURES
Following spectral and spatial calibration of the imaging
system, hyperspectral line‐scan images were acquired for
5549 wholesome chicken carcasses and 93 unwholesome
chicken carcasses on a 140‐bpm commercial processing line
in March 2007. The wholesome or unwholesome condition
of the birds on the line was identified by an FSIS veterinarian
who observed the birds before they passed through the
illuminated IFOV, where the imaging system acquired
55‐band hyperspectral data for the chicken carcasses. These
hyperspectral images were analyzed for Region of Interest
(ROI) optimization and selection of one key wavelength and
two ratio wavebands based on average spectral differences
between wholesome and unwholesome birds.
Using only the key wavelength and ratio wavebands that
were selected by the hyperspectral analysis described above,
random track mode on the same imaging system was
implemented for multispectral inspection. LabView software
(National Instruments Corp., Austin, Tex.) was used to
develop software modules for detecting the starting point
(SP) and ending point (EP) of each bird, and for implement‐
ing classification algorithms based on fuzzy logic. During
two 8‐h shifts in July 2007, the imaging system conducted
multispectral inspection for over 100,000 birds at a commer‐
cial processing line. A FSIS veterinary medical officer
identified bird conditions during several 30‐ to 40‐min
periods for verification of system performance.
HYPERSPECTRAL IMAGE ANALYSIS
Analysis of the hyperspectral relative reflectance images
began with removal of the background. A relative reflectance
threshold value of 0.1 was set for the 620‐nm waveband. For
any spatial pixel in the hyperspectral reflectance image, the
pixel was identified as a background pixel if its reflectance
at 620 nm was lower than the 0.1 threshold value. The value
of the relative reflectance for every pixel identified as a
background pixel was re‐assigned to be zero, thus removing
these pixels from further image analysis.
The background‐removed relative reflectance line‐scan
images were compiled to form hyperspectral image cubes of
entire wholesome and unwholesome chicken carcasses.
Using MATLAB software (MathWorks, Natick, Mass.), the
hyperspectral chicken images were then analyzed to opti‐
mize the spatial ROI within the chicken images. The
optimized ROI was one which provided the greatest spectral
difference between averaged wholesome pixels and averaged
unwholesome pixels across all 55 wavebands, which was
obtained as follows. Within a bird image, the potential ROI
area spanned from an upper border across the breast of the
bird to a lower border at the lowest non‐background spatial
pixel in each line scan, or to the last (512th) spatial pixel if
there were no background pixels present at the lower edge of
the image. The average relative reflectance spectrum was
calculated across all ROI pixels for all wholesome chicken
images, and the average relative reflectance spectrum was
calculated across all ROI pixels for all unwholesome chicken
images. The difference spectrum between the wholesome
and unwholesome average spectra was calculated. This
calculation was performed for potential ROIs of varying size,
as defined by the number of ROI pixels and their vertical
coordinate locations within each line‐scan, to optimize the
ROI size and location by selecting the ROI that produced the
greatest maximum value in its difference spectrum. Using the
optimized ROI, the waveband corresponding to the greatest
spectral difference between averaged wholesome chicken
pixels and averaged unwholesome chicken pixels was
identified as a key waveband for differentiation of whole‐
some and unwholesome chicken carcasses by relative
reflectance intensity. Again using the optimized ROI, the
average wholesome and average unwholesome spectra were
analyzed and potential two‐waveband ratios were identified
as several ratios using wavebands at which the average
wholesome and average unwholesome chicken pixel spectra
showed local maxima and local minima. The value of each
potential band ratio was calculated for the average whole‐
some chicken pixels and for the average unwholesome
chicken pixels. The two‐waveband ratio showing the greatest
difference in ratio value between average wholesome and
average unwholesome chicken pixels was identified for use
in differentiating wholesome and unwholesome chicken
carcasses. Multispectral imaging inspection used the key
wavelength and the two‐waveband ratio to differentiate
between wholesome and unwholesome chicken carcasses.
MULTISPECTRAL INSPECTION OF CHICKEN CARCASS
Effective multispectral imaging inspection of wholesome
and unwholesome chicken carcasses on a processing line
required the capacity for detecting individual bird carcasses,
classifying the condition of the chicken carcass, and
generating a corresponding output useful for process control,
at speeds compatible with online processing line operations.
LabVIEW 8.0 (National Instruments Corp., Austin, Tex.)
was used to control the spectral imaging system to perform
the tasks required for multispectral inspection of chicken
carcasses on a poultry processing line. The line‐by‐line mode
of operation was the basis of the following algorithm that was
developed to detect the entry of a bird carcass into the IFOV.
Figure 2 shows the line‐by‐line algorithm for multispec‐
tral inspection to detect and classify wholesome and
unwholesome chicken carcasses on a processing line. First,
a line‐scan image was acquired that contains only raw
reflectance values at the two key wavebands needed for
intensity and ratio differentiation, the raw reflectance data
was converted into relative reflectance data, and background
478 APPLIED ENGINEERING IN AGRICULTURE
Figure 2. A flowchart of the method for online multispectral line‐scan imaging inspection of chickens for wholesomeness.
pixels were removed from the image (fig. 2, Box 2.1). The
line‐scan image was checked for the presence of the Starting
Point (SP) of a new bird (fig. 2, Box 2.2); if no SP was present,
no further analysis was performed for this line‐scan image
and a new line‐scan image was acquired. If the line‐scan was
found to contain an SP, then the ROI pixels were located
(fig. 2, Box 2.3) and the decision output value, Do, was
calculated for each pixel in the ROI of the line‐scan image
fig. 2, Box 2.4). With each new line‐scan image acquired
(fig. 2, Box 2.5), the ROI pixels were located, and the
decision output value of Do was calculated for each pixel,
until the Ending Point (EP) was detected (fig. 2, Box 2.6),
indicating no additional line‐scan images to be analyzed for
the bird carcass. The average Do value for the bird was
calculated (fig. 2, Box 2.9) and compared to the threshold
value (fig. 2, Box 2.10) for the final determination of
wholesomeness or unwholesomeness for the bird carcass
(fig. 2, Boxes 2.11 and 2.12).
With the acquisition of each new line‐scan image at the
start of the detection algorithm, (fig. 2, Box 2.1), the relative
reflectance at 620 nm was examined for each of the first
(uppermost) 256 pixels of the line‐scan image. The value of
the relative reflectance at 620 nm was always at a low‐
intensity (below 0.1) for these pixels when there was no
chicken carcass present in the IFOV. When the relative
reflectance at 620 nm increased above 0.1 for any single pixel
among the uppermost 256 pixels in the line‐scan image, this
indicated that a chicken carcass had entered the IFOV. This
indication assumed that the inverted chicken carcass was
correctly hung from the processing line shackle by both legs
and that the entry of the first leg into the IFOV was triggering
the detection. The detection algorithm examined only the
uppermost 256 pixels in order to disregard carcass wings
which were always overlapped between adjacent carcasses
on the processing line. After detecting a line‐scan image with
a single pixel among the uppermost 256 exhibiting relative
479Vol. 24(4): 475‐485
reflectance greater than 0.1 at 620 nm, the subsequent
line‐scan images were monitored as additional pixels within
the 256 pixels began showing relative reflectance values
greater than 0.1 (fig. 2, Box 2.2). Between the first detected
pixel and the 256th pixel, pixels below the first detected pixel
began increasing in relative reflectance as the chicken
continues to move across the field of view. There would
eventually be a line‐scan image with one (or several)
remaining low‐intensity pixel located below the first de‐
tected pixel, and above or at the 256th pixel, which was
immediately followed by another line‐scan in which the
previous line‐scan's last low‐intensity pixel(s) had increased
above 0.1. The last low‐intensity pixel, or the pixel in the
center of the last contiguous group of remaining low‐
intensity pixels, was identified as the Starting Point (SP) of
the bird carcass and represented the junction between the
thigh and the abdomen on the leading edge of the carcass.
Similar to the above algorithm, the following algorithm
was developed to detect the last relevant line‐scan image for
each bird as it passed through the IFOV (fig. 2, Box 2.6).
After the SP was detected, each subsequent line‐scan image
was analyzed to determine if the relative reflectance intensity
at 620 nm for the pixel matching the vertical coordinate of the
SP was above or below 0.1. When a line‐scan image was
acquired for which that pixel had a relative reflectance
intensity at 620 nm that was below 0.1, this pixel was
identified as the Ending Point (EP) of the bird carcass,
indicating that the main body of the bird had already passed
through the IFOV and no further line‐scans should be
analyzed for that specific bird carcass.
After the initial identification of the SP for a bird carcass,
the line‐scan image containing the SP and subsequent
line‐scan images up to the one containing the EP were
analyzed, line‐by‐line (fig. 2, Boxes 2.3 through 2.8), using
the following algorithm to classify the bird carcass. For each
line‐scan image, fuzzy logic membership functions were
used to produce two decision outputs for each non‐
background pixel in the line‐scan image that was located
within the ROI, using the ROI and waveband parameters
previously determined through hyperspectral imaging analy‐
sis. For each pixel, two fuzzy logic membership functions
were used to generate wholesome and unwholesome fuzzy
membership values w1 and u1, corresponding to wholesome
and unwholesome chickens, from the key wavelength
reflectance intensity value for that pixel. Two additional
fuzzy logic membership functions were used to generate
wholesome and unwholesome fuzzy membership values w2
and u2, corresponding to wholesome and unwholesome
chickens, from the ratio value for that pixel. The fuzzy
inference engine executed a min‐max operation (Chao et al.,
1999) to obtain a decision output Do for each pixel based on
the n membership functions as follows, where n is the number
of criteria input used (in this case, n = 2):
Do = max [min {w1 0 wn }, min {u1 0 un }]
For each pixel, the value of Do was in the range between
0 and 1, where 0 indicates 100% possibility of wholesome‐
ness and 1 indicated 100% possibility of unwholesomeness.
When the EP for that bird carcass was encountered, the
average Do value for all ROI pixels for that bird was
calculated (fig. 2, Box 2.9). The bird carcass was identified
as being unwholesome if the average Do value was greater
than 0.6; otherwise the chicken carcass was identified as
being wholesome (fig. 2, Boxes 2.10, 2.11, 2.12).
RESULTS AND DISCUSSION
ANALYSIS OF IN‐PLANT HYPERSPECTRAL IMAGES
The hyperspectral images were analyzed to optimize the
ROI size and location and the key wavebands for differenti‐
ation by reflectance intensity and by waveband ratio.
Figure 3 shows a contour image of two examples of chicken
carcasses with the SP and EP marked and connected by a line
on each. The possible size and location of the ROI is
described by parameters m and n, which extended below the
Figure 3. Contour images of two chicken carcasses marked with example locations of the SP, EP, m, and n parameters used for locating the region of
interest.
480 APPLIED ENGINEERING IN AGRICULTURE
SP‐EP line. The values of m and n indicated, by percentage
of the pixel length between the SP‐EP line and the furthest
non‐background pixel below the SP‐EP line, the location of
the upper and lower ROI borders. The possible locations of
the upper ROI border ranged between a 10% and 40%
distance below the SP‐EP line, and the possible locations of
the lower ROI border range between a 60% and 90% distance
below the SP‐EP line.
For each possible ROI, the average spectrum across all
ROI pixels from the 5549 wholesome chicken carcasses, and
the average spectrum across all ROI pixels from the
93 unwholesome chicken carcasses, were calculated. The
difference between the average wholesome and average
unwholesome value at each of the 55 bands was calculated
and their range for each possible ROI is shown in figure 4.
Because the 40% to 60% ROI showed the range with the
greatest difference values between the average wholesome
and unwholesome spectra, this ROI was considered the
optimized ROI to be used for multispectral inspection. As
shown in figure 5, the 30th band showed the greatest
Figure 4. The range, for possible ROIs, of difference values between average wholesome and average unwholesome chicken spectra, for optimizing
the ROI to be used for inspection of chickens.
Figure 5. The averaged wholesome and average unwholesome chicken spectra, highlighting the 580 nm key waveband that can be used for intensity-
based differentiation of wholesome and unwholesome chickens.
481Vol. 24(4): 475‐485
difference between the average wholesome and the average
unwholesome spectra from among all 55 bands for the
optimized ROI; this band, corresponding to 580 nm, was
selected as the key waveband to be used for intensity‐based
differentiation of wholesome and unwholesome chicken
carcasses.
Figure 6 shows the average wholesome and average
unwholesome chicken spectra, marked with the wavebands
that were investigated for differentiation of wholesome and
unwholesome chicken carcasses by a two‐waveband ratio.
The average wholesome and average unwholesome ratio
values were calculated for three possible two‐waveband
ratios, using wavebands at 440 and 460 nm, 500 and 540 nm,
and 580 and 620 nm. The following differences were then
calculated:
W440/W460 – U440/U460 = 0.003461
W500/W540 – U500/U540 = 0.038602
W580/W620 – U580/U620 = 0.115535
The last ratio, using the 580‐ and 620‐nm wavebands,
showed the greatest difference between the average whole‐
some and average unwholesome chicken spectra and was
thus selected for use in differentiation by two‐waveband
ratio.
The optimized ROI and key wavebands determined from
the hyperspectral data analysis were used for multispectral
inspection of over 100,000 chickens on a 140‐bpm process‐
ing line during two 8‐h shifts at a commercial poultry plant.
Figure 7 shows examples of chicken images highlighting the
ROI that was used for online inspection. The inspection
program specifically determined the 40% to 60% ROI for
each bird, which was clearly affected by the size and position
of the bird. The ROI was a regular rectangular area for a bird
whose body extended past the lower edge of the image, such
as the first bird in figure 7. For other birds, the presence of
background pixels near the lower edge of the image resulted
in irregularly shaped ROIs.
Table 1 shows the mean and standard deviation values for
relative reflectance at 580 nm for wholesome and unwhole‐
some birds in three data subsets drawn from the hyperspectral
data analysis using the 40% to 60% ROI and each of the two
Figure 6. The averaged wholesome and average unwholesome chicken spectra, for possible key wavebands that can be used for two‐waveband ratio
differentiation of wholesome and unwholesome chickens.
Figure 7. Nine chicken images with the optimized ROI highlighted on each chicken.
482 APPLIED ENGINEERING IN AGRICULTURE
Table 1. Mean and standard deviation (SD) values for reflectance
intensity at 580 nm for wholesome and unwholesome chicken images.
Wholesome Unwholesome
Mean SD Mean SD
Hyperspectral analysis 0.378 0.088 0.243 0.076
Inspection shift 1 0.419 0.115 0.253 0.069
Inspection shift 2 0.398 0.083 0.253 0.075
Table 2. Mean and standard deviation (SD) values for two‐waveband
ratio using 580 and 620 nm for wholesome and
unwholesome chicken images.
Wholesome Unwholesome
Mean SD Mean SD
Hyperspectral analysis 0.948 0.037 0.904 0.052
Inspection shift 1 0.958 0.033 0.918 0.048
Inspection shift 2 0.941 0.038 0.919 0.048
inspection shifts. Table 2 shows the mean and standard
deviation values for the two‐waveband ratio using 580 and
620 nm for wholesome and unwholesome birds for the same
three data subsets. Paired t‐tests showed no significant
differences (P = 0.05) between the three data sets for the
wholesome means, and similarly no significant difference
between the three data sets for the unwholesome means. This
demonstrates that when the spectral imaging system is
appropriately and consistently operated to maintain proper
distance and illumination conditions, hyperspectral data
collected by the system can be appropriately used for
multispectral inspection conducted at different times and
locations.
IN‐PLANT TESTING OF MULTISPECTRAL INSPECTION
For multispectral classification, fuzzy logic membership
functions were built based on the mean and standard
deviation values for the 580‐nm key waveband from the
hyperspectral analysis data subset, and on the mean and
standard deviation values for the 580‐ and 620‐nm two‐
waveband ratio, again from the hyperspectral analysis data
subset. Figure 8 shows the structure of the fuzzy logic
membership functions. These functions were used to classify
each ROI pixel within an image as either wholesome or
unwholesome, by using each pixel's 580‐nm intensity value
and its ratio value using 580 and 620 nm as inputs to obtain
a decision output value Do between 0 and 1. The average Do
value for a bird was used to determine a wholesome or
unwholesome assignment by comparison with a threshold
value. Figure 9 first shows a masked image of nine chickens
with all ROI pixels highlighted for each chicken (top), and
then another image highlighting only those ROI pixels that
were classified as wholesome pixels (bottom), i.e., Do values
of individual pixels were each compared to the 0.6 threshold
value. The fourth chicken from the left is an unwholesome
bird and all of its ROI pixels were identified as unwholesome,
consequently not appearing in the second image (bottom).
Figures 10 and 11 show scatterplots of the imaging
system's decision outputs against the number of ROI pixels
for each chicken imaged during inspection shifts 1 and 2. The
total numbers of wholesome and unwholesome chickens
identified by the system are shown in table 3, compared with
numbers drawn from FSIS tally sheets created by three
inspection stations on the same processing line during those
two inspection shifts. Although direct bird‐to‐bird compari‐
son between the imaging inspection system and the inspec‐
tors was not performed, the percentages indicated that the
relative numbers of wholesome and unwholesome identified
by the imaging inspection system and by the processing line
inspectors were similar.
Figure 8. The structure of the fuzzy logic membership functions which use the intensity‐based input value (I580) and ratio‐based input value (I580/620)
to create pixel‐based decision outputs for wholesomeness classification. Wm: wholesome mean; Um: unwholesome mean; Wm‐sd : wholesome mean mi‐
nus one standard deviation; Um+sd: wholesome mean plus one standard deviation.
483Vol. 24(4): 475‐485
Figure 9. A masked image (top) of nine chickens that highlights the ROI pixels to be analyzed for each chicken, and a second image (bottom) highlighting
the ROI pixels for each chicken that were classified as wholesome.
Figure 10. A scatterplot graph showing the distribution of chicken carcasses imaged during inspection shift 1, by the number of ROI pixels and the final
decision output for each chicken.
484 APPLIED ENGINEERING IN AGRICULTURE
Figure 11. A scatterplot graph showing the distribution of chicken carcasses imaged during inspection shift 2, by the number of ROI pixels and the final
decision output for each chicken.
Table 3. Wholesome and unwholesome birds identified during inspection shifts by processing line inspectors and by the imaging inspection system.
Line Inspectors Imaging Inspection System
Wholesome Unwholesome Total Wholesome Unwholesome Total
Shift 1 53563 (99.84%) 84 (0.16%) 53647 (100%) 45305 (99.37%) 288 (0.63%) 45593 (100%)
Shift 2 64972 (99.89%) 71 (0.11%) 65043 (100%) 60922 (99.84%) 98 (0.16%) 61020 (100%)
A veterinarian also conducted several periods of system
verification, each lasting approximately 30 to 40 min. The
veterinarian conducted bird‐by‐bird observation of chicken
carcasses immediately before they entered the IFOV of the
imaging system. The imaging system output was observed
for agreement with the veterinarian's identifications. The
veterinarian observed 16,174 wholesome birds and 43 un‐
wholesome birds over 4 verification periods during inspec‐
tion shift 1. Of these birds, the imaging system incorrectly
identified only 118 wholesome birds (99.27% correct) and
2 unwholesome birds (95.35% correct). The veterinarian
observed 27,626 wholesome birds and 35 unwholesome
birds over 6 verification periods during inspection shift 2. Of
these birds, the imaging system incorrectly identified only
46 wholesome birds (99.83% correct) and 1 unwholesome
bird (97.14% correct). These results, together with the
percentages listed in table 3, strongly suggest that the
imaging inspection system can perform successfully on a
commercial poultry processing line.
For multispectral inspection conducted on a 140‐bpm
processing line was performed for this study, the imaging
system acquired about 30 to 40 line‐scan images between the
SP and EP for each chicken inspected. Previous testing of the
imaging system on a 70‐bpm processing line (Chao et al.,
2007) demonstrated similar performance in identification of
wholesome and unwholesome birds with the analysis of
about 70‐ to 80‐line‐scan images for each chicken. Because
the unwholesome birds exhibit a systemic unwholesome
condition affecting the entire body of the bird, this line‐scan
imaging system is able to identify such birds at even higher
speeds; on a 200‐bpm processing line, for example, the
system would perform similarly in identifying wholesome
and unwholesome birds by analyzing about 20‐ to 25‐line‐
scan images for each chicken.
CONCLUSIONS
An online line‐scan imaging system capable of both
hyperspectral and multispectral visible/near‐infrared reflec‐
tance was developed to inspect freshly slaughtered chickens
on a processing line for wholesomeness. In‐plant testing
results indicated that the imaging inspection system achieved
over 99% accuracy in identifying wholesome chickens and
over 96% accuracy in identifying unwholesome diseased
chickens. With appropriate methods of hyperspectral analy‐
sis and algorithms for online image processing, a machine
vision system utilizing an EMCCD camera for multispectral
inspection can satisfy both the food safety performance
485Vol. 24(4): 475‐485
standards and the high‐speed production requirements (e.g.,
at least 140 bpm) of commercial chicken processing. Use of
the imaging system may also help to improve product safety
by preventing most unwholesome birds from entering the
evisceration line and lowering the risk of cross‐
contamination. In addition, use of the system can help reduce
the routine workload imposed upon FSIS inspectors working
in HIMP processing plants, allowing them opportunities to
perform more meaningful tasks to enhance the public health
safety for poultry products.
REFERENCES
Chao, K., Y. R. Chen, H. L. Early, and B. Park. 1999. Color image
classification systems for poultry viscera inspection. Applied
Engineering in Agriculture 15(4): 363‐369.
Chao, K., Y. R. Chen, W. R. Hruschka, and F. B. Gwozdz. 2002.
On‐line inspection of poultry carcasses by a dual‐camera system.
J. of Food Engineering 51(3): 185‐192.
Daley, W., R. Carey, and C. Thompson. 1994. Real‐time color
grading and defect detection of food products. In Optics in
Agriculture, Forestry, and Biological Processing, SPIE 2345:
403‐411. Bellingham, Wash.: SPIE.
Delwiche, S. R. 2003. Classification of scab‐ and other
mold‐damaged wheat kernels by near‐infrared reflectance
spectroscopy. Transactions of the ASAE 46(3): 731‐738.
Jing, H., X. Chen, Y. Tao. 2003. Analysis of factors influencing the
mapping accuracy of x‐ray and laser range images in a bone
fragment inspecting system. ASAE Paper No. 033086. St.
Joseph, Mich.: ASAE.
Kim, M. S., A. M., Lefcourt, and Y. R. Chen. 2003. Multispectral
laser‐induced fluorescence imaging system for large biological
samples. Applied Optics 42(19): 3927‐2934.
Lu, R. 2007. Nondestructive measurement of firmness and soluble
solids content for apple fruit using hyperspectral scattering
images. Sensing and Instrumentation for Food Quality and
Safety 1(1): 19‐27.
Park, B., and Y. R. Chen. 2000. Real‐time dual‐wavelength image
processing for poultry safety inspection. J. Food Proc. Engr.
23(5): 329‐351.
Tao, Y., J. Shao, K. Skeeles, and Y. R. Chen. 1998. Detection of
poultry disease and enlarged spleen by computer imaging.
ASAE Paper No. 983017. St. Joseph, Mich.: ASAE.
Windham, W. R., D. P. Smith, B. Park, K. C. Lawrence, and P. W.
Feldner. 2003. Algorithm development with
visible/near‐infrared spectra for detection of poultry feces and
ingesta. Transactions of the ASAE 46(6): 1733‐1738.
USDA. 1996. Pathogen reduction: Hazard analysis and critical
control point (HACCP) systems. Final rule. Fed. Reg. 61:
38805‐38989.
USDA. 1997. HACCP‐based inspection models project (HIMP).
Proposed rule. Fed. Reg. 62:31553‐31562.
USDA. 2007. Poultry production and value – 2006 summary.
National Agricultural Statistics Service. Washington, D.C.
486 APPLIED ENGINEERING IN AGRICULTURE
