Advertisement

Analytical and Bioanalytical Chemistry

, Volume 410, Issue 22, pp 5663–5673 | Cite as

Through-packaging analysis of butter adulteration using line-scan spatially offset Raman spectroscopy

  • Santosh Lohumi
  • Hoonsoo Lee
  • Moon S. Kim
  • Jianwei Qin
  • Byoung-Kwan Cho
Research Paper
Part of the following topical collections:
  1. Food Safety Analysis

Abstract

Spectroscopic techniques for food quality analysis are limited to surface inspections and are highly affected by the superficial layers (skin or packaging material) of the food samples. The ability of spatially offset Raman spectroscopy (SORS) to obtain chemical information from below the surface of a sample makes it a promising candidate for the non-destructive analysis of the quality of packaged food. In the present study, we developed a line-scan SORS technique for obtaining the Raman spectra of packaged-food samples. This technique was used to quantify butter adulteration with margarine through two different types of packaging. Further, the significant commercial potential of the developed technique was demonstrated by its being able to discriminate between ten commercial varieties of butter and margarine whilst still in their original, unopened packaging. The results revealed that, while conventional backscattering Raman spectroscopy cannot penetrate the packaging, thus preventing its application to the quality analysis of packaged food, SORS analysis yielded excellent qualitative and quantitative analyses of butter samples. The partial least-square regression analysis predictive values for the SORS data exhibit correlation coefficient values of 0.95 and 0.92, associated with the prediction error 3.2 % and 3.9 % for cover-1 & 2, respectively. The developed system utilizes a laser line (ca. 14-cm wide) that enables the simultaneous collection of a large number of spectra from a sample. Thus, by averaging the spectra collected for a given sample, the signal-to-noise ratio of the final spectrum can be enhanced, which will then have a significant effect on the multivariate data analysis methods used for qualitative and/or qualitative analyses. This recently presented line-scan SORS technique could be applied to the development of high-throughput and real-time analysis techniques for determining the quality and authenticity various packaged agricultural products.

Keywords

Food safety Food authenticity Through-packaging analysis Raman imaging SORS 

Introduction

Foodstuffs have always been vulnerable to adulteration with cheaper, inferior substances, or chemicals, usually for economic gain. Therefore, the quality and safety testing of food requires the development of effective techniques and is of concern particularly for food-related regulatory organizations. Several techniques are commonly used for the detection of food adulteration and are critically reviewed in the literature [1, 2]. Raman spectroscopy has a long history of application to food science, given that its excellent chemical sensitivity and specificity to structural changes in molecules make it a promising tool for examining the quality and authenticity of food products [3]. Moreover, Raman spectroscopic techniques, along with imaging, have been successfully applied as a reliable analytical technique in a range of sectors spanning food, pharmaceuticals, biology, and materials science [4, 5].

Conventional backscattering Raman spectroscopy is suitable for surface or minimal-subsurface analysis; however, the analysis of greater depths within a material is limited by the interference of the fluorescence signals emitted from the surface [6]. Moreover, although the transmission geometry of Raman spectroscopy can be exploited to overcome the surface layer fluorescence when attempting to analyze internal material, the mixed Raman information retrieved from the internal layers cannot be resolved. However, a new means of deep-probing such highly turbid layered food and biological samples has emerged with the advent of spatially offset Raman spectroscopy (SORS) [7], permitting the deep non-invasive analysis of layered samples and the subsequent recovery of pure Raman signatures of individual layers by numeric processing [8]. The concept of SORS was pioneered by Matousek et al. [7] and has become an established technique for deep analysis in a range of studies in different sectors such as the through-skin analysis of salmon [9], the evaluation of the internal maturity of tomatoes [10], the non-invasive detection of gelatin-encapsulated powders [11], medical diagnosis and disease monitoring [12], and the analysis of hidden painted images [13].

Although SORS is a depth-sensitive Raman spectroscopic technique and has proven effective in the detection of the depth-dependent Raman spectra of layered samples, current SORS measurement methods are based on fiber-optic probes and therefore have poor depth resolution [14]. The use of a fiber probe array generally relies on the fixed positions of the fibers, and thus, only the Raman spectra for a fixed offset distance can be calculated; hence, there is no instrumental flexibility. To overcome these limitations of SORS measurements, some groups have used a point laser as an illumination source and have collected the SORS spectra by offsetting the point by a given distance from the Raman probe. These customized SORS systems provide instrumental flexibility and allow the user to select a range of offsets depending upon the sample of interest. Recently, Qin et al. (2011) used such a customized SORS system to investigate the maturity level of tomatoes [10, 15]. The same system was further adopted for the detection of gelatin-encapsulated powders [11]. Although the developed system provides instrumental flexibility, the use of a point laser incurs a long spectral acquisition time for each point. Therefore, the sampling time remains too long for those samples with large surface areas. Another limitation of the point-scan-based SORS technique stems from its low signal-to-noise (S/N) ratio of the collected Raman offset spectra. Therefore, multiple accumulations are usually collected with a higher exposure time and then averaged to compensate for the S/N ratio. Thus, they ultimately require a long data collection time to obtain a reasonable S/N ratio for the final (averaged) spectrum.

Moreover, despite the current advances in the SORS technique, very few applications have been found within the field of food analysis. Nowadays, because the packaged food market is growing rapidly in response to consumer demand, the quality analysis of packaged food has become important for both the food industry and regulatory organizations. This can only be achieved by developing a technique that can see through any superficial layer (of packaging) and which can retrieve the signals from the food inside. Therefore, one potential and interesting application of SORS in the food sector could be the analysis of packaged food quality such as the detection of packaged food contamination and the authenticity analysis of (adulteration-prone) packaged high-value dairy products (i.e., butter and cheese).

In the dairy products sector, butter is a rich source of fat-soluble vitamins and therefore attracts a relatively high price in comparison with other dairy products. Therefore, it is prone to being adulterated with vegetable fat (margarine) or with other animal fats for economic profit [16]. Moreover, given the price difference and compositional similarity, the adulteration of butter with margarine continues to be a threat to consumers, particularly in developing countries [17, 18]. The potential of Raman spectroscopy as a means of detecting the adulteration of butter with margarine was recently investigated [17]. In that study, conventional backscattering Raman spectral data for margarine-adulterated butter samples were collected by exposing the adulterated samples to the laser points. However, in terms of the ability to simplify the control of food, and to evaluate the authenticity of packaged food, analytical techniques should be able to see through the different (plastic) packaging layers and retrieve physio-chemical information from the food inside. Therefore, considering the existing limitations of the SORS technique, and for advancing its application to food-quality analysis, in the present study we propose a macro-scale line-scan SORS measurement technique which is able to see through superficial layers and/or packaging. Unlike previous studies in which the point laser was offset, the present study used a full laser line (ca. 14-cm wide), offset with a narrow offset interval and broad offset range to generate multiple sets of Raman data with a single scan of different parts of the sample surface/sub-surfaces along the laser line. This facilitates the high-throughput SORS analysis of food and biological samples. This study paves the way for the detection and reconstruction of signals from packaged food through the packaging layer which can be further analyzed for quality- and authenticity-analysis purposes. The main objectives of the present study were to (1) develop a means of acquiring spatially offset Raman spectra using a macro-scale line-scan Raman spectroscopic system; (2) detect butter adulteration with margarine through two different types of packaging using the developed SORS system; (3) perform a chemometric analysis for quantitative detection utilizing the partial least-squares regression (PLSR) analysis model; and (4) perform through-container measurement and discrimination between different commercial butter and margarine samples.

Materials and methods

Line-scan SORS system

The developed line-scan SORS system utilizes a 785-nm laser line, an optical assembly to expand the laser-line and ensure laser uniformity on to the sample surface, and a sensing module to collect Raman signals and generate SORS data. A NIR laser system (OptiGrate Corp., USA) was used to generate high-energy laser light combined from 19 diode emitters. The generated laser line was filtered using a 785-nm band pass filter and a laser beam polarizer (CM1-PBS 252; ThorLabs, Hans-Boecker, Germany). A polarizer is a kind of optical fiber in which the light transmission depends on the polarization state. Normally, light with linear polarization in a certain direction is passed, while the light polarized in the orthogonal direction is blocked [19]. A cylindrical lens (f = 200 mm) was used to expend the laser line. The resulting laser beam was passed through an engineered diffuser (ED1-L4100; ThorLabs, Germany) to obtain a homogeneous laser line. The shaped laser line was projected onto a 785-nm dichromic beam splitter (Semrock, USA) mounted at 35° to project the laser beam on to the sample surface. In addition, the beamsplitter was mounted on an XYZ translation stage, which was manually controlled to move the beam splitter back and forth. The laser power applied to the sample surface was approximately 450 mW/25 mm, as measured by a digital power meter (PM100D, ThorLabs, Germany). A schematic representation of the developed SORS system is shown in Fig. 1a and a photograph of the system is shown in Fig. 1b. To generate Raman spectra over a broad offset range, the entire laser line was offset by different distances from the imaging slit by manually moving the beam splitter. The XYZ translation stage for the beam splitter allows the user to easily change the range and interval of the offset for different applications. One advantage of this setup is that SORS signals from a large number of points along the laser line can be collected simultaneously, rather than point-by-point signal collection as in the case of point-scan SORS. The SORS signals produced by the sample were collected by a detection module consisting of an objective lens with a focal length of 23 mm, two Rayleigh filters to remove the effect of Rayleigh scattering, and an imaging spectrograph to disperse the incoming signals onto the CCD detector, thus creating a 2D spectral–spatial image.
Fig. 1

Schematic (a) and photograph (b) of line-scan SORS system for acquiring Raman data from different sample depths

Experimental samples and SORS data collection

Commercial butter and margarine samples were purchased from a supermarket. To demonstrate the potential of the developed SORS system, three sample sets were prepared. First, to select the optimal offset distance for effectively seeing through the packaging cover, pure butter and margarine samples were prepared and packed into an aluminum sample holder (interior dimensions 10 mm × 20 mm × 6 mm). Further, two different plastic sheets (1-mm and 0.8-mm thick), cut from the two different original containers, were placed on top of the aluminum sample holder to completely cover the butter and margarine samples for SORS measurements. A second type of sample was prepared to prove the effectiveness of the SORS technique for the through-packaging quantitative analysis of butter adulterated with margarine. To this end, both butter and margarine were weighed in several proportions (from 0 to 50% w/w margarine at 5% intervals), and the mixtures were subjected to oven heating at 50 °C for 1 h. The samples were then transferred to glass beakers, and finally mixed well by subjecting them to vortex mixing.

The prepared mixtures were stored at 2 °C so that they would attain their commercial solid form. The solid butter samples were packed into an aluminum sample holder. Note that sample preparation was required only to develop a calibration model. Commercial samples could be directly analyzed in their packaging using the proposed technique. To mimic the real-world conditions presented by packaged commercial butter, a piece of plastic was cut from each of the original containers and placed over the sample holder to completely cover the sample (as shown in Fig. 2), prior to starting the SORS measurements at a selected (optimal) offset distance.
Fig. 2

Detection of butter adulteration through packaging: a butter packed into sample holder; b, c two different packaging plastics cut from the original container, placed on the top of the samples

Finally, to demonstrate the potential of our SORS system for through-container measurement and discrimination, different commercial butter and margarine samples were purchased from supermarkets and scanned with the SORS system while still in their unopened packaging. The purchased butter and margarine samples used in the present study were packaged in plastic containers with a thin covering layer of aluminum complex. Raman data could not be collected through the aluminum complex because it would reflect most of the laser light back. Therefore, the SORS data were collected from the other side of the packaged samples, where the aluminum-complex packaging would not interfere with the measurements.

The developed line-scan SORS system, which covers a spectral range of 740–1010 nm, corresponding to Raman shifts of − 763–2837 cm−1 with an average spectral resolution of 3.5 cm−1, was combined with a height-adjustable stage to position the samples for the SORS measurements. The sample-to-beam splitter distance was set to 16 cm while the lens-to-sample distance was 23 cm. With this setting, a uniform laser-line of ca. 140 mm could be formed, and the length of the instantaneous field of view (IFOV) of the Raman system was determined to be 140 mm. Since the CCD has an area array detector of 1024 × 1024 pixels, the maximum spatial resolution along the scanning line can be estimated to be 140 mm/1024 pixels = 0.14 mm/pixel. First, to select the optimal offset distance for cover 1 (Fig. 2b), both butter and margarine were separately covered with cover 1 and the Raman spectra were obtained at a 0-mm offset (which is equal to conventional backscattering geometry). SORS data were collected with an offset range of 2–10 mm with an offset interval of 2 mm. A single scan for each offset range was acquired with an exposure time of 8 s. The same methodology was also used to select the optimal offset for cover 2 (Fig. 2c).

For the quantitative analysis of butter adulteration, SORS data for the covered pure and adulterated butter samples were collected using the optimal offset distance with the above instrumentation settings. The mixture samples were packed into a sample holder and seven scans at intervals of 0.5 mm for each mixture and each packaging cover were acquired. To compare the performances of the conventional and SORS geometries of Raman spectroscopy for butter authenticity analysis through the packaging, the conventional Raman spectrum (0-mm offset) of each mixture sample was also collected using the same instrumental settings, except for the offset of the laser line.

The third group of samples, consisting of different varieties of butter and margarine, was scanned by the SORS system after being placed on a conveyor unit at a set distance from the camera lens. A total of 10 scans for each variety (sample) were collected with an offset distance of 5 mm and an exposure time of 8 s. A dark current image was collected with the laser off and the camera lens covered with its cap. This image was subtracted from the original sample image. Subsequently, only the corrected data were used for further data analysis.

Raman data analysis

The spatial profile of each sample was visualized by selecting a band image, and spectral data were then extracted from the region of interest. The major signals corresponding to the butter and margarine used in this study were observed mainly in the range of 800–1800 cm−1. Therefore, the data size was further reduced by retaining only the informative spectral range. The baseline drift in the Raman spectra caused by autofluorescence always blurs or even swamps the signals and adversely affects the analytical results, particularly in multivariate data analysis [20]. Since fluorescence signals are usually generated during laser-sample interaction in Raman measurement, the selected spectral data were fluorescence-corrected using the adaptive iteratively reweighted penalized least squares (airPLS) method [20]. For pure butter and margarine, SORS data were collected for an optimal offset, after which 10 pixels were selected for each scan and the airPLS method was applied to the spectrum of each pixel. Thus, the fitted baseline was calculated and subtracted from the original Raman spectra to obtain fluorescence-free Raman data. Finally, the fluorescence-corrected Raman spectra for 10 pixels were averaged for each offset measurement.

As seven scans with an increment of 0.5 mm were acquired using the optimal offset distance for mixture samples, 15 pixels for each scan were averaged and regarded as being an individual sample. Hence, 77 samples (7 samples × 11 groups) were arranged in a matrix and were then subjected to fluorescence correction using the airPLS method. A multivariate data analysis method of partial least squares regression (PLSR) was used to establish a quantitative model between the fluorescence-corrected Raman spectral data and the added value of the margarine. For this purpose, a Y-vector of the reference values of the actual adulteration percentage was given to the spectral data matrix. To illustrate the proficiency of the PLSR model and the developed algorithm, the spectral data set was divided into a calibration set consisting of 42 samples and a validation set consisting of 35 samples. A summary of the descriptive statistics for the calibration and validation data sets is given in Table 1. For the third group of data, principal component analysis (PCA) was carried out on preprocessed SORS data to identify the natural discrimination between the different groups of samples. All the chemometric analyses and computations were executed using MATLAB software (The Mathworks, USA).
Table 1

Summary of descriptive statistics for the calibration and validation data sets (excluding number of samples, all units in %, Std: standard deviation)

 

Number of samples

Minimum

Maximum

Range

Mean ± Std

Calibration set

42

0

50

50

25 ± 17.28

Validation set

35

5

45

40

25 ± 14.35

Results and discussion

SORS measurements for optimal offset selection

The SORS technique depends on variables such as the layer depth, sample absorption, and scattering parameters [9]. As has been stated previously, the number of Raman-scattered photons reaching the detector decreases with an increase in the offset [21], thus introducing more noise to the Raman spectra, consequently decreasing the signal-to-noise (S/N) ratio. The S/N ratio is very important in multivariate data analysis; as the S/N ratio increases, the model performance improves. Therefore, the optimal offset distances at which the Raman signals for butter and margarine can be recovered effectively through the packaging covers 1 and 2 were determined.

Butter and margarine samples with both covers 1 and 2 were measured at different spatial offsets, including a 0-mm offset that is equal to the conventional backscattering geometry of Raman spectroscopy. A set of spatially offset Raman spectra acquired by placing covers 1 and 2 on the butter samples is shown in Fig. 3a and b. The same procedure was repeated for margarine and the resulting spectra are shown in Fig. 3c and d. The pure spectra of both covers and those of butter and margarine were plotted with the offset spectra for comparison. The most intense Raman peak for butter and margarine was found at 1442 cm−1. However, both covers produced high-intensity peaks near this region: the high-intensity peak for cover 1 appeared at 1459 cm−1 and that for cover 2 appeared at around 1431 cm−1. For cover 1, for a 0-mm offset, the dominating features of the spectrum were the broad spectral features related to the packaging cover. As the offset distance increases, the relative intensities of the 1459-cm−1 peaks gradually diminished, whereas those of the 1442-cm−1 peaks were gradually augmented. Moreover, a significant reduction in the minor peaks of packaging cover 1 with an increase in the offset distance can be seen in Fig. 3a and c. At the same time, the minor peaks of either butter or margarine become dominant. An extensive visual analysis of Fig. 3a and c shows that the Raman information from the top layer of the cover and the bottom layer of butter/margarine are mixed in the spatially offset spectra collected at offset distances of up to 4 mm. However, starting at an offset of 6 mm, a greater spectral contribution from the bottom layer of the butter/margarine can be seen, as the Raman peak of the butter/margarine at 1442 cm−1 has a higher intensity than the peak at 1459 cm−1, which is related to the cover. Moreover, the minor interfering peaks of cover 1 almost disappeared.
Fig. 3

Spatially offset Raman spectra of the cover 1 on butter (a), cover 2 on butter (b), cover 1 on margarine (c), and cover 2 on margarine samples (d). Spectra are normalized and separated along the intensity axis for clarity (spatial offset range from 0 mm (standard Raman backscattering) to 10 mm)

In Fig. 3b and d, a corresponding measurement series obtained through packaging cover 2 is provided. The spectral features of cover 2 are different from those of cover 1, and given that cover 2 is thinner, the laser beam is expected to be able to penetrate the covering surface. Therefore, the spectral features of butter/margarine along with the spectral features of cover 2 can be seen, even at a 0-mm offset, whereas the spectral interference of cover 1 almost disappears once the offset exceeds 2 mm.

Calculation of the base materials’ (butter and margarine) S/N ratio involved the background noise value as determined for the spectral range of 1550–1600 cm−1, which is free from the Raman signals of both the surface (covers) and subsurface materials. Based on the results obtained for the signal intensity and S/N ratio, the optimal offset distances for covers 1 and 2 were set to 6 mm and 4 mm, respectively. The larger offset distance for cover 1 was selected given its thickness and strong Raman features. Therefore, based on the results obtained in the optimal offset selection study, an offset of 6 mm for cover 1 and 4 mm for cover 2 were used for further quantitative measurements.

The efficacy of the SORS technique was proven through a series of experiments in which the Raman spectra of covered butter and margarine were acquired with conventional backscattering and optimal offset. A summary of the results is shown in Fig. 4. Figure 4a shows the actual Raman signature for butter and margarine. A significant difference in the peak intensities and a small variation in spectral pattern are evident from Fig. 4a. The highest intensity peaks at 1442 cm−1 and 1300 cm−1 arise from the δ (C–H) scissoring vibration and the C–H twisting vibration of the –CH2 group, respectively. The peak observed at 1268 cm−1 is assigned to the δ (C–H) bending vibration at the cis double bond in R–HC=CH–R, and the two bands at 1083 and 1063 cm−1 originate from the (C–C) stretching vibration. At the end of the spectra, margarine exhibits comparatively higher peaks than butter at 1667 cm−1 originating from the (C=C) cis double-bond stretching vibration of RHC=CHR [17, 22, 23].
Fig. 4

Preprocessed Raman spectra for butter and margarine acquired with conventional backscattering geometry and SORS spectra obtained at optimal offsets

Further investigations were undertaken to determine whether the conventional backscattering geometry of Raman spectroscopy can be used for the through-packaging analysis of butter adulteration. Unlike Fig. 4a–c shows no significant difference between the Raman spectra of butter and margarine. The Raman data collected with cover 1 seems highly overweighted from the superficial layer of the cover; however, small differences can be seen throughout the spectra in the regions where there are spectral differences between butter and margarine. In addition, the conventional Raman spectra of butter and margarine collected with cover 2 shows some peaks related to the base materials; however, no notable difference can be observed (Fig. 4c). This result confirms that conventional backscattering Raman spectroscopy cannot be used effectively for the authentication of commercially packaged butter samples. On the other hand, when the same samples were measured using the SORS technique at the optimal offset, a significant difference between the Raman spectra of butter and margarine can be seen, as shown in Fig. 4d and e. These spectral differences are similar to those shown in Fig. 4a. Nevertheless, the SORS data at the optimal offset distance were still slightly influenced by the surface of the packaging covers. The spectral differences, however, are still abundant to separate the packaged butter and margarine samples, which is not possible with the conventional geometry of Raman spectroscopy.

Quantitative analysis

An important issue related to butter authenticity analysis is the quantitative analysis of adulterants in commercially packaged butter samples, which can be considered as a truly non-destructive analysis method. Therefore, based on the results of optimal offset selection, offsets of 6 mm and 4 mm were selected for cover 1 and cover 2, respectively, and all the quantitative measurements were conducted using the optimal offsets. To prove the efficacy and superiority of the developed SORS system, the same samples were also scanned with conventional backscattering geometry (0-mm offset) and the results were compared. The collected SORS data of all the samples for each cover were arranged in a matrix, as previously discussed, and then categorized according to their adulteration concentration and divided into calibration and validation sets. A PLSR model was first developed with the SORS data and then with non-SORS (conventional backscattering Raman) data. Therefore, four PLSR models were developed: SORS data collected with covers 1 and 2, and non-SORS data collected with covers 1 and 2. The optimum number of latent variables to be used with the PLSR algorithms is an important parameter for achieving better prediction performance. The number of latent variables used in PLSR models was chosen to optimize the model performance and minimize the model error, such as the error incurred by under- and over-fitting the data, based on the lowest root-mean-square error (RMSE) for the cross-validation set [24].

The results obtained with all the PLSR models developed with the four data sets are summarized in Table 2. The PLSR results obtained from the quantification analysis of the validation sets developed with the conventional backscattering Raman data for adulterated butter samples do not exhibit a strong relationship between the actual and predicted concentrations as they yielded low correlation values (R2) of 0.84 and 0.64 with a high standard error of prediction (SEP) of 5.8 and 8.4% (Fig. 5a, b) for the PLSR models developed with covers 1 and 2, respectively. However, it should be noted that the PLSR model yielded better prediction result for cover 1 than it did for cover 2. This is possibly because conventional Raman spectroscopy can better see through cover 1 and can differentiate between the Raman signatures of butter and margarine as shown in Fig. 4b; therefore, the variation in the spectral signature reflects the adulterant (margarine) concentration as a result of the superior performance of the PLSR model.
Table 2

Prediction results for the pure and margarine adulterated butter samples from the PLSR models developed with conventional Raman and SORS data sets

 

Cover 1

Cover 2

Parameters

No offset

Offset

No offset

Offset

R 2 v

0.846

0.948

0.64

0.922

SEC

3.4

1.2

4.6

1.8

SEP

5.8

3.2

8.4

3.9

Factors

6

5

5

6

Fig. 5

Regression plot of actual versus calculated percentages of margarine in butter in the validation sets of conventional Raman data: (a) cover 1 and (b) cover 2, and SORS data: (c) cover 1 and (d) cover 2

The PLSR prediction values for SORS of adulterated samples are shown in Fig. 5c and d. The obtained results show an excellent agreement between the SORS-predicted concentrations and the actual concentration of margarine in butter samples. However, the PLSR results obtained for the SORS data with cover 1 have a better linear regression and lower error for validation set (R2 = 0.948, SEP = 3.2%) compared to the SORS data with cover 2 (R2 = 0.922, SEP = 3.9%). In comparison with conventional Raman data, the PLSR models developed with SORS data demonstrate an extensive enhancement in prediction accuracy and a significant decrease in model error. For data collected with cover 2, the R2 value increased from 0.64 to 0.92 and the error value decreased from 8.4 to 3.9% when the SORS geometry used.

The PLSR prediction values for non-SORS data scattered over a wide range, whereas for the SORS data, most of the prediction values fell into the regression line (Fig. 5). Moreover, Fig. S1 demonstrating the residuals were randomly dispersed throughout the x-axis (fitted values) therefore the residuals have constant variance. In addition, the beta coefficients obtained from the PLSR model developed with the conventional data for covered samples exhibited some high-intensity peaks. However, most of the region appears to be affected by noise, probably related to the covering materials (Fig. S2 (a, b)). On the other hand, the beta coefficient obtained for the SORS data points to a spectral difference between the groups of samples, exhibiting some meaningful bands that can be attributed to the variation between the Raman spectra of butter and margarine (Fig. S2 (c, d)).

Qualitative analysis of commercial butter and margarine samples

To test whether the developed SORS method could discriminate between commercial butter and margarine samples while still in their as-sold unopened form, butter and margarine samples of different varieties and origins purchased from supermarkets were scanned under the SORS system with an offset of 5 mm. In addition, all the butter and margarine samples were subsequently measured without the packaging cover in an attempt to identify the natural similarity and differences between the data and further enable the system’s use as a ground tool. Moreover, the conventional Raman data for all the samples were also collected with the package covers in place to determine whether the conventional geometry (0-mm offset) can see through the packaging cover and discriminate between the different groups of samples. The preprocessed spectra for all the butter and margarine samples are shown in Fig. S3 (a). It was notable that there were observable differences between some butter spectra. However, a significant difference can be seen between the data for the butter and margarine. Figure S2(b) shows the preprocessed spectra for all the packaging covers. The main difference between the cover spectra because of the composition of the packaging material from the same company exhibits a significant similarity in the cover spectra, such as covers 3 and 4 (BC3 and BC4), and covers 5 and 6 (BC5 and BC6) in Fig. S3 (b).

The data collected for seven different varieties of butter (B1–B7) and three different varieties of margarine (M1–M3) were arranged in a matrix and a PCA analysis was performed on the preprocessed data. The PCA score plot (Fig. S4) of the Raman data, collected with the conventional geometry and without covers from the butter and margarine samples, displayed a separation of the samples according to their natural (chemical) characteristics. Therefore, the method can be further considered for use as a tool by evaluating the loading and scores plots of PCA for two other data sets; the conventional Raman through-packaging data and the SORS through-packaging data. PCA was further executed on two other data sets and the resulting score plots are shown in Fig. 6. A full evaluation of the PCA score plot for conventional Raman through-packaging data shows that the data are classified according to the packaging cover rather than the contained butter/margarine composition because the samples marked B3 and B4 (butter 3 and 4) appear to overlap, while the same overlapping can also be seen for B5 and B6 (Fig. 6a). As stated previously, there is a significant similarity in the packaging covers of these samples, and a correlation matrix which shows the similarities between Raman spectra of the different covers is also given in Fig. S5. The similarity between the aforementioned covers can be seen with a correlation value (R) > 0.99.
Fig. 6

The PCA scores plots: a conventional Raman data collected through packaging and b SORS data collected through packaging. Different symbols represent the different butter (B1–B7) and margarine (M1–M3) samples

On the other hand, the PCA score plot for the SORS data collected for the butter and margarine samples through the packaging covers exhibits data grouping (Fig. 6b) and follows a data scattering pattern similar to that shown in Fig. S4 (ground tool). Moreover, unlike in Fig. 6a, there is no overlapping data because similarities in the Raman spectra of packaging covers were identified. The discrimination between the different butter and margarine samples within their packaging using the SORS technique is certainly a result of variations in the contents (butter and margarine) rather than any variations in the material of the covers. It is also evident from the PCA loading plots (Fig. 7) that the peak at 1442 cm−1 is a significant contributor to the separation which can be seen in the Raman spectra of the butter and margarine samples (Fig. 4a). Moreover, the peaks marked with the dashed lines in Fig. 7a and b are caused by the spectral variations between different groups of samples. A visual comparison of the loading plots for three different data sets reveals that the loading for the SORS data features peaks that are similar to the loadings used as a ground tool (non-SORS—unpacked samples), and these peaks cannot be seen in the loading plot for the conventional Raman data collected for the packaged samples (non-SORS—packed samples). However, it should be noted that the PCA loadings for SORS data are not entirely a result of variations in the contents but also are somehow affected by the superficial layer of the packaging cover. Therefore, they exhibit some peaks in the same spectral region of non-SORS packed samples which are due to the variations in the packaging covers.
Fig. 7

The first two principal component loadings: a loading 1 and b loading 2 calculated for three different data sets

Based on the results of this study, we envision the developed line-scan system being readily applied to the high-throughput quality and authenticity analysis of plastic-packaged butter/margarine samples. The ultimate advantages of this macro-scale SORS technique over benchtop or handheld SORS techniques is speed because the use of a wider (ca. 14 cm) laser line allows a user to scan a large area along the laser-line at one time and each scan can further be averaged to increase the S/N ratio of the final spectrum. In addition, relative to the closely related transmission Raman spectroscopy (TRS), our technique is more effective for this particular application because TRS cannot separate the information from individual layers. Thus, the yielded spectra will attain equal information from both packaging layers and contents which ultimately affects the performance of the multivariate data analysis method used for qualitative or quantitative analysis. Moreover, in this particular application, the Raman signals collected on opposite sides using TRS would not actually be representative of the contents (butter/margarine) because of the thin packaging layer of the complex aluminum over the top of the package. Therefore, the present line-scan-based macro-scale SORS technique could be an effective and direct approach for the qualitative and quantitative analysis of (plastic) packaged commercial butter/margarine samples.

Conclusion

The existing systems for SORS measurements either do not provide instrumental flexibility by restricting the selection of optimal offset distance and interval (fiber probe array) or are limited to small-scale analysis (point-scan-based SORS). In the present study, we have proposed and demonstrated the extension of spatially offset Raman spectroscopic techniques for the collection of Raman spectra from deep layers under the packaging. The present work has several important implications for the simultaneous collection of a series of Raman spectra using an offset laser line: the total scanning time for SORS measurements can be significantly decreased and a large number of samples with large surface areas can be qualitatively and quantitatively analyzed rapidly.

The potential of the developed SORS system was evaluated for the quantification of butter adulteration with margarine through two different packaging covers. The collected data were preprocessed and PLSR models were developed with each data set. The comparison of PLSR results obtained for through-packaging analysis of butter adulteration using conventional and SORS geometries of Raman spectroscopy showed that the SORS geometry of Raman spectroscopy has the potential to see effectively through the packaging covers. In addition, employing the multivariate analysis method of PLSR, margarine concentration in packaged butter samples can be predicted with high accuracy. Next, to test whether the method had the potential to be transposable to commercially packaged butter and margarine samples, the SORS data for ten (seven butter and three margarine) different commercial varieties of butter and margarine samples were collected and used to develop a PCA model. This produced promising results and, therefore, proved that the developed technique was an effective, non-destructive, and rapid means of qualitatively and quantitatively analyzing commercially packaged butter/margarine samples.

In summary, we propose a macro-scale line-scan SORS technique as a noninvasive tool to probe the Raman spectrum of butter and margarine through the packaging layer. This study paves the way for the future Raman analysis of a range of packaged food products in a high-throughput (macro-scale) manner, which is not possible with existing benchtop or handheld SORS and TRS systems for the aforementioned reasons. A particularly interesting challenge is to recover detailed spectral information from samples packaged with traditional aluminum complexes because the aluminum packaging seems to reflect most of the incident laser light back thus does not allow the Raman signals to be generated from the contents. Nevertheless, since only minor modifications to our previously developed Raman imaging system are required to upgrade it into a SORS system, the developed system can also be used for SORS imaging, which will have several potential applications in diverse disciplines where micro-scale SORS imaging is limited because of the scanning speed, such as the quantitative prediction and spatial visualization of active pharmaceutical ingredients in tablets, discovering hidden painted images with large surface areas, and in biological analyses where it is necessary to see through the skin and visualize the deep tissue. Overall, the examples given herein illustrate the wide range of potential applications of the developed line-scan SORS system.

Notes

Acknowledgements

This research was supported by research fund of Chungnam National University.

Compliance with ethical standards

The samples used in this study were purchased from a supermarket, and no commercial or financial relationship with the products’ brand used. Informed consent was provided by all individuals involved in this study.

Conflict of interest

The authors declare that they have no conflict of interest.

Supplementary material

216_2018_1189_MOESM1_ESM.pdf (616 kb)
ESM 1 (PDF 616 kb)

References

  1. 1.
    Reid LM, O’donnell CP, Downey G. Recent technological advances for the determination of food authenticity. Trends Food Sci Technol. 2006;17:344–53.CrossRefGoogle Scholar
  2. 2.
    Ellis DI, Brewster VL, Dunn WB, Allwood JW, Golovanov AP, Goodacre R. Fingerprinting food: current technologies for the detection of food adulteration and contamination. Chem Soc Rev. 2012;41:5706–27.CrossRefPubMedGoogle Scholar
  3. 3.
    Li-Chan ECY. The application of Raman spectroscopy in food science. Trends Food Sci Technol. 1996;7:361–70.CrossRefGoogle Scholar
  4. 4.
    Lohumi S, Kim MS, Qin J, Cho B-K. Raman imaging from microscopy to macroscopy: quality and safety control of biological materials. Trends Anal Chem. 2017;93:183–98.CrossRefGoogle Scholar
  5. 5.
    Qin J, Kim MS, Chao KCB-K. Raman chemical imaging technology for food and agricultural applications. J Biosyst Eng. 2017;42:170–89.Google Scholar
  6. 6.
    Matousek P. Deep non-invasive Raman spectroscopy of living tissue and powders. Chem Soc Rev. 2007;36:1292–304.CrossRefPubMedGoogle Scholar
  7. 7.
    Matousek P, Clark IP, Draper ERC, Morris MD, Everall N, Towrie M, et al. Subsurface probing in diffusely scattering media using spatially offset Raman spectroscopy. Appl Spectrosc. 2005;59:393–400.CrossRefPubMedGoogle Scholar
  8. 8.
    Conti C, Colombo C, Realini M, Zerbi G, Matousek P. Subsurface Raman analysis of thin painted layers. Appl Spectrosc. 2014;68:686–91.  https://doi.org/10.1366/13-07376.CrossRefPubMedGoogle Scholar
  9. 9.
    Afseth NK, Bloomfield M, Wold JP, Matousek P. A novel approach for subsurface through-skin analysis of salmon using spatially offset Raman spectroscopy (SORS). Appl Spectrosc. 2014;68:255–62.CrossRefPubMedGoogle Scholar
  10. 10.
    Qin J, Chao K, Kim MS. Nondestructive evaluation of internal maturity of tomatoes using spatially offset Raman spectroscopy. Postharvest Biol Technol. 2012;71:21–31.CrossRefGoogle Scholar
  11. 11.
    Chao K, Dhakal S, Qin J, Peng Y, Schmidt W, Kim M, et al. A spatially offset Raman spectroscopy method for non-destructive detection of gelatin-encapsulated powders. Sensors. 2017;17:1–12.CrossRefGoogle Scholar
  12. 12.
    Matousek P, Stone N. Development of deep subsurface Raman spectroscopy for medical diagnosis and disease monitoring. Chem Soc Rev. 2016;45:1794–802.CrossRefPubMedGoogle Scholar
  13. 13.
    Botteon A, Conti C, Realini M, Colombo C, Matousek P. Discovering hidden painted images: subsurface imaging using microscale spatially offset Raman spectroscopy. Anal Chem. 2017;89:792–8.CrossRefPubMedGoogle Scholar
  14. 14.
    Liu W, Ong YH, Yu XJ, Ju J, Perlaki CM, Liu LB, et al. Snapshot depth sensitive Raman spectroscopy in layered tissues. Opt Express. 2016;24:28312–25.CrossRefPubMedGoogle Scholar
  15. 15.
    Qin J, Chao K, Kim MS. Investigation of Raman chemical imaging for detection of lycopene changes in tomatoes during postharvest ripening. J Food Eng. 2011;107:277–88.CrossRefGoogle Scholar
  16. 16.
    Fadzlillah NA, Rohman A, Ismail A, Mustafa S, Khatib A. Application of FTIR-ATR spectroscopy coupled with multivariate analysis for rapid estimation of butter adulteration. J Oleo Sci. 2013;62:555–62.CrossRefPubMedGoogle Scholar
  17. 17.
    Nedeljković A, Rösch P, Popp J, Miočinović J, Radovanović M, Pudja P. Raman spectroscopy as a rapid tool for quantitative analysis of butter adulterated with margarine. Food Anal Methods. 2016;9:1315–20.CrossRefGoogle Scholar
  18. 18.
    Koca N, Kocaoglu-Vurma NA, Harper WJ, Rodriguez-Saona LE. Application of temperature-controlled attenuated total reflectance-mid-infrared (ATR-MIR) spectroscopy for rapid estimation of butter adulteration. Food Chem. 2010;121:778–82.CrossRefGoogle Scholar
  19. 19.
    Gat N. Imaging spectroscopy using tunable filters: a review. Proc SPIE. 2000;4056:50–64.  https://doi.org/10.1117/12.381686.CrossRefGoogle Scholar
  20. 20.
    Zhang Z-M, Chen S, Liang Y-Z. Baseline correction using adaptive iteratively reweighted penalized least squares. Analyst. 2010;135:1138–46.CrossRefPubMedGoogle Scholar
  21. 21.
    Qin J, Kim MS, Schmidt WF, Cho B-K, Peng Y, Chao K. A line-scan hyperspectral Raman system for spatially offset Raman spectroscopy. J Raman Spectrosc. 2016;47:437–43.CrossRefGoogle Scholar
  22. 22.
    El-Abassy RM, Eravuchira PJ, Donfack P, Von Der Kammer B, Materny A. Fast determination of milk fat content using Raman spectroscopy. Vib Spectrosc. 2010;56:3–8.CrossRefGoogle Scholar
  23. 23.
    Gallier S, Gordon KC, Jiménez-Flores R, Everett DW. Composition of bovine milk fat globules by confocal Raman microscopy. Int Dairy J. 2011;21:402–12.CrossRefGoogle Scholar
  24. 24.
    Seo Y-W, Ahn CK, Lee H, Park E, Mo C, Cho B-K. Non-destructive sorting techniques for viable pepper (Capsicum annuum L.) seeds using Fourier transform near-infrared and Raman spectroscopy. J Biosyst Eng. 2016;41:51–9.CrossRefGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Authors and Affiliations

  • Santosh Lohumi
    • 1
  • Hoonsoo Lee
    • 2
  • Moon S. Kim
    • 2
  • Jianwei Qin
    • 2
  • Byoung-Kwan Cho
    • 1
  1. 1.Department of Biosystems Machinery Engineering, College of Agricultural and Life ScienceChungnam National UniversityDaejeonSouth Korea
  2. 2.Environmental Microbial and Food Safety Laboratory, Agricultural Research ServiceU.S. Department of AgricultureBeltsvilleUSA

Personalised recommendations