A temperature measurement method and device, computer equipment and storage medium

Abstract

The application provides a temperature measurement method and device, computer equipment and a storage medium, and belongs to the field of temperature measurement. The method comprises the following steps: acquiring sample data; constructing a plurality of luminance value and temperature relationship formulas according to a plurality of spectral response logarithm ratios and a two-color temperature measurement formula; constructing a unit structure of luminance value mapping to predicted temperature; determining physical constraints according to the plurality of luminance value and temperature relationship formulas; constructing a neural network model of luminance value mapping to predicted temperature according to the unit structure and the physical constraints; training the neural network model through the sample data, determining unknown parameters, and then obtaining a temperature prediction model; acquiring a temperature field image of a target object; inputting the temperature field image of the target object into the temperature prediction model to obtain the temperature of the target object. The neural network system constructed through physical constraints reduces the dependence of the normalization process on the true temperature value, and improves the convenience of temperature measurement.

Description

Technical Field

[0001] This invention belongs to the field of temperature measurement, and specifically relates to a temperature measurement method, device, computer equipment, and storage medium. Background Technology

[0002] In existing technologies, temperature measurement via images mainly employs the two-color thermometry method. However, the RGB three-channel brightness values ​​of color images captured by CCD cameras vary due to differences in the characteristics of the CCD camera itself and the combination of aperture and shutter speed. This necessitates standardizing the two-color thermometry formula before using it to calculate temperature, so that it can adapt to the RGB three-channel brightness values ​​of the input image, thereby calculating a more accurate temperature value.

[0003] However, the standardization process essentially involves calibrating an unknown quantity within the two-color thermometry formula. This calibration process requires manually measuring a sufficient number of true temperature values, which is extremely difficult in many industrial settings. For example, measuring the temperature at a specific location within a large furnace is very challenging; temperature measurements can only be performed at a limited number of locations on the furnace surface by inserting thermocouples.

[0004] Furthermore, the RGB three-channel brightness values ​​of a color image are typically between [0, 255]. When using a CCD camera to capture flame images of a temperature field, to ensure more accurate temperature calculations, camera parameters are usually adjusted so that the three-channel brightness values ​​are not equal to 0 and 255, thus avoiding distortion. However, each time the camera parameters are adjusted, the dual-color thermometry formula needs to be re-normalized, which greatly hinders the real-time application of existing technology in different situations or with large temperature distribution ranges. Summary of the Invention

[0005] To address the inconvenience caused by the need for multiple standardization processes in image temperature measurement in existing technologies, this invention provides a temperature measurement method, apparatus, computer equipment, and storage medium.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] First, a temperature measurement method is provided, the method comprising:

[0008] Obtain the sample temperature field image and its corresponding sample data of the real temperature;

[0009] The three primary color brightness values ​​of the sample temperature field image are extracted. The three primary color brightness values ​​and multiple unknown parameters are used to characterize the logarithmic ratios of various spectral responses. Then, based on the multiple logarithmic ratios of spectral responses and the formula of the two-color thermometry method, multiple relationships between brightness values ​​and temperature are constructed. The logarithmic ratio of the spectral response is the logarithmic ratio of the spectral response intensity of one primary color channel to that of another primary color channel.

[0010] Construct an equation relating the multiple brightness values ​​to temperature, and use the equation as a physical constraint.

[0011] A neural network model mapping brightness values ​​to predicted temperature is constructed using the unknown parameters as network weights; the neural network model is trained using the sample temperature field image and physical constraints to determine the unknown parameters, and then the unknown parameters are substituted into the relationship between brightness values ​​and temperature to obtain a temperature prediction model;

[0012] Obtain the temperature field image of the target object;

[0013] The temperature field image of the target object is input into the temperature prediction model to obtain the temperature of the target object.

[0014] Optionally, the multiple spectral response log ratios include the red-blue spectral response log ratio and the red-green spectral response log ratio;

[0015] The logarithmic ratio of the red-blue spectral response is:

[0016] K rb =ln(K) r / K b );

[0017] The logarithmic ratio of the red-green spectral response is:

[0018] K rg =ln(K) r / K g );

[0019] Among them, K r K represents the spectral response coefficient of the red channel. g K represents the spectral response coefficient of the green channel. b This represents the spectral response coefficient for the blue channel.

[0020] Optionally, the step of constructing multiple relationships between brightness values ​​and temperature based on various spectral response logarithmic ratios and the two-color thermometry formula includes:

[0021] The formula for the two-color temperature measurement method is:

[0022]

[0023] Where C2 represents the second radiation constant; λr , λ g , λ b These represent the wavelengths of the three primary colors in the temperature field image; T rg and T rb The results are the temperature measurement results of the red-green bicolor thermometry method and the red-blue bicolor thermometry method, respectively; R e G e B e These represent the brightness values ​​of the R, G, and B channels, respectively; ε(λ,T) is a dimensionless parameter representing the spectral emissivity.

[0024] By characterizing the logarithmic ratios of the red-blue and red-green spectral responses using the luminance values ​​of the three primary colors and several unknown parameters, the relationship between these luminance values ​​and temperature is determined as follows:

[0025]

[0026] Among them, a rg’ b rg’ c rg’ a rb’ b rb’ and c rb’ The unknown parameter is...

[0027] Optionally, constructing a neural network model that maps brightness values ​​to predicted temperature using the unknown parameters as network weights includes:

[0028] Corresponding to the relationship between brightness value and temperature, a neural network model is constructed to map brightness value to predicted temperature. The single-channel expression of the neural network model is as follows:

[0029]

[0030] Where g1, g2, and g3 all represent activation functions, a rg b rg c rg a rb b rb and c rb C represents the network weights to be solved corresponding to the unknown parameters. rg and C rb T' represents a constant. rg and T' rb These represent the output of a single channel:

[0031] Corresponding to the number of unknown parameters, a multi-channel neural network model is constructed based on the single-channel expression.

[0032] Optionally, the neural network model is trained using the sample temperature field image and physical constraints to determine the unknown parameters. These unknown parameters are then substituted into the relationship between brightness and temperature to obtain a temperature prediction model, including:

[0033] The sample temperature field image is input into the neural network model to obtain the predicted temperature;

[0034] Based on the physical constraints, the physical loss between multiple predicted temperatures is determined, and the neural network model is trained with the goal of minimizing the physical loss to determine the values ​​of the unknown parameters;

[0035] The unknown parameter values ​​are substituted into the relationship between brightness and temperature, and weighting coefficients are determined based on multiple sample data to obtain a temperature prediction model.

[0036] Optionally, determining the weighting coefficients based on multiple sample data to obtain the temperature prediction model includes:

[0037] Substitute multiple sample data into the relationship between brightness value and temperature and solve them simultaneously to determine the weighting coefficient of the relationship between brightness value and temperature.

[0038] Then, the weighting coefficients are substituted into the relationship between brightness value and temperature to determine the temperature prediction model.

[0039] Secondly, a temperature measuring device is also provided, the device comprising:

[0040] The acquisition module is used to acquire the sample temperature field image and the sample data of the corresponding real temperature;

[0041] A construction module is used to extract the three primary color brightness values ​​from a sample temperature field image. These brightness values, along with multiple unknown parameters, characterize various spectral response logarithmic ratios. Based on these logarithmic ratios and the two-color thermometry formula, multiple relationships between brightness values ​​and temperature are constructed. The spectral response logarithmic ratio is the logarithmic ratio of the spectral response intensity of one primary color channel to that of another. Equations are constructed between these multiple brightness value-temperature relationships, serving as physical constraints. A neural network model mapping brightness values ​​to predicted temperature is built using the unknown parameters as network weights. This neural network model is trained using the sample temperature field image and the physical constraints to determine the unknown parameters. These unknown parameters are then substituted into the brightness value-temperature relationships to obtain a temperature prediction model.

[0042] The acquisition module is also used to acquire a temperature field image of the target object;

[0043] The prediction module is used to input the temperature field image of the target object into the temperature prediction model to obtain the temperature of the target object.

[0044] In addition, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the above-described temperature measurement method.

[0045] In addition, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described temperature measurement method.

[0046] The temperature measurement method provided by this invention has the following beneficial effects:

[0047] First, a correspondence between brightness values ​​and temperature is established by constructing multiple unknown parameters, reducing the number of parameters that need to be calibrated in temperature measurement in existing technologies. Second, a neural network system is constructed through physical constraints, and the unknown parameters are determined through the neural network system. The loss used in this process is all physical constraint loss and does not involve regression loss, thus greatly reducing the dependence of the temperature prediction model on sample data. This not only helps to improve the accuracy of temperature measurement, but also reduces the dependence of normalization processing on calibration parameters, improves the convenience of normalization processing, eliminates the obstacles to the real-time application of image temperature measurement in different occasions or in situations with large temperature distribution ranges, and improves the convenience of temperature measurement. Attached Figure Description

[0048] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic flowchart of a temperature measurement method provided by the present invention according to an exemplary embodiment.

[0050] Figure 2 This is a schematic diagram of a unit structure for mapping brightness values ​​to predicted temperature according to an exemplary embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of a physical constraint-based neural network model provided by the present invention according to an exemplary embodiment.

[0052] Figure 4 This is a block diagram of a temperature measuring device provided by the present invention according to an exemplary embodiment. Detailed Implementation

[0053] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0054] Radiation thermometry methods mainly include three types: luminance thermometry, total radiation thermometry, and two-color thermometry. With the rapid development of CCD cameras and computer technology, an image-based temperature measurement method has emerged. This method acquires images of the combustion flame in the temperature field and uses the radiation thermometry mechanism to measure the temperature distribution of the temperature field.

[0055] Image-based radiation thermometry has many advantages: (1) It has low thermal inertia and can achieve transient temperature measurement; (2) It is a non-contact temperature measurement method, which can avoid direct contact between the equipment and the harsh environment and protect the equipment from corrosion; (3) It can measure the temperature distribution of a large temperature field in real time and realize the visualization of the measured temperature field.

[0056] By adding a monochrome filter to a CCD camera, relatively accurate RGB three-channel brightness values ​​of a color image can be quickly collected. This characteristic can be utilized in dual-color thermometry, combining the principles of dual-color thermometry with the RGB three-channel brightness values ​​of a color image to measure the temperature of the area captured by the CCD camera, thereby visualizing the temperature distribution.

[0057] This invention adds physical constraints to the two-color temperature measurement method, improving the convenience of image temperature measurement.

[0058] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0059] First, this invention provides a temperature measurement method, specifically as follows: Figure 1 As shown, it includes the following steps:

[0060] S101. Obtain the sample temperature field image and the corresponding sample data of the real temperature.

[0061] Specifically, a CCD camera can be used to capture a temperature field image, the captured image can be denoised, and the actual temperature corresponding to the temperature field image can be measured.

[0062] S102. Construct multiple relationships between brightness values ​​and temperature based on various spectral response logarithmic ratios and the formula for two-color thermometry.

[0063] Specifically, the brightness values ​​of the three primary colors in the sample temperature field image can be extracted. The brightness values ​​of the three primary colors and multiple unknown parameters are used to characterize the logarithmic ratios of various spectral responses. Then, based on the logarithmic ratios of various spectral responses and the formula of the two-color thermometry method, multiple relationships between brightness values ​​and temperature can be constructed. The logarithmic ratio of the spectral response is the logarithmic ratio of the spectral response intensity of one primary color channel to that of another primary color channel.

[0064] In one embodiment, the multiple spectral response log ratios include the red-blue spectral response log ratio and the red-green spectral response log ratio.

[0065] The logarithmic ratio of the red-blue spectral response is determined using the following formula:

[0066] K rb =ln(K) r / K b );

[0067] The logarithmic ratio of the red-green spectral response is determined using the following formula:

[0068] K rg =ln(K) r / K g );

[0069] Among them, K r K represents the spectral response coefficient of the red channel. g K represents the spectral response coefficient of the green channel. b The blue channel spectral response coefficient is affected by the CCD spectral response characteristics and can be regarded as an error correction coefficient. It is a nonlinear function of the temperature value and is also a quantity that needs to be calibrated in the two-color temperature measurement formula.

[0070] In another embodiment, the formula for measuring the temperature of the two-color hair is determined as follows:

[0071]

[0072] Where C2 represents the second radiation constant, C2 = 1.4338 × 10⁻² m·K; λ r =700.0nm, λ g =546.1nm, λ b =435.8nm represent the RGB wavelengths of a color CCD; T rg T rb and T gb The two-color method yields temperature measurements of the radiator. When measuring the temperature of the same object, ideally, T... rg =T rb =T gbThe unit is K; Re, Ge, and Be represent the pixel values ​​of the R, G, and B channels, respectively, also called luminance values; ε(λ,T) is a dimensionless parameter representing spectral emissivity.

[0073] GB two-color thermometry suffers from distortion due to photogenerated charge saturation in the R primary color potential well. Therefore, this invention selects the RG and RB two-color thermometry method for research. Since flames are mainly composed of solid particles with continuous radiation spectra, when measuring the temperature distribution of a combustion flame, it can be approximated as an ash body, i.e., ln(ε(λ)). r ,T) / ε(λ g ,T))=0,ln(ε(λ r ,T) / ε(λ b ,T))=0. Because K rg and K rb Since each of these values ​​has a quadratic functional relationship with the natural logarithm of its corresponding luminance ratio, the logarithmic ratios of the red-blue and red-green spectral responses can be characterized by the luminance values ​​of the three primary colors and several unknown parameters, respectively. The relationship between these multiple luminance values ​​and temperature can then be determined as follows:

[0074]

[0075] Among them, a rg’ b rg’ c rg’ a rb’ b rb’ and c rb’ This is the unknown parameter.

[0076] S103. Construct an equation relating the multiple brightness values ​​to temperature, and use this equation as a physical constraint.

[0077] In this step, considering that the existing technology for measuring temperature through images requires multiple normalization processes and training of the model using a large amount of sample data, resulting in very cumbersome preprocessing when applied across different scenarios, this invention proposes the concept of physical constraints based on the above problems. It constructs equations between multiple brightness values ​​and temperature relationships, and mutually constrains the temperatures predicted by the brightness values ​​of different primary colors, thereby constructing a self-constrained model that does not require actual temperature measurement.

[0078] Specifically, considering that the pixel values ​​of the three color channels of the same pixel can only take integer values ​​in the range [0, 255], there will inevitably be slight errors. The above relationship between brightness and temperature can be transformed into the following formula to improve accuracy:

[0079]

[0080] Where, Δb rgand Δb rb Let η be the error quantity, which is very small, approaching 0, and η is... rg and η rb The error coefficient is approximately equal to 1.

[0081] Next, temperature equations for the red-blue bicolor temperature measurement method and the red-green bicolor temperature measurement method are constructed; the mapping relationship of the unit structure is substituted into the temperature equation to obtain the physical constraint.

[0082] Specifically, multiple equations relating brightness values ​​to temperature are constructed, as follows:

[0083]

[0084] Due to Δb rg and Δb rb It is a very small quantity that approaches 0, and η rg and η rb Since it is approximately equal to 1, the above equation can be simplified to:

[0085]

[0086] Substituting the element structure, we obtain the physical constraint T. r ′ g ≈T r ′ b The physical constraint is a constraint equation constructed using different two-color thermometry formulas.

[0087] S104. Construct a neural network model that maps brightness values ​​to predicted temperatures using the unknown parameters as network weights.

[0088] Specifically, a neural network model that maps brightness values ​​to predicted temperatures can be constructed based on the relationship between brightness values ​​and temperature.

[0089] First, a basic unit structure can be constructed that maps brightness values ​​to predicted temperatures. This basic unit structure is as follows: Figure 2 As shown.

[0090] Using this basic unit structure as a single channel of the neural network model, corresponding to the number of unknown parameters, a multi-channel neural network model is constructed based on the single-channel expression.

[0091] The single-channel expression is as follows:

[0092]

[0093] Where g1, g2, and g3 all represent activation functions, g1(·) = ln(·), g2(·) = (·). 2 , a rg b rgc rg a rb b rb and c rb Represents the network weights to be solved, respectively, and a rg’ b rg’ c rg’ a rb’ b rb’ and c rb’ One-to-one correspondence, C rg and C rb Represents a constant. T' rg and T' rb These represent the outputs of the unit structure.

[0094] For example, when the RGB three-channel brightness values ​​of a pixel are known, a single pixel can be used to construct one of the above single-channel expressions. Since this single-channel expression has six unknowns, theoretically, six pixels at different temperatures are needed to construct six different formulas to solve for these six unknowns. Each unit structure can input one pixel at a time, so using six parallel unit structures simultaneously can construct a physically constrained neural network model, as shown below. Figure 3 As shown.

[0095] Since the red-blue and red-green bicolor thermometry methods theoretically produce equal temperature values, this invention constructs a sub-model for each method based on its mechanistic formula. Theoretically, both sub-models, when inputting brightness values, will output temperature values, and these values ​​should theoretically be equal. However, each sub-model requires training with three unknowns. Before proper training, the output temperature values ​​of the two sub-models are not equal, resulting in a physical constraint loss. This physical constraint is used as the loss function to construct a six-channel neural network model that maps brightness values ​​to predicted temperatures.

[0096] S105. Train the neural network model to obtain the temperature prediction model.

[0097] Specifically, the neural network model is trained using the sample temperature field image and physical constraints to determine the unknown parameter. Then, the unknown parameter is substituted into the relationship between brightness value and temperature to obtain the temperature prediction model.

[0098] In this step, the sample temperature field image from the sample data can be input into the neural network model to obtain the predicted temperature; the physical loss between multiple predicted temperatures is determined according to the physical constraint, and the neural network model is trained with the goal of minimizing the physical loss to determine the value of the unknown parameter; the value of the unknown parameter is substituted into the relationship between the brightness value and the temperature, and the weight coefficients are determined according to multiple sample data to obtain the temperature prediction model.

[0099] In this model, the input is a known quantity. At the same time, six pixels with different temperatures are input into the neural network model. After multiple time intervals, the model will gradually adjust the network weights due to the loss of physical constraints, that is, adjust the six unknown parameters, so that the physical loss of the model gradually approaches 0, and finally obtain the relatively accurate solution of the six unknown parameters.

[0100] Since the loss used in training the model is purely physical constraint loss and does not involve regression loss, this method can significantly reduce the network model's dependence on the true value. Furthermore, because solving for these six parameters only satisfies the equation between brightness and temperature, simply solving for these six parameters is insufficient for temperature measurement. That is, multiplying both sides of the equation by the unknown coefficients still maintains the equation, as shown below:

[0101]

[0102] Where β is an unknown coefficient, the above equation can be broken down to obtain:

[0103]

[0104] Where, β rg and β rg α represents the weighting coefficients of the different luminance-temperature relationships constructed based on the red-green bicolor thermometry method and the red-blue bicolor thermometry method, respectively. rg and α rb These represent the error values ​​in the relationship between different brightness values ​​and temperature.

[0105] Since the physical constraints are constructed through equations relating different brightness values ​​to temperature, the resulting physical loss is not the loss between the predicted result and the true value, but rather the loss between the predicted results of the mapping relationship between different brightness values ​​and temperature. This causes the network weights trained by the neural network model to be different from the true network weights, but rather to be related values ​​that have a multiple or other relationship with the true network weights.

[0106] To further determine the true network weights, i.e. the true values ​​of the unknown parameters, multiple sample data can be substituted into the relationship between brightness and temperature and combined to determine the weight coefficients of the relationship between brightness and temperature; then, the weight coefficients can be substituted into the relationship between brightness and temperature to determine the temperature prediction model.

[0107] For example, after substituting the value of the unknown parameter into the temperature equation, multiple actual measured sample data are then substituted into the temperature equation and combined to determine the weighting coefficient and error of the relationship between the brightness value and temperature; the weighting coefficient and error are then substituted into the relationship between the brightness value and temperature to determine the temperature prediction model.

[0108] Therefore, by substituting multiple sample data into the relationship between brightness value and temperature, the weighting coefficient can be obtained through simultaneous calculation.

[0109] Finally, the calculated weighting coefficients are substituted into the relationship between brightness value and temperature to obtain the temperature prediction model.

[0110] S106. Obtain the temperature of the target object through a temperature prediction model.

[0111] Specifically, a temperature field image of the target object can be obtained; the temperature field image of the target object can be input into the temperature prediction model to obtain the temperature of the target object.

[0112] In addition, to further improve the accuracy of this temperature prediction model, T can also be taken as... rg and T rb The average value is taken as the final measurement result.

[0113] The above method first establishes a correspondence between brightness values ​​and temperature through multiple unknown parameters, reducing the number of parameters that need to be calibrated in temperature measurement in existing technologies. Secondly, a neural network system is constructed through physical constraints, and the unknown parameters are determined through the neural network system. The loss used in this process is all physical constraint loss and does not involve regression loss, thus greatly reducing the dependence of the temperature prediction model on sample data. This not only helps to improve the accuracy of temperature measurement, but also reduces the dependence of normalization processing on calibration parameters, improves the convenience of normalization processing, eliminates the obstacles to the real-time application of image temperature measurement in different occasions or in situations with large temperature distribution ranges, and improves the convenience of temperature measurement.

[0114] Secondly, the present invention also provides a temperature measuring device, such as... Figure 2 As shown, it includes:

[0115] The acquisition module 201 is used to acquire the sample temperature field image and the sample data of the corresponding real temperature.

[0116] Module 202 is used to extract the three primary color brightness values ​​of the sample temperature field image. The three primary color brightness values ​​and multiple unknown parameters are used to characterize various spectral response logarithmic ratios. Then, based on these logarithmic ratios and the two-color thermometry formula, multiple relationships between brightness values ​​and temperature are constructed. The spectral response logarithmic ratio is the logarithmic ratio of the spectral response intensity of one primary color channel to another. A unit structure mapping brightness values ​​to predicted temperature is constructed. Based on these multiple brightness value-temperature relationships, the mapping equation of this unit structure is determined, and physical constraints are determined based on this mapping equation. These physical constraints are constraint equations constructed using different two-color thermometry formulas. A neural network model mapping brightness values ​​to predicted temperature is constructed based on this unit structure and physical constraints. The neural network model is trained using the sample data to determine the unknown parameters, thereby obtaining a temperature prediction model.

[0117] The acquisition module 201 is also used to acquire temperature field images of the target object.

[0118] The prediction module 203 is used to input the temperature field image of the target object into the temperature prediction model to obtain the temperature of the target object.

[0119] Using the aforementioned device, the correspondence between brightness values ​​and temperature is first established through multiple unknown parameters, reducing the number of parameters that need to be calibrated in temperature measurement in existing technologies. Secondly, a neural network system is constructed through physical constraints, and the unknown parameters are determined through the neural network system. The loss used in this process is all physical constraint loss and does not involve regression loss, thus greatly reducing the dependence of the temperature prediction model on sample data. This not only helps to improve the accuracy of temperature measurement, but also reduces the dependence of normalization processing on calibration parameters, improves the convenience of normalization processing, eliminates the obstacles to the real-time application of image temperature measurement in different occasions or in situations with large temperature distribution ranges, and improves the convenience of temperature measurement.

[0120] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The steps of the provided temperature measurement method.

[0121] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for various operations. The processor reads the corresponding computer program from the non-volatile memory into the memory and then executes it to achieve the above-mentioned functions. Figure 1 The steps of the provided temperature measurement method.

[0122] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0123] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] It should be noted that the above-described specific embodiments enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A temperature measurement method, characterized in that, The method includes: Obtain the sample temperature field image and its corresponding sample data of the real temperature; The three primary color brightness values ​​of the sample temperature field image are extracted. The three primary color brightness values ​​and multiple unknown parameters are used to characterize the logarithmic ratios of various spectral responses. Then, based on the multiple logarithmic ratios of spectral responses and the formula of the two-color thermometry method, multiple relationships between brightness values ​​and temperature are constructed. The logarithmic ratio of the spectral response is the logarithmic ratio of the spectral response intensity of one primary color channel to that of another primary color channel. Construct an equation relating the multiple brightness values ​​to temperature, and use the equation as a physical constraint. A neural network model mapping brightness values ​​to predicted temperature is constructed using the unknown parameters as network weights; The neural network model is trained using the sample temperature field image and physical constraints to determine the unknown parameters. Then, the unknown parameters are substituted into the relationship between brightness value and temperature to obtain the temperature prediction model. Obtain the temperature field image of the target object; The temperature field image of the target object is input into the temperature prediction model to obtain the temperature of the target object; The various spectral response log ratios include the red-blue spectral response log ratio and the red-green spectral response log ratio; The logarithmic ratio of the red-blue spectral response is: K rb =ln(K r / K b ); The logarithmic ratio of the red-green spectral response is: K rg =ln(K r / K g ); Among them, K r K represents the spectral response coefficient of the red channel. g K represents the spectral response coefficient of the green channel. b The blue channel spectral response coefficient; The formulas for constructing multiple relationships between brightness values ​​and temperature based on various spectral response logarithmic ratios and dual-color thermometry include: The formula for the two-color temperature measurement method is: ； ； in, C 2 represents the second radiation constant; λ r , λ g , λ b These represent the wavelengths of the three primary colors in the temperature field image; T rg and T rb The results are the temperature measurement results of the red-green bicolor thermometry method and the red-blue bicolor thermometry method, respectively. R e 、G e 、B e These represent the brightness values ​​of the R, G, and B channels, respectively. ε(λ,T) It is a dimensionless parameter representing spectral emissivity; By characterizing the logarithmic ratios of the red-blue and red-green spectral responses using the luminance values ​​of the three primary colors and several unknown parameters, the relationship between these luminance values ​​and temperature is determined as follows: ； ； in, a rg’ , b rg’ , c rg’ , a rb’ , b rb’ and c rb’ For the unknown parameter; The neural network model that uses the unknown parameters as network weights to map brightness values ​​to predicted temperatures includes: Corresponding to the relationship between brightness value and temperature, a neural network model is constructed to map brightness value to predicted temperature. The single-channel expression of the neural network model is as follows: ； Where g1, g2, and g3 all represent activation functions. a rg , b rg , c rg , a rb , b rb and c rb This represents the network weights to be solved corresponding to the unknown parameters. C rg and C rb Represents a constant. T’ rg and T’ rb These represent the output of a single channel; Corresponding to the number of unknown parameters, a multi-channel neural network model is constructed based on the single-channel expression.

2. The temperature measurement method according to claim 1, characterized in that, The neural network model is trained using the sample temperature field image and physical constraints to determine the unknown parameters. These unknown parameters are then substituted into the relationship between brightness and temperature to obtain a temperature prediction model, including: The sample temperature field image is input into the neural network model to obtain the predicted temperature; Based on the physical constraints, the physical loss between multiple predicted temperatures is determined, and the neural network model is trained with the goal of minimizing the physical loss to determine the values ​​of the unknown parameters; The unknown parameter values ​​are substituted into the relationship between brightness and temperature, and weighting coefficients are determined based on multiple sample data to obtain a temperature prediction model.

3. The temperature measurement method according to claim 2, characterized in that, The step of determining weighting coefficients based on multiple sample data to obtain the temperature prediction model includes: Substitute multiple sample data into the relationship between brightness value and temperature and solve them simultaneously to determine the weighting coefficient of the relationship between brightness value and temperature. Then, the weighting coefficients are substituted into the relationship between brightness value and temperature to determine the temperature prediction model.

4. A temperature measuring device, characterized in that, The apparatus for using a temperature measurement method as described in any one of claims 1-3 comprises: The acquisition module is used to acquire the sample temperature field image and the sample data of the corresponding real temperature; A construction module is used to extract the three primary color brightness values ​​from a sample temperature field image. These brightness values, along with multiple unknown parameters, characterize various spectral response logarithmic ratios. Based on these logarithmic ratios and the two-color thermometry formula, multiple relationships between brightness values ​​and temperature are constructed. The spectral response logarithmic ratio is the logarithmic ratio of the spectral response intensity of one primary color channel to that of another. Equations are constructed between these multiple brightness value-temperature relationships, serving as physical constraints. A neural network model mapping brightness values ​​to predicted temperature is built using the unknown parameters as network weights. This neural network model is trained using the sample temperature field image and the physical constraints to determine the unknown parameters. These unknown parameters are then substituted into the brightness value-temperature relationships to obtain a temperature prediction model. The acquisition module is also used to acquire a temperature field image of the target object; The prediction module is used to input the temperature field image of the target object into the temperature prediction model to obtain the temperature of the target object.

5. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 3.

6. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 3.

Images