The quality monitoring of frozen marine products has become essential in the fishery industry, where efficient and effective quality assurance is becoming increasingly important. In this study, we proposed a novel method of evaluating fish quality by combining the fluorescence excitation-emission matrix (EEM) with imaging techniques to visualize the spatial-temporal changes of freshness indices such as K-value and taste component IMP content. The result showed that the distribution of K-value and IMP content could be visualized with accuracy of R2 = 0.78 and R2 = 0.83, respectively. Furthermore, this innovative approach was applied to differentiate burnt meat, which is a type of abnormal meat found in many types of fish, and it was found that burnt meat could be detected even when in a frozen condition.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Freshness is regarded as one of the vital parameters for the quality assessment of fish and fish products. With the increasing globalization of the sale of fish products, the demand for frozen marine products, such as tuna, mackerel, cod, salmon etc., is increasing day by day. Therefore, the quality monitoring of frozen marine products has become essential in the fishery industry, and efficient and effective quality assurance is becoming increasingly important. Improved methods for determining freshness and quality are sought by processors, consumers, and regulatory officials . Nowadays, attention is focused on developing rapid, reliable and non-destructive techniques at moderate costs for monitoring seafood quality and freshness to verify that it is safe for human consumption. A number of sensory and instrument methods have been proposed to evaluate the state of fish freshness [2, 3]. Sensory methods required trained personnel and is somewhat time consuming, and therefore are considered costly and not always practical for large-scale commercial purposes. As chemical and biochemical methods for the evaluation of freshness eliminate personal opinions in quality scoring based on organoleptic changes occurring as fish storage time is extended, they are, accordingly, considered more reliable and accurate than sensory methods.
In the chemical methods, concentrations of adenosine 5′-triphosphate (ATP) and its breakdown products, which are adenosine 5′-diphosphate (ADP), adenosine 5′-monophosphate (AMP), inosine 5′-monophosphate (IMP), inosine (HxR), and hypoxanthine (Hx), respectively, are used as indices of freshness quality in a wide variety of fish [4–9]. The ratio among all or some of these nucleotide breakdown compounds are commonly used as indicators of freshness quality. The IMP content increases after decomposition of ATP and decreases after maintaining for a certain period of time and is well known as a component strongly related to the “umami taste” of fish . K-value, which is defined as the ratio of non-phosphorylated ATP metabolites to the total ATP breakdown products, was suggested in 1958 by a Japanese research group as an objective index of fish freshness .
Traditional methods using either sensory or instrumental evaluation, can provide reliable information about fish quality, however, these methods are destructive, expensive, time-consuming, and require highly skilled personnel. Note that, once fresh and un-fresh fish are frozen, they generally look the same and it would be rather difficult to differentiate them with the naked eye. Moreover, we must consider the influence of thawing when frozen fish is evaluated. The only way to discover the difference between fresh and un-fresh states either optically or destructively is by using conventional chemical analyses on frozen fish. An excitation-emission matrix (EEM), also known as fluorescence landscape  or fluorescence fingerprint , is a set of fluorescence spectra acquired at consecutive excitation wavelengths to create a three-dimensional diagram. The EEM has been widely applied for nondestructive measurement of the physical and chemical properties of objects [14–17]. In fish quality assessment, EEM has been applied to estimate freshness indices (such as K-value) and ATP content in frozen fish with high precision [18–20]. However, this approach is a point measurement which can only estimate quality at one point, and the freshness condition of other parts of fish body could not be tracked. Therefore, the previous methods are not practical for large-scale commercial purposes such as that of mackerel, which involves a vast number of fish and whose freshness declines quickly, or a big fish such as tuna, where the progress of freshness change varies for individual fish, lots, and parts.
In this study, we focus on a novel fluorescence imaging method, in which EEM is combined with imaging techniques, to visualize the distribution of fish quality such as K-value and IMP content. This approach based on expanding the point estimation of EEM to image, where each point in EEM corresponds to one image measured under a specific excitation-emission wavelength. We also propose an optimization method to reducing the dimensions of the EEM by selecting the most efficient excitation wavelength, which allow us to visualize both K-value and IMP content using only one excitation light. Firstly, we prepared fish samples with different freshness conditions, then measured the fluorescence spectra (EEM) and the fish quality (K-value and IMP content) of the samples. By using the measured EEM data to estimate K-value and IMP content, the most efficient excitation wavelength was selected for visualization. After that, we measured fluorescence images under the most efficient excitation wavelength and then built the visualization model from the measured fluorescence images.
In experiment 2, the obtained visualization model was applied to solve the on-site problem. In previous studies, it has been reported that the quality of fish depends on the killing procedures. In general, fishes in the market are caught with fishing nets and some of them likely die while struggling in them. The result of a violent struggle during capture makes muscle soften faster and could not be eaten raw in the struggled samples compared to instantly killed samples . This problem, which known as “burnt meat” or, in Japanese, as “yake niku”, is a type of abnormal meat that occurs in many types of fish such as tuna, amberjack, mackerel etc. Instead of being translucent, firm and possessing a delicate flavor, burnt meat is pale, exudes a clear fluid, and has a soft texture and slightly sour taste . However, burnt fish and instantly killed fish generally look the same when frozen and is somewhat difficult to differentiate them using the naked eye. In experiment 2, the visualization model obtained in experiment 1 was applied to identify burnt fish samples.
2. Material and methods
2.1. Experiment 1
The first group of alive spotted mackerel (Scomber australasicus) with an average weight of 341 ± 72.3 g and length of 29.9 ± 1.9 cm was harvested from a fish cage (Kamaishi City, Iwate Prefecture, Japan). Twenty four fresh fish were immediately killed by neck breaking, put in slurry ice for blood removal as well as to minimize the changes of the freshness condition and were then transferred to the laboratory. All fish samples were beheaded, gutted, had their tail cut and then vacuum packed. Three samples were filleted, vacuum packed and immediately frozen, while the rest were stored in iced water in a low temperature room at 1 °C for 3.5 h, 1, 2, 3, 5, 7, 9 days to simulate the different freshness conditions, then filleted, vacuum packed and frozen by air blast freezer at −60 °C. There were eight different freshness conditions and three spotted mackerel were used for each condition, yielding 24 fillets of the left-side and 24 fillets of the right-side. The left-side fillets were used to measure EEM and ATP-related compounds, and the right-side fillets were used to measure fluorescence image.
Fluorescence spectra (EEM) measurement
The fluorescence spectra of the frozen fillets were measured by using a fluorescence spectrophotometer F-7000 (Hitachi High-Tech Science Corporation). In this experiment, the left-side fillets were used and placed inside the freezer with dry ice to maintain the temperature of samples and environment below −30 °C. EEM data at two points (A, B) as shown in Fig. 1 were measured using an external Y-type fiber optic probe. At each point, EEM was obtained by measuring the emission intensity in 10 nm intervals between 250 ~ 800 nm while scanning the excitation wavelengths from 250 ∼ 800 nm in 10 nm steps. The slit width was set at 20 nm for both excitation and emission and scan speed was set at 30,000 nm/min. The photomultiplier voltage (PMT voltage) was adjusted to 350 V throughout the entire experiment.
After obtaining the fluorescence spectra of all frozen fillet samples, the cylindrical subsamples were cut from the EEM acquired positions of the frozen fillet (Fig. 1) for the analysis of ATP-related compounds. The muscle extraction was performed according to Ehira and Uchiyama (1986) . The dissecting of subsamples was accomplished inside a cold room (4.5 °C), and the frozen fillets and cutting tools were kept cool using dry ice. The solution was frozen and stored at −60 °C until the HPLC analysis.
According to Maeda et al. , ATP, ADP, AMP, IMP, HxR and Hx in the muscle extracts of frozen fillets were determined using a high-performance liquid chromatography (HPLC) system. After acquiring the data of all ATP-related compounds from HPLC, the K-value was calculated according to Saito et al.  by the following equation:
Fluorescence image measurement
Figure 2 shows the experimental setup for measuring fluorescence images. In this experiment, the right-side fillets were used for the image measurement. Each frozen fillet sample was placed in the middle of a styrofoam box with dry ice inside to minimize temperature changes of the samples. The temperature inside the styrofoam box was kept under −50 °C. The samples were illuminated by an excitation at 340 nm, which is the most efficient excitation wavelength for freshness prediction of spotted mackerel in a frozen condition (described at 3.1), using MAX-303 light source (ASAHI SPECTRA). The fluorescence emitted from the sample was filtered through a band-pass filter to obtain a fluorescence image at a specific emission wavelength. The emission wavelength was controlled by changing the band-pass filter at the filter system. The BU-56DUV CCD camera (Bitran) was used for measuring the fluorescence image at every 10 nm from 380 to 630 nm. The 2×2 binning  was used in the measurement with the exposure time set to 1500 ms for each image. A total of 31 images were measured in each sample. The measured fluorescence images contained dark noise [26, 27] which was corrected by subtracting a dark image acquired by closing the camera shutter and covering the lens with a lens cap so that no light could enter through the lens.
The fluorescence images obtained depend on camera sensitivity. The intensity of fluorescence images is different for different wavelengths and changes if other cameras are used. Device-independent fluorescence images can be obtained by using a spectroradiometer and a camera to measure the same object, and then compare the intensity in each wavelength to calibrate the camera sensitivity. In this study, MAX-303 (ASAHI SPECTRA) was used as the light source with the light covered from 380∼780 nm. We used SR-3AR (TOPCON) as the spectroradiometer and measured the white patch in Macbeth ColorChecker (X-rite) from 380 ~ 780 nm. The correction function for BU-56DUV CCD camera (Bitran), which was used for measuring fluorescence images, was calculated and used to calibrate the fluorescence images obtained from measurement.
2.2. Experiment 2
The second group of alive spotted mackerel (Scomber australasicus) were harvested from the same fish cage (Kamaishi City, Iwate Prefecture, Japan). Eight fresh fish with an average weight of 309.8±90.2 g and length of 30.3±5.2 cm were killed by struggle in air for 30 min. All fish samples were stored in iced water at 0 °C for 2, 4, 24, 40 h (4 different freshness conditions) to stimulate the different freshness conditions, then filleted, vacuum packed and frozen by air blast freezer at −60 °C. There were two struggle spotted mackerel for each condition.
Fluorescence image measurement
The fluorescence images of each of the samples were measured with the condition the same as in experiment 1. The samples were illuminated by an excitation at 340 nm. The fluorescence emitted from the sample was filtered through a band-pass filter to obtain a fluorescence image at every 10 nm from 380 to 630 nm.
3. Results and discussion
3.1. Experiment 1
EEM preprocessing and analysis for predicting the fish freshness
Figure 3(a) shows the EEM spectra obtained from the fish fillet measurement. The EEM was formed by recording fluorescence intensities at an emission wavelength range of 250×800 nm under the same excitation wavelength range of 250×800 nm. The raw data of EEM includes some parts that do not contain a fluorescence property. There is hypothetically no emission below the excitation based on Stokes’ shift . Besides, owing to light scattering effects such as the Raman and Rayleigh effects, a typical scattering problem normally exists in any excitation-emission matrix [29, 30]. Scattering signals and those areas whose emission wavelengths are shorter than the excitation wavelengths do not carry relevant chemical information and should be entirely excluded from the EEM before commencing the building of the calibration models. The preprocessed EEM spectra masked only the fluorescence area after removing those irrelevant areas is shown in Fig. 3(b).
After EEM preprocessing (i.e. removed irrelative area), the data size of each EEM was reduced to 1054 from 3136 variables (each variable corresponds to one excitation-emission wavelength combination of original spectra). The reduced EEM data (1054 variables) was used to predict the measured K-value and IMP content by a partial least squares (PLS) regression model. The PLS model was built under a leave-one-out cross-validation that used one sample as the validation and the remaining samples as the training set. The prediction accuracy is compared quantitatively using the coefficient of determination (R2), standard error of prediction (SEP) and latent variable (LV). The PLS models revealed that EEM could be used to predict K-value and IMP content with high accuracy (R2 = 0.86 and R2 = 0.84, respectively. This is very similar to the result of ElMasry et al. , which used all fluorescence intensity at 1054 variables to predict K-value, with a prediction accuracy of R2 = 0.86.
Selecting the most efficient excitation wavelength
In the PLS prediction models, we used all fluorescence intensity at 1054 variables on each EEM as predictors, where each variable corresponds to one excitation-emission wavelength combination. Accordingly, when expanding this prediction model to the image, it is necessary to measure a fluorescence image at 1054 excitation-emission wavelength combinations for each target, which mean 1054 images. Therefore, in order to visualize the frozen fish quality expressed as K-value and IMP content more competently, reducing the high dimensionality of EEM data is required.
In this study, the dimension of EEM was reduced by performing repeated PLS regression modeling using all emission wavelengths in the range under each excitation wavelength to identify the most effective excitation wavelength. The results of wavelength selection for K-value and IMP content were shown in Figs. 4(a) and 4(b), respectively and average of them in Fig. 4(c). The prediction accuracy differs for each excitation wavelength and freshness index (K-value and IMP content). In the graph of average, at the excitation wavelength range of 300~350 nm, the prediction of these freshness indices is lower than those of all 1054 variables though accuracy is still high (R2 is more than 0.79). The highest prediction accuracy is at an excitation wavelength of 310 nm (R2 = 0.83) and the next higher prediction accuracy is at an excitation wavelength of 340 nm (R2 = 0.80).
Due to the limitations of the equipment used in the experiment, there is no spectroradiometer that can measure whole emission wavelength (350∼570 nm) under an excitation wavelength of 310 nm. For this reason, 340 nm was chosen as the most efficient excitation wavelength instead of 310 nm to visualize fish freshness. Note that, the prediction accuracy under 340 nm is almost equivalent to 310 nm.
Visualization of the quality of frozen fish
In samples preparation (see Section 2.1), the fish samples were filleted after being stored in a refrigerator for a specific period, just before being vacuum packed and frozen in a freezer. Thus, in this study, we assumed that the freshness (K-value and IMP content) of the left-side fillets are the same as the freshness of the right-side fillets at the corresponding point.
In order to visualize the frozen fish freshness expressed by K-value and IMP content, we need to build a prediction model from the measured fluorescence images. The areas corresponded to the fluorescence measurement and the chemical analysis (Fig. 1) were used for analysis. For each fluorescence image, those areas were masked and the pixels value inside the mask was averaged. There were 31 fluorescence images, yielding 31 variables obtained for each point. In this study, we have 24 fish samples and two points were measured for EEM for each fish sample, consequently 31 variables at 48 points were obtained. Those 31 variables were used for predicting the K-value and IMP content at corresponding points by PLS regression model. The PLS model was built utilizing leave-one-out cross-validation which used data at one point as the validation and the remaining points as the training set. The prediction accuracy is evaluated by using the coefficient of determination (R2) and standard error of prediction (SEP).
The prediction accuracy of the PLS models from the measured fluorescence images were R2 = 0.78 for K-value and R2 = 0.82 for IMP content. These are lower than those prediction models using all 1054 variables of EEM (R2 = 0.86 for K-value and R2 = 0.84 for IMP content). However, the aim of the visualization is to make visible the spatial-temporal freshness changes of a whole sample with acceptable accuracy. Thus, the prediction accuracy of these PLS models (R2 = 0.78 for K-value and R2 = 0.82 for IMP content) is high enough and are acceptable for the visualization.
From the obtained prediction models, the distributions of K-value and IMP content can be estimated by combining the parameters of the prediction models with fluorescence image at corresponding emission wavelength to obtain the freshness distribution image as follows:Fig. 5. There were three samples for each storage time condition, which correspond to three rows in figure. The samples in Fig. 5(a) are same as samples in Fig. 5(b), which are result of visualize K-value and IMP content, respectively. As a result, K-value increased in accordance with ice storage time. On the other hand, IMP content rapidly increased for the storage period from 0 to 1 day, and then gradually decreased from 2 to 9 days.
The measured K-value and IMP content obtained from chemical analysis are shown in Figs. 6(a) and 6(b), respectively. As a result, K-value increased linearly with storage time. On the other hand, IMP content increased dramatically from 0% to 90% for the samples from 0 to 1 day, and then gradually decreased for the samples from 2 to 9 days. From the visualization result obtained in Fig. 5, the average of estimated K-value and IMP content in each of the fish samples were calculated. The estimated K-value and IMP content are shown in Figs. 6(c) and 6(d), respectively. As expected, the changes of K-value and IMP content obtained from chemical analysis and estimated from fluorescence images showed the same trend and as same as the previous studies [11, 31, 32], ATP inside the muscle is decomposed rapidly. After the decomposition of ATP, ADP and AMP are also decomposed quickly and advance to IMP in a stroke, occur instantaneous accumulation of IMP. Furthermore, IMP thus formed is slowly converted to HxR and then to Hx, which means the slowly increasing of K-value. Note that, the change in both K-value and IMP content differ for each part of the fish. The dorsal part, which was used for chemical analysis, showed the highest K-value compared to other parts. Therefore, the average of estimated K-value and IMP content change is slightly smoother than the measured value, which was point measurement.
This result suggested that the purposed method could accurately visualize the distribution of both K-value and IMP content using only one excitation light and be practical for large-scale commercial purposes with a vast number of fish or large fish such as tuna.
3.2. Experiment 2
Figure 7 shows the K-value distributions of the neck break sample. There were two samples for each storage time condition, which correspond to two rows in figure. As expected, the changing of K-value in the struggle samples is different compared to the neck break samples (instantly killed). Concretely, the K-value of the struggle samples is much higher for meat around the backbone compared to the neck break samples. In general, while the fish is struggling, muscles around the backbone move more intensely. For this reason, meat around the backbone tends to be burnt meat and the K-value is considered to be increased faster .
Figure 8(a) shows the average of K-value in area around the backbone and remaining area. As a result, from the average K-value of the area around the backbone and the average K-value of the remaining area, Fig. 8(b) shows the zoom in on the neck break samples and the struggle samples, which have almost the same K-value. As expected, the neck break samples and the struggle samples can be divided into two groups. It is suggested that burnt meat could be detected even in frozen condition by visualizing the distributions of K-value. As the freshness condition of frozen fish meat cannot be tracked without thawing the sample, this visualization method could be a promising tool to solve the problem.
In this study, we presented a novel approach for evaluating frozen fish quality and freshness by combining the fluorescence EEM with an imaging technique. Visualization was successfully performed to reveal the spatial-temporal changes of frozen fish quality such as K-value and IMP content with accuracy of R2 = 0.78 and R2 = 0.82, respectively using only one excitation light. The results also showed that the changing trends of both K-value and IMP content in fish differ between each part. Besides, this new method was applied to differentiate burnt meat and suggested that burnt meat could be detected even in a frozen condition. The proposed method offers a more practical way for large-scale commercial purposes. Further development will be needed to visualize the distribution of fish quality from the skin side, which will be a more pragmatic approach in actual use.
Leading Graduate School Program R03 of MEXT; Science and Technology Research Promotion Program for Agriculture, Forestry, Fisheries and Food Industry provided by Ministry of Agriculture, Forestry and Fisheries, Japan.
The authors are grateful to Prof. Toshimichi Maeda, National Fisheries University and staffs of Iwate University Sanriku Reconstruction Regional Creation Promotion Organization Kamaishi Satellite for assistance with the samples preparation. We also thankful to the members of Tokyo University of Marine Science and Technology, Food Processing Laboratory for helping us with the EEM and chemical measurement.
1. L. M. Nollet and F. Toldrá, Handbook of Seafood and Seafood Products Analysis (Chemical Rubber Company, 2009). [CrossRef]
2. G. Olafsdottir, E. Martinsdóttir, J. Oehlenschläger, P. Dalgaard, B. Jensen, I. Undeland, I. Mackie, G. Henehan, J. Nielsen, and H. Nilsen, “Methods to evaluate fish freshness in research and industry,” Trends Food Sci. & Technol. 8, 258–265 (1997). [CrossRef]
3. X. Huang, J. Xin, and J. Zhao, “A novel technique for rapid evaluation of fish freshness using colorimetric sensor array,” J. Food Eng. 105, 632–637 (2011). [CrossRef]
4. L. F. Jacober and A. Rand, “Biochemical evaluation of seafood,” Chem. Biochem. Mar. Food Prod. 1982347–365 (1982).
5. M. E. Surette, T. A. Gill, and P. J. LeBlanc, “Biochemical basis of postmortem nucleotide catabolism in cod (Gadus morhua) and its relationship to spoilage,” J. Agric. Food Chem. 36, 19–22 (1988). [CrossRef]
6. T. A. Gill, “Objective analysis of seafood quality,” Food Rev. Int. 6, 681–714 (1990). [CrossRef]
7. D. Greene, J. Babbitt, and K. Reppond, “Patterns of nucleotide catabolism as freshness indicators in flatfish from the gulf of alaska,” J. Food Sci. 55, 1236–1238 (1990). [CrossRef]
8. T. Haitula, M. Kiesvaara, and M. Moran, “Freshness evaluation in European whitefish (Coregonus wartmanni) during chill storage,” J. Food Sci. 58, 1212–1215 (1993). [CrossRef]
9. C. Handumrongkul and J. Silva, “Aerobic counts, color and adenine nucleotide changes in CO2 packed refrigerated striped bass strips,” J. Food Sci. 59, 67–69 (1994). [CrossRef]
10. H. Hong, J. M. Regenstein, and Y. Luo, “The importance of ATP-related compounds for the freshness and flavor of post-mortem fish and shellfish muscle: A review,” Critical Rev. Food Sci. Nutr. 57, 1787–1798 (2017).
11. T. Saito, “A new method for estimating the freshness of fish,” Bull. Jpn. Soc. for Sci. Fish 24, 749–750 (1959). [CrossRef]
13. I. M. Warner, G. D. Christian, E. R. Davidson, and J. B. Callis, “Analysis of multicomponent fluorescence data,” Anal. Chem. 49, 564–573 (1977). [CrossRef]
14. M. J. Sorrell, J. Tribble, L. Reinisch, J. A. Werkhaven, and R. H. Ossoff, “Bacteria identification of otitis media with fluorescence spectroscopy,” Lasers Surg. Medicine 14, 155–163 (1994). [CrossRef]
15. G. Wolf, J. S. Almeida, C. Pinheiro, V. Correia, C. Rodrigues, M. A. Reis, and J. G. Crespo, “Two-dimensional fluorometry coupled with artificial neural networks: A novel method for on-line monitoring of complex biological processes,” Biotechnol. Bioeng. 72, 297–306 (2001). [CrossRef] [PubMed]
16. H. Sterenborg, M. Motamedi, R. Wagner, M. Duvic, S. Thomsen, and S. Jacques, “In vivo fluorescence spectroscopy and imaging of human skin tumours,” Lasers Med. Sci. 9, 191–201 (1994). [CrossRef]
18. G. ElMasry, H. Nagai, K. Moria, N. Nakazawa, M. Tsuta, J. Sugiyama, E. Okazaki, and S. Nakauchi, “Freshness estimation of intact frozen fish using fluorescence spectroscopy and chemometrics of excitation–emission matrix,” Talanta 143, 145–156 (2015). [CrossRef] [PubMed]
19. G. ElMasry, N. Nakazawa, E. Okazaki, and S. Nakauchi, “Non-invasive sensing of freshness indices of frozen fish and fillets using pretreated excitation–emission matrices,” Sensors Actuators B: Chem. 228, 237–250 (2016). [CrossRef]
20. M. Shibata, G. ElMasry, K. Moriya, M. M. Rahman, Y. Miyamoto, K. Ito, N. Nakazawa, S. Nakauchi, and E. Okazaki, “Smart technique for accurate monitoring of ATP content in frozen fish fillets using fluorescence fingerprint,” LWT-Food Sci. Technol. 92, 258–264 (2018). [CrossRef]
21. M. Ando, M. Joka, S. Mochizuki, K.-I. Satoh, Y. Tsukamasa, and Y. Makinodan, “Influence of death struggle on the structural changes in chub mackerel muscle during chilled storage,” Fish. Sci. 67, 744–751 (2001). [CrossRef]
22. C. Watson, R. Bourke, and R. W. Brill, “A comprehensive theory on the etiology of burnt tuna,” Fish. Bull. 86, 367–372 (1988).
23. S. Ehira, K. Saito, and H. Uchiyama, “Accurracy of measuring K value, an index for estimating freshness of fish, by freshness testing paper,” Bull. Tokai Reg. Fish. Res. Lab. (Japan) (1986).
24. T. Maeda, A. Yuki, H. Sakurai, K. Watanabe, N. Itoh, E. Inui, K. Seike, Y. Mizukami, Y. Fukuda, and K. Harada, “Alcohol brine freezing of japanese horse mackerel (Trachurus japonicus) for raw consumption,” Transactions Jpn. Soc. Refrig. Air Cond. Eng. 24, 323–330 (2007).
25. H. Nasibov, A. Kholmatov, B. Akselli, A. Nasibov, and S. Baytaroglu, “Performance analysis of the CCD pixel binning option in particle-image velocimetry measurements,” IEEE/ASME Transactions on Mechatronics 15, 527–540 (2010). [CrossRef]
28. B. Valeur and J.-C. Brochon, New Trends in Fluorescence Spectroscopy: Applications to Chemical and Life Sciences, vol. 1 (Springer Science & Business Media, 2012).
29. A. A. Rinnan and C. M. Andersen, “Handling of first-order Rayleigh scatter in PARAFAC modelling of fluorescence excitation–emission data,” Chemom. Intell. Lab. Syst. 76, 91–99 (2005). [CrossRef]
30. C. M. Andersen and R. Bro, “Practical aspects of PARAFAC modeling of fluorescence excitation-emission data,” J. Chemom. 17, 200–215 (2003). [CrossRef]
31. K. Fijisawa and M. Yoshino, “Activities of adenylate-degrading enzymes in muscles from vertebrates and invertebrates,” Comp. biochemistry physiology. B, Comp. biochemistry 86, 109–112 (1987). [CrossRef]
32. N. Jones and J. Murray, “Rapid measures of nucleotide dephosphorylation in iced fish muscle. their value as indices of freshness and of inosine 5′-monophosphate concentration,” J. Sci. Food Agric. 15, 684–690 (1964). [CrossRef]
33. Y. Kodani, A. Kato, M. Honda, and Y. Ishihara, “Studies on the “Yakeniku” of Thunnus thynnus unloaded at the Sakaiminato fishing port,” Tottori Inst. Ind. Technol. Res. Rep. 20101–10 (2010).