Multispectral radiation thermometry inversion method based on improved bayesian optimization generalized regression neural network

Abstract

The application discloses a multispectral radiation temperature measurement inversion method based on an improved Bayesian optimization generalized regression neural network, and comprises the following steps: carrying out pretreatment on an infrared multispectral image sequence, corresponding the pretreated data to a target real temperature, and constructing a data set; establishing a generalized regression neural network model for real temperature regression inversion of a target temperature measurement area; based on the data set, adopting a bionic optimization algorithm as a sampling method of Bayesian optimization, and optimizing hyperparameters in the generalized regression neural network model by using an improved Bayesian optimization algorithm, so as to inversely calculate the target real temperature by using the optimized generalized regression neural network model. The method does not need to assume an emissivity model, has higher temperature measurement efficiency and temperature measurement precision, and has wider applicability.

Description

Technical Field

[0001] This invention belongs to the field of infrared thermometry technology, specifically relating to a multispectral radiometric thermometry inversion method based on an improved Bayesian optimized generalized regression neural network. Background Technology

[0002] Multispectral radiometric thermometry is a multi-wavelength thermometry method that combines multispectral imaging with radiometric temperature measurement. It obtains the true temperature and spectral emissivity of a target by processing spectral information from multiple different spectral bands acquired by a multispectral imaging system. This method has advantages such as being non-contact, not affecting the temperature distribution of the measured target, having high measurement speed, and a large measurement range. It is particularly suitable for measuring temperature and emissivity in high-temperature and ultra-high-temperature targets in non-contact settings. In recent years, the rapid development of multispectral thermometry technology has significant scientific and practical value for high-temperature measurement in hazardous environments, monitoring of industrial production processes such as new material preparation and metal smelting and welding, and has important economic and social significance for promoting the development of scientific research and industry.

[0003] In current multispectral radiometric temperature measurement algorithms, the accurate calculation of spectral emissivity is both a key focus and a challenge. Existing methods mainly rely on establishing an emissivity coefficient model. This involves creating a system of equations based on the n spectral channels of a multispectral thermometer, containing n equations but with n+1 unknowns. This system of equations is underdetermined. By assuming an emissivity coefficient model, the true temperature and emissivity of the target can be fitted.

[0004] However, emissivity coefficient models are easily constrained by different target parameters, leading to significant uncertainties and resulting in low accuracy of the calculated spectral emissivity, thus affecting temperature measurement accuracy. Furthermore, existing data processing procedures for establishing emissivity coefficient models are quite complex, impacting temperature measurement efficiency. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a multispectral radiometric thermometry inversion method based on an improved Bayesian optimized generalized regression neural network. The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] In a first aspect, this invention proposes a multispectral radiometric thermometry inversion method based on an improved Bayesian optimized generalized regression neural network, comprising:

[0007] Step 1: Preprocess the infrared multispectral image sequence and correlate the preprocessed data with the actual temperature of the target to construct a dataset;

[0008] Step 2: Establish a generalized regression neural network model for the regression and inversion of the true temperature of the target temperature measurement area;

[0009] Step 3: Based on the dataset, a biomimetic optimization algorithm is used as the sampling method for Bayesian optimization, and the improved Bayesian optimization algorithm is used to optimize the hyperparameters in the generalized regression neural network model so as to use the optimized generalized regression neural network model to invert the target true temperature.

[0010] Secondly, this invention proposes a multispectral radiometric thermometry inversion device based on an improved Bayesian optimized generalized regression neural network, used to implement the method steps proposed in the first aspect of this invention, including:

[0011] The preprocessing module is used to preprocess the infrared multispectral image sequence and correlate the preprocessed data with the target's true temperature to construct a dataset;

[0012] The model building module is used to build a generalized regression neural network model for the regression inversion of the true temperature of the target temperature measurement area;

[0013] The optimization module is used to optimize the hyperparameters in the generalized regression neural network model based on the dataset, using a biomimetic optimization algorithm as the sampling method for Bayesian optimization, and using an improved Bayesian optimization algorithm to optimize the hyperparameters in the generalized regression neural network model, so as to use the optimized generalized regression neural network model to retrieve the target true temperature.

[0014] Thirdly, the present invention proposes an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0015] Memory, used to store computer programs;

[0016] A processor is used to execute a program stored in memory to implement the method steps proposed in the first aspect of the present invention.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method steps proposed in the first aspect of the present invention.

[0018] The beneficial effects of this invention are:

[0019] 1. The multispectral radiometric temperature measurement inversion method based on an improved Bayesian optimized generalized regression neural network provided by this invention constructs a generalized regression neural network model and introduces Bayesian optimization to optimize the hyperparameters of the network model. This allows the optimized model to achieve regression inversion of the true temperature of the target temperature measurement area. Specifically, a biomimetic optimization algorithm is used as the sampling method for Bayesian optimization. Biomimetic optimization algorithms can handle large-scale hyperparameter spaces and find the global optimum, while Bayesian optimization can better handle high-dimensional and non-convex problems. Combining the two improves the efficiency and accuracy of hyperparameter tuning, thereby enhancing the model's inversion performance. Compared to traditional data processing methods, this method does not require an emissivity model, has higher temperature measurement efficiency and accuracy, and wider applicability.

[0020] 2. The multispectral radiometric temperature inversion method based on the improved Bayesian optimized generalized regression neural network provided by this invention uses the weighted product of Gaussian kernel function and exponential kernel function as the kernel function when constructing the generalized regression neural network model. This retains both the wide application range of Gaussian kernel function and the low sensitivity of exponential kernel function to parameters, thereby further improving the temperature measurement accuracy.

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 A schematic diagram of a multispectral radiometric thermometry inversion method based on an improved Bayesian optimized generalized regression neural network provided in an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram illustrating the acquisition of a target infrared multispectral image sequence by an infrared multispectral imaging system provided in an embodiment of the present invention;

[0024] Figure 3 A flowchart illustrating the process of optimizing model hyperparameters using an improved Bayesian optimization method, provided in an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the process of Bayesian optimization of sampling point selection using a genetic algorithm, provided in an embodiment of the present invention.

[0026] Figure 5 The structural block diagram of the multispectral radiometric temperature inversion device based on an improved Bayesian optimized generalized regression neural network provided in the embodiments of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1

[0029] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a multispectral radiometric temperature inversion method based on an improved Bayesian optimized generalized regression neural network provided in an embodiment of the present invention. The multispectral radiometric temperature inversion method based on an improved Bayesian optimized generalized regression neural network provided in this embodiment specifically includes the following steps:

[0030] Step 1: Preprocess the infrared multispectral image sequence and correlate the preprocessed data with the target's true temperature to construct a dataset.

[0031] Understandably, it is necessary to obtain the infrared multispectral image sequence before preprocessing it.

[0032] In this embodiment, an infrared multispectral imaging system can be used to acquire infrared multispectral image sequences. Figure 2 The diagram illustrates the acquisition of a target's infrared multispectral image sequence by an infrared multispectral imaging system. Generally, an infrared multispectral imaging system mainly includes two modules: an optical system module and a data processing module. The optical system module is primarily used to acquire the target's infrared multispectral image sequence and can include multiple data acquisition channels. This embodiment uses a 6-channel example for illustration.

[0033] After acquiring the infrared multispectral image sequence, it can be processed using subsequent data processing modules.

[0034] In this embodiment, the infrared multispectral image sequence is preprocessed, specifically including:

[0035] 11) Based on the target temperature measurement area, spatial correlation calculation is performed on the infrared multispectral image sequence, and the images in the infrared multispectral image sequence are shifted according to the calculation results to obtain the shifted data.

[0036] Since the rotation of the filter wheel may cause the images of each channel to be misaligned, in order to accurately measure the temperature, this embodiment calculates the correlation using formula (1):

[0037]

[0038] In the formula, x and y represent images from different channels, r(x,y) represents the correlation coefficient between images x and y, and x i and y i These are the gray values ​​of image x and y at the i-th pixel position, respectively. and are the mean gray values ​​of image x and y, respectively, and n is the total number of pixels in an image.

[0039] Specifically, for an infrared multispectral image sequence, a reference channel is selected, and the absolute cross-correlation matrix between the reference channel and several other channels is calculated based on the target temperature measurement area, and the maximum value is found.

[0040] Based on the maximum value, spatial correlation calculation is performed on the first sequence of the infrared multispectral image sequence for each channel to obtain the offset value for different channels.

[0041] The offset value is used to offset the subsequent image sequence of the corresponding channel to obtain the offset data.

[0042] 12) Perform maximum and minimum value normalization on the offset data to obtain the preprocessed data.

[0043] Specifically, in order to reduce the impact of large differences in the numerical range of different input features on the model, the following formula (2) is used to perform maximum and minimum value normalization on the offset data:

[0044]

[0045] Where X is the original data, X min and X max These represent the minimum and maximum values ​​of the data, respectively.

[0046] After obtaining the preprocessed data, it is mapped to the actual temperature to construct a dataset, denoted as D = {(X...} i ,Y i )}, i=1,2,…,n,X i ∈R N Y i ∈R, where X represents the preprocessed spectral intensity vector and Y represents the corresponding true temperature.

[0047] Furthermore, the constructed dataset can be randomly divided into a training set and a test set. The training set is used to optimize and train the model, while the test set is used to test whether the model meets the requirements.

[0048] Step 2: Establish a generalized regression neural network model for the regression inversion of the true temperature of the target temperature measurement area.

[0049] Specifically, based on the constructed dataset, a corresponding regression network model can be set up, as shown below:

[0050] f(x) = ωφ(x) + b (3);

[0051] In the formula, ω is the weight vector, b is the bias, and φ(x) is a function used to map the input data from a low-dimensional space to a high-dimensional space.

[0052] The problem of solving for the unknown parameters in model (3) can be transformed into the problem of minimizing the generalized function shown in formula (4):

[0053]

[0054] In the formula, R(f) represents a generalized function. Let L represent the regularization term, C represent the penalty factor controlling the magnitude of ||ω||, and L represent the regularization term. ε (x i f) is the ε-insensitive loss function, and its expression is:

[0055]

[0056] After considering the fitting error and introducing a relaxation factor based on the principle of structural risk minimization, i.e., L... ε (f(x i ),y i )=ξ i ω and b can be obtained by solving the optimization problem of formula (6).

[0057]

[0058] In formula (6), y i For the actual temperature, ξ i and ξ i * It is a relaxation factor.

[0059] Equations (5) and (6) can be solved by introducing Lagrange multipliers, resulting in equation (7):

[0060]

[0061] For ω, b, ξ i ξ i * Taking the partial derivative and setting it to zero, we get:

[0062]

[0063] Substituting these values ​​yields the dual problem:

[0064]

[0065] Equation (9) can then be transformed into equation (10):

[0066]

[0067] Thus, the generalized regression neural network model for the regression inversion of the true temperature in the target temperature measurement area is obtained, expressed as the following equation (11):

[0068]

[0069] In the formula, f() represents the regression function of the generalized regression neural network model, x represents the training samples, and α and α' are... * Both represent Lagrange multipliers, α i and Let K(x) represent the Lagrange multiplier corresponding to the i-th training sample. i ,x j The kernel function is defined as the weighted product of the Gaussian kernel function and the exponential kernel function, where x is the weighted product of the Gaussian kernel function and the exponential kernel function. i ,x j Let b represent different training samples, and s represent bias. i t represents the constraint, C represents the penalty factor, and n represents the number of samples in the training set.

[0070] Optionally, as an implementation method, the kernel function used in model (11) takes the following form:

[0071]

[0072] In the formula, σ is the hyperparameter controlling the width of the Gaussian kernel, μ is the hyperparameter of the weighted product coefficient of the Gaussian kernel function, and ν is the hyperparameter of the weighted product coefficient of the exponential kernel function.

[0073] This embodiment uses a weighted product of the Gaussian kernel function and the exponential kernel function as the kernel function. The larger μ is, the closer the mixed kernel function is to the Gaussian kernel function; the smaller μ is, the closer the mixed kernel function is to the exponential kernel function. When μ takes an appropriate value, this kernel function retains both the wide application range of the Gaussian kernel function and the low sensitivity to parameters of the exponential kernel function, further improving the temperature measurement accuracy.

[0074] Step 3: Based on the dataset, a biomimetic optimization algorithm is used as the sampling method for Bayesian optimization, and the improved Bayesian optimization algorithm is used to optimize the hyperparameters in the generalized regression neural network model so as to use the optimized generalized regression neural network model to invert the target true temperature.

[0075] Bayesian optimization is a global optimization algorithm based on probability distributions, where the probabilistic model is represented as a combination of mean and variance. The principle of Bayesian optimization is to estimate the posterior distribution of the objective function using Bayes' theorem, then establish an alternative objective function based on past evaluation results, thereby finding the next combination of hyperparameters that minimizes the objective function. A typical Bayesian optimization process includes the following steps: 1. Establishing a Gaussian process; 2. Selecting the next sampling point, i.e., the parameter combination; 3. Conducting experiments and updating the Gaussian process; 4. Repeating steps 2-3 until the iteration terminates.

[0076] This embodiment improves upon existing Bayesian optimization by employing a biomimetic optimization algorithm to implement the sampling method within existing Bayesian optimization algorithms. Optionally, the biomimetic optimization algorithm can employ genetic algorithms, particle swarm optimization, etc. This embodiment preferably utilizes a genetic algorithm to implement the sampling method in the Bayesian optimization algorithm.

[0077] Please see Figure 3 , Figure 3 The flowchart illustrating the optimization of model hyperparameters using the improved Bayesian optimization method provided in this embodiment of the invention specifically includes step 3:

[0078] 31) Initialize the generalized regression neural network model, define the hyperparameters to be optimized, and randomly generate the prior distribution of the hyperparameters.

[0079] Specifically, we define the hyperparameter search space and prior distribution of the hyperparameters for the generalized regression neural network model, determine the hyperparameters to be optimized, including C, σ, ε, μ, ν, define the search range for each hyperparameter, and construct the prior distribution of the hyperparameters.

[0080] 32) Based on the prior distribution of hyperparameters, a Gaussian process is established using the dataset.

[0081] Specifically, a Gaussian process can be viewed as a stochastic process composed of an infinite number of Gaussian random variables, and is a nonparametric model. When x∈Ω and the objective function is f(x), then any finite number of x = {x1, x2, ..., x...} n A Gaussian process can be composed of a mean function m(x) and a covariance function k(x,x′), expressed by the following formula:

[0082] f(x)~GP(m(x),k(x,x')) (13);

[0083] In the formula, m(x) is the mean function, m(x)=E[f(x)].

[0084] k(x,x′) is the covariance function, which depends on the distance between points and can be expressed as a translation-invariant distance measure. Here, the Gaussian covariance function is used, which is also the kernel function in the generalized regression neural network model. The formula is shown in equation (12) above.

[0085] The hyperparameters C, σ, ε, μ, ν initialized in step 31), and the dataset D = {(X i ,Y i Substituting the training set in} into the generalized regression neural network model, i.e., equation (11), we can obtain each α i and Then we can obtain ω and b, and by solving for ω and b, we can obtain the mean and variance of the Gaussian process on the test set.

[0086] 33) Use a genetic algorithm to select the next evaluation point and obtain a new combination of hyperparameters.

[0087] Please see Figure 4 , Figure 4 The flowchart illustrating the Bayesian optimization sampling point selection using a genetic algorithm provided in this embodiment of the invention specifically includes:

[0088] 33a) Initialize a population with a combination of hyperparameters based on the hyperparameters of the current Gaussian process and the regression results of the current Gaussian process.

[0089] First, based on the results of the Gaussian process regression, namely the mean and variance, the range coefficient is calculated using the formula: range coefficient a = (mean - variance) / (mean + variance).

[0090] Then, using the hyperparameters C, σ, ε, μ, ν from the previous Gaussian process, an initial population is initialized (e.g., C∈[1-a,1+a]*C), and each individual in the population is a combination of hyperparameters.

[0091] At the same time, it is also necessary to set parameters such as the population size and maximum number of iterations for the genetic algorithm.

[0092] 33b) Calculate the fitness value of each individual in the population.

[0093] Specifically, in this embodiment, the reciprocal of the root mean square error between the predicted temperature and the actual temperature of the generalized regression neural network model is used as the fitness value to evaluate the predictive performance of each individual.

[0094] 33c) Determine if the iteration termination condition of the genetic algorithm is met. If it is met, output the optimal individual as the new hyperparameter combination; otherwise, proceed to step 33d.

[0095] Optionally, this embodiment uses common stopping conditions for genetic algorithms, such as reaching the maximum number of iterations or reaching a certain degree of convergence.

[0096] 33d) Perform selection, crossover, and mutation operations on the current population, and return to step 33b).

[0097] The specific processes of selection, crossover, and mutation operations in genetic algorithms can be found in existing related technologies, and will not be described in detail in this embodiment.

[0098] New combinations of hyperparameters can be obtained through iterative genetic algorithms.

[0099] 34) Update the Gaussian process using new combinations of hyperparameters.

[0100] Specifically, the prior hyperparameters are updated using the newly selected individuals, and the posterior distribution of the hyperparameters is obtained again. The new combination of hyperparameters updates the probabilistic model of the Bayesian optimization algorithm. The updated probabilistic model reflects the new estimate and uncertainty of the objective function.

[0101] 35) Repeat steps 33)-34) until the iteration termination condition of the Bayesian optimization algorithm is met, and the optimal combination of hyperparameters is obtained.

[0102] Specifically, in this embodiment, the iteration termination condition of the Bayesian optimization algorithm can be set to the mean squared error of the test set being less than a certain value. If this condition is not met, the iteration returns to the genetic algorithm to continue. If the condition is met, the optimal hyperparameter combination is selected from the last generation of the population. This set of hyperparameters gives the Bayesian model the highest posterior probability.

[0103] After iteration, the optimal hyperparameter combination of the model was obtained, thus completing model optimization. Next, the infrared multispectral image of the target to be detected is preprocessed, following the same preprocessing procedure as in step 1 above. The preprocessed image is then input into the optimized model to retrieve the target's true temperature.

[0104] This invention provides a multispectral radiometric temperature measurement inversion method based on an improved Bayesian-optimized generalized regression neural network. On one hand, it constructs a generalized regression neural network model; on the other hand, it introduces Bayesian optimization to optimize the hyperparameters of the network model, enabling the optimized model to achieve regression inversion of the true temperature of the target temperature measurement area. Specifically, a biomimetic optimization algorithm is used as the sampling method for Bayesian optimization. Biomimetic optimization algorithms can handle large-scale hyperparameter spaces and find the global optimum, while Bayesian optimization can better handle high-dimensional and non-convex problems. By combining the two, the efficiency and accuracy of hyperparameter tuning can be improved, thereby enhancing the model's inversion performance. Compared to traditional data processing methods, this method does not require an emissivity model, has higher temperature measurement efficiency and accuracy, and has wider applicability.

[0105] Example 2

[0106] Based on the first embodiment described above, and using the same inventive concept, this embodiment provides a multispectral radiometric thermometry inversion device based on an improved Bayesian optimized generalized regression neural network. Please refer to... Figure 5 , Figure 5 A structural block diagram of a multispectral radiometric thermometry inversion device based on an improved Bayesian optimized generalized regression neural network, provided in an embodiment of the present invention. The device specifically includes:

[0107] The preprocessing module is used to preprocess the infrared multispectral image sequence and correlate the preprocessed data with the target's true temperature to construct a dataset;

[0108] The model building module is used to build a generalized regression neural network model for the regression inversion of the true temperature of the target temperature measurement area;

[0109] The optimization module is used to optimize the hyperparameters in the generalized regression neural network model based on the dataset, using a biomimetic optimization algorithm as the sampling method for Bayesian optimization, and using an improved Bayesian optimization algorithm to optimize the hyperparameters in the generalized regression neural network model, so as to use the optimized generalized regression neural network model to retrieve the target true temperature.

[0110] The device provided in this embodiment can implement the method described in Embodiment 1 above. Detailed procedures can be found in Embodiment 1 above, and will not be repeated here. Therefore, this device also has high temperature measurement efficiency and accuracy.

[0111] In addition, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0112] Memory is used to store computer programs;

[0113] When the processor executes the program stored in the memory, it implements the method steps described in Embodiment 1 above. The detailed process will not be repeated here.

[0114] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed, can implement the method steps described in Embodiment 1 above.

[0115] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus (devices), or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems." Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The computer program may be stored / distributed in a suitable medium, provided with or as part of other hardware, or may take other distribution forms, such as via the Internet or other wired or wireless telecommunications systems.

[0116] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A multispectral radiation thermometry inversion method based on improved Bayesian-optimized generalized regression neural network, characterized in that, include: Step 1: Preprocess the infrared multispectral image sequence and correlate the preprocessed data with the actual temperature of the target to construct a dataset; Step 2: Establish a generalized regression neural network model for regressing and inverting the true temperature of the target temperature measurement area; the generalized regression neural network model is expressed as: ； wherein, represents a regression function expression of a generalized regression neural network model, represents a training sample, and both represent a Lagrange multiplier, and represents a Lagrange multiplier corresponding to the i training sample, represents a kernel function, defined as a weighted product of a Gaussian kernel function and an exponential kernel function, represents different training samples, represents a bias, represents a constraint condition, represents a penalty factor, represents a total number of samples of the training set; Step 3: Based on the dataset, a biomimetic optimization algorithm is used as the sampling method for Bayesian optimization, and the improved Bayesian optimization algorithm is used to optimize the hyperparameters in the generalized regression neural network model so as to use the optimized generalized regression neural network model to invert the target true temperature. Specifically, it includes: 31) Initialize the generalized regression neural network model, define the hyperparameters to be optimized, and randomly generate the prior distribution of the hyperparameters; 32) Based on the prior distribution of the hyperparameters, a Gaussian process is established using the dataset; 33) Use a genetic algorithm to select the next evaluation point and obtain a new combination of hyperparameters; 34) Update the Gaussian process using the new hyperparameter combination; 35) Repeat steps 33)-34) until the iteration termination condition of the Bayesian optimization algorithm is met, and the optimal combination of hyperparameters is obtained.

2. The multispectral radiometric temperature inversion method based on improved Bayesian optimization generalized regression neural network according to claim 1, characterized in that, In step 1, the infrared multispectral image sequence is preprocessed, specifically including: 11) Based on the target temperature measurement area, perform spatial correlation calculation on the infrared multispectral image sequence, and shift the images in the infrared multispectral image sequence according to the calculation results to obtain the shifted data; 12) Perform maximum and minimum value normalization on the offset data to obtain preprocessed data.

3. The multispectral radiometric temperature inversion method based on improved Bayesian optimization generalized regression neural network according to claim 2, characterized in that, Step 11) specifically includes: For an infrared multispectral image sequence, a reference channel is selected, and the absolute cross-correlation matrix between the reference channel and several other channels is calculated based on the target temperature measurement area, and the maximum value is found. Based on the maximum value, spatial correlation calculation is performed on the first sequence of the infrared multispectral image sequence for each channel to obtain the offset value for different channels. The offset value is used to offset the subsequent image sequence of the corresponding channel to obtain the offset data.

4. The multispectral radiometric temperature inversion method based on improved Bayesian optimization generalized regression neural network according to claim 1, characterized in that, The expression for the kernel function is: ； wherein is a hyperparameter that controls the width of the Gaussian kernel, is a hyperparameter that controls the weighting product coefficient of the Gaussian kernel function, is a hyperparameter that controls the weighting product coefficient of the exponential kernel function.

5. The multispectral radiometric temperature inversion method based on improved Bayesian optimization generalized regression neural network according to claim 1, characterized in that, Step 33) specifically includes: 33a) Initialize a population with a combination of hyperparameters based on the hyperparameters of the current Gaussian process and the regression results of the current Gaussian process; 33b) Calculate the fitness value of each individual in the population; wherein the fitness value is the reciprocal of the root mean square error between the predicted temperature and the actual temperature of the generalized regression neural network model; 33c) Determine whether the iteration termination condition of the genetic algorithm is met. If it is met, output the optimal individual as the new hyperparameter combination; otherwise, execute step 33d). 33d) Perform selection, crossover, and mutation operations on the current population, and return to step 33b).

6. A multispectral radiometric thermometry inversion device based on an improved Bayesian optimized generalized regression neural network, used to implement the method steps of any one of claims 1-5, characterized in that, include: The preprocessing module is used to preprocess the infrared multispectral image sequence and correlate the preprocessed data with the target's true temperature to construct a dataset; The model building module is used to build a generalized regression neural network model for the regression inversion of the true temperature of the target temperature measurement area; The optimization module is used to optimize the hyperparameters in the generalized regression neural network model based on the dataset by using a biomimetic optimization algorithm as the sampling method for Bayesian optimization, and by using an improved Bayesian optimization algorithm to optimize the hyperparameters in the generalized regression neural network model, so as to use the optimized generalized regression neural network model to invert the target true temperature.

7. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus; among which, the processor, communication interface, and memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to execute a program stored in the memory to implement the steps of the method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, performs the steps of the method described in any one of claims 1-5.

Images