Assessment of fish (cod) freshness by VIS/NIR spectroscopy

F. Sigernes, M. Esaiassen, K. Heia, G. Eilertsen,

J.P. Wold and N.K. Sørensen

Norwegian Institute of Fisheries and Aquaculture,

N-9005 Tromsø, Norway


The use of near-infrared spectroscopy has previously been proven promising for assessing aspects of fish quality, e.g. fat, water, protein and salt contents. Initially, transmission measurements in the wavelength range 860 to 920 nm have been conducted on samples of cod muscle. A correlation between attenuated light and storage time of cod fillets on ice were found. New instrumentation using diffuse reflectance now allows measurements to be done in a non-destructive way on whole fillets . By recording spectra from 5 skinless fillets daily over a period of 14 days in the visible and near-infrared wavelength region, a model for predicting stored time was constructed using Partial Least Square Regression. The correlation coefficient between measured and predicted time was 0.95, and the standard deviation was found to be close to 28 hrs.



During the last decade near-infrared (NIR) spectroscopy has become extensively used in the agricultural and food processing industries for measuring protein, moisture, and starch in grain and forage crops. The use of NIR spectroscopy have provided a rapid and sensitive method for analysis of organic materials with little or no sample preparation /1,2/.

The method is based on the simple fact that organic molecules absorb light. The absorption bands are the results of overtones or combinations of overtones originating in the fundamental mid-range infrared region of the spectrum. Especially, overtones due to hydrogenic stretching vibrations or combinations involving stretching and bending modes of C-H, O-H, N-H or C=O groups are common /1,2/. As a result, NIR spectra are very complex, and any peak of interest is almost surely overlapped by one or more interfering peaks. The bandwidths approaching the visible part of the spectrum may be up to 100-150 nm wide. In order to handle this complexity, multivariate data analysis is used to calibrate the spectra and to quickly identify, test quality, or quantify constituent levels in samples.

Despite that NIR spectroscopy has gained a foothold as a quantitative method in food analysis, relatively little is known concerning the applicability to fish freshness. And to our knowledge, assessment of fish freshness by NIR measurements has never been performed. However, both NIR transmittance and transflectance have shown to be well suited for simultaneously analysis of specific quality parameters like fat and moisture in salmon /3,4,5/. Anyway, it is well known that fish freshness is no distinct parameter. It is a complex interplay of several parameters where each one to some extent may explain freshness /6/. Due to this complexity and the fact that a NIR measurement actually represents a spectral footprint of a whole range of processes, we wanted to investigate if it could be used as a tool to assess fish freshness. And, since the reactions that occurs in a fish post mortem is dependent on time and temperature, our first task was to search for a correlation between the spectral changes and storage time on ice.



3.1 Instrumentation

In this study two types of instruments are used. NIR transmission spectra were collected by the well known and widely used Infratec 1255-Food & Feed Analyzer. It consists basically of a light source, a monochromator, a sample container, and a detector. The detector measures light emanating from the sample (cod) with the light source and the monochromator located on the opposite side. In other words, the monochromatic light passing through the sample is recorded. This straight line approach is the most common in spectroscopic measurements. Figure 1 shows a schematic of the main parts of the instrument.

Fig. 1 - Reproduction of the schematic of the Infratec 1255-Food & Feed Analyzer from the instrument manual of the Perstorp Analytical Company. The monochromator (1) contains a 50 W Tungsten lamp and a diffraction grating. As the grating turns, monochromatic light passes through the sample in the sample cup (2) and reaches the detector (3).

If the samples measured by this technique turn out to be cloudy, opaque, or just to highly absorbing, another approach called diffuse reflectance may prove useful. In this mode, the incident monochromatic beam of light strikes the sample perpendicular to the surface. The light penetrate the sample and is diffusely reflected in all directions. It is the diffuse back-scattered light of the sample we detect. The detector and light source is then on the same side of the sample. NIR diffuse reflectance spectra were obtained by the NIRSystem 6500 monochromator, using a fiber probe as front optics. The probe itself consists of splitted fibers grouped to receive or transmit light. Figure 2 shows the schematics of the instrument together with the experimental setup using the fiber probe.


Fig. 2 - The left panel shows a simple schematic of the NIRSystem 6500 monochromator. The fiber probe and the experimental setup is shown in the right panel. In the left panel (A) is turnable grating, (B) order sorting filters, (C) flat mirror, (D) tungsten halogen lamp at entrance slit (with power supply and regulator board), and (E) exit slit. In the right panel (F) is fiber probe, (G) teflon plate, (I) opal glass plate, and (H) skinless fish fillet. The measurements were carried out at points labeled (1), (2), (3), (4) and (5) for both sides of the fillets.

The ratio between the detected sample intensity I and the incident intensity J is in our case named Reflectance and is represented as

Reflectance(l)=R(l)=I(l)/J(l), (1)

where l is the wavelength in nm. Absorbance is then related to reflectance by the following equation

Absorbance(l) = A(l)=log[1/R(l)] . (2)

The logarithm used in eq. (2) is related to the Beer / Bougier / Lambert Law. It state that high absorbance means low reflectance or, low diffuse back-scattered light. The incident intensity or the reference spectrum is obtained by holding a piece of white diffuse plate of teflon in front of the probe instead of the actual sample. It is taken before and after each measured cycle to make sure that the lamp has been stable during operation. The lamp is temperature stabilized.

Note that if we use similar logic, R(l) can be replaced in eq. (1) with the term Transmittance, T(l). Eq. (2) then becomes A(l)=log[1/T(l)]. The output from both the Infratec 1255-Food & Feed Analyzer and the NIRSystem 6500 are recorded in terms of Absorbance.

3.2 Sample preparation

In the transmission experiment 9 cod were gutted and stored on ice. Measurements were conducted on day number 1, 4, 7 and 11. Samples of intact tissue shaped as cylinders, 23 mm in diameter and 10 mm high, were stamped out from the loin of the cods. One sample per fish was used per day. A total of 36 spectra were recorded.

In the diffuse reflectance experiment 5 cod were slaughtered and filleted. One skinless fillet per fish was selected and put on ice. First day of measurements was 24 hours after death. 5 spectral samples on both sides of the fillet were recorded. The measured points were selected according to figure 2 along the lateral line of the loin part. The fillets were then put back to storage on ice. The above procedure was repeated daily over a period of 14 days. A total of 350 spectra were recorded.



Figure 3 shows the results of the transmission measurements. The transmittance of the cod samples was decreasing with increasing number of days on ice. The relative high values of the absorbance may explain why the error bars are as high as 2 days. The intensity of the light reaching the detector is only 10 to 20% of the incident intensity. As mentioned above, if the samples are absorbing too much light, then a diffuse reflectance setup may prove more useful. However, the measurements motivated us to believe that there is a connection between NIR spectra and fish freshness, assessed as day on ice.

Fig. 3 - The mean attenuation at the wavelengths 860, 890, 920 and 940 nm as a function of days on ice is given. These transmission measurements were carried out with the Infratec 1255-Food & Feed Analyzer.

During the diffuse reflectance or the transflectance experiment the sample intensities reaching the detector were much higher. Each stored spectrum is an average of 32 spectra. The instrument provides auto bias and gain to obtain a maximum signal to noise ratio. It is also frequently checked to evaluate performance and stability.

First of all, it was easy to see the difference between dark- and white- fish muscle in the diffuse reflectance spectra. Most of the dark muscle is located on the skin-side of the fillet (skin is removed), especially towards the tail. The in-side of the fillet is more or less pure white. Figure 4 shows a typical data set from a fillet on day 3. The dark muscle has a high absorbance in the visible compared to the white muscle. The white muscle scatter visible light better than the dark muscle. In the near infrared region the two muscle types follow the same trend. The diffusion becomes smaller with increasing wavelength.

Fig. 4 - Typical diffuse reflectance spectra (absorbance) from cod-fillet on day 3. The curve marked (1) is from the skin-side of the fillet measured at point (1) according to figure 2. The curve marked (2) is obtained from the in-side of the fillet at point (5).

During the experiment, it was noticed that the back-scattered light from the in-side fillet close to the probe was kind of orange/yellow. This light had no problem penetrating the fillet. If we turned the fillet over to the skin-side the light turned more weak and reddish. This eye observation is consistent with figure 4's minimum in-side absorbance above the blue part of the spectrum and the more red shifted minimum at the skin-side of the fillets.

The most characteristic or dominant spectral change over time was seen from the skin-side tail region of the fillets. The probe’s field of view covered mostly the dark muscle in this region. The samples showed a loss of spectral structure with increasing time. The dip in the spectra at approximately 500 nm became less and less dominant. The peak close to 550 nm became wider and less dominant with increasing time. Note, that on the skin-side of the fillet, the tail was darkest and as we moved up it became more white. The in-side of the fillets consisted of white muscle only, and it was hard to see any change in spectral characteristics due to time from this region.

Principal Component Analysis (PCA) and Partial Least Square Regression (PLSR) have been applied to this data set to reduce any trends in time. One of the main goals was to evaluate where to sample the fillets to obtain a optimized correlation with storage time. The results showed no significant difference between the correlation obtained at the skin-side compared to the in-side of the fillets. Figure 5 shows the results obtained from the in-side of the fillets at point (5) according to figure 2.

Fig. 5 - Correlation between predicted versus measured storage time on ice assessed from the diffuse reflectance spectra of the thick in-side of the cod-fillets. This plot is generated by the software Unscrambler version 6.1 Full cross validation is used. The residual validation varians decreased monotonously from 15 to 1 with 5 principal components. The data set is composed of spectra in the wavelength region 400-1100 nm.

According to the above model, storage time of cod-fillets on ice can be predicted with an error less than 30 hours. Nevertheless, further investigations are underway with a larger amount of samples (fish) in order to reduce individual variation errors. The measurements will also be supported with chemical analyses. As a start point, these measurements state that VIS/NIR spectroscopy may serve a tool to assess freshness of cod, measured as days on ice.


Osborne B.G., Fearn T. Near Infrared Spectroscopy in Food Analysis. Longman Scientific & Technical, Essex, England, (1988), 200p.

Hildrum K.I., Isaksson T., Naes T., Tandberg A. Near Infra-red Spectroscopy. Ellis Horwood Ltd., West Sussex, England, (1992), 473p.

Sollid H., Solberg C. Salmon fat-content estimation by near-infrared transmission spectroscopy. J. Food Science, 57, (1992), 792-793.

Wold J.P., Jakobsen T., Krane L. Atlantic salmon average fat content estimated by near-infrared transmittance spectroscopy. J. Food Science, 61, (1996), 74-77.

Wold J.P., Isaksson T. Non-destructive determination of fat and moisture in whole Atlantic salmon by near-infrared diffuse spectroscopy. J. Food Science, 62, (1997), 734-736.

Olavsdottir G., Martinsdottir E., Oehlenschläger J., Dalgaard P., Jensen B., Undeland I., Mackie I.M., Henehan G., Nielsen J., Nilsen H. Methods to evaluate fish freshness in research and industry. Trends in Food Science & Technology, 8, (1997), 258-265.