Spectral analysis method, device and system

Abstract

To provide a spectrum analysis device, method and system which, while utilizing spatial information in three-dimensional spectrum data, realize high accuracy quantitative determination / identification in which influences of noise are reduced and measurement time reductions by search of an effective wavelength band in an excitation wavelength or a detection wavelength unit.SOLUTION: A spectrum analysis device comprises: an input unit for accepting three-dimensional spectrum data as input; a wavelength unit feature quantity extraction unit for extracting the detection intensity of two or more detection wavelength bands in each excitation wavelength band from the three-dimensional spectrum data, or a feature quantity from the detection intensity of two or more detection wavelength bands in each detection wavelength band; an estimation unit for performing quantitative determination / identification on the basis of the feature quantity generated by the wavelength unit feature quantity extraction unit; an effective wavelength band search unit for searching for a wavelength band (effective wavelength band) that is used for the quantitative determination / identification in excitation wavelength or detection wavelength units; and an output unit for outputting a result of the quantitative determination / identification and the effective wavelength band.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a spectral analysis method, apparatus and system. [Background technology]

[0002] Spectral analyzers obtain a spectrum by irradiating a sample with excitation light and measuring the light detected from the sample. Some spectral analyzers obtain a three-dimensional spectrum composed of three axes: excitation wavelength, detection wavelength, and detection intensity. For example, a spectrofluorometer measures a three-dimensional fluorescence spectrum known as a fluorescence fingerprint. When a sample is irradiated with excitation light of a specific wavelength, the excited sample emits fluorescence of various wavelengths. This is called a fluorescence spectrum, and the intensity of the fluorescence at each fluorescence wavelength is called fluorescence intensity. A sample emits different fluorescence spectra by changing the wavelength of the excitation light. Therefore, by measuring the fluorescence spectrum while sequentially changing the wavelength of the excitation light, a three-dimensional spectrum with three axes: excitation wavelength, fluorescence wavelength, and fluorescence intensity is obtained. The aforementioned detection wavelength corresponds to the fluorescence wavelength, and detection intensity corresponds to the fluorescence intensity. Analyzing the shape of the obtained three-dimensional spectrum makes it possible to quantify and identify the types of components contained in the sample.

[0003] Conventional methods for analyzing spectral data include, for example, a method of performing regression analysis using only peak values ​​in a three-dimensional spectrum. Furthermore, Patent Document 1 discloses a method of providing a rectangular measurement window for a fluorescent fingerprint and performing multivariate analysis using the integral value of the fluorescent intensity within the measurement window. Furthermore, Patent Document 2 discloses a method for estimating the content of a target substance while limiting the excitation wavelengths used for estimation by performing sparse estimation using two-dimensional spectral data and optical absorption characteristic values ​​(detection intensities) at each of multiple excitation wavelengths as explanatory variables. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2015-180895 A [Patent Document 2] JP 2018-013418 A Summary of the Invention [Problem to be solved by the invention]

[0005] In linear regression using only peak values, once the wavelength band in which the peak appears is identified, only that wavelength band is measured and analyzed, aiming for short measurement and analysis times. However, because only peak information is used, there are issues such as difficulty in quantification / identification using spatial information such as peak width or distance between peaks, and susceptibility to the influence of point noise. Hereafter, in spectral analysis, the wavelength band used for quantification and / or identification is referred to as the effective wavelength band.

[0006] In addition, the technology described in Patent Document 1 addresses the above issues by providing a rectangular measurement window and performing multivariate analysis using the integral value of the fluorescence intensity within the measurement window. This aims to limit the measurement wavelength range while reducing the influence of noise. However, the measurement window must be manually designed or a regression equation must be constructed from the fluorescence intensity within a randomly set measurement window, and the measurement window must be repeatedly reset based on the quantification results. It is difficult to visually determine the wavelength range effective for quantification / identification of the analyte. Furthermore, random trials can result in a large measurement window, making it difficult to adequately limit the effective wavelength range. Therefore, it can be difficult to build a highly accurate quantification / identification model in a short time or to limit the effective wavelength range to achieve efficient measurement.

[0007] The technology described in Patent Document 2 automatically selects wavelength bands effective for quantification / identification by performing sparse estimation using all detected intensities in two-dimensional spectral data as explanatory variables. However, because the selected wavelength bands are points rather than regions, there are issues such as difficulty in extracting spatial information and susceptibility to point noise, similar to when only peak information is used.

[0008] In view of the above-mentioned problems, the present invention aims to provide a spectral analysis device, method, and system that utilizes spatial information in three-dimensional spectral data while reducing the influence of noise and achieving highly accurate quantification and / or identification. [Means for solving the problem]

[0009] In order to achieve the above object, the spectral analysis device disclosed in the present invention comprises an input unit that receives spectral data of a sample as an input, a wavelength-unit feature extraction unit that extracts features of the sample from the spectral data, an estimation unit that estimates characteristics of the sample based on the features, and an output unit that outputs the estimation results obtained by the estimation unit, wherein the spectral data includes a plurality of excitation wavelengths and optical spectra indicating detection wavelengths and detection intensities for the excitation wavelengths, and is data discretized over a predetermined wavelength width, the wavelength-unit feature extraction unit extracts features from the detection intensities of two or more detection wavelength bands in each excitation wavelength band, or the detection intensities of two or more excitation wavelength bands in each detection wavelength band, and the estimation unit performs estimation processing to quantify and / or identify the characteristics of the sample. The spectral analysis method disclosed in the present invention includes an input step of receiving spectral data of a sample as input, a wavelength-unit feature extraction step of extracting features of the sample from the spectral data, an estimation step of estimating characteristics of the sample based on the features, and an output step of outputting the estimation results obtained by the estimation step, wherein the spectral data includes a plurality of excitation wavelengths and optical spectra indicating detection wavelengths and detection intensities for the excitation wavelengths, and is data discretized over a predetermined wavelength width, and the wavelength-unit feature extraction step extracts features from the detection intensities of two or more detection wavelength bands in each excitation wavelength band, or from the detection intensities of two or more excitation wavelength bands in each detection wavelength band, and the estimation step performs estimation processing to quantify and / or identify characteristics of the sample. [Effects of the Invention]

[0010] According to the present invention, it is possible to provide a spectral analysis device, method, and system that utilizes spatial information in three-dimensional spectral data, achieves highly accurate quantification and / or identification with reduced noise influence, and can limit the wavelength band used for quantification and / or identification in units of excitation wavelength or detection wavelength. Problems, configurations, and advantages other than those described above will become clear from the description of the embodiments below. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 illustrates an example of a hardware configuration of a spectrum analysis device according to a first embodiment. [Figure 2] FIG. 1 is a diagram illustrating an example of a functional block diagram of a spectrum analysis device according to a first embodiment. [Figure 3] FIG. 10 is a diagram showing an example of three-dimensional spectral data. [Figure 4] FIG. 10 is a diagram illustrating an example of a method of inputting data to a wavelength-unit feature extraction unit. [Figure 5] FIG. 2 is a diagram illustrating an example of a processing flow of a spectrum analysis method according to the first embodiment. [Figure 6] FIG. 10 is a diagram illustrating an example of a functional block diagram of a spectrum analysis device according to a third embodiment. [Figure 7] FIG. 10 is a diagram showing an example of displaying the relationship between quantitative / identification accuracy and the number of effective wavelength bands. [Figure 8] FIG. 10 is a diagram showing an example of displaying an effective wavelength band. [Figure 9] FIG. 11 is a diagram illustrating an example of a processing flow of a spectrum analysis method according to a third embodiment. [Figure 10] FIG. 10 is a diagram illustrating an example of a hardware configuration of a spectrum analysis system according to a fourth embodiment. [Figure 11] FIG. 10 is a diagram illustrating an example of a functional block diagram of a spectrum analysis device according to a second embodiment. [Figure 12] FIG. 10 is a diagram illustrating an example of a processing flow of a spectrum analysis method according to a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of a spectrum analysis method, device, and system according to the present invention will be described with reference to the accompanying drawings. In the following description and accompanying drawings, components having the same functional configuration are designated by the same reference numerals, and redundant description will be omitted. [Example]

[0013] <Hardware configuration of spectrum analyzer> In Example 1, we describe a spectral analysis device that utilizes spatial information within three-dimensional spectral data to perform quantification / identification while reducing the effects of noise, and that can limit the wavelength band (effective wavelength band) used for quantification / identification in units of excitation wavelength or detection wavelength. In this embodiment, the three-dimensional spectral data is data that includes a plurality of excitation wavelengths and optical spectra (spectra consisting of detection wavelengths and detection intensities) and is discretized in a predetermined wavelength range. Furthermore, quantification / discrimination refers to quantification and / or discrimination, and a quantification / discrimination model is a model for estimating data used for at least one of quantification and discrimination.

[0014] The hardware configuration of a spectrum analysis device according to a first embodiment will be described with reference to Fig. 1. The spectrum analysis device 100 includes an interface unit 110, a calculation unit 111, a memory 112, and a bus 113. The interface unit 110, the calculation unit 111, and the memory 112 transmit and receive information via the bus 113.

[0015] Each component of the spectrum analysis device 100 will now be described. The interface unit 110 is a communication device that transmits and receives signals to and from devices external to the spectrum analysis device 100. Devices that transmit and receive signals to and from the interface unit 110 include a control device 120 that controls a spectrum measurement device 121, the spectrum measurement device 121 that measures three-dimensional spectrum data, and a display device 122 such as a monitor or printer that displays the processing results of the spectrum analysis device 100.

[0016] The calculation unit 111 is a device that executes various processes within the spectrum analysis device 100, and is, for example, a CPU (Central Processing Unit) or an FPGA (Field-Programmable Gate Array). The functions executed by the calculation unit 111 will be described later.

[0017] The memory 112 is a device that stores the programs executed by the calculation unit 111, parameters of the quantitative / discrimination models, processing results, etc., and is an HDD, SSD, RAM, ROM, flash memory, or the like.

[0018] <Functional Configuration of Spectrum Analysis Device 100> 2 is an example of a functional block diagram according to an embodiment of the spectrum analysis device 100. Each of these functional units may be realized by software that runs on the calculation unit 111, or may be realized by dedicated hardware.

[0019] The spectral analysis device 100 includes, as functional units, an input unit 201, a wavelength-unit feature extraction unit 202, an estimation unit 203, and an output unit 204. Each functional unit will be described below. The input unit 201 receives one or more pieces of discretized three-dimensional spectral data input from the interface unit 110 . The wavelength-unit feature extracting unit 202 extracts features from the three-dimensional spectral data based on the detected intensities of two or more detection wavelength bands in each excitation wavelength band, or from the detected intensities of two or more excitation wavelength bands in each detection wavelength band.

[0020] The estimation unit 203 reads one of the quantification / identification models stored in the memory 112, and performs quantification / identification based on the feature amount. The output unit 204 outputs the quantification / classification results. In the case of quantification, in addition to the estimated quantification results for each sample, evaluation values ​​of the quantitative model such as the coefficient of determination and RMSE (Root Mean Squared Error) may be output, or graph data such as a calibration curve may be output. In the case of classification, in addition to the estimated classification results for each sample, evaluation values ​​of the classification model such as the classification rate, precision, and recall may be output. These outputs are displayed on the display device 122 via the interface unit 110.

[0021] Note that the above functions do not need to be configured exactly as the functional units in Figure 2, as long as they can achieve processing corresponding to the operation of each functional block. Figure 5 shows an example of a processing flow diagram according to the first embodiment. Each step corresponds to each element in the functional block diagram shown in Figure 2. In the input step 501, one or more pieces of discretized three-dimensional spectral data are received from the interface unit 110. In wavelength unit feature extraction step 502, features are extracted from the three-dimensional spectral data from the detected intensities of two or more detection wavelength bands in each excitation wavelength band, or from the detected intensities of two or more excitation wavelength bands in each detection wavelength band.

[0022] In the estimation step 503, one of the quantitative / discriminative models stored in the memory 112 is read, and quantitative / discriminative analysis is performed based on the feature amount. The output step 504 outputs the quantification / identification results. In the case of quantification, in addition to the estimated quantification results for each sample, evaluation values ​​of the quantitative model such as the coefficient of determination and RMSE (Root Mean Squared Error) may be output, or graph data such as a calibration curve may be output. In the case of identification, in addition to the estimated identification results for each sample, evaluation values ​​of the identification model such as the identification rate, precision, and recall may be output. Hereinafter, detailed operations will be described with each functional unit as the subject, but it is also acceptable to interpret each step corresponding to each functional unit as the subject.

[0023] <Configuration and operation of each part> The operations of the input unit 201, wavelength-unit feature extraction unit 202, and estimation unit 203 among the functional units will be described in detail below. The input unit 201 receives the discretized three-dimensional spectral data via the interface unit 110 and stores it in the memory 112 .

[0024] An example of fluorescence fingerprint data is shown in Figure 3. Discretized fluorescence fingerprint data 300 is an example of fluorescence fingerprint data that has been discretized and stored. Fluorescence fingerprint data visualization example 301 shows the detection intensity for each combination of excitation and detection wavelengths displayed as contour data, with the excitation light wavelength on the vertical axis and the detection wavelength on the horizontal axis. As shown in discretized fluorescence fingerprint data 300, the fluorescence fingerprint data is measured and recorded in a discretized form. Discretized fluorescence fingerprint data 300 shows an example in which the excitation light / detection light was measured from 200 nm to 700 nm at 5 nm intervals. For example, the detection intensity for 200 nm detection light with 200 nm excitation light is recorded in row 1, column 1 of discretized fluorescence fingerprint data 300; the detection intensity for 205 nm detection light with 200 nm excitation light is recorded in row 1, column 2; and the detection intensity for 200 nm detection light with 205 nm excitation light is recorded in row 2, column 1 of discretized fluorescence fingerprint data 300. The number of columns and rows of the discretized fluorescent fingerprint data 300 is 100x100 because 500 nm is recorded at 5 nm intervals.

[0025] Fluorescence fingerprint data visualization example 301 is an example in which discretized fluorescence fingerprint data 300 has been converted into a two-dimensional image as contour data, and when humans observe fluorescence fingerprint data, they often confirm it in a display such as fluorescence fingerprint data visualization example 301. In addition to contour data, it may also be displayed as a gray image or heat map image in which detection intensity is assigned to brightness values ​​or colors, or as a bird's-eye view. The shape of the fluorescence fingerprint differs depending on the properties of the sample to be measured, and these shape differences are used to identify the sample or quantify specific components.

[0026] The wavelength-unit feature extraction unit 202 extracts features from the three-dimensional spectral data based on the detection intensities of two or more detection wavelength bands in each excitation wavelength band, or the detection intensities of two or more excitation wavelength bands in each detection wavelength band. This makes it possible to select and limit wavelengths effective for identification and quantification in excitation or detection wavelength units while extracting spatial features in the detection wavelength direction or excitation wavelength direction.

[0027] The advantages of the method disclosed in this example will be explained in comparison with two reference methods. The first reference method is a method of performing regression using each detection intensity, i.e., the value of each square in the discretized fluorescence fingerprint data 300, as an explanatory variable. In the first reference method, a regression equation is constructed by assigning a high weight to the detection intensity at a combination of excitation wavelength and detection wavelength that is effective for identification and quantification. After constructing the regression equation, measurement time is shortened by measuring only the explanatory variables assigned a high weight, i.e., the combination of excitation / detection wavelength, during identification and quantification. However, there are issues such as the difficulty of extracting spatial features such as the width of peak components, and the direct influence of point noise on identification and quantification.

[0028] The second reference method is to extract spatial features using a convolutional neural network (CNN). CNN is a machine learning method frequently used in the field of image processing, and extracts features by automatically learning multiple filters. Using a CNN makes it possible to extract spatial features and also improves robustness against point noise. However, because CNNs generally calculate features comprehensively from the entire two-dimensional data, i.e., the values ​​of all squares of the discretized fluorescence fingerprint data 300, they tend to require detection intensities over a wide range of wavelength bands to calculate features, which can make it difficult to limit the effective wavelength band.

[0029] In contrast to these reference methods, the method disclosed in this example extracts features independently from the excitation wavelength, i.e., each row of the discretized fluorescence fingerprint data 300, or from the detection wavelength, i.e., each column of the discretized fluorescence fingerprint data 300. This makes it possible to extract features that have spatial information in the detection wavelength direction or excitation wavelength direction and are robust to point noise, and further makes it possible to select and limit the effective wavelength band in units of excitation or detection wavelength.

[0030] FIG. 4 illustrates an example of a feature extraction range. FIG. 4(a) shows an example of extracting features from the entire detection wavelength band for each excitation wavelength band. FIG. 4(b) shows an example of extracting features from the entire excitation wavelength band for each detection wavelength band. FIG. 4(c) shows an example of extracting features from a portion of the detection wavelength band for each excitation wavelength band. FIG. 4(d) shows an example of extracting features from a portion of the excitation wavelength band for each detection wavelength band. The rectangles depicted within the 3D spectral data in FIGS. 4(a), 4(b), 4(c), and 4(d) represent input units to the feature extraction model for extracting features. The wavelength-based feature extraction unit 202 extracts features independently for each excitation wavelength band, as shown in FIGS. 4(a) and 4(c), or for each detection wavelength band, as shown in FIGS. 4(b) and 4(d). In this case, features may be extracted from the entire detection or excitation wavelength band, as shown in FIGS. 4(a) and 4(b), or may be extracted from only a portion of the detection or excitation wavelength band, as shown in FIGS. 4(c) and 4(d).

[0031] 4(a) and 4(c) show that feature values ​​are extracted from the same detection wavelength for each excitation wavelength, but the detection wavelength band from which feature values ​​are extracted may be changed depending on the excitation wavelength. For example, in a fluorescence fingerprint, noise called scattered light is measured mainly in the region where the excitation wavelength and detection wavelength are equal, but the region from which feature values ​​are extracted may be changed depending on the excitation wavelength to avoid this scattered light region. The same applies when the detection wavelength and excitation wavelength are swapped.

[0032] Feature extraction methods include, for example, 1DCNN (1-Dimensional Convolutional Neural Network). 1DCNN is a CNN for one-dimensional data that extracts multiple features for each excitation wavelength or detection wavelength. Feature learning methods include, for example, connecting one or more 1DCNNs for feature extraction with a fully connected layer that uses the output of the 1DCNN as input for quantification or classification, and then using a dataset created for each quantification or classification problem to update the 1DCNN and the fully connected layer using machine learning techniques to learn features. Feature learning methods are not limited to this; they can also be used to learn common features by simultaneously training multiple tasks, or unsupervised learning methods such as autoencoders. Additionally, handcrafted features such as histograms of oriented gradients (HOG) and haar-like features can be used, as well as statistical analysis methods such as principal component analysis (PCA) and partial least squares (PLS).

[0033] The wavelength-unit feature extraction unit 202 extracts information on the region (detection intensity measured at two or more wavelengths) as shown in Fig. 4, thereby enabling the estimation unit 203, described later, to provide a quantification / identification model with improved robustness against noise while utilizing spatial information. In addition, by extracting features independently for each excitation wavelength or each detection wavelength as shown in Fig. 4 and using each feature as an input for quantification / identification in the estimation unit 203, it becomes possible to provide a quantification / identification model that can limit the effective wavelength band.

[0034] The estimation unit 203 reads any quantification / identification model stored in the memory 112 and inputs the feature obtained by the wavelength-unit feature extraction unit 202 into the quantification / identification model, thereby performing quantification / identification.

[0035] For the quantitative / discriminative model, a sparse estimation model such as Lasso regression, Elastic Net, or Group Lasso may be used, or a machine learning model such as Neural Network, Random Forest, or Support Vector Machine may be used. Here, an example using Lasso regression is explained.

[0036] Lasso regression estimates based on a regression equation such as Equation 1, where x ij is a feature in wavelength band unit output by wavelength unit feature extraction unit 202, i is the index of excitation or detection wavelength, and j is the index of feature. Note that the number of excitation or detection wavelengths (i.e., the number of rows or columns in discretized fluorescence fingerprint data 300) is N, and the number of features extracted from each wavelength is M. ij represents the weight for each feature, b represents the bias value, and y^ represents the output of the regression model.

number

number

number

number

[0037] The weights w constructed by the above method are stored in advance in the memory 112, and the estimation unit 203 reads the weights w when the spectrum analysis device 100 is started or when the weights w are updated, and executes the quantification / identification process.

[0038] As described above, the wavelength-unit feature extraction unit 202 extracts features from the detection intensities of two or more detection wavelengths at each excitation wavelength, or from the detection intensities of two or more excitation wavelengths at each detection wavelength, and the estimation unit 203 performs quantification / identification using the features as explanatory variables. This makes it possible to provide a spectral analysis device and method that utilizes spatial information within the spectral data to perform highly accurate quantification / identification with reduced influence of noise, and that can limit the wavelength band used for quantification / identification by excitation wavelength or detection wavelength unit. [Example]

[0039] In Example 2, we will explain a spectral analysis device that adds a quantitative / identification model generation function to the spectral analysis device described in Example 1, thereby enabling highly accurate identification and selection of effective wavelength bands for quantitative / identification problems set by the user.

[0040] <Hardware configuration of spectrum analyzer> The hardware configuration of the spectrum analysis device according to the second embodiment is similar to the hardware configuration of the spectrum analysis device according to the first embodiment shown in FIG. 1, and therefore a description thereof will be omitted.

[0041] <Functional configuration of spectrum analyzer> FIG. 11 is an example of a functional block diagram of a spectrum analysis device according to the second embodiment. The spectral analysis device 1100 includes, as functional units, an input unit 1101, a wavelength-unit feature extraction unit 202, an estimation unit 1103, and an output unit 204. The wavelength-unit feature extraction unit 202 and the output unit 204 are the same as those in the first embodiment, and therefore their explanation will be omitted. The input unit 1101 and the estimation unit 1103 will be explained below. The input unit 1101 receives, via the interface unit 110, three-dimensional spectral data alone or a combination of three-dimensional spectral data and teaching information. The estimation unit 1103 generates a quantitative / discriminative model using a set of multiple three-dimensional spectral data and teaching information stored in the memory 112.

[0042] Note that the above functions do not need to be configured exactly as the functional units in Figure 11, as long as they can realize the processing that realizes the operation of each functional block. Figure 12 shows an example of a processing flow diagram according to the second embodiment. Each step corresponds to each element in the functional block diagram shown in Figure 11. The wavelength-unit feature extraction step 502 and output step 504 in FIG. 12 are the same as those described in the first embodiment, and therefore a description thereof will be omitted. In the input step 1201, three-dimensional spectral data alone or a combination of three-dimensional spectral data and teaching information is accepted via the interface unit 110. The estimation step 1203 uses a set of multiple three-dimensional spectral data and teaching information stored in the memory 112 to generate a quantitative / discriminative model. Hereinafter, detailed operations will be described with each functional unit as the subject, but it is also acceptable to interpret each step corresponding to each functional unit as the subject.

[0043] <Configuration and operation of each part> The operations of the input unit 1101 and the estimation unit 1103 will now be described in detail. The input unit 1101 receives three-dimensional spectral data alone or a combination of three-dimensional spectral data and training information via the interface unit 110. When receiving three-dimensional spectral data alone, the input unit 1101 performs the same processing as the input unit 201 described in the first embodiment. When receiving a combination of three-dimensional spectral data and training information, the input unit 1101 stores the received combination of three-dimensional spectral data and training information in the memory 112.

[0044] When a certain condition is satisfied, the estimation unit 1103 generates a quantitative / discriminative model using a set of multiple three-dimensional spectral data and teaching information stored in the memory 112. The condition for starting the generation of the quantitative / discriminative model is the set of multiple three-dimensional spectral data newly stored in the memory 112. and The quantitative / discrimination model may be updated when the number of sets of teaching information exceeds a predetermined number, when a request is made by the user via the interface unit 110, or every time a predetermined time period has elapsed.

[0045] As described in the first embodiment, the quantitative / discriminative model may be a sparse estimation model such as Lasso regression, Elastic Net, or Group Lasso, or a machine learning model such as Neural Network, Random Forest, or Support Vector Machine. Here, an example using Lasso regression will be described.

[0046] As described in Example 1, Lasso regression constructs the regression equation shown in Equation 2b and derives the weight w based on the constraints shown in Equation 3. Here, among the multiple pairs of three-dimensional spectral data and training information described above, the three-dimensional spectral data group is designated X, and the corresponding training information group is designated y, and the weight w that satisfies Equation 3 is derived. The weight is derived using ISTA (Iterative Shrinkage Thresholding Algorithm), which is one of the solutions for Lasso regression. This allows a quantitative / discriminative model to be constructed using a data set independently collected by the user. A quantitative / discriminative model can also be constructed using a similar method when using other sparse estimation models or regression models. When a machine learning model is used, the quantitative / discriminative model is trained in accordance with the respective machine learning algorithms using the three-dimensional spectral data group and the corresponding training information group. Furthermore, the initial value of the weight of the quantitative / identification model may be set to a fixed value or a random value, or may be set according to a probability distribution such as a normal distribution, or the weight of the quantitative / identification model stored in memory 112 may be set.

[0047] In addition, when the wavelength-unit feature extraction unit 202 uses a machine learning method such as a neural network or a statistical analysis method such as principal component analysis or PLS as a feature extraction model, the feature extraction model may be updated using not only the quantitative / discrimination model but also the above-mentioned multiple pieces of three-dimensional spectral data or a combination of multiple pieces of three-dimensional spectral data and training information.

[0048] For example, when using Auto Encoder, a type of machine learning method, or principal component analysis, a type of statistical analysis method, no supervised information is required to update the feature extraction model. Therefore, features are updated by unsupervised learning or statistical analysis using only multiple pieces of 3D spectral data. Also, when using 1DCNN, a type of machine learning method described in Example 1, or PLS, a type of statistical analysis method, the feature extraction model is updated by supervised learning using a set of multiple pieces of 3D spectral data and a supervised signal, or by performing statistical analysis using the supervised information as a response variable.

[0049] As a result of the above, it is possible to construct a quantitative / identification model using a data set that the user has collected independently, and it is possible to provide a spectral analysis device and method that can perform highly accurate identification and select effective wavelength bands for quantitative / identification problems that the user has set independently. [Example]

[0050] In the third embodiment, a spectral analysis device will be described in which a function for automatically determining an effective wavelength band or a function for supporting a user in determining an effective wavelength band is added to the spectral analysis device 100 described in the second embodiment.

[0051] <Hardware configuration of spectrum analyzer> The hardware configuration of the spectrum analysis device according to the third embodiment is similar to that of the spectrum analysis device according to the first embodiment shown in FIG. 1, and therefore a description thereof will be omitted.

[0052] <Functional configuration of spectrum analyzer> FIG. 6 is an example of a functional block diagram of a spectrum analysis device according to the third embodiment. The spectral analysis device 600 includes, as functional units, an input unit 601, a wavelength-unit feature extraction unit 202, an estimation unit 1103, an output unit 604, and an effective wavelength band search unit 605. The wavelength-unit feature extraction unit 202 is the same as in the first embodiment, and the estimation unit 1103 is the same as in the second embodiment, so their explanations will be omitted. The input unit 601, the effective wavelength band search unit 605, and the output unit 604 will be explained below.

[0053] The input unit 601, like the input unit 1101 described in the second embodiment, receives three-dimensional spectral data alone or a combination of three-dimensional spectral data and teaching information via the interface unit 110. If necessary, the input unit 601 also receives an effective wavelength band from an effective wavelength band search unit 605 (described later) and processes the spectral data to be output to the wavelength-unit feature extraction unit 202. The effective wavelength band search unit 605 automatically determines the effective wavelength band based on the weight of the quantitative / identification model constructed by the estimation unit 1103, or assists the user in determining the effective wavelength band. The output unit 604 displays the effective wavelength band determined by the effective wavelength band search unit 605 together with the quantification / identification results, and outputs information to assist the user in determining the effective wavelength band, as necessary.

[0054] Note that the above functions do not need to be configured exactly as the functional units in Figure 6, as long as they can realize the processing that realizes the operation of each functional block. Figure 9 shows an example of a processing flow diagram according to the third embodiment. Each step corresponds to each element in the functional block diagram shown in Figure 6.

[0055] The wavelength-unit feature extraction step 502 shown in FIG. 9 is the same as that in the first embodiment, and the estimation step 1203 is the same as that in the second embodiment, so the description thereof will be omitted. In the input step 901, similar to the input step 1201 described in the second embodiment, three-dimensional spectral data alone or a combination of three-dimensional spectral data and training information is accepted via the interface unit 110. Furthermore, if necessary, an effective wavelength band is accepted from an effective wavelength band search step 905 (described later) and the spectral data to be output to the wavelength-unit feature extraction step 502 is processed. In the effective wavelength band search step 905, the effective wavelength band is automatically determined based on the weight of the quantitative / identification model constructed in the estimation step 1203, or the user is assisted in determining the effective wavelength band. In the output step 904, the effective wavelength band determined in the effective wavelength band search step 905 is displayed together with the quantification / identification results, and information is output to assist the user in determining the effective wavelength band, if necessary. Hereinafter, detailed operations will be described with each functional unit as the subject, but it is also acceptable to interpret each step corresponding to each functional unit as the subject.

[0056] <Configuration and operation of each part> The input unit 601 receives the three-dimensional spectral data alone or a combination of the three-dimensional spectral data and teaching information via the interface unit 110, and performs the same processing as the input unit 1101 described in Example 2. Alternatively, the input unit 601 may receive effective wavelength band information output by the effective wavelength band search unit 605, and input only the spectral data corresponding to the effective wavelength band from the three-dimensional spectral data to the wavelength-unit feature extraction unit 202, thereby shortening the feature extraction time.

[0057] The effective wavelength band search unit 605 automatically determines the effective wavelength band in units of excitation wavelength or detection wavelength based on the weights of the quantification / identification model constructed by the estimation unit 1103, and outputs the effective wavelength band or displays information to assist the user in determining the effective wavelength band on the display device 122 via the output unit 604 and the interface unit 110.

[0058] First, a method for automatically determining an effective wavelength band will be described. As described in the second embodiment, the estimation unit 1103 constructs a quantitative / identification model using a sparse estimation model such as Lasso regression, Elastic Net, or Group Lasso, or a machine learning model such as Neural Network, Random Forest, or Support Vector Machine. For example, when a sparse estimation method such as Lasso regression, Elastic Net, or Group Lasso is used, the coefficient multiplied by multiple explanatory variables (feature amounts) has a characteristic value of 0, and therefore the excitation or detection wavelength band from which a feature amount corresponding to a non-zero coefficient is extracted becomes the wavelength band used for quantification / identification. Therefore, the effective wavelength band search unit 605 receives the weight value of the quantitative / identification model constructed by the estimation unit 1103 and determines the effective wavelength band based on the weight value. For example, when Lasso regression is used as the quantitative / identification model, w in Equation 1 is 11 ,w 12 ···w nm and the weight w corresponding to the feature extracted from wavelength i. i0 ,···w im If any one of has a value other than 0, wavelength i is considered to be an effective wavelength band, and weight w i0 ,···w im If all of are 0, wavelength i is excluded from the effective wavelength band. Also, when using a decision tree such as random forest or gradient boosting, any number of explanatory variables are selected in descending order of importance, and the wavelengths corresponding to those explanatory variables are selected as the effective wavelength band.

[0059] We also describe a method for automatically determining hyperparameters for controlling sparseness. For example, in Lasso regression, the number of features that become 0 changes depending on the value of the regularization strength α. As α increases, the number of features that become 0 increases, allowing for a smaller number of effective wavelength bands. However, because the number of features used for classification decreases, the quantification / classification accuracy tends to decrease as α increases. To determine the strength of α, for example, multiple pairs of 3D spectral data and training information for calibration are received from the user and stored in memory 112. Evaluation is performed using calibration data while varying α, and α is automatically determined by adopting an α that minimizes the number of effective wavelength bands while maintaining accuracy above a predetermined or user-specified threshold for the calibration data, or by adopting an α that maximizes classification accuracy while maintaining effective wavelength bands below the threshold. In addition to Lasso regression, it is also possible to automatically determine hyperparameters based on the quantification / classification accuracy of the calibration data.

[0060] Next, a method for determining an effective wavelength band when using a fully connected layer or a support vector machine, which is a type of neural network, will be described. Although the fully connected layer and the support vector machine differ from Lasso regression and the like in the weight optimization method, they are similar in that the quantification / identification results are calculated in accordance with Equation 1. Therefore, the effective wavelength band search unit 605 determines the effective wavelength band by, for example, sorting the weights of the quantification / identification model in descending order of absolute value of the weights and extracting the feature quantities and wavelength bands corresponding to the top weights. For example, a threshold value TH is set for determining the wavelength band to be adopted as the effective wavelength band, and a wavelength band corresponding to a weight having an absolute value exceeding the threshold value TH is adopted as the effective wavelength band. The method for determining the effective wavelength band is not limited to this, and may be, for example, a threshold value TH for the ranking of the sorted results regarding the absolute values. R , and sort the absolute values ​​from the top to the TH RThe wavelength bands corresponding to the first weight may be adopted as the effective wavelength bands, or a threshold TH for the ratio of the sorted results of the absolute values ​​may be set. P By setting up a higher TH P %. R , T.H. P These correspond to the hyperparameters for controlling the sparseness mentioned above, so the automatic hyperparameter determination method mentioned above can be used to determine the thresholds TH and TH R , T.H. P The quantitative / identification model may be updated again using only the spectral data corresponding to the adopted effective wavelength band.

[0061] The above method makes it possible to automatically determine the effective wavelength band based on the weights of the quantitative / identification model. The effective wavelength band search unit 605 may also automatically determine whether to extract features for each excitation wavelength or each detection wavelength. One method for this determination is to use the calibration data described above. Using calibration data consisting of a set of multiple sets of three-dimensional spectral data and training information, the quantification / identification accuracy and number of effective wavelength bands are obtained for each of the methods for extracting features for each excitation wavelength and for extracting features for each detection wavelength. Based on the quantification / identification accuracy and number of effective wavelength bands for both methods, the unit automatically determines which method to adopt. The automatic determination method may simply adopt the method with the highest quantification / identification accuracy, or it may adopt the method with the lowest number of effective wavelength bands and a quantification / identification accuracy equal to or greater than a predetermined or user-specified threshold, or it may adopt the method with the highest quantification / identification accuracy within a predetermined or user-specified number of effective wavelength bands.

[0062] Next, we will explain how to assist users in determining effective wavelength bands. As mentioned above, constructing a quantitative / identification model requires determining, for example, hyperparameters for controlling sparseness and a feature extraction method. For example, in Lasso regression, the regularization strength α must be determined as a hyperparameter. As mentioned above, as the value of α increases, the number of effective wavelength bands decreases, but the quantitative / identification accuracy also tends to decrease. The necessary information is presented to help users determine an appropriate α. For example, using the calibration data mentioned above, the number of effective wavelength bands and quantitative / identification accuracy are plotted when the value of α is changed. A graph showing the relationship between the number of effective wavelength bands and quantitative / identification accuracy, as shown in Figure 7, is generated and presented to the user via the output unit 604 (described below). The example in Figure 7 is a graph showing the relationship between effective wavelength bands and quantitative / identification accuracy in Lasso regression. It is assumed that the user selects one quantitative / identification model from the graph using a mouse or other device via the interface unit 110. The regularization strength α, quantitative / identification accuracy, and number of effective wavelength bands for the selected quantitative / identification model are displayed in "α," "Accuracy," and "Wavelength." This allows the user to select the hyperparameter value taking into account the required measurement time and quantification / identification accuracy. Furthermore, by presenting a graph showing the relationship between the number of effective wavelength bands and quantification / identification accuracy when extracting features for each excitation wavelength and when extracting features for each detection wavelength, the user can be assisted in selecting whether to extract features for each excitation wavelength or each detection wavelength. Furthermore, when using a quantification / identification model that does not have a hyperparameter such as the regular intensity α, the quantification / identification accuracy when extracting features for each excitation wavelength and when extracting features for each detection wavelength may be presented to the user, allowing the user to decide which feature extraction method to adopt.

[0063] The output unit 604 displays the quantification / identification accuracy, the number of effective wavelength bands, specific wavelength bands, etc. on the display device 122 via the interface unit 110. Fig. 8 shows an example of an effective wavelength band display. The example in Fig. 8 indicates that excitation wavelengths of 200 nm, 220 nm, 290 nm, and 315 nm have been selected as effective wavelength bands. As mentioned above, a GUI showing the relationship between effective wavelength bands and quantification / identification accuracy, as shown in Fig. 7, may be presented to the user to assist in determining the effective wavelength band. [Example]

[0064] In Example 4, a spectral analysis system is described that uses the spectral analysis device 600 described in Example 3 to achieve highly accurate quantification / identification of three-dimensional spectral data while shortening measurement time by searching for effective wavelength bands in units of excitation wavelength or detection wavelength. The spectral analysis system described in Example 4 has two phases: a training phase in which a quantification / identification model is constructed using a training dataset provided by a user and an effective wavelength band is searched for; and an evaluation phase in which only the spectrum of the effective wavelength band identified in the training phase is measured for an evaluation sample and quantification / identification is performed using the quantification / identification model constructed in the training phase. Below, the hardware configuration will be described, followed by a detailed description of each phase.

[0065] <Hardware Configuration of Spectral Analysis System 1000> 10 shows a hardware configuration diagram according to the fourth embodiment. The spectrum analysis system 1000 according to the fourth embodiment includes a control device 120, a spectrum measurement device 121, a spectrum analysis device 600, and a display device 122.

[0066] The control device 120 sets the wavelength band of the excitation light irradiated onto the sample and the detection wavelength band of the light reflected, transmitted, and absorbed by the sample in the spectrum measurement device 121 described below. The spectrum measuring device 121 measures the spectra of the training sample and the evaluation sample corresponding to the wavelengths set by the control device 120 , and inputs the discretized three-dimensional spectrum data to the spectrum analyzing device 600 . The spectral analysis device 600 is the spectral analysis device described in Example 3, and in the training phase, it constructs training data from a combination of three-dimensional spectral data of a training sample output by the spectral measurement device 121 and teacher information input by a user, and generates a quantification / identification model and searches for an effective wavelength band using the training data set.In addition, in the evaluation phase, it uses the quantitative / identification model generated in the training phase to perform quantification / identification using the three-dimensional spectral data (evaluation data) of an evaluation sample output by the spectral measurement device 121 as input. The display device 122 displays the effective wavelength band and the quantification / identification results output by the spectrum analysis device 600 to the user.

[0067] Each device operates differently in the training phase, where a quantification / identification model is constructed and the effective wavelength band is determined based on the training sample, and in the estimation phase, where quantification / identification of the evaluation sample is performed. Therefore, the operation of each device will be explained separately for the training phase and the estimation phase.

[0068] <Operation of each device (training phase)> The control device 120 sets the wavelength band of the excitation light to be irradiated onto the training sample and the detection wavelength bands of reflected light, transmitted light, absorbed light, etc. to be detected from the sample in the spectrum analysis device 600. In the training phase, the entire excitation wavelength band and detection wavelength band that can be candidates for the effective wavelength band are set as the measurement wavelength band.

[0069] The spectrum measurement device 121 measures the spectrum of a training sample for obtaining training data, obtains discretized three-dimensional spectral data, and inputs it to the spectrum analysis device 600. The spectral analysis device 600 receives three-dimensional spectral data of a training sample measured by the spectrometer 121 and teaching information for the three-dimensional spectral data, and stores the data as a training data set in the memory 112 within the spectral analysis device 600. A quantitative / identification model is constructed using the training data set stored in the memory 112 by the method described in Example 2. Furthermore, by the method described in Example 3, an effective wavelength band is automatically determined, or information is presented to the user via the display device 122 to assist in determining the effective wavelength band. The determined effective wavelength band is sent to the control device 120 and set as the excitation wavelength band or detection wavelength band to be measured in the evaluation phase.

[0070] The display device 122 displays a GUI for presenting to the user the quantification / identification accuracy for the training data set output by the spectrum analysis device 600, the number of effective wavelength bands as shown in Fig. 8, specific wavelength bands, etc. Also, as described in the third embodiment, a GUI for assisting in determining effective wavelength bands as shown in Fig. 7 may be presented to the user.

[0071] <Operation of each device (evaluation phase)> The control device 120 sets the effective wavelength band output by the spectrum analysis device 600 in the training phase as the wavelength band of the excitation light to be irradiated onto the evaluation sample and the wavelength band of the reflected light, transmitted light, absorbed light, etc. to be detected from the sample. The spectrum measuring device 121 measures the spectrum of the evaluation sample corresponding to the wavelength set by the control device 120, obtains three-dimensional spectral data discretized in the same manner as in the training phase, and inputs the data to the spectrum analyzing device 600.

[0072] The spectrum analysis device 600 receives the three-dimensional spectral data of the evaluation sample measured by the spectrum measurement device 121, and performs quantification / identification using the quantification / identification model constructed in the training phase in the manner described in the first embodiment. The display device 122 presents to the user the quantification / identification results of the evaluation sample output by the spectrum analysis device 600. Furthermore, the display device 122 may visualize and display the acquired spectrum as needed.

[0073] <Modification> In each embodiment, the input unit 201 (601, 1101) may perform any preprocessing on the discretized three-dimensional spectral data after receiving the data. For example, the input unit 201 (601, 1101) may perform centering or standardization based on statistics such as the average or standard deviation for each wavelength band of the spectral data, frequency filtering such as a high-pass filter, a low-pass filter, or a band-pass filter, or linear / nonlinear functions. Furthermore, the detected intensity of a specific region, such as the aforementioned scattered light, or a detected intensity above a certain value may be replaced with a fixed value. Furthermore, such preprocessing may be performed by the wavelength-based feature extraction unit 202.

[0074] In Example 2, a method for generating a quantitative / discriminative model using a user's own data was described. However, any data augmentation method may be used to augment the training dataset provided by the user. Examples of data augmentation methods include adding random noise, performing a weighted linear sum of multiple spectral data, and varying the value range. A quantitative / discriminative model may be created by adding the augmented data to the training dataset, or the quantitative / discriminative model may be created using only the augmented data. Furthermore, the augmented data may be used as calibration data, or the calibration data may be augmented to perform a more detailed evaluation.

[0075] In each embodiment, the wavelength-unit feature extraction unit 202 may extract features by combining two or more detection intensities input in units of excitation wavelength or detection wavelength with other information, such as differential values ​​of spectral data, coordinate information of each detection intensity (information on pairs of excitation wavelength and detection wavelength), light absorption characteristics, experimental conditions, experimental environment, etc.

[0076] In the second embodiment, a method for generating a new quantitative / identification model for a quantitative / identification problem independently set by a user was described. However, calibration data may be prepared separately from the training data, and the newly generated model and models already stored in memory 112 may be evaluated. Based on the evaluation results, a model to be adopted may be determined. The evaluation results may include not only the quantitative / identification results but also the number of effective wavelength bands. For example, a method may be used to select a model with a quantitative / identification accuracy equal to or higher than a predetermined number of effective wavelength bands and the lowest number of effective wavelength bands, or a model with a quantitative / identification accuracy equal to or lower than a predetermined number of effective wavelength bands, or a method may be used to select a model based on some index calculated from the quantitative / identification accuracy and the number of effective wavelength bands. Alternatively, the evaluation results may be presented to the user, allowing the user to decide which model to adopt.

[0077] In Example 3, a method for searching for an effective wavelength band when the estimation unit 1103 constructs a new quantitative / identification model was described. However, it is also possible to search for an effective wavelength band using a quantitative / identification model stored in advance in the memory 112, and present the effective wavelength band to the user as shown in Figure 8.

[0078] In the fourth embodiment, a method has been described in which the three-dimensional spectral data measured by the spectrometer 121 is used as training data. However, three-dimensional spectral data prepared in advance may be stored in the memory 112 and used as part of the training data.

[0079] In each embodiment, the input unit 201 (601, 1101) may receive discretized three-dimensional spectral data and then perform scaling using nearest neighbor interpolation or linear interpolation. For example, in Example 4, if a quantitative / identification model is constructed using discretized fluorescence fingerprint data 300 measured at 1-nm intervals as a training data set, and then it becomes necessary to perform quantification / identification on evaluation data measured at 5-nm intervals, the input unit 201 may apply the quantitative / identification model generated from the training data set by performing scaling on the discretized fluorescence fingerprint data 300 of the evaluation data so that the height (number of excitation wavelength bands) and width (number of detection wavelength bands) are five times larger. The same applies when scaling is performed. Note that the scaling process may be performed by the estimation unit 203 (1103). For example, in the example of the discretized fluorescence fingerprint data 300, the estimation unit 203 (1103) may apply the quantification / identification model generated from the training data set by enlarging the excitation wavelength or detection wavelength unit feature of the discretized fluorescence fingerprint data 300 obtained by the wavelength unit feature extraction unit 202 so that the size in the wavelength band direction or the detection wavelength band direction is five times larger. The same applies when performing reduction processing.

[0080] As described above, the disclosed spectral analysis device includes an input unit 201 that receives spectral data of a sample as an input, a wavelength-unit feature extraction unit 202 that extracts features of the sample from the spectral data, an estimation unit 203 that estimates properties of the sample based on the features, and an output unit 204 that outputs the estimation results obtained by the estimation unit 203. The spectral data includes a plurality of excitation wavelengths and optical spectra indicating detection wavelengths and detection intensities for the excitation wavelengths, and is data discretized over a predetermined wavelength range, the wavelength-unit feature extraction unit 202 extracts features from the detection intensities of two or more detection wavelength bands in each excitation wavelength band, or from the detection intensities of two or more excitation wavelength bands in each detection wavelength band, and the estimation unit 203 performs estimation processing to quantify and / or identify properties of the sample. This configuration and operation makes it possible to utilize spatial information within the three-dimensional spectral data while achieving highly accurate quantification and / or identification with reduced noise effects, and to limit the wavelength bands used for quantification and / or identification in units of excitation wavelength or detection wavelength.

[0081] Furthermore, as disclosed in the second embodiment, the input unit 1101 receives one or more pairs of the spectral data and the teacher information corresponding to the spectral data, and the estimation unit 1103 can construct the quantification and / or identification model based on the features extracted from the spectral data by the wavelength-unit feature extraction unit 202 and the teacher information. With this configuration, it is possible to construct a model to be used for quantification and identification as appropriate.

[0082] Furthermore, as disclosed in the third embodiment, the apparatus may further include an effective wavelength band search unit 605 that searches for an effective wavelength band, which is a wavelength band corresponding to a feature used for the quantification and / or identification, based on the parameters of the quantification and / or identification model obtained by the estimation unit 1103. The effective wavelength band search unit 605 searches for an effective wavelength band in units of excitation wavelengths when the wavelength-unit feature extraction unit 202 has extracted a feature for each detection intensity of two or more detection wavelength bands in each excitation wavelength band, or in units of detection wavelengths when the wavelength-unit feature extraction unit 202 has extracted a feature for each detection intensity of two or more excitation wavelength bands in each detection wavelength band, and the output unit 604 outputs the effective wavelength band searched by the effective wavelength band search unit 605 in addition to the estimation result. According to this configuration and operation, in addition to the results of quantification and identification, information on wavelength bands effective for quantification and identification can be obtained.

[0083] Furthermore, the wavelength-unit feature extraction unit 202 may extract features that can be commonly used for a plurality of quantifications and / or identifications by analyzing or learning from a plurality of training data sets. According to this operation, it is possible to obtain feature amounts that can be used in common for various types of estimation, and various types of estimation can be performed efficiently.

[0084] In addition, the effective wavelength band search unit 605 acquires, as evaluation information, the quantification and / or identification accuracy and the number of effective wavelength bands when the wavelength-unit feature extraction unit 202 extracts features for each detection intensity of two or more detection wavelength bands in each excitation wavelength band, and when the wavelength-unit feature extraction unit 202 extracts features for each detection intensity of two or more excitation wavelength bands in each detection wavelength band, and automatically determines the effective wavelength band based on the evaluation information, or supports the user in determining the effective wavelength band by presenting the evaluation information to the user. Therefore, quantification and identification can be performed efficiently using an appropriate number of effective wavelength bands according to the desired identification accuracy.

[0085] In addition, the effective wavelength band search unit 605 acquires, as evaluation information, the relationship between the quantification and / or identification accuracy and the number of effective wavelength bands when a hyperparameter that affects the quantification and / or identification accuracy and the number of effective wavelength bands in the estimation unit 1103 is changed, and automatically determines the value of the hyperparameter based on the evaluation information, or presents the evaluation information to the user to assist the user in determining the value of the hyperparameter. Therefore, hyperparameters can be appropriately set according to the desired classification accuracy, enabling efficient quantification and classification.

[0086] The input unit 601 also receives the effective wavelength band output by the effective wavelength band search unit 605, and outputs only the spectrum corresponding to the effective wavelength band from the spectral data received as input to the wavelength-unit feature extraction unit 202. Therefore, the effective wavelength band can be applied to the subsequent feature extraction processing and estimation processing, thereby realizing the efficiency of the feature extraction processing and estimation processing.

[0087] The disclosed system also includes a control device 120 that sets the wavelength band of excitation light to be irradiated onto a sample and the wavelength band of detection light, a spectroscopic measurement device 121 that measures spectral data for the sample corresponding to the wavelength bands set by the control device, a spectral analysis device 600 that analyzes the spectral data measured by the spectroscopic measurement device 121, and a display device 122 that displays estimation results obtained by the spectral analysis device. The spectral analysis device 600 includes an input unit 601 that receives spectral data of the sample as input, a wavelength-unit feature extraction unit 202 that extracts feature quantities of the sample from the spectral data, an estimation unit 1103 that estimates characteristics of the sample based on the feature quantities, and an output unit 604 that outputs the estimation results obtained by the estimation unit 1103. The spectral data includes a plurality of excitation wavelengths and a spectrum indicating detection wavelengths and detection intensities for the excitation wavelengths, and is data discretized at a predetermined wavelength width. The wavelength-unit feature extraction unit 202 extracts features from the detection intensities of two or more detection wavelength bands in each excitation wavelength band, or from the detection intensities of two or more excitation wavelength bands in each detection wavelength band, and the estimation unit 1103 performs estimation processing to quantify and / or identify the characteristics of the sample. In this way, a system having a spectral analysis device can utilize the spatial information in the three-dimensional spectral data to achieve highly accurate quantification and / or identification with reduced noise effects, and can limit the wavelength band used for quantification and / or identification in units of excitation wavelength or detection wavelength.

[0088] The spectral analysis device 600 further includes an effective wavelength band search unit 605 that searches for an effective wavelength band, which is a wavelength band corresponding to a feature used for the quantification and / or identification, based on the parameters of the quantification and / or identification model obtained by the estimation unit 1103. The effective wavelength band search unit 605 searches for an effective wavelength band in excitation wavelength units if the wavelength-unit feature extraction unit 202 has extracted a feature for each detection intensity of two or more detection wavelength bands in each excitation wavelength band, or in detection wavelength units if the wavelength-unit feature extraction unit 202 has extracted a feature for each detection intensity of two or more excitation wavelength bands in each detection wavelength band. The output unit 604 outputs the effective wavelength band searched by the effective wavelength band search unit 605 in addition to the estimation result. The control device 120 receives the effective wavelength band output by the spectral analysis device 600 and sets the effective wavelength band as the wavelength band of excitation light to be irradiated on the sample or the wavelength band of detection light to be detected. According to this configuration and operation, the time required for spectrum measurement device 121 to acquire spectrum data can be reduced.

[0089] The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, not only can the configurations be deleted, but also replacements and additions of configurations are possible. [Explanation of symbols]

[0090] 100: Spectral analysis device, 110: Interface unit, 111: Calculation unit, 112: Memory, 113: Bus, 120: Control device, 121: Spectral measurement device, 122: Display device, 201: Input unit, 202: Wavelength unit feature extraction unit, 203: Estimation unit, 204: Output unit, 300: Discretized fluorescence fingerprint data, 301: Example of visualization of fluorescence fingerprint data, 500: Spectral analysis method, 501: Input step, 502: Wavelength unit feature extraction step, 503: Estimation step, 504: output step, 600: spectrum analysis device, 601: input unit, 604: output unit, 605: effective wavelength band search unit, 900: spectrum analysis method, 901: input step, 904: output step, 905: effective wavelength band search step, 1000: spectrum analysis system, 1100: spectrum analysis device, 1101: input unit, 1103: estimation unit, 1200: spectrum analysis method, 1201: input step, 1203: estimation step

Claims

1. an input unit that accepts spectral data of a sample as input; a wavelength-unit feature extraction unit that extracts a feature of the sample from the spectrum data; an estimation unit that estimates properties of the sample based on the feature amount; an output unit that outputs the estimation result obtained by the estimation unit; Equipped with the spectral data includes a plurality of excitation wavelengths and a spectrum indicating a plurality of detection wavelengths and detection intensities for each of the plurality of excitation wavelengths, and is discretized at a predetermined wavelength width; The wavelength-unit feature extraction unit For each of the plurality of excitation wavelengths, a distribution of the detection intensity in an entire region or a part of a region of a detection wavelength band formed by the plurality of detection wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each excitation wavelength; or For each of a plurality of detection wavelengths, a distribution of the detection intensity in an entire region or a part of a region of an excitation wavelength band formed by the plurality of excitation wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each detection wavelength; the estimation unit performs an estimation process to quantify and / or identify characteristics of the sample; an effective wavelength band search unit that searches for an effective wavelength band that is a wavelength band corresponding to a feature used for the quantification and / or identification based on parameters of the quantification and / or identification model obtained by the estimation unit; the effective wavelength band search unit searches for an effective wavelength band in units of excitation wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each excitation wavelength, or in units of detection wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each detection wavelength; the output unit outputs the effective wavelength band searched by the effective wavelength band search unit in addition to the estimation result; The effective wavelength band search unit acquires, as evaluation information, the quantification and / or identification accuracy and the number of effective wavelength bands when the wavelength-unit feature extraction unit extracts features from the one-dimensional data independently for each excitation wavelength and when the wavelength-unit feature extraction unit extracts features from the one-dimensional data independently for each detection wavelength, and automatically determines the effective wavelength band based on the evaluation information, or presents the evaluation information to the user to assist the user in determining the effective wavelength band.

2. an input unit that accepts spectral data of a sample as input; a wavelength-unit feature extraction unit that extracts a feature of the sample from the spectrum data; an estimation unit that estimates properties of the sample based on the feature amount; an output unit that outputs the estimation result obtained by the estimation unit; Equipped with the spectral data includes a plurality of excitation wavelengths and a spectroscopic spectrum indicating a plurality of detection wavelengths and detection intensities for each of the plurality of excitation wavelengths, and is discretized at a predetermined wavelength width; The wavelength-unit feature extraction unit For each of the plurality of excitation wavelengths, a distribution of the detection intensity in an entire region or a part of a region of a detection wavelength band formed by the plurality of detection wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each excitation wavelength; or For each of a plurality of detection wavelengths, a distribution of the detection intensity in an entire region or a part of a region of an excitation wavelength band formed by the plurality of excitation wavelengths is acquired as one-dimensional data, and feature extraction is performed independently for each detection wavelength from the one-dimensional data; the estimation unit performs an estimation process to quantify and / or identify characteristics of the sample; an effective wavelength band search unit that searches for an effective wavelength band that is a wavelength band corresponding to a feature used for the quantification and / or identification based on the parameters of the quantification and / or identification model obtained by the estimation unit; the effective wavelength band search unit searches for an effective wavelength band in units of excitation wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each excitation wavelength, or in units of detection wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each detection wavelength; the output unit outputs the effective wavelength band searched by the effective wavelength band search unit in addition to the estimation result; The effective wavelength band search unit acquires, as evaluation information, the relationship between the quantification and / or identification accuracy and the number of effective wavelength bands when the parameters that affect the quantification and / or identification accuracy and the number of effective wavelength bands in the estimation unit are changed, and automatically determines the values ​​of the parameters based on the evaluation information, or presents the evaluation information to a user to assist the user in determining the values ​​of the parameters.

3. A spectral analysis device according to claim 1 or 2, the input unit receives one or more pairs of the spectral data and teacher information corresponding to the spectral data; The spectral analysis device is characterized in that the estimation unit constructs the quantitative and / or identification model based on the features extracted from the spectral data by the wavelength-unit feature extraction unit and the teaching information.

4. A spectrum analysis device according to claim 1 or 2, The spectral analysis device is characterized in that the wavelength-unit feature extraction unit extracts features that can be commonly used for multiple quantifications and / or identifications by analyzing or learning from multiple training data sets.

5. A spectral analysis device according to claim 1 or 2, the input unit receives the effective wavelength band output by the effective wavelength band search unit, and outputs only a spectrum corresponding to the effective wavelength band from the spectral data received as input to the wavelength-unit feature extraction unit.

6. an input step of accepting spectral data of a sample as input; a wavelength-unit feature extraction step of extracting a feature of the sample from the spectrum data; an estimation step of estimating properties of the sample based on the feature amount; an output step of outputting the estimation result obtained by the estimation step; Including, the spectral data includes a plurality of excitation wavelengths and a spectroscopic spectrum indicating a plurality of detection wavelengths and detection intensities for each of the plurality of excitation wavelengths, and is discretized at a predetermined wavelength width; The wavelength-unit feature extraction step includes: For each of the plurality of excitation wavelengths, a distribution of the detection intensity in an entire region or a part of a region of a detection wavelength band formed by the plurality of detection wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each excitation wavelength; or a distribution of the detection intensity in an entire region or a part of a region of an excitation wavelength band formed by the plurality of excitation wavelengths is acquired as one-dimensional data for each of the plurality of detection wavelengths, and a feature amount is extracted from the one-dimensional data independently for each detection wavelength; The estimation step includes performing an estimation process to quantify and / or identify characteristics of the sample; The method further includes an effective wavelength band searching step of searching for an effective wavelength band that is a wavelength band corresponding to a feature used for the quantification and / or identification based on the parameters of the quantification and / or identification model obtained by the estimation step, the effective wavelength band searching step searches for an effective wavelength band in units of excitation wavelengths if feature extraction from the one-dimensional data is performed independently for each excitation wavelength in the wavelength-unit feature extraction step, or in units of detection wavelengths if feature extraction from the one-dimensional data is performed independently for each detection wavelength; the output step outputs the effective wavelength band searched in the effective wavelength band search step in addition to the estimation result; The effective wavelength band search step obtains, as evaluation information, the quantitative and / or identification accuracy and the number of effective wavelength bands when features are extracted from the one-dimensional data independently for each excitation wavelength in the wavelength-unit feature extraction step, and when features are extracted from the one-dimensional data independently for each detection wavelength, and automatically determines the effective wavelength band based on the evaluation information, or supports the user in determining the effective wavelength band by presenting the evaluation information to the user.

7. an input step of accepting spectral data of a sample as input; a wavelength-unit feature extraction step of extracting a feature of the sample from the spectrum data; an estimation step of estimating properties of the sample based on the feature amount; an output step of outputting the estimation result obtained by the estimation step; Including, the spectral data includes a plurality of excitation wavelengths and a spectroscopic spectrum indicating a plurality of detection wavelengths and detection intensities for each of the plurality of excitation wavelengths, and is discretized at a predetermined wavelength width; The wavelength-unit feature extraction step includes: For each of the plurality of excitation wavelengths, a distribution of the detection intensity in an entire region or a part of a region of a detection wavelength band formed by the plurality of detection wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each excitation wavelength; or a distribution of the detection intensity in an entire region or a part of a region of an excitation wavelength band formed by the plurality of excitation wavelengths is acquired as one-dimensional data for each of the plurality of detection wavelengths, and a feature amount is extracted from the one-dimensional data independently for each detection wavelength; The estimation step includes performing an estimation process to quantify and / or identify characteristics of the sample; an effective wavelength band searching step of searching for an effective wavelength band that is a wavelength band corresponding to a feature used for the quantification and / or identification based on the parameters of the quantification and / or identification model obtained by the estimation step; the effective wavelength band searching step searches for an effective wavelength band in units of excitation wavelengths if feature extraction from the one-dimensional data is performed independently for each excitation wavelength in the wavelength-unit feature extraction step, or in units of detection wavelengths if feature extraction from the one-dimensional data is performed independently for each detection wavelength; the output step outputs the effective wavelength band searched in the effective wavelength band search step in addition to the estimation result; The effective wavelength band search step acquires, as evaluation information, the relationship between the quantification and / or identification accuracy and the number of effective wavelength bands when the parameters that affect the quantification and / or identification accuracy and the number of effective wavelength bands in the estimation step are changed, and automatically determines the values ​​of the parameters based on the evaluation information, or presents the evaluation information to a user to assist the user in determining the values ​​of the parameters.

8. A spectral analysis method according to claim 6 or 7, the input step includes receiving one or more pairs of the spectral data and teacher information corresponding to the spectral data; A spectral analysis method characterized in that the estimation step constructs the quantitative and / or discriminative model based on the features extracted from the spectral data in the wavelength-unit feature extraction step and the teaching information.

9. A spectral analysis method according to claim 6 or 7, A spectral analysis method characterized in that the wavelength-unit feature extraction step extracts features that can be commonly used for multiple quantifications and / or identifications by analyzing or learning from multiple training data sets.

10. a control device that sets the wavelength band of the excitation light to be irradiated onto the sample and the wavelength band of the detection light; a spectrometer for measuring spectral data corresponding to the wavelength band set by the control device for the sample; a spectrum analyzer that analyzes the spectrum data measured by the spectrum measuring device; a display device that displays the estimation result obtained by the spectrum analysis device; and The spectrum analysis device an input unit that accepts spectral data of a sample as input; a wavelength-unit feature extraction unit that extracts a feature of the sample from the spectrum data; an estimation unit that estimates properties of the sample based on the feature amount; an output unit that outputs the estimation result obtained by the estimation unit; Equipped with the spectral data includes a plurality of excitation wavelengths and a spectrum indicating a plurality of detection wavelengths and detection intensities for each of the plurality of excitation wavelengths, and is discretized at a predetermined wavelength width; The wavelength-unit feature extraction unit For each of the plurality of excitation wavelengths, a distribution of the detection intensity in an entire region or a part of a region of a detection wavelength band formed by the plurality of detection wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each excitation wavelength; or a distribution of the detection intensity in an entire region or a part of a region of an excitation wavelength band formed by the plurality of excitation wavelengths is acquired as one-dimensional data for each of the plurality of detection wavelengths, and a feature amount is extracted from the one-dimensional data independently for each detection wavelength; the estimation unit performs an estimation process to quantify and / or identify characteristics of the sample; The spectrum analysis device further comprising an effective wavelength band search unit that searches for an effective wavelength band that is a wavelength band corresponding to a feature used for the quantification and / or identification based on the parameters of the quantification and / or identification model obtained by the estimation unit; the effective wavelength band search unit searches for an effective wavelength band in units of excitation wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each excitation wavelength, or in units of detection wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each detection wavelength; the output unit outputs the effective wavelength band searched by the effective wavelength band search unit in addition to the estimation result; The effective wavelength band search unit acquires, as evaluation information, the quantification and / or identification accuracy and the number of effective wavelength bands when the wavelength-unit feature extraction unit extracts features from the one-dimensional data independently for each excitation wavelength and when the wavelength-unit feature extraction unit extracts features from the one-dimensional data independently for each detection wavelength, and automatically determines the effective wavelength band based on the evaluation information, or presents the evaluation information to the user to assist the user in determining the effective wavelength band.

11. a control device that sets the wavelength band of the excitation light to be irradiated onto the sample and the wavelength band of the detection light; a spectrometer for measuring spectral data corresponding to the wavelength band set by the control device for the sample; a spectrum analyzer that analyzes the spectrum data measured by the spectrum measuring device; a display device that displays the estimation result obtained by the spectrum analysis device; and The spectrum analysis device an input unit that accepts spectral data of a sample as input; a wavelength-unit feature extraction unit that extracts a feature of the sample from the spectrum data; an estimation unit that estimates properties of the sample based on the feature amount; an output unit that outputs the estimation result obtained by the estimation unit; Equipped with the spectral data includes a plurality of excitation wavelengths and a spectroscopic spectrum indicating a plurality of detection wavelengths and detection intensities for each of the plurality of excitation wavelengths, and is discretized at a predetermined wavelength width; The wavelength-unit feature extraction unit For each of the plurality of excitation wavelengths, a distribution of the detection intensity in an entire region or a part of a region of a detection wavelength band formed by the plurality of detection wavelengths is acquired as one-dimensional data, and feature extraction is performed from the one-dimensional data independently for each excitation wavelength; or a distribution of the detection intensity in an entire region or a part of a region of an excitation wavelength band formed by the plurality of excitation wavelengths is acquired as one-dimensional data for each of the plurality of detection wavelengths, and a feature amount is extracted from the one-dimensional data independently for each detection wavelength; the estimation unit performs an estimation process to quantify and / or identify characteristics of the sample; The spectrum analysis device further comprising an effective wavelength band search unit that searches for an effective wavelength band that is a wavelength band corresponding to a feature used for the quantification and / or identification based on the parameters of the quantification and / or identification model obtained by the estimation unit; the effective wavelength band search unit searches for an effective wavelength band in units of excitation wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each excitation wavelength, or in units of detection wavelengths when the wavelength-unit feature extraction unit performs feature extraction from the one-dimensional data independently for each detection wavelength; the output unit outputs the effective wavelength band searched by the effective wavelength band search unit in addition to the estimation result; The effective wavelength band search unit acquires, as evaluation information, the relationship between the quantification and / or identification accuracy and the number of effective wavelength bands when the parameters that affect the quantification and / or identification accuracy and the number of effective wavelength bands in the estimation unit are changed, and automatically determines the values ​​of the parameters based on the evaluation information, or presents the evaluation information to a user to assist the user in determining the values ​​of the parameters.

Images