Method and system for detecting and authenticating tracers in labels via surface-enhanced Raman spectroscopy

Abstract

The present invention relates to a method and a corresponding system for checking the presence of a genuine SERS or SERRS tracer having a unique characteristic surface-enhanced scattering signature on a machine-readable marking applied to a valuable security, using a Raman spectrometer suitable for Raman spectroscopic analysis of the marking. The method according to the invention enables reliable and rapid detection of the presence of SERS / SERRS tracers and is particularly suitable for checking the authenticity of valuable security documents, such as banknotes, that are being moved at a given speed, possibly at high speed, relative to a Raman spectrometer or are briefly exposed to the Raman spectrometer.

Description

Technical Field

[0001] The present invention relates to the technical field of detecting tracers present in markings on substrates (e.g. banknotes) by surface-enhanced Raman spectroscopy (SERS) or by surface-enhanced resonance Raman spectroscopy (SERRS). The tracers are of the SERS or SERRS type and therefore have unique characteristic surface enhancement signatures (i.e. surface-enhanced Raman scattering signatures or surface-enhanced resonance Raman scattering signatures, respectively) which enable their detection using standard Raman spectrometers. Background Art

[0002] As is well known to those skilled in the art, a SERS or SERRS tracer comprises an aggregate of nanoparticles presenting a plasmonic surface and Raman-active reporter molecules adsorbed on the surface of the nanoparticles. The nanoparticles presenting a plasmonic surface are responsible for generating the electric field required for Raman amplification, while the Raman-active reporter molecules provide the unique vibrational fingerprint of the SERS tracer. The SERS or SERRS tracer may further comprise an outer coating layer that isolates the aggregate of nanoparticles having the Raman-active molecules adsorbed on the surface from the external medium. Thus, the outer coating layer: a) isolates the SERS / SERRS tracer from the external medium, thereby preventing the Raman-active reporter molecules from being leached out of the SERS / SERRS tracer and protecting the SERS / SERRS tracer from contamination by the external medium that may produce false peaks; b) increases the colloidal stability of the SERS / SERRS tracer; and c) provides a convenient surface for further chemical functionalization. The outer coating layer includes silica and polymers such as poly(ethyleneimine) (PEI), poly(styrene-alt-maleic acid) sodium salt (PSMA), poly(diallyldimethylammonium chloride) (PDADMAC), and the like.

[0003] Raman spectroscopy is widely used for quantitative pharmaceutical analysis, but common barriers to its use are, e.g. Figure 1 As shown, due to the fact that the Raman signal has a much shorter duration than the fluorescence signal, the sample fluorescence usually masks the scattered Raman signal. Figure 1The Raman intensity signal (10) (relative intensity values) resulting from illumination with a 600 ps laser pulse (a 1 ns threshold is shown with a vertical dashed line), and various luminescence (fluorescence) intensity signals (11, 12, 13, and 14) (having lifetimes of 1 ns, 5 ns, 10 ns, and 50 ns, respectively) are shown. It is known that time gating provides an instrument-based method for rejecting a large portion of the fluorescence signal by the temporal resolution of the spectral signal, and time gating enables the acquisition of Raman spectra of fluorescent materials. An additional practical advantage is that spectral signal analysis can be performed even in ambient lighting. Conventional partial least squares (PLS) regression enables spectral signal quantification with improved performance (based on visual inspection) Raman active time domain selection. Model performance is further improved by using kernel-based regularized least squares (RLS) regression and greedy feature selection (i.e., "forward selection" by selecting the best features one by one, or "backward selection" by removing the worst features one by one), where data usage in both the Raman shift dimension and the time dimension is statistically optimized. In particular, overall time-gated Raman spectroscopy with optimized data analysis in both the spectral and temporal dimensions shows potential for sensitive and relatively routine quantitative analysis of photoluminescent materials, such as drugs during pharmaceutical development and manufacturing.

[0004] Raman spectra are obtained by measuring the intensity distribution of Raman scattered photons as a function of wavelength, which are received from a substrate containing a substance of interest and illuminated by a monochromatic light source. Quantitative determination is based on the concentration of the substance of interest, which is proportional to the integrated intensity of its characteristic Raman band. However, overlapping peaks of different compounds in a mixture present on the substrate and experimental effects that are not related to the sample concentration often complicate signal analysis. In this case, multivariate analysis that can include a large amount of spectral data is more reliable than methods that only consider one or a few spectral features. Several multivariate methods have been established for interpreting Raman spectra. The purpose of such methods is to: (i) extract spectral information that quantifies the substance of interest, (ii) estimate the uncertainty of the quantification, and (iii) evaluate the performance of the constructed model.

[0005] Partial least squares (PLS) regression is one of the most widely used chemometric methods for the quantitative analysis of spectra. PLS contacts the information in the two data matrices X (e.g., spectral variation) and Y (e.g., sample components) by maximizing the covariance of the two data matrices in the multivariate model. Kernel-based regularized least squares (kernel-based RLS) regression is another method, which has the ability to learn functions from nonlinear data features, which, when combined with feature selection algorithms such as greedy forward feature selection, optimizes the use of the information provided by the data features. PLS and RLS are very similar because the two are intended to shrink the solution away from the ordinary least squares solution in the direction of the variable space with a large sample dispersion of lower variability.

[0006] Known sources of error in the quantitative analysis of powder mixtures using Raman spectroscopy include intra- and inter-day variations in the Raman instrument, changes in room temperature and humidity, sample fluorescence, mixing, packing, and positioning, as well as sample particle size and compactness. While most issues can be addressed with appropriate spectral processing and data analysis methods, complete subtraction of fluorescence without any instrument-based methods is difficult, even with sophisticated algorithms.

[0007] Furthermore, in many potential applications, the measured Raman spectra are obscured by a strong fluorescence background. This is because the probability of Raman (cross-sectional) scattering is much lower than that of fluorescence. In other words, Raman scattering and fluorescence emission are two competing phenomena, and the spectrum is dominated by the most likely phenomenon, which is usually fluorescence. This will therefore cause a continuous background in the residual spectrum and, in particular, increase the photon shot noise, reducing the signal-to-noise ratio. This leads to uncertainty in both material identification and concentration measurement.

[0008] However, Raman and fluorescence scattered photons have different lifetimes. Raman photons are observed very quickly during excitation (with a laser), while fluorescence photons can still be detected nanoseconds or even milliseconds later, so if scattered photons are collected only during the short Raman scattering phase, the fluorescence background can be suppressed. This can be achieved by irradiating the sample with short, intense laser pulses (with a pulse width much smaller than the fluorescence lifetime) rather than conventional continuous wave ("CW") radiation and recording the sample response only during these short pulses. Therefore, by synchronizing the measurement with the period of the laser pulse, the probability of detecting fluorescence photons can be reduced, because fluorescence photons are mainly emitted after Raman scattered photons. In addition, the accuracy of the baseline of the Raman spectrum is improved, which also leads to higher accuracy in both material identification and quantitative analysis. The synchronization (or gating) signal is a digital signal or pulse (sometimes called a "trigger") that provides a time window so that a specific event or signal among many events or signals will be selected and other events or signals will be eliminated or discarded.

[0009] Synchronization can be achieved with various detection systems (such as time-resolved photomultiplier tubes, high-speed optical gates based on Kerr cell enhanced charge-coupled devices, quantum dot resonant tunneling diodes, and complementary metal oxide semiconductor single photon avalanche diodes (CMOS SPADs). One of the basic advantages of CMOS SPAD is the ability to suppress both photoluminescence tails and photon noise. SPAD is implemented in standard CMOS technology and contains a pn junction that is reverse biased above its breakdown voltage, which means that even the entry of a single photon can trigger avalanche breakdown, which can then be recorded. The width and position of the time gate need to be appropriately selected. Current CMOS single photon avalanche diodes are compact and cheap while being able to achieve sufficient time resolution (sub-nanoseconds). CMOS SPAD detectors have been used to evaluate fluorescence lifetime. Recently, the applicability of CMOS SPAD for fluorescence suppression in Raman spectroscopy of pharmaceutical products has also been demonstrated.

[0010] Some early studies have implemented this "time-gating" technique using high-speed optical shutters based on Kerr cells or mode-locked lasers with spectrometers and intensified CCDs (ICCDs, "intensified charge-coupled devices"). In addition, some analysis was performed to determine the appropriate gate positions for ICCDs and CCDs to achieve optimal fluorescence suppression efficiency. However, these devices are either highly complex, physically large and expensive, or can only measure a single wavelength band of the spectrum at a time, making them unsuitable for field applications due to the long measurement times they require and inability to use in situations where the sample is moving relative to the Raman spectrometer. To overcome these problems, CCDs and ICCDs should be replaced with more suitable detectors.

[0011] A problem arises when a Raman spectrometer is used to authenticate a SERS tracer or SERRS tracer present in a marking (e.g. a pattern printed with ink containing a SERS / SERRS tracer) applied to a security (e.g. a banknote). More specifically, the spectrum measured by the Raman spectrometer comprises a tracer "fingerprint" (i.e. spectral characteristics that uniquely identify the tracer) as well as additional interference or background information. The SERS or SERRS tracer (spectral) fingerprint comprises vibrational bands represented by multiple peaks of different widths and having the shape of a Gaussian / Lorentzian distribution at different positions in the spectrum. The positions of the peaks in the spectrum are not absolute and will depend on the wavelength of the laser excitation light (due to shifts from the laser wavelength). Raman and SERS / SERRS signals are physical effects that are different from fluorescence: the substrate of the security (e.g. the paper of the banknote) as well as the marking (e.g. the ink present on the banknote) have fluorescence spectra that can be measured by a Raman spectrometer. In case different inks (e.g. multiple prints on a banknote), substrates (e.g. paper) and tracers are present in the same measurement track of the spectrometer, the spectral content thus obtained is cumulative. Therefore, the measurement from a Raman spectrometer usually consists of a plurality of spectral information resulting from the cumulative effect. Some of this spectral information is known ("known spectral data"), such as ink, paper, tracer etc., and is stable over time (depending on the banknote design). However, some of this spectral information is unknown ("unknown spectral data") and is due to (constantly changing) external conditions during the measurement process (e.g. polluting fumes (e.g. the presence of human sweat, or even beer, or food traces...) or the presence of stains on the carrier of the tracer etc.). These unknown spectral information are added during the circulation of the banknote and cannot be anticipated. Furthermore, these issues are even more relevant in the case of measurements on high-speed moving documents requiring very short integration times (e.g. 100 to 500 μs), such as in the case of banknotes transported at several meters / second (e.g. 10 to 12 m / s or more) in banknote sorting devices with a high spatial resolution (e.g. a few millimeters).

[0012] Under such harsh conditions, current state of the art solutions involve, for example, as disclosed in US 10,417,856 B2, using a large number (i.e. 100 or more) of spectral channels to measure the entire Raman spectrum, together with a small entrance slit (the higher the spectral resolution, the smaller the slit must be, and thus the less light on the CCD sensor), which may be coupled to an absorbing wall in the Raman spectrometer (for partially absorbing the interfering Rayleigh scattered excitation light). The problem addressed in this patent is the situation where a constructed banknote has to be authenticated by detecting the SERS spectrum of a security tracer. The solution disclosed is to map out the complete banknote by using multiple small measurements transported along the banknote. This requires integration times of several hundred microseconds and as a result the readable signal is very low under this mechanism (which is why the spectral resolution needs to be compromised). An improved differentiation between the Raman spectrum of the tracer and the spectrum caused by other components of the banknote is disclosed in US 2007 / 0165209 A1. However, there is still a need for faster detection of Raman spectra with higher signal levels to provide a more reliable diagnosis. Summary of the Invention

[0013] The present invention relates to a method and a corresponding system capable of checking the presence of a genuine SERS or SERRS tracer having unique characteristic surface enhancement features on a machine-readable mark applied to a valuable security (e.g., a banknote or label having a mark printed with an ink containing the tracer) by using a Raman spectrometer adapted to perform Raman spectroscopy (RS) analysis of the mark. The present invention can be used to authenticate valuable security or articles marked with a SERS or SERRS tracer according to various processes, such as the following:

[0014] The tracer(s) may be present in a specific area within a portion of a substrate of a valuable document or article: for example, in the case of a paper substrate (e.g. a banknote), the tracer(s) may be fixed to the fibers of the paper in the specified area. In this case, the marking comprising the tracer(s) is the portion of the substrate impregnated with the tracer(s).

[0015] The tracer(s) may be mixed with an ink printed on a specific area of ​​a substrate of a security or article. In this case, the marking containing the tracer(s) is a portion of the substrate printed with the ink containing the tracer(s).

[0016] The tracer(s) may be mixed (e.g., as a layer) with a material (e.g., a varnish) applied to a specific area of ​​a substrate of a security document or article. In this case, the marking comprising the tracer(s) is the portion of the substrate to which the material is applied.

[0017] The tracer(s) may be mixed with the specific material of the coating layer applied to the plastic carrier.

[0018] In all cases, the marking applied to the document or article comprises a material comprising (one or more than one) SERS or SERRS tracers (e.g. a portion of the substrate itself comprising tracer fibers, or an ink printed on the substrate, or a varnish layer applied to the substrate, ...).

[0019] The method according to the invention enables the reliable and rapid detection of the presence of genuine SERS or SERRS tracers and is particularly suitable for checking the authenticity of valuable documents (e.g. banknotes, etc.) marked with said tracers, which are moving relative to a Raman spectrometer at a given speed and possibly at high speed (e.g. 10 m / s or more) or are only briefly exposed to the Raman spectrometer (e.g. as in a sorting machine).

[0020] In order to overcome the above-mentioned shortcomings of the prior art, the present invention relates to a method for authenticating a mark, the mark being applied to a substrate and having a composition comprising a first material, the first material comprising a SERS tracer or a SERRS tracer, the method comprising the following steps:

[0021] - defining a complete model of the Raman spectrum of a true marker applied on a true substrate and having a component comprising a true first material as a first weighted sum of: a reference Raman spectrum of the true tracer collected when the true tracer, a reference true substrate and a reference true first material are illuminated with excitation light, respectively; a reference Raman spectrum of the reference true substrate not marked with the true tracer; and a reference Raman spectrum of the reference true first material not including the true tracer, the true first material comprising a true SERS tracer or a true SERRS tracer;

[0022] - defining a simplified model of the Raman spectrum of a simplified marker as a second weighted sum of the reference Raman spectrum of the reference true substrate and the reference Raman spectrum of the reference true first material, the simplified marker differing from the true marker only in that the components of the simplified marker do not include the true tracer;

[0023] - when irradiating the marker with the excitation light, measuring a corresponding Raman light signal scattered by the marker via a Raman spectrometer to obtain a measured Raman spectrum of the marker;

[0024] - fitting the measured Raman spectrum to the complete model of the Raman spectrum by calculating values ​​of weights in the complete model that minimize the difference between the complete model and the measured Raman spectrum subject to the non-negativity constraint of the weights and obtaining corresponding first residuals;

[0025] - fitting the measured Raman spectrum to the simplified model of the Raman spectrum by calculating the following values ​​of the weights in the simplified model, which minimize the difference between the simplified model and the measured Raman spectrum under the constraint of non-negativity of the weights, and obtaining corresponding second residuals;

[0026] - calculating an F value corresponding to an F test comparing the complete model with the simplified model for the measured Raman spectrum based on the obtained first and second residuals; and

[0027] - deciding whether the tracer is present in the label based on the calculated F value.

[0028] Therefore, in the case where the F value is compatible with the presence of a true SERS or SERRS tracer in the test mark, the mark is considered to be authentic. In the case where the F value is incompatible with the presence of a true SERS or SERRS tracer in the test mark, the mark can be considered to be counterfeit or at least suspicious. The reference to the true substrate differs from the true substrate only in that it is not marked with a true (SERS or SERRS) tracer. Similarly, the reference to the true first material differs from the true first material only in that it does not include a true (SERS or SERRS) tracer. Of course, in the case where the mark to be checked is actually authentic, its first material and its tracer also correspond to the true first material that includes the true tracer. The above-mentioned reference to the true substrate represents the corresponding true substrate without marking (for example, the paper substrate of a banknote before it is printed), and the reference to the true first material represents the corresponding true first material that does not include any tracer.

[0029] The method according to the present invention is particularly suitable for the case where, during the operation of measuring the Raman light signal scattered by the label, the label is moving relative to the Raman spectrometer.

[0030] In the above method, the labeled component may include a second material, and the weighted sums of the complete model and the simplified model may further include a reference spectrum of the corresponding true second material collected when the true second material is illuminated by the excitation light, with corresponding weights. The second material (e.g., ink) is generally different from the first material including the tracer and does not include the tracer.

[0031] In a preferred mode, the Raman spectrometer has a plurality of spectral channels, and the operation of measuring the Raman light signal scattered by the label comprises:

[0032] - dispersing the collected Raman light into the plurality of spectral channels, and acquiring a two-dimensional digital image of the dispersed spectral data using an imaging unit;

[0033] - Pre-processing the acquired two-dimensional digital image by performing the following operations using a processing unit equipped with a memory:

[0034] - transforming the two-dimensional spectral data into one-dimensional spectral data via line binning and conversion of the binned data into wavelength data;

[0035] - resampling the one-dimensional spectral data to obtain a one-dimensional spectrum having data points with equal distances in terms of dimensionality;

[0036] - calibrating the one-dimensional spectrum relative to a reference white light spectrum stored in the memory to obtain a calibrated spectrum;

[0037] - filtering the calibrated spectrum using a low-pass filter to obtain a filtered spectrum; and

[0038] - aligning the filtered spectrum in terms of wavelength with a reference spectrum of the tracer stored in the memory, thereby obtaining a pre-processed spectrum; and

[0039] - performing an operation of calculating the first residual and the second residual by using the preprocessed spectrum as the measured Raman spectrum.

[0040] The optics and gratings of a Raman spectrometer cause typical (two-dimensional) deformations of the Raman lines formed on a two-dimensional image (the Raman lines are bent and compressed). Line binning and calibration operations are performed to compensate for these deformations of the Raman lines. Calibration operations are typically performed using (reference) excitation light delivered by an argon lamp to calculate the two-dimensional deformation of the Raman lines by comparison with the image of the observed argon lines.

[0041] According to the above preferred mode, the method may include:

[0042] - defining a spectral measurement vector as the vector corresponding to the obtained pre-processed spectrum;

[0043] - defining a first spectral vector as the product of a first weight vector and a complete design matrix having columns respectively representing reference spectral data of the complete model, and determining respective non-negative components of the first weight vector, the first weight vector minimizing a first residual vector corresponding to a difference between the first spectral vector and the spectral measurement vector via a least squares method;

[0044] - defining a second spectral vector as a product of a second weight vector and a simplified design matrix having columns respectively representing reference spectral data of the simplified model, and determining respective non-negative components of the second weight vector, the second weight vector minimizing a second residual vector corresponding to a difference between the second spectral vector and the spectral measurement vector via a least squares method;

[0045] - calculating a first residual sum of squares RSS1 of errors corresponding to the first weight vector, the first weight vector having p1 non-negative components;

[0046] - calculating a second residual sum of squares RSS2 of errors corresponding to the second weight vector, the second weight vector having p2 non-negative components; and

[0047] - The F value is calculated as the ratio of the difference between the second residual sum of squares RSS2 and the first residual sum of squares RSS1 divided by the difference between the numbers p2 and p1, and the first residual sum of squares RSS1 divided by the difference between the number N of components of the spectral measurement vector and the number p1, i.e. F = ((RSS2-RSS1) / (p1-p2)) / (RSS1 / (N-p1)).

[0048] Furthermore, the operation of determining each non-negative component of the first weight vector and the second weight vector may include:

[0049] - expressing the first weight vector that minimizes the first residual vector as the product of the pseudo-inverse matrix of the complete design matrix and the spectral measurement vector, and expressing the second weight vector that minimizes the second residual vector as the product of the pseudo-inverse matrix of the simplified design matrix and the spectral measurement vector; and

[0050] In the case where a component of the first weight vector or the second weight vector, respectively, has a negative value:

[0051] - modifying the full design matrix or the reduced design matrix accordingly by removing spectral vectors corresponding to negative components from the full design matrix or the reduced design matrix;

[0052] - setting the negative value components to zero; and

[0053] - recalculating the pseudo-inverse matrix of the modified full design matrix or the modified reduced design matrix, respectively, until the obtained components of the first weight vector and the second weight vector have only non-negative values.

[0054] The present invention also relates to a system operable to implement the steps of the above method, the system being for authenticating a mark, the mark being applied to a substrate and having a composition comprising a first material, the first material comprising a SERS tracer or a SERRS tracer, the system comprising a light source, a Raman spectrometer, an imaging unit, and a control unit having a processing unit and a memory, the light source being controlled by the control unit via a current loop to deliver calibrated excitation light, the system being configured to perform the following operations:

[0055] - illuminating the marking with the excitation light delivered by the light source controlled by the control unit; and

[0056] - collecting Raman light generated from the marker, dispersing the collected Raman light in the Raman spectrometer having a plurality of spectral channels, acquiring a two-dimensional digital image of corresponding spectral data using the imaging unit, and storing the acquired spectral data in the memory as a measured Raman spectrum of the marker;

[0057] in,

[0058] the memory stores a complete model of the Raman spectrum of a true marker applied on a true substrate and having a component comprising a true first material as a first weighted sum of: a reference Raman spectrum of the true tracer collected when the true tracer, a reference true substrate and a reference true first material are illuminated with excitation light, respectively; a reference Raman spectrum of the reference true substrate not marked with the true tracer; and a reference Raman spectrum of the reference true first material not including the true tracer, the true first material comprising a true SERS tracer or a true SERRS tracer;

[0059] the memory stores a simplified model of the Raman spectrum of a simplified marker as a second weighted sum of the reference Raman spectrum of the reference true substrate and the reference Raman spectrum of the reference true first material, the simplified marker differing from the true marker only in that the components of the simplified marker do not include the true tracer; and

[0060] The system is further configured to perform the following operations via the processing unit:

[0061] - fitting a measured Raman spectrum stored in the memory to the stored complete model of the Raman spectrum by calculating the following values ​​of the weights in the complete model, which minimize the difference between the complete model and the measured Raman spectrum under the constraint of non-negativity of the weights, and obtaining corresponding first residuals and storing them in the memory;

[0062] - fitting the measured Raman spectrum stored in the memory to the stored simplified model of the Raman spectrum by calculating the following values ​​of the weights in the simplified model and obtaining corresponding second residuals and storing them in the memory;

[0063] - calculating an F value corresponding to an F test comparing the complete model with the simplified model for the measured Raman spectrum based on the stored first and second residuals and storing the calculated value in the memory; and

[0064] - deciding whether the tracer is present in the label based on the stored F value, and delivering a signal representative of the result of the decision.

[0065] In a preferred embodiment of the system, during the operation of measuring the Raman light signal scattered by the marker, the marker is moving relative to the Raman spectrometer, and the control unit synchronizes the illumination of the marker with the light source and the acquisition of the measured Raman spectrum via the Raman spectrometer and the imaging unit with the movement of the marker.

[0066] In the above system, when the component of the mark includes a second material, each weighted sum of the complete model and the simplified model also includes a reference spectrum of the true second material with a corresponding weight, collected when the corresponding true second material is illuminated by the excitation light and stored in the memory. For example, in the case of a printed mark, the true second material can correspond to a set of inks used for the printed mark but does not include a SERS or SERRS tracer.

[0067] In the above system, the processing unit may be configured to pre-process the stored two-dimensional digital image by performing the following operations:

[0068] - transforming the two-dimensional spectral data into one-dimensional spectral data via line binning and conversion of the binned data into wavelength data;

[0069] - resampling the one-dimensional spectral data to obtain a one-dimensional spectrum having data points equally spaced in wavelength;

[0070] - calibrating the one-dimensional spectrum relative to a reference white light spectrum stored in the memory to obtain a calibrated spectrum;

[0071] - filtering the calibrated spectrum using a low-pass filter to obtain a filtered spectrum;

[0072] - aligning the filtered spectrum in terms of wavelength with a reference spectrum of the tracer stored in the memory, thereby obtaining a pre-processed spectrum and storing it in the memory; and

[0073] - performing an operation of calculating the first residual and the second residual by using the pre-processed spectrum stored in the memory as the measured Raman spectrum.

[0074] In addition, the processing unit may be further configured to:

[0075] - defining a spectral measurement vector as the vector corresponding to the obtained pre-processed spectrum;

[0076] - defining a first spectral vector as the product of a first weight vector and a complete design matrix having columns respectively representing reference spectral data of the complete model, and determining respective non-negative components of the first weight vector, the first weight vector minimizing a first residual vector corresponding to a difference between the first spectral vector and the spectral measurement vector via a least squares method;

[0077] - defining a second spectral vector as a product of a second weight vector and a simplified design matrix having columns respectively representing reference spectral data of the simplified model, and determining respective non-negative components of the second weight vector, the second weight vector minimizing a second residual vector corresponding to a difference between the second spectral vector and the spectral measurement vector via a least squares method;

[0078] - calculating a first residual sum of squares RSS1 of errors corresponding to the first weight vector, the first weight vector having p1 non-negative components, and storing the calculated first residual sum of squares RSS1 and the number p1 in the memory;

[0079] - calculating a second residual sum of squares RSS2 of errors corresponding to the second weight vector, the second weight vector having p2 non-negative components, and storing the calculated second residual sum of squares RSS2 and the number p2 in the memory; and

[0080] - The F value is calculated as the ratio of the difference between the stored second residual sum of squares RSS2 and the stored first residual sum of squares RSS1 divided by the difference between the stored quantities p2 and p1, and the stored first residual sum of squares RSS1 divided by the difference between the number N of components of the spectral measurement vector and the number p1, i.e. F = ((RSS2-RSS1) / (p1-p2)) / (RSS1 / (N-p1)).

[0081] The processing unit may be further configured to determine the non-negative components of the first weight vector and the second weight vector by:

[0082] - expressing the first weight vector that minimizes the first residual vector as the product of the pseudo-inverse matrix of the complete design matrix and the spectral measurement vector;

[0083] - expressing the second weight vector that minimizes the second residual vector as the product of the pseudo-inverse matrix of the simplified design matrix and the spectral measurement vector; and

[0084] In the case where a component of the first weight vector or the second weight vector, respectively, has a negative value:

[0085] - modifying the full design matrix or the reduced design matrix accordingly by removing spectral vectors corresponding to negative components from the full design matrix or the reduced design matrix;

[0086] - setting the negative value components to zero; and

[0087] - recalculating the pseudo-inverse matrix of the modified complete design matrix or the modified reduced design matrix accordingly until the obtained components of the first weight vector and the second weight vector have only non-negative values, and storing the obtained components in the memory.

[0088] The present invention will be described more fully hereinafter with reference to the accompanying drawings, which illustrate salient aspects and features of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 The relative lifetimes of Raman and photoluminescence (including fluorescence) signals are illustrated (not to scale).

[0090] Figure 2 The Raman spectrum of the SERS tracer is illustrated to show the effect of the intensity enhancement of the Raman scattered light due to the structure of the SERS particle.

[0091] Figure 3 Raman spectra of SERRS tracers are shown.

[0092] Figure 4 is a flow chart illustrating an embodiment of a method according to the present invention.

[0093] Figure 5 is a flowchart illustrating a non-negativity constraint method according to the present invention.

[0094] Figure 6 is a schematic diagram of a system including a Raman spectrometer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0095] In order to overcome the above-mentioned shortcomings of the prior art and to detect the presence of SERS or SERRS tracers in a tag applied to a substrate to be authenticated, and also to reliably quantify the amount of signal from the SERS / SERRS tracer fingerprint (i.e., a very specific peak on their Raman spectrum) within the raw spectral data from the tag measured using a Raman spectrometer, the method according to the invention compares the measured spectral data from the test tag with reference Raman spectral models of the various individual materials forming the corresponding authentic tag and with reference Raman spectra of authentic substrates, and uses a robust quality model that can reliably determine whether a SERS / SERRS tracer has been identified in the tag. If the tracer in the tag is identified as authentic, the tag itself is considered authentic, and more generally, the security (applied to the substrate of the security) comprising the tag is considered authentic.

[0096] The additional / unwanted spectral information in the raw spectral data acquired by the Raman spectrometer is divided into two sub-spectral categories, respectively related to the aforementioned "known spectral data" and "unknown spectral data", to improve the signal-to-noise ratio (SNR) and provide a rapid and reliable detection of the presence of SERS / SERRS tracers on labels applied to securities compatible with high-speed sorting devices. The known spectral data is used to model the measured spectral information, while the unknown spectral data, which is expected to be "low-frequency" data, is modeled using only polynomials (e.g., Legendre polynomials, or Jacobi, Gegenbauer, Zernike, Chebyshev, Romanovski polynomials).

[0097] Raman spectroscopy and SERS spectroscopy (compared to cm -1 The scattered intensity is plotted in units of Raman shift) Figure 2 The above example illustrates the spectral enhancement effect caused by the structure of the SERS tracer itself, and Figure 3 The above illustrates the spectral enhancement effect due to the structure itself of an example of a SERRS tracer (where the Raman spectrum is scaled by a factor of 8 to better overlap with the SERRS spectrum). In both figures, the characteristic enhanced Raman scattering intensity peaks are clearly visible and are specific to the structure itself of the nanoparticles forming the tracer, making these peaks constitute identifying features (i.e., these peaks are the "fingerprint" of the tracer).

[0098] According to an exemplary embodiment of the present invention, the mark (pattern) to be authenticated is printed on the paper substrate of the banknote using several inks. In the case where the mark (and thus the banknote) is authentic, the components of each (authentic) ink are known, and a authentic SERS tracer with a known (reference) Raman spectrum has been added to one of these inks to be printed on the banknote. The ink including the SERS tracer corresponds to the first material mentioned above, and the second material mentioned above corresponds to (one or more than one) other inks. In this particular embodiment, there are four different inks (each with its specific components) in the mark, and each ink, if authentic and without a tracer, has a known (reference) Raman spectrum. When the authentic SERS tracer, the reference authentic paper substrate, and each of the four reference authentic inks are irradiated with excitation light (here laser) respectively, the reference Raman spectra of the authentic SERS tracer, the reference authentic paper substrate (of the corresponding authentic banknote), and each of the four reference authentic inks can be measured using a Raman spectrometer. These reference Raman spectra are then used to derive a complete model of the Raman spectrum of a universal true marker as a linear combination of different reference spectra. Each reference spectrum corresponds to a certain number of scattered light intensity values ​​at different wavelengths acquired via a Raman spectrometer. Thus, for each of the above-mentioned true SERS tracers, the reference true paper substrate, and the four reference true inks, an interpolated reference spectral curve can be obtained that gives the measured scattered intensity I (i.e., I(λ)) as a function of wavelength λ. For simplicity, we assume that the same number n (e.g., n=1024) of reference intensity values ​​(corresponding to n different wavelength values) are extracted from each reference spectral curve.

[0099] In a complete model of the Raman spectrum of a generic genuine mark (with four genuine inks) applied to a (genuine) substrate, the (discrete) representation of the spectral curve comprises n Raman intensity values ​​I (taken along the spectral curve) i (i=1,…,n,), and each intensity value I i Modeled as the (p1-1) reference Raman intensity value X i2 ,…,X ip1 A linear combination of X i1 ≡1, where i=1,…,n), where p1=7 in the specific embodiment (p1 is the number of independent variables in the model). Therefore, we have:

[0100] I i =β1X i1 +β2X i2 +…+β7X i7 , where β1,…,β7 are weights, and:

[0101] -X i2(i = 1, ..., n) are the n intensity values ​​at selected representative points along the (normalized) reference Raman spectrum of the true SERS tracer. The selected points are in a wavelength band of approximately 150 nm width within the NIR range (near infrared, 750 to 1400 nm). The normalization of the spectral curve is obtained by taking the difference between the measured value and the minimum of the measured values ​​and by setting the highest peak value to 1000 to remove the offset value (the data is usually not centered on 0 on the vertical axis).

[0102] -X i3 (i=1, ..., n) are n intensity values ​​at selected representative points along a (normalized) reference Raman spectrum of a reference real paper; and

[0103] -X i4 ,…,X i7 (i=1,…,n) are n intensity values ​​at selected representative points along the (normalized) reference spectrum of each of the four reference authentic inks used to print the authentic mark (each of the four reference authentic inks considered alone (i.e., without including the SERS tracer)).

[0104] In vector notation, a vector I can be represented by n scalar components I i (i=1,…,n); the vector β can be associated with p1 (here p1=7) scalar weights β1,β2,…,β p1 associated, and the (n×p1) matrix X can be associated with the full model in which the first column includes n values ​​X i1 = 1 (i = 1, ..., n), and the second column to the p1th column are respectively composed of components X i2 (i=1,…,n),…,X ip1 (i=1,…,n) is formed.

[0105] Therefore, the representation of the Raman spectrum in the complete model is: I = Xβ.

[0106] According to the present invention, a "simplified" marking is a marking applied to a (real) paper substrate which differs from the real marking only by the fact that it does not comprise a (real) SERS tracer. In a simplified model of the Raman spectrum of such a simplified marking, we thus have n Raman intensity values ​​J obtained on the spectral curve i (i=1,…,n), these n Raman intensity values ​​are modeled as (p2-1) reference Raman intensity values ​​Z i2 ,…,Z ip2 A linear combination of (as Z i1 ≡1, where i=1,…,n), where p2=6. So we have:

[0107] J i =μ1Zi1 +μ2Z i2 +…+μ6Z i6 , where μ1,…,μ6 are weights, and:

[0108] -Z i2 (i=1, ..., n) are n intensity values ​​at selected representative points along a (normalized) reference Raman spectrum of a reference real paper; and

[0109] -Z i3 ,…,Z i6 (i=1,...,n) are the respective n intensity values ​​at selected representative points along the four (normalized) reference spectra of the respective four inks used to print the simplified mark (excluding the SERS tracer, of course).

[0110] In fact, by defining the simplified model, we have here (for i=1,…,n): Z i1 =X i1 =1, and Z ik =X i(k+1) , where k = 2,…,p2.

[0111] In vector notation, a vector J can be represented by n scalar components J i (i=1,…,n); the vector μ can be associated with p2 (here, p2=6) scalar weights μ1,μ2,…,μ p2 and the (n×p2) matrix Z can be associated with a simplified model in which the first column includes n values ​​Z i1 =1(i=1,…,n), and the second column to the p1th column are respectively composed of components Z i2 (i=1,…,n),…,Z ip2 (i=1,…,n) is formed.

[0112] Therefore, the representation of the Raman spectrum in the simplified model is: J=Zμ.

[0113] The mark on the banknote to be authenticated is illuminated with laser excitation light and the corresponding Raman light signal scattered by the mark is measured using a Raman spectrometer to obtain a measured Raman spectrum of the mark. Preferably, a Raman spectrometer equipped with a multimode laser source (MML) is used. In fact, even if it is common practice to use a single-mode fiber (SML) source to obtain the best possible resolution, experience has shown that using an MML source actually increases the detection speed. For example, the laser power can be increased by a factor of 10 relative to an SML source (without any compromise), while the integration time of the measurement is reduced by a factor of 10 (for example, we can achieve 0.2ms instead of 2ms). This is due to two main differences between SML and MML sources: the laser power (for example, the SML is about 100mW at 760nm when the MML is much higher (for example, about 1W)) and the line width (the SML is about 0.02nm when the MML is 0.08nm).

[0114] The measured Raman spectrum gives the measured (Raman) scattered light intensity Y as a function of the scattered light wavelength λ, i.e., Y(λ). The Raman spectrometer has multiple spectral channels, and the Raman light signal scattered by the marker and collected by the spectrometer is first dispersed in these spectral channels (via a grating), and the imaging unit (CCD) acquires a two-dimensional digital image of the corresponding dispersed spectral data as a two-dimensional array of intensity values ​​vs. wavelength (i.e., two-dimensional spectral data). Since the two-dimensional spectral data acquired from the Raman spectrometer is raw, the two-dimensional spectral data is further pre-processed, primarily to reduce the amount of data subsequently analyzed by the processing unit (to reduce processing time and be compatible with banknote detection in high-speed sorting machines), improve the signal-to-noise ratio (SNR), and accurately locate the Raman bands of the tracer fingerprint.

[0115] The pre-processing step of the two-dimensional digital image acquired by the imaging unit is performed by a processing unit equipped with a memory and includes the following operations:

[0116] 1) The acquired two-dimensional spectral data is converted into one-dimensional spectral data by linear binning and conversion of the binned data to wavelength data. This transformation significantly reduces the amount of data to be processed and improves the signal-to-noise ratio (SNR) (noise typically decreases by a factor of the square root of the number of pixels in a column of a two-dimensional digital image).

[0117] 2) The obtained one-dimensional spectral data is resampled to form a one-dimensional spectrum with data points that are equally spaced in wavelength. This operation is performed via spline or polynomial interpolation of the spectral data. This resampling has the advantage of reducing spectral compression along the horizontal axis and also provides linear resolution of the spectrum, which allows the use of well-known signal processing tools (low-pass filtering by FFT convolution, FIR convolution, etc.).

[0118] 3) Calibrate the resampled one-dimensional spectrum against a reference white light spectrum stored in memory (e.g., from a quartz tungsten-halogen lamp to balance the sensitivity of the Raman spectrometer) to obtain a calibrated (one-dimensional) spectrum. This operation makes it possible to balance the light intensity delivered by the Raman spectrometer (typically, a spectrometer outputs different values ​​for the same light intensity at different wavelengths).

[0119] 4) Filter the calibrated spectrum using a low-pass filter to obtain a filtered spectrum. In practice, unwanted very high-frequency noise in spectral data is primarily due to the imaging unit (i.e., its image sensor and its circuitry) and is known to be a measurement artifact. This filtering can be performed using various methods, such as a moving average filter, an FFT (Fast Fourier Transform) filter, or a Savitzky-Golay filter. Preferably, FFT filtering is used (since this method can also be used for spectral alignment).

[0120] 5) Align the filtered spectrum in wavelength with the reference spectrum of the true tracer stored in memory. In practice, due to many possible reasons (e.g., expansion of the spectrometer, temperature changes affecting the wavelength of the light source and / or the grating, mechanical disturbances due to vibrations, etc.), the stored reference Raman spectrum of the true tracer is usually not aligned with the Raman spectrum measured from the marker. Therefore, in order to have the best possible verification of the tracer fingerprint, the measured Raman spectrum obtained from the marker is aligned in wavelength with the reference spectrum. This alignment can be achieved by different methods, such as the following:

[0121] - by running the algorithm at different shift increments and picking the best position on the wavelength axis;

[0122] - by running the algorithm at different shift increments and interpolating to find the optimal position on the wavelength axis;

[0123] - preferably by performing convolution with the tracer fingerprint in the frequency domain;

[0124] - By monitoring the position of the light source during measurements with markers.

[0125] As a result of the above operation, a preprocessed Raman spectrum is obtained from the two-dimensional spectrum data acquired by the imaging unit.

[0126] The (discrete) representation of the spectral curve Y(λ) of the pre-processed Raman spectrum comprises n (pre-processed) Raman intensity values ​​Y (obtained along the spectral curve) i (i=1,…,n), and the (n-dimensional) vector Y can be combined with n scalar components Y i ,i=1,…,n are associated.

[0127] To fit the (pre-processed) Raman spectra to the complete model, the spectral (measurement) vector Y is decomposed into Y = I + ε (linear regression analysis) with a first spectral vector I = Xβ, where X is the n × p1 (design) matrix of the complete model, β is the corresponding first weight vector and ε is a vector with components ε i The component β of the first weight vector β that minimizes the error vector ε can be determined by various known optimization methods. k (k=1,…,p1; here, p1=7). For example, the residual vector can be calculated (iteratively) for a number of selected values ​​of the components of the vector β, and the vector β corresponding to the residual vector with the lower norm can be selected. Another approach is to use a well-known optimization algorithm, such as Dantzig's simplex algorithm. Preferably, we use the least squares residual (LSR) method, which has the advantage of being less intensive in terms of CPU computation and is therefore more suitable for authenticating marks on banknotes in high-speed sorting machines.

[0128] In the same way, to fit the (preprocessed) Raman spectrum to the simplified model, the vector Y is decomposed into Y = J + ε' with a second spectral component J = Zμ, where Z is the n × p2 (design) matrix of the simplified model, μ is the corresponding second weight vector, and ε' is the vector with component ε' i The component μ of the second weight vector μ that minimizes the error vector ε′ can be determined by the least squares residual (LSR) method. m (m=1,…,p2; here, p2=(p1-1)=6).

[0129] According to the LSR method, the least squares parameter estimates of β (or μ for the simplified model) of the full model taking into account the measured values ​​Y are obtained from the following p1 (or p2) normal equation:

[0130] ε i =Y i -β1X i1 +β2X i2 +…+β p1 X ip1 (i=1,…,n), that is, ε=Y-Xβ, and

[0131] j=1,…,p1;

[0132] Or respectively,

[0133] ε' i =Y i -μ1Z i1 +μ2Z i2 +…+μp2 Z ip2 (i=1,…,n), that is, ε'=Y-Zμ, and

[0134] j=1,…,p2.

[0135] The LSR method provides a solution that minimizes the squared residual, which is And for the simplified model If we assume that the columns of the design matrix for the full model X are linearly independent, we can use X + =(X T X) -1 X T The (left) pseudo-inverse solution X of the design matrix X using the full model + , where X T is the transpose of X and is denoted by (and we get X + X = Id). If the rows of the design matrix of the full model X are linearly independent, we can use X + =X T (XX T ) -1 Use the (right) pseudo-inverse solution X of the model X + , and is still recorded as (and we get XX + =Id). In practice, we use the singular value decomposition (SVD) method for computing the pseudo-inverse solution of the design matrix to have a stable and fast calculation. In the same way, we compute the pseudo-inverse solution Z of the design matrix of the simplified model Z + And recorded as These pseudo-inverse matrices are preferably pre-computed (once the corresponding design matrices are known) and stored in the memory of the processing unit. Once the first weight vector is determined and the second weight vector The statistical significance of these estimated weights for the two models (given the same measurement vector Y) can be checked by performing a classic F-test, i.e. the quality of the full model relative to the reduced model.

[0136] However, the problem with the above LSR method is that the LSR method does not consider whether the solution obtained is "feasible". In fact, if the solution involves the weight components β of the vector β j (j∈{2,…,7}) or the weight component μ of the vector μ rIf the value of (r∈{2,…,6}) is negative, the intensity of the associated spectral component should be negative, which is physically impossible (this would constitute an infeasible solution). It has been observed that the authentication method is more robust than when using a specific minimization method to comply with the non-negativity constraints (NNC) on the values ​​of the weights. Some methods that integrate said non-negativity constraints are known: for example, the active set method (described in detail in the book "Solving Least Square Problems" by Charles L. Lawson and Richard J. Hanson, SIAM 1995) or the gradient descent method of Landweber. According to the present invention, the LSR method is combined with Figure 5 The following methods are combined to meet the non-negativity constraint. This will be explained in the case of the full model with p1 = 7 weights and can be directly transposed to the case of the simplified model (where p2 = 6 weights) after necessary modifications. The weight vector is calculated The value of the p1 components β1,…,β7 is obtained by (S1) according to the pseudo-inverse matrix X + and the spectrum measurement vector Y stored in the memory of the processing unit (ie, using ) and checks (S2) to determine whether the weight vector Are there any negative weight values ​​in ? Figure 5 In the example shown, the two weights β2 and β6 have negative values ​​(corresponding to the SERS tracer and the third ink, respectively), and then the value of the weight β2 is set to zero (S3), and the corresponding column of the design matrix X (i.e., the column corresponding to the Raman spectrum of the (true) SERS tracer) (with the component X 12 ,…,X n2 ) is removed (S4) from the (initial) design matrix X, and thus a new n×(p1-1) design matrix X' is obtained. Then, the corresponding new pseudo-inverse matrix X' is calculated (S5) + , and the new pseudo-inverse matrix X' + Used to calculate (S6) the new weight vector in This new weight vector has only (p1-1) components β'1, β'3, β'4, β'5, β'6 and β'7 (because we set β2 to 0). A check (S7) is then performed to determine whether the calculated weight vector Whether there are any negative weight values ​​(Yes "Y") or no weight values ​​(No "N"). Figure 5 In the example shown, one weight β'6 has a negative value (corresponding to the third ink), the value of the weight β'6 is set to 0 (S8), and the corresponding column of the design matrix X' (i.e., the column corresponding to the Raman spectrum of the (reference true) third ink) (with component X16 ,…,X n6 ) is removed (S9) from the design matrix X, and thus a new n×(p1-2) design matrix X" is obtained. Then the corresponding new pseudo-inverse matrix X" is calculated (S10). + , and the new pseudo-inverse matrix X' + Used to calculate (S11) the new weight vector in This new weight vector has only (p1-2) components β'1, β'3, β'4, β'5 and β'7 (because we set β2 and β'6 to 0). A check (S12) is then performed to determine whether the calculated weight vector Whether there are any negative weight values ​​(yes "Y") or no negative weight values ​​(no "N"). Figure 5 In the example shown, the remaining values ​​of the weight components β"1, β"3, β"4, β"5, and β"7 are all positive. As a result (S13), the final p1 values ​​of the weight components obtained via the LSR method under the non-negativity constraint (i.e., the LSR-NNC method) are β"1, 0, β"3, β"4, β"5, 0, and β"7, and the calculation stops (S14). If there is a negative value (i.e., Y) at step S12, steps (S8) to (S12) are performed accordingly. The LSR-NNC method is also applied to the second weight vector Calculation of the (non-negative) values ​​of the components of .

[0137] In the first step of adding weight vector and the second weight vector Now that we have obtained reliable values ​​for the components of (i.e., non-negative values ​​corresponding to physically possible values), we can now perform a reliable F-test to compare the quality of the full model with respect to the reduced model. To do this, the F-value is calculated as the ratio of the obtained second residual sum of squares of the reduced model and the first residual sum of squares of the full model The difference between the two values ​​divided by the difference between the quantities p2 and p1 (p2-p1), and the first residual sum of squares RSS1 divided by the difference between the number n (here, n=1024) of components of the spectral measurement vector Y and the number p1, F=((RSS2-RSS1) / (p1-p2)) / (RSS1 / (n-p1)). Therefore, F=[(RSS2-RSS1) / RSS1]×K, where the factor K≡(n-p1) / (p1-p2). In the example considered, we get the same number of data points (i.e., n) for both models. The full model (model 1) has one more parameter relative to the simplified model (model 2). As always, the model with more parameters will always be able to fit the data at least as well as the model with fewer parameters, and the F-test will determine whether the full model will give a significantly better fit to the data than the simplified model (without tracer). From the classical formula above, we obtain the value of the factor K given by (n-p1) / (p1-p2)=(n-7) / 1=1017. The value of F is a number that represents the probability of having a true SERA tracer in the marker.

[0138] In general, the value of F depends on the SNR as follows:

[0139] - With low SNR and presence of SERS tracer in the label: low value F. This is normal, since random noise has the same influence on the discriminative fingerprint of the SERS tracer.

[0140] - With low SNR and no (true) SERS tracer present in the label: the value of F is low.

[0141] - With high SNR and the presence of a (true) SERS tracer in the label: the value of F is high.

[0142] - With high SNR and no (true) SERS tracer present in the label: the value of F is low.

[0143] In the case where the trend between SNR and the value F is linear, when the value of F is between 8000 and 1000000, it is not very suitable for deciding the authenticity of the mark. A further "compression" step can be applied to modify the value of F in order to create a plateau on the curve representing the dependence of the value F on SNR. In this embodiment, the modified ("compressed") value F' is obtained via the transformation F'=constant×Log(F) (for example, the constant factor has a value of 5).

[0144] Through a series of experiments, we can reliably conclude that:

[0145] A low threshold (LTV) value of -F' below about 20 (eg, between 1 and 20) corresponds to the absence of a (genuine) SERS tracer in the tag, and a negative decision is made indicating that the corresponding banknote is not authentic. - .

[0146] A high threshold (HTV) value of F' above about 50 (eg, between 50 and 80) indicates the presence of a SERS tracer in the tag, and a positive decision is made that the corresponding banknote is authentic. + .

[0147] - whereas intermediate values ​​of F' (e.g. between the low threshold LTV and the high threshold HTV) do not allow conclusions to be drawn (the result strongly depends on the level of SNR). In this latter case, since it cannot be decided whether a SERS tracer is present in the tag, it cannot be decided whether the banknote is authentic: the banknote is retained (R) for a more detailed (e.g. electronic forensic) analysis.

[0148] exist Figure 4 The steps of the above preferred embodiment of the method for authenticating a marking applied to a substrate and having a composition comprising ink and a SERS tracer (or SERRS tracer) are summarized above. The method starts at (M0) and specifies the values ​​p1 (where p1 ≥ 4) and p2 = (p1 - 1) of the number of reference Raman spectra in the complete model and the simplified model and stores these values ​​in the memory of the processing unit (M1), and takes a number n of points on the measured Raman spectrum. In step (M2), the corresponding Raman spectrum X of the complete model is specified. i2 ,…,X ip1 (i=1,…,n) and the corresponding Raman spectrum Z of the simplified model i2 ,…,Z ip2 , and store the corresponding complete design matrix X and simplified design matrix Z. In step (M3), calculate and store the corresponding pseudo-inverse X of the complete design matrix + and the pseudo-inverse Z of the simplified design matrix + Then, in step (M4), a measurement Raman spectrum is acquired from the mark (when the mark is irradiated with the excitation laser) via a two-dimensional image obtained by the imaging unit of the Raman spectrometer, and the measurement Raman spectrum is preprocessed to obtain a one-dimensional spectrum and form a corresponding spectral measurement vector Y having n components. In step (M5), the LSR method and the NNC method (i.e., LSR-NNC) are performed to calculate a first weight vector corresponding to the complete model and the second weight vector corresponding to the simplified model These two minimize the square of the first residual vector ε (i.e., Y-Xβ) of the complete model and the square of the second residual vector ε' (i.e., Y-Zμ) of the simplified model, respectively. Then, the first residual sum of squares is calculated and the second residual sum of squares And in step (M6) the corresponding F value is obtained, where F = K(RSS2-RSS1) / RSS1 (and K = (n-p1) / (p1-p2)). In step (M7), the compressed value F' is calculated (for example, using the transformation F' = 5Log(F)). Finally, a decision is made taking into account the compressed value F' and the storage values ​​HTV (high threshold) and LTV (low threshold) for convenience of marking:

[0149] - In step (M8), the value F' is compared with the value HTV: if F' is greater than HTV (condition c1), a positive decision D is made in step (M9) + , i.e. the banknote with the mark is genuine (and the calculation stops (M9')); if F' is less than or equal to HTV (condition c2), then,

[0150] - In step (M10), the value of F' is compared with the value LTV: if F' is less than LTV (condition c3), a negative decision D is made in step (M11) - , i.e. the banknote with the mark is not true (and the calculation stops at (M11')); if F' is greater than or equal to LTV (condition c4), the banknote is retained (R) for further analysis in step (M12) (and the calculation stops at (M12')).

[0151] In the case where the label comprises multiple SERS tracers or SERRS tracers, it may not be reliable enough to decide the authenticity based on a single F value alone. According to the present invention, multiple different simplified models can be used and different F values ​​can be calculated to compare the complete model of the true label (i.e., multiple reference spectra including various tracers) with each of the simplified models. For example, different simplified models can correspond to the following label, which differs from the true label only in that one of the different tracers of the true label is not present. These F values ​​are obtained from the (pre-processed) spectral vector Y, which is obtained from the measured Raman spectrum of the label to be authenticated by applying the above-mentioned LSR-NNC method to find different weight vectors that minimize the square of the corresponding residual vector. The decision about the authenticity of the label must involve different threshold rules for each of the calculated F values, which leads to a certain complexity. In this case, the decision about the authenticity can preferably be based on a decision tree containing said threshold rules.

[0152] The present invention also relates to a system (60) (specific embodiments of which are Figure 6As shown in FIG, the system (60) comprises a light source (61), a Raman spectrometer (62), an imaging unit (63), a processing unit (64), a memory unit (65) and a control unit (66). When the (moving) mark reaches the height of the imaging unit (65), the control unit (66) controls the light source (61) (here a laser) via a current loop to deliver calibrated excitation light and illuminate the mark (67) on the banknote (68) to be authenticated. The laser excitation light is delivered to the mark (67) via a dichroic mirror (69). In response to the illumination, the Raman light is scattered from the mark, collected via the dichroic mirror (69), and dispersed toward the CCD sensor (71) of the imaging unit (63) via a grating (70). A corresponding two-dimensional digital image of the collected Raman spectrum is formed by the imaging unit (63) and constitutes a 2D measured Raman spectrum stored in the memory unit (65). As described above, the memory unit (65) stores the complete model of the genuine mark (applied on the genuine substrate of the genuine banknote), that is, the number n of points of the selected reference spectrum, the number p1 of weights forming the first weight vector β, and the number p2 of weights forming the second weight vector μ. The reference spectrum of the complete model is stored as a component of the complete (design) matrix X, and the reference spectrum of the simplified model is stored as a component of the simplified (design) matrix Z. The memory unit (65) also stores the simplified model, the pre-calculated pseudo-inverse X of the matrix X and the matrix Z, respectively. + and Z + As described above, the stored two-dimensional measured Raman spectrum is pre-processed via the processing unit (64) to obtain a (one-dimensional) pre-processed spectrum in the form of a spectral measurement vector Y with n components stored in the memory unit (65). The processing unit (64) then calculates a first weight vector corresponding to the complete model (which minimizes the square of the first residual vector ε=Y-Xβ), and calculates the second weight vector (which minimizes the square of the second residual vector ε'=Y-Zμ), and the calculated residual vector is stored in the memory unit (65). The memory unit also stores the values ​​of the parameters HTV and LTV corresponding to the high threshold and the low threshold to be considered in the F test for the full model and the reduced model, respectively. The first residual sum of squares associated with the full model and the second model, respectively, is calculated by the processing unit (64). and the second residual sum of squares And the corresponding F value of the F test is further calculated by the processing unit (64) as F=K(RSS2-RSS1) / RSS1, where K=(n-p1) / (p1-p2). The processing unit (64) calculates the compressed F' value as F'=5Log(F) and stores the value in the memory unit (65). The processing unit (64) makes a final decision (preferably displayed on the screen) based on the stored value F' and the stored values ​​of the parameters HTV and LTV when comparing the value with HTV and LTV (as described above). If the flag is considered not true (with a negative decision D - ), the corresponding banknote is retained as counterfeit. If F' ≥ LTV, the banknote is considered suspicious and retained for further (electronic forensic) analysis.

[0153] The above disclosed subject matter is to be considered illustrative rather than restrictive and is intended to provide a better understanding of the present invention as defined by the independent claims.

Claims

1. A method for authenticating a marking, the marking being applied to a substrate and having a composition comprising a first material, the first material comprising a SERS tracer or a SERRS tracer, the method being characterized in that it comprises the following steps performed by a system comprising a light source, a Raman spectrometer, an imaging unit, and a control unit having a processing unit and a memory, wherein: The light source is controlled by the control unit via a current loop to deliver calibrated excitation light: storing in the memory a complete model of the Raman spectrum of a true marker applied on a true substrate and having a component including a true first material as a first weighted sum of: a reference Raman spectrum of the true tracer collected when the true tracer, a reference true substrate, and a reference true first material are irradiated with excitation light, respectively; a reference Raman spectrum of a reference true substrate not marked with the true tracer; and a reference Raman spectrum of a reference true first material not including the true tracer, the true first material including a true SERS tracer or a true SERRS tracer; storing in the memory a simplified model of the Raman spectrum of a simplified marker as a second weighted sum of a reference Raman spectrum of the reference true substrate and a reference Raman spectrum of the reference true first material, the simplified marker differing from the true marker only in that a component of the simplified marker does not include the true tracer; When the marker is illuminated with the excitation light, measuring a corresponding Raman light signal scattered by the marker via a Raman spectrometer to obtain a measured Raman spectrum of the marker; By the processing unit: fitting the measured Raman spectrum to a complete model of the Raman spectrum by calculating values ​​of weights in the complete model that minimize a difference between the complete model and the measured Raman spectrum subject to the non-negativity constraint of the weights and obtaining corresponding first residuals; fitting the measured Raman spectrum to the simplified model of the Raman spectrum by calculating the following values ​​of weights in the simplified model, which minimize the difference between the simplified model and the measured Raman spectrum under the non-negativity constraint of the weights, and obtaining corresponding second residuals; calculating an F value corresponding to an F test for comparing a complete model with a simplified model for the measured Raman spectrum based on the obtained first residual and second residual; as well as The presence or absence of the tracer in the label is determined based on the calculated F value.

2. The method according to claim 1, wherein During operation of measuring a Raman light signal scattered by the tag, the tag is moving relative to the Raman spectrometer.

3. The method according to claim 1 or 2, wherein The labeled component includes a second material, and each weighted sum of the full model and the simplified model further includes a reference spectrum of the true second material with a corresponding weight, collected when the corresponding true second material is irradiated with the excitation light.

4. The method according to claim 1 or 2, wherein: The Raman spectrometer has a plurality of spectral channels, and the operation of measuring the Raman light signal scattered by the marker includes: Dispersing the collected Raman light in the plurality of spectral channels, and acquiring a two-dimensional digital image of the dispersed spectral data using the imaging unit; The acquired two-dimensional digital image is preprocessed by performing the following operations using the processing unit: transforming the two-dimensional spectral data into one-dimensional spectral data via line binning and conversion of the binned data into wavelength data; resampling the one-dimensional spectral data to obtain a one-dimensional spectrum having data points equally spaced in wavelength; calibrating the one-dimensional spectrum relative to a reference white light spectrum stored in the memory to obtain a calibrated spectrum; filtering the calibrated spectrum using a low-pass filter to obtain a filtered spectrum; and aligning the filtered spectrum in wavelength with a reference spectrum of a tracer stored in the memory, thereby obtaining a pre-processed spectrum; and The operation of calculating the first residual and the second residual is performed by using the preprocessed spectrum as the measured Raman spectrum.

5. The method according to claim 4, comprising: The spectral measurement vector is defined as the vector corresponding to the obtained preprocessed spectrum; defining a first spectral vector as a product of a first weight vector and a complete design matrix having columns respectively representing reference spectral data of the complete model, and determining respective non-negative components of the first weight vector that minimizes a first residual vector corresponding to a difference between the first spectral vector and the spectral measurement vector via a least squares method; defining a second spectral vector as a product of a second weight vector and a simplified design matrix having columns respectively representing reference spectral data of the simplified model, and determining respective non-negative components of the second weight vector that minimizes a second residual vector corresponding to a difference between the second spectral vector and the spectral measurement vector via a least squares method; Calculating a first residual sum of squares RSS1 of errors corresponding to the first weight vector, wherein the first weight vector has p1 non-negative components; Calculating a second residual sum of squares RSS2 of errors corresponding to the second weight vector, the second weight vector having p2 non-negative components; as well as The F value is calculated as the ratio of the difference between the second residual sum of squares RSS2 and the first residual sum of squares RSS1 divided by the difference between the numbers p2 and p1, and the first residual sum of squares RSS1 divided by the difference between the number N of components of the spectral measurement vector and the number p1, that is, F = ((RSS2-RSS1) / (p1-p2)) / (RSS1 / (N-p1)).

6. The method according to claim 5, wherein: Determining each non-negative component of the first weight vector and each non-negative component of the second weight vector includes: expressing the first weight vector that minimizes the first residual vector as a product of a pseudo-inverse matrix of the complete design matrix and the spectral measurement vector, and expressing the second weight vector that minimizes the second residual vector as a product of a pseudo-inverse matrix of the simplified design matrix and the spectral measurement vector; and In case a component of the first weight vector or the second weight vector respectively has a negative value: modifying the full design matrix or the reduced design matrix accordingly by removing spectral vectors corresponding to negative components from the full design matrix or the reduced design matrix; Set negative components to zero; and The pseudo-inverse matrix of the modified complete design matrix or the modified reduced design matrix is ​​recalculated accordingly until the obtained components of the first weight vector and the second weight vector have only non-negative values.

7. A system for authenticating a label, the label being applied to a substrate and having a composition comprising a first material, the first material comprising a SERS tracer or a SERRS tracer, the system comprising a light source, a Raman spectrometer, an imaging unit, and a control unit having a processing unit and a memory, the light source being controlled by the control unit via a current loop to deliver calibrated excitation light, the system being configured to: irradiating the mark with the excitation light delivered by the light source controlled by the control unit; and collecting Raman light generated from the marker, dispersing the collected Raman light in the Raman spectrometer having a plurality of spectral channels, acquiring a two-dimensional digital image of corresponding spectral data using the imaging unit, and storing the acquired spectral data in the memory as a measured Raman spectrum of the marker; The system is characterized by: The memory stores a complete model of the Raman spectrum of a true marker applied on a true substrate and having a component including a true first material as a first weighted sum of: a reference Raman spectrum of the true tracer collected when the true tracer, a reference true substrate, and a reference true first material are irradiated with excitation light, respectively; a reference Raman spectrum of a reference true substrate not marked with the true tracer; and a reference Raman spectrum of a reference true first material not including the true tracer, the true first material including a true SERS tracer or a true SERRS tracer; The memory stores a simplified model of the Raman spectrum of a simplified marker as a second weighted sum of a reference Raman spectrum of the reference true substrate and a reference Raman spectrum of the reference true first material, the simplified marker differing from the true marker only in that a component of the simplified marker does not include the true tracer; and The system is further configured to perform the following operations via the processing unit: fitting the measured Raman spectrum stored in the memory to the stored complete model of the Raman spectrum by calculating the following values ​​of weights in the complete model, which minimize the difference between the complete model and the measured Raman spectrum under the non-negativity constraint of the weights, and obtaining corresponding first residuals and storing them in the memory; Fitting the measured Raman spectrum stored in the memory to the stored simplified model of the Raman spectrum by calculating the following values ​​of weights in the simplified model, minimizing the difference between the simplified model and the measured Raman spectrum under the non-negativity constraint of the weights, and obtaining corresponding second residuals and storing them in the memory; calculating an F value corresponding to an F test for comparing a complete model with a simplified model for the measured Raman spectrum based on the stored first residual and second residual, and storing the calculated value in the memory; as well as A decision is made based on the stored F value whether the tracer is present in the label, and a signal representing the result of the decision is delivered.

8. The system according to claim 7, wherein: During operation of measuring a Raman light signal scattered by the marker, the marker is moving relative to the Raman spectrometer, and the control unit synchronizes illumination of the marker with the light source and acquisition of the measured Raman spectrum via the Raman spectrometer and the imaging unit with the movement of the marker.

9. The system according to claim 7 or 8, wherein: The labeled component includes a second material, and each weighted sum of the complete model and the simplified model also includes a reference spectrum of the true second material with corresponding weights, which is collected when the corresponding true second material is irradiated with the excitation light and stored in the memory.

10. The system according to claim 7 or 8, wherein: The processing unit is configured to pre-process the stored two-dimensional digital image by: transforming the two-dimensional spectral data into one-dimensional spectral data via line binning and conversion of the binned data into wavelength data; resampling the one-dimensional spectral data to obtain a one-dimensional spectrum having data points equally spaced in wavelength; calibrating the one-dimensional spectrum relative to a reference white light spectrum stored in the memory to obtain a calibrated spectrum; filtering the calibrated spectrum using a low-pass filter to obtain a filtered spectrum; aligning the filtered spectrum in wavelength with a reference spectrum of the tracer stored in the memory, thereby obtaining a preprocessed spectrum and storing it in the memory; as well as The operation of calculating the first residual and the second residual is performed by using the preprocessed spectrum stored in the memory as the measured Raman spectrum.

11. The system according to claim 10, wherein: The processing unit is configured to: The spectral measurement vector is defined as the vector corresponding to the obtained preprocessed spectrum; defining a first spectral vector as a product of a first weight vector and a complete design matrix having columns respectively representing reference spectral data of the complete model, and determining respective non-negative components of the first weight vector that minimizes a first residual vector corresponding to a difference between the first spectral vector and the spectral measurement vector via a least squares method; defining a second spectral vector as a product of a second weight vector and a simplified design matrix having columns respectively representing reference spectral data of the simplified model, and determining respective non-negative components of the second weight vector that minimizes a second residual vector corresponding to a difference between the second spectral vector and the spectral measurement vector via a least squares method; Calculating a first residual sum of squares RSS1 of errors corresponding to the first weight vector, the first weight vector having p1 non-negative components, and storing the calculated first residual sum of squares RSS1 and the quantity p1 in the memory; Calculating a second residual sum of squares RSS2 of errors corresponding to the second weight vector, the second weight vector having p2 non-negative components, and storing the calculated second residual sum of squares RSS2 and the quantity p2 in the memory; and The F value is calculated as the ratio of the difference between the stored second residual sum of squares RSS2 and the stored first residual sum of squares RSS1 divided by the difference between the stored quantities p2 and p1, and the stored first residual sum of squares RSS1 divided by the difference between the number N of components of the spectral measurement vector and the number p1, F = ((RSS2-RSS1) / (p1-p2)) / (RSS1 / (N-p1)).

12. The system according to claim 11, wherein The processing unit is configured to determine each non-negative component of the first weight vector and each non-negative component of the second weight vector by: expressing the first weight vector that minimizes the first residual vector as a product of a pseudo-inverse matrix of the complete design matrix and the spectral measurement vector; expressing the second weight vector that minimizes the second residual vector as a product of a pseudo-inverse matrix of the simplified design matrix and the spectral measurement vector; as well as In case a component of the first weight vector or the second weight vector respectively has a negative value: modifying the full design matrix or the reduced design matrix accordingly by removing spectral vectors corresponding to negative components from the full design matrix or the reduced design matrix; Set negative components to zero; as well as The pseudo-inverse matrix of the modified complete design matrix or the modified reduced design matrix is ​​recalculated accordingly until the obtained components of the first weight vector and the second weight vector have only non-negative values, and the obtained components are stored in the memory.

Images