Focused linear model correction and linear model correction for multivariate calibration model maintenance

Abstract

The invention relates to focused linear model correction and linear model correction for multivariate calibration model maintenance. The device may obtain a master beta coefficient of a master calibration model associated with the master instrument. The main beta coefficient may be at a grid of the target instrument. The device may perform constraint optimization of the objective function according to a set of constraints in order to determine a pair of pass beta coefficients. The constrained optimization may be performed based on a pair of initial transfer [beta] coefficients, a main [beta] coefficient, and a spectrum associated with a scout set. The device may determine a transfer [beta] coefficient based on the pair of transfer [beta] coefficients. The device may determine a final transfer [beta] coefficient based on a set of transfer [beta] coefficients including the transfer [beta] coefficient. The final transfer beta coefficient may be associated with generating a transfer calibration model corresponding to the primary calibration model for use by the target instrument.

Description

[0001] Division Application Description

[0002] This application is a divisional application of an application with an application date of July 10, 2019, an application number of 202210541483.0, and an invention title of "Focused Linear Model Calibration and Linear Model Calibration for Multivariate Calibration Model Maintenance", wherein the application with an application number of 202210541483.0 is a divisional application of an application with an application date of July 10, 2019, an application number of 201910621447.3, and an invention title of "Focused Linear Model Calibration and Linear Model Calibration for Multivariate Calibration Model Maintenance".

[0003] Background

[0004] Spectral instruments can be configured with calibration models for calibrating spectral measurements performed by the spectral instruments. The calibration models are typically generated based on reference values corresponding to known samples and spectra corresponding to the known samples measured by the spectral instruments.

[0005] Overview

[0006] According to some possible embodiments, a method may include: obtaining, by a device, main β (beta) coefficients of a main calibration model associated with a main instrument, wherein the main β coefficients are at a grid of a target instrument; performing, by the device, constrained optimization of an objective function according to a set of constraints to determine a pair of transfer β coefficients, wherein the constrained optimization is performed based on a pair of initial transfer β coefficients, the main β coefficients, and spectra associated with a scouting set; determining, by the device and based on the pair of transfer β coefficients, transfer β coefficients; and determining, by the device, final transfer β coefficients based on a set of transfer β coefficients including the transfer β coefficients, wherein the final transfer β coefficients are associated with generating a transfer calibration model corresponding to the main calibration model for use by the target instrument.

[0007] According to some possible embodiments, a method may include: determining, by a device, that a grid of a main instrument associated with a main calibration model does not match a grid of a target instrument for which a transfer calibration model corresponding to the main calibration model is to be generated; interpolating, by the device and based on determining that the grid of the main instrument does not match the grid of the target instrument, β coefficients of the main calibration model to the grid of the target instrument; and determining, by the device, main β coefficients associated with generating the transfer calibration model based on a result of interpolating the β coefficients of the main calibration model to the grid of the target instrument.

[0008] According to some possible embodiments, a method may include: obtaining, by a device, a scout set associated with updating a calibration model, where the scout set includes spectra associated with a set of samples and the calibration model will be updated based on the set of samples; determining, by the device, a β coefficient associated with the calibration model; determining, by the device and based on the β coefficient and using a linear model correction (LMC) technique, an updated β coefficient associated with the updated calibration model; and updating, by the device, the calibration model based on the updated β coefficient.

[0009] Aspects of the present disclosure may be implemented in one or more of the following embodiments:

[0010] 1) A method, including:

[0011] obtaining, by a device, a main β coefficient of a main calibration model associated with a main instrument, where the main β coefficient is located at a grid of a target instrument;

[0012] performing, by the device, constrained optimization of an objective function according to a set of constraints to determine a pair of transfer β coefficients, where the constrained optimization is performed based on a pair of initial transfer β coefficients, the main β coefficient, and spectra associated with a scout set;

[0013] determining, by the device and based on the pair of transfer β coefficients, a transfer β coefficient; and

[0014] determining, by the device, a final transfer β coefficient based on a set of transfer β coefficients including the transfer β coefficient, where the final transfer β coefficient is associated with generating a transfer calibration model corresponding to the main calibration model for use by the target instrument.

[0015] 2) The method according to 1), wherein the transfer calibration model is generated based on the final transfer β coefficient.

[0016] 3) The method according to 1), wherein the set of transfer β coefficients includes at least one other transfer β coefficient, and each other transfer β coefficient is determined based on a corresponding execution of the constrained optimization of the objective function according to a corresponding pair of initial transfer β coefficients.

[0017] 4) The method according to 1), wherein obtaining the main β coefficient includes:

[0018] determining that the grid of the main instrument matches the grid of the target instrument; and

[0019] identifying the β coefficient of the main calibration model as the main β coefficient.

[0020] 5) The method according to 1), wherein obtaining the main β coefficient includes:

[0021] Determine that the grid of the master instrument does not match the grid of the target instrument;

[0022] Based on determining that the grid of the master instrument does not match the grid of the target instrument, interpolate the master calibration set to the grid of the target instrument to create interpolated calibration data;

[0023] Generate a regression model based on the interpolated calibration data; and

[0024] Determine the master β coefficient as the β coefficient of the regression model.

[0025] 6) The method according to 1), wherein obtaining the master β coefficient includes:

[0026] Determine that the grid of the master instrument does not match the grid of the target instrument;

[0027] Based on determining that the grid of the master instrument does not match the grid of the target instrument, interpolate the β coefficient of the master calibration model to the grid of the target instrument; and

[0028] Determine the master β coefficient based on the result of interpolating the β coefficient of the master calibration model to the grid of the target instrument.

[0029] 7) The method according to 6), wherein based on determining that the master calibration set associated with the master calibration model is not available, interpolate the β coefficient of the master calibration model to the grid of the target instrument.

[0030] 8) The method according to 1), wherein the set of constraints includes correlation constraints associated with each of the master β coefficient and each of the pair of transfer β coefficients.

[0031] 9) The method according to 1), wherein the set of constraints includes slope constraints associated with each of the master β coefficient and each of the pair of transfer β coefficients.

[0032] 10) The method according to 1), wherein the set of constraints includes calibration range constraints of predicted values associated with the reconnaissance set.

[0033] 11) The method according to 1), further comprising generating the pair of initial transfer β coefficients based on at least one of the following:

[0034] Random generation of the pair of initial transfer β coefficients,

[0035] Applying a linear function associated with a random value to the master β coefficient, or

[0036] Adding a random value to the master β coefficient.

[0037] 12) A method, comprising:

[0038] Determining, by a device, that a grid of a master instrument associated with a master calibration model does not match a grid of a target instrument, and generating a transfer calibration model corresponding to the master calibration model for the target instrument;

[0039] Interpolating, by the device and based on determining that the grid of the master instrument does not match the grid of the target instrument, β coefficients of the master calibration model to the grid of the target instrument; and

[0040] Determining, by the device, a master β coefficient associated with generating the transfer calibration model based on a result of interpolating the β coefficients of the master calibration model to the grid of the target instrument.

[0041] 13) The method according to 12), wherein, based on determining that a master calibration set associated with the master calibration model is unavailable, the β coefficients of the master calibration model are interpolated to the grid of the target instrument.

[0042] 14) The method according to 12), further comprising:

[0043] Performing constrained optimization of an objective function according to a set of constraints to determine a pair of transfer β coefficients, wherein the constrained optimization is performed based on a pair of initial transfer β coefficients, the master β coefficient, and a spectrum associated with a reconnaissance set;

[0044] Determining a transfer β coefficient based on the pair of transfer β coefficients;

[0045] Determining a final transfer β coefficient based on a set of transfer β coefficients including the transfer β coefficient; and

[0046] Generating the transfer calibration model based on the final transfer β coefficient.

[0047] 15) The method according to 14), wherein the set of constraints includes:

[0048] A correlation constraint associated with each of the master β coefficient and each transfer β coefficient of the pair of transfer β coefficients,

[0049] A slope constraint associated with each of the master β coefficient and each transfer β coefficient of the pair of transfer β coefficients, and

[0050] A calibration range constraint of predicted values associated with the reconnaissance set.

[0051] 16) The method according to 14), further comprising generating the pair of initial transfer β coefficients based on at least one of the following:

[0052] Random generation of the pair of initial transfer β coefficients,

[0053] Apply a linear function associated with a random value to the primary β coefficient, or

[0054] Add a random value to the primary β coefficient.

[0055] 17) The method according to 12), further comprising:

[0056] Determine a transfer β coefficient associated with generating the transfer calibration model based on the primary β coefficient and using a linear model correction (LMC) technique; and

[0057] Generate the transfer calibration model based on the transfer β coefficient.

[0058] 18) The method according to 17), wherein a reference value regarding a reconnaissance set associated with using the LMC technique is predicted based on the primary calibration model and the primary transfer set.

[0059] 19) A method, comprising:

[0060] Obtain, by a device, a reconnaissance set associated with updating a calibration model, wherein the reconnaissance set includes spectra associated with a set of samples, and the calibration model will be updated based on the set of samples;

[0061] Determine, by the device, a β coefficient associated with the calibration model;

[0062] Determine, by the device and based on the β coefficient and using a linear model correction (LMC) technique, an updated β coefficient associated with updating the calibration model; and

[0063] Update, by the device, the calibration model based on the updated β coefficient.

[0064] 20) The method according to 19), wherein the update of the calibration model is performed during operation of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figures 1A - 1C Is an overview diagram of an example embodiment described herein.

[0066] Figure 2 Is a diagram of an example environment in which the systems and / or methods described herein may be implemented.

[0067] Figure 3 Is Figure 2 A diagram of example components of one or more devices of.

[0068] Figure 4is a flowchart of an example process of a focused linear model calibration technique as described herein, which technique is associated with determining transfer β coefficients for generating a transfer calibration model.

[0069] Figures 5A - 5C is associated with Figure 4 an example illustration of a focused linear model calibration technique.

[0070] Figure 6 is a flowchart of an example process as described herein for interpolating β coefficients of a primary calibration model to a grid of a target instrument to determine primary β coefficients for use in a focused linear model calibration technique or a linear model calibration technique.

[0071] Figures 7A - 7C and Figure 8A and Figure 8B are illustrations associated with interpolating β coefficients of a primary calibration model to a grid of a target instrument and using the LMC technique and the fLMC technique respectively associated with performing calibration model transfer.

[0072] Figures 9A - 9D , Figure 10A , Figure 10B , Figure 11A and Figure 11B are illustrations showing example results associated with achieving standardization of a calibration model among multiple instruments.

[0073] Figure 12 is a flowchart of an example process as described herein for performing model update using a linear model calibration technique.

[0074] Figure 13A and Figure 13B are illustrations showing example results of performing calibration model update using a linear model calibration technique.

[0075] Figure 14A and Figure 14B are illustrations showing example results associated with predicting reference values using a primary calibration model and a primary transfer set.

[0076] DETAILED DESCRIPTION

[0077] The following detailed description of example embodiments refers to the accompanying drawings. Like reference numerals in different figures may identify the same or similar elements.

[0078] Calibration model transfer and calibration model update are two important areas of multivariate calibration model maintenance for spectral applications such as applications in the near-infrared (NIR) region.

[0079] In some cases, when using a multivariate calibration model developed on a first spectroscopic instrument (or under one environmental condition) to predict calibration attributes of samples measured on a second spectroscopic instrument (or by the first spectroscopic instrument under different environmental conditions), the results are unacceptable. Additionally, even for the same spectroscopic instrument, the signal may drift over time, meaning that existing calibration models will need to be updated. When updating calibration models, calibration transfer techniques can be implemented to avoid the cumbersome and costly task of re-collecting data and re-calibrating existing calibration models, so as to transfer the calibration model from one condition to another, regardless of the source of the drift.

[0080] A requirement of general calibration model transfer techniques is to obtain a transfer dataset that includes spectra of the same set of samples collected by a first instrument (e.g., the master instrument from which the calibration model will be transferred, or a given instrument under the original condition) and a second instrument (e.g., the target instrument to which the calibration model will be transferred, or a given instrument under the target condition). In some cases, obtaining the transfer dataset is difficult or impossible. For example, when the calibration model of a perishable material needs to be transferred from a master instrument in one country to a target instrument in another country, it may be impossible to obtain the transfer dataset.

[0081] The Linear Model Correction (LMC) technique can solve this problem by only requiring a small number of spectra collected solely by the target instrument. The set of spectra used by the LMC technique is called the scout set. However, for the LMC technique to work, the reference values of the scout set (e.g., the actual values measured in a chemical laboratory) are needed. In some cases, obtaining these reference values can be very time-consuming and / or expensive.

[0082] Some embodiments described herein provide a Focused LMC (fLMC) technique that can be used in association with performing calibration model transfer. Similar to the LMC technique, the fLMC technique only requires a scout set collected by the target instrument. However, different from the LMC technique, the fLMC technique does not require the reference values of the scout set. Therefore, using the fLMC technique in association with calibration model transfer reduces the cost, difficulty, and / or complexity of calibration model transfer (e.g., compared to the LMC technique and the above general calibration model transfer techniques).

[0083] In addition, calibration model transfer from a primary instrument having a relatively high spectral resolution and / or a relatively wide wavelength range to a target instrument having a relatively low spectral resolution and / or a relatively narrow wavelength range is often encountered (e.g., when transferring a calibration model from a bench-top instrument to a portable instrument). For calibration model transfer in such cases, general calibration transfer techniques require a complete primary calibration set (e.g., a collection of spectra associated with a set of samples measured by the primary instrument) in order to initiate the calibration model transfer process. Here, the spectra of the primary calibration set are interpolated into the grid of the target instrument, and then an intermediate model is developed for transfer to the target instrument.

[0084] However, it is not always possible to access the primary calibration set. Even when the primary calibration set is accessible, in some cases, the primary database may be large and / or may have a long maintenance history. Thus, it may be difficult and / or time-consuming to obtain a clean primary calibration set from the database.

[0085] Some embodiments described herein provide a process in which the fLMC technique or the LMC technique is used in association with performing calibration model transfer using the β coefficients of the primary calibration model without requiring a primary calibration set. The use of the β coefficients (instead of the primary calibration set) reduces the cost, difficulty, and / or complexity of calibration model transfer.

[0086] In addition, when a calibration model developed on a primary instrument is to be deployed on multiple other instruments (e.g., multiple different target instruments) that may have inter-instrument variations, it may be difficult to perform calibration model transfer using conventional calibration model transfer techniques (e.g., when the target instruments are located at a great distance from the primary instrument). In some embodiments, to address this issue, the LMC technique or the fLMC technique can be configured on multiple target instruments. When the primary calibration model is transferred to a target instrument, the user only needs to collect a scout set (e.g., spectra from several samples associated with a given application). The calibration model can use these spectra for automatic correction in combination with the LMC technique (e.g., when reference values are available) or in combination with the fLMC technique (e.g., regardless of whether reference values are available).

[0087] In addition, as described above, after a calibration model is deployed on a given instrument, calibration model update may be required (e.g., due to changes in samples, measurement environment, etc.). A general technique for performing calibration model update is to add new samples to the existing calibration set and then reconstruct the calibration model. However, this technique may require a large number of samples to fit the calibration model to the new samples or new conditions. In addition, this technique requires all calibration data to be available. Further, when the calibration database is large, especially when the spectral range is wide and the spectral resolution is high, reconstructing the calibration model may consume a large amount of time and / or resources (e.g., processor resources, battery power, etc.). Thus, it may be impossible to update the calibration model during the online operation of the instrument.

[0088] Some embodiments described herein provide techniques for performing calibration model updates using LMC techniques. In essence, LMC techniques require a relatively small number of samples to perform calibration model updates. In some embodiments, the update set (i.e., the scout set associated with performing the calibration model update) can include samples representing different conditions of future samples in order to make future predictions more accurate. In addition, calibration model updates using LMC techniques are relatively faster than the general update techniques described above. For example, calibration model updates using the LMC procedure can be performed within a few seconds, making it possible to update the calibration model during online operation.

[0089] In addition, in some cases, transfer sets from both the master instrument and the target instrument may be available, while the reference values of the transfer sets may not be available. In such cases, as described herein, the transfer set from the target instrument can be used as the scout set and the reference value predicted by the master calibration model of the transfer set from the master instrument can be used as the reference value to perform the LMC technique. Thus, when spectral data is available, the LMC technique can be performed, otherwise the spectral data can be used to perform other conventional calibration transfer techniques.

[0090] Figures 1A - 1C is an illustration of an example embodiment described herein. Figure 1A and Figure 1B is an illustration of an example embodiment 100 associated with using a focused linear model correction (fLMC) correction technique, which is associated with generating a transfer calibration model corresponding to a master calibration model associated with a master instrument for configuration on a target instrument.

[0091] For Figure 1A and Figure 1B the purposes of the example embodiment 100, the master calibration model configured on the master instrument will be transferred to the target instrument. In other words, a transfer calibration model corresponding to the master calibration model configured on the master instrument will be generated for use by the target instrument. The example embodiment 100 describes the use of the fLMC technique associated with generating the transfer calibration model.

[0092] As shown by reference numeral 105 in Figure 1A a modeling device (e.g., a device associated with generating the transfer calibration model) can obtain the master β coefficients of the master calibration model at the grid of the target instrument.

[0093] The master β coefficients can include a set of coefficients associated with the master calibration model. For example, the master β coefficients can include a vector of regression coefficients associated with a partial least squares (PLS) regression calibration model configured on the master instrument.

[0094] As described above, the main beta coefficient lies on the grid of the target instrument. The grid of the target instrument is a parameter of the target instrument defined by the spectral resolution and wavelength range of the target instrument. Similarly, the grid of the main instrument is a parameter of the main instrument defined by the spectral resolution and wavelength range of the main instrument. In some embodiments, the grid of the main instrument may be different from the grid of the target instrument (e.g., when the main instrument has a relatively higher spectral resolution and / or a wider wavelength range compared to the target instrument). Alternatively, the grid of the main instrument may match the grid of the target instrument (e.g., when the spectral resolution and wavelength range of the main instrument match the spectral resolution and wavelength range of the target instrument within a threshold amount).

[0095] In some embodiments, the manner in which the modeling device obtains the main beta coefficient may be based on whether the grid of the main instrument matches the grid of the target instrument.

[0096] For example, the modeling device may determine (e.g., based on information provided by the main instrument and / or the target instrument, based on information stored or accessible by the modeling device) whether the grid of the main instrument matches the grid of the target instrument. In some embodiments, if the modeling device determines that the grid of the main instrument matches the grid of the target instrument, the modeling device may identify the beta coefficient of the main calibration model as the main beta coefficient. In other words, when the grid of the main instrument matches the grid of the target instrument, the modeling device may directly use the beta coefficient of the main instrument as the main beta coefficient (e.g., because the beta coefficient of the main calibration model is already on the grid of the target instrument). In this case, the beta coefficient of the main calibration model may be used as the main beta coefficient regardless of whether the main calibration set associated with the main calibration model is available.

[0097] In some embodiments, if the modeling device determines that the grid of the main instrument does not match the grid of the target instrument, the main instrument may obtain the main beta coefficient based on the main calibration set associated with the main calibration model. For example, if the grid of the main instrument does not match the grid of the target instrument, the modeling device may interpolate the main calibration set onto the grid of the target instrument in order to create interpolated calibration data (i.e., the spectra of the main calibration set interpolated onto the grid of the target instrument). Here, the modeling device may generate a regression model (e.g., a PLS model, a principal component regression (PCR) model, etc.) based on the interpolated calibration data, and may determine the main beta coefficient as the beta coefficient of the regression model. In some embodiments, when the main calibration set is available, the modeling device may obtain the main beta coefficient in this manner. For example, the modeling device may determine that the main calibration set is available (e.g., accessible, not exceeding a threshold size or complexity level), and may proceed as described above.

[0098] In some embodiments, if the modeling device determines that the grid of the main instrument does not match the grid of the target instrument, the main instrument may obtain the main beta coefficient based on the beta coefficient of the main calibration model, an example of which is in Figure 1Cshown in

[0099] Figure 1C is an illustration of an example implementation 150 associated with interpolating the β - coefficients of a master calibration model onto the grid of a target instrument to obtain master β - coefficients. As shown by reference numeral 155, the modeling device can determine that the grid of the master instrument does not match the grid of the target instrument. As shown by reference numeral 160, based on this determination, the modeling device can interpolate the β - coefficients of the master calibration model onto the grid of the target instrument. As shown by reference numeral 165, the result of interpolating the β - coefficients of the master calibration model onto the grid of the target instrument can be used as the master β - coefficients. In some implementations, when the master calibration set is not available, the modeling device can obtain the master β - coefficients in this way. For example, the modeling device can determine that the master calibration set is not available (e.g., inaccessible, exceeds a threshold size or complexity level), and can proceed as described above.

[0100] In some implementations, the modeling device can use the fLMC technique (as described in connection with example implementation 100) passed for the calibration model in association with interpolating the β - coefficients of the master calibration model onto the grid of the target instrument to determine the master β - coefficients. Additionally or alternatively, the modeling device can use the LMC technique in association with interpolating the β - coefficients of the master calibration model onto the grid of the target instrument to determine the master β - coefficients. In other words, the interpolation of the β - coefficients of the master calibration model onto the grid of the target instrument can be used in association with performing the fLMC technique or the LMC technique passed for the calibration model.

[0101] Returning to the fLMC technique associated with example implementation 100, in some implementations, the modeling device can determine the final transferred β - coefficients based on a set of transferred β - coefficients. The final transferred β - coefficients are the β - coefficients used to generate the transferred calibration model. In some implementations, as described below, the modeling device can determine each of the set of transferred β - coefficients based on the corresponding iteration of the constrained optimization of an objective function.

[0102] In some implementations, for each iteration, the modeling device can perform the constrained optimization of the following objective function:

[0103]

[0104] with the following constraints:

[0105] corr(b transA ,b 主 )≥r (1)

[0106] corr(b transB ,b 主 )≥r (2)

[0107] slope(b transA,b 主 )≥r (3)

[0108] slope(b transB ,b 主 )≥r (4)

[0109] minY cal <<X 侦察 b transA <<maxY cal (5)

[0110] minY cal <<X 侦察 b transB <<maxY cal (6)

[0111] where X 侦察 is a reconnaissance set (e.g., the spectrum of the reconnaissance set measured by a target instrument), b transA and b transB are a pair of transfer β coefficients associated with a given iteration, b 主 is the main β coefficient, r is a constraint threshold, and minY cal and maxY cal define a calibration range associated with the target instrument.

[0112] In some embodiments, the constraint threshold r (e.g., the relevant constraints and / or slope constraints as described in the above equations) can be optimized using a validation set. In this case, a set of constraint thresholds r can be used iteratively, and the best r (e.g., determined based on the root mean square error of prediction (RMSEP) of the validation set) can be used in association with determining the resulting transfer β coefficients. In some embodiments, this constraint threshold optimization can be used in association with the fLMC technique or the LMC technique.

[0113] To establish an objective function, the concept of reproducibility is introduced. Assuming that each of a pair of transfer β coefficients b transA and b transB is capable of fitting the reconnaissance set, the difference in the predicted values of the reconnaissance set using b transA and b transB should be small. Thus, the objective function is used to minimize the squared difference in the predicted values of the reconnaissance set using b transA and b transB By using this concept of reproducibility, the need for a reference value of the reconnaissance set is eliminated. In other words, due to this concept of reproducibility, the fLMC technique does not require a reference value of the reconnaissance set (unlike the LMC technique).

[0114] To obtain meaningful results, the minimization of the objective function needs to be carried out under a set of constraints. For example, the set of constraints can include correlation constraints associated with the main β coefficient (b 主 ) and each of a pair of transfer β coefficients (b transA and b transB ), where the pair of transfer β coefficients is associated with a given iteration of the constrained optimization of the objective function. According to this correlation constraint, the correlation between b transA and b 主 , and the correlation between b transB and b 主 should satisfy a threshold (e.g., indicated by equations (1) and (2) respectively, where, for example, the value r can be greater than or equal to 0.95).

[0115] As another example, the set of constraints can include slope constraints associated with the main β coefficient and each of a pair of transfer β coefficients, where the pair of transfer β coefficients is associated with a given iteration of the constrained optimization of the objective function. According to this slope constraint, the slope between b transA and b 主 , and the slope between b transB and b 主 should satisfy a threshold (e.g., indicated by equations (3) and (4) respectively, where, for example, the value r can be greater than or equal to 0.95).

[0116] As another example, the set of constraints can include calibration range constraints for the predicted values associated with the reconnaissance set. According to this calibration constraint, the values of the reconnaissance set predicted using b transA (i.e., X 侦察 b transA ) and the values of the reconnaissance set predicted by b transB (i.e., X 侦察 b transB ) should be within the calibration range (e.g., indicated by equations (5) and (6) respectively), or within a range close to the reference value of the reconnaissance set.

[0117] To initiate a given iteration of the above-mentioned constrained optimization process, initial values of b transA and b transB are required (i.e., b transA0 and b transB0 respectively). In some embodiments, the modeling device can generate a pair of initial transfer β coefficients based on the random generation of a pair of initial transfer β coefficients. Additionally or alternatively, the modeling device can apply a linear function associated with a random value to the main β coefficient (e.g., b transA0 , b transB0 = m × b 主+n, where m and n are random numbers) to generate an initial pair of transfer β coefficients. Additionally or alternatively, the modeling device may generate an initial pair of transfer β coefficients based on adding a random value to the main β coefficient (e.g., b transA0 , b transB0 = b 主 +n, where n is a random number).

[0118] For a given iteration of constraint optimization, the modeling device may generate an initial pair of transfer β coefficients (e.g., b transAi0 and b transBi0 for iteration i, and b transAk0 and b transBk0 for iteration k), and may perform constraint optimization of the objective function to determine a pair of transfer β coefficients (e.g., b transAi and b transBi for iteration i, and b transAk and b transBk for iteration k). Then, the modeling device may determine the transfer β coefficients based on this pair of transfer β coefficients (e.g., b transi for iteration i and b transk for iteration k). For example, as shown by reference numeral 110 with respect to iteration i, the modeling device may generate b transAi0 and b transBi0 , perform constraint optimization of the objective function to determine b transAi and b transBi , and determine the transfer β coefficient associated with iteration i (b transAi and b transBi ) based on this pair of transfer β coefficients (e.g., based on the average of b transi ). As another example, as shown by reference numeral 115 with respect to iteration k, the modeling device may generate b transAk0 and b transBk0 , perform constraint optimization of the objective function to determine b transAk and b transBk , and determine the transfer β coefficient associated with iteration k (b transAk and b transBk ) based on this pair of transfer β coefficients (e.g., based on the average of b transk ). Here, b transi and b transk are included in a set of transfer β coefficients, and the modeling device may determine the final transfer β coefficient (b trans ) based on this set of transfer β coefficients.

[0119] In some embodiments, the modeling device may be configured to perform multiple (e.g., 5 times, 20 times, 100 times, etc.) iterations of constrained optimization of the objective function (e.g., to avoid biased results based on the randomized nature of an initial pair of transfer β coefficients).

[0120] As Figure 1B shown by reference numeral 120, the modeling device may determine a final transfer β coefficient (b trans ) based on the set of transfer β coefficients. For example, the modeling device may determine the final transfer β coefficient to be equal to the average, median, mode, etc. of the set of transfer β coefficients (e.g., b transi to b transk ).

[0121] As shown by reference numeral 125, the modeling device may generate a transfer calibration model based on the final transfer β coefficient. For example, the modeling device may generate a regression model (e.g., a PLS model, a PCR model, etc.) based on the final transfer β coefficient. As shown by reference numeral 130, the modeling device may provide the transfer calibration model to the target instrument (e.g., such that the target instrument can use the transfer calibration model). In this way, the modeling device may be configured to use the fLMC technique, which allows the modeling device to generate a transfer calibration model using spectra associated with a scout set without the need for reference values of the scout set.

[0122] As described above, Figures 1A - 1C is provided only as an example. Other examples are possible and may be different from the examples described with respect to Figures 1A - 1C this description.

[0123] Figure 2 is a diagram of an example environment 200 in which the systems and / or methods described herein may be implemented. As Figure 2 shown, the environment 200 may include a primary instrument 205, a target instrument 210, a modeling device 215, and a network 220. The devices of the environment 200 may be interconnected via a wired connection, a wireless connection, or a combination of wired and wireless connections.

[0124] The master instrument 205 includes a device configured with a master calibration model that can perform spectral measurements on a sample. For example, the master instrument 205 can include a benchtop (i.e., non-portable) spectrometer device that performs spectroscopy (e.g., vibrational spectroscopy such as near-infrared (NIR) spectroscopy, mid-infrared spectroscopy (mid-IR), Raman spectroscopy, etc.). In some embodiments, the master instrument 205 can be capable of obtaining spectral measurement results at a higher resolution than the spectral measurement results obtained by the target instrument 210 (i.e., the master instrument 205 can be a high-resolution device while the target instrument 210 can be a low-resolution device). For example, the master instrument 205 can be capable of obtaining spectral measurement results on 400 channels, while the target instrument 210 may be capable of obtaining spectral measurement results on 125 channels. In some embodiments, the master instrument 205 can be configured with a master calibration model for calibrating the spectral measurement results obtained by the master instrument 205. In some embodiments, the master instrument 205 can receive information from and / or transmit information to another device (such as the modeling device 215) in the environment 200.

[0125] The target instrument 210 includes a device capable of performing spectral measurements on a sample based on a target calibration model, where the target calibration model can be generated based on information associated with the master calibration model associated with the master instrument 205 as described herein. For example, the target instrument 210 can include a portable spectrometer device or a handheld spectrometer device that performs spectroscopy. In some embodiments, the target instrument 210 can be capable of obtaining spectral measurement results at a lower resolution than the spectral measurement results obtained by the master instrument 205. In some embodiments, the target instrument 210 can receive information from and / or transmit information to another device (such as the modeling device 215) in the environment 200.

[0126] The modeling device 215 includes a device capable of performing operations associated with the following actions: transferring the master calibration model from the master instrument 205 to the target instrument 210 (i.e., generating a transfer calibration model corresponding to the master calibration model) and / or updating the calibration model configured on a given instrument (such as the master instrument 205 or the target instrument 210) as described herein. For example, the modeling device 215 can include a server, a group of servers, a computer, a cloud computing device, etc. In some embodiments, the modeling device 215 can receive information from and / or transmit information to another device (such as the master instrument 205 and / or the target instrument 210) in the environment 200. In some embodiments, the modeling device 215 and the master instrument 205 can be implemented within a single device. Alternatively, in some embodiments, the modeling device 215 and the target instrument 210 can be implemented within a single device.

[0127] Network 220 includes one or more wired and / or wireless networks. For example, network 220 may include a cellular network (e.g., a New Radio (NR / 5G) network, a Long-Term Evolution (LTE) network, a 3G network, a Code Division Multiple Access (CDMA) network, etc.), a Public Land Mobile Network (PLMN), a Local Area Network (LAN), a Wide Area Network (WAN), a Metropolitan Area Network (MAN), a telephone network (e.g., a Public Switched Telephone Network (PSTN)), a private network, an ad hoc network, an intranet, the Internet, a fiber-optic based network, a cloud computing network, etc. and / or a combination of these or other types of networks.

[0128] Figure 2 The number and arrangement of the devices and networks shown are provided as an example. In fact, compared to Figure 2 the devices and networks shown, there may be additional devices and / or networks, fewer devices and / or networks, different devices and / or networks, or differently arranged devices and / or networks.

[0129] In addition, Figure 2 two or more of the devices shown may be implemented within a single device, or Figure 2 the single device shown may be implemented as multiple distributed devices. For example, although the main instrument 205 and the modeling device 215 are described as two separate devices, the main instrument 205 and the modeling device 215 may be implemented within a single device. As another example, the target instrument 210 and the modeling device 215 may be implemented within a single device. Additionally or alternatively, a set of devices (e.g., one or more devices) of the environment 200 may perform one or more functions described as being performed by another set of devices of the environment 200.

[0130] Figure 3 is an illustration of example components of a device 300. The device 300 may correspond to the main instrument 205, the target instrument 210, and / or the modeling device 215. In some embodiments, the main instrument 205, the target instrument 210, and / or the modeling device 215 may include one or more devices 300 and / or one or more components of the device 300. As Figure 3 shown, the device 300 may include a bus 310, a processor 320, a memory 330, a storage component 340, an input component 350, an output component 360, and a communication interface 370.

[0131] The bus 310 includes components that allow communication among the components of the device 300. The processor 320 is implemented in hardware, firmware, or a combination of hardware and software. The processor 320 takes the form of a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), a microprocessor, a microcontroller, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or another type of processing component. In some embodiments, the processor 320 includes one or more processors that can be programmed to perform functions. The memory 330 includes random access memory (RAM), read only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic memory, and / or optical memory) that stores information and / or instructions for use by the processor 320.

[0132] The storage component 340 stores information and / or software related to the operation and use of the device 300. For example, the storage component 340 may include a hard disk (e.g., a magnetic disk, an optical disk, a magneto-optical disk, and / or a solid state disk), a compact disc (CD), a digital versatile disc (DVD), a floppy disk, a cartridge, a magnetic tape, and / or another type of non-transitory computer-readable medium along with corresponding drives.

[0133] The input component 350 includes components that allow the device 300 to receive information, for example, via user input (e.g., a touch screen display, a keyboard, a keypad, a mouse, a button, a switch, and / or a microphone). Additionally or alternatively, the input component 350 may include sensors for sensing information (e.g., a global positioning system (GPS) component, an accelerometer, a gyroscope, and / or an actuator). The output component 360 includes components that provide output information from the device 300 (e.g., a display, a speaker, and / or one or more light emitting diodes (LEDs)).

[0134] The communication interface 370 includes transceiver-like components (e.g., a transceiver and / or separate receivers and transmitters) that enable the device 300 to communicate with other devices, for example, via a wired connection, a wireless connection, or a combination of wired and wireless connections. The communication interface 370 may allow the device 300 to receive information from another device and / or provide information to another device. For example, the communication interface 370 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, a universal serial bus (USB) interface, a Wi-Fi interface, a cellular network interface, etc.

[0135] Device 300 may perform one or more of the processes described herein. Device 300 may perform these processes based on software instructions stored by a non-transitory computer-readable medium (such as memory 330 and / or storage component 340) and executed by processor 320. A computer-readable medium is defined herein as a non-transitory memory device. A memory device includes storage space within a single physical storage device or storage space spread across multiple physical storage devices.

[0136] The software instructions may be read into memory 330 and / or storage component 340 from another computer-readable medium or from another device via communication interface 370. When executed, the software instructions stored in memory 330 and / or storage component 340 may cause processor 320 to perform one or more of the processes described herein. Additionally or alternatively, hardwired circuitry may be used in place of or in combination with software instructions to perform one or more of the processes described herein. Accordingly, the embodiments described herein are not limited to any specific combination of hardware circuitry and software.

[0137] Figure 3 The number and arrangement of the components shown are provided as an example. In fact, compared to Figure 3 the components shown, device 300 may include additional components, fewer components, different components, or components in a different arrangement. Additionally or alternatively, a set of components (such as one or more components) of device 300 may perform one or more functions described as being performed by another set of components of device 300.

[0138] Figure 4 is a flowchart of an example process 400 of a focused linear model correction (fLMC) technique associated with determining transfer beta coefficients for generating a transfer calibration model as described herein. In some embodiments, Figure 4 one or more of the process blocks of Figure 4 may be performed by modeling device 215. In some embodiments,

[0139] As Figure 4 shown, process 400 may include obtaining the master beta coefficients of a master calibration model associated with a master instrument, where the master beta coefficients are at the grid of a target instrument (block 410). For example, modeling device 215 may obtain the master beta coefficients of a master calibration model associated with master instrument 205, where as described above, the master beta coefficients are at the grid of target instrument 210.

[0140] As Figure 4As further shown in, process 400 may include performing constrained optimization of an objective function according to a set of constraints to determine a pair of transfer β coefficients, where the constrained optimization is performed based on a pair of initial transfer β coefficients, a primary β coefficient, and a spectrum associated with a scout set (block 420). For example, modeling device 215 may perform constrained optimization of an objective function according to a set of constraints to determine a pair of transfer β coefficients, where as described above, the constrained optimization is performed based on a pair of initial transfer β coefficients, a primary β coefficient, and a spectrum associated with a scout set.

[0141] As Figure 4 further shown in, process 400 may include determining transfer β coefficients based on the pair of transfer β coefficients (block 430). For example, as described above, modeling device 215 may determine transfer β coefficients based on the pair of transfer β coefficients.

[0142] As Figure 4 further shown in, process 400 may include determining final transfer β coefficients based on a set of transfer β coefficients including the transfer β coefficients, where the final transfer β coefficients are associated with generating a transfer calibration model corresponding to a primary calibration model for use by a target instrument (block 440). For example, modeling device 215 may determine final transfer β coefficients based on a set of transfer β coefficients including the transfer β coefficients, where the final transfer β coefficients are associated with generating a transfer calibration model corresponding to a primary calibration model for use by target instrument 210.

[0143] Process 400 may include additional embodiments, such as any single embodiment or any combination of embodiments described below and / or in combination with one or more other processes described elsewhere herein.

[0144] In some embodiments, modeling device 215 and / or target instrument 210 may generate a transfer calibration model based on the final transfer β coefficients.

[0145] In some embodiments, a set of transfer β coefficients includes at least one other transfer β coefficient, each other transfer β coefficient being determined based on a corresponding execution of constrained optimization of an objective function according to a corresponding pair of initial transfer β coefficients.

[0146] In some embodiments, when obtaining the primary β coefficient, modeling device 215 may determine that the grid of primary instrument 205 matches the grid of target instrument 210 and identify the β coefficient of the primary calibration model as the primary β coefficient.

[0147] In some embodiments, when obtaining the master β coefficient, the modeling device 215 may determine that the grid of the master instrument 205 does not match the grid of the target instrument 210; based on determining that the grid of the master instrument 205 does not match the grid of the target instrument 210, interpolate the master calibration set to the grid of the target instrument 210 to create interpolated calibration data; generate a regression model based on the interpolated calibration data; and determine the master β coefficient as the β coefficient of the regression model.

[0148] In some embodiments, when obtaining the master β coefficient, the modeling device 215 may determine that the grid of the master instrument 205 does not match the grid of the target instrument 210; based on determining that the grid of the master instrument 205 does not match the grid of the target instrument 210, interpolate the β coefficient of the master calibration model to the grid of the target instrument 210; and determine the master β coefficient based on the result of interpolating the β coefficient of the master calibration model to the grid of the target instrument 210. In some embodiments, based on the determination that the master calibration set associated with the master calibration model is unavailable, the β coefficient of the master calibration model is interpolated to the grid of the target instrument 210.

[0149] In some embodiments, in addition to the calibration range constraints of the predicted values associated with the reconnaissance set, a set of constraints includes correlation constraints associated with each of the master β coefficient and a pair of transfer β coefficients, and / or slope constraints associated with each of the master β coefficient and a pair of transfer β coefficients.

[0150] In some embodiments, the modeling device 215 may generate a pair of initial transfer β coefficients based on the random generation of a pair of initial transfer β coefficients, applying a linear function associated with a random value to the master β coefficient, and / or adding a random value to the master β coefficient.

[0151] Although Figure 4 illustrates example blocks of process 400, in some embodiments, process 400 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks compared to those depicted in Figure 4 In addition or alternatively, two or more blocks of process 400 may be executed in parallel.

[0152] To illustrate the effectiveness of the fLMC technique, the PLS regression model of the Brix of sugarcane was transferred from a bench-top FOSS NIR master instrument to a portable MicroNIR target instrument. Figures 5A - 5C is a diagram associated with the results of the transfer of this example calibration model using the fLMC technique.

[0153] A total of 1712 FOSS spectra were used to build the master calibration model. These spectra were first interpolated to the MicroNIR grid. The interpolated calibration data were used to build an intermediate master calibration model, and the resulting β coefficients were used as b主 . The MicroNIR instrument collected 126 spectra, of which 15 spectra were randomly selected as the scout set for performing fLMC. The remaining 111 spectra were used as an external validation set to validate the transfer calibration model. Using the FOSS validation set from the same 111 samples, the prediction performance of the transfer calibration model was compared with that of the primary calibration model.

[0154] As Figure 5A shown, without performing transfer of the calibration model, when using the intermediate primary calibration model to predict the MicroNIR validation set, the root mean square error of prediction (RMSEP) was high. However, as Figure 5B shown, when using the fLMC technique for transfer of the calibration model, the RMSEP decreased significantly. The RMSEP using the original FOSS model for the FOSS validation set was also calculated and used as a benchmark for evaluating the performance of the transfer calibration model. In Figure 5C , it can be seen that the residuals between the predicted Brix values and the laboratory Brix values for the validation set by the transfer calibration model remained within approximately ±2RMSEP of the original FOSS primary calibration model, indicating that the confidence that the transferred MicroNIR calibration lies within the original bounds of the FOSS calibration is approximately 95%. These results indicate that the performance of the transfer calibration model generated using the fLMC technique is close to that of the original FOSS primary calibration model (e.g., having a relatively wide wavelength range and relatively high spectral resolution).

[0155] In addition, for comparison, the same FOSS primary calibration model was transferred to MicroNIR using the mean difference correction (MDC) technique and the piecewise direct standardization (PDS) technique (which are two typical techniques for transfer of calibration models). To apply these two techniques, a transfer set consisting of 15 spectra from the primary instrument and the target instrument was used. These spectra were from the same samples as those used in the scout set when using the fLMC technique. Using the transfer calibration models with MDC and PDS, the RMSEP for the same validation set was 1.80 and 0.72, respectively. Thus, in this case, the fLMC technique performed better than the MDC technique and worse than the PDS technique. However, different from the PDS technique, the fLMC technique does not require a transfer set on the primary instrument, making the fLMC technique relatively lower in cost and / or complexity while achieving similar performance.

[0156] As described above, Figures 5A - 5C is provided only as an example. Other examples are possible and may be different from the example described with respect to Figures 5A - 5C .

[0157] Figure 6FIG. 600 is a flow chart of an example process for interpolating the β coefficients of a master calibration model to the grid of a target instrument to determine master β coefficients for use with an fLMC technique or an LMC technique. In some embodiments, Figure 6 one or more process blocks of Figure 6 may be performed by the modeling device 215. In some embodiments,

[0158] As Figure 6 shown, process 600 may include determining that the grid of the master instrument associated with the master calibration model does not match the grid of the target instrument, and generating a transfer calibration model corresponding to the master calibration model for the target instrument (block 610). For example, as described above, the modeling device 215 may determine that the grid of the master instrument 205 associated with the master calibration model does not match the grid of the target instrument 210, and generate a transfer calibration model corresponding to the master calibration model for the target instrument 210.

[0159] As Figure 6 further shown in

[0160] As Figure 6 further shown, process 600 may include interpolating the β coefficients of the master calibration model to the grid of the target instrument based on determining that the grid of the master instrument does not match the grid of the target instrument (block 620). For example, as described above, the modeling device 215 may interpolate the β coefficients of the master calibration model to the grid of the target instrument 210 based on determining that the grid of the master instrument 205 does not match the grid of the target instrument 210.

[0161] Process 600 may include additional embodiments, such as any single embodiment or any combination of embodiments of one or more other processes described below and / or combined elsewhere herein.

[0162] In some embodiments, based on a determination that the master calibration set associated with the master calibration model is not available, the β coefficients of the master calibration model are interpolated to the grid of the target instrument 210.

[0163] In some embodiments, the modeling device 215 may perform constrained optimization of an objective function according to a set of constraints to determine a pair of transfer β coefficients, where the constrained optimization is performed based on a pair of initial transfer β coefficients, a primary β coefficient, and a spectrum associated with a scout set. Here, the modeling device may determine the transfer β coefficient based on the pair of transfer β coefficients; the final transfer β coefficient may be determined based on a set of transfer β coefficients including the transfer β coefficient. In other words, in some embodiments, the modeling device 215 may use the fLMC technique to determine the final transfer β coefficient. In some embodiments, the set of constraints includes correlation constraints associated with the primary β coefficient and each of the pair of transfer β coefficients, slope constraints associated with the primary β coefficient and each of the pair of transfer β coefficients, and calibration range constraints of predicted values associated with the scout set. In some embodiments, the modeling device 215 may generate a pair of initial transfer β coefficients based on random generation of a pair of initial transfer β coefficients, applying a linear function associated with a random value to the primary β coefficient, or adding a random value to the primary β coefficient.

[0164] In some embodiments, the modeling device 215 may determine transfer β coefficients associated with generating a transfer calibration model based on the primary β coefficient and using linear model correction (LMC) techniques. In other words, in some embodiments, the modeling device 215 may use the LMC technique to determine the final transfer β coefficient. In some embodiments, reference values of a scout set associated with using the LMC technique are predicted based on a primary calibration model and a primary transfer set.

[0165] Although Figure 6 example blocks of process 600 are shown, in some embodiments, process 600 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks compared to those depicted in Figure 6 In addition or alternatively, two or more blocks of process 600 may be executed in parallel.

[0166] In some embodiments, as described above, the β coefficients of the primary calibration model may be interpolated into the grid of the target instrument 210 and used as the primary β coefficient. For example, in some embodiments, such techniques may be used in combination with the LMC technique or the fLMC technique. Figures 7A - 7C And Figure 8A And Figure 8B are diagrams associated with interpolating the β coefficients of the primary calibration model into the grid of the target instrument and using the LMC technique and the fLMC technique associated with performing transfer of the calibration model, respectively.

[0167] Using the same dataset as described above with respect to Figures 5A - 5C performing the LMC technique using the result of interpolating the β coefficients of the primary calibration model into the grid of the target instrument as the primary β coefficient, the results of which are shown inFigures 7A - 7C Here, since there is no available master calibration set, it is impossible to construct an intermediate master calibration model. As Figure 7A shown, when directly using the interpolated β coefficients to predict the validation set on the target instrument, the resulting RMSEP is high. As Figure 7B shown, when using the interpolated β coefficients as the master β coefficients and performing the LMC technique, the RMSEP is significantly reduced. In addition, as Figure 7C shown, the residuals between the predicted Brix values and the laboratory Brix values for the validation set of the transfer calibration model remain within ±2RMSEP of the original FOSS master calibration model, except for limited exceptions.

[0168] In addition, using the same dataset as described above regarding Figures 5A - 5C and performing the fLMC technique using the result of interpolating the β coefficients of the master calibration model to the grid of the target instrument as the master β coefficients, the results are shown in Figure 8A and Figure 8B Although there is a slight performance degradation compared to using the LMC technique, the RMSEP is significantly reduced compared to the case where no calibration model transfer is performed (as shown in Figure 5A ). As Figure 8A shown, the RMSEP is quite low, where the normalized RMSEP is 7.7% (normalized to the average Brix value of the validation set). In addition, as Figure 8B shown, most of the residuals between the predicted Brix values and the laboratory Brix values for the validation set of the transfer calibration model remain within ±2RMSEP of the original FOSS master calibration model. It should be noted that the fLMC technique is the only technique that can be used in the following situations: the master calibration set is not available, the grids between the master instrument and the target instrument are different, only the scout set collected by the target instrument is used for transfer, and there are no reference values for the scout set. In some embodiments, the performance of the transfer calibration model can be further improved as the calibration model is updated.

[0169] As described above, Figures 7A - 7C and Figure 8A and Figure 8B are provided only as examples. Other examples are possible and can be different from the examples regarding 7A- Figure 7C and Figure 8A and Figure 8B described.

[0170] In some embodiments, the techniques described herein can be used to standardize calibration models among multiple instruments. As described above, instruments or devices of the same type typically encounter inter-instrument differences. Thus, when a calibration model is developed on one instrument but needs to be deployed on multiple (e.g., hundreds, millions, etc.) instruments, the inter-instrument differences can lead to inconsistent performance. For this problem, it may be impractical to use general methods to perform calibration model transfer, especially when the instruments are located in different locations. To solve this problem, the LMC technique and the fLMC technique can be configured on the instruments. Here, when the master calibration model is transferred to the target instrument, only the spectra from a small number of samples need to be collected. The calibration model can be automatically corrected using the LMC technique (e.g., when the reference values of the scout set are available) or using the fLMC technique (e.g., regardless of whether the reference values of the scout set are available).

[0171] Figures 9A - 9D , Figure 10A , Figure 10B , Figure 11A and Figure 11B are diagrams showing example results associated with standardizing calibration models among multiple instruments. In the examples associated with Figures 9A - 9D , Figure 10A , Figure 10B , Figure 11A and Figure 11B , the raw data from the MicroNIR device was calibrated in two different ways (Data A and Data B) to simulate inter-instrument variations. 759 spectra from 38 mixture samples were used to build a calibration model to predict caffeine content. 200 spectra from another 10 mixture samples were used as a validation set. As Figure 9A and Figure 9B show, when using calibration model A to predict validation data A, or when using calibration model B to predict validation data B, the performance is similar. However, as Figure 9C shows, when using calibration model A to predict validation data B, the performance deteriorates. As Figure 9D shows, many of the residuals between the predicted values and the laboratory values for validation set B exceed the ±2RMSEP benchmark for using model A to predict validation A.

[0172] Ten samples with three replicate spectra were randomly selected from calibration set B as a scout set to perform the LMC technique and the fLMC technique. As Figure 10A shows, using the LMC technique, the RMSEP is significantly reduced. As Figure 10B shows, all the prediction residuals are within the ±2RMSEP benchmark for using model A to predict validation A. As Figure 11A shows, using the fLMC technique, the RMSEP is similarly reduced. As Figure 11BAs shown, most of the prediction residuals are within the ±2RMSEP benchmark for predicting Validation A using Model A. Therefore, it is effective to correct the inter-instrument differences in model performance using the LMC technique or the fLMC technique.

[0173] In fact, the LMC technique and the fLMC technique are effective when using as few as 8 samples as the scout set. It is worth noting that Figure 10A 、 Figure 10B 、 Figure 11A and Figure 11B The results in are examples of medium performance. The scout samples are randomly selected to simulate the real test scenarios on the user side. The final performance results are affected by which samples are used as the scout set. Similarly, the fLMC technique performs worse than the LMC technique. However, when there is no reference value available for the scout set, the fLMC technique is the only technique available for calibrating model transfer.

[0174] As described above, Figures 9A - 9D 、 Figure 10A 、 Figure 10B 、 Figure 11A and Figure 11B are provided only as examples. Other examples are possible and may differ from the examples described with respect to Figures 9A - 9D 、 Figure 10A 、 Figure 10B 、 Figure 11A and Figure 11B .

[0175] As described above, in some cases, the LMC technique can be applied to calibrate model updates by using updated samples as the scout set for the LMC technique. It is worth noting that this does not require the use of all calibration data (e.g., as required by the general model update technique of adding updated samples to the calibration set and recalibrating the model), and it takes a relatively small amount of time, such that calibrating model updates can be performed during the online operation of the instrument (e.g., the host instrument 205, the target instrument 210).

[0176] Figure 12 is a flowchart of an example process 1200 for performing a calibration model update using the LMC technique. In some embodiments, Figure 12 One or more process blocks of can be executed by the modeling device 215. In some embodiments, Figure 12 One or more process blocks of can be executed by another device or a group of devices (such as the host instrument 205 and / or the target instrument 210) that is separate from or includes the modeling device 215.

[0177] As Figure 12As shown, process 1200 may include obtaining a scout set associated with updating a calibration model, where the scout set includes spectra associated with a set of samples based on which the calibration model will be updated (block 1210). For example, modeling device 215 may obtain a scout set associated with updating a calibration model, where the scout set includes spectra associated with a set of samples based on which the calibration model will be updated.

[0178] As Figure 12 further shown, process 1200 may determine a β coefficient associated with the calibration model (block 1220). For example, modeling device 215 may determine a β coefficient associated with the calibration model.

[0179] As Figure 12 further shown, process 1200 may include determining an updated β coefficient associated with updating the calibration model based on the β coefficient and using LMC techniques (block 1230). For example, modeling device 215 may determine an updated β coefficient associated with updating the calibration model based on the β coefficient and using LMC techniques.

[0180] As Figure 12 further shown, process 1200 may include updating the calibration model based on the updated β coefficient (block 1240). For example, modeling device 215 may update the calibration model based on the updated β coefficient (e.g., such that the updated calibration model uses the updated β coefficient associated with performing calibration).

[0181] Process 1200 may include additional implementations, such as any single implementation or any combination of implementations described below and / or in combination with one or more other processes described elsewhere herein.

[0182] In some implementations, the update of the calibration model is performed during the operation of an instrument (e.g., the primary instrument 205, the target instrument 210) without taking the device offline.

[0183] While Figure 12 example blocks of process 1200 are shown, in some implementations, process 1200 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks compared to those depicted Figure 12 herein. Additionally or alternatively, two or more blocks of process 1200 may be executed in parallel.

[0184] Figure 13A And Figure 13B are diagrams showing example results of performing calibration model updates using linear model correction techniques.

[0185] For updating (as described above in connection with Figure 7BThe Brix model of sugarcane described, using an additional 30 MicroNIR spectra as the update set. Here, the LMC technique is applied to update the calibration model. Figure 13A is for the same validation set used in Figure 7B and shows a graphical illustration of the prediction results associated with this update. As shown, the model performance is improved, where as Figure 13B shown, the RMSEP decreases and the prediction residuals decrease.

[0186] As indicated above, Figure 13A and Figure 13B are provided only as examples. Other examples are possible and may be different from the examples described with respect to Figure 13A and Figure 13B described.

[0187] As described above, the LMC technique requires reference values for the scout set. When transfer sets from the master instrument 205 and the target instrument 210 are both available, but the reference values for these samples are not available, the master calibration model and the master transfer set can be used to predict the reference values so that the LMC technique can be made available.

[0188] Figure 14A and Figure 14B are graphical illustrations showing example results associated with predicting reference values using the master calibration model and the master transfer set. As Figure 14A shown, using the same data set as those associated with Figures 5A - 5C , when using the true reference values, the RMSEP is 0.44. As Figure 14B shown, when using the reference values predicted using the master calibration model and the master transfer set, the RMSEP is 0.80. Although the performance of the LMC technique using predicted reference values is not as good as that of the LMC technique using true reference values, as described above, the performance is improved compared to using the MDC technique or the fLMC technique and is similar to using the PDS technique.

[0189] As indicated above, Figure 14A and Figure 14B are provided only as examples. Other examples are possible and may be different from the examples described with respect to Figure 14A and Figure 14B described.

[0190] Some embodiments described herein provide a focused LMC (fLMC) technique that can be used in connection with performing calibration model transfer. Similar to the LMC technique, the fLMC technique only requires a scout set collected by the target instrument. However, different from the LMC technique, the fLMC technique does not require a reference value for the scout set. Thus, using the fLMC technique in connection with calibration model transfer reduces the cost, difficulty, and / or complexity of calibration model transfer (e.g., as compared to the LMC technique and the general calibration model transfer techniques described above).

[0191] Some embodiments described herein provide a process in which the fLMC technique or the LMC technique uses the β coefficients of the primary calibration model associated with performing calibration model transfer without requiring a primary calibration set.

[0192] Some embodiments described herein provide a process for model update using the LMC technique.

[0193] The foregoing disclosure provides illustration and description, but is not intended to be exhaustive or to limit the embodiments to the precise forms disclosed. Modifications and variations are possible or can be obtained from practice of the embodiments, given the above disclosure.

[0194] As used herein, the term component is defined to be broadly interpreted as hardware, firmware, and / or a combination of hardware and software.

[0195] Some embodiments are described herein in connection with a threshold. As used herein, meeting a threshold can refer to a value that is greater than the threshold, more than the threshold, higher than the threshold, greater than or equal to the threshold, less than the threshold, fewer than the threshold, lower than the threshold, less than or equal to the threshold, equal to the threshold, etc.

[0196] It will be apparent that the systems and / or methods described herein can be implemented in different forms of hardware, firmware, or a combination of firmware and software. The actual specific control hardware or software code used to implement these systems and / or methods is not a limitation on the embodiments. Thus, in describing the operation and behavior of the systems and / or methods herein without reference to specific software code, it should be understood that software and hardware can be designed to implement the systems and / or methods based on the description herein.

[0197] Although specific combinations of features are recited in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of possible embodiments. In fact, many of these features can be combined in ways not specifically recited in the claims and / or not disclosed in the specification. Although each dependent claim listed below may directly depend on only one claim, the disclosure of possible embodiments includes each dependent claim in combination with every other claim in the claim set.

[0198] No element, act, or instruction used herein shall be construed as critical or essential unless so expressly described. Additionally, as used herein, the articles "a" and "an" are intended to include one or more items and may be used interchangeably with "one or more." Further, as used herein, the term "set" is intended to include one or more items (e.g., related items, unrelated items, combinations of related and unrelated items, etc.) and may be used interchangeably with "one or more." The term "one" or similar language is used where only one item is intended to be illustrated. Additionally, as used herein, the terms "has," "have," "having," etc. are intended to be open-ended terms. Further, unless expressly stated otherwise, the phrase "based on" is intended to mean "at least partially based on."

Claims

1. A method, comprising: After a calibration model is deployed to a spectroscopic instrument, receiving, by a device, an update set that includes new samples representing different conditions; And After the calibration model is deployed to the spectroscopic instrument and during online operation of the spectroscopic instrument, using, by the device, a linear model correction LMC technique and the update set to update the calibration model, Wherein the calibration model is updated based on β coefficients associated with the calibration model.

2. The method according to claim 1, wherein the calibration model is updated without using all calibration data of an existing calibration set.

3. The method according to claim 1, wherein the calibration model is updated without adding the new samples to an existing calibration set.

4. The method according to claim 1, wherein the update set is a scout set associated with performing calibration model update.

5. The method according to claim 1, further comprising: Receiving a master calibration model from a master instrument; Generating a transfer calibration model corresponding to the master calibration model; And Transmitting the transfer calibration model to a target instrument, Wherein the spectroscopic instrument is the master instrument or the target instrument, and Wherein the calibration model is the master calibration model or the transfer calibration model.

6. The method according to claim 1, wherein the spectroscopic instrument is implemented within the device.

7. The method according to claim 1, further comprising: Determining, by the device, the β coefficients associated with the calibration model.

8. The method according to claim 1, further comprising: Determining the β coefficients based on another β coefficient, using the LMC technique and the update set.

9. A device, comprising: One or more memories; And One or more processors, coupled to the one or more memories, configured to cause the device to: After a calibration model is deployed to a spectroscopic instrument, receive an update set that includes new samples representing different conditions; And After the calibration model is deployed to the spectroscopic instrument and during online operation of the spectroscopic instrument, update the calibration model based on β coefficients associated with the calibration model, using a linear model correction LMC technique and the update set.

10. The device according to claim 9, wherein the one or more processors are configured to cause the device to update the calibration model as follows: Update the calibration model without using all calibration data of an existing calibration set.

11. The device according to claim 9, wherein the one or more processors are configured to cause the device to update the calibration model as follows: Update the calibration model without adding the new samples to an existing calibration set.

12. The device according to claim 9, wherein the update set is a scout set associated with performing calibration model update.

13. The device according to claim 9, wherein the spectroscopic instrument is a target instrument, and Wherein the calibration model is a transfer calibration model.

14. The device according to claim 9, wherein the spectroscopic instrument is implemented within the device.

15. The device according to claim 9, wherein the one or more processors are further configured to cause the device to: Receive a master calibration model from a master instrument; Generate a transfer calibration model corresponding to the master calibration model; and Transmit the transfer calibration model to a target instrument, wherein the spectral instrument is the master instrument or the target instrument, and wherein the calibration model is the master calibration model or the transfer calibration model.

16. The device according to claim 9, wherein the one or more processors are further configured to cause the device to: Determine the β coefficient associated with the calibration model.

17. The device according to claim 9, wherein the one or more processors are further configured to cause the device to: Determine the β coefficient based on another β coefficient, using the LMC technique and the update set.

18. A non-transitory computer-readable medium storing a set of instructions, the set of instructions including: One or more instructions that, when executed by one or more processors of a device, cause the device to: Receive an update set after a calibration model is deployed to a spectral instrument, the update set including new samples representing different conditions; And During online operation of the spectral instrument after the calibration model is deployed to the spectral instrument and, based on a β coefficient associated with the calibration model, use the linear model correction LMC technique and the update set to update the calibration model.

19. The non-transitory computer-readable medium according to claim 18, wherein the one or more processors are configured to cause the device to update the calibration model as follows: Update the calibration model without using all calibration data of an existing calibration set and without adding the new samples to the existing calibration set.

20. The non-transitory computer-readable medium according to claim 18, wherein the spectral instrument is a master instrument, and wherein the calibration model is a master calibration model.

Images