Journal of Food Engineering 391 (2025) 112459

Contents lists available at ScienceDirect

Journal of Food Engineering

journal homepage: www.elsevier.com/locate/jfoodeng

Assessment of woody breast in broiler breast fillets using

structured-illumination reflectance imaging coupled with
surface profilometry
Jiaxu Cai , Yuzhen Lu *
Department of Biosystems & Agricultural Engineering, Michigan State University, East Lansing, MI, 48824, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Woody breast (WB) myopathy impairs the quality and marketability of poultry products, leading to significant
Sinusoidal illumination economic losses for poultry industries worldwide due to product downgrading and consumer complaints. WB-
Phase analysis affected broiler breast fillets are characterized by abnormal tissue hardness, muscle rigidity, and irregular
Machine learning
shape. Manual evaluation based on tactile palpation and visual examination is the current practice for WB
Poultry
Woody breast
assessment at poultry processing facilities, but it is subjective, labor-intensive, and may induce contamination.
Structured-illumination reflectance imaging (SIRI), which is capable of depth-resolved tissue characterization
and three-dimensional sample topography, has emerged a potential means for poultry defect detection, as
opposed to conventional uniform illumination-based imaging. This study presents a novel effort to assess WB in
broiler meat using SIRI coupled with phase-measuring profilometry. An in-house assembled, broadband SIRI
platform with phase-shifted sinusoidal illumination patterns at different spatial frequencies was used to acquire
images from normal and WB-affected boneless chicken fillets. Acquired pattern images at each spatial frequency
were demodulated into 1) one phase difference image that depicts sample surface geometry and 2) two intensity
images [i.e., direct component (DC) and amplitude component (AC)]. Hand-crafted geometric and textural
features were extracted from the phase difference and intensity images respectively, and used for differentiating
between normal and WB-affected samples by regularized linear discriminant analysis models. The features from
the phase difference images were more effective than the texture features from either DC or AC images, and the
highest overall classification accuracy of nearly 93% was achieved by modeling an optimized set of features from
the phase difference images, representing improvements of about 8%–11% over the accuracy obtained using the
features from DC and AC images. This study demonstrated the effectiveness of sample surface topographic
profiles, revealed by the phase difference images, for WB assessment. SIRI coupled with surface profilometry
offers a useful tool for WB assessment of broiler breast meat. The dataset of this study has been made publicly
available1 to enourage the research into SIRI for poultry quality assessment.

1. Introduction broilers has been implemented over decades, enabling substantial in­
creases in growth rate, weight gain, and feed efficiency (USDA-ERS,
Poultry meat is a primary source of animal proteins and essential 2024). The average size and weight of broilers on the market have
nutrients such as vitamins, iron, zinc, etc. (Marangoni et al., 2015). The doubled over the past 30 years for exceptional growth performance and
global demand for poultry products has steadily increased in recent breast meat yield (NCC, 2022b). However, the rapid growth of broilers
years due to their accessibility and associated health benefits (Ellsworth has brought about the prevalence of muscular myopathies or defects in
et al., 2023). In the United States (U.S.), per capita consumption of poultry meat (Petracci et al., 2015; Tijare et al., 2016), such as white
poultry meat has grown from 82.9 lbs to 99.5 lbs over the past decade striping, woody or wooden breast (WB), and spaghetti meat (Che et al.,
and is projected to increase to 107.5 lbs by 2033 (NCC, 2022a). To meet 2022). These myopathies impair the quality and consumer acceptance of
the growing demand for poultry meat, intense genetic selection of poultry products (da Silva et al., 2017), leading to an annual loss of

* Corresponding author.
E-mail address: [email protected] (Y. Lu).
1
https://doi.org/10.5281/zenodo.14558847

https://doi.org/10.1016/j.jfoodeng.2024.112459
Received 7 October 2024; Received in revised form 18 December 2024; Accepted 25 December 2024
Available online 26 December 2024
0260-8774/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

hundreds of millions of dollars for the U.S. poultry industry (Barbut, differentiating normal and WB-affected samples. As a mesoscopic im­
2019). aging modality, OCT acquires information from small sample areas, e.g.,
WB, which is among the most challenging quality issues in the global 5 mm × 1.7 mm (lateral × depth) in Ekramirad et al. (2024), which
poultry industry (Caldas-Cueva & Owen, 2020), is a muscle anomaly could be limited in effectively characterizing the spatial heterogeneity of
with elevated levels of fibrosis, collagen, lipidosis, etc. (Sihvo et al., poultry samples. Using line-scanning hyperspectral imaging for broiler
2014; Soglia et al., 2016), although the etiology remains to be fully meat, Pallerla et al. (2024) reported an overall accuracy of 95% for
ascertained. Previous studies have reported a high incidence of WB in classifying samples of three WB degrees (“normal”, “mild WB”, and
broiler fillets of about 60%–90%, depending on factors such as flocks, “severe WB”). Considering that WB-affected tissues have altered tissue
dietary treatments, bird age, and among others (de Almeida Mallmann, structure and potentially optical scattering properties, Cai and Lu (2024)
2019; Tijare et al., 2016; Xing et al., 2020). WB-affected meat is pri­ proposed light scattering imaging for poultry WB assessment, where
marily characterized by abnormal tissue hardness, muscle rigidity, and scattering image features were modeled for discriminating between
irregular shape profiles (Caldas-Cueva & Owen, 2020). Currently, at normal and defective meat samples, resulting in an overall accuracy of
processing facilities, WB assessment is done manually based on hand 92.3%, although there was no quantification of optical properties of
palpation combined with visual examination by trained personnel (Pang poultry samples.
et al., 2020). This approach is labor-intensive, costly, and prone to Other optical sensing technologies have been also investigated for
human assessment errors. Instrumental texture analysis methods (Xiong poultry quality assessment. Wold et al. (2017, 2019) used near-infrared
et al., 2006; Morey and Owens, 2017), such as Meullenet-Owens razor spectroscopy (NIRS) to assess WB conditions of broiler fillets, where a
shear (MORS), blunt MORS (BMORS), and Warner-Bratzler shear, have commercial 15-band NIR scanning system [QVision500 (TOMRA Sort­
been applied to correlate texture or mechanical property measurements ing Solutions, Leuven, Belgium)] originally used for meat chemical
(e.g., peak count, shear force, energy) of samples with WB conditions composition analysis was adapted for online WB detection. The NIR
(Bowker and Zhuang, 2019), but they are destructive, time-consuming, scanner yielded an accuracy of 91.1% in classifying normal and
and not suited for online detection of bulk samples. WB-affected chicken samples, while a separate benchtop NIRS instru­
Siddique et al. (2021, 2022) performed the bioelectrical impedance ment achieved a higher accuracy of 96.6% (Wold and Løvland, 2020).
analysis (BIA) as a non-destructive method to assess WB conditions in Geronimo et al. (2019) utilized NIRS combined with a computer vision
broiler breast meat, obtaining accuracies of 66.7%–88.9% in differen­ system for classifying normal and WB-affected breast samples, obtaining
tiating normal and defective samples using support vector machine overall accuracies of 96.3% and 91.8%, respectively. However, the
(models. The performance of BIA depended on the severity of WB degree limited number of samples (40 normal and 40 defective) used in tthe
(i.e., normal, moderate, and severe) and instrument configurations (i.e., research casts doubt on the reliability of performance evaluations. Li
hand-held and plate BIA). Parajuli et al. (2024) reported on the use of a et al. (2022) applied NIRS and compression speed models for WB
force-sensing robot device to characterize the spatial distribution of WB detection and obtained an accuracy of 82.58% and 82.14% in recog­
in broiler fillets based on compression force measurements. Either the nizing normal and WB-affected samples, respectively. NIRS is well suited
BIA or force sensing approach has noticeable limitations given the for online applications, but as a point sensing technique, it acquires
requirement of close contact with samples, which is susceptible to food information from a small area, which may restrict its detection capacity;
contamination, relatively slow for sample evaluation, and hence may currently, few NIRS units are commercially adopted by poultry in­
not be suited for in-line detection. Sun et al. (2021) applied an active dustries (Barbut et al., 2024).
air-deformation system, originally developed for tenderness evaluation Structured-illumination reflectance imaging (SIRI), which employs
of poultry meats based on shape deformation analysis (Lee et al., 2008), spatially modulated light in the sinewave form for sample imaging, is
to assess WB conditions in broiler fillets, where the best correlation capable of depth-resolved characterization for enhanced tissue defect
coefficient of − 0.86 was achieved between the sample deformation detection as well as three-dimensional (3-D) surface reconstruction (Lu
diameter and WB categories (i.e., normal, mild, and severe). Despite and Cai, 2023), which can be difficult to achieve by conventional uni­
these investigations, since no imaging technology was involved, the lack form illumination imaging. In SIRI, acquired pattern images are
of or limited spatial information on poultry samples could hinder the demodulated into two sets of intensity images, i.e., direct component
effective discrimination between normal and myopathic samples. (DC) and amplitude component (AC) (Lu et al., 2016a); DC corresponds
Computer vision or imaging technology has received widespread to the image due to uniform illumination, while AC is unique to sinu­
attention in meat quality evaluation (Taheri-Garavand et al., 2019; soidally modulated structured light, which provides better contrast and
Modzelewska-Kapituła & Jun 2022; Park, 2016), with a handful of resolution for physical features, depending on the spatial frequency of
recent studies focused on poultry WB assessment. Caldas-Cueva et al. illumination patterns (Lu and Lu, 2019). Numerous studies demon­
(2021) reported an accuracy of 84% for distinguishing normal from strated the effectiveness of SIRI for enhanced detection of defects in
WB-affected broiler carcasses using color images. The authors extracted horticultural products, such as bruises in apples (Lu and Lu, 2017,
a small set of the features describing carcass structural information (e.g., 2018a, 2021, 2018b; Lu et al., 2016b; Li et al., 2018), chilling and
length, width, area, etc.) and modeled them using logistic regression mechnical injuries in cucumbers (Lu et al., 2021; Lu and Lu, 2020, 2021;
analysis, in which broiler carcasses affected by WB showed greater Lu et al., 2020b), and early decay in peaches (Sun et al., 2019) and or­
breast width at cranial and caudal regions (P < 0.05) as well as greater anges (Li et al., 2023, 2024).
angles at the tip of keel and breast areas in the caudal section (P < 0.05). Recently, the application of SIRI was extended to poultry and meat
Yoon et al. (2022) developed a side-view machine vision system for products (Olaniyi et al., 2023a, 2023b, 2024; Cai et al., 2024). Partic­
online detection of WB-affected broiler breast fillets, where the bending ularly, in classifying normal and white striping (WS)–affected chicken
characteristics of each sample fillet falling off a belt conveyor were fillets, AC images resulted in consistently better accuracies with im­
analyzed. Three different conveyor speeds of 10–100 feet per second provements up to 12% or higher compared to DC images, depending on
were tested, and an average overall accuracy of about 95% was achieved modeling approaches (Olaniyi et al., 2023b). In SIRI, because light
in classifying two-class (normal vs defective) samples. These studies patterns projected onto samples are deformed by the sample geometry,
demonstrated the relevance of sample geometric morphology (e.g., di­ such deformation can be utilized for 3-D surface reconstruction on a
mensions, bending curvature, etc.) to WB assessment, although there pixel level using demodulated phase images combined with existing
was room for improvement in differentiating between moderate and profilometry techniques in optical metrology (Zhang, 2018a; Zuo et al.,
severe WB categories. Ekramirad et al. (2024) employed optical coher­ 2018), as demonstrated in a prior study on surface profiling of apples (Lu
ence tomography (OCT) to image the subsurface tissues of broiler meats and Lu, 2018c). WB-affected chicken fillets tend to exhibit ridge-like
for WB detection, achieving an overall accuracy of 95% for bulges along the cranial to caudal region varying with severities of

2
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

WB, as opposed to normal (WB-free) fillets that have relatively flat and force (N) and energy (N.mm). The samples were sheared seven times by
uniform shape profiles throughout the surface (Caldas-Cueva & Owen, the BMORS probe on the ventral surface at the cranial region (Bowker
2020; Kuttappan et al., 2016). Additionally, WB-affected fillets are often and Zhuang, 2019; Pang et al., 2020) with a test speed of 5 mm/s, pre-
affected by WS defects due to the interrelated histological changes in the and post-speed of 10 mm/s, and a trigger force of 10 g, up to a maximum
affected muscle tissues (Bowker et al., 2019; Aguirre et al., 2020). These puncture depth of 20 mm. Table 1 summarizes the measurement sta­
characteristics of WB make SIRI potentially effective for WB assessment tistics of broiler fillets used in this study. The WB-affected samples had
of poultry breast fillets, given the demonstrated efficacy of the imaging higher values of shear force and energy compared to normal (non-WB)
technique for WS detection (Olaniyi et al., 2023a, 2023b) and surface samples, which is in agreement with previous observations on WB
geometry profiling of samples (Lu and Lu, 2018c). assessment (Bowker and Zhuang, 2019; Sun et al., 2021; Yoon et al.,
This study was therefore to make the first attempt to apply SIRI 2022); though, these measurements were overall lower than those pre­
coupled with phase-measuring profilometry to the assessment of WB viously reported, probably due to the relatively small samples (in terms
conditions in broiler breast meat. Specific objectives were to: 1) acquire of weight) collected in this study. Differences in bird strains, flocks, and
SIRI pattern images from breast fillets of varying WB conditions under meat status (e.g., fresh non-frozen vs. frozen-thawed meat) contribute to
sinusoidal illumination at varied spatial frequencies, 2) develop an al­ variations in texture measurements (Bowker and Zhuang, 2019; Petracci
gorithm pipeline that extracts geometric features from the phase- et al., 2015; Tijare et al., 2016).
derived surface profile map and the texture features from the intensity
(DC and AC) images of SIRI for WB assessment in broiler fillets; 3) build
and evaluate discriminant models based on the extracted features to 2.2. Image acquisition
distinguish between normal (no WB) and WB-affected broiler fillets.
A benchtop broadband SIRI system, as schematically shown in Fig. 2,
2. Materials and methods was assembled for poultry imaging. The system mainly consisted of a
quartz tungsten halogen (QTH) light source (Newport Corporation,
2.1. Broiler breast samples Irvine, CA, USA) connected with a power supply controller, a digital
light projector (DLP) (DLi6500 Optics Bundle, DLi, Austin, TX, USA) that
A total of 168 broiler breast fillets were collected in two batches from uses a digital micromirror device of 1920 × 1080 pixels for spatial light
the deboning line of a commercial poultry processing plant (Orland, IN, modulation, a monochromatic camera (Edge 4.2, PCO, Kelheim, Ger­
USA). The broiler fillets were assessed by trained personnel at the pro­ many) with a resolution of 2048 × 2048 pixels, equipped with a 35 mm
cessing facilities and classified into two categories: 72 “Normal” (WB- fixed focal length lens (Edmund Optics Inc., Barrington, NJ, USA), and a
free) and 96 “Defective” (affected by moderate and severe WB) samples, desktop computer. The light from the QTH lamp passed through a
based on hand palpation as well as visual examination (Bowker and visible-near-infrared liquid light guide (Newport Corporation, Irvine,
Zhuang, 2019; Pang et al., 2020). These fillets were packed in coolers CA, USA) to the DLP for light pattern illumination onto the broiler breast
filled with ice and immediately transported to the laboratory for this samples. The DLP and camera were synchronized by a trigger cable to
study. Upon arrival, they were further evaluated to correct potential allow for simultaneous pattern projection and image acquisition. The
errors in prior categorization (less than 5% error rate), and then indi­ camera was positioned perpendicularly downward at about 50 cm above
vidually packed in a foam tray, wrapped in film, and stored in a freezer the sample, with a pixel resolution of about 0.1 mm/pixel. The projector
at − 20 ◦ C before image collection. The frozen samples were transferred was set up at an angle of about 10◦ with respect to the camera and 40 cm
to cold storage at around 2 ◦ C for 24–48 h prior to imaging to allow them from the sample, illuminating an area of about 25 cm × 15.5 cm. A pair
to thaw. The effects of the freeze-thaw process on the broiler samples of crossed linear polarizers were attached in front of the projector and
were not considered in this study (Villegas-Cayllahua et al., 2023; Soglia the camera lens to minimize specular reflection from the sample surface.
et al., 2019). Prior to image acquisition, soft tissue paper was used to The imaging system was operated in an enclosed dark chamber for
absorb surface moisture from the broiler samples to minimize glare ef­ sample illumination and image collection to avoid the interference of
fects otherwise caused by specular reflection from the meat surface. ambient light.
Fig. 1 shows photographs of two representative normal and WB-affected In SIRI, three phase-shifted sinusoidal patterns with phase offsets of
samples. It can be seen that the defective sample presents an out-bulging
shape profile at the cranial and caudal ends as compared with the Table 1
normal (WB-free) one. Characteristics (mean ± standard deviation) of broiler breast fillets.
Right after imaging (described in Session 2.2), each sample was “Normal” “Defective”
weighted and subjected to texture measurements using a texture Average weight (g) 191.68 ± 63.89 290.02 ± 48.02
analyzer (TA.XT2i, Stable Micro Systems, Surrey, UK) equipped with a Average shear force (N) 8.35 ± 1.68 21.04 ± 8.60
blunt Meullenet-Owens razor shear (BMORS) probe. The texture anal­ Average shear energy (N.mm) 69.24 ± 12.73 161.83 ± 59.80
ysis provided quantitative measures of samples in terms of the shear

Fig. 1. (a) Top and (b) side views of two example normal and woody breast-affected broiler breast fillets.

3
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

Fig. 2. Schematic of a broadband structured-illumination reflectance imaging system for assessing woody breast of broiler meat.

0, 2π/3, and 4π/3 at a given spatial frequency are normally used due to patterns were generated as 8-bit bitmaps in Matlab R2022R (The
the ease of three-phase demodulation to retrieve intensity and phase MathWorks Inc, USA) and then loaded to the graphical interface soft­
images (Lu and Lu, 2019). In this study, images were acquired under the ware (LightCrafter 6500/9000 GUI, Texas Instruments, Dallas, Texas) of
illumination of a sequence of three-phase-shifted patterns at eight the projector for sample illumination. The broiler samples were indi­
different spatial frequencies of 0.015–0.150 cycles/mm. These spatial vidually placed on a sample stage and imaged under consecutive
frequencies differ from those in our previous studies on WS assessment switching of a total of 24 preloaded sinusoidal illumination patterns (24
of broiler meat (Olaniyi et al., 2023a, 2023b, 2024), given the different = 3 phase-shifted patterns per spatial frequency × 8 spatial frequencies)
characteristics of WB and WS defects, and the actual size of the illumi­ triggered by the camera bundled software (camware v4.2, PCO, Kel­
nated patterns was factored more accurately in implementing illumi­ heim, Germany) with an exposure time of 500 ms, resulting in the
nation patterns at the intended spatial frequencies in this study. These acquisition of 24 reflectance pattern images per sample and hence 4032

Fig. 3. The pipeline of image processing to obtain the phase-difference, direct component (DC), and amplitude component (AC) images using three phase-shifted
patterned images at each spatial frequency.

4
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

images for all the 168 samples. The acquired dataset has been made be uniquely determined using the following equation (Huntley and
publicly available in the Zenodo repository (Lu, 2024). Saldner, 1993; Zhao et al., 1994):
( )
(fh /fl )ϕl − φh
k = Round (5)
2.3. Image processing 2π

where Round() is a function that returns the closest integer value, fh and
The acquired pattern images were first processed by a low-pass
fl represent the spatial frequency of high- and low-frequency fringe
Gaussian filter of 3 × 3 pixels in kernel size to suppress image noise.
patterns, respectively, and φh and ϕl are the corresponding wrapped and
Then, as illustrated in Fig. 3, the three phase-shifted pattern images at
absolute phases at the high and low spatial frequencies, respectively.
each spatial frequency were demodulated into DC, AC, and phase images
Here the spatial frequencies of 0.03 cycles/mm (low-frequency) and
as follows:
0.15 cycles/mm (high-frequency) were used to retrieve the phase map.
1 First, the ϕl was coarsely restored by comparing the wrapped phase with
IDC = (I1 + I2 + I3 ) (1)
3 the estimated reference phase using the method proposed by Wang et al.
√̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (2023), which suffers from low SNR and resolution in the phase map due
2 to the sparse fringe distribution at a lower frequency. According to Eq.
IAC = (I1 − I2 )2 + (I1 − I3 )2 + (I2 − I3 )2 (2)
3 (5), though, the low-frequency phase ϕl can be used to determine the
( √̅̅̅ ) fringe order k of the wrapped phase at the high frequency. With the
3(I1 − I3 ) fringe order k obtained, the high-frequency absolute phase ϕh with
φ = arctan (3)
(2I2 − I1 − I3 ) improved measurement resolution and precision was then calculated
using Eq. (4). Thereafter, a phase-difference map was attained by sub­
where I1, I2, and I3 are the reflectance pattern images acquired under the
tracting the estimated reference map from the absolute phase ϕh, which
illustration of three phase-shifted patterns at the same spatial frequency,
depicts the topographic surface profiles of a sample due to the propor­
and the demodulation operations are performed on a pixel basis. The
tional relationship between the phase difference and surface height (Lu
intensity images, DC and AC, characterize the optical properties of a
and Cai, 2023; Lu and Lu, 2018c). Finally, the resultant phase difference
sample (Lu and Lu, 2019), while the phase image depicts the surface
maps were subjected to a median filter of 5 × 5 pixels in kernel size to
geometry of the sample (Fig. 3).
reduce noise, before subsequent feature extraction and modeling.
It is important to note that the arctangent function used in Eq. (3) for
phase retrieval results in phase values wrapped to the range of – π to π,
with 2π discontinuities or jumps in the resultant map, which is referred 2.4. Feature analysis and classification modeling
to as the wrapped phase (Zhang, 2018b). Further phase unwrapping is
needed to restore the wrapped phase to a continuous phase map (also The two sets of image data obtained, i.e., the phase images (Dataset I)
known as the absolute phase). The relationship between the absolute and the intensity DC and AC images (Dataset II), were utilized separately
phase (ϕ) and wrapped phase (φ) is provided: for feature extraction and WB classification modeling. Our previous
ϕ = φ + 2πk (4) studies have shown the superiority of using pre-trained deep-learning
(DL) models over hand-crafted feature engineering in extracting textural
where k is the integer number to represent fringe orders of the wrapped features from intensity images for WS assessment of poultry meat
phase, which remains to be determined using a phase unwrapping (Olaniyi et al., 2023a, 2023b). Here, the pretrained ResNeXt-101 (Xie
technique. et al., 2017) was used as a feature extractor for both phase and intensity
Among numerous absolute phase recovery techniques, the two- images. The output from the last convolutional layer of the model was
frequency method (Zhang, 2018b), which employs the phase informa­ taken as the features, which is a vector of 2048 features for each given
tion obtained at a low frequency to unwrap the high-frequency phase, input image, as done in a previous study on beef marbling assessment by
was adopted in this study given its high-resolution [or signal-noise-ratio SIRI (Cai et al., 2024). Although pretrained DL models were effective for
(SNR)] and precision for absolute phase retrieval. The fringe order k can extracting discriminative features from the intensity (DC and AC

Fig. 4. The procedures for classifying normal and woody breast (WB)-affected broiler meat samples using features extracted from intensity and phase-difference
images. DC and AC denote direct component and amplitude component, respectively.

5
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

images) for assessing physical characteristics of samples (e.g., WS in Feature analysis and classification modeling were performed using
poultry meat and beef marbling) (Olaniyi et al., 2023a, 2023b; Cai et al., the DL framework PyTorch (version 1.13.1) and the scikit-learn library
2024), the chance is that these models like ResNeXt-101 may not be well (version 1.0.2) in Python (version 3.8.16) on a desktop computer with
suited for representing the phase difference images in this study, an Intel® Core™ i9-10900x CPU (128 GB RAM) and an NVIDIA RTX
because the phase difference images that depict the surface geometry A6000 GPU (with 48 GB RAM). Given a specific set of feature data,
differ dramatically from texture intensity images or other types of nat­ model training and testing took less than 10 min for 50 replications on
ural scene images in the ImageNet database used for pre-training the DL the desktop computer.
models.
Given the potential risk in using DL-based feature extraction for 3. Results and discussion
phase difference images, due attention was given to hand-crafted feature
engineering, which may lead to more effective features than those ob­ 3.1. Demodulated images
tained from generic DL models, for capturing sample geometry. The
histogram of gradient (HOG) (Dalal and Triggs, 2005) and local binary Fig. 5 shows example phase-difference images of normal and WB-
patterns (LBP) (Ojala et al., 1996) are the image descriptors that capture affected chicken fillets. These images reflect the topographic surface
local gradients and spatial structures (e.g., shape or contour) in images, profiles of samples. The defective broiler examples (second row), with
which have been used to extract features from 3-D depth maps for the color bar ranging from blue to red representing surface geometry/
human face/action recognition (Ali et al., 2018; Tatarenkov and height, show ridge-like bulges along the cranial to caudal region varying
Buchatsky, 2018; Chen et al., 2015). Inspired by LBP, binarized statis­ with the severity of WB, while the normal (WB-free) samples presented
tical image features (BSIF) provide meaningful representations of im­ with relatively flat and uniform shape profiles (first row, Fig. 5). These
ages by implementing filters learned using statistics of images for surface profiles characterize the physical shape differences visually
improved pattern classification performance (Kannala and Rahtu, observed, as shown in Fig. 1. Hence, the image (geometric) features that
2012). represent the sample surface profiles are likely to be effective for
In this study, as indicated in Fig. 4, all three types of features (HOG, differentiating normal and WB-affected samples.
LBP, and BSIF) were extracted from the phase difference images to build Fig. 6 shows examples of demodulated intensity (DC and AC) images
the set of hand-crafted features as opposed to the DL-based features for of normal and WB-affected samples. Compared with DC, AC images vary
WB assessment. These features, which were extracted using an open- with spatial frequency and appear to provide better visualization of
source Matlab Toolbox Balu (Mery, 2011), included 1800 HOG, 59 textural characteristics (e.g., white striations) of samples, except that the
LBP, and 128 BSIF. The HOG feature set was constructed using 20 × 10 intensity of AC images diminishes as the spatial frequency increases.
spaced cells in horizontal and vertical directions with 9 histogram bins, This observation on intensity images is in agreement with previous
resulting in 1800 features. The LBP features were calculated for 58 poultry and beef imaging (Olaniyi et al., 2023a, 2023b, 2024; Cai et al.,
rotation-invariant uniform patterns and all non-uniform patterns 2024). However, the intensity images reveal little information on the
defined in an 8-pixel neighborhood (Ojala et al., 2002). For the BSIF geometry of samples. It is reported that WS is co-occurrent with WB in
descriptor, the features were extracted with a filter size of 5 × 5 pixels broiler meat to a varying degree due to the interrelated histological
and 7 bit length, which is the default setting of the Balu toolbox (Mery, changes in the affected muscles (Bowker et al., 2019; Aguirre et al.,
2011). Since both filter size and the length of the bit string affect the 2020). The co-occurrence of WS in the WB-affected samples could
resultant feature set and downstream model performance (Kannala and potentially contribute to differentiating between normal and defectuve
Rahtu, 2012), two parameters were empirically tuned in extracting samples in modeling texture features extracted from the intensity im­
features from phase-difference images for improved model performance ages. It should be noted that, in this study, most of the normal (WB-free)
by varying the filter size from 3 to 17 (in an increment of 2), followed by samples showed no incidence of WS defects, while only a small per­
adjusting the bit length from 5 to 12 with the optimized filter size (see centage (around 25% through visual inspection) of WB-affected samples
Session 3.2.2). The extracted hand-crafted features were concatenated (e.g., the last row example in Fig. 6) were affected by WS defects.
in different combinations for model development and performance
evaluation. 3.2. Classification of normal and WB-affected samples
For each sample, the DL-based features were extracted from the in­
tensity (DC and AC) images at different spatial frequencies, while the 3.2.1. Modeling using phase-difference images
hand-crafted (HOG, LBP, and BSIF) features were from the phase- The two types of features extracted from the phased-difference im­
difference image obtained using pattern images at 0.03 cyles/mm ages, i.e., 1) DL-based and 2) hand-crafted geometric features, were used
(low-frequency) and 0.15 cycles/mm (high-frequency). Classification for building RLDA models for differentiating between normal and WB-
models were built for WB assessment by using the intensity and phase- affected samples. Fig. 7 (a) shows the overall classification accuracy
difference image features separately as well as their combination. In based on the DL-based features and different sets of hand-crafted fea­
each modeling scenario, the candidate feature dataset was randomly tures. The DL-based features yielded an overall accuracy of 75.73%,
partitioned into training and test sets according to the ratio of 65%: 35% while the hand-crafted features achieved better performance (given
(corresponding to 109 : 59 fillet samples). Regularized linear discrimi­ higher accuracy and smaller variations), with overall accuracies ranging
nant analysis (RLDA), which is an efficient classifier designed for from 80.75% to 91.32%, representing significant improvements of
handling high-dimensional features in small datasets (Guo et al., 2007), 5.02%–15.59% (P < 0.05). Among the three types of hand-crafted fea­
was employed to differentiate between normal and WB-affected sam­ tures, LBP and BSIF outperformed HOG substantially, yielding the
ples. The shrinkage hyperparameter of RLDA models was optimized classification accuracy of 90.54% and 90.95%, respectively, and the
through grid search with 10-fold cross-validation on training data. To ensemble of LBP and BSIF produced a slight improvement to 91.32%.
obtain a reliable estimate of model performance, a 50-repeated holdout Principal component analysis (PCA) (Jolliffe, 2002) was applied to the
validation was carried out for model training/testing, where the feature combined feature set for sample visualization. The scattering plot based
dataset was subjected to random dataset partition in each modeling on the first two principal components of the LBP and BSIF combined
replication, and the averaged accuracy over the 50 replications was features, as shown in Fig. 7 (b), confirmed the separation between
calculated as the primary metric for model performance evaluation and normal and WB-affected samples. It is noted that the combination of all
comparison (via a paired-t test). The choice of the 50 replications here three types (HOG, LBP, and BSIF) of hand-crafted features produced a
was empirically based on preliminary experimentation with different mediocre accuracy of 84.92%, which suggests that LBP and BSIF were
replications to seek stabilized model performance. more effective than HOG for WB assessment.

6
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

Fig. 5. Examples of demodulated phase-difference (or phase-depth) images (best viewed in color) of normal and defective (woody breast-affected) broiler
breast fillets.

Among the three types of hand-crafted features, the BSIF had the 0.055 cycles/mm) still fell behind the hand-crafted features, i.e., the best
most discriminative power. To exploit BSIF for improved WB assess­ BSIF and LBP combined, of the phase difference images, with a notice­
ment, its two parameters, including the filter size and bit length able accuracy gap of 4.68% from the best accuracy (92.95%) achieved
(Kannala and Rahtu, 2012), were tuned to extract better feature sets by the latter. These results suggest that the phase image-derived geo­
from the phase difference images. For the sake of optimization effi­ metric features contributed primarily to the success in differentiating
ciency, instead of conducting an exhaustive search for the best param­ between normal and WB-affected samples while the AC image-derived
eter pair, the filter size was first tuned with the bit length set to default, textural features were less effective.
and thereafter the bit length was tuned by applying the optimized filter
size. Fig. 8 shows the effect of varying the parameters on the classifi­ 3.3. Discussion
cation accuracies. It seems the accuracy decreased with relatively larger
filter sizes and increased with relatively larger bit lengths. Tuning the This study presents the first proof-of-concept validation of the
filter size resulted in an overall accuracy of 91.86% with a filter size of 7 applicability of SIRI coupled with phase analysis for differentiating be­
× 7 pixels, and further an improved accuracy of 92.78% was obtained tween normal and WB-affected poultry meat. The SIRI technique allows
with the bit length set to 8. Despite the local optimization, the resultant reconstruction of the surface topography of chicken samples, which is
BSIF features produced an improvement of 1.83% in the overall accu­ reflected in the retrieved phase-difference images, for enhanced WB
racy over that achieved before optimization. Moreover, the optimized assessment in comparison with using intensity images, which differs
BSIF features were concatenated with the LBP features, yielding an even from 2D computer vision techniques reported on poultry WB assessment
better overall accuracy of 92.95%, which is a statistically significant (P (Geronimo et al., 2019; Caldas-Cueva et al., 2021; Yoon et al., 2022).
< 0.01) improvement of 1.63% over the accuracy (91.32%) obtained by The best classification accuracy of nearly 93% obtained by phase dif­
the ensemble of the default BSIF (before parameter tuning) and LBP ference image features was respectable, although some studies reported
features. higher accuracies of around 95% (Yoon et al., 2022; Ekramirad et al.,
2024). These previous studies however examined a comparatively
3.2.2. Modeling using intensity images smaller collection of chicken samples (fewer than 50) and often lacked
Fig. 9 shows the overall classification accuracies using the intensity repeated validations in their modeling pipeline, which could lead to an
image (DC and AC) features and their combination with the hand- overestimate of classification performance.
crafted features from the phase-difference images. The feature combi­ Recently, light scattering imaging (LSI) was instrumented and
nation was done by concatenating the optimized BSIF features and LBP investigated for assessing WB conditions in broiler meat, which is
features with the features of AC images at individual spatial frequencies described in our seperate study (Cai and Lu, 2024), where scattering
to produce the ensemble of AC-phase image features. The DC images images were acquired from the same set of samples as in this study,
produced a baseline overall accuracy of 82.10%, while the AC images under the illumination using a highly focused light beam. The LSI
yielded better accuracies up to 85.02% at 0.15 cycles/mm, despite a technique is based on the assumption that WB-affected tissues differ
slight but insignificant performance drop at 0.015 cycles/mm [no sig­ from normal tissues in their light-scattering characteristics because of
nificant difference (P > 0.5) from the accuracy (82.10%) by DC]. The tissue structural alterations due to the prescence of WB, which was
combined features of AC and phase difference images yielded significant validated from the visual inspection of acquired scattering images of
improvements at all the spatial frequencies, with the overall classifica­ normal and defective samples. Extracting scattering image features
tion accuracy ranging from 86.81% to 88.27%, outperforming the AC followed by discriminant modeling yielded an overall accuracy
image features used alone. These results validated the positive effects of exceeding 92% (Cai and Lu, 2024), which is comparable to the perfor­
the geometric features from the phase-difference images for WB mance of the SIRI technique. Deviating from conventional imaging
assessment. The best feature combination (with 88.27% accuracy at techniques operating under uniform illumination, both techniques

7
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

Fig. 6. Examples of demodulated intensity images, i.e., direct component (DC) and amplitude component (AC), at the spatial frequencies of 0.015–0.15 cycles/mm
(from left to right) for normal and woody breast-affected (defective) broiler breast samples.

capture different yet meaningful aspects of WB characteristics relevant features, will be beneficial for improved sample classification. Since the
to evaluation purposes. present SIRI acquires only broadband images, given the efficacy of
The pipeline for feature engineering and model development of this spectral sensing for WB assessment (Pallerla et al., 2024; Wold et al.,
study has room for further optimization and improvement in the WB 2017, 2019; Wold and Løvland, 2020), it is possible to enhance WB
classification, although the BSIF feature extraction was locally opti­ assessment by implementing SIRI in multi-/hyper-spectral mode.
mized. Dedicated efforts to select the most relevant features or pursue Further research into SIRI will involve collecting an expanded, more
dimension reduction via PCA, instead of modeling the entire set of balanced set of broiler samples to enhance model performance while

8
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

Fig. 7. (a) Overall classification accuracies of deep-learning-based and hand-crafted features extracted from 3-D phase images; (b) scatter plot of the LBP-BSIF
combined features in the space spanned by the first two principal components (PCs) for two WB categories of broiler meat samples. HOG, LBP, and BSIF denote
histograms of gradient, local binary pattern, and binarized statistical image features, respectively.

Fig. 8. Overall classification accuracies by modeling binarized statistical image features (BSIF) extracted for different filter sizes (left) and bit length (right). The red
bar indicates the overall accuracy obtained using the BSIF features extracted with defect parameter settings. (For interpretation of the references to color in this figure
legend, the reader is referred to the Web version of this article.)

Fig. 9. Comparison of overall accuracies for the intensity image [direct component (DC) and amplitude component (AC)] features and the combination of AC image
features with the phase difference image features.

9
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

classifying poultry samples into three categories according to WB con­ the work reported in this paper.
ditions. Notably, in addition to the phase analysis-based SIRI technique
in this study, other 3-D imaging techniques such as time of flight and Acknowledgments
stereovision can also provide surface topography of poultry samples. It
would be worthwhile to pursue a dedicated study on performance This research was possible through funding by the Michigan Alliance
evaluation of 3-D imaging techniques for WB assessment of broiler meat, for Animal Agriculture (M-AAA) and USDA-NIFA (Grant No.
especially given the availability of cost-effective, consumer-grade depth 2022–67018-36625) and the support of Miller Poultry (Orland, IN) who
cameras. The off-the-shelf depth cameras may not deliver high-precision provided chicken samples.
3-D maps but are compact standalone imagers that can be readily
implemented for high-speed food quality detection (Xu et al., 2024). It is Data availability
also important to point out that the SIRI technique can be applied to
quantify and map optical properties (absorption and scattering) of The datast of this study is publicly available at https://doi.
poultry tissues using the computational methods developed in the org/10.5281/zenodo.14558847.
spatial frequency domain imaging (Cuccia et al., 2009; Bigio and Fan­
tini, 2016; Lu et al., 2020a), which would provide more insights into WB References
assessment of poultry meat.
Moving forward, the SIRI system of this study, which was only Aguénounon, E., Dadouche, F., Uhring, W., Giux, S., 2019. Single snapshot of optical
properties image quality improvement using anisotropic two-dimensional windows
applied to stationary samples, has yet to be upgraded for real-time filtering. J. Biomed. Opt. 24 (7), 071611. https://doi.org/10.1117/1.JBO.24.7.071
quality detection of moving samples, which is challenged by the 611.
lengthy acquisitions of multiple (three in this study) phase-shifted si­ Aguirre, M.E., Leyva-Jimenez, H., Travis, R., Lee, J.T., Athrey, G., Alvarado, C.Z., 2020.
Evaluation of growth production factors as predictors of the incidence and severity
nusoidal pattern images at a given spatial frequency. Single-shot image of white striping and woody breast in broiler chickens. Poultry Sci. 99 (7),
demodulation, which has been researched in the field of optical 3723–3732. https://doi.org/10.1016/j.psj.2020.03.026.
metrology and biomedical optics (Feng et al., 2019; Aguénounon et al., Ali, H.H., Moftah, H.M., Youssif, A.A., 2018. Depth-based human activity recognition: a
comparative perspective study on feature extraction. Future Computing and
2019), can be a viable solution to the implementation of real-time SIRI if Informatics Journal 3 (1), 51–67. https://doi.org/10.1016/j.fcij.2017.11.002.
the intensity (DC and AC) and phase images can be retrieved from Barbut, S., 2019. Recent myopathies in broiler’s breast meat fillets. World Poultry Sci. J.
single-phase images with minimal artifacts. Moreover, generative arti­ 75 (4), 559–582. https://doi.org/10.1017/S0043933919000436.
Barbut, S., Mitchell, R., Hall, P., Bacon, C., Bailey, R., Owens, C.M., Petracci, M., 2024.
ficial intelligence has recently received considerable attention in image
Myopathies in broilers: supply chain approach to provide solutions to challenges
generation. These generative models (Liu et al., 2024; Zuo et al., 2024) related to raising fast growing birds. Poultry Sci., 103801
have the potential to achieve single-frame image demodulation, facili­ Bigio, I.R., Fantini, S., 2016. Quantitative Biomedical Optics: Theory, Methods, and
tating the online implementation of SIRI. Applications. Cambridge University Press, Cambridge, United Kingdom.
Bowker, B., Zhuang, H., Yoon, S.C., Tasoniero, G., Lawrence, K., 2019. Relationships
between attributes of woody breast and white striping myopathies in commercially
4. Conclusions processed broiler breast meat. J. Appl. Poultry Res. 28 (2), 490–496. https://doi.org/
10.3382/japr/pfz007.
Bowker, B., Zhuang, H., 2019. Detection of razor shear force differences in broiler breast
This study presents a new evaluation of a broadband SIRI system for meat due to the woody breast condition depends on measurement technique and
assessing WB conditions in broiler breast fillets. The surface profiles of meat state. Poultry Sci 98 (11), 6170–6176. https://doi.org/10.3382/ps/pez334.
samples obtained using a phase demodulation and processing procedure Cai, J., Lu, Y., 2024. Detection of wood breast condition in broiler breast fillets using
light scattering imaging. Journal of the ASABE 67 (in press). https://doi.org/10.1
captured the shape differences between normal and WB-affected 3031/ja.16108.
chicken samples, while the demodulated intensity (DC and AC) images Cai, J., Lu, Y., Olaniyi, E., Wang, S., Dahlgren, C., Devost-Burnett, D., Dinh, T., 2024.
revealed only the surface texture characteristics of samples. An algo­ Beef marbling assessment by structured-illumination reflectance imaging with deep
learning. J. Food Eng. 369, 111936. https://doi.org/10.1016/j.jfoodeng.2024.111
rithm pipeline extracted different types of features from the phase dif­ 936.
ference images (that depict the sample geometry) and the intensity Caldas-Cueva, J.P., Mauromoustakos, A., Sun, X., Owens, C.M., 2021. Use of image
images and built models for differentiating between normal and WB- analysis to identify woody breast characteristics in 8-week-old broiler carcasses.
Poultry Sci 100 (4), 100890. https://doi.org/10.1016/j.psj.2020.12.003.
affected samples. The hand-crafted features of the phase difference im­
Caldas-Cueva, J.P., Owens, C.M., 2020. A review on the woody breast condition,
ages were more effective for WB assessment than the textural features of detection methods, and product utilization in the contemporary poultry industry.
either DC or AC images. The highest classification accuracy of 92.95% J. Anim. Sci. 98 (8), skaa207. https://doi.org/10.1093/jas/skaa207.
was achieved by an optimized set of BSIF and LBP features from the Che, S., Wang, C., Iverson, M., Varga, C., Barbut, S., Bienzle, D., Susta, L., 2022.
Characteristics of broiler chicken breast myopathies (spaghetti meat, woody breast,
phase difference images, representing improvements of 7.93%–11.05% white striping) in Ontario, Canada. Poultry Sci 101 (4), 101747. https://doi.
over the accuracy obtained by textural features from DC and AC images. org/10.1016/j.psj.2022.101747.
The ensemble of the optimized phase-derived features and the intensity Chen, C., Jafari, R., Kehtarnavaz, N., 2015. Action recognition from depth sequences
using depth motion maps-based local binary patterns. In: 2015 IEEE Winter
image features yielded higher accuracies than using the features of in­ Conference on Applications of Computer Vision, pp. 1092–1099.
tensity images alone but still underperformed the phased-derived fea­ Cuccia, D.J., Bevilacqua, F., Durkin, A.J., Ayers, F.R., Tromberg, B.J., 2009. Quantitation
tures, implying that the sample geometry was more relevant than the and mapping of tissue optical properties using modulated imaging. J. Biomed. Opt.
14 (2), 024012. https://doi.org/10.1117/1.3088140, 024012.
surface texture characteristics to WB assessment. This study has da Silva, D.C.F., de Arruda, A.M.V., Gonçalves, A.A., 2017. Quality characteristics of
demonstrated the efficacy of SIRI with surface profilometry for WB broiler chicken meat from free-range and industrial poultry system for the
assessment in broiler breast meat. consumers. J. Food Sci. Technol. 54, 1818–1826. https://doi.org/10.1007/s13197-0
17-2612-x.
Dalal, N., Triggs, B., 2005. Histograms of oriented gradients for human detection. 2005
CRediT authorship contribution statement IEEE computer society conference on computer vision and pattern recognition
(CVPR’05) 1, 886–893.
de Almeida Mallmann, B., 2019. Prediction and Evaluation of Breast Myopathy (PhD
Jiaxu Cai: Writing – original draft, Formal analysis, Data curation.
Dissertation). University of Arkansas, Fayetteville, Arkansas.
Yuzhen Lu: Writing – review & editing, Writing – original draft, Su­ Ekramirad, N., Yoon, S.C., Bowker, B.C., Zhuang, H., 2024. Nondestructive assessment of
pervision, Investigation, Funding acquisition, Conceptualization. woody breast myopathy in chicken fillets using optical coherence tomography
imaging with machine learning: a feasibility study. Food Bioprocess Technol 1–18.
https://doi.org/10.1007/s11947-024-03369-1.
Declaration of Competing Interest Ellsworth, S., Crandall, P.G., Seo, H.S., O’Bryan, C.A., 2023. Consumers’ willingness to
pay for safer, more environmentally friendly smoke flavored chicken breasts.
The authors declare that they have no known competing financial J. Sensory Stud. 38 (2), e12812. https://doi.org/10.1111/joss.12812.

interests or personal relationships that could have appeared to influence

10
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

Feng, S., Chen, Q., Gu, G., Tao, T., Zhang, L., Hu, Y., et al., 2019. Fringe pattern analysis Lu, Y., Cai, J., 2023. Structured-light imaging. In: Zhang, Q. (Ed.), Encyclopedia of Smart
using deep learning. Advanced photonics 1 (2), 025001. https://doi.org/10.1117/1. Agriculture Technologies. New York: Springer, Cham. https://doi.org/10.1007/978-
AP.1.2.025001, 025001. 3-030-89123-7_166-1.
Geronimo, B.C., Mastelini, S.M., Carvalho, R.H., Júnior, S.B., Barbin, D.F., Marangoni, F., Corsello, G., Cricelli, C., Ferrara, N., Ghiselli, A., Lucchin, L., Poli, A.,
Shimokomaki, M., Ida, E.I., 2019. Computer vision system and near-infrared 2015. Role of poultry meat in a balanced diet aimed at maintaining health and
spectroscopy for identification and classification of chicken with wooden breast, and wellbeing: an Italian consensus document. Food Nutr. Res. 59 (1), 27606. https://
physicochemical and technological characterization. Infrared Phys. Technol. 96, doi.org/10.3402/fnr.v59.27606.
303–310. https://doi.org/10.1016/j.infrared.2018.11.036. Mery, D., 2011. BALU: a Matlab toolbox for computer vision, pattern recognition and
Guo, Y., Hastie, T., Tibshirani, R., 2007. Regularized linear discriminant analysis and its image processing. http://dmery.ing.puc.cl/index.php/balu.
application in microarrays. Biostatistics 8 (1), 86–100. https://doi.org/10.1093/bios Morey, A., Owens, C.M., 2017. In: Methods for measuring meat texture. In: Poultry
tatistics/kxj035. Quality Evaluation. Woodhead Publishing, pp. 115–132. https://doi.org/10.10
Huntley, J.M., Saldner, H., 1993. Temporal phase-unwrapping algorithm for automated 16/B978-0-08-100763-1.00005-2.
interferogram analysis. Applied optics 32 (17), 3047–3052. https://doi.org/10.1 Modzelewska-Kapituła, M., Jun, S., 2022. The application of computer vision systems in
364/AO.32.003047. meat science and industry–A review. Meat Sci 192, 108904. https://doi.org/10.10
Jolliffe, I.T., 2002. Principal Component Analysis (Second Edition), . Springer, New York, 16/j.meatsci.2022.108904.
USA. NCC, 2022a. Per capita consumption of poultry and livestock, 1965 to forecast 2022. In:
Kannala, J., Rahtu, E., 2012. BSIF: binarized statistical image features. In: Proceedings of in Pounds. National Chicken Council. https://www.nationalchickencouncil.org/ab
the 21st International Conference on Pattern Recognition (ICPR2012), out-the-industry/statistics/per-capita-consumption-of-poultry-and-livestock-1965
pp. 1363–1366. -to-estimated-2012-in-pounds/. Retrieved from: (accessed December 26, 2024).
Kuttappan, V.A., Hargis, B.M., Owens, C.M., 2016. White striping and woody breast NCC, 2022b. U.S. Broiler Performance, 2022. National Chicken Council. Retrieved from.
myopathies in the modern poultry industry: a review. Poultry Sci 95 (11), https://www.nationalchickencouncil.org/about-the-industry/statistics/u-s-broiler
2724–2733. https://doi.org/10.3382/ps/pew216. -performance/ (accessed December 26, 2024).
Lee, Y.S., Owens, C.M., Meullenet, J.F., 2008. A novel laser air puff and shape profile Ojala, T., Pietikäinen, M., Harwood, D., 1996. A comparative study of texture measures
method for predicting tenderness of broiler breast meat. Poultry Sci 87 (7), with classification based on featured distributions. Pattern Recogn 29 (1), 51–59.
1451–1457. https://doi.org/10.3382/ps.2007-00463. https://doi.org/10.1016/0031-3203(95)00067-4.
Li, Z., Wu, L., Yang, Z., Zhang, Y., Wang, P., Xu, X., 2022. Improved identification and Ojala, T., Pietikainen, M., Maenpaa, T., 2002. Multiresolution gray-scale and rotation
classification accuracy of wooden breast by jointly using near-infrared spectroscopy invariant texture classification with local binary patterns. IEEE Trans. Pattern Anal.
and compression speed. Food Res. Int. 161, 111795. https://doi.org/10.1016/j. Mach. Intell. 24 (7), 971–987. https://doi.org/10.1109/TPAMI.2002.1017623.
foodres.2022.111795. Olaniyi, E.O., Lu, Y., Cai, J., Sukumaran, A.T., Jarvis, T., Rowe, C., 2023a. Feasibility of
Li, J., Lu, Y., Lu, R., 2023. Detection of early decay in navel oranges by structured- imaging under structured illumination for evaluation of white striping in broiler
illumination reflectance imaging combined with image enhancement and breast fillets. J. Food Eng. 342, 111359. https://doi.org/10.1016/j.jfoodeng.20
segmentation. Postharvest Biol. Technol. 196, 112162. https://doi.org/10.1016/j.po 22.111359.
stharvbio.2022.112162. Olaniyi, E., Lu, Y., Sukumaran, A.T., Jarvis, T., Rowe, C., 2023b. Non-destructive
Li, J., Lu, Y., Lu, R., 2024. Identification of early decayed oranges using structured- assessment of white striping in broiler breast meat using structured-illumination
illumination reflectance imaging coupled with fast demodulation and improved reflectance imaging with deep learning. J. ASABE 66 (6), 1437–1447. https://doi.or
image processing algorithms. Postharvest Biol. Technol. 207, 112627. https://doi. g/10.13031/ja.15667.
org/10.1016/j.postharvbio.2023.112627. Olaniyi, E.O., Lu, Y., Zhang, X., Sukumaran, A.T., Thames, H.T., Pokhrel, D., 2024. Non-
Li, R., Lu, Y., Lu, R., 2018. Structured illumination reflectance imaging for enhanced destructive assessment of microbial spoilage of broiler breast meat using structured
detection of subsurface tissue bruising in apples. Trans. ASABE 61 (3), 809–819. illumination reflectance imaging with machine learning. Food Anal. Methods 1–12.
https://doi.org/10.13031/trans.12243. https://doi.org/10.1007/s12161-024-02605-w.
Liu, H., Yan, N., Shao, B., Yuan, S., Zhang, X., 2024. Deep learning in fringe projection: a Pallerla, C., Feng, Y., Owens, C.M., Bist, R.B., Mahoudi, S., Sohrabipour, P., Davar, A.,
review. Neurocomputing 581, 127493. https://doi.org/10.1016/j.neucom.20 Wang, D., 2024. Neural network architecture search enabled wide-deep learning
24.127493. (NAS-WD) for spatially heterogenous property awared chicken woody breast
Lu, R., Van Beers, R., Saeys, W., Li, C., Cen, H., 2020a. Measurement of optical properties classification and hardness regression. Artif. Intell. Agric. 14, 73–82. https://doi.
of fruits and vegetables: a review. Postharvest Biol. Technol. 159, 111003. htt org/10.1016/j.aiia.2024.11.003.
ps://doi.org/10.1016/j.postharvbio.2019.111003. Pang, B., Bowker, B., Yang, Y., Zhang, J., Zhuang, H., 2020. Relationships between
Lu, Y.. SIRI_PoultryWB: structured-illumination reflectance imaging data for poultry instrumental texture measurements and subjective woody breast condition scores in
woody breast assessment [Data set]. Zenodo. https://doi.org/10.5281/zenodo.1455 raw broiler breast fillets. Poultry Sci 99 (6), 3292–3298. https://doi.org/10.1016/j.
8847. psj.2019.12.072.
Lu, Y., Li, R., Lu, R., 2016a. Structured-illumination reflectance imaging (SIRI) for Park, B., 2016. In: Quality evaluation of poultry carcass. In Sun, D.W. (ed.), Computer
enhanced detection of fresh bruises in apples. Postharvest Biol. Technol. 117, 89–93. Vision Technology for Food Quality Evaluation, pp. 213–241. https://doi.org/10.10
https://doi.org/10.1016/j.postharvbio.2016.02.005. https://doi.org/10.1016/j.po 16/B978-0-12-802232-0.00009-8.
stharvbio.2016.02.005. Parajuli, P., Yoon, S.C., Zhuang, H., Bowker, B., 2024. Characterizing the spatial
Lu, Y., Li, R., Lu, R., 2016b. Fast demodulation of pattern images by spiral phase distribution of woody breast condition in broiler breast fillet by compression force
transform in structured-illumination reflectance imaging for detection of bruises in measurement. J. Food Meas. Char. 18 (3), 1991–2003. https://doi.org/10.1007
apples. Comput. Electron. Agric. 127, 652–658. /s11694-023-02330-8.
Lu, Y., Lu, R., 2017. Histogram-based automatic thresholding for bruise detection of Petracci, M., Mudalal, S., Soglia, F., Cavani, C., 2015. Meat quality in fast-growing
apples by structured-illumination reflectance imaging. Biosyst. Eng. 160, 30–41. broiler chickens. World Poultry Sci. J. 71 (2), 363–374. https://doi.org/10.101
https://doi.org/10.1016/j.compag.2016.07.012. 7/S0043933915000367.
Lu, Y., Lu, R., 2018a. Detection of surface and subsurface defects of apples using Siddique, A., Shirzaei, S., Smith, A.E., Valenta, J., Garner, L.J., Morey, A., 2021.
structured-illumination reflectance imaging with machine learning algorithms. Acceptability of artificial intelligence in poultry processing and classification
Trans. ASABE 61 (6), 1831–1842. https://doi.org/10.13031/trans.12930. efficiencies of different classification models in the categorisation of breast fillet
Lu, Y., Lu, R., 2018b. Fast Bi-dimensional empirical mode decomposition as an image myopathies. Front. Physiol. 12, 712649. https://doi.org/10.3389/fphys.2021.
enhancement technique for fruit defect detection. Comput. Electron. Agric. 152, 712649.
314–323. Siddique, A., Herron, C.B., Valenta, J., Garner, L.J., Gupta, A., Sawyer, J.T., Morey, A.,
Lu, Y., Lu, R., 2018c. Structured-illumination reflectance imaging coupled with phase 2022. Classification and feature extraction using supervised and unsupervised
analysis techniques for surface profiling of apples. J. Food Eng. 232, 11–20. https machine learning approach for broiler woody breast myopathy detection. Foods 11
://doi.org/10.1016/j.compag.2018.07.025. (20), 3270. https://doi.org/10.3390/foods11203270.
Lu, Y., Lu, R., 2019. Structured-illumination reflectance imaging for the detection of Sihvo, H.K., Immonen, K., Puolanne, E., 2014. Myodegeneration with fibrosis and
defects in fruit: analysis of resolution, contrast and depth-resolving features. Biosyst. regeneration in the pectoralis major muscle of broilers. Vet. Pathol. 51 (3), 619–623.
Eng. 180, 1–15. https://doi.org/10.1016/j.biosystemseng.2019.01.014. https://doi.org/10.1177/0300985813497488.
Lu, Y., Saeys, W., Kim, M., Peng, Y., Lu, R., 2020b. Hyperspectral imaging technology for Soglia, F., Mudalal, S., Babini, E., Di Nunzio, M., Mazzoni, M., Sirri, F., et al., 2016.
quality and safety evaluation of horticultural products: a review and celebration of Histology, composition, and quality traits of chicken Pectoralis major muscle
the past 20-year progress. Postharvest Biol. Technol. 170, 111318. https://doi. affected by wooden breast abnormality. Poultry Sci 95 (3), 651–659. https://doi.or
org/10.1016/j.postharvbio.2020.111318. https://doi.org/10.1016/j.postharvbio.20 g/10.3382/ps/pev353.
20.111318. Soglia, F., Silva, A.K., Lião, L.M., Laghi, L., Petracci, M., 2019. Effect of broiler breast
Lu, Y., Lu, R., 2020. Enhancing chlorophyll fluorescence imaging under structured abnormality and freezing on meat quality and metabolites assessed by 1 H-NMR
illumination with automatic vignetting correction for detection of chilling injury in spectroscopy. Poultry Sci 98 (12), 7139–7150. https://doi.org/10.3382/ps/pez514.
cucumbers. Comput. Electron. Agric. 168, 105145. https://doi.org/10.1016/j. Sun, X., Maynard, C.J., Caldas-Cueva, J.P., Coon, C.N., Owens, C.M., 2021. Using air
compag.2019.105145. deformation of raw fillet surfaces to identify severity of woody breast myopathy in
Lu, Y., Lu, R., 2021. Detection of chilling injury in pickling cucumbers using dual-band broiler fillets. LWT 141, 110904. https://doi.org/10.1016/j.lwt.2021.110904.
chlorophyll fluorescence imaging. Foods 10 (5), 1094. https://doi.org/10.3390/food Sun, Y., Lu, R., Lu, Y., Tu, K., Pan, L., 2019. Detection of early decay in peaches by
s10051094. structured-illumination reflectance imaging. Postharvest Biol. Technol. 151, 68–78.
Lu, Y., Lu, R., Zhang, Z., 2021. Detection of subsurface bruising in fresh pickling https://doi.org/10.1016/j.postharvbio.2019.01.011.
cucumbers using structured-illumination reflectance imaging. Postharvest Biol.
Technol. 180, 111624. https://doi.org/10.1016/j.postharvbio.2021.111624.

11
J. Cai and Y. Lu Journal of Food Engineering 391 (2025) 112459

Taheri-Garavand, A., Fatahi, S., Omid, M., Makino, Y., 2019. Meat quality evaluation Xing, T., Zhao, X., Zhang, L., Li, J.L., Zhou, G.H., Xu, X.L., Gao, F., 2020. Characteristics
based on computer vision technique: a review. Meat Sci 156, 183–195. https://doi. and incidence of broiler chicken wooden breast meat under commercial conditions
org/10.1016/j.meatsci.2019.06.002. in China. Poultry Sci 99 (1), 620–628. https://doi.org/10.3382/ps/pez560.
Tatarenkov, D.A., Buchatsky, A.N., 2018. Feature extraction from a depth map for human Xiong, R., Cavitt, L.C., Meullenet, J.F., Owens, C.M., 2006. Comparison of Allo–Kramer,
detection. In: 2018 Systems of Signal Synchronization, Generating and Processing in Warner–Bratzler and razor blade shears for predicting sensory tenderness of broiler
Telecommunications. pp. 1–4. https://doi.org/10.1109/SYNCHROINFO.2018. breast meat. J. Texture Stud. 37 (2), 179–199. https://doi.org/10.1111/j.1745-460
8456984. 3.2006.00045.x.
Tijare, V.V., Yang, F.L., Kuttappan, V.A., Alvarado, C.Z., Coon, C.N., Owens, C.M., 2016. Xu, J., Lu, Y., Olaniyi, E., Harvey, L., 2024. Online volume measurement of
Meat quality of broiler breast fillets with white striping and woody breast muscle sweetpotatoes by a LiDAR-based machine vision system. J. Food Eng. 361, 111725.
myopathies. Poultry Sci 95 (9), 2167–2173. https://doi.org/10.3382/ps/pew129. https://doi.org/10.1016/j.jfoodeng.2023.111725.
USDA-ERS, 2024. USDA Agricultural Projections to 2033. U.S. Department of Agriculture Yoon, S.C., Bowker, B.C., Zhuang, H., Lawrence, K.C., 2022. Development of imaging
Economic Research Service (USDA-ERS), Washington, D.C. https://www.usda.go system for online detection of chicken meat with wooden breast condition. Sensors
v/sites/default/files/documents/USDA-Agricultural-Projections-to-2033.pdf 22 (3), 1036. https://doi.org/10.3390/s22031036.
(accessed December 26, 2024). Zhang, S., 2018a. High-speed 3D shape measurement with structured light methods: a
Villegas-Cayllahua, E.A., de Mello, J.L.M., Dutra, D.R., de Oliveira, R.F., Cavalcanti, É.F., review. Opt Laser. Eng. 106, 119–131. https://doi.org/10.1016/j.optlaseng.2018.0
Pereira, M.R., et al., 2023. Effect of freezing on the quality of breast meat from 2.017.
broilers affected by wooden breast myopathy. Poultry Sci 102 (8), 102702. htt Zhang, S., 2018b. Absolute phase retrieval methods for digital fringe projection
ps://doi.org/10.1016/j.psj.2023.102702. profilometry: a review. Opt Laser. Eng. 107, 28–37. https://doi.org/10.1016/j.op
Wang, Y., Xu, H., Zhu, H., Chen, X., Wang, Y., 2023. Pixel-wise phase unwrapping with tlaseng.2018.03.003.
adaptive reference phase estimation for 3-D shape measurement. IEEE Trans. Zhao, H., Chen, W., Tan, Y., 1994. Phase-unwrapping algorithm for the measurement of
Instrum. Meas. 72, 1–9. https://doi.org/10.1109/TIM.2023.3241971. three-dimensional object shapes. Applied optics 33 (20), 4497–4500. https://doi.
Wold, J.P., Veiseth-Kent, E., Høst, V., Løvland, A., 2017. Rapid on-line detection and org/10.1364/AO.33.004497.
grading of wooden breast myopathy in chicken fillets by near-infrared spectroscopy. Zuo, C., Feng, S., Huang, L., Tao, T., Yin, W., Chen, Q., 2018. Phase shifting algorithms
PLoS One 12 (3), e0173384. https://doi.org/10.1371/journal.pone.0173384. for fringe projection profilometry: a review. Opt Laser. Eng. 109, 23–59. https://doi.
Wold, J.P., Måge, I., Løvland, A., Sanden, K.W., Ofstad, R., 2019. Near-infrared org/10.1016/j.optlaseng.2018.04.019.
spectroscopy detects woody breast syndrome in chicken fillets by the markers Zuo, R., Wei, S., Wang, Y., Kam, M., Opfermann, J.D., Hsieh, M.H., Krieger, A., Kang, J.
protein content and degree of water binding. Poultry Sci 98 (1), 480–490. https:// U., 2024. Deep learning-based single-shot fringe projection profilometry. Advanced
doi.org/10.3382/ps/pey351. Biomedical and Clinical Diagnostic and Surgical Guidance Systems XXII 12831,
Xie, S., Girshick, R., Dollár, P., Tu, Z., He, K., 2017. Aggregated residual transformations 1283106. https://doi.org/10.1117/12.3001837.
for deep neural networks. In: Proceedings of the IEEE Conference on Computer Wold, J.P., Løvland, A., 2020. NIR spectroscopic techniques for quality and process
Vision and Pattern Recognition, pp. 1492–1500. control in the meat industry. Meat Muscle Biol. 4(2), 8, 1–8. https://doi.org/
10.22175/mmb.10020.

12