A method for monitoring a temperature field of a sample in a neutron scattering experiment based on a graph neural network

Abstract

The present application relates to a kind of based on graph neural network's neutron scattering experimental sample temperature field monitoring method, comprising the following steps: step one, research the attenuation law of high-energy particle environment to sample surface radiation energy, step two, construct the hyperspectral imaging model of sample surface radiation, step three, using multispectral radiation thermometry theory, complete the numerical reconstruction of sample surface temperature field under high-energy particle environment;Step four, combined with hyperspectral numerical image data, construct the surface temperature field measurement proxy model based on graph neural network;The purpose of the present application is to provide a kind of based on graph neural network proxy model's neutron scattering experimental sample surface temperature field monitoring method, realize the accurate measurement of sample surface temperature field under high-energy particle environment.

Description

Technical Field

[0001] This invention relates to the field of sample surface temperature field measurement, and in particular to a method for monitoring the surface temperature field of neutron scattering experiments based on a graph neural network surrogate model. Background Technology

[0002] Internationally, large-scale scientific facilities are conducting numerous high-energy particle experiments. High-energy physics facilities such as the Spallation Neutron Source and the Synchrotron Radiation Facility use high-energy particles (protons p, electrons e, π± mesons, neutrons n, gamma photons, etc., with speeds approaching the speed of light) as "projectiles" to bombard samples in order to analyze the basic structure of the samples. Since the basic structural characteristics of materials, including electron spin, atomic thermal vibration, crystal phase transformation, and the formation of microstructures, are all closely related to temperature, in high-energy particle scattering experiments, especially in the most commonly used high-temperature variable-temperature environment experiments, the real-time and accurate measurement of the sample surface temperature field is crucial to ensuring the accuracy of the experimental results.

[0003] Current technologies for accurate online measurement of sample surface temperature fields in high-energy particle environments still face the following challenges: 1. Currently, contact methods are primarily used to measure sample surface temperature in high-energy particle environments. These methods suffer from long response times and low accuracy, failing to meet the real-time and precision requirements of high-energy physics experiments. In contrast, non-contact temperature measurement methods offer advantages such as short response times, wide temperature ranges, and the ability to measure temperature fields over long distances. However, research on the application of non-contact temperature measurement methods in extreme high-energy particle environments is limited, and the impact mechanism of high-energy particle environments on non-contact temperature measurement errors remains unclear. 2. While hyperspectral imaging methods for measuring object surface temperature fields have made some progress, frequent sample changes in high-energy particle scattering experiments often result in unknown surface emissivity, which varies significantly with sample temperature and spectral bands, leading to radiation-induced temperature measurement errors. Therefore, further research is needed to eliminate the influence of surface emissivity and achieve accurate measurement of sample surface temperature fields based on hyperspectral imaging in high-energy particle environments. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring the surface temperature field of neutron scattering experimental samples based on a graph neural network surrogate model, so as to achieve accurate measurement of the surface temperature field of samples in a high-energy particle environment.

[0005] To achieve the above objectives, the technical solution adopted by this invention is as follows: a method for monitoring the surface temperature field of a neutron scattering experimental sample based on a graph neural network surrogate model, comprising the following steps:

[0006] Step 1: Investigate the attenuation law of radiation energy on the sample surface by the high-energy particle environment;

[0007] Step 2: Construct a hyperspectral imaging model of the sample surface radiation;

[0008] Step 3: Using multispectral radiometric thermometry theory, numerical reconstruction of the surface temperature field of the sample under high-energy particle environment is completed;

[0009] Step 4: Combine hyperspectral numerical image data to construct a proxy model for surface temperature field measurement based on graph neural networks.

[0010] Preferably, the study of the mechanism of high-energy particle environment attenuation of sample surface radiation energy specifically involves: investigating typical sample materials for high-energy particle scattering experiments through literature analysis and empirical investigation, and establishing a functional relationship between sample surface emissivity and wavelength and temperature; analyzing the accuracy and response time requirements of sample environments such as high temperature, ultra-high temperature, high temperature in-situ deformation, temperature jump, and electromagnetic levitation for online measurement of sample surface temperature field; analyzing the main factors interfering with hyperspectral radiation thermometry of sample surface under high-energy particle environment, and using Maxwell's theory to analyze the influence of strong electromagnetic radiation environment on radiation transmission; for high-temperature and high-pressure absorbing gas media, using collisional broadening theory to study the pressure and temperature broadening effects of gas absorption bands, assuming the spectral lines are Lorentz lines, and using the latest HITRAN2012 high-resolution spectral database and its high-temperature version HITEMP2010 database, using the exponential-tail reciprocal statistical narrow band model (also known as the Malkmus model) to obtain the average transmittance of the gas medium bands.

[0011]

[0012] In the formula, L is the path length, X is the gas mole fraction, and P is the gas pressure. The average absorption coefficient within the spectral band. The average spectral line density within the spectral band. The average half-width of the spectral lines within the band is given; the radiation characteristic parameters of dispersed particles (including high-energy particles) are calculated using MIE scattering theory; according to MIE scattering theory, the radiation characteristics of spherical particles are related to the particle size parameter χ and the complex refractive index m, and the attenuation factor Q of spherical particles is given. e Scattering factor Q s Scattering albedo ωp and scattering phase function Φ p It can be represented as:

[0013]

[0014] In the formula, G is the projected area of ​​the particle, Θ is the angle between the scattering direction and the incident direction, and a n and b nS1 and S2 are scattering coefficients and radiation amplitude functions, respectively. Based on the radiation characteristics of the high-energy particle environment, Planck's law and Wien's law are combined to quantitatively calculate the target radiation attenuation rate at different bands and temperatures for parameters such as the concentration of the interference medium, the medium temperature, and the emissivity of the sample surface in the high-energy particle environment. The optimal response band of the hyperspectral imager in the high-energy particle environment is determined according to the principle of minimum attenuation rate.

[0015] Preferably, a hyperspectral imaging model of sample surface radiation is constructed. Specifically, this involves utilizing the geometrical optics theory and Fourier optics theory of hyperspectral imaging, combined with the geometric structure of the hyperspectral imaging optical lens and related parameters such as object-side focal length, image-side focal length, and aperture angle, and using an image transmission fiber as the light transmission medium to establish the position of the sample surface radiation point corresponding to each pixel in the hyperspectral numerical image and the spatial direction of the radiation rays reaching each pixel; based on radiometry and colorimetry theories, the correspondence between sample surface radiation at different temperatures and wavelengths and the gray values ​​of the hyperspectral numerical image is established; using digital imaging methods, the image gray values ​​are converted from analog quantities to discrete quantities to obtain an array structure composed of gray values ​​of different pixels; for a uniform temperature field, a planar precision mathematical model is used to analyze the differences between the measurement results of different pixels in the hyperspectral image and to compensate for their consistency; the influence of factors such as imaging non-uniformity of the hyperspectral optical system, measurement distance, field of view, dark current, and noise on radiation measurement errors is analyzed, and radiation measurement errors are corrected in conjunction with blackbody furnace calibration experiments; based on the above research scheme, the relationship between the gray value G of the hyperspectral numerical image and the radiation intensity distribution I of the sample surface can be expressed as:

[0016]

[0017] In the formula, the A1 matrix can be determined by the radiation measurement error calibration experiment, and G = (g1, g2...g k ...g n ), g k denoted by k, where k represents the gray value of the spectral band, n is the total number of spectral bands in the hyperspectral image, i represents a pixel in the hyperspectral image, N is the total number of pixels in the hyperspectral image, j represents a pixel on the sample surface image, and M is the total number of image points.

[0018] Preferably, the numerical reconstruction of the sample surface temperature field under high-energy particle environment specifically involves: using the calculation results of the radiation characteristics of the high-energy particle environment and the theory of medium radiation transfer, establishing a set of radiation transfer equations, radiation energy equations, and radiation boundary conditions for the high-energy particle environment; numerically solving the above equations using the backward Monte Carlo method; parametrically analyzing the influence of measurement uncertainties such as the temperature, concentration, and measurement distance of the interfering medium in the high-energy particle environment on the accuracy of hyperspectral radiometric temperature measurement, clarifying the evolution law of hyperspectral radiometric temperature measurement error with each influencing factor, and establishing a sample surface temperature calibration algorithm to eliminate interference from the high-energy particle environment based on error analysis theory; combining the functional relationship between the sample surface emissivity and wavelength and temperature, utilizing the high-dimensional spectral information contained in the hyperspectral image and the theory of multispectral radiometric temperature measurement, eliminating the influence of the unknown emissivity of the sample surface, and obtaining the correspondence between the measured radiation intensity I and the true temperature T of the sample surface.

[0019]

[0020] The A2 matrix can be determined by a calibration algorithm that corrects for the influence of the high-energy particle environment and the unknown emissivity of the sample surface. Combined with the formula relating the gray value G of the hyperspectral numerical image to the radiation intensity distribution I of the sample surface, the numerical reconstruction of the temperature field T of the sample surface from the gray value G of the hyperspectral numerical image can be achieved.

[0021] Preferably, the construction of a graph neural network temperature field measurement model based on hyperspectral numerical images specifically involves: using hyperspectral image sample data obtained through theoretical model calculations and experimental tests to train a graph neural network model to achieve regression measurement of the temperature field; considering the characteristics of hyperspectral numerical image data, researching corresponding data dimensionality reduction and feature extraction methods; transforming the preprocessed hyperspectral numerical image into a graph structure, where each node in the graph structure corresponds to a pixel in the hyperspectral numerical image; then training a fully connected deep learning neural network for each node in the graph structure, decomposing the original global temperature field reconstruction problem into several simpler local reconstruction problems; considering that the temperature of each pixel is highly correlated with the temperatures of its surrounding pixels (up, down, left, right, and diagonally), the network input parameters are the spectral gray values ​​(Gi) of the pixel and its surrounding pixels (up, down, left, right, and diagonally). l ,G2,....G k-1 G k ), measuring optical path distance L, and concentration of interfering gas medium X g Interfering gas medium pressure P g Temperature T of the interfering gas medium g Interfering particle medium concentration X P Interfering particle medium temperature T P The particle size D of the interfering particle mediumP Sample surface emissivity ε, measurement ambient temperature T S The network output parameter is the sample surface temperature T corresponding to that pixel:

[0022] T = f(P) f ;h(P h ;D))

[0023] In the formula, f(·) represents the nonlinear regression measurement function, and P f The parameters of the nonlinear regression measurement function are represented by h(·), which represents the feature extraction function based on the graph neural network. h The parameters of the unit nodes within the graph neural network domain are represented by , and D represents the sample data of the graph neural network model. The technical effects of this invention are as follows: First, it studies the attenuation law of radiant energy on the sample surface from the high-energy particle environment, establishes and solves a hyperspectral imaging measurement theoretical model for the sample surface temperature field, eliminates the influence of the high-energy particle environment and surface emissivity, and establishes the basic theory and algorithm for the correspondence between hyperspectral numerical images and the sample surface temperature field; then, using the hyperspectral image sample data obtained from theoretical model calculations and experimental tests, it combines a high-performance graph neural network model to construct a surrogate model for rapidly measuring the sample surface temperature field, aiming to improve the accuracy of experimental tests at large-scale high-energy physics facilities. Detailed Implementation

[0024] The key technical solutions of this application are as follows:

[0025] 1. Mechanism study of the attenuation of radiative energy on sample surface by high-energy particle environment

[0026] Through literature review and empirical investigation, this study investigates typical sample materials used in high-energy particle scattering experiments, establishes a functional relationship between sample surface emissivity and wavelength and temperature, and analyzes the accuracy and response time requirements of online temperature field measurement on sample surfaces under various environmental conditions, including high temperature, ultra-high temperature, high-temperature in-situ deformation, temperature jump, and electromagnetic levitation. It also analyzes the main factors interfering with hyperspectral radiometric thermometry of sample surfaces under high-energy particle environments. Maxwell's theory is proposed to analyze the influence of strong electromagnetic radiation environments on radiative transmission. For high-temperature, high-pressure absorbing gas media, collisional broadening theory is proposed to study the pressure and temperature broadening effects of gas absorption bands. Assuming the spectral lines are Lorentz-shaped, the latest HITRAN 2012 high-resolution spectral database and its high-temperature version HITEMP 2010 database are used. An exponential-tailed reciprocal statistical narrow band model (also known as the Malkmus model) is proposed to obtain the average transmittance of the gas medium.

[0027]

[0028] In the formula, L is the path length, X is the gas mole fraction, and P is the gas pressure. The average absorption coefficient within the spectral band. The average spectral line density within the spectral band. The average half-width of the spectral lines within the band is given; the radiation characteristic parameters of dispersed particles (including high-energy particles) are calculated using MIE scattering theory; according to MIE scattering theory, the radiation characteristics of spherical particles are related to the particle size parameter χ and the complex refractive index m, and the attenuation factor Q of spherical particles is given. e Scattering factor Q s Scattering albedo ωp and scattering phase function Φ p It can be represented as:

[0029]

[0030] In the formula, G is the projected area of ​​the particle, Θ is the angle between the scattering direction and the incident direction, and a n and b n S1 and S2 are scattering coefficients and radiation amplitude functions, respectively. Based on the radiation characteristics of the high-energy particle environment, Planck's law and Wien's law are combined to quantitatively calculate the target radiation attenuation rate at different bands and temperatures for parameters such as the concentration of the interference medium, the medium temperature, and the emissivity of the sample surface in the high-energy particle environment. The optimal response band of the hyperspectral imager in the high-energy particle environment is determined according to the principle of minimum attenuation rate.

[0031] 2. Construct a hyperspectral imaging model of sample surface radiation.

[0032] Utilizing the geometric optics and Fourier optics theories of hyperspectral imaging, combined with the geometric structure of the hyperspectral imaging lens and related parameters such as object-side focal length, image-side focal length, and aperture angle, and considering the use of optical fiber as the light transmission medium, this study establishes the position of the sample surface radiation point corresponding to each pixel in the hyperspectral numerical image and the spatial direction of the radiation rays reaching each pixel. Based on radiometry and colorimetry theories, the correspondence between sample surface radiation at different temperatures and wavelengths and the gray values ​​of the hyperspectral numerical image is established. A digital imaging method is used to convert the image gray values ​​from analog to discrete quantities, obtaining an array structure composed of gray values ​​from different pixels. For uniform temperature fields, a planar precision mathematical model is proposed to analyze the differences between measurement results of different pixels in the hyperspectral image and to perform consistency compensation. The influence of factors such as imaging non-uniformity of the hyperspectral optical system, measurement distance, field of view, dark current, and noise on radiation measurement errors is analyzed, and radiation measurement errors are corrected by combining blackbody furnace calibration experiments. Based on the above research scheme, the relationship between the gray value G of the hyperspectral numerical image and the radiation intensity distribution I of the sample surface can be expressed as:

[0033]

[0034] In the formula, the A1 matrix can be determined by the radiation measurement error calibration experiment, and G = (g1, g2...g k ...g n ), g k denoted by k, where k represents the gray value of the spectral band, n is the total number of spectral bands in the hyperspectral image, i represents a pixel in the hyperspectral image, N is the total number of pixels in the hyperspectral image, j represents a pixel on the sample surface image, and M is the total number of image points.

[0035] 3. Numerical reconstruction of the surface temperature field of samples under high-energy particle environment

[0036] Based on the calculation results of the radiation characteristics of the high-energy particle environment and the theory of radiative transfer in the medium, a set of radiative transfer equations, a radiation energy equation, and radiation boundary conditions for the high-energy particle environment are established. The inverse Monte Carlo method is used to numerically iterate and solve these equations. The influence of measurement uncertainties caused by factors such as the temperature, concentration, and measurement distance of the interfering medium in the high-energy particle environment on the accuracy of hyperspectral radiometric temperature measurement is analyzed parametrically. The evolution law of hyperspectral radiometric temperature measurement error with each influencing factor is clarified. Based on error analysis theory, a sample surface temperature calibration algorithm to eliminate interference from the high-energy particle environment is established. Combining the functional relationship between sample surface emissivity and wavelength and temperature, and utilizing the high-dimensional spectral information contained in the hyperspectral image and the theory of multispectral radiometric temperature measurement, the influence of unknown emissivity of the sample surface is eliminated, and the correspondence between the measured radiation intensity I and the true sample surface temperature T is obtained.

[0037]

[0038] The A2 matrix can be determined by a calibration algorithm that corrects for the influence of the high-energy particle environment and the unknown emissivity of the sample surface. Combined with the formula relating the gray value G of the hyperspectral numerical image to the radiation intensity distribution I of the sample surface, the numerical reconstruction of the temperature field T of the sample surface from the gray value G of the hyperspectral numerical image can be achieved.

[0039] 4. Research Scheme for a surrogate model for measuring the surface temperature field of samples in a high-energy particle environment

[0040] Using hyperspectral image sample data obtained from theoretical model calculations and experimental tests, a graph neural network model is trained to achieve temperature field regression measurement. Considering the characteristics of hyperspectral numerical image data, corresponding data dimensionality reduction and feature extraction methods are studied. The preprocessed hyperspectral numerical image is transformed into a graph structure, where each node corresponds to a pixel in the hyperspectral numerical image. Then, a fully connected deep learning neural network is trained for each node in the graph structure, decomposing the original global temperature field reconstruction problem into several simpler local reconstruction problems. Considering the significant correlation between the temperature of each pixel and the temperatures of its surrounding pixels (up, down, left, right, and diagonally), the network input parameters are the spectral gray values ​​(Gi, Gi, Gi, Gi) of the pixel and its surrounding pixels. l ,G2,....G k-1 G k ), measuring optical path distance L, and concentration of interfering gas medium X g Interfering gas medium pressure P g Temperature T of the interfering gas medium g Interfering particle medium concentration X P Interfering particle medium temperature T P The particle size D of the interfering particle medium P Sample surface emissivity ε, measurement ambient temperature T S The network output parameter is the sample surface temperature T corresponding to that pixel:

[0041] T = f(P) f ;h(P h ;D))

[0042] In the formula, f(·) represents the nonlinear regression measurement function, and P f The parameters of the nonlinear regression measurement function are represented by h(·), which represents the feature extraction function based on the graph neural network. h This represents the parameters of the unit nodes within the graph neural network domain, where D represents the sample data of the graph neural network model.

[0043] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0044] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring the temperature field of a neutron scattering experiment sample based on a graph neural network, characterized in that, Includes the following steps: Step 1: Investigate the attenuation law of radiation energy on the sample surface by the high-energy particle environment; Step 2: Construct a hyperspectral imaging model of the sample surface radiation; Step 3: Using multispectral radiometric thermometry theory, numerical reconstruction of the surface temperature field of the sample under high-energy particle environment is completed; Step 4: Combine hyperspectral numerical image data to construct a proxy model for surface temperature field measurement based on graph neural networks; The specific mechanism for studying the attenuation of radiative energy on sample surfaces by high-energy particle environments involves: investigating typical sample materials used in high-energy particle scattering experiments through literature analysis and empirical investigation; establishing a functional relationship between sample surface emissivity and wavelength and temperature; analyzing the accuracy and response time requirements of online temperature field measurement on sample surfaces under conditions of high temperature, ultra-high temperature, high-temperature in-situ deformation, temperature jump, and electromagnetic levitation; analyzing the main factors interfering with hyperspectral radiometric temperature measurement of sample surfaces under high-energy particle environments; and using Maxwell's theory to analyze the influence of strong electromagnetic radiation environments on radiative transmission. For high-temperature, high-pressure absorbing gas media, the collisional broadening theory is proposed to study the pressure and temperature broadening effects of gas absorption bands. Assuming the spectral lines are Lorentz-shaped, the latest HITRAN 2012 high-resolution spectral database and its high-temperature version HITEMP 2010 database are used. An exponential-tailed reciprocal statistical narrow band model is proposed to obtain the average transmittance of the gas medium. In the formula, L is the path length, X is the gas mole fraction, and P is the gas pressure. The average absorption coefficient within the spectral band. The average spectral line density within the spectral band. The average half-width of the spectral lines within the band; the radiation characteristic parameters calculated using MIE scattering theory; where the dispersed particles include high-energy particles, and according to MIE scattering theory, the radiation characteristics of spherical particles are related to the particle size parameter χ and the particle's complex refractive index m, and the attenuation factor Q of the spherical particles. e Scattering factor Q s scattering albedo ω p and scattering phase function Φ p It can be represented as: In the formula, G is the projected area of ​​the particle, Θ is the angle between the scattering direction and the incident direction, and a n and b n S1 and S2 are scattering coefficients and radiation amplitude functions, respectively. Based on the radiation characteristics of the high-energy particle environment, Planck's law and Wien's law are combined to quantitatively calculate the target radiation attenuation rate at different bands and temperatures caused by the concentration of the interference medium, the medium temperature, and the emissivity of the sample surface in the high-energy particle environment. The optimal response band of the hyperspectral imager in the high-energy particle environment is determined according to the principle of minimum attenuation rate.

2. The method for monitoring the temperature field of a neutron scattering experiment sample based on a graph neural network according to claim 1, characterized in that, A hyperspectral imaging model of sample surface radiation is constructed. Specifically, it utilizes the geometric optics theory and Fourier optics theory of hyperspectral imaging, combined with the geometry of the hyperspectral imaging optical lens and the object-side focal length, image-side focal length, and aperture angle, and uses an image transmission fiber as the light transmission medium to establish the position of the sample surface radiation point corresponding to each pixel in the hyperspectral numerical image and the spatial direction of the radiation rays reaching each pixel; based on radiometry and colorimetry theories, the correspondence between sample surface radiation at different temperatures and wavelengths and the gray values ​​of the hyperspectral numerical image is established. Digital imaging methods were used to convert image grayscale values ​​from analog to discrete quantities, obtaining an array structure composed of grayscale values ​​from different pixels. For a uniform temperature field, a planar precision mathematical model was employed to analyze the differences in measurement results between different pixels in the hyperspectral image, and consistency compensation was applied. The effects of imaging non-uniformity of the hyperspectral optical system, measurement distance, field of view, dark current, and noise on radiation measurement errors were analyzed, and blackbody furnace calibration experiments were used to correct these errors. Based on the above research, the relationship between the grayscale value G of the hyperspectral numerical image and the radiation intensity distribution I on the sample surface can be expressed as: In the formula, the A1 matrix can be determined by the radiation measurement error calibration experiment, and G = (g1, g2...g k ...g n ), g k denoted by k, where k represents the gray value of the spectral band, n is the total number of spectral bands in the hyperspectral image, i represents a pixel in the hyperspectral image, N is the total number of pixels in the hyperspectral image, j represents a pixel on the sample surface image, and M is the total number of image points.

3. The method for monitoring the temperature field of a neutron scattering experiment sample based on a graph neural network according to claim 2, characterized in that, The numerical reconstruction of the sample surface temperature field under high-energy particle environment involves establishing a set of high-energy particle environment radiation transfer equations, radiation energy equations, and radiation boundary conditions using the calculation results of the radiation characteristics of the high-energy particle environment and the theory of medium radiation transfer. The inverse Monte Carlo method is then used to numerically iterate and solve these equations. The influence of measurement uncertainties related to the temperature, concentration, and measurement distance of the interfering medium in the high-energy particle environment on the accuracy of hyperspectral radiometric temperature measurement is analyzed parametrically. The evolution law of hyperspectral radiometric temperature measurement error with each influencing factor is clarified. Based on error analysis theory, a sample surface temperature calibration algorithm to eliminate interference from the high-energy particle environment is established. Finally, combining the functional relationship between sample surface emissivity and wavelength and temperature, the influence of unknown emissivity on the sample surface is eliminated using the high-dimensional spectral information contained in the hyperspectral image and multispectral radiometric temperature measurement theory, thus obtaining the correspondence between the measured radiation intensity I and the true sample surface temperature T. The A2 matrix can be determined by a calibration algorithm that corrects for the influence of the high-energy particle environment and the unknown emissivity of the sample surface. Combined with the formula relating the gray value G of the hyperspectral numerical image to the radiation intensity distribution I of the sample surface, the numerical reconstruction of the temperature field T of the sample surface from the gray value G of the hyperspectral numerical image can be achieved.

4. The method for monitoring the temperature field of a neutron scattering experiment sample based on a graph neural network according to claim 1, characterized in that, The construction of a graph neural network temperature field measurement model based on hyperspectral numerical images involves: using hyperspectral image sample data obtained from theoretical model calculations and experimental tests, training a graph neural network model to achieve regression measurement of the temperature field; considering the characteristics of hyperspectral numerical image data, researching corresponding data dimensionality reduction and feature extraction methods; transforming the preprocessed hyperspectral numerical image into a graph structure, where each node corresponds to a pixel in the hyperspectral numerical image; then training a fully connected deep learning neural network for each node in the graph structure, decomposing the original global temperature field reconstruction problem into several simpler local reconstruction problems; considering the significant correlation between the temperature of each pixel and the temperatures of its surrounding pixels (up, down, left, right, and diagonally), the network input parameters are the spectral gray values ​​(Gi) of the pixel and its surrounding pixels (up, down, left, right, and diagonally). l ,G2,....G k-1 G k ), measuring optical path distance L, and concentration of interfering gas medium X g Interfering gas medium pressure P g Temperature T of the interfering gas medium g Interfering particle medium concentration X P Interfering particle medium temperature T P The particle size D of the interfering particle medium P Sample surface emissivity ε, measurement ambient temperature T S The network output parameter is the sample surface temperature T corresponding to that pixel: T=f(P f ;h(P h ;D)) In the formula, f(·) represents the nonlinear regression measurement function, and P f The parameters of the nonlinear regression measurement function are represented by h(·), which represents the feature extraction function based on the graph neural network. h This represents the parameters of the unit nodes within the graph neural network domain, where D represents the sample data of the graph neural network model.

Images