Use of Raman Microscopy and Band-Target Entropy Minimization

Jan 9, 2008 - Abstract. Raman microscopy was used in mapping mode to collect more than 1000 spectra in a 100 μm × 100 μm area from a commercial sta...
0 downloads 11 Views 252KB Size
Download PDF
Anal. Chem. 2008, 80, 729-733

Use of Raman Microscopy and Band-Target Entropy Minimization Analysis To Identify Dyes in a Commercial Stamp. Implications for Authentication and Counterfeit Detection Effendi Widjaja and Marc Garland*

Process Science and Modeling, Institute of Chemical and Engineering Sciences, 1 Pesek Rd, Jurong Island, Singapore 627833

Raman microscopy was used in mapping mode to collect more than 1000 spectra in a 100 µm × 100 µm area from a commercial stamp. Band-target entropy minimization (BTEM) was then employed to unmix the mixture spectra in order to extract the pure component spectra of the samples. Three pure component spectral patterns with good signal-to-noise ratios were recovered, and their spatial distributions were determined. The three pure component spectral patterns were then identified as copper phthalocyanine blue, calcite-like material, and yellow organic dye material by comparison to known spectral libraries. The present investigation, consisting of (1) advanced curve resolution (blind-source separation) followed by (2) spectral data base matching, readily suggests extensions to authenticity and counterfeit studies of other types of commercial objects. The presence or absence of specific observable components form the basis for assessment. The present spectral analysis (BTEM) is applicable to highly overlapping spectral information. Since a priori information such as the number of components present and spectral libraries are not needed in BTEM, and since minor signals arising from trace components can be reconstructed, this analysis offers a robust approach to a wide variety of material problems involving authenticity and counterfeit issues. Due to widespread counterfeiting of various commercial objects, the need for a nondestructive and accurate authentication is obvious. A wide range of analytical methods and instrumental techniques for screening and identifying the suspected counterfeits have been developed.1-6 Raman microscopy is one of the favored techniques, since it is typically nondestructive and it provides a wealth of molecular-level information on the materials * To whom correspondence should be addressed. E-mail: marc_garland@ ices.a-star.edu.sg. (1) Eliasson, C.; Matousek, P. Anal. Chem. 2007, 79, 1696-1701. (2) Sanchez, H. J.; Valentinuzzi, M. C. X-Ray Spectrom. 2006, 35, 379-382. (3) Reid, L. M.; O’Donnell, C. P.; Downey, G. Trends Food Sci. Technol. 2006, 17, 344-353. (4) Olsen, B. A.; Kiehl, D. E. Am. Pharm. Rev. 2006, 9, 115-118. (5) Tran, C. D.; Cui, Y.; Smirnov, S. Anal. Chem. 1998, 70, 4701-4708. (6) Chan, T. W. D.; But, P. P. H.; Cheng, S. W.; Kwok, I. M. Y.; Lau, F. W.; Xu, H. X. Anal. Chem. 2000, 72, 1281-1287. 10.1021/ac701940k CCC: $40.75 Published on Web 01/09/2008

© 2008 American Chemical Society

or objects being analyzed.7-9 As such, identification of specific analytes leading to authentication is made possible. Raman array microscopy in particular,10 which integrates spectroscopy with mapping and imaging technology, has the capability to generate hundreds or even thousands of Raman spectra from one specific area of sample. However, visual examination of all the collected spectra in one area is neither practical nor sufficient for many purposes, and thus, more sophisticated techniques for rapid but detailed objective analysis of the data is needed. Multivariate statistical analysis can certainly play an important role since all the collected spectral information is included simultaneously in the numerical analysis.11-13 Band-target entropy minimization (BTEM) is one multivariate analysis tool which could fit this purpose.14,15 BTEM is a blind source separation algorithm, capable of extracting pure component spectra from matrices of mixture spectra without any need for a priori information. It has been successfully applied to the analysis of various data matrices generated from a wide variety of spectroscopic instruments including infrared,16-18 NMR,19 Raman,20,21 mass spectrometry,22 (7) Wise, D.; Wise, A. J. Raman Spectrosc. 2004, 35, 710-718. (8) Aponick, A.; Marchozzi, E.; Johnston, C.; Wigal, C. T. J. Chem. Educ. 1998, 75, 465-466. (9) Perez-Alonso, M.; Castro, K.; Madariaga, J. M. Curr. Anal. Chem. 2006, 2, 89-100. (10) Cooke, P. M. Anal. Chem. 1998, 70, 385R-423R. (11) Lavine, B. K.; Davidson, C. E.; Ritter, J.; Westover, D. J.; Hancewicz, T. Microchem. J. 2004, 76, 173-180. (12) Medendorp, J.; Lodder, R. A. J. Chemom. 2005, 19, 533-542. (13) Lin, W. Q.; Jiang, J. H.; Yang, H. F.; Ozaki, Y.; Shen, G. L.; Yu, R. Q. Anal. Chem. 2006, 78, 6003-6011. (14) Widjaja, E.; Garland, M. In Proceeding of the International Conference on Scientific & Engineering Computation (IC-SEC). Recent Advances in Computational Sciences and Engineering; Lee, H. P., Kumar, K., Eds.; Imperial College Press: London, 2002; pp 62-66. (15) Widjaja, E. Development of band-target entropy minimization (BTEM) and associated software tools. Ph.D. Thesis, National University of Singapore, 2002. (16) Widjaja, E.; Li, C. Z.; Garland, M. Organometallics 2002, 21, 1991-1997. (17) Chew, W.; Widjaja, E.; Garland, M. Organometallics 2002, 21, 1982-1990. (18) Widjaja, E.; Li, C. Z.; Chew, W.; Garland, M. Anal. Chem. 2003, 75, 44994507. (19) Guo, L. F.; Wiesmath, A.; Sprenger, P.; Garland, M. Anal. Chem. 2005, 77, 1655-1662. (20) Allian, A. D.; Widjaja, E.; Garland, M. Dalton Trans. 2006, 35, 4211-4217. (21) Widjaja, E.; Tan, Y. Y.; Garland, M. Org. Proc. Res. Dev. 2007, 11, 98-103.

Analytical Chemistry, Vol. 80, No. 3, February 1, 2008 729

Figure 1. Raman spectra of “blue area” from Singapore stamp at two distinct spatial positions.

and X-ray diffraction.23 BTEM is particularly suitable for determining the pure component spectra of constituents present in minor or trace amounts, frequently at the ppm or sub-ppm concentration levels.24,25 It has been extensively used to analyze spectra from organometallic and fine chemical syntheses, where 10 or more underlying spectral patterns from a single set of observations (experiment) have been obtained.26,27 The utility of BTEM spectral reconstruction has been clearly and repeatedly shown in the context of identifying previously unknown chemical compounds. In the above-mentioned reaction studies, the pure component spectral patterns were then compared to first principles density functional theory spectral estimates, in order to determine the identity of the nonisolatable and previously unreported intermediates present.20 Since BTEM has demonstrated utility and does not require a priori knowledge, applications outside traditional chemistry fields appear worthwhile. In the present study, confocal Raman array microscopy is applied to a commercial object in order to collect a large series of Raman spectra as a function of spatial position, and then multivariate BTEM analysis is performed. Once the underlying pure component spectra are reconstructed, these spectra can be compared to previously known or observed components or spectral libraries in order to confirm the identity of the constituents present. The authenticity of the commercial object can then be more readily evaluated by the presence or absence of certain components. As an example of this methodological approach, we report herein the use of this technique to determine the constituents present in a commercial Singapore stamp.

was irradiated with a 785-nm near-infrared diode laser, and a 50× objective lens was used to collect the backscattered light. The stamp was placed under the microscope objective, and laser power up to 6 mW was shone onto the surface of the stamp. Raman pointby-point mapping with a step size of 2.5 µm in both the x and y directions was performed on the area of 100 µm × 100 µm. Scans were collected using a static 1800 groove/mm dispersive grating in an spectral window from 1052 to 1700 cm-1, and acquisition time for each spectrum was ∼15 s. The final Raman imaging data cube collected from this experiment had dimension 41 × 41 × 1682.

EXPERIMENTAL SECTION The commercial object used in this study was a new and unused 40-cent Singapore stamp purchased from a post office. Raman spectra were measured using a Raman microscope (InVia Reflex, Renishaw) equipped with near-infrared enhanced deepdepleted thermoelectrically Peltier cooled CCD array detector (576 × 384 pixels) and a high-grade Leica microscope. The sample

COMPUTATIONAL ASPECTS Unlike most self-modeling curve resolution techniques, BTEM was especially developed to resolve one pure spectrum at a time from a set of mixture spectra. The first step in the analysis is the use of singular value decomposition (SVD) to extract the singular vectors that describe the variance of the original mixture data set. With the present data set, in order to extract an estimate of a pure scattering coefficient, a selected band in the first few nonnoise right singular vectors is targeted. The BTEM algorithm then retains this feature and, at the same time, returns an entire fulllength spectrum containing this band and all correlated bands, which have minimum entropy. This routine is repeated for all important observable physical features in the selected right singular vectors. A superset of reconstructed pure component scattering coefficients is obtained, and this set is reduced to eliminate redundancies. This results in an enumeration of all observable pure component spectra. As mentioned, the targeted bands are retained during the reconstruction. As part of the process, the resulting pure spectral patterns are returned in a normalized form. When all normalized observable pure component spectra have been reconstructed, relative contributions of these signals can be calculated by projecting them onto the original data set. For a detailed description of the BTEM algorithm, readers are referred to refs 15-18.

(22) Zhang, H. J.; Garland, M.; Zeng, Y. Z.; Wu, P. J. Am. Soc. Mass Spectrom. 2003, 14, 1295-1305. (23) Guo, L. F.; Kooli, F.; Garland, M. Anal. Chim. Acta 2004, 517, 229-236. (24) Li, C. Z.; Widjaja, E.; Chew, W.; Garland, M. Angew. Chem., Int. Ed. 2002, 41, 3785-3789. (25) Li, C. Z.; Widjaja, E.; Garland, M. J. Am. Chem. Soc. 2003, 125, 55405548. (26) Li, C. Z.; Widjaja, E.; Garland, M. J. Catal. 2003, 213, 126-134. (27) Widjaja, E.; Li, C. Z.; Garland, M. J. Catal. 2004, 223, 278-289

RESULTS Raman point-by-point mapping was applied to a small blue area of a 40-cent Singapore stamp. As can be seen in Figure 1, there is a significant difference in the measured Raman spectra. The two Raman spectra, measured from two different spots, ∼140 µm apart, were significantly different. As such, it can be assumed that the spectral variation or constituent heterogeneity on this mea-

730

Analytical Chemistry, Vol. 80, No. 3, February 1, 2008

Figure 2. The 1681 preprocessed Raman spectra from Singapore stamp.

sured stamp is quite high. Such high spectral variation among the mixture spectra is a good prerequisite for spectral reconstruction using BTEM. The collected Raman mapping spectra were first reorganized (deconcatenated) from a three-way array to a conventional twoway array. This array unfolding was done by stacking up each data column of the image abscissa into a single column. Since the experimental raw Raman spectra acquired from the stamp were the combination of (a) Raman scattering signals, (b) spikes due to cosmic rays, and (c) some autofluorescence background, spectral preprocessing was performed to generate Raman scatter spectra alone. Spikes were removed in the first step followed by noise reduction and baseline correction using five-point adjacent smoothing and third-order modified polynomial fitting. The preprocessed Raman spectra in stack-plot form are shown in Figure 2. SVD was performed on the preprocessed Raman spectra, and the resulting first 12 right singular vectors or basis vectors are shown in Figure 3. These basis vectors contain abstract information on the pure component spectra of the observable components present in the system and are ordered according to their contribution to the total variance in the observations. Hence, the first few vectors are associated with the chemically and physically significant information in the system. The remaining vectors are primarily associated with the random instrumental and experimental error. As can be seen in Figure 3, the first six vectors have very clear localized signals and little noise. The next three vectors still possess localized signals but considerably more noise. The 10th-12th vectors were essentially noise with few localized signals. The remaining vectors (not shown) were essentially white noise. Accordingly, the first 12-16 basis vectors were sufficient in the subsequent BTEM spectral reconstructions. The first few vectors were inspected for suitable targets to be used in the BTEM algorithm. From the first vector, the most significant band at 1529-1532 cm-1 labeled as “1” in Figure 3 was chosen as a target for spectral reconstruction. This feature was also seen in the subsequent first few vectors. The prominent spectral feature in the second vector was in the region of 10841089 cm-1, labeled as “2”, and it can be seen also in the fourth and fifth vector. In the third and fourth vector, another prominent feature was in the region of 1597-1601 cm-1, labeled as “3”, and was also chosen as a target for BTEM.

These three chosen band-targets were subjected to BTEM analysis, and subsequently, three pure component spectra were resolved from the targeted features. The full spectral range pure component spectral estimates are shown in Figure 4a. The relative contributions from each component were then obtained by projecting the spectral estimates onto the preprocessed normalized mixture Raman spectra. The mixture spectra were normalized (L1 norm) to overcome the problem of pixel-to-pixel Raman intensity variation during mapping measurements. The resulting relative contributions were then reorganized into a map or spatial distribution for each component, which are shown in Figure 4b. Note that the axes for spatial distribution are in pixel number, which can be directly converted to distance by multiplying by 2.5 µm for each pixel. Combined, the three spectral patterns account for more than 99% of the measured mixture spectral signals. The pure component spectral estimates obtained from BTEM analysis (without the use of any a priori information) were then compared to known spectral libraries associated with the dye and paper-making industries. It was found that the first spectral estimate (1) is quite similar to porphyrin-based dyes,28,29 including phthalocyanine blue, CuC32H16N8.30,31 Phthalocyanine blue is a frequently used blue dye, which has found use in inkjet printers.32 The Raman band assignment for this component is shown in Table 1. The second spectral estimate (2) recovered from BTEM analysis is similar to the Raman spectrum of calcite, containing peaks at 1086 s, 1258 w, 1450 m, and 1602 m. Calcite is commonly used as one of the pigments for paper.33,34 It has strong Raman peaks at 713 and 1087 cm-1 due to the vibrations of ν4-symmetric CO3 deformation and ν1-symmetric CO3 stretching, respectively, and a weak Raman peak at 1435 cm-1 due to ν3-asymmetric CO3 stretching.32,35,36 In the current BTEM estimate, the intense peak at 1086 cm-1 is very close to the ν1-symmetric CO3 stretching of calcite, but peaks at 1258, 1450, and 1602 cm-1, which are also found, certainly do not belong to calcite. It is suggested that these additional peaks may belong to impurities naturally present or added to modify the calcite. These minor signals are collinear with the calcite signal and, hence, cannot be decoupled from the present sample spectra. The third spectral estimate (3) from the BTEM analysis yields peaks at 1258 m, 1287 w, 1400 m, 1450 w, 1527 m, and 1598 s. This spectrum is very similar to the spectrum obtained by Chaplin et al. when a nongenuine Hawaiian missionary stamp was being investigated using Raman microscopy.30 Chaplin et al. assigned this spectrum as an unidentified organic yellow material that was possibly used to simulate aged paper.30 In our current estimate, an additional peak at 1527 cm-1 was also recovered together with (28) Lu, F.-L.; Rintoul, L.; Sun, X.; Arnold, D. P.; Zhang, X.-X.; Jiang, J.-Z. J. Raman Spectrosc. 2004, 35, 860-868. (29) Aroca, R.; Zeng, Z.-Q.; Mink, J. J. Phys. Chem. Solids 1990, 51, 135-139. (30) Lu, F.-L.; Rintoul, L.; Sun, X.; Arnold, D. P.; Zhang, X.-X.; Jiang, J.-Z. J. Raman Spectrosc. 2004, 35, 860-868. (31) Aroca, R.; Zeng, Z.-Q.; Mink, J. J. Phys. Chem. Solids 1990, 51, 135-139. (32) Chaplin Tracey, D.; Clark Robin, J. H.; Beech David, R. J. Raman Spectrosc. 2002, 33, 424-428. (33) He, P.; Bitla, S.; Bousfiled, D.; Tripp, C. P. Appl. Spectrosc. 2002, 56, 11151121. (34) Burgio, L.; Clark, R. J. H. Spectrochim. Acta, Part A 2001, 57, 1491-1521. (35) Degen, I. A.; Newman, G. A. Spectrochim. Acta, Part A 1993, 49A, 859887. (36) Gunasekaran, S.; Anbalagan, G.; Pandi, S. J. Raman Spectrosc. 2006, 37, 892-899.

Analytical Chemistry, Vol. 80, No. 3, February 1, 2008

731

Figure 3. First 12 right singular vectors from the Raman mapping measurements on a stamp : (a) first vector, (b) second vector, (c) third vector, ..., (l) 12th vector.

Figure 4. (a) Pure component Raman spectral estimates from BTEM. (b) Score image for each pure component that represents its spatial distribution.

the other peaks. This peak is clearly an artifact, as evidenced by the asymmetric sigmoid profile of this band. It could belong to copper-phthalocyanine blue and arises due to imperfect spectral reconstruction. The unidentified organic dye is also a minor component compared to copper-phthalocyanine blue and calcite. The intensities of this organic dye as shown in the score image in Figure 4b are ∼10 times lower than copper-phthalocyanine blue. DISCUSSION In this study, 1681 Raman spectra were collected from a small region of an inexpensive commercial object (40-cent Singapore stamp). Multivariate curve resolution analysis was applied to these 732

Analytical Chemistry, Vol. 80, No. 3, February 1, 2008

spectra, and three pure component spectra were recovered with good signal-to-noise ratios. The three components were then identified by comparison to known spectral libraries. Although this was a simple example, the implications of the present methodological approach for authentication of more valuable commercial objects are obvious. Given appropriate nondestructive and information-rich spectroscopies, multiple independent and meaningful spectra of almost any commercial object can be obtained, and these can be analyzed to obtain pure constituent information. Subsequently, authentication is primarily an issue of having extensive and appropriate/relevant spectral libraries for comparison. The presence of expected constituents and the absence of counterfeit-associated constituents form the

Table 1. Characteristic Raman Bands of Spectral Estimate No. 1 from BTEM and Their Corresponding Band Assignments Showing Typical Porphyrin Modes peak position, cm-1

Raman band assignment

1008 w 1040 w 1109 m 1144 m 1158 w 1170 w 1186 w 1196 m 1210 w 1217 m 1308 m

C-H bending C-H bending C-H bending pyrrole breathing C-H bending C-H bending C-H bending C-H bending C-H bending C-H bending C)C pyrrole and benzene stretching isoindole stretching isoindole stretching isoindole stretching isoindole stretching isoindole stretching C)C pyrrole stretching C)C pyrrole stretching C)N aza stretching benzene stretching benzene stretching

1342 s 1372 w 1414 w 1430 w 1452 s 1471 w 1483 w 1530 s 1598 w 1611 w

primary basis for authentication. At another level of analysis, the relative concentrations of constituents provide a more refined basis for further inquiry. Only 1681 spectra were measured in this study, and these were obtained in a very small area. For more valuable commercial objects, the acquisition of many more spectra, from many different regions of the object, would be prudent and easily justified. Moreover, it is well-known that significant increases in the number of spectra taken leads to significantly improved spectral estimates and the detection of additional minor or trace constituents. This

improvement in spectral reconstruction is due in part to the enlarged set of observations (and hence more signal variance), but it is also related to numerical/statistical issues associated with the determination of the basis vectors. Thus, with increased number of observations, signals that are normally lost in the noise are revealed. Obviously, with sufficient consideration of the spectroscopy and spectral window used, applications for this methodology are numerous. These range from the authentication of ceramics, to paper items (manuscripts, books, bank notes), to paintings, frescos, etc. CONCLUSIONS The present Raman array analysis of a Singapore stamp demonstrates the combination of effective multivariate curve resolution with spectral library matching for commercial objects. In this study, the pure component spectra of three components were recovered with good signal-to-noise ratios, without a priori knowledge. These three components were then identified by subsequent comparison to spectral libraries. This rather general methodology is applicable to a host of spectroscopic measurements taken from a wide range of commercial items. Such a methodology provides a firm scientific basis for evaluating authenticity of commercial objects, based on the presence or absence of specific components. ACKNOWLEDGMENT The authors thank the Institute of Chemical and Engineering Sciences for support of this work.

Received for review September 17, 2007. Accepted October 29, 2007. AC701940K

Analytical Chemistry, Vol. 80, No. 3, February 1, 2008

733