J fS C: food Chemistry and Toxicology

Classification of Fresh and Frozen-thawed

Fish by Near-infrared Spectroscopy
M. UDDIN AND E. OKAZAKI

ABSTRACT: Frozen fish usually have a much lower market price than fresh fish: therefore, adulteration could
occur. This article focuses on the use of near infrared (NIR) spectroscopy to detect whether fish has been frozen-
thawed because NIR spectroscopy has demonstrated the potential for addressing some authenticity issue in
foods and is known to be a nondestructive rapid technique. Horse mackerel (n = 162) were evaluated as fresh and
frozen-thawed fish sample. Dry extract spectroscopy by infrared reflection (DESIR) of fresh and frozen-fhawed
fish samples was performed on the meat juices then discriminated by principal component analysis ( P a ) and
multiple linear regressions (MLR). In DESIR specwa, the overall absorbance level was found to decrease in
frozen-thawed samples, indicating the different chemical composition of juice, amount of dry matter, particle
size, and their scattering properties. The spectral changes that take place between fresh and frozen-thawed
samples are clearly seen in the 1920- to 2350-nm region. The spectra are dominated by peaks attributed to
proteins, in particular, peaks at 1510, 1700, 1738, 2056, 2176,2298, and 2346 nm. It was found that fresh and
frozen-thawed fish could be separated 100%correctly by DESIR technique.
Keywords: fresh and frozen-thawed fish, NIR spectroscopy, DESIR, MLR, PCA

Introduction fers the possibility of measuring physical and chemical properties.

Once calibrated, the NIR spectrometer is simple to operate (Bun-
P roduct authenticity is an emerging area of concern within the
food industry. It is generally accepted that fresh and frozen-
thawed fish are types of products that should be differentiated. For
ing-Pfaue 2003). It has been widely used in the food industry and
is based on the electromagnetic absorption of organic compounds
the benefit of the consumer and prevention of unfair competition (Misra and others 2000).
in the trade of fishery products, correct labeling of frozen-thawed During freezing and thawing of meat, ice crystal growth causes
fish or fillets is desirable. With regard to fish and fishery products, biochemical and physical changes. The later result in the disruption
1 of the major issues is the substitution of frozen-thawed for fresh. of cellular organelles and release of their contents into the meat
The Food and Agricultural Organization (FA0 1982) Code of Practice drip juice. Spectroscopic methods, particularly NIR spectroscopy,
defines fresh fish as "freshly caught fish, which have received no have demonstrated the potential for addressing some authentic-
preserving treatment or which have been preserved only by chill- ity issue in foods and are known to be rapid and noninvasive tech-
ing" and frozen fish as"fish, which have been subjected to a freez- niques (Ding and Xu 1999 Cozzolino and others 2002;Downey and
ing process sufficient to reduce the temperature of the whole prod- others 2003). The physical and chemical changes of frozen-thawed
uct to a level low enough to preserve the inherent quality of the fish fish meat more specifically in juice level might be detected by NIR
and which have been maintained at this low temperature during spectroscopy. Therefore, an attempt has been made to classify
transportation, storage and distribution up to and including the fresh and frozen-thawed fish by NIR spectroscopy, which is'also
time of final sale." Consequently, control of labeling is possible only known to be a nondestructive technique. It is fast, easy to operate,
if rapid and reliable methods exit, which allow food control author- and needs no chemicals. These make it possible to run on- or at-line
ities to distinguish between fresh and frozen-thawed fish or fillets. in the fish processing plant.
Several methods to discriminate between unfrozen and frozen
fish or fillets have been reported (Love 1956; Yoshioka and Kitami- Materials and Methods
kado 1988; Sakaguchi and others 1989; Kitamikado and others
1990; Rehbein 1992). However, all the methods reported are either Fish sample
time-consuming or have limitations for practical uses. NIR spectros- One hundred sixty-two live horse mackerel 7kachum.sjuponkus
copy offers a number of important advantages over traditional were purchased from Kanazawa prefecture, Japan, and were trans-
chemical methods. It is a physical and very fast method, requiring ported with sea water to the Natl. Research Inst. of Fisheries Science,
minimal or no sample preparation and its precision can be high. In Yokohama, Japan. The fish used in this study were between 88.2 and
contrast with traditional chemical analysis, no reagents are required 236.7 g (average, 144.6 g) and had a fork length between 16.7 and
and no wastes are produced. It is a multianalytical technique: sev- 25.3 cm (average, 20.3 cm). The fish were killed in ice-cold water
eral determinations can be made simultaneously. The method of- shock in a big plastic bucket, which ailowed us to keep them i n j q -
free within 10 min. Fish were divided into 3 equal groups and used
MS 20040305 Submitted 511 1/04,Revised 6/1/04,Accepted 7/7/04.The au- for further evaluation. For fresh or unfrozen fish, 54 samples were
thors are with Diu. of Food Technology and Biochemktv, Natl. Research used soon after they were killed, followed by temperature equilibra-
h t . of Fisheries Science, 2-12-4 Fukuura, Kanarawa, Yokohama236-8648, tion at 20 "C while the 2nd lot of 54 fish was kept at -40 "C.The re-
Japan. Direct inquiries to author Uddin (E-mail:musleh@affrc.ao.la).
maining 3rd lot of 54 fish was kept as a filleted state at -40 "C. After

(D 2004 Institute of Food Technologists Vol. 69, Nr. 8, 2W-JOWNAL OP QOOO BCWNCB co65
Fultherreproduction without permissin is prohblted Published on Web 10/14/2004
fresh a n d frozen-thawed fish by NIR ...
10 d, frozen stored fish and fillets were removed and thawed over- 1993).The classifiers were assigned a score for each sample on the
night at 5 "C and then used as frozen-thawed samples. basis of its NIR spectral data. A fish sample with a score below 1.5
was classified as fresh, whereas a sample with a score greater than
NIR measurement 1.5 was classified as frozen-thawed. PCA and MLR operations were
Dry extract spectroscopy by infrared reflection (DESIR), modi- preformed by using The Unscrambler (Version 8.05, Camo, NJ.,
fied from Alfaro and others (1990),was performed on the meat U.S.A.) and Vision Spectral Software packages (version 2.1 1. NIR-
juice or natural drip juice from fresh/frozen fish or fillets. Meat juic- Systems, Md., U.S.A.), respectively.
es were collected by specially constructed centrifuge tubes in
which 15 g fish meat was centrifuged for 10 min at 10000 rpm with Results and Discussion
a disc of a 25-mm dia, perforated with 446 circular holes of 0.25-mm
dia and 30 mm above the bottom. An aliquot of 0.6 mL extract or 0 ne hundred and sixty-two average meat juices or drip spectra
were collected from the same numbers of individual fish sam-
natural drip juice (collected from each polyethylene bag) was ples in which 54 samples of fresh, frozen whole, and frozen fillets
poured onto the center of a glass fiber filter paper (Whatman Glass equal. To identify fresh fish from frozen-thawed fish, spectral chang-
Filter, GFlE 5.5 cm) and then shaken. After waiting for diffusion of es induced by frozen and then thawed fish have been studied in 3
the liquid through the filter paper (normally 1.5 to 2 min), the filter individual groups. It was found that both frozen-thawed meat juice
paper was dried at 30 "C for 30 min using an temperature-con- and frozen-thawed natural drip juice had identical spectral patterns
trolled air-circulated incubator (Eyela, LTI-GOlSD, Tokyo, Japan) that overlapped each other. Spectra of single samples of fresh and
and then removed and cooled at room temperature in desiccators frozen-thawed centrifuged juices are shown in Figure 2.These sam-
with silica gel to protect moisture absorption. The filter paper was ples were chosen at random and may be expected to be representa-
then scanned in diffuse reflectance mode. Duplicate filter papers tive of the entire collection with regard to alterations in spectral fea-
were prepared and scanned for each sample. The measurements tures. The spectra in Figure 2 are dominated by peaks attributed to
were performed, in the 1100 to 2500 nm range, on the deposit side protein (Murrayand Williams 1990;Osbome and others 1993),in par-
in a NIR spectrophotometer (model 6500, NIRSystems Inc., Silver ticular, peaks at 1510, 1700,1738,2056,2176,2298,and 2350 nm
Spring, Md., U.S.A.), with a Standard Sample Cup, which was man- (Shenk and others 1992; Kuenstner and Norris 1994). The overall
ually rotated in 5 steps. NIR reflectance spectra were recorded in 2- absorbance level was found to decrease in frozen-thawed sample.
nm steps. Figure 1 show the experimental flow diagram used in this The major differences between the spectra involve a baseline shift,
study to differentiate fresh and frozen-thawed fish. which may be due to variation in the quantity of dry materials on
The protein concentration of extract and drip juice was deter- each filter paper. In addition to an additive effect, this variation also
mined according to Lowry and others (1951) using a commercial may be expected to alter the scattering properties of filter paper, pro-
preparation (DC Protein Assay Reagents, BIO-RAD Laboratories, ducing the slope change evident in Figure 2. Irreversible denatur-
Calif., U.S.A.). ation and physical damage of proteins could be important predictors
in this respect. Smaller fractions will in general give increased light
Multivariate data analysis scattering. This might also affect the meat juice, causing a lower ab-
'Mo multivariate analytical models were used in the classifica- sorbance of frozen-thawed than fresh DESIR samples.
tion of fresh and frozen-thawed fish: principal component analysis Visual inspections of the original spectra (Figure 2) did not show
(PCA) and a dummy regression by stepwise multiple linear regres- any obvious special characters that could be used for classification
sion (MLR). PCA is a linear modeling method and compresses major analysis otherwise the log (1 IR)values decreased at all wavelengths
variations in original data into principal components. Data sets in frozen-thawed as opposed to fresh samples. However, 2nd deriv-
were subjected to PCA and the scores plots were examined (Byme ative spectra showed special characters and there were clear differ-
and others 1998).In MLR analysis, spectra (n= 162) from fresh and ences in terms of intensity at specific wavelengths among the spec-
frozen-thawed samples were randomly selected and assigned to a tra of fresh and frozen-thawed DESIR samples (Figure 3). It is well
calibration set and validation set. Fresh samples in the calibration known that the derivative treatment useful for noise reduction in-
set were assigned a dummy score of 1, and the frozen-thawed sam-
ples were assigned a dummy score of 2. Classification equations
were developed using stepwise MLR technique (Chen and Massie

I I
Flgum 1-An experiment8l flow dlagnm for c l 8 s s l t i ~ o n Agum 2-Wd rdlectanco spoctrm of froah and frozen-
of fresh and froun-th8wed flsh thawed horse mackerel Julces

C666 JOURNAL OF FOOD SCIENCE-Vol. 69, Nr. 8,2004

Fresh and frozen-thawed fish by NIR ...
creases resolution of spectrum peaks. The 2nd derivative operation Table 1-Resuits of the calibration and prediction for de-
was chosen in this study, which has been widely used for NIR spec- tecting fresh and frozen-thawed hone mackerel centri-
fuged Juice by multiple linear remreuions IMLRP
tral data analysis because it presents distinct absorption peaks in
the same location as they appear in the log (1IR) spectrum (Chen Wavelength selected (nm)
and Massie 1993;Ding andxu 1999; Uddin and others 2002). In 2nd A1 A2 A3 R SEC (%I SEP(%) Bias (%I
derivative form, the peak amplitudes were greater for fresh DESIR 2180 0.953 0.085 0.063 -0.002
sample, suggesting that differencesin absorbance are related to.the 2180 2056 0.982 0.057 0.033 -0.005
frozen-then-thawing treatment. Distinct separation of fresh and 2180 2056 2298 0.994 0.051 0.024 -0.001
frozen-thawed samples occurred at several absorption bands asample number: 54 for calibration set and 54 for prediction set.
throughout the spectra; however, they are more clearly visible be- bbias = the average of difference between actual value and near infrared
(“3) value; R = multiple correlation coefficient; SEC = standard error of
tween the 1920 and 2350 nm region. Spectral changes between calibration; SEP = bias corrected standard error of prediction.
fresh and frozen-thawed samples were recorded by computer and
a discrimination model was developed using MLR and PCA tech-
niques from the 2nd derivative spectrum because it produces dis-
tinct peaks. samples was also performed on the basis of PCA scores. PCA in-
MLR analysis allowed us to select appropriate wavelengths relat- volves decomposing 1 data matrix, X, into a “structure” part and a
ed to specific known chemical bonds without interference. MLR is “noise”part. The most commonly used plot in multivariate data
also useful because the noise reduction from omitting noninforma- analysis is the score vector for PC1 versus the score vector for PC2.
tive parts of the spectrum has a grater advantage than the advan- This is easy to understand because these are the 2 directions along
tage often gained by using all data. The wavelengths selected by a which the data swarm exhibits the largest and the 2nd largest vari-
stepforward-stepreverse regression in this study to provide the ances. Spectra from both sets of fresh and frozen-thawed samples
calibration model with the lowest standard error of calibration were merged into a single dataset. On examination by PCA,the
(SEC) and highest correlation coefficients of calibration (R)are giv- spectra were seen to form 2 separate clusters along the 1st compo-
en in Table 1. To protect overfitting of the calibration equation, nent axis (Figure 5). The score plots for PC1 versus PC2 reveal out-
only 3 specific wavelengths were calculated. The 1st wavelength se- lying samples in the sample score distributions. PCA was able to
lection, which constitutes the most important spectral region and separate 100%of fresh and frozen-thawed samples,which was also
1 of the major steps for MLR analysis, at 2180 nm due to the absor- performed by MLR. The major portion of dry matter in drip or cen-
bance of protein, was selected manually. This 2180 nm known to be trifuged juice is protein that was found to be 106,129, and 131 mgl
a key absorption band is related to protein in NIR spectroscopic mL for fresh juice, frozen-thawed juice, and natural drip juice, re-
analysis for agricultural products (Shenk and others 1992; Osborne
and others 1993). As the number of wavelengths selected for the
discrimination model increased, the classification accuracy was also
increased. Scatter plots of the test set for the MLR model using lin-
ear functions of reflectance spectra are shown in Figure 4. The nu-
merical value represents scores for fresh and frozen-thawed fish
samples described previously in the “Multivariate data analysis”
section. Selecting only 3 specific wavelengths (2180,2056,and 2298
nm), which are extremely related to protein (Murray and Williams
1990; S h e d and others 1992), fresh and frozen-thawed fish were
clearly separated by NIR reflectance spectroscopy. Using the MLR
technique, 100%were correctly classified as fresh (n= 54) or frozen-
thawed (n= 54) fish samples.
The classification analysis between fresh and frozen-thawed

Flgure I-scattOr plat.of th.k.t .aUdng the rofkctanii

of the fresh for cladfylng trorm-umwodhorse mackerel.
(a) Fresh Juice8nd frozen-tluwed natural drlp; (b) froah and
frozen-thawed Juice.

Figure 6-Prlnclpal component .~ll(.l~ -4 score plots

ot the tn.h for cl8dlyiW --th.md horrr
(a) Fresh Jula;. and trorm-thawod ruhwd arP; (b)fmd~
and
-.
Flaun S-Socond doriv.tlve spectra of fresh and frozen-
thawed h o w mackerel Julces frozen-thawed Juice.

Vol. 69, Nr. 8, 2004-JOlJftNAL OP eoOD 8 C W B Coo7

fresh and frozen-thawed fish by NIR ...
spectively (data not shown). The nature of dry matter in filter pa-
per, particle size, and their scattering properties might lead to clas-
sify each other by PCA,which compresses major variations in orig-
inal data into principal components.
To differentiate fresh and frozen-thawed fish, measurement of
the electric properties of fish tissues, visual inspection of the eye lens,
judgments of the integrity of red blood cells by microscopy,or estima-
tion of the hematocrit value were proposed (Yoshioka 1983; Rehbein
1992).In practice, proposed methods cannot be applied to those fish
or fillets possessing no blood, eye lens, or skin. Determination of the
release of enzymes originally bound to mitochondria, lysosomes, or
red blood cells were also investigated (Rehbein and Cakli 2000);
however, all techniques mentioned previously are tedious and time-
consuming. On the contrary DESIR could minimize those aforemen-
tioned limitations and also be a faster screening technique. Using
natural drip loss is nondestructive and can be performed with a por-
table drying unit when small amounts of drip are available. In a pre-
liminary study, we applied this DESIR technique to differentiate
fresh and frozen-thawed bigeye tuna (Thunnusobesus) and amber-
jack (Seriolapurpurscens) with a small number of sample sets, which
were also 100% identified. Results obtained in this work require ex-
tension to a larger set of different fish samples for validation, but
they strongly suggest the utility of this approach, at the very least, as
a screening procedure. A further study using a surface interactance
fiber-optic accessory is in progress.
Conclusions
LR and PCA analysis of NIR reflectance spectra have been
M shown to possess the potential for distinguishingbetween
fresh and frozen-thawed fish samples under the experimental con-
dition of this work. It was found that fresh and frozen-thawedfish
could be 100% correctly separated by the DESIR technique. With a
simple press mechanism, pressed juice from a small piece ofmeat
could also be used with a portable drying unit, which allows this
analysis within 10 to 15 min.
Acknowledgments
‘The authors would like to acknowledge Dr. Sumio Kawano and Dr.
Srinnapa Saranwong for technical assistance in data analysis.
C668 JOURNAL O f QOOD SCIENCE-Vol. 69, Nr. 8, 2004
References
Alfaro G , Meurens M. Birth GS. 1990. Liquid analysis by dry extract near-infra-
red reflectance on fiberglass. Appl Spectrosc 44:979-86.
Buning-pfaue H. 2003. Analysis of water in food by near infrared spectroscopy.
Food Chem 82:107-15.
Byrne CE, Downey C, Troy DI, Buckley D1. 1998. Non-destructive prediction of
selected quality attributes of beef by near infrared reflectance spectroscopy
between 750 and 1098 nm. Meat Sci 49:399-409.
Chen YR. Massie DR. 1993. Visible/near infrared reflectance and interactance spec-
troscopy for detection of abnormal poultry carcasses. Trans ASAE 36:863-9.
Cozzolino D, Martins V. Murray 1. 2002. Visible and near infrared spectroscopy
of beef longissirnus dorsi muscle as a means of discriminating between pas-
ture and corn silage feeding regimes. J Near Infrared Spectrosc 10:187-93.
Ding HB. Xu RJ. 1999. Differentiation of beef and kangaroo meat by visible/
near infrared reflectance spectroscopy. I Food Sci 64:814-7.
Downey G , Fouratier V, Kelly ID. 2003. Detection of honey adulteration by addi-
tion of fructose and glucose using near infrared transflectance spectroscopy.
J Near Infrared Spectrosc 11:447-56.
[FA01 Food and Agriculture Organization of the United Nations. 1982. Refer-
ence manual to codes of practice for fish and fishery products. FA0 Fish Circ.
750. 217.
Kitamikado M. Yuan C, lleno R. 1990. An enzymatic method designated to differ-
entiate between fresh and frozen-thawed fish. J Food Sci 55:74-6.
Kuenstner TT,Norris KH. 1994. Spectrophotometry of human hemoglobin in the
near infrared region from I000 to 2500 nm. J Near Infrared Spectrosc 259-65.
Love RM. 1956. Post-mortem changes in the lenses ot fish eyes. I I . Effect of freez-
ing. and their usefulness in determining the past history of the fish. 1 Sci Food
Agric 7:220-6.
Lowry OH, Rosebrough N J , Farr AL, Randall RI. 1951. Protein measurement with
the fohn phenol reagent. 1 Biol Chem 193:265-75.
Misra JB. Mathur RS. Bhatt DM. 2000. Near-infrared spectroscopy: a potential
tool for non-destructive determination of oil content in groundnuts. J Sci Food
Agric 80237-40.
Murray I , Williams PC. 1990. Chemical principles of near-infrared technology.
in: Williams PC. Norris KH, editors. Near-infrared technology in the agricul-
tural and food industries. St. Paul, Minn.: American Assn. of Cereal Chemists.
Oshorne B G Fearn T, Hindle PH. 1993. Practical spectroscopy with application in
food and beverage analysis. Harlow, U.K.: Longman Scientific and Technical.
Rehbein H. 1992. Physical and biochemical methods for the differentiation be-
tween fresh and frozen-thawed fish or fillets. ltal I Food Sci 275-86.
Rehbein H,Cakli S. 2000. The lysosomal enzyme activities of fresh, cooled, fro-
zen and smoked salmon fish (Onchorhyncus kera and Salrno snlar). Turk I Vet
Anim Sci 24:103-8.
Sakaguchi M, Murata M. Kim JB. 1989. The effects of repeated freeze-thaw cycle
on torrymeter readings of carp fillets. Nippon Suisan Gakkaishi 55:1665-9.
Shenk IS.Workman JJ. Westerhaus MO. 1992. Application of N I R spectroscopy to
agricultural products. In: Burns DA. Ciurczak EW, editors. Handbook of near-
infrared analysis. New York: Marcel Dekker. Inc.
llddin M, lshizaki S, Okazaki E. Tanaka M. 2002. Near-infrared reflectance spec-
troscopy for determining end-point temperature of heated fish and shellfish
meats. J Sci Food Agric 82286-92.
Yoshioka K. 1983. Differentiation of freeze-thawed fish from fresh fish by the
determination of hematocrit value. Bull lpn SOCSci Fish 49:149.
Yushioka K, Kitamikado M. 1988. Differentiation of freeze-thawed fish fillet from
fresh fish fillet by the examination of erythrocyte. Nipp Suis Cakkai 54:1221-5.