A joint inversion method for temperature and emissivity in multispectral thermometry

Abstract

A multispectral thermometry method for joint temperature and emissivity inversion is disclosed, relating to non-contact thermometry of high-temperature targets. To address the shortcomings of existing technologies, which cannot achieve joint emissivity inversion without prior knowledge of the emissivity, this invention provides the following technical solution: establishing a multispectral thermometry model; constructing temperature calculation models for each channel; outputting multiple sub-objective functions for optimization; constructing a weighted objective function for optimization; outputting the global optimal solution as the initial point; constructing an optimization model with a penalty function, outputting the target temperature and spectral emissivity; performing simulation verification based on the inversion results; evaluating errors by setting typical emissivity curves and real temperatures; and outputting the algorithm's accuracy index. This method is suitable for non-contact, high-precision joint temperature and emissivity measurement of high-temperature targets in complex spectral environments.

Description

Technical Field

[0001] In the field of non-contact temperature measurement of high-temperature targets, specifically involving the joint inversion of temperature and emissivity in multispectral temperature measurement. Background Technology

[0002] In the field of non-contact temperature measurement of high-temperature targets, multispectral radiation thermometry has become a hot topic in current research and engineering applications due to its strong anti-interference capability and high-temperature adaptability. This type of method typically relies on multiple infrared measurement channels in different wavelength bands. By receiving the radiation intensity of the target object at different wavelengths and combining it with radiation theory models, the true temperature of the target is indirectly derived. However, in practical applications, because the target emissivity is unknown or variable, traditional temperature inversion methods based on single-channel or two-channel ratio methods are easily affected by factors such as systematic errors and differences in channel response, resulting in low accuracy of the inversion results and even instability.

[0003] In recent years, researchers have proposed strategies such as multi-objective optimization, nonlinear fitting, and intelligent optimization algorithms to improve inversion accuracy. For example, some literature utilizes intelligent algorithms such as genetic algorithms and particle swarm optimization to jointly invert temperature and emissivity parameters, improving the global convergence capability of the results; other studies have introduced statistical methods such as Bayesian estimation into temperature inversion to suppress the influence of random measurement noise. However, the above methods often have the following problems: on the one hand, the construction of the model objective function does not fully consider the requirements for measurement consistency and suppression of abnormal extrema between different channels; on the other hand, the optimization algorithms either lack global search capability and are prone to getting trapped in local optima (such as SQP-type methods), or have slow convergence speed and low local accuracy (such as standard PSO and other heuristic methods), making it difficult to balance accuracy and efficiency.

[0004] Furthermore, in some existing studies, the stability and generalizability of the inversion algorithm have not been fully verified on typical materials or real measurement data. This results in the scheme performing well in simulation environments, but its accuracy drops significantly under complex working conditions, which limits its practical application.

[0005] In summary, existing technologies have the drawback of being unable to achieve high-precision and high-stability joint inversion of temperature and emissivity without prior knowledge of the emissivity. Summary of the Invention

[0006] To address the shortcomings of existing technologies, which cannot achieve high-precision and high-stability joint inversion of temperature and emissivity without prior knowledge of the emissivity, the present invention provides the following technical solution:

[0007] A method for joint inversion of temperature and emissivity using multispectral thermometry includes:

[0008] The steps include establishing a multispectral temperature measurement model, collecting the output voltage, center wavelength, and blackbody calibration parameters of each spectral channel, constructing temperature calculation models for each channel, and outputting multiple sub-objective functions for optimization.

[0009] Three types of sub-objective functions are constructed: temperature uniformity, extreme value suppression, and channel difference constraint. These are then integrated into a single objective function through weighted summation, and the steps of the weighted objective function for optimization are output.

[0010] Based on the weighted objective function, the particle swarm optimization algorithm is used to perform a global search, and the global optimal solution is output as the initial point.

[0011] Using the global optimal solution as the initial point, an improved sequential quadratic programming algorithm is introduced to construct an optimization model with a penalty function. The Hessian matrix is ​​updated using a quasi-Newton method, and the Armojo line search strategy is combined to output the steps of the target temperature and spectral emissivity.

[0012] The steps involve performing simulation verification based on the above inversion results, evaluating errors by setting typical emissivity curves and real temperatures, and outputting the algorithm accuracy index.

[0013] Furthermore, a preferred implementation is provided in which the temperature consistency objective function is constructed based on the squared difference between the temperature measured in each channel and the average temperature.

[0014] Furthermore, a preferred implementation is provided in which the extreme value suppression objective function is used to penalize abnormal fluctuations in individual channel measurements.

[0015] Furthermore, a preferred embodiment is provided, wherein the channel difference constraint function is used to constrain the temperature difference between any two channels to not exceed a set threshold.

[0016] Furthermore, a preferred implementation is provided in which the particle swarm optimization algorithm sets the population size, inertia weight, and acceleration factor parameters, and uses the maximum number of iterations to limit the search range.

[0017] Furthermore, a preferred implementation is provided, in which the sequential quadratic programming algorithm updates the Hessian matrix using the BFGS method and embeds inequality and equality constraints into the objective function using the penalty function method.

[0018] A multispectral thermometry device for joint temperature and emissivity inversion is also provided, comprising:

[0019] A multispectral temperature measurement model is established, the output voltage, center wavelength and blackbody calibration parameters of each spectral channel are collected, the temperature calculation model under each channel is constructed, and the module for multiple sub-objective functions for optimization is output.

[0020] Three types of sub-objective functions are constructed: temperature uniformity, extreme value suppression, and channel difference constraint. These are then integrated into a single objective function through weighting, and a module is output to provide the weighted objective function for optimization.

[0021] This module is based on a weighted objective function and uses the particle swarm optimization algorithm to perform a global search, outputting the global optimal solution as the initial point.

[0022] Using the global optimal solution as the initial point, an improved sequential quadratic programming algorithm is introduced to construct an optimization model with a penalty function. The Hessian matrix is ​​updated using a quasi-Newton method, and the Armojo line search strategy is combined to output the module of target temperature and spectral emissivity.

[0023] Based on the above inversion results, simulation verification is performed, and error assessment is conducted by setting typical emissivity curves and real temperatures. The module outputs the algorithm accuracy index.

[0024] A computer storage medium is also provided for storing a computer program, which, when read by the computer, executes the method.

[0025] A computer is also provided, including a processor and a storage medium, wherein the computer executes the method when the processor reads a computer program stored in the storage medium.

[0026] A computer program product is also provided, which, when executed, implements the method described.

[0027] Compared with the prior art, the advantages of the technical solution provided by the present invention are as follows:

[0028] By introducing three types of objective functions—"temperature consistency," "extreme value suppression," and "channel difference constraint"—and integrating them into a unified weighted single objective function, temperature inversion not only pursues the minimum deviation of the average temperature but also possesses the ability to suppress outliers and constrain the consistency of errors within channels. Compared to traditional studies that construct a single objective function based solely on the average temperature error, this approach can more comprehensively measure the algorithm's optimization effect and significantly reduce systematic errors caused by fluctuations in individual channels in multi-channel measurements.

[0029] A particle swarm optimization (PSO) algorithm is used to perform a global search of the initial parameter space, obtaining a relatively optimal solution as the initial point. Then, a sequential quadratic programming (SQP) algorithm is introduced for local fine-tuning, improving the convergence and accuracy of the optimization results. This hybrid strategy combining global and local optimization effectively avoids the drawback of the traditional SQP algorithm getting trapped in local optima and also compensates for the shortcomings of the PSO algorithm in convergence accuracy and speed. Thus, the algorithm's stability and global optimization capability are improved without increasing the computational burden.

[0030] In terms of constraint handling, a penalty function method is introduced, along with BFGS to update the Hessian matrix and an Armojo line search mechanism, ensuring the convergence and iterative stability of the algorithm under constraints. Unlike some literature that directly uses standard optimization solvers for black-box processing, this scheme makes targeted improvements to each step of the SQP process, thereby enhancing the numerical stability under complex objective functions and demonstrating stronger adaptability to high-dimensional and nonlinear constrained problems.

[0031] Finally, in the simulation verification section, four representative typical emissivity curves (monotonically increasing / decreasing, convex, and concave) were selected, with the maximum error controlled within 10K and the error rate as low as 0.6%. Unlike traditional methods that often use ideal models for verification, this approach is closer to engineering practice and can more effectively test the algorithm's adaptability to complex emissivity variations, demonstrating the algorithm's engineering feasibility and practical value in complex multispectral high-temperature scenarios. It is suitable for non-contact, high-precision joint temperature and emissivity measurement of high-temperature targets in complex spectral environments. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method.

[0033] Figure 2 The simulation verification diagram shows four complex emissivity variations.

[0034] Figure 3 The diagram shows four typical emissivity curves. Detailed Implementation

[0035] To make the advantages and benefits of the technical solution provided by the present invention clearer, the technical solution provided by the present invention will now be described in further detail with reference to the accompanying drawings, specifically:

[0036] Implementation Method 1: This implementation method provides a joint inversion method for temperature and emissivity in multispectral thermometry, including:

[0037] The steps include establishing a multispectral temperature measurement model, collecting the output voltage, center wavelength, and blackbody calibration parameters of each spectral channel, constructing temperature calculation models for each channel, and outputting multiple sub-objective functions for optimization.

[0038] Three types of sub-objective functions are constructed: temperature uniformity, extreme value suppression, and channel difference constraint. These are then integrated into a single objective function through weighted summation, and the steps of the weighted objective function for optimization are output.

[0039] Based on the weighted objective function, the particle swarm optimization algorithm is used to perform a global search, and the global optimal solution is output as the initial point.

[0040] Using the global optimal solution as the initial point, an improved sequential quadratic programming algorithm is introduced to construct an optimization model with a penalty function. The Hessian matrix is ​​updated using a quasi-Newton method, and the Armojo line search strategy is combined to output the steps of the target temperature and spectral emissivity.

[0041] The steps involve performing simulation verification based on the above inversion results, evaluating errors by setting typical emissivity curves and real temperatures, and outputting the algorithm accuracy index.

[0042] The temperature consistency objective function is constructed based on the squared difference between the temperature measured in each channel and the average temperature.

[0043] The extreme value suppression objective function is used to punish abnormal fluctuations in individual channel measurements.

[0044] The channel difference constraint function is used to constrain the temperature difference between any two channels to not exceed a set threshold.

[0045] The particle swarm optimization algorithm sets parameters for population size, inertia weight, and acceleration factor, and uses a maximum number of iterations to limit the search range.

[0046] The sequential quadratic programming algorithm uses the BFGS method to update the Hessian matrix and embeds inequality and equality constraints into the objective function through the penalty function method.

[0047] Implementation Method Two: This implementation method further defines the technical solution provided in Implementation Method One. Specifically:

[0048] A high-precision joint temperature and emissivity inversion method without prior knowledge of emissivity is proposed, applicable to target temperature identification and emissivity fitting problems in multispectral non-contact temperature measurement systems. This method mainly includes the following four stages: establishing a multispectral temperature measurement model, constructing a multi-objective function and integrating it into a single objective function, solving for the optimal solution using a PSO-SQP hybrid optimization algorithm, and performing simulation and verification with actual data.

[0049] The first step is to establish a multispectral temperature measurement model.

[0050] This step is the foundation of the entire inversion method. Its core is to construct a radiative transfer model for multiple spectral channels using the law of radiation, accurately expressing the physical relationship between temperature, emissivity, and the channel measurement signals.

[0051] First, a multispectral radiation thermometer was used to acquire the radiation voltage output signal of the target in multiple bands. Each channel corresponds to a center wavelength, and its measurement results are affected by the target surface temperature, material spectral emissivity, and environmental factors. To isolate the influence of environmental interference and system response, blackbody calibration data under known laboratory conditions were incorporated into the temperature measurement model, including the standard radiation intensity or voltage response value of each channel at the reference temperature.

[0052] Secondly, based on Planck's radiation law and the proportional calibration method, the inversion expression for the target temperature in each channel is derived. Since the emissivity is unknown, the model can only provide an indirect temperature expression with emissivity as the variable. Different channels correspond to a common true temperature at the same measurement time, but due to instrument errors, random noise, and differences in channel sensitivity, the independently calculated temperature values ​​for each channel deviate. Therefore, in subsequent optimization, a unified multi-objective control mechanism needs to be introduced to ensure the consistency of the temperature results output by each channel, thereby improving the final inversion accuracy.

[0053] Finally, the indirect temperature expression and its mutual error characteristics for each channel are output, providing input basis for the design of the objective function.

[0054] The second step is to construct a multi-objective function and integrate it into a single-objective optimization model.

[0055] To address the issues of large deviations in multi-channel measurement results and poor inversion stability, this step designs multiple physically meaningful optimization objective functions based on the temperature model established in the first step, and integrates them into a unified optimization index.

[0056] Specifically, the objective functions constructed include:

[0057] (1) Temperature consistency objective function, used to measure the difference between the independent inversion temperature of each channel and its mean, reflecting the overall measurement consistency;

[0058] (2) The extreme value suppression objective function is used to identify and suppress abnormal deviations of individual channels caused by noise or system imbalance, thereby improving robustness to occasional errors;

[0059] (3) Channel difference constraint function, used to limit the temperature difference between any channel within a reasonable physical range, thereby enhancing the overall constraint of the inversion model.

[0060] The three objective functions described above are integrated using a linear weighting method to construct a weighted single objective function. To maintain the algorithm's universality and parameter-tuning-free characteristics, the weights are initially set to equal weight, meaning each sub-objective has an equal impact on the final optimized value. In practical applications, the weights can be adjusted according to different measurement scenarios to adapt to specific constraints or preferences.

[0061] Ultimately, the integrated objective function is constructed as a standard constrained optimization problem, with optimization variables including a set of emissivity parameters and a common target temperature, which serve as the unified input for subsequent global and local optimization algorithms.

[0062] The third step is to use the PSO algorithm to obtain initial values.

[0063] Considering that the integrated objective function has a nonlinear and multimodal structure, traditional gradient-based optimization algorithms are insufficient in global search capabilities and are prone to getting trapped in local minima. This step introduces Particle Swarm Optimization (PSO) as a global search method to find optimal initial parameter values ​​within the feasible region.

[0064] The PSO algorithm simulates the foraging behavior of bird flocks, achieving optimal solution approximation through the cooperative search of multiple particles in the solution space. First, the initial positions and velocities of the particle swarm are randomly initialized, and key algorithm parameters are set, including the number of particles, maximum number of iterations, individual learning factor, swarm learning factor, and inertia factor.

[0065] Subsequently, guided by the objective function, the particle swarm continuously updates its individual positions and velocities, records its individual historical best values ​​and the swarm's global best value, and iterates until it reaches a stable stage. In each iteration, the particles choose their next direction of movement based on the objective function value at their current position and historical experience information.

[0066] Once the preset accuracy condition or the maximum number of iterations is reached, the algorithm outputs the initial emissivity and initial temperature values ​​corresponding to the globally optimal particle. This initial solution is usually located in a high-quality solution space, providing a better starting point for subsequent gradient methods and effectively avoiding getting trapped in non-optimal stable points.

[0067] The fourth step is to perform local fine-tuning using the improved SQP algorithm.

[0068] After obtaining a high-quality initial solution from the PSO output, this step further introduces the Sequential Quadratic Programming (SQP) algorithm to perform a locally refined solution to the problem under constraints. SQP is an efficient nonlinear programming algorithm based on the Lagrange multiplier method and quasi-Newton method, suitable for handling constrained continuous variable optimization problems.

[0069] First, a Lagrange function is constructed to represent the original optimization problem, combining the objective function with all constraints. In each iteration, the original problem is linearly approximated as a quadratic programming subproblem. The gradient information at the current point is used as the linear term, and the Hessian matrix is ​​estimated using a quasi-Newton method (such as the BFGS algorithm) as the quadratic term, thus avoiding the direct calculation of the second derivative of the objective function.

[0070] Considering that both inequality and equality constraints must be satisfied during the solution process, this step uses the penalty function method to process all constraints, embedding the cost of violating the constraints into the objective function, and controlling the cost intensity of the constraint violation by setting a penalty factor.

[0071] To enhance the numerical stability and convergence efficiency of the algorithm, this step introduces an Armojo line search strategy. In each update round, the step size is controlled by setting a descent factor and a reduction coefficient, ensuring that the objective function monotonically decreases after each update, avoiding oscillations or ineffective moves. Simultaneously, a dynamic step size adjustment mechanism is used to adaptively adjust the current iteration amplitude based on historical step size trends, further improving the convergence speed under complex models.

[0072] Finally, when the optimization process meets the set termination conditions (such as the objective function change being less than a threshold, the maximum number of iterations being reached, and the gradient norm approaching zero), the current emissivity solution and the target temperature solution are output. This result is the final solution to the inversion problem, possessing high accuracy and high stability, and can be directly used for subsequent simulation verification or engineering measurement tasks.

[0073] Implementation Method 3: Combination Figure 1 This embodiment describes the technical solution provided above in further detail through specific examples. Specifically:

[0074] This method aims to accurately deduce the true temperature and spectral emissivity of a target without prior knowledge of its emissivity. Its core process is as follows:

[0075] Establish a multi-objective temperature inversion model → Construct the objective function and unify it into a single-objective optimization → Introduce PSO to obtain initial values ​​→ Use an improved SQP to solve for the optimal solution → Apply it to simulation and experimental verification.

[0076] Detailed steps outlined below:

[0077] 1. Establishment of a multispectral thermometry model

[0078] Output: Objective function .

[0079] A channel measurement model was established based on the radiation equation;

[0080] Introduce temperature uniformity objective F1, extreme value suppression objective F2, and channel difference constraint F3;

[0081] For a multispectral radiation thermometer with N spectral channels, the output signal of the Nth channel is acquired.

[0082] At reference temperature T ’ Next, collect the blackbody radiation received by the i-th channel;

[0083] Based on the ratio, a temperature measurement model is established:

[0084] Assume the temperature measured in the nth channel is T. n ,but

[0085]

[0086] blackbody temperature T n ’ (Given) It is the unknown quantity in the formula. This represents the target temperature obtained from the inversion calculation in the nth channel. This is the fitting constant for the nth channel at the reference temperature, which is related to the channel wavelength and calibration data. This is the emissivity influence coefficient corresponding to the nth channel. This represents the target emissivity parameter to be determined for the nth channel.

[0087] Since the target's temperature value is unique at a specific time, theoretically, the temperature measured by each channel should be equal. However, random errors in the measurement cause differences in the temperatures measured by each channel, meaning they are not completely equal. According to error theory, the smaller the standard deviation of the target temperature measured by different channels, the higher the reliability and accuracy of the measurement. Assume the target temperature measured by each channel is T. n Establish an objective function

[0088]

[0089] in, Describing the objective function of temperature uniformity

[0090] N represents the total number of spectral channels.

[0091] x is the parameter vector to be optimized, including variables such as emissivity.

[0092] The average inversion temperature for all channels:

[0093]

[0094] For F1, if the temperature of one channel deviates significantly from the true value at an iteration point, the average temperature of that channel will also deviate significantly from the true value. This negatively impacts subsequent calculations. To avoid this situation, a second objective function F2 is proposed:

[0095]

[0096] This represents the objective function for extreme value suppression.

[0097] The temperature retrieved from all channels except the nth channel is denoted as .

[0098] Ideally, the temperatures of any two channels are equal. This means the difference between them is zero.

[0099]

[0100] Therefore, the following objective function F3 can be constructed:

[0101]

[0102] This represents the objective function for channel difference constraints.

[0103] Using weighted method They are integrated into a single comprehensive objective function F', and a constrained optimization problem is constructed.

[0104] Based on the above analysis, for real-world temperature inversion problems, it is difficult to evaluate the accuracy of the algorithm results based solely on the value of a single objective function. Therefore, a linear weighted method is adopted to combine the advantages and disadvantages of various objective functions, according to F... i (x)

[0105] The importance of this lies in how the linear weighted method transforms the objective optimization problem into:

[0106]

[0107] This represents the weighted overall objective function. Let be the i-th sub-objective function, corresponding to temperature uniformity, extreme value suppression, and channel difference constraints, respectively. The weight coefficients are those corresponding to the i-th objective function.

[0108] This represents a typical single-objective optimization problem. Under the condition... When this condition is met, the optimal solution is an effective solution to the corresponding multi-objective optimization problem.

[0109]

[0110] When the importance of each sub-objective function is unknown, the equal weighting of F'(x) is a relatively compromise method, that is, all ωi are taken as 1 / 3.

[0111] In summary, the final generated single-objective function, and the following constrained optimization problem, have the following final form:

[0112]

[0113] This represents the integrated single-objective optimization function.

[0114] Let the emissivity of the nth channel be the variable to be optimized.

[0115] These are the calibration coefficient and emissivity weighting coefficient for the nth channel, respectively.

[0116] The first term corresponds to the temperature consistency objective function, which calculates the sum of squared deviations between the temperature of each channel and the average temperature.

[0117] The second term corresponds to the extreme value suppression objective function, which suppresses the deviation between the temperature of a certain channel and the average temperature of the other channels.

[0118] The third term corresponds to the channel difference constraint function, which constrains the temperature deviation between any two channels.

[0119] Three objective functions with equal weighting coefficients It is composed of weighted combinations.

[0120] As input for the next step: Formulate a mathematical expression for the constrained optimization problem, preparing for algorithm optimization.

[0121] 2. Design of PSO-SQP Hybrid Optimization Algorithm

[0122] PSO has strong global search capabilities, and the algorithm is optimized by combining it with SQP.

[0123] 2.1 PSO Algorithm: Providing Initial Values

[0124] Initialize the particle swarm's position and velocity;

[0125] Set parameters (such as c1, c2, w, popsize, velocity boundary, etc.);

[0126] Search for the globally optimal particle as the initial value for SQP.

[0127] 2.2 Improved SQP Algorithm: Fine-grained Local Search

[0128] Construct a quadratic programming subproblem;

[0129] Update the Hessian matrix using BFGS;

[0130] Armojo line search and dynamic step size are introduced to ensure stability and convergence;

[0131] The constraints are converted into penalty terms and embedded in the objective function.

[0132] As input for the next step: to form a stable and efficient optimization framework to invert the true temperature and emissivity.

[0133] The SQP algorithm is an efficient method for solving constrained optimization problems. Under certain approximate conditions, the SQP algorithm simplifies the original problem into a quadratic programming problem and attempts to find the optimal solution. If the solution is not optimal, the SQP algorithm uses this solution as a new approximate solution to construct a new quadratic programming problem and continues iterating. The corresponding Lagrangian function is:

[0134]

[0135] Let represent the Lagrangian function, used to construct constrained optimization problems.

[0136] The original objective function is...

[0137] Let be the Lagrange multiplier vector, corresponding to the constraints in the optimization problem.

[0138] x is the vector of variables to be optimized.

[0139] This represents the inner product term of the Lagrange multipliers and the variable, used to integrate constraints into the objective function.

[0140] This function is used in the SQP algorithm to calculate gradients and Hessian by combining the main function and constraint information.

[0141] The basic iterative format of the SQP algorithm is:

[0142]

[0143] It is an approximate solution. The step size is obtained through linear search. It is the solution to the following quadratic programming problem.

[0144]

[0145]

[0146] It is a Lagrange Hessian matrix. Because It might not be positive definite, therefore the direction of descent is not necessarily descent. Therefore, a quasi-Newton method for unconstrained optimization problems is introduced. Using... Approximate matrix replace The algorithm adjusts the approximation of the Hessian matrix based on the current step size and update direction, and performs correction using the Broyden-Fletcher-Goldfarb-Shanno (BFGS) formula. The problem being corrected is:

[0147]

[0148]

[0149] Furthermore, a quadratic programming subproblem, Armojo line search, and dynamic step size adjustment strategy are introduced. By defining a custom optimization problem, the constraints are incorporated into the objective function to form a comprehensive value function. :

[0150]

[0151] in It is a penalty coefficient used to control the intensity of the penalty for violating constraints. and It involves inequality constraints and equality constraints. To ensure the effectiveness of the step size, an Armojo line search strategy is introduced, combined with a dynamic step size adjustment strategy for optimization. Specifically, the goal is to find a suitable step size. This effectively reduces the objective function, significantly improving the algorithm's stability and convergence speed. Especially in high-dimensional, heavily constrained optimization problems, it effectively avoids common numerical instability issues and ensures solution quality. The Armojo line search strategy is as follows:

[0152]

[0153] in It is the current step size. It is a constant that controls the step size update.

[0154] Considering the complementary advantages and disadvantages of PSO and SQP algorithms in global and local optimization processes, a hybrid optimization algorithm combining PSO and SQP is proposed, based on the SQP algorithm. Traditional SQP trajectory optimization requires a good initial trajectory to achieve good optimization results; its global search capability is weak, but its local search capability is strong. It can converge to local optima in a relatively short time. However, for objective functions with multiple local optima, SQP is very sensitive to the given initial point and easily gets trapped in local optima, failing to obtain the global optimum. Therefore, it is combined with the PSO algorithm. The initial point is obtained through iteration using PSO, and the global optimum is obtained using SQP.

[0155] Implementation method four details the verification process and significant effects of the multispectral radiometric thermometry inversion algorithm, highlighting its superior performance in real-world engineering scenarios. Combined with... Figure 3 The explanation is as follows:

[0156] To comprehensively test the algorithm's adaptability to complex and unknown emissivity variations, this implementation selected four types of actual or highly simulated emissivity curves with typical morphological characteristics as verification samples: Model A: Monotonically decreasing emissivity; Model B: Monotonically increasing emissivity; Model C: Convex emissivity variation (increasing first, then decreasing); Model D: Concave emissivity variation (decreasing first, then increasing). These four types of curves basically cover the main emissivity variation modes that high-temperature materials may encounter in actual working conditions, providing a rigorous testing benchmark for the algorithm's robustness.

[0157] For the four typical emissivity curves mentioned above, the emissivity results obtained by the algorithm inversion were compared and analyzed with the target emissivity. The inversion results are in high agreement with the true values. The maximum absolute temperature error was strictly controlled within 10K, and the error rate was as low as 0.6% under some operating conditions. This level of accuracy is significantly better than many traditional methods that rely on ideal models for verification, fully demonstrating the effectiveness and high accuracy of the algorithm in handling complex emissivity variations.

[0158] The above description of several specific embodiments further details the technical solution provided by the present invention in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, combinations of embodiments, and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for joint inversion of temperature and emissivity using multispectral thermometry, characterized in that, include: The steps include establishing a multispectral temperature measurement model, collecting the output voltage, center wavelength, and blackbody calibration parameters of each spectral channel, constructing temperature calculation models for each channel, and outputting multiple sub-objective functions for optimization. Three types of sub-objective functions are constructed: temperature uniformity, extreme value suppression, and channel difference constraint. These are then integrated into a single objective function through weighted summation, and the steps of the weighted objective function for optimization are output. Based on the weighted objective function, the particle swarm optimization algorithm is used to perform a global search, and the global optimal solution is output as the initial point. Using the global optimal solution as the initial point, an improved sequential quadratic programming algorithm is introduced to construct an optimization model with a penalty function. The Hessian matrix is ​​updated using a quasi-Newton method, and the Armojo line search strategy is combined to output the steps of the target temperature and spectral emissivity. The steps involve performing simulation verification based on the inversion results, evaluating errors by setting typical emissivity curves and real temperatures, and outputting the algorithm's accuracy index.

2. The method for joint inversion of temperature and emissivity in multispectral thermometry according to claim 1, characterized in that, The temperature consistency objective function is constructed based on the squared difference between the temperature measured in each channel and the average temperature.

3. The method for joint inversion of temperature and emissivity in multispectral thermometry according to claim 1, characterized in that, The extreme value suppression objective function is used to punish abnormal fluctuations in individual channel measurements.

4. The method for joint inversion of temperature and emissivity in multispectral thermometry according to claim 1, characterized in that, The channel difference constraint function is used to constrain the temperature difference between any two channels to not exceed a set threshold.

5. The method for joint inversion of temperature and emissivity in multispectral thermometry according to claim 1, characterized in that, The particle swarm optimization algorithm sets parameters for population size, inertia weight, and acceleration factor, and uses a maximum number of iterations to limit the search range.

6. The method for joint inversion of temperature and emissivity in multispectral thermometry according to claim 1, characterized in that, The sequential quadratic programming algorithm uses the BFGS method to update the Hessian matrix and embeds inequality and equality constraints into the objective function through the penalty function method.

7. A multispectral thermometry device for joint inversion of temperature and emissivity, characterized in that, include: A multispectral temperature measurement model is established, the output voltage, center wavelength and blackbody calibration parameters of each spectral channel are collected, the temperature calculation model under each channel is constructed, and the module for multiple sub-objective functions for optimization is output. Three types of sub-objective functions are constructed: temperature uniformity, extreme value suppression, and channel difference constraint. These are then integrated into a single objective function through weighting, and a module is output to provide the weighted objective function for optimization. This module is based on a weighted objective function and uses the particle swarm optimization algorithm to perform a global search, outputting the global optimal solution as the initial point. Using the global optimal solution as the initial point, an improved sequential quadratic programming algorithm is introduced to construct an optimization model with a penalty function. The Hessian matrix is ​​updated using a quasi-Newton method, and the Armojo line search strategy is combined to output the module of target temperature and spectral emissivity. The module performs simulation verification based on the inversion results, evaluates errors by setting typical emissivity curves and real temperatures, and outputs the algorithm accuracy index.

8. A computer storage medium for storing computer programs, characterized in that, When the computer program is read by the computer, the computer executes the method of claim 1.

9. A computer, comprising a processor and a storage medium, characterized in that, When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1.

10. A computer program product, as a computer program, is characterized by: When the computer program is executed, it implements the method of claim 1.

Images