A spectral model matching method based on component absorption peaks

Abstract

The present invention mainly relates to the field of near-infrared spectroscopy. In order to solve the problem that the consistency of different near-infrared spectroscopy devices of the same model is poor and the spectral models established by different devices are not universal, resulting in unstable and inaccurate prediction results of the content of certain components of the measured spectrum, the present invention provides a spectral model matching method based on component absorption peaks, the core idea of ​​which is: first, multiple near-infrared spectroscopy devices of the same model are used to collect spectral data of samples of known components, and then the spectral data of the samples to be tested are reconstructed according to the wavelength range and number of the component absorption peaks, and a reconstructed spectral model library is established, and then the characterization coefficients of the absorption peaks of different components of the sample reconstructed spectral data are calculated, and the matching deviation value is calculated in combination with the characterization coefficient, and the spectral model is accurately matched according to the matching deviation value, and finally the matching spectral model is called to predict and analyze the content of the components of unlabeled samples to complete the spectral model matching analysis.

Description

Technical Field

[0001] The invention relates to the technical field of infrared spectroscopy, and in particular to a spectral model matching method based on component absorption peaks. Background Art

[0002] With the development of micro-electromechanical technology, in recent years, miniaturized near-infrared spectrometers have been widely used in various fields such as food, beverages, medicine, coal, etc. Compared with traditional large-scale Fourier transform spectral analysis systems, they have the advantages of simple structure, low cost, easy to carry, fast detection, and real-time visible results. At present, they have become a popular product in the field of spectroscopy. In particular, miniaturized near-infrared spectrometers have the ability to predict multi-component indicators in parallel. Only one spectral data of the sample to be tested needs to be collected. Through specific algorithm processing, the parallel prediction of the multi-component content of the sample to be tested can be achieved. The component content of the sample can be predicted at the same time in a very short time. For example, if the sample to be tested is Sichuan pepper, the content of components such as Sichuan pepper moisture, volatile oil, and hemp can be obtained at the same time. If the sample to be tested is wheat, the content of components such as wheat moisture, protein, and starch can be obtained at the same time. The application scenarios of spectral detection technology are greatly enhanced by the detection characteristics of miniaturized near-infrared spectrometer equipment.

[0003] Compared with large-scale near-infrared spectroscopic equipment in laboratories, miniaturized near-infrared spectrometers are affected by light sources, detectors, manufacturing processes, usage methods, environmental conditions, etc. The consistency, stability, reproducibility and other indicators of the miniaturized near-infrared spectrometers are significantly reduced, especially the consistency of the equipment. Different miniaturized near-infrared spectroscopic equipment of the same model has poor consistency, and the spectral models established by different equipment are not universal. In practical applications, each miniaturized near-infrared spectroscopic equipment needs to be modeled separately, which wastes a lot of manpower and material resources and greatly hinders the development of miniaturized near-infrared spectroscopic analysis technology.

[0004] At present, the conventional method used in the market to deal with the problem of non-universality of spectral models of miniaturized near-infrared spectroscopy equipment is spectral model transfer, and its algorithms include slope-intercept method (S / B), Shenks algorithm, direct standardization (DS) and piecewise standardization (PDS). Although these algorithms can solve the problem of non-universality of spectral models to a certain extent, for the prediction of component content of multi-component samples, due to the singleness of the model transfer algorithm, targeted spectral model transfer will not be performed for each component, resulting in the problem that the prediction effect of some component contents is better and the prediction effect of some component contents is poor. For example, for pepper samples, the prediction effect after processing with the model transfer algorithm is prone to the problem that the prediction effect of water content is better, but the prediction effect of cannabinoids and volatile oil contents is poor. Summary of the invention

[0005] The technical problems to be solved by the present invention are:

[0006] A spectral model matching method based on component absorption peaks is provided to solve the problem that the prediction results of certain component contents are unstable and inaccurate due to the poor consistency of different near-infrared spectral equipment of the same model and the lack of universality of spectral models established by different equipment.

[0007] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0008] A spectrum model matching method based on component absorption peaks comprises the following steps:

[0009] Step 1: Use multiple spliced ​​multi-sensor near-infrared spectrometers of the same model to collect spectral data of the sample;

[0010] Step 2: Reconstruct the sample spectrum data according to the component absorption peak range and number, and establish a reconstructed spectrum model library;

[0011] Step 3: Calculate the characterization coefficients of the absorption peaks of different components of the reconstructed spectral data;

[0012] Step 4: Calculate the matching deviation value according to the characterization coefficient, and perform accurate matching of the spectral model according to the matching deviation value;

[0013] Step 5: Call the matching spectral model to perform prediction analysis on the component content of the sample with unknown component content, and complete the spectral model matching analysis.

[0014] Furthermore, the near-infrared spectroscopy device described in step 1 is randomly divided into two parts, one part is used to establish a spectral model library, and the other part is used to match the spectral model.

[0015] Furthermore, there are N devices for establishing a spectral model library and 1 device for matching the spectral model. The N spectral devices for establishing the spectral model library are used to collect spectra of M samples with known component contents respectively, and the N devices collect a total of (M×N) spectral data; the 1 device for matching the spectral model is used to collect the spectrum of a single sample with unknown component content.

[0016] Furthermore, the specific method of step 2 is: select X absorption peaks according to the spectral absorption characteristics of the sample components, and reconstruct the multi-sensor spectral data into single-sensor spectral data based on the X absorption peaks. The specific reconstruction method is: calculate the number of wavelength points contained in each single-sensor spectral data according to the wavelength range and resolution of the single-sensor near-infrared device, calculate the number of wavelength points and the spectral wavelength range contained in each absorption peak in combination with the number of selected absorption peaks, select the same number of wavelength points and wavelength range from the multi-sensor spectral data to reconstruct it into single-sensor spectral data, and complete the data reconstruction;

[0017] Each of the N devices for establishing spectral models collects M spectral data. After reconstructing the M data, they are matched one by one with the component contents of M samples with known component calibration values. The partial least squares method is used for modeling to obtain a single component spectral model. A total of N component spectral models are established for the N devices.

[0018] Furthermore, the specific method of step 3 is: obtain the absorbance of the wavelength point included in each absorption peak, add the absorbance of the wavelength point included in each absorption peak, and make a quotient with the absorbance of the wavelength point within the spectral wavelength range included in all absorption peaks, so as to obtain the characterization coefficients corresponding to the X absorption peaks of the reconstructed spectral data, and record the characterization coefficients as α1, α2, ..., α X .

[0019] Furthermore, the specific method of step 4 is: firstly, the mean value of the reconstructed spectral data of M samples collected by a single device in the spectral model library is calculated, and the absorbance of the M samples in a certain wavelength range included in the absorption peak is respectively (k1, k2, ... k M ), perform averaging on it, and obtain the mean value of M absorbances at this wavelength:

[0020]

[0021] Use the same method to calculate the mean absorbance of all wavelength points included in each absorption peak, and record them as L represents the number of wavelength points contained in each absorption peak;

[0022] Calculate the matching deviation value of the reconstructed spectrum data of the sample with unknown component content compared with the reconstructed spectrum data of the sample with known component content, wherein the calculation method of the matching deviation value c1 of the reconstructed spectrum data of the sample with unknown component content acquired by the spectrum model device to be matched compared with the reconstructed spectrum data of the sample with known component content acquired by the first spectrum database establishment device is:

[0023]

[0024] Among them, K1 i , K2 i ,……,KX i are the absorbances of the samples with unknown component content and the samples with known component content at the corresponding wavelengths in each absorption peak;

[0025] Similarly, the matching deviations of the reconstructed spectral data of the sample with unknown component content acquired by the spectral model equipment to be matched and the reconstructed spectral data of the sample with known component content acquired by the N spectral database establishment equipment are calculated respectively (c1, c2, ... c N), from the matching deviation value (c1, c2, ... c N ) are selected as matching devices, and the spectral model based on the reconstructed spectral data in the selected devices is directly matched to the spectral model device to be matched, so as to complete the spectral model matching between the two types of devices.

[0026] Beneficial effects of the present invention:

[0027] The spectral model matching method based on component absorption peaks described in the present invention reconstructs spectral data according to the wavelength range and number of component absorption peaks, reduces the amount of spectral data while retaining the component characterization capability of the sample to be tested, improves the efficiency of spectral analysis, and simultaneously calculates the matching deviation value through the characterization coefficients between different absorption peaks to achieve accurate matching of spectral models between different spectral devices, which not only solves the problem that spectral models of miniaturized near-infrared spectral devices are not universal, but also matches different spectral models for accurate prediction of different component contents of the sample to be tested, effectively improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a spectral model matching method based on component absorption peaks described in the present invention. DETAILED DESCRIPTION

[0029] like Figure 1 As shown, a spectral model matching method based on component absorption peaks described in the present invention first uses multiple miniaturized near-infrared devices of the same model to collect spectral data of the sample to be tested, then reconstructs the spectral data of the sample to be tested according to the wavelength range and number of the component absorption peaks, and establishes a reconstructed spectral model library, then calculates the characterization coefficients of different component absorption peaks of the reconstructed spectral data of the sample to be tested, calculates the matching deviation value in combination with the characterization coefficient, accurately matches the spectral model according to the matching deviation value, and finally calls the matching spectral model to predict and analyze the component content of the unknown group component sample to complete the spectral model matching analysis.

[0030] Example:

[0031] This embodiment takes the near-infrared spectrum detection of water as an example to specifically introduce the spectrum model matching method based on component absorption peaks of the present invention;

[0032] In this embodiment, a plurality of miniaturized near-infrared spectrometers of the same model are specifically classified as (N+1), wherein N devices are spectral model library establishment devices, and 1 device is a spectral model device to be matched, which are used to implement the spectral model matching method based on component absorption peaks. In specific practical applications, the number of spectral model library establishment devices is selected to be 10. Too many devices will consume a lot of manpower and material resources for spectral data collection and modeling. Too few devices will cover fewer spectral models, resulting in lower matching accuracy and larger prediction errors. The number of spectral model devices to be matched can be (1~+∞), and the spectral model matching of all devices can be achieved by processing them using the same spectral model matching method. The specific number of samples with known component content and samples with unknown component content is (M+1), of which M samples have known component content and 1 sample has unknown component content. In specific practical applications, the number of samples with known component content is 200. The spectral model established by this number of samples has high stability and strong prediction ability, which meets the requirements of miniaturized near-infrared spectral analysis technology. Too many samples will cause waste of manpower and material resources, and the ability to improve the effect of the spectral model is very limited. The spectral model established by too few samples has poor stability, is very easy to overfit, and has weak prediction ability; the number of samples with unknown component content can be (1~+∞), and the spectral data can be collected and processed by the processing method of a single unknown component content sample to complete the composition prediction of all samples with unknown component content. The specific collection method is: use N devices to collect M spectra of samples with known component content respectively, and a single device collects M spectral data, and a total of (M×N) spectral data are collected; use 1 spectral model device to be matched to collect spectral data of 1 sample with unknown component content to obtain a single spectral data.

[0033] The commonly used single-sensor miniaturized near-infrared spectroscopy equipment covers a band range of 410 nanometers. The miniaturized near-infrared spectroscopy equipment used in this embodiment is a multi-sensor splicing type. The equipment includes 4 single sensors with band ranges of 800nm~1210nm, 1220nm~1630nm, 1640nm~2050nm, and 2060nm~2470nm. The use of a spliced ​​multi-sensor miniaturized near-infrared spectroscopy equipment can expand the single sensor equipment that initially only covers a band range of 410 nanometers to cover 1670 nanometers, which almost covers the entire near-infrared band range. This can effectively expand the band range of the equipment and cover the absorption peaks of various components of the sample to be tested as comprehensively as possible.

[0034] The spectral data of the samples to be tested are reconstructed according to the wavelength range and number of the component absorption peaks, and a reconstructed spectral model library is established. The component composition of different samples to be tested is different. For different components of the samples to be tested, the spectral data is reconstructed according to the wavelength range and number of the absorption peaks of the components. Only a certain number of wavelength points around the absorption peak are retained to replace the original spectral data. The amount of spectral data is greatly reduced while retaining the characteristic information of the spectral data, thereby improving the efficiency of spectral analysis.

[0035] In this embodiment, in the practical application of near-infrared spectroscopy rapid detection, taking the water absorption peak as an example, in the near-infrared band range, water has strong absorption peaks at wavelengths of 1940nm, 1450nm, 1190nm, and 970nm, of which 1450nm and 970nm are the double frequency absorption bands of OH, and 1940nm and 1190nm are the combined frequency absorption bands of OH. In the near-infrared absorption peak of water, the absorption rate at the wavelength of 1190nm is small, which is generally used for water measurement with high humidity content. The absorption rate at the wavelength of 970nm is too large, and the measurement range is narrow, which is generally suitable for small-range high-precision water content measurement. The applicability of the above two wavelengths is greatly limited, and the absorption rates of the two wavelengths of 1940nm and 1450nm are more suitable, which are often used for water measurement in various fields. Therefore, in this patent, the wavelength range of the water absorption peak is selected as 1940nm and 1450nm, and the number is 2.

[0036] Reconstructing the spliced ​​multi-sensor spectral data and reconstructing the multi-sensor spectral data into single-sensor spectral data can not only effectively reduce the amount of spectral data, but also use conventional single-sensor spectral data modeling methods for spectral modeling, effectively improving the universality of spectral data and spectral analysis efficiency. The resolution of the miniaturized spliced ​​multi-sensor spectral equipment used in this patent is 10nm, and the band range is 800nm~2470nm. It can be calculated that the wavelength points contained in each multi-sensor spectral data are X1=1+(2470-800) / 10=168, and the band range of a single sensor is 410nm. It can be calculated that the wavelength points contained in each single-sensor spectral data are X2=1+410 / 10=42. At the same time, combined with the wavelength range of the water absorption peak of 1940nm and 1450nm, the number is 2, then it can be further calculated that the number of wavelength points distributed in the two absorption peaks of the reconstructed spectral data based on the water absorption peak is X3=42 / 2=21, the band range with a wavelength of 1940nm water absorption peak as the center point and including 21 wavelength points is 1840nm~2040nm, and the band range with a wavelength of 1450nm water absorption peak as the center point and including 21 wavelength points is 1350nm~1550nm. From the above, it can be obtained that the band range of the reconstructed spectral data includes two small bands, namely 1350nm~1550nm and 1840nm~2040nm, which contains a total of 42 wavelength points. Compared with the 168 wavelength points of the multi-sensor spectral data, on the basis of retaining the moisture characterization capability of the spectral data, the amount of spectral data is greatly reduced, and the efficiency of spectral analysis is improved.

[0037] Furthermore, the (M×N) spectral data collected by the spectral model building device are spectrally reconstructed, and a reconstructed spectral model library is established based on the reconstructed spectral data. For a single device among the N devices, M spectral data are collected, and the reconstructed spectral data are also M. The reconstructed spectral data are matched one by one with the M known moisture calibration values, and the partial least squares method is used for spectral modeling to obtain a single moisture spectral model. Similarly, the same modeling process is performed on the remaining (N-1) spectral model library building devices to obtain (N-1) moisture spectral models. Therefore, the N spectral model library building devices have established N moisture spectral models, that is, the spectral model library contains N moisture spectral models.

[0038] In this embodiment, the characterization coefficient of the absorption peak of different components is actually expressed as the characterization ability of the component by the reconstructed small-band spectral data based on the absorption peak of the component. Since the moisture content of the sample with unknown component content is predicted in the actual application, the spectrum data of a single sample with unknown component content collected by the spectral model device to be matched is used as a benchmark when calculating the characterization coefficient. The band ranges of the reconstructed small-band spectral data based on the moisture absorption peak are 1350nm~1550nm and 1840nm~2040nm respectively. The absorbances of the reconstructed spectral data at 21 wavelength points in the band range of 1350nm~1550nm are set to be (W1, W2, ... W 21 ), the absorbances at 21 wavelength points in the wavelength range of 1840nm to 2040nm are (S1, S2, ... S 21 ), since the absorbance value directly reflects the absorption capacity of water at this wavelength to near-infrared light, the larger the absorbance value, the stronger the ability of the water absorption peak to characterize the water content. The calculation formulas for calculating the characterization coefficients α1 and α2 corresponding to the two water absorption peaks of the reconstructed spectral data are:

[0039]

[0040] Spectral model matching is performed based on the matching deviation value. For the same component content, the smaller the matching deviation value of the spectral data reconstructed by the two types of miniaturized near-infrared spectroscopy devices, the more consistent the trend between the two spectral data, and thus it can be seen that their spectral models are also tending to be consistent. Therefore, the two devices with the smallest matching deviation values ​​in the two types of devices are selected for spectral model matching.

[0041] In this embodiment, the matching deviation value is actually expressed as the deviation value between the spectral data reconstructed by the spectral model library establishment device for the sample to be tested and the spectral data reconstructed by the spectral model to be matched device for the sample to be tested. First, the average value of the spectral data reconstructed by a single device in the spectral model library establishment device for M samples to be tested is calculated. The specific implementation process is: taking the wavelength range of 1350nm as an example, the absorbance values ​​of the M samples in this wavelength range are set to be (k1, k2, ..., k M ), perform averaging on it, and obtain the mean value of M absorbances in this wavelength range for:

[0042]

[0043] Similarly, the same data processing method is performed on each wavelength point in the band range of 1350nm~1550nm and 1840nm~2040nm, and the average absorbance corresponding to these wavelength points is obtained as and

[0044] Calculate the matching deviation value of the reconstructed spectrum data of the sample with unknown component content compared with the reconstructed spectrum data of the sample with known component content, wherein the calculation method of the matching deviation value c1 of the reconstructed spectrum data of the sample with unknown component content acquired by the spectrum model device to be matched compared with the reconstructed spectrum data of the sample with known component content acquired by the first spectrum database establishment device is:

[0045]

[0046] Among them, K1 i , K2 i ,……,KX i are the absorbances of the samples with unknown component content and the samples with known component content at the corresponding wavelengths in each absorption peak;

[0047] Similarly, the matching deviations of the reconstructed spectral data of the sample with unknown component content acquired by the spectral model equipment to be matched and the reconstructed spectral data of the sample with known component content acquired by the N spectral database establishment equipment are calculated respectively (c1, c2, ... c N ).

[0048] For the reconstructed spectral data, the smaller the trend offset of the reconstructed spectral data collected by the two types of near-infrared spectral equipment is, the more consistent the trends between the two spectral data are, and thus it can be seen that their spectral models are also consistent. From the above, from the matching deviation values ​​(c1, c2, ... c N ) are selected as the matching devices with the smallest matching deviation value, and the spectral models based on the reconstructed spectral data established by the spectral model building devices in the two devices are directly matched to the spectral model device to be matched, so as to complete the spectral model matching between the two types of devices.

[0049] The matching spectral model is called to predict and analyze the component content of the sample with unknown component content, and the spectral model matching analysis is completed. For a component of the sample to be tested, after matching the spectral model of the two types of spectral equipment by matching the deviation value, the matching spectral model is called to predict and analyze the reconstructed spectral data of the sample with unknown component content, and the predicted value of the component content is obtained to complete the spectral model matching analysis.

[0050] In this embodiment, after the spectral model to be matched device collects spectral data of a single sample with unknown component content and reconstructs it, the moisture spectral matching model of the reconstructed spectral data is called to perform predictive analysis on the spectral data of the unknown component content to obtain a moisture prediction analysis value, which is the moisture content prediction value of the sample with unknown component content, thereby completing the spectral matching analysis.

Claims

1. A spectral model matching method based on component absorption peaks, characterized in that: The following steps are involved: Step 1: Use multiple spliced ​​multi-sensor near-infrared spectrometers of the same model to collect spectral data of the sample; Step 2: Reconstruct the sample spectrum data according to the component absorption peak range and number, and establish a reconstructed spectrum model library; According to the spectral absorption characteristics of the sample components, X absorption peaks are selected, and the number of wavelength points contained in each single sensor spectrum data is calculated according to the wavelength range and resolution of the single sensor near-infrared device. The number of wavelength points and the spectral wavelength range contained in each absorption peak are calculated respectively in combination with the number of selected absorption peaks. The same number of wavelength points and wavelength range are selected from the multi-sensor spectrum data to reconstruct the single sensor spectrum data; the reconstructed spectrum data is matched one by one with the sample component content of the known component calibration value, and the partial least squares method is used to model the single component spectrum model. Each device in the N devices for building spectral models collects M spectral data, reconstructs the M data and corresponds them one by one with the component contents of M samples with known component calibration values, adopts partial least squares method to build a model, obtains a single component spectral model, and N components are built by the N devices in total; Step 3: Calculate the characterization coefficients of the absorption peaks of different components of the reconstructed spectral data; obtain the absorbance of the wavelength points included in each absorption peak, add the absorbance of the wavelength points included in each absorption peak, and make a quotient with the absorbance of the wavelength points within the spectral wavelength range included in all absorption peaks to obtain the characterization coefficients corresponding to the X absorption peaks of the reconstructed spectral data. The characterization coefficients are recorded as α1, α2, ..., α X ; Step 4: Calculate the matching deviation value according to the characterization coefficient, and perform accurate matching of the spectral model according to the matching deviation value; the specific method of step 4 is: first calculate the mean of the reconstructed spectral data of M samples collected by a single device in the spectral model library, and the absorbance of the M samples in a certain wavelength range included in the absorption peak is (k1, k2, ...k M ), perform averaging on it, and obtain the mean value of M absorbances at this wavelength: Use the same method to calculate the mean absorbance of all wavelength points included in each absorption peak, and record them as L represents the number of wavelength points contained in each absorption peak; Calculate the matching deviation value of the reconstructed spectrum data of the sample with unknown component content compared with the reconstructed spectrum data of the labeled sample, where the calculation method of the matching deviation value c1 of the reconstructed spectrum data of the sample with unknown component content acquired by the spectrum model device to be matched compared with the reconstructed spectrum data of the labeled sample acquired by the first spectrum database establishment device is: Among them, K1 i , K2 i ,……,KX i They are the absorbance of the sample with unknown component content and the labeled sample at the corresponding wavelength points in each absorption peak; Similarly, the matching deviations of the reconstructed spectral data of the sample with unknown component content acquired by the spectral model equipment to be matched and the reconstructed spectral data of the labeled sample acquired by the equipment to establish the N spectral databases are calculated respectively (c1, c2, ... c N ), from the matching deviation value (c1, c2, ... c N ) select multiple devices with the smallest matching deviation value as matching devices, directly match the spectral model based on the reconstructed spectral data in the selected devices to the spectral model device to be matched, and complete the spectral model matching between the two types of devices; Step 5: Call the matching spectral model to perform prediction analysis on the component content of the sample with unknown component content, and complete the spectral model matching analysis.

2. The spectral model matching method based on component absorption peak according to claim 1, characterized in that: The near-infrared spectroscopy equipment described in step 1 is randomly divided into two parts, one part is used to establish a spectral model library, and the other part is used to match the spectral model.

3. The spectral model matching method based on component absorption peak according to claim 2, characterized in that: There are N devices for establishing a spectral model library and 1 device for matching the spectral model. The N spectral devices for establishing the spectral model library are used to collect spectra of M samples with known component contents respectively, and the N devices collect a total of (M×N) spectral data; the 1 device for matching the spectral model is used to collect a single spectrum of a sample with unknown component content.

Images