An Adaptive Selection Method for Multispectral Radiation Thermometry Emissivity Model

By adaptively selecting the emissivity model using a multispectral imaging system and the AIC evaluation index, the problem of model mismatch in traditional radiation thermometry methods is solved, achieving high-precision temperature field reconstruction and accurate characterization of spectral emissivity.

CN123111475APending Publication Date: 2026-11-17XIAN TECH UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611228427.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Under complex or variable operating conditions, traditional radiation thermometry methods rely on fixed emissivity models, which can lead to model mismatch, overfitting, or underfitting, making it difficult to achieve high-precision temperature sensing and real-time monitoring.

Method used

A multispectral imaging system was used to acquire radiance signals, and a set of candidate emissivity models was constructed. By combining Planck's radiation law and the nonlinear least squares algorithm, the optimal emissivity model was selected using the AIC evaluation index. Adaptive selection was achieved by quantitatively evaluating the sum of squared residuals and the degrees of freedom of the model.

Benefits of technology

It achieves adaptive matching of emissivity models without manual prediction, reduces fitting residuals, prevents overfitting and underfitting, improves the accuracy and stability of temperature calculation, and adapts to temperature field reconstruction under complex working conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111475A_ABST
    Figure CN123111475A_ABST
Patent Text Reader

Abstract

This invention discloses an adaptive selection method for emissivity models in multispectral radiometric thermometry, solving the problems of "underfitting" and "overfitting" of emissivity models caused by nonlinear changes in the spectral emissivity of material surfaces under high-temperature industrial and extreme environments. In this invention, the AIC information criterion is introduced into multi-band radiometric thermometry, establishing a quantitative evaluation and decision framework of "sum of squared fitting residuals - model coefficients." This allows for pixel-by-pixel adaptive dynamic matching of the emissivity model without manual prediction or rigid specification of model shape. The sum of squared fitting residuals accurately captures the spectral characteristics of strongly non-gray body regions such as flame edges and areas of drastic changes in combustion products, eliminating the underfitting distortion of traditional gray body models. Simultaneously, relying on the penalty mechanism of the AIC information criterion on the degrees of freedom of the emissivity model, meaningless high-order redundant parameters are automatically eliminated, ensuring a reduction in fitting residuals while preventing abnormal fluctuations in temperature calculation results in edge regions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to an adaptive selection method for a multispectral radiometric emissivity model. Background Technology

[0002] With the rapid development of high-temperature industries, metallurgical manufacturing, aerospace, and additive manufacturing, achieving non-contact, high-precision temperature sensing and real-time monitoring under complex or variable operating conditions has become a core research challenge. Radiation thermometry, with its advantages of not interfering with the temperature field, fast response speed, and wide measurement range, is widely used in various extreme or inaccessible high-temperature environments. However, in practical applications, the accuracy of radiation thermometry is highly dependent on the spectral emissivity of the target surface. In complex industrial environments, the emissivity of material surfaces is not a constant but dynamically changes with various factors such as temperature, wavelength, surface roughness, oxidation level, and phase transition state. Accurately establishing or selecting an emissivity model that fits the actual situation is the core bottleneck for eliminating the interference of "unknown emissivity" on radiation thermometry and achieving accurate true temperature restoration.

[0003] Traditional methods for processing emissivity in radiometric thermometry mainly fall into three categories: constant assumption method, pre-defined empirical function fitting, and colorimetric / multi-band proportional restoration method. Simple monochromatic or dual-color thermometry typically assumes that emissivity is constant or independent of wavelength. However, when materials undergo high-temperature oxidation or phase transitions, this assumption can lead to significant temperature measurement errors. While multi-band radiometric thermometry introduces multi-wavelength information, it generally relies on a pre-defined, fixed emissivity model. These methods primarily depend on prior assumptions or fixed mathematical fitting. Although they may achieve certain results in ideal or specific single scenarios, in complex and variable real-world measurement environments, rigidly using a single, fixed emissivity model can easily lead to "model mismatch" problems. If the pre-defined model is too simple, it cannot accurately represent the spectral changes on the material surface; if the pre-defined model is too complex, it can cause severe ill-conditioned solutions and overfitting phenomena when solving multi-band nonlinear equations, resulting in serious distortion of temperature and emissivity calculation results and a lack of adaptability to dynamic operating conditions and complex surface features.

[0004] Patent CN113776675A discloses a multispectral radiation thermometry method based on a generalized inverse neural network, attempting to classify data and emissivity models by comparing and contrasting them using a neural network. However, this method heavily relies on a large amount of prior simulation data for network training, making the training process cumbersome and limiting its generalization and online adaptive capabilities when facing complex and unknown real-world conditions. Patent CN110017904B discloses a CCD camera multispectral algorithm for solving true temperature and emissivity, but its solution process is based on a fixed model algorithm framework, lacking a decision-making mechanism for dynamically evaluating and switching model forms based on real-time spectral data. This makes it difficult to guarantee solution accuracy when there are drastic changes in material or surface conditions. Neither of these methods addresses the model mismatch and overfitting / underfitting problems caused by fixed models or rigid classifications at their mathematical root. Summary of the Invention

[0005] The purpose of this invention is to solve the problem of "underfitting" or "overfitting" of the emissivity model selected or established due to the nonlinear change of the spectral emissivity of the material surface under high temperature industrial and extreme environments, and to provide an adaptive selection method for multispectral radiation thermometric emissivity model.

[0006] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0007] An adaptive selection method for a multispectral radiometric emissivity model, characterized by the following steps:

[0008] Step 1: Use a multispectral imaging system to acquire the radiance signals of the high-temperature target under test at multiple specific wavelengths, and perform photoelectric conversion and preprocessing to obtain the true monochromatic radiance at multiple specific wavelengths;

[0009] Step 2: Based on the material physical properties of the high-temperature target being measured, construct a set of candidate emissivity models with different complexities and function forms;

[0010] Step 3: For each candidate emissivity model in the candidate emissivity model set, establish an overdetermined system of equations based on Planck's radiation law, input the real monochromatic radiance into the overdetermined system of equations, use the nonlinear least squares algorithm to solve for the optimal estimated temperature and optimal model coefficients under the model, and calculate the sum of squared fitting residuals.

[0011] Step 4: Calculate the AIC evaluation index corresponding to each candidate emissivity model based on the sum of squared residuals, and correct the AIC evaluation index. Select the optimal emissivity model based on the corrected AIC evaluation index.

[0012] Furthermore, the multispectral imaging system includes a collimating focusing lens, a filter array, and a photodetector connected in sequence. The data output terminal of the photodetector is connected to a PC processing terminal, which is used to execute steps 2-4.

[0013] Furthermore, step 1 specifically includes:

[0014] Step 1.1: Set the collimating and focusing lens opposite to the high-temperature target being measured. The thermal radiation emitted by the high-temperature target is focused by the collimating and focusing lens onto the filter array, and then split by the filter array to obtain narrowband monochromatic light.

[0015] Step 1.2: Narrowband monochromatic light is incident on the photodetector and converted into a photocurrent signal. The photocurrent signal is then processed sequentially through transimpedance current-to-voltage conversion, programmable amplification, filtering and noise reduction, dark current compensation, and impedance matching to obtain the true monochromatic radiance of multiple specific wavelengths, which is then input to the PC processing terminal.

[0016] Furthermore, the filter array employs N narrowband filters with different center wavelengths, covering a wavelength range of... or .

[0017] Further, in step 2, the expression for the candidate emissivity model set is: ;

[0018] in, Represents the i-th specific wavelength The corresponding unknown coefficient vector is The candidate emissivity models; where m represents the ordinal number of the candidate emissivity models and M represents the total number of candidate emissivity models;

[0019] Unknown coefficient vector The expression is:

[0020] ;

[0021] Indicates the first The number of parameters to be determined in each candidate emissivity model Let each of the following represent the first, ..., second emissivity values ​​of the m-th candidate emissivity model: Model coefficients.

[0022] Furthermore, step 3 specifically involves:

[0023] Step 3.1: For the m-th candidate emissivity model, expand the vector of unknown coefficients to be solved into unknowns that include the true temperature. , represented as: ;in, Represents the actual temperature; unknown quantity quantity Represented as: The number of channels N in a multispectral imaging system satisfies: ;

[0024] Step 3.2: For the m-th candidate emissivity model, establish an overdetermined set of equations using Planck's radiation law;

[0025] Step 3.3: Input the true monochromatic radiance into the overdetermined equation system, and use the nonlinear least squares algorithm to solve for the minimum value of the m-th candidate emissivity model. To obtain the optimal estimated temperature and optimal model coefficients The solution formula is:

[0026] ;

[0027] in, This represents the true monochromatic radiance at the i-th specific wavelength obtained by the multispectral imaging system. This represents the m-th candidate emissivity model. This represents the i-th specific wavelength. This represents the theoretical radiance calculated according to Planck's blackbody radiation law;

[0028] Step 3.4: Based on the optimal estimated temperature and optimal model coefficients The sum of squared residuals is calculated using the following formula:

[0029] .

[0030] Furthermore, the expression for the theoretical radiance calculated according to Planck's blackbody radiation law is as follows:

[0031] ;

[0032] In the formula, The first radiation constant has a value of [value missing]. , The second radiation constant has a value of .

[0033] Furthermore, step 4 specifically involves:

[0034] Step 4.1: Calculate the AIC evaluation index for the m-th candidate emissivity model. The calculation formula is:

[0035] ;

[0036] Step 4.2: Set the number of channels N and the unknowns. quantity Proportional threshold When satisfied The modified Akaike Information Content Criterion was used to evaluate the AIC indicators. For high-precision fine-tuning, the fine-tuning formula is:

[0037] ;

[0038] This indicates the revised AIC assessment metrics;

[0039] Step 4.3: Select the modified minimum AIC evaluation index, the expression of which is:

[0040] ;

[0041] This represents the ordinal number of the candidate emissivity model corresponding to the minimum AIC evaluation metric;

[0042] Step 4.4, place the first The candidate emissivity model is selected as the optimal emissivity model.

[0043] Compared with the prior art, the present invention has the following beneficial technical effects:

[0044] 1. The present invention provides an adaptive selection method for emissivity models in multispectral radiometric thermometry, which introduces AIC information criteria into the field of multiband radiometric thermometry, establishes a quantitative evaluation decision framework of "fitting residual sum of squares (RSS) - model degrees of freedom (model coefficients)", and realizes pixel-by-pixel adaptive dynamic matching of emissivity models without manual prediction or rigid specification of model shape.

[0045] 2. The present invention provides an adaptive selection method for the emissivity model of multispectral radiometric thermometry. It uses residual terms to accurately capture the spectral characteristics of strong non-gray body regions such as the flame edge and the region of drastic changes in combustion products, which can eliminate the underfitting distortion of traditional gray body models. At the same time, relying on the penalty mechanism of the AIC information criterion on the degrees of freedom of the emissivity model, it automatically eliminates meaningless high-order redundant parameters, which can prevent abnormal fluctuations in the temperature calculation results in the edge region while ensuring the reduction of fitting residuals.

[0046] 3. The present invention provides an adaptive selection method for emissivity models in multispectral radiometric thermometry, which constructs a computational closed loop of "candidate emissivity model set → nonlinear least squares algorithm solution → minimization judgment of modified AIC evaluation index". It has the advantages of fast convergence speed and low computational overhead, and can perfectly meet the needs of high-precision pixel-by-pixel reconstruction of complex two-dimensional / three-dimensional temperature fields from single-point radiometric thermometry.

[0047] 4. The present invention provides an adaptive selection method for emissivity models of multispectral radiometric thermometry, which can adaptively select the best-fitting emissivity model pixel by pixel. Compared with the traditional full-field gray body model, which cannot take into account spatial non-uniformity, the method of the present invention significantly improves the accuracy of edge temperature field reconstruction. Attached Figure Description

[0048] Figure 1 A flowchart illustrating an embodiment of an adaptive selection method for a multispectral radiometric emissivity model provided by the present invention;

[0049] Figure 2 This is a schematic diagram of the structure of the multispectral imaging system in an embodiment of the present invention;

[0050] Figure 3 This is a grayscale image of a candle flame in an embodiment of the present invention;

[0051] Figure 4 This is a distribution diagram of the full-field emissivity model adaptively selected based on the AIC information criterion in this embodiment of the invention;

[0052] Figure 5 This is a pseudo-color image of the reconstructed two-dimensional temperature field distribution of a candle flame in an embodiment of the present invention;

[0053] Figure 6 This is a histogram comparing the fitting residuals of the full-field grayscale model and the emissivity model adaptively selected by the AIC information criterion in this embodiment at typical pixels at the edge of the flame.

[0054] Figure label:

[0055] 1-The high-temperature target to be measured; 2-Multispectral imaging system; 3-Collimating and focusing lens; 4-Filter array; 5-Photodetector; 6-PC processing terminal. Detailed Implementation

[0056] To make the objectives, advantages, and features of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are merely used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0057] This embodiment provides an adaptive selection method for emissivity models in multispectral radiometric thermometry. It innovatively introduces the Akaike Information Criterion (AIC) from statistics into the emissivity model decision-making process for radiometric thermometry, constructing an adaptive selection framework for emissivity models based on the AIC evaluation mechanism. By quantitatively balancing the model fitting residuals and model parameter complexity, it achieves the effect of automatically selecting the optimal emissivity model from the candidate emissivity model set without manual prediction or rigid specification of the model form. Compared with traditional fixed-model methods, it achieves a dual breakthrough in scene adaptability and true-temperature restoration accuracy.

[0058] To better achieve high-precision synchronous calculation of true temperature and emissivity under complex operating conditions, this embodiment deeply optimizes the adaptive selection and solution architecture based on AIC, building upon the candidate emissivity model set. At the evaluation and decision-making stage, the AIC information criterion automatically applies mathematical penalties to overly complex candidate emissivity models, effectively preventing overfitting risks caused by blindly increasing model complexity in pursuit of minimal residual fit. Simultaneously, goodness-of-fit evaluation is combined to prevent underfitting due to overly simple models, ensuring that the selected model accurately represents the true spectral emissivity trend while possessing high robustness. Furthermore, in the overdetermined equations solution stage, the adaptively selected optimal model reduces the ill-conditioned nature of the coefficient matrix, significantly improving the convergence speed and noise resistance stability of the joint calculation of true temperature and spectral emissivity, completely resolving the solution distortion problem caused by improper model selection in traditional multi-band radiation thermometry methods under complex conditions. This embodiment provides an adaptive selection method for multi-spectral radiation thermometry emissivity models, such as... Figure 1 As shown, it includes the following steps:

[0059] Step 1: Acquire the radiance signals of the high-temperature target under test at multiple specific wavelengths using a multispectral imaging system 2, and perform photoelectric conversion and preprocessing to obtain multiple true monochromatic radiance values; specifically:

[0060] Step 1.1, as follows Figure 2 As shown, a collimating and focusing lens 3, a filter array 4, and a photodetector 5 are connected in sequence. The data output terminal of the photodetector 5 is connected to a PC processing terminal 6 to form a multispectral imaging system 2. The collimating and focusing lens 3 is positioned opposite the high-temperature target 1 being measured. The thermal radiation emitted by the high-temperature target 1 is focused by the collimating and focusing lens 3 onto the filter array 4, and then dispersed by the filter array 4 to obtain narrowband monochromatic light. The filter array 4 uses N narrowband filters with different center wavelengths, covering a wavelength range of... or .

[0061] Step 1.2: Narrowband monochromatic light is incident on photodetector 5 and converted into a photocurrent signal. The photocurrent signal is then processed sequentially through transimpedance current-to-voltage conversion, programmable amplification, filtering and noise reduction, dark current compensation, and impedance matching to obtain multiple true monochromatic radiance values. These values ​​are then input to PC processing terminal 6 for further processing.

[0062] In this embodiment, the photodetector 5 can be a CCD / CMOS or InGaAs detector.

[0063] Step 2: To address the complex emissivity variations of different materials (such as metals, ceramics, and oxides) at high temperatures, this embodiment constructs a set of M candidate emissivity models with different complexities and function forms based on the possible physical properties of the material of the high-temperature target 1 being tested. The expression for the candidate emissivity model set is: ;in, Indicates wavelength as And the vector of unknown coefficients is The candidate emissivity models; where m represents the ordinal number of the candidate emissivity models and M represents the total number of candidate emissivity models;

[0064] Unknown coefficient vector The expression is:

[0065] ;

[0066] Indicates the first The number of parameters to be determined in each candidate emissivity model Let each of the following represent the first, ..., second emissivity values ​​of the m-th candidate emissivity model: Model coefficients.

[0067] Each candidate emissivity model corresponds to a complexity and function form.

[0068] Step 3: For each candidate emissivity model in the candidate emissivity model set, establish an overdetermined system of equations based on Planck's radiation law. Input the actual monochromatic radiance into the overdetermined system of equations, and use a nonlinear least squares algorithm to solve for the optimal estimated temperature and optimal model coefficients under that model. Calculate the sum of squared fitting residuals. Specifically:

[0069] Step 3.1: For the m-th candidate emissivity model, expand the vector of unknown coefficients to be solved into unknowns that include the true temperature. , represented as: ;in, Represents the actual temperature; unknown quantity quantity Represented as: The number of channels N in the multispectral imaging system 2 satisfies: .

[0070] Step 3.2: Input the true monochromatic radiance of each specific wavelength into the m-th candidate emissivity model, and establish an overdetermined set of equations in conjunction with Planck's radiation law.

[0071] Step 3.3: Use the Levenberg-Marquardt (LM) nonlinear least squares algorithm to find the minimum value of the m-th candidate emissivity model. To obtain the optimal estimated temperature and optimal model coefficients The solution formula is:

[0072] ;

[0073] in, This represents the true monochromatic radiance at the i-th specific wavelength obtained by the multispectral imaging system 2. This represents the m-th candidate emissivity model. Let represent the i-th specific wavelength, and N represent the number of channels in the multispectral imaging system 2. In this embodiment, the multispectral imaging system 2 selected has 9 channels. This represents the theoretical radiance calculated according to Planck's blackbody radiation law; according to Planck's radiation law. The expression is:

[0074] ;

[0075] In the formula, The first radiation constant has a value of [value missing]. , The second radiation constant has a value of .

[0076] Step 3.4: Based on the optimal estimated temperature and optimal model coefficients Calculate the residual sum of squares (RSS) of the fitted data. The calculation formula is:

[0077] .

[0078] Step 4, relying solely on Minimization can lead to a tendency to select the candidate emissivity model with the most model coefficients (a complex, high-order model), thus misdiagnosing high-frequency noise as emissivity peaks / troughs and causing severe overfitting. Conversely, forcibly specifying a candidate emissivity model with fewer model coefficients (a low-order model) can result in insufficient model expressive power and underfitting. This embodiment introduces the Akaike Information Criterion (AIC) from statistics. It calculates the AIC evaluation index for each candidate emissivity model based on the sum of squared residuals, corrects the AIC evaluation index, and selects the optimal emissivity model based on the corrected AIC evaluation index. Specifically:

[0079] Step 4.1: Calculate the AIC evaluation index for the m-th candidate emissivity model. The calculation formula is:

[0080] ;

[0081] In the formula: the first term This term measures the goodness of fit of the model (the smaller the residuals, the smaller this term); the second term This is a complexity penalty term (the more unknown parameters the model has, the greater the penalty).

[0082] Step 4.2: In this embodiment, the number of channels N=9, which falls within the small sample size range. The number of channels N and the unknown quantity are set. quantity The ratio threshold is 40, when the condition is met. To avoid the limitation of the standard still favoring complex models under small sample conditions, a modified Akaike information content criterion is adopted for the AIC evaluation index. For high-precision fine-tuning, the fine-tuning formula is:

[0083] ;

[0084] This indicates the revised AIC assessment metrics;

[0085] Step 4.3: Select the modified minimum AIC evaluation index, the expression of which is:

[0086] ;

[0087] This represents the ordinal number of the candidate emissivity model corresponding to the minimum AIC evaluation metric;

[0088] Step 4.4, place the first The candidate emissivity models are selected as the optimal emissivity model. This decision-making mechanism can automatically achieve the optimal balance between "pursuing the minimum fitting residual" and "restricting the model's degrees of freedom", completely preventing overfitting and underfitting.

[0089] By employing the above steps, the most suitable emissivity model can be adaptively selected pixel-by-pixel based on the radiance signal acquired by the multispectral imaging system 2. Combined with Planck's blackbody radiation law, the true temperature distribution and full-band spectral emissivity of the measured high-temperature target 1 can be finally obtained through inversion. The adaptive selection method for the multispectral radiation thermometry emissivity model in this embodiment is used for the following verification:

[0090] In the radiometric calibration and two-dimensional temperature field measurement experiment, the dataset includes baseline data acquired through spectral radiometric calibration using a calibration halogen lamp, and measured spectral radiometric data (including monochromatic radiance) of the two-dimensional temperature field of a candle flame. In the two-dimensional temperature measurement scenario of the candle flame, the grayscale image of the candle flame is shown below. Figure 3 As shown, due to the significant differences in the concentration of combustion products, distribution of soot particles, and temperature gradient between the flame center and the edge region, especially in the pixel region at the flame edge where the spectral emissivity exhibits drastic non-gray body characteristics, traditional global assumption methods are prone to causing severe model mismatch.

[0091] The set of emissivity models for candle flames includes:

[0092] Candidate emissivity model 1 (grey volume constant model, ):

[0093] .

[0094] Candidate emissivity model 2 (first-order linear model, ):

[0095] .

[0096] Candidate Emissivity Model 3 (Second-order Polynomial Model) ):

[0097] .

[0098] Candidate Emissivity Model 4 (Exponential Degradation Model) ):

[0099] .

[0100] Using the method described in this embodiment, the emissivity model for multispectral radiometric thermometry is adaptively selected, resulting in the following emissivity model distribution across the entire field: Figure 4 As shown, a pseudo-color image of the two-dimensional temperature field distribution of the candle flame reconstructed based on the overall emissivity model is presented. Figure 5As shown in the figure, at typical pixels along the flame edge, the histogram comparing the fitting residuals of the full-field grayscale model and the emissivity model adaptively selected based on the AIC information criterion in this embodiment is shown in the figure. Figure 6 As shown.

[0101] according to Figure 6 As can be seen, the method in this embodiment achieves spatial adaptive matching of the emissivity model for each pixel in a two-dimensional image by introducing the AIC information criterion. In the complex and highly nonlinear region of flame edge pixels, compared with the traditional global assumption gray body model (uniform gray body assumption across the entire field), the fitting residual of the radiometric measurement equation by this method is significantly reduced by 16.65%. At the same time, due to the complexity penalty term in the AIC criterion, the algorithm effectively suppresses the overfitting phenomenon caused by blindly fitting high-order models to low signal-to-noise ratio pixels at the edge due to noise interference, thus significantly improving the convergence and boundary continuity of the reconstructed temperature field at the flame edge.

[0102] Table 1 shows the comparison results of the fitting residuals of pixels in different regions of the candle flame under different emissivity models.

[0103] Table 1

[0104]

[0105] according to Figures 4-6 As shown in Table 1, in the two-dimensional temperature field measurement of a candle flame, the central region of the flame typically exhibits near-gray-body characteristics, while edge pixels, influenced by combustion products and air mixing, display non-gray-body characteristics. Traditional full-field gray-body models cannot account for spatial inhomogeneity, while this algorithm can adaptively select the best-fitting emissivity model for each pixel, directly reducing the fitting residual of edge pixels by 16.65%, significantly improving the accuracy of edge temperature field reconstruction. Since the radiation signal of flame edge pixels is relatively weak and has a low signal-to-noise ratio, the traditional method of rigidly using high-order polynomial fitting is prone to misdiagnosing edge noise as emissivity spectrum fluctuations (overfitting). This embodiment relies on the AIC information criterion's penalty mechanism for model degrees of freedom to automatically eliminate meaningless high-order redundant parameters, ensuring a reduction in fitting residuals while preventing abnormal fluctuations in temperature calculation results in the edge region.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. An adaptive selection method for a multispectral radiometric emissivity model, characterized in that, Includes the following steps: Step 1: Use a multispectral imaging system (2) to collect the radiance signals of the high-temperature target (1) at multiple specific wavelengths, and perform photoelectric conversion and preprocessing to obtain the true monochromatic radiance at multiple specific wavelengths; Step 2: Based on the material physical properties of the high-temperature target (1) being tested, construct a set of candidate emissivity models with different complexities and function forms; Step 3: For each candidate emissivity model in the candidate emissivity model set, establish an overdetermined system of equations based on Planck's radiation law, input the real monochromatic radiance into the overdetermined system of equations, use the nonlinear least squares algorithm to solve for the optimal estimated temperature and optimal model coefficients under the model, and calculate the sum of squared fitting residuals. Step 4: Calculate the AIC evaluation index corresponding to each candidate emissivity model based on the sum of squared residuals, and correct the AIC evaluation index. Select the optimal emissivity model based on the corrected AIC evaluation index.

2. The adaptive selection method for a multispectral radiometric emissivity model according to claim 1, characterized in that, The multispectral imaging system (2) includes a collimating focusing lens (3), a filter array (4), and a photodetector (5) connected in sequence. The data output terminal of the photodetector (5) is connected to a PC processing terminal (6), which is used to execute steps 2-4.

3. The adaptive selection method for a multispectral radiometric emissivity model according to claim 2, characterized in that, Step 1 is as follows: Step 1.1: Set the collimating and focusing lens (3) opposite to the high temperature target (1) to be measured. The thermal radiation emitted by the high temperature target (1) is focused by the collimating and focusing lens (3) onto the filter array (4) and then split by the filter array (4) to obtain narrowband monochromatic light. Step 1.2: Narrow-band monochromatic light is incident on the photodetector (5) and converted into a photocurrent signal. The photocurrent signal is then processed sequentially by transimpedance current to voltage conversion, programmable amplification, filtering and noise reduction, dark current compensation, and impedance matching to obtain the true monochromatic radiation brightness of multiple specific wavelengths. The signal is then input to the PC processing terminal (6).

4. The adaptive selection method for a multispectral radiometric emissivity model according to claim 3, characterized in that, The filter array (4) employs N narrowband filters with different center wavelengths, covering a wavelength range of or .

5. The adaptive selection method for a multispectral radiometric emissivity model according to claim 1, characterized in that, In step 2, the expression for the candidate emissivity model set is: ; in, Represents the i-th specific wavelength The corresponding unknown coefficient vector is The candidate emissivity models; where m represents the ordinal number of the candidate emissivity models and M represents the total number of candidate emissivity models; Unknown coefficient vector The expression is: ; Indicates the first The number of parameters to be determined in each candidate emissivity model Let each of the following represent the first, ..., second emissivity values ​​of the m-th candidate emissivity model: Model coefficients.

6. The adaptive selection method for a multispectral radiometric emissivity model according to claim 5, characterized in that, Step 3 specifically involves: Step 3.1: For the m-th candidate emissivity model, expand the vector of unknown coefficients to be solved into unknowns that include the true temperature. , is represented as: ;in, Represents the actual temperature; unknown quantity quantity Represented as: The number of channels N in the multispectral imaging system (2) satisfies: ; Step 3.2: For the m-th candidate emissivity model, establish an overdetermined set of equations using Planck's radiation law; Step 3.3: Input the true monochromatic radiance into the overdetermined equation system, and use the nonlinear least squares algorithm to solve for the minimum value of the m-th candidate emissivity model. To obtain the optimal estimated temperature and optimal model coefficients The solution formula is: ; in, This represents the true monochromatic radiance at the i-th specific wavelength obtained by the multispectral imaging system (2). This represents the m-th candidate emissivity model. This represents the i-th specific wavelength. This represents the theoretical radiance calculated according to Planck's blackbody radiation law; Step 3.4: Based on the optimal estimated temperature and optimal model coefficients The sum of squared residuals is calculated using the following formula: 。 7. The adaptive selection method for a multispectral radiometric emissivity model according to claim 6, characterized in that, The expression for the theoretical radiance calculated according to Planck's blackbody radiation law is as follows: ; In the formula, The first radiation constant has a value of [value missing]. , The second radiation constant has a value of .

8. The adaptive selection method for a multispectral radiometric emissivity model according to claim 6, characterized in that, Step 4 specifically involves: Step 4.1: Calculate the AIC evaluation index for the m-th candidate emissivity model. The calculation formula is: ; Step 4.2: Set the number of channels N and the unknowns. quantity proportional threshold When satisfied The modified Akaike Information Content Criterion was used to evaluate the AIC indicators. For high-precision fine-tuning, the fine-tuning formula is: ; This indicates the revised AIC evaluation metrics; Step 4.3: Select the modified minimum AIC evaluation index, the expression of which is: ; This represents the ordinal number of the candidate emissivity model corresponding to the minimum AIC evaluation metric; Step 4.4, place the first The candidate emissivity model is selected as the optimal emissivity model.

Citation Information

Patent Citations

  • A Multispectral Radiometric Thermometry Method Based on CCD Camera

    CN110017904B

  • Multispectral radiation temperature measurement inversion calculation method based on generalized inverse-neural network, computer and storage medium

    CN113776675A