Spectroscopic multivariate analysis device and multivariate analysis method

Abstract

To solve the problem of prior art that when analyzing a substance whose components are unknown, it is not easy to appropriately determine a pure spectrum, which is an obstacle to improving the reliability of an MCR method.SOLUTION: A multivariable analyzer 1 comprises: a multivariable curve decomposition unit 10 for separating provisional f pieces of provisional pure spectra (k1, k2 through kf) from m measured spectra (a1, a2 through am) and calculating the provisional concentration value (c) of the provisional pure spectra and each of the provisional pure spectra; a provisional image generation unit 22 for generating, for each provisional pure spectrum (k), the distribution information of provisional concentration values (c) with respect to the positions of m measurement points, on the basis of the calculated provisional concentration values (c); and a grouping unit 24 for grouping the provisional f pieces of provisional concentration distribution image on the basis of the coincidence degree of images and determining the number (f') of multiple components of the sample on the basis of the number of groups.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a multivariate analysis device that can calculate the pure spectrum and concentration value of each component for each measurement point from the measured spectrum data set when measuring the spectroscopic spectrum of a substance containing multiple components at multiple measurement points, and in particular to a multivariate analysis device that can perform calculations based on the multivariate curve resolution (MCR) method. [Background technology]

[0002] When analyzing foreign matter attached to a product substrate, for example, if an infrared microscope is used to measure the absorption spectra at multiple points on the substrate containing the foreign matter, a multivariate analysis device can be used to calculate multiple pure spectra and their respective concentration values from the measured spectral data set, and these can be plotted at the positions of each measurement point to create a concentration distribution image (chemical image).In addition, an existing spectral library can be used to identify the component names corresponding to each pure spectrum, and the component names can be displayed together with the concentration distribution image.

[0003] One of the multivariate analysis methods is multivariate curve resolution (MCR), which has the advantage of being able to effectively separate the pure spectra of each component from the measured spectrum even if the sample contains a large number of components. Patent Document 1 discloses a technique that further improves the MCR method to improve the separation accuracy of pure spectral data when the number of components is large. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6080577 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in such a multivariate curve resolution (MCR) method, if the number of components is not set appropriately, an appropriate pure spectrum cannot be obtained, so the number of components must be set carefully.

[0006] Setting a large number of components can result in over-separation of spectra that represent important components, such as those with high concentrations, resulting in multiple spectra with lower concentration data than before separation. For example, a spectrum with a 70% contribution rate may be separated into an over-separated spectrum with a 40% contribution rate and an over-separated spectrum with a 30% contribution rate, causing other components with lower contribution rates to appear higher in the analysis results. Furthermore, setting a small number of components makes it impossible to detect all necessary components. When analyzing substances with unknown components, such as in the case of foreign matter analysis, it is not easy to set an appropriate number of components, which has been an obstacle to improving the reliability of the MCR calculation results.

[0007] The present invention has been made in consideration of the problems of the prior art, and its object is to provide a multivariate analysis device and a multivariate analysis method that can calculate an appropriate pure spectrum regardless of the number of components set in the multivariate curve resolution (MCR) method. [Means for solving the problem]

[0008] In order to achieve the above object, the multivariate analysis device according to the present invention comprises: A multivariate analysis device that performs multivariate curve resolution on a plurality of measurement spectra (a) obtained by spectroscopically measuring a large number of measurement points (m points) of a sample consisting of a plurality of components, a multivariate curve decomposition unit that separates a provisionally set number (f) of provisional pure spectra (k1, k2, ... kf) from the multiple measured spectra (a1, a2, ... am) on the assumption that the measured spectrum (a) is expressed as a linear combination (a1 = c11k1 + c12k2 + ... c1fkf) of multiple (f) pure spectra (k) multiplied by the concentration value (c) for each of the pure spectra, and calculates the provisional pure spectra (k1, k2, ... kf) and the provisional concentration value (c) for each of the provisional pure spectra; a provisional concentration distribution generating unit that generates distribution information of the provisional concentration values (c) with respect to the positions of the measurement points (m points) for each provisional pure spectrum (k) based on the calculated provisional concentration values (c); The method is characterized by comprising a grouping unit that groups a provisional number (f pieces) of provisional concentration distribution information generated for each provisional pure spectrum (k) based on the degree of agreement of the provisional concentration distribution information, and determines the number (f' pieces) of multiple components based on the number of groups.

[0009] Furthermore, the image processing device may include a concentration value determining unit that determines the concentration value for each component by adding up the provisional concentration values (c) of a plurality of pieces of provisional concentration distribution information that have been grouped in the grouping unit.

[0010] The image processing device may also include a concentration distribution generating unit that generates concentration distribution information for each component based on the concentration values determined by the concentration value determining unit, and an image output unit that forms a visualized image by superimposing the generated concentration distribution information.

[0011] The apparatus may further include a pure spectrum determination unit configured to determine, as a pure spectrum, a provisional pure spectrum having the highest contribution rate from among a plurality of provisional pure spectra corresponding to a plurality of provisional concentration distribution information grouped in the same group by the grouping unit, or to determine a pure spectrum by combining a plurality of provisional pure spectra corresponding to a plurality of provisional concentration distribution information grouped in the grouping unit.

[0012] The multivariate curve resolution unit may be further configured to calculate a pure spectrum and a concentration value based on the number (f') of the multiple components determined by the grouping unit.

[0013] The multivariate analysis method according to the present invention comprises: A multivariate analysis method for performing multivariate curve resolution on a plurality of measurement spectra (a) obtained by spectroscopically measuring a large number of measurement points (m points) of a sample consisting of multiple components, using a computer, comprising: a multivariate curve decomposition step of separating a provisionally set number (f) of tentative pure spectra (k1, k2, ... kf) from the multiple measured spectra (a1, a2, ... am) on the assumption that the measured spectrum (a) is expressed as a linear combination (a1 = c11k1 + c12k2 + ... c1fkf) of multiple (f) pure spectra (k) multiplied by the concentration value (c) for each of the pure spectra, and calculating the provisional pure spectra (k1, k2, ... kf) and the provisional concentration value (c) for each of the provisional pure spectra; a provisional concentration distribution generation step of generating distribution information of the provisional concentration values (c) with respect to the positions of the measurement points (m points) for each provisional pure spectrum (k) based on the calculated provisional concentration values (c); and a grouping step of grouping a provisional number (f pieces) of provisional concentration distribution information generated for each provisional pure spectrum (k) based on the degree of similarity of the provisional concentration distribution information, and determining the number (f' pieces) of multiple components based on the number of groups.

[0014] In the multivariate analysis device according to the present invention, the provisional concentration distribution generating unit is configured to generate concentration distribution information for each of the number (f) of provisional pure spectra (provisional components) provisionally calculated by the multivariate curve resolution method.The grouping unit is configured to group the provisional number (f) of provisional concentration distribution information based on the degree of agreement of the provisional concentration distribution information, and determine the original number (f') of multiple components based on the number of groups.

[0015] <Significance of grouping provisional concentration distribution> When conventional multivariate curve decomposition methods decompose measured spectral data into a number of provisional pure spectra greater than the actual number of components, this means that a pure spectrum that was supposed to represent one component has been decomposed into multiple provisional pure spectra. The inventors noticed that if too many provisional components are set, provisional concentration distributions created for each provisional pure spectrum will have provisional concentration distributions with similar or identical conditions. Then, by determining that provisional concentration distributions with similar or identical conditions represent the same component, the number of provisional components can be reduced, and as a result, an appropriate pure spectrum can be determined. [Effects of the Invention]

[0016] According to the multivariate analysis device and multivariate analysis method having the configuration of the present invention, it is possible to determine the pure spectrum appropriately, and to obtain concentration distribution information based on such an appropriate pure spectrum. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a diagram showing the configuration of a multivariate analysis apparatus according to an embodiment of the present invention. [Figure 2] FIG. 1 is a process flow diagram showing a multivariate analysis method according to an embodiment of the present invention. [Figure 3] This is an observation image of a powder sample consisting of three components attached to a substrate. [Figure 4] FIG. 4 is a diagram showing a concentration distribution image of provisional number (20) created using provisional concentration values obtained by subjecting the measured spectrum of the powder sample of FIG. 3 to MCR processing. [Figure 5] 5 is a diagram illustrating a grouping process for the provisional density distribution image of FIG. 4. FIG. [Figure 6] FIG. 10 is a diagram showing the results of the grouping process. [Figure 7] 4 is a chemical image of all components of the powder sample of FIG. 3 displayed using the provisional concentration values determined after the grouping process. [Figure 8]This figure shows a provisional concentration distribution image (20) created using provisional concentration values obtained by MCR processing of measurement spectra acquired under different measurement conditions for the powder sample in Figure 3. [Figure 9] 9A to 9C are diagrams illustrating a grouping process for the provisional density distribution image of FIG. 8. DETAILED DESCRIPTION OF THE INVENTION

[0018] A preferred embodiment of the present invention will now be described with reference to the drawings. Fig. 1 shows a schematic configuration of a multivariate analysis apparatus 1 according to one embodiment of the present invention and the configuration of its peripheral devices. The multivariate analysis apparatus 1 shown in the figure is composed of an arithmetic processing device such as a computer, and includes an MCR processing unit 10 and an image processing unit 20. Fig. 1 also shows, as peripheral devices, a spectrum measurement unit 2, an MCR condition input unit 4, an image output unit 30, a display unit 40, a database search unit 50, and a spectrum library 60.

[0019] The MCR processing unit 10 includes, as functional blocks, a setting unit 12 that sets a measured spectral matrix A based on a measured spectral data set from an external spectrum measurement unit 2, a setting unit 14 that sets an initial value of a provisional pure spectral matrix K, and a storage and calculation unit 16 that calculates convergent values of the provisional concentration matrix C and the provisional pure spectral matrix K by iterative calculation using constraint conditions. The calculated values of the provisional concentration matrix C and the provisional pure spectral matrix K are passed to an image processing unit 20.

[0020] The processing target of the MCR processing unit 10 is a measurement spectrum data set of a sample composed of a plurality of components measured by various analytical methods such as scattered light analysis, IR analysis, ultraviolet-visible light analysis, fluorescence analysis, etc. The data set includes a plurality of measurement spectrum data obtained by mapping and measuring a large number of measurement points of the sample. The external spectrum measurement unit 12 is, for example, a microscopic Raman spectrometer (a device that irradiates a sample with laser light and measures the spectrum of scattered light from the sample), a microscopic infrared device (a device that irradiates a sample with an infrared interference wave, detects the reflected light or transmitted light from the sample, and obtains an absorption spectrum by Fourier transform), a microscopic ultraviolet-visible-near infrared spectrometer (a device that irradiates a sample with ultraviolet, visible, and near infrared light and measures the absorption spectrum of the reflected light or transmitted light from the sample), etc., and may be linked with the MCR processing unit 10.

[0021] The image processing unit 20 includes, as functional blocks, a provisional image generation unit 22 (corresponding to the provisional concentration distribution generation unit of the present invention) that creates a concentration distribution image for each provisional component based on the numerical values of the provisional concentration matrix C and the numerical values of the provisional pure spectrum matrix K from the MCR processing unit 10, a grouping storage / operation unit 24 that further groups these provisional concentration distribution images, a determination unit 26 for the provisional concentration value, and a determination unit 28 for the pure spectrum.

[0022] <Overview of MCR Processing> The multivariate analysis method according to the present embodiment will be conceptually described using FIG. 2 and the determinant (1) described later. First, in step S12, the measurement spectrum matrix setting unit 12 sets a measurement spectrum matrix "A" having, as matrix components, m measurement spectrum sequences corresponding to m measurement points of the sample. If one measurement spectrum sequence is represented by "a", the measurement spectrum a is a column of n numerical values, that is, it is composed of spectrum values at n wavenumber points. And the measurement spectrum sequences of the m measurement points are represented by "a1, a2,... am", and the measurement spectrum matrix A has these as matrix components. The horizontal direction of the matrix A corresponds to the wavenumber direction of the spectrum, and the vertical direction corresponds to the measurement position.

[0023] Next, in step S14, the temporary pure spectral matrix setting unit 14 sets the initial value of a temporary pure spectral matrix "K," whose matrix elements are f pure spectral sequences corresponding to the f provisional components that have been provisionally set. If one temporary pure spectral sequence is represented by "k," the temporary pure spectrum k is composed of a sequence of n numerical values, i.e., n spectral values. The pure spectral sequences of the f provisional components are represented by "k1, k2, ... kf," and the temporary pure spectral matrix K has these as matrix elements. The vertical direction of matrix K corresponds to the number of components, and the horizontal direction corresponds to the spectral wavenumber direction.

[0024] In step S14, for example, the provisional pure spectral matrix setting unit 14 may perform a principal component analysis (PCA) process on the measured spectral matrix A to estimate the provisional pure spectral matrix K. The number (f) of provisional components contained in the sample can be determined by referring to the magnitude of the cumulative contribution ratio R in Patent Document 1, for example. Then, the provisional pure spectral matrix K can be set using the provisional principal component spectra adopted based on the cumulative contribution ratio R as initial values.

[0025] There is no particular limitation on how to set the initial value of the provisional pure spectrum. The initial value may be set by a known method. The MCR condition input unit 4 may be configured so that the user can specify the setting conditions for the initial value of the provisional pure spectrum.

[0026] For example, if the result of PCA processing on the measured spectrum data set indicates that there are three principal components, a larger number of components may be provisionally set, for example, 6 to 10. Here, if the provisional number of components f is set to 10, the 10 principal component spectra calculated by PCA will be used as the initial values of the provisional pure spectrum of this embodiment.

[0027] Alternatively, the maximum number of principal components that can be calculated by PCA processing (for example, 20) may be set as the number of components f of the tentative pure spectrum, and all principal component spectra from the preliminary analysis may be used as the initial values of the tentative pure spectrum.

[0028] Alternatively, without using PCA, spectra close to standard spectra of components that are expected to be included may be extracted from the measured spectrum data set as appropriate and used as the initial value of the provisional pure spectrum. In any case, in this embodiment, it is sufficient to set the number of components provisionally, and there is no problem even if the number of components is set larger than expected.

[0029] The provisional concentration matrix "C" has the concentration values of the provisional components contained at the measurement points as matrix elements. If the concentration value of one provisional component contained in the substance at one measurement point is represented by "c", the concentration value of each provisional component at the first measurement point is represented by "c11, c12, ... c1f", and the concentration value of each provisional component at the mth measurement point is represented by "cm1, cm2, ... cmf". The provisional concentration matrix C has these provisional concentration values as matrix elements. The horizontal direction of matrix C corresponds to the number of provisional components, and the vertical direction corresponds to the measurement position.

[0030] As shown in steps S16, S18, and S20, the multivariate curve resolution (MCR) method assumes that the measured spectrum a is expressed as a linear combination of multiple provisional pure spectra k multiplied by provisional concentration values c, and then uses the measured spectrum matrix A, provisional concentration matrix C, and provisional pure spectrum matrix K to calculate the

[0031]

number

[0032] If a substance is composed of f provisional components, each of which is distributed two-dimensionally or three-dimensionally within the substance, the spectroscopic spectrum of the substance at m measurement points can be expressed as a linear combination of provisional pure spectra weighted according to the concentration of the provisional components distributed there. This will be explained using the above-mentioned m sequence of measured spectra "a1, a2, ... am", f sequence of provisional pure spectra "k1, k2, ... kf", and the concentration values of the f provisional components at the m-th measurement point "cm1, cm2, ... cmf".

[0033] The measured spectrum sequence a1 at the first measurement point is a1=c11×k1+c12×k2+...+c1f×kf...(2) The provisional concentration value c11 is the concentration of the pure spectrum k1 of the first provisional component that constitutes the measured spectrum a1 at the first measurement point, and is a scalar quantity. The measured spectrum sequence a1 is expressed as a linear combination of f provisional pure spectrum sequences k1, k2, ... kf and f provisional concentration values c11, c12, ... c1f.

[0034] Similarly, the measured spectrum sequence a2 at the second measurement point is a2=c21×k1+c22×k2+...+c2f×kf...(3) The measured spectrum sequence am at the m-th measurement point is am=cm1×k1+cm2×k2+...+cmf×kf ···(4) It is expressed as:

[0035] Therefore, it can be seen that the relational expression A=CK in equation (1) is a matrix representation of the above equations (2) to (4). The temporary concentration matrix / temporary pure spectrum matrix storage / calculation unit 16 in Fig. 1 is a means for separating matrices C and K from matrix A, and here, the well-known multivariate curve resolution (MCR) method can be adopted (for example, the method described in Patent Document 1).

[0036] In the provisional loop processing (S16 to S20), first, in step S16, a provisional concentration matrix C is calculated from matrices A and K. The relationship between the measured spectrum matrix A, provisional concentration matrix C, and provisional pure spectrum matrix K is expressed by the following equation, which is an extension of the Beer-Lambert equation to multiple components and multiple wavelengths (multiple wavenumbers). A=CK, AK T =C(KK T ) is derived, and further, AK T (K.K. T ) -1 =CKK T (K.K. T ) -1 Therefore, the provisional concentration matrix C is expressed by the following equation (5). C=AK T (K.K. T ) -1 ···(5) In step S16, the provisional concentration matrix C is calculated from equation (5). At this time, if the provisional concentration matrix C has a negative element, a constraint is applied that the negative element is replaced with 0.

[0037] Next, in step S18, the provisional pure spectral matrix K is temporarily set to blank, and the provisional pure spectral matrix K is calculated using the matrices C and A according to the following equation (6). K=(C T C) -1 C T A (6) Here, if K has negative elements, a constraint is applied that replaces the negative elements with 0.

[0038] The above-described process (S16) of blanking out the provisional concentration matrix C, determining the provisional concentration matrix C using the provisional pure spectral matrix K and the measured spectral matrix A, and applying the constraint conditions, and the process (S18) of blanking out the pure spectral matrix K, determining the provisional pure spectral matrix K using the provisional concentration matrix C and the measured spectral matrix A, and applying the constraint conditions, are repeated until the termination condition (S20) is satisfied. As a result, the converged values of the provisional concentration matrix C and the provisional pure spectral matrix K are obtained.

[0039] The termination condition is preferably a condition that sufficiently reduces fluctuations in the provisional concentration matrix C and the provisional pure spectral matrix K. For example, the termination condition for the MCR process may be that the processes from S16 to S20 are repeated a predetermined number of times (for example, 100 times).

[0040] <Image processing> Next, in step S22, the provisional image generator 22 uses the number f of provisional pure spectral sequences from the MCR process as the "number of provisional components" and creates a provisional concentration distribution image for each provisional component based on the provisional concentration value c, which is a component of the provisional concentration matrix C. Here, the provisional concentration value cmf indicates the concentration of the fth provisional component contained in the material at the mth measurement point. In other words, since each column of the provisional concentration matrix C indicates the concentration distribution of each provisional component, the provisional concentration matrix C can be used to display the concentration distribution of each provisional component on a monitor in an easy-to-understand manner.

[0041] Here, for convenience, the provisional concentration distribution is referred to as an image, but in computer processing, it is not necessary to visualize the provisional concentration distribution as an image. Therefore, instead of a provisional concentration distribution image, a "sequence of provisional concentration values" arranged so that the distribution state of provisional concentration values at each measurement point can be set as provisional concentration distribution information. If it is desired to visualize the provisional concentration distribution image, individual provisional chemical images, numbered f, where f is the number of provisional components, can be created. In this case, the concentration distribution of the provisional components can be visualized using a saturation (shade) display, color display (RGB), contour display, or the like based on the provisional concentration value c.

[0042] The procedure will be explained in detail using the powder sample shown in Figure 3. This powder sample consists of three-component powders (PS, PMMA, and ODS) attached to a Kbr substrate. Ten thousand measurement points were set within a measurement range of 100 μm in length and width, and the Raman scattered light from each was mapped and measured to obtain a Raman spectrum dataset. The MCR process described above was then performed to obtain a provisional concentration matrix C. Figure 4 shows provisional concentration distribution images of 20 components based on the provisional concentration matrix C. Provisional concentration distribution images 1 to 20 in Figure 4 are ranked by the contribution rate of the provisional pure spectra. If the number of provisional pure spectra set, f, were set to "3" based on the contribution rate, only three provisional concentration distribution images would be generated, potentially overlooking the fourth ODS component.

[0043] Therefore, as shown in FIG. 4, 20 provisional pure spectra, which is considerably more than the actual number of components, are set, and the provisional image generating unit 22 is made to obtain 20 provisional concentration distribution images.

[0044] Next, in step S24, the grouping storage and calculation unit 24 divides the 20 provisional concentration distribution images into groups based on the image similarity (e.g., correlation coefficient), and determines the number of components (f') contained in the sample based on the number of groups. A known image matching method can be applied to the image similarity. In this embodiment, the normalized cross-correlation (NCC) method is used as the evaluation function for the similarity. In the NCC method, the provisional concentration distribution images are treated as vectors, and the correlation coefficient is calculated by calculating the inner product. Therefore, the closer the directions of the two vectors are, the larger the correlation coefficient becomes, with a maximum of 1 and a minimum of 0. A correlation coefficient of 0.3 or higher can be said to indicate a certain degree of correlation. Other known image matching methods include SSD (a method for evaluating the sum of squared differences), SAD (a method for evaluating the sum of absolute values of differences), and ZNCC (a method for calculating normalized cross-correlation after subtracting the average value). These may be used appropriately by setting a threshold value according to each evaluation function.

[0045] A specific example of grouping is shown using Figure 5. First, image 1 with the highest contribution rate to the tentative pure spectrum is extracted. Then, the correlation coefficient between extracted image 1 and the remaining images is calculated. Images with correlation coefficients equal to or greater than a threshold value (e.g., 0.3) are extracted. The extracted images are set as the first group. In this example, six images (images 1, 7, 9, 12, 15, and 18) are extracted as the first group.

[0046] Next, Image 2, which had the highest contribution rate of the provisional pure spectrum, was extracted from the remaining 14 images. Similarly, the correlation coefficient between Image 2 and the remaining images was calculated. In this case, there were no images with a correlation coefficient of 0.3 or higher, so only Image 2 was placed in the second group.

[0047] Next, image 3, which has the highest contribution rate of the provisional pure spectrum, is extracted from the remaining 13 images. Similarly, the correlation coefficient between image 3 and the remaining images is calculated, and images with a correlation coefficient of 0.3 or higher are extracted and set as the third group. In this example, five images (images 3, 5, 6, 8, and 16) are extracted as the third group.

[0048] Similarly, image 4, which has the highest contribution rate of the provisional pure spectrum, is extracted from the remaining eight images. Similarly, the correlation coefficient between image 4 and the remaining images is calculated. In this case, there were no images with a correlation coefficient of 0.3 or higher, so only image 4 was set to the fourth group.

[0049] Next, image 5, which has the highest contribution rate of the provisional pure spectrum, is extracted from the remaining seven images. Similarly, the correlation coefficient between image 5 and the remaining images is calculated. In this case, there were no images with a correlation coefficient of 0.3 or higher.

[0050] In this embodiment, if there is only one image set in a group twice in a row, the group set up to the last group (the fourth group) is considered valid, and the remaining seven images are deleted. As a result, the four groups shown in Figure 6 are determined to be valid, and the number of multiple components of the sample (f') is determined to be four.

[0051] As described above, in the MCR processing, the provisional pure spectral matrix K is set with a provisional number of components f. However, in the image processing, the original number of multiple components (f') can be determined by performing grouping of provisional concentration distribution images with the number of components f.

[0052] In this embodiment, furthermore, in step S26, the concentration value determination unit 26 determines provisional concentration values for the multiple provisional pure spectra grouped together, for example, by summing them. The determined concentration value (c') becomes the analysis result of the MCR method. Then, in step S28, the provisional image generation unit 22 generates a chemical image with the original number of multiple components f' based on the determined concentration value (c').

[0053] Next, in step S30, the image output unit 30 superimposes the f' individual chemical images generated by the provisional image generation unit 22 for all components and displays them on the display unit 40. Figure 7 shows a color-coded diagram of the chemical images of the powder sample determined to be four components.

[0054] In this embodiment, in step S32, the pure spectrum determination unit 28 selects the provisional pure spectrum with the highest contribution rate from among the provisional pure spectra in the same group and deletes the remaining provisional pure spectrum sequences in the same group. This is performed for each group to determine a pure spectrum with component number f'.

[0055] Alternatively, in step 32, the pure spectrum determiner 28 determines a pure spectrum with the number of components f' by combining a plurality of pure spectra that have been grouped together.

[0056] The data set of the pure spectrum determined by the pure spectrum determining unit 28 is treated as the analysis result of the MCR method.

[0057] In step S32, the MCR processor 10 may re-execute the MCR process based on the number of components f' determined by the image processor 20, and determine the pure spectral matrix and the concentration matrix.

[0058] Then, in step S34, the database search unit 50 uses the pure spectrum with the component number f' determined by any of the above methods to search for a corresponding substance in the spectrum library 60, thereby enabling the components of the pure spectrum to be identified with a high accuracy rate. The image output unit 30 can then display the names of the identified components in a chemical image.

[0059] The multivariate analysis device 1 has a storage device (not shown) that stores an MCR processing program and an image processing program for realizing the respective functional blocks of the MCR processing unit 10 and the image processing unit 20. Each processing unit executes these processing programs as appropriate to achieve its functions.

[0060] A Raman spectrum data set was acquired using different measurement conditions for the same powder sample as in Figure 3. Figure 8 shows 20 provisional concentration distribution images created based on the provisional concentration values obtained through MCR processing. The order of the 20 images PC1 to PC20 in Figure 8 is random. Table 1 shows the calculation results of the correlation coefficients for all combinations of provisional concentration distribution images PC1 to PC20.

[0061] For convenience, Table 1 shows the correlation coefficients for all combinations, but in the grouping of this embodiment, it is also possible to calculate only the correlation coefficients for necessary combinations. The grouping procedure will be explained using Table 1. First, extract the image PC1 with the highest contribution rate of the provisional pure spectrum. Then, calculate the correlation coefficients between the extracted image PC1 and the remaining images (first row of Table 1). Extract images PC7, PC13, and PC19 with correlation coefficients of 0.3 or more. Set the extracted images to the first group.

[0062] [Table 1]

[0063] Next, from the remaining 16 images, image PC3 with the highest contribution rate of the provisional pure spectrum is extracted. Similarly, the correlation coefficient between image PC3 and the remaining images is calculated (third row of Table 1). Of course, images already extracted as the first group are excluded from the comparison. Images PC11 and PC18 with correlation coefficients of 0.3 or higher are extracted. The extracted images are set as the second group.

[0064] Next, image PC9 with the highest contribution rate of the provisional pure spectrum is extracted from the remaining 13 images. Similarly, the correlation coefficient between image PC9 and the remaining images is calculated (line 9 of Table 1), and image PC12 with a correlation coefficient of 0.3 or more is extracted and set as the third group.

[0065] Similarly, image PC4 with the highest contribution rate of the provisional pure spectrum is extracted from the remaining 11 images. Similarly, the correlation coefficient between image PC4 and the remaining images is calculated (fourth row of Table 1). In this case, there are no images with a correlation coefficient of 0.3 or higher, and only image PC4 is set to the fourth group.

[0066] Then, from the remaining 10 images, image PC5 with the highest contribution rate of the provisional pure spectrum was extracted, and the correlation coefficient was calculated in the same way, but no images fell into the same group.

[0067] The result of this grouping is shown in FIG. 9. The number of components is determined to be four by grouping. In this embodiment, the pure spectrum determination unit 28 selects the provisional pure spectrum (PC1) with the highest contribution rate from the provisional pure spectra of the images PC1, PC7, PC13, and PC19 in the first group, and deletes the remaining provisional pure spectrum sequences in the same group. The remaining provisional pure spectrum of PC1 is determined to be the first pure spectrum. Alternatively, the pure spectrum determination unit 28 may determine the first pure spectrum by combining the provisional pure spectra of the images PC1, PC7, PC13, and PC19 in the same group. The pure spectra of the other groups are processed in the same way. As a result, a pure spectrum with the determined number of components, "4," can be obtained. Of course, a pure spectrum may also be determined by re-executing MCR processing based on the determined number of components.

[0068] The apparatus and method of the present invention are effectively applied to multivariate analysis of various multi-point spectral data sets and time-lapse spectral data sets. The present invention is effectively applied to, for example, a microscopic Raman spectrometer that irradiates a sample with laser light and measures the spectrum of the scattered light, a microscopic infrared spectrometer that irradiates a sample with infrared interference waves and detects and Fourier transforms the reflected and transmitted light to obtain an absorption spectrum, and a microscopic ultraviolet-visible-near-infrared spectrometer that irradiates a sample with ultraviolet, visible, and near-infrared light and measures the absorption spectrum of the reflected and transmitted light. [Explanation of symbols]

[0069] 1. Multivariate analysis device 4. MCR condition input section 10. MCR processing section 12. Measurement spectrum matrix setting section 14. Provisional pure spectral matrix setting section 16: Storage and calculation section for temporary concentration matrix and temporary pure spectrum matrix (temporary loop processing section) 20. Image processing section 22...Temporary image generation unit (temporary density distribution generation unit) 24 Grouping memory and calculation unit 26. Density value determination unit 28... Pure spectrum determination section 30 Image output unit 40...Display section 50 Database Search Section 60 Spectral Library

Claims

1. A multivariate analysis device that performs multivariate curve resolution on a plurality of measurement spectra obtained by spectroscopically measuring a large number of measurement points of a sample consisting of a plurality of components, a multivariate curve decomposition unit that separates a provisionally set number of tentative pure spectra from the plurality of measured spectra, on the assumption that the measured spectrum is expressed as a linear combination of a plurality of pure spectra multiplied by a concentration value for each of the pure spectra, and calculates the tentative pure spectra and a tentative concentration value for each of the tentative pure spectra; a provisional concentration distribution generating unit that generates distribution information of provisional concentration values with respect to the positions of the measurement points for each provisional pure spectrum based on the calculated provisional concentration values; a grouping unit that groups the provisional concentration distribution information generated for each of the provisional pure spectra based on the degree of similarity of the provisional concentration distribution information, and determines the number of the multiple components based on the number of groups.

2. 2. The multivariate analysis apparatus according to claim 1, further comprising a concentration value determination unit that determines a concentration value for each component by adding up provisional concentration values of a plurality of pieces of provisional concentration distribution information that have been grouped together by the grouping unit.

3. 3. The multivariate analysis apparatus according to claim 2, a concentration distribution generating unit that generates concentration distribution information for each component based on the concentration values determined by the concentration value determining unit; an image output unit that forms a visualized image by superimposing the generated concentration distribution information; A multivariate analysis device comprising:

4. 4. The multivariate analysis apparatus according to claim 1, The multivariate analysis apparatus further comprises a pure spectrum determination unit that determines the provisional pure spectrum having the highest contribution rate as the pure spectrum from among a plurality of provisional pure spectra corresponding to a plurality of provisional concentration distribution information grouped in the grouping unit.

5. 4. The multivariate analysis apparatus according to claim 1, The multivariate analysis apparatus further comprises a pure spectrum determination unit that determines a pure spectrum by synthesizing a plurality of provisional pure spectra corresponding to a plurality of provisional concentration distribution information grouped in the grouping unit.

6. 6. The multivariate analysis apparatus according to claim 1, The multivariate analysis apparatus is characterized in that the multivariate curve resolution unit is further configured to calculate a pure spectrum and a concentration value based on the number of multiple components determined by the grouping unit.

7. A multivariate analysis method for performing multivariate curve resolution on a plurality of measurement spectra obtained by spectroscopically measuring a large number of measurement points of a sample consisting of a plurality of components, using a computer, comprising: a multivariate curve decomposition step of separating a provisionally set number of tentative pure spectra from the plurality of measured spectra and calculating the tentative pure spectra and tentative concentration values for each of the tentative pure spectra, on the assumption that the measured spectrum is expressed as a linear combination of a plurality of pure spectra multiplied by a concentration value for each of the pure spectra; a provisional concentration distribution generating step of generating distribution information of provisional concentration values with respect to the positions of the measurement points for each provisional pure spectrum based on the calculated provisional concentration values; a grouping step of grouping the provisional concentration distribution information generated for each of the provisional pure spectra based on the degree of similarity of the provisional concentration distribution information, and determining the number of the multiple components based on the number of groups.

Images