A Bayesian optimization-based calibration method for spectral response functions
By optimizing the spectral response function model using Bayesian optimization methods combined with high and low spectral resolution data, the problems of error and light source stability in the calibration of the spectrometer's spectral response function were solved, and high-precision spectral response function calibration was achieved.
Patent Information
- Application Number
- CN202311736285.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-18
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-12-18
AI Technical Summary
Existing technologies for obtaining the spectral response function of a spectrometer suffer from large deviations in calculation methods and are affected by the stability of narrow-linewidth light sources. There is a lack of calibration methods that can both correct errors and remain unaffected by the stability of the light source.
A Bayesian optimization-based approach is adopted, combining high-spectral-resolution and low-spectral-resolution data to optimize the model parameters of the spectral response function. The model is constructed using a Gaussian function, and the accurate spectral response function is obtained through Bayesian optimization.
Accurate calibration of the spectral response function has been achieved, meeting the requirements of high-precision cross-calibration of spaceborne spectral remote sensors and signal restoration of the spectrometer itself.
Smart Images

Figure CN117723153B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of spectral response function calibration methods for spectrometers, specifically relating to a spectral response function calibration method based on Bayesian optimization. Background Technology
[0002] Accurate measurement of the spectral response function of spectrometers is crucial for high-precision cross-calibration of spaceborne spectroscopic remote sensors and for restoring the spectrometer's own spectral signals. It addresses challenges in cross-calibration such as differences in spectral response characteristics between different payloads and spectral signal aliasing caused by insufficient spectral resolution. Currently, direct calculation and measurement methods are primarily used to obtain the instrument's spectral response function. The calculation method obtains the instrument's design parameters and uses the optical dispersion equation to solve for an approximate solution to the spectral response function. However, due to various errors in the system, the calculated results often deviate significantly from the actual values. The measurement method uses light sources that produce narrow-linewidth spectra as input sources, directly measuring the spectral impact response curve and obtaining the spectral response function through curve fitting. However, this method is significantly affected by the energy magnitude and stability of the narrow-linewidth spectral source. Currently, there is no spectral response function calibration method that can correct for various errors in the spectrometer's optical system while remaining unaffected by the stability of wide-spectrum light sources. Therefore, developing a complete spectral response function calibration method that combines the advantages of both direct calculation and measurement methods remains a critical issue that urgently needs to be addressed in the field of spectrometer spectral response function calibration. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention provides a Bayesian optimization-based spectral response function calibration method. This method establishes a theoretical model of the spectral response function using the spectrometer's design parameters, determines the parameters of the model to be optimized, and then uses a Bayesian optimization method combined with measurements of high-resolution and low-resolution spectral data to optimize the model parameters, resulting in an accurate spectral response function curve. This method meets the high-precision cross-calibration requirements of spaceborne spectral remote sensors and the accurate calibration requirements of the spectral response function for spectrometer self-spectral signal restoration.
[0004] To achieve the above objectives, the present invention provides the following technical solution:
[0005] A Bayesian optimization-based calibration method for spectral response functions includes the following steps:
[0006] Step 1: Measure the spectrum of the light source using a high-spectral-resolution reference spectrometer to obtain the spectral dataset H. spectrum ;
[0007] Step 2: Measure the spectrum of the same light source using a low-spectral-resolution uncalibrated spectrometer to obtain the spectral dataset L. spectrum ;
[0008] Step 3: Using a Gaussian function as the spectral response function model for the spectrometer to be calibrated, the model parameters to be optimized are determined as spectral resolution FWHM and the center wavelength λ of the spectral response function. c ;
[0009] Step 4: Randomly select initial values for the parameters FWHM1 and the center wavelength λ of the spectral response function. c,1 ;
[0010] Step 5: Calculate the spectral response function value using the spectral response function model, and calculate the target function return value using the measurement model of the spectrometer to be calibrated;
[0011] Step 6: Update the Bayesian optimized probabilistic surrogate model using the existing evaluation parameter dataset, and use the Bayesian optimized acquisition function to acquire the next set of evaluation parameters FWHM. i and the center wavelength λ of the spectral response function c,i , where i is the number of iterations;
[0012] Step 7: Determine if the number of iterations has been reached. If the number of iterations has not been reached, repeat steps 5 and 6.
[0013] Step 8: Output the optimal parameter values and calculate the spectral response function values using the spectral response function model.
[0014] Furthermore, in step 3, the spectral response function model is:
[0015]
[0016] Among them, R SRF Let λ be the spectral response function, and λ represent the spectral response wavelength. λ is the standard deviation. c λ is the center wavelength of the spectral response function.
[0017] Furthermore, in step 5, the measurement model of the spectrometer to be calibrated is as follows:
[0018]
[0019] Where * represents convolution operation and i represents the number of iterations.
[0020] Furthermore, in step 5, the return value of the objective function is:
[0021]
[0022] Where j is the index of the discretized spectral data point, N is the total number of discretized spectral data points, and λ j This represents the wavelength of the j-th data point.
[0023] Furthermore, in step 6, the Bayesian optimized probabilistic surrogate model is:
[0024]
[0025] Where w represents the parameters in the probabilistic proxy model, and D 1:t Let p(w|D) represent the observed data set. 1:t ) indicates that the observed data D 1:t The posterior probability distribution of w after correction, p(D) 1:t |w) is the likelihood distribution of the observed data given parameter w, p(w) is the prior probability distribution of w, and p(D) is the probability distribution of the observed data given parameter w. 1:t ) represents the marginal likelihood distribution of marginalized w.
[0026] Furthermore, in step 6, the Bayesian optimized acquisition function is:
[0027]
[0028] Where, x t+1 Let t be the size of the observed dataset, χ be the input space, x be the evaluation point within the input space, φ(·) be the cumulative density function of the standard normal distribution, v* represent the current optimal function value, and μ be the next evaluation point. t (x) is the mean of the function values, σ t (x) is the standard deviation of the function values.
[0029] The beneficial effects of this invention are:
[0030] This invention uses a Gaussian function to model the spectral response function, and uses a Bayesian optimization method combined with measured spectral data to obtain various parameter values required to calculate the spectral response function. This satisfies the high-precision cross-calibration requirements of spaceborne spectral remote sensors and the accurate calibration requirements of the spectral response function for the restoration of the spectral signal of the spectral instrument itself. Attached Figure Description
[0031] Figure 1 This is a schematic diagram of a Bayesian optimization-based spectral response function calibration method according to the present invention.
[0032] Figure 2 It is a high-spectral-resolution reference spectral data plot;
[0033] Figure 3 This is a spectral data graph of a low-spectral-resolution spectrometer to be calibrated;
[0034] Figure 4 This is a graph showing the relationship between the number of iterations and the minimum objective value;
[0035] Figure 5 This is a graph showing the calibration results of the spectral response function of the present invention. Detailed Implementation
[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific implementations.
[0037] like Figure 1 As shown, this invention provides a spectral response function calibration method based on Bayesian optimization, the specific steps of which are as follows:
[0038] Step 1: Extract TSIS-1SIM HSRS spectra with a resolution of 0.005 nm and a wavelength range of 700–800 nm. Figure 2 The mixed solar reference spectrum shown is used as the spectrum measured by the hyperspectral resolution reference spectrometer;
[0039] Step 2: Using a measurement model, the high-resolution spectral data generated in Step 1 is compared with the data with a spectral resolution FWHM of 20 nm and a spectral response function center wavelength λ. c Convolution is performed on the spectral response function at 750 nm to generate Figure 3 The spectrum measured by the low spectral resolution spectrometer to be calibrated is shown below.
[0040] Step 3: Input the data obtained in Step 1 and Step 2 into the Bayesian optimization model and optimize the spectral resolution FWHM and the center wavelength λ of the spectral response function. c parameter;
[0041] Step 4: Randomly generate initial values for the spectral resolution and the center wavelength of the spectral response function, FWHM0 and λ. c,0 ;
[0042] Step 5: Calculate the specific spectral response function values corresponding to the parameters in Step 4 using the spectral response function model, and combine this with the measurement model to calculate the return value of the objective function;
[0043] Step 6: Update the Bayesian optimized probabilistic surrogate model using the existing evaluation parameter dataset, and use the Bayesian optimized acquisition function to acquire the next set of spectral resolution and spectral response function center wavelength values (FWHM). i and λ c,i ;
[0044] Step 7: Repeat steps 5 and 6 until the maximum number of iterations is reached and the result is generated. Figure 4 The graph showing the relationship between the number of iterations and the minimum objective value is shown below.
[0045] Step 8: Based on the relationship diagram generated in Step 7, determine the optimal spectral resolution and the center wavelength of the spectral response function as 750.10 nm and 21.71 nm, respectively. Then, calculate the spectral response function value using the spectral response function model. The calculation results are as follows: Figure 5 As shown.
[0046] Specifically, the hyperspectral resolution reference spectral data extracted in step 1 is as follows:
[0047] H spectrum
[0048] The low-spectral-resolution spectrometer spectral data extracted in step 2 are as follows:
[0049] L spectrum
[0050] In steps 2 and 5, the measurement model is:
[0051]
[0052] Where * represents convolution operation and i represents the number of iterations.
[0053] In steps 5 and 8, the spectral response function model is as follows:
[0054]
[0055] Among them, R SRF Let λ be the spectral response function, and λ represent the spectral response wavelength. λ is the standard deviation. c λ is the center wavelength of the spectral response function.
[0056] In step 5, the return value of the objective function is:
[0057]
[0058] Where j is the index of the discretized spectral data point, N is the total number of discretized spectral data points, and λ j This represents the wavelength of the j-th data point.
[0059] In step 6, the Bayesian optimized probabilistic surrogate model is:
[0060]
[0061] Where w represents the parameters in the probabilistic proxy model, and D 1:t Let p(w|D) represent the observed data set. 1:t ) indicates that the observed data D 1:t The posterior probability distribution of w after correction, p(D) 1:t |w) is the likelihood distribution of the observed data given parameter w, p(w) is the prior probability distribution of w, and p(D) is the probability distribution of the observed data given parameter w. 1:t ) represents the marginal likelihood distribution of marginalized w.
[0062] In step 6, the Bayesian optimized acquisition function is:
[0063]
[0064] Where, x t+1 Let t be the size of the observed dataset, χ be the input space, x be the evaluation point within the input space, φ(·) be the cumulative density function of the standard normal distribution, v* represent the current optimal function value, and μ be the next evaluation point. t (x) is the mean of the function values, σ t (x) is the standard deviation of the function values.
[0065] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Those skilled in the art can still make various modifications to the foregoing technical solutions within the scope of their knowledge without departing from the spirit of the present invention.
Claims
1. A calibration method for spectral response functions based on Bayesian optimization, characterized in that, The method includes the following steps: Step 1: Measure the spectrum of the light source using a high-spectral-resolution reference spectrometer to obtain the spectral dataset Hspectrum; Step 2: Measure the spectrum of the same light source using a low-spectral-resolution uncalibrated spectrometer to obtain the spectral dataset Lspectrum; Step 3: Using a Gaussian function as the spectral response function model for the spectrometer to be calibrated, the parameters of the spectral response function model to be optimized are determined as the spectral resolution FWHM and the center wavelength λ of the spectral response function. c ; Step 4: Randomly select initial values for the parameters FWHM0 and the center wavelength λ of the spectral response function. c,0 ; Step 5: Calculate the spectral response function value using the spectral response function model, and calculate the target function return value using the measurement model of the spectrometer to be calibrated; Step 6: Update the Bayesian optimized probabilistic surrogate model using the existing evaluation parameter dataset, and use the Bayesian optimized acquisition function to acquire the next set of evaluation parameters FWHM. i and the center wavelength λ of the spectral response function c,i , where i is the number of iterations; Step 7: Determine if the number of iterations has been reached. If the number of iterations has not been reached, repeat steps 5 and 6. Step 8: Output the optimal parameter values and calculate the spectral response function values using the spectral response function model.
2. The spectral response function calibration method based on Bayesian optimization according to claim 1, characterized in that, In step 3, the spectral response function model is: Among them, R SRF Let λ be the spectral response function, and λ represent the spectral response wavelength. λ is the standard deviation. c λ is the center wavelength of the spectral response function.
3. The spectral response function calibration method based on Bayesian optimization according to claim 1, characterized in that, In step 5, the measurement model of the spectrometer to be calibrated is as follows: Where * represents convolution operation and i represents the number of iterations.
4. The spectral response function calibration method based on Bayesian optimization according to claim 1, characterized in that, In step 5, the return value of the objective function is: Where j is the index of the discretized spectral data point, N is the total number of discretized spectral data points, and λ j This represents the wavelength of the j-th data point.
5. The spectral response function calibration method based on Bayesian optimization according to claim 1, characterized in that, In step 6, the Bayesian optimized probabilistic surrogate model is: Where w represents the parameters in the probabilistic proxy model, and D 1:t Let p(w|D) represent the observed data set. 1:t ) indicates that the observed data D 1:t The posterior probability distribution of w after correction, p(D) 1:t |w) is the likelihood distribution of the observed data given parameter w, p(w) is the prior probability distribution of w, and p(D) is the probability distribution of the observed data given parameter w. 1:t ) represents the marginal likelihood distribution of marginalized w.
6. The spectral response function calibration method based on Bayesian optimization according to claim 1, characterized in that, In step 6, the Bayesian optimized acquisition function is: Where, x t+1 Let t be the size of the observed dataset, χ be the input space, x be the evaluation point within the input space, φ(·) be the cumulative density function of the standard normal distribution, v* represent the current optimal function value, and μ be the next evaluation point. t (x) is the mean of the function values, σ t (x) is the standard deviation of the function values.
Citation Information
Patent Citations
Infrared thermal imaging temperature high-precision calibration method based on Bayesian inference
CN109932059A
Spectrometer calibration method and device, electronic equipment and storage medium
CN116007754A
