Infrared thermometry compensation and correction method, system and storage medium for integrating environmental meteorological data
By constructing an infrared quantization value-irradiation response mechanism model and an asymmetric focal plane temperature hysteresis response model, and combining a temperature measurement deviation correction and resolution enhancement network, the problem of temperature measurement accuracy and resolution under the influence of multiple factors in infrared thermometry was solved, and high-precision and high-definition temperature image reconstruction was achieved.
Patent Information
- Application Number
- CN202611510862.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing infrared thermometry methods struggle to maintain accuracy under the combined influence of multiple factors. They fail to consider the spatial heterogeneity of the focal plane temperature field and thermal inertial lag, and the temperature image resolution is insufficient, lacking physical constraints.
An infrared quantization value-irradiance response mechanism model integrating environmental meteorological data is constructed. Combined with an asymmetric focal plane temperature hysteresis response model, a temperature measurement deviation correction network and a resolution enhancement network are used to introduce radiation residuals and energy conservation regularization terms to output high-definition temperature distribution images.
It effectively compensates for temperature measurement deviations caused by environmental and meteorological factors, solves the problem of thermal inertia hysteresis, improves temperature measurement accuracy and generalization ability, and achieves accurate reconstruction of high-definition temperature distribution images.
Smart Images

Figure CN123111485A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of infrared thermometry technology, specifically relating to an infrared thermometry compensation and correction method, system, and storage medium that integrates environmental meteorological data. Background Technology
[0002] Infrared thermal imagers are widely used in industrial monitoring, equipment fault diagnosis, medical screening, and drone inspection due to their advantages such as non-contact operation, fast response, and the ability to perform area array imaging. However, uncooled infrared thermal imagers are significantly affected by environmental and meteorological conditions in actual temperature measurements: changes in ambient temperature cause focal plane temperature drift, wind speed and solar irradiance lead to uneven spatial distribution of the focal plane temperature field, and atmospheric attenuation changes the radiative transfer characteristics with variations in humidity and visibility. The coupled effect of these factors makes it difficult for traditional calibration models to maintain temperature measurement accuracy.
[0003] Existing compensation methods mainly fall into two categories: one is based on nonlinear fitting to establish a functional relationship between error and environmental variables, but it mostly models a single error source and is difficult to integrate the coupled effects of multiple factors; the other is based on neural networks to directly fit the mapping relationship between gray values and temperature, but it does not fully consider the spatial heterogeneity and thermal inertia hysteresis effect of the focal surface temperature field, and lacks physical constraints, resulting in limited generalization ability. In addition, existing methods do not jointly model environmental meteorological data with the focal surface temperature hysteresis response, making it difficult to adapt to conditions of rapid changes in environmental temperature; the process of improving temperature image resolution also lacks energy consistency constraints, which easily leads to distortion of temperature distribution.
[0004] Therefore, there is an urgent need for an infrared thermometry compensation and correction method that integrates environmental meteorological data, takes into account both physical mechanisms and data-driven approaches, and simultaneously improves temperature measurement accuracy and enhances temperature image resolution. Summary of the Invention
[0005] To address the problems of existing infrared temperature measurement compensation methods, such as difficulty in integrating the coupled effects of multiple factors, failure to consider the spatial heterogeneity and thermal inertia lag of the focal plane temperature field, lack of physical constraints, and insufficient temperature image resolution, this invention provides an infrared temperature measurement compensation and correction method, system, and storage medium that integrates environmental meteorological data.
[0006] In a first aspect, embodiments of this application provide an infrared thermometry compensation and correction method that integrates environmental meteorological data, the method comprising:
[0007] S1. Construct an infrared quantization value-irradiance response mechanism model that incorporates spatial heterogeneity of the focal surface temperature field and corrections by meteorological elements, and output the initial temperature estimation feature vector.
[0008] S2. Simultaneously acquire the focal surface temperature, temperature change rate and meteorological elements, construct an asymmetric focal surface temperature hysteresis response mechanism model, and output the true working temperature of the focal surface.
[0009] S3. Input the initial temperature estimation feature vector and the actual working temperature of the focal plane into the temperature measurement deviation correction network. After dimensionality reduction, temporal fusion, downsampling and channel enhancement, and the introduction of radiation residual and energy conservation regularization terms, the target temperature value is output.
[0010] S4. Input the target temperature value into the resolution enhancement network, extract high-frequency components and rearrange pixels to output a high-definition temperature distribution image;
[0011] S5. Identify the coefficients of the mechanism model described in S1 and S2 and the weights of the network described in S3 and S4 in stages, and output the optimal parameter set.
[0012] S6. Load the optimal parameter set, acquire infrared quantization values, focal surface temperature, temperature change rate and meteorological elements in real time, and output the corrected temperature value and high-definition temperature distribution image after processing by S1 to S4 in sequence.
[0013] Secondly, embodiments of this application provide an infrared temperature measurement compensation and correction system that integrates environmental meteorological data, applied to the method described in the first aspect, the system comprising:
[0014] The mechanism modeling module is used to construct an infrared quantized value-irradiance response mechanism model that incorporates spatial heterogeneity of the focal surface temperature field and corrections by meteorological elements, and outputs an initial temperature estimation feature vector.
[0015] The hysteresis response module is used to synchronously acquire the focal surface temperature, temperature change rate and meteorological elements, construct an asymmetric focal surface temperature hysteresis response mechanism model, and output the actual working temperature of the focal surface.
[0016] The temperature measurement deviation correction module is used to input the initial temperature estimation feature vector and the actual working temperature of the focal surface into the temperature measurement deviation correction network and output the target temperature value.
[0017] The resolution enhancement module is used to input the target temperature value into the resolution enhancement network, and after high-frequency component extraction and pixel rearrangement, output a high-definition temperature distribution image.
[0018] The parameter identification module is used to identify the coefficients of the mechanism model in the mechanism modeling module and the hysteresis response module, as well as the weights of the network in the temperature measurement deviation correction module and the resolution enhancement module in stages, and output the optimal parameter set.
[0019] The real-time correction module is used to load the optimal parameter set and acquire infrared quantization values, focal plane temperature, temperature change rate and meteorological elements in real time. After being processed by the mechanism modeling module, hysteresis response module, temperature measurement deviation correction module and resolution enhancement module, the corrected temperature value and high-definition temperature distribution image are output.
[0020] Thirdly, embodiments of this application provide a computer-readable storage medium storing a program that instructs a device to perform the infrared temperature measurement compensation and correction method for fusing environmental meteorological data as described in the first aspect.
[0021] Compared with existing technologies, the present invention has the following advantages: by introducing spatial heterogeneity of focal surface temperature field and meteorological element correction, it effectively compensates for temperature measurement deviation caused by environmental meteorological factors; by using an asymmetric focal surface temperature hysteresis response mechanism model, it solves the problem of thermal inertia hysteresis; by integrating physical constraints into the temperature measurement deviation correction network, it improves the temperature measurement accuracy and generalization ability under complex working conditions; and by using a resolution enhancement network and energy consistency constraints, it achieves accurate reconstruction of high-definition temperature distribution images. Attached Figure Description
[0022] Figure 1 This is a schematic flowchart of an infrared thermometry compensation and correction method for fusing environmental meteorological data, provided in an embodiment of this application.
[0023] Figure 2 This is a diagram illustrating the architecture of an infrared temperature measurement compensation and correction system that integrates environmental meteorological data, provided in one embodiment of this application.
[0024] Figure 3 A schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.
[0026] It should be noted that in the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.
[0027] Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0028] Example 1
[0029] Figure 1 This is a schematic flowchart illustrating an infrared thermometry compensation and correction method for fusing environmental meteorological data, provided as an embodiment of this application. Figure 1 As shown, the method includes:
[0030] S1. Construct an infrared quantization-irradiance response mechanism model that incorporates spatial heterogeneity of the focal plane temperature field and corrections from meteorological factors, and output an initial temperature estimation feature vector. Based on the physical laws of infrared radiation transmission, the detector focal plane is partitioned according to differences in thermal inertia, and the attenuation effect of environmental meteorological factors on radiation propagation is introduced to establish a physical mapping relationship from infrared quantization values to temperature estimates, providing physically meaningful initial temperature estimation features for subsequent processing.
[0031] Specifically, in this embodiment, S1 includes:
[0032] S1.1 Divide the detector focal plane into multiple thermal inertia quantum regions, and establish an infrared quantization value-irradiance response equation for each sub-region. The thermal inertia quantum region refers to a sub-region with similar thermal response characteristics, divided from different areas of the focal plane based on differences in thermal inertia caused by variations in the support structure, packaging method, and surrounding electronic components. An independent response equation is established for each sub-region between the infrared quantization value (i.e., the raw digital quantity output by the detector) and the incident irradiance to reflect the unique radiation-quantization conversion relationship of that sub-region.
[0033] S1.2. An atmospheric attenuation correction term is introduced into the response equation for each sub-region. This atmospheric attenuation correction term is calculated in real time based on ambient temperature, humidity, wind speed, and solar irradiance. The atmospheric attenuation correction term characterizes the loss of infrared radiation energy in the atmospheric path between the target and the thermal imager due to factors such as water vapor absorption, aerosol scattering, and turbulent disturbances. Ambient temperature affects atmospheric transmittance, humidity determines the intensity of water vapor absorption, wind speed changes turbulence intensity and aerosol distribution, and solar irradiance indirectly affects refraction and scattering characteristics by heating the atmospheric path. These meteorological factors collectively determine the real-time value of the correction term.
[0034] Specifically, the atmospheric attenuation correction term is calculated as follows:
[0035] ,
[0036] in, Atmospheric transmittance, dimensionless, with a value of 0~1; The distance between the target and the thermal imager, in meters (m). For measuring angles, the unit is rad; The water vapor absorption coefficient is a function of ambient temperature, expressed in meters (m). -1 ; Here is the aerosol scattering coefficient, in m. -1 ; The wind speed correction index ranges from 0.5 to 1.0. The ambient temperature. This refers to wind speed. This refers to relative humidity.
[0037] Obtained by looking up a table, using ambient temperature The value is obtained by interpolation from a preset water vapor absorption coefficient table, which serves as an index. and The data was obtained through experimental calibration. The calibration method was as follows: at a fixed distance and angle, the wind speed was varied, the change in atmospheric transmittance was measured, and the result was obtained through fitting. and This formula allows the atmospheric attenuation correction term to simultaneously consider ambient temperature, humidity, wind speed, and measurement geometry, unlike existing correction methods that only consider a single factor.
[0038] S1.3. Concatenate the outputs of the response equations for each sub-region according to their spatial positions on the focal plane to obtain the initial temperature estimation feature vector. Concatenation according to spatial positions on the focal plane means arranging the temperature estimates output by the response equations of each thermal inertia quantum region into a vector form corresponding to the pixel distribution on the focal plane, based on the row and column coordinates of each region on the focal plane. This feature vector preserves the differences in the physical responses of each sub-region and constitutes the input feature format required for subsequent network processing.
[0039] S2. Simultaneously acquire focal surface temperature, temperature change rate, and meteorological elements to construct an asymmetric focal surface temperature hysteresis response mechanism model and output the true working temperature of the focal surface. Addressing the issue of temperature sensor response lagging behind the actual temperature change of the focal surface, this model utilizes the asymmetric influence of the focal surface temperature change rate and environmental meteorological conditions on the heating and cooling process to calculate the actual working temperature of the focal surface, providing a more accurate temperature reference for subsequent networks.
[0040] Specifically, in this embodiment, S2 includes:
[0041] S2.1. Simultaneously acquire focal plane temperature, focal plane temperature change rate, ambient wind speed, and solar irradiance. The focal plane temperature is read in real time by a temperature sensing element integrated on the detector's focal plane; the focal plane temperature change rate is obtained by differential calculation of the focal plane temperature at adjacent time points; ambient wind speed and solar irradiance are synchronously acquired by an external meteorological sensing unit. Synchronous acquisition means that all the above parameters are sampled at the same time reference to ensure that the input quantities are aligned in time during subsequent hysteresis response modeling.
[0042] S2.2 Using the rate of change of focal surface temperature and meteorological forcing as inputs, an asymmetric hysteresis response equation for focal surface temperature is constructed, where the heating and cooling responses are asymmetrical. The meteorological forcing term refers to the external thermal forcing effect exerted by ambient wind speed and solar irradiance on the heat exchange process of the focal surface; wind speed enhances convective heat dissipation, while solar irradiance brings additional heat flow input. The asymmetric hysteresis response equation considers the inconsistency in time constants caused by differences in heat capacity, thermal conductivity, and convective heat transfer coefficients during the heating and cooling processes of the focal surface. Separate response equations are established for the heating and cooling branches to more accurately describe the hysteretic behavior of the focal surface temperature in response to environmental changes.
[0043] Specifically, the asymmetric focal surface temperature hysteresis response equation is constructed as follows:
[0044] ,
[0045] ,
[0046] ,
[0047] in, for The actual operating temperature of the focal surface at any given time, in degrees Celsius (°C). for The focal plane sensor outputs temperature at any given time, in degrees Celsius (°C). for The rate of change of focal surface temperature over time, in °C / s. The sampling time interval is in seconds. The overall hysteresis coefficient, in seconds. Reference hysteresis coefficient, unit: seconds. : Asymmetric factor, dimensionless, ranging from 0.1 to 0.5. : Symbolic function, Wind speed coupling coefficient, unit: s·m -1 ·s, Solar irradiance coupling coefficient, unit: s·m 2 ·W -1 , Solar irradiance.
[0048] , , , The identification was obtained through the least squares method based on blackbody calibration data in the first stage of S5. The identification method was as follows: under the condition of known true blackbody temperature, the temperature of the focal plane sensor and its rate of change were collected, and... The objective is to minimize the difference between the actual temperature of the blackbody and the above coefficients. This formula is obtained by... To achieve asymmetry in heating and cooling responses, by... This allows the hysteresis coefficient to change dynamically with meteorological conditions, which differs from existing modeling methods that use a fixed hysteresis coefficient.
[0049] S2.3 Solve the asymmetric focal surface temperature hysteresis response equation to obtain the true working temperature of the focal surface considering thermal inertia hysteresis and its confidence level. Thermal inertia hysteresis refers to the phenomenon that the actual temperature of the focal surface lags behind the sensor output temperature due to the heat capacity effect. After solving the equations for the heating and cooling branches respectively, the branch solution corresponding to the current rate of change of the focal surface temperature is taken as the true working temperature of the focal surface. The confidence level is a reliability index comprehensively evaluated based on the equation residuals, the fluctuation amplitude of the meteorological forcing term, and the magnitude of the rate of change, used to characterize the reliability of the temperature estimate.
[0050] S3. The initial temperature estimation feature vector and the actual working temperature of the focal plane are input into the temperature measurement deviation correction network. After dimensionality reduction, temporal fusion, downsampling, and channel enhancement, radiation residuals and energy conservation regularization terms are introduced to output the target temperature value. Based on the temperature estimation features output by the physical model and the actual focal plane temperature after hysteresis compensation, the law of residual temperature measurement deviation is learned through a data-driven approach. During the learning process, radiation physics constraints and energy conservation constraints are applied to ensure that the output temperature value is both accurate and conforms to physical laws.
[0051] The temperature measurement deviation correction network in S3 includes a dimension reduction module, a time-series fusion module, a downsampling module, a channel enhancement module, and a regression output module connected in sequence.
[0052] The dimensionality reduction module performs principal component analysis (PCA) on the input features consisting of the initial temperature estimation feature vector and the actual working temperature of the focal surface to extract the latent variables that dominate the temperature measurement error. PCA is a statistical dimensionality reduction method that transforms potentially correlated multiple variables into a few linearly uncorrelated composite variables, i.e., latent variables, through orthogonal transformation. Here, the input features include both the initial temperature estimation feature vector and the actual working temperature of the focal surface. These two components may contain redundant information. PCA retains the top principal components with the largest variance contribution, thus reducing the feature dimensionality while preserving the key information that has the greatest impact on the temperature measurement error.
[0053] The temporal fusion module employs a combination of sequential convolution and dense connections to capture the dynamic evolution of focal plane thermal hysteresis. Sequential convolution refers to a one-dimensional convolution operation along the time dimension, used to extract local patterns of input features changing over time. Dense connections pass the output of each layer to all subsequent layers, enabling feature reuse across layers. The dynamic evolution of focal plane thermal hysteresis refers to the lag, accumulation, and asymmetric response characteristics exhibited by the focal plane temperature as it changes over time. The combination of sequential convolution and dense connections effectively captures these temporal dependencies.
[0054] The downsampling module employs exponentially weighted downsampling to suppress transient temperature overshoot noise at the focal plane. Exponentially weighted downsampling assigns exponentially decaying weights to each element within the window during the downsampling process, resulting in elements closer to the center of the window having higher weights and those at the edges having lower weights. Transient temperature overshoot noise at the focal plane refers to short-term spikes in noise caused by thermal inertia and sensor response characteristics when the focal plane temperature changes rapidly. Exponentially weighted downsampling smooths out this type of noise while preserving the overall trend of temperature changes.
[0055] The channel enhancement module strengthens the representation capabilities of meteorological elements and focal surface temperature channels through adaptive channel weight enhancement. Adaptive channel weight enhancement applies learnable attention weights to each channel of the feature map, with the weight values dynamically adjusted based on the dependencies between channels. The meteorological element channel and the focal surface temperature channel respectively carry environmental meteorological information and focal surface thermal state information. Adaptive enhancement increases the contribution of these channels in subsequent regression and suppresses interference from irrelevant channels.
[0056] The regression output module maps the enhanced features to the target temperature space and outputs the target temperature value. This module typically consists of a fully connected layer that maps the high-dimensional features, after dimensionality reduction, temporal fusion, downsampling, and channel enhancement, into a single temperature value through linear or nonlinear transformations, enabling direct prediction of the target temperature.
[0057] Specifically, in this embodiment, step S3 introduces radiation residuals and energy conservation regularization terms, including:
[0058] S3.1 Construct the Planck radiation residuals The goal is to ensure that the radiance of the target temperature value output by the constraint network, converted using Planck's radiation law, matches the radiance corresponding to the measured infrared quantized value. Planck's radiation law describes the quantitative relationship between the radiance emitted by a blackbody at different temperatures and its wavelength, forming the physical basis of infrared thermometry. The Planck radiation residual term calculates the theoretical radiance by substituting the network's predicted target temperature value into Planck's radiation law, compares it with the radiance corresponding to the measured infrared quantized value, and uses the difference as a penalty term. This forces the network's output temperature value to conform to the physical laws of infrared radiation transmission, avoiding physical inconsistencies caused by purely data-driven approaches.
[0059] S3.2 Constructing the energy conservation regularization term This constrains the energy balance between the target temperature value output by the network and the actual working temperature of the focal plane, as well as meteorological factors. Energy conservation refers to the energy balance between the target surface, the focal plane, and the surrounding environment during heat exchange. The energy conservation regularization term calculates the net heat flow to the target surface based on the target temperature, the actual working temperature of the focal plane, and meteorological factors (wind speed, solar irradiance, etc.), compares this net heat flow with the heat flow predicted by the theoretical heat balance equation, and uses the difference as a penalty term, thus constraining the network output to conform to the thermodynamic law of energy conservation.
[0060] S3.3. Add the Planck radiation residual term and the energy conservation regularization term to the total loss function. The total loss function is a weighted sum of the mean square error terms between the predicted and actual temperatures, the Planck radiation residual term, and the energy conservation regularization term. The mean square error term ensures that the predicted and actual temperatures are close in value; the Planck radiation residual term ensures that the output satisfies the physical laws of radiative transfer; and the energy conservation regularization term ensures that the output satisfies thermodynamic energy conservation. Weighting coefficients are used to adjust the relative contributions of each term to the total loss.
[0061] ,
[0062] in, To calculate the mean square error between the predicted temperature and the actual temperature, These are the weighting coefficients.
[0063] Specifically, the Planck radiation residual term and the energy conservation canonical term are calculated as follows:
[0064] ,
[0065] ,
[0066] ,
[0067] in, The total number of samples; For the first Measured infrared quantization values of each sample; For the first The model predicts the infrared quantization value for each sample; To prevent small quantities with a denominator of zero; The quantization value-radiance conversion coefficient is determined by the detector gain and bias. The first radiation constant is 3.7418 × 10⁻⁶. -16 W·m 2 ; The second radiation constant is 1.4388 × 10⁻⁶. -2 m·K; For the first The model predicts the target temperature for each sample, in °C. The convective heat transfer coefficient is expressed in W·m. -2 ·K -1 ; The Stefan-Boltzmann constant is 5.6697 × 10⁻⁶. -8 W·m -2 ·K -4 ; The target emission rate is dimensionless. Heat flux absorbed by solar radiation, unit: W·m -2 It is determined by solar irradiance and target absorptivity. For weights.
[0068] The theoretical quantized value is obtained by substituting the network-predicted temperature into Planck's formula and comparing it with the measured quantized value. The net heat flow is obtained by calculating the convective heat transfer, radiative heat transfer, and solar irradiation absorption heat flow between the target surface and the focal plane, and the environment, and comparing it with the theoretical expected value. The search range was determined by grid search, with a step size of 0.01 to 1. Make the network output satisfy Planck's radiation law. This ensures that the network output satisfies the thermodynamic energy conservation principle, and the two together constitute a physical constraint, which is different from the purely data-driven network training method.
[0069] S4. Input the target temperature value into the resolution enhancement network. After high-frequency component extraction and pixel rearrangement, a high-resolution temperature distribution image is output. Spatial detail enhancement is performed on the bias-corrected target temperature value. By extracting high-frequency information of the temperature field and applying energy consistency constraints, the low-resolution temperature data is reconstructed into a high-resolution temperature distribution image, improving the representation of spatial details of the temperature field.
[0070] Specifically, in this embodiment, the resolution enhancement network in S4 includes a high-frequency component extraction module, a sub-pixel thermal diffusion prior module, a feature fusion module, and a pixel rearrangement module.
[0071] The high-frequency component extraction module extracts high-frequency detail components from the temperature field by subtracting upsampling from downsampling of the low-resolution temperature map composed of target temperature values. Downsampling involves performing pooling or downsampling operations on the low-resolution temperature map to obtain a smoothed low-frequency temperature distribution; upsampling restores the low-frequency temperature distribution to its original size; and subtraction involves subtracting the original low-resolution temperature map from the restored low-frequency temperature distribution pixel by pixel, with the residual being the high-frequency detail component. The high-frequency detail component represents spatial detail information in the temperature field, such as local temperature differences, edge transitions, and subtle thermal features.
[0072] The sub-pixel thermal diffusion prior module constrains the spatial distribution of high-frequency detail components based on the thermal diffusion equation, suppressing non-physical temperature abrupt changes. The thermal diffusion equation describes the physical laws governing the propagation of heat from high-temperature regions to low-temperature regions in a medium, and its solution exhibits spatial smoothness and continuity. The sub-pixel thermal diffusion prior applies the thermal diffusion equation as a constraint to the high-frequency detail components, suppressing temperature gradients between adjacent pixels that exceed physically reasonable ranges. This avoids non-physical temperature abrupt changes caused by noise or model errors, ensuring that the reconstructed temperature field conforms to the laws of heat conduction.
[0073] The feature fusion module fuses high-frequency detail components with low-resolution temperature features through skip connections. Skip connections refer to directly passing features from earlier layers of the network to later layers, where they are superimposed or stitched together with the processed deeper features. This fusion of high-frequency detail components and low-resolution temperature features ensures that the reconstruction process simultaneously preserves the overall distribution trend of the temperature field and local detail information, avoiding the loss of high-frequency information during deep network processing.
[0074] The pixel rearrangement module rearranges the fused features by channel to generate a high-resolution temperature distribution image. Pixel rearrangement refers to rearranging different channel elements at the same spatial location in the feature map according to predetermined rules, thereby improving resolution. Specifically, if a magnification of r is required, the feature values of r² channels are rearranged into an r×r spatial block, and these blocks are stitched together to form a high-resolution temperature distribution image. This method achieves pixel-level reconstruction without interpolation and has high computational efficiency.
[0075] Specifically, the sub-pixel thermal diffusion prior module constrains the spatial distribution of high-frequency detail components in the following manner:
[0076] ,
[0077] ,
[0078] in, This is the a priori loss term for thermal diffusion; This represents the total number of pixels in the high-frequency detail component. for High-frequency temperature component at the location, in °C; The region represents the spatial region containing high-frequency detail components, where D is the thermal diffusivity in meters. 2 ·s -1 ; For the Laplace operator; This represents the rate of change of the high-frequency temperature component over time. Represents the spatial coordinates of the temperature image; It is the Laplacian operator for the high-frequency temperature component in space, representing the degree of deviation of the pixel temperature from the average temperature of its neighborhood. This represents the second spatial derivative along the direction of travel. The second-order spatial derivatives along the column direction are represented by the two-dimensional Laplace operator. The thermal diffusivity D is determined by the target material properties and obtained through a lookup table. The Laplace operator is discretized using a five-point difference scheme. The temporal partial derivatives are calculated by differentiating the high-frequency components of adjacent frames. This formula applies the thermal diffusivity equation as a constraint to the high-frequency detail components, suppressing non-physical temperature abrupt changes, unlike super-resolution reconstruction methods that rely solely on data-driven approaches.
[0079] When outputting the high-resolution temperature distribution image in step S4, the method further includes applying an energy consistency constraint to the high-resolution temperature distribution image: calculating the average energy of the high-resolution temperature distribution image in each local region; calculating the average energy of the original low-resolution temperature map in the corresponding local region; constructing an energy consistency loss term based on the difference between the two, and adding it to the total loss function of the resolution enhancement network, so that the average energy of the high-resolution temperature distribution image in the local region is consistent with that of the original low-resolution temperature map. Figure 1 To.
[0080] The energy consistency constraint refers to the constraint conditions applied to the high-resolution temperature distribution image during the resolution upscaling process to avoid local temperature energy shifts or distortions in the reconstructed image. The average energy within a local region refers to the average of the pixel temperature values within a spatial region of the temperature map, summed in an equivalent radiant energy manner, to characterize the overall thermal radiation level of that region. Since pixel rearrangement and convolution operations during super-resolution reconstruction may alter the distribution of local temperature values, leading to some regions being overly hot or underheated, it is necessary to use the average energy of the corresponding region in the original low-resolution temperature map as a benchmark for constraint.
[0081] Specifically, the high-resolution temperature distribution image is first divided into multiple local regions, and the average energy of each local region is calculated. Then, following the same spatial correspondence, corresponding regions are divided in the original low-resolution temperature image, and their average energy is calculated. Finally, an energy consistency loss term is constructed based on the difference between the two and added to the total loss function of the resolution upscaling network. During network training, this loss term forces the average energy of the high-resolution temperature distribution image in any local region to remain consistent with that of the original low-resolution temperature image, thereby ensuring that the super-resolution reconstruction does not introduce additional energy shifts and ensuring the physical reliability of the reconstructed temperature field at the radiation energy level.
[0082] Specifically, the energy consistency loss term is calculated as follows:
[0083] ,
[0084] ,
[0085] ,
[0086] in, This is the energy consistency loss term; This represents the total number of units in the local area. This represents the total radiant energy of the k-th local region in the high-resolution temperature map, expressed in W·m³. -2 ; This represents the total radiative energy of the k-th local region in the low-resolution temperature map, in W·m³. -2 ; Let be the set of pixels corresponding to the kth local region; This represents the temperature value of the j-th pixel in the high-resolution temperature map, in °C. This represents the temperature value of the j-th pixel in the low-resolution temperature map, in °C. It is the Stefan-Boltzmann constant; Target emission rate; : To prevent small quantities with a denominator of zero.
[0087] First, the high-resolution temperature map and the low-resolution temperature map are divided into K local regions according to the same spatial correspondence. Then, the total radiant energy of each region is calculated according to the formula mentioned above. Finally, the squared relative error is added as a loss term to the total loss function. This formula makes the total radiant energy of the high-resolution temperature map in the local region equal to that of the low-resolution temperature map. Figure 1 This approach aims to prevent energy shifts introduced during super-resolution reconstruction, unlike traditional super-resolution loss functions that only focus on pixel values that are close together.
[0088] S5. Identify the coefficients of the mechanism model described in S1 and S2 and the weights of the network described in S3 and S4 in stages, and output the optimal parameter set. Using multi-condition experimental data, determine the coefficients in the physical model and the weights in the network in stages, so that the entire method has a usable optimal parameter configuration, providing a parameter basis for actual operation.
[0089] Specifically, in this embodiment, S5 includes:
[0090] S5.1 Collect multi-condition samples covering different ambient temperatures, humidity levels, wind speeds, solar irradiance, and target emissivity to construct training and testing sets. The multi-condition samples refer to infrared quantization values, focal surface temperature, temperature change rate, meteorological elements, and corresponding real-world temperature labels collected under different combinations of ambient temperatures, humidity levels, wind speeds, solar irradiance, and target surface emissivity. The training set is used for model parameter learning, and the testing set is used to evaluate the model's generalization ability; the samples in the two sets do not overlap.
[0091] S5.2, First Stage: Fix the weights of the temperature measurement deviation correction network described in S3 and the resolution enhancement network described in S4. Based on the blackbody calibration data, use the least squares method to identify the coefficients of the infrared quantization value-irradiation response mechanism model described in S1 and the asymmetric focal plane temperature hysteresis response mechanism model described in S2. The blackbody calibration data refers to data collected under laboratory conditions with a high-precision blackbody as the target. Its true temperature is known and its emissivity is close to the ideal value, making it suitable for identifying the coefficients of the physical mechanism model. The least squares method is a mathematical optimization method that solves for undetermined coefficients by minimizing the sum of squared residuals between predicted and measured values. In this stage, the network weights are not updated; only the coefficients of the physical model are determined to ensure the physical rationality of the mechanism model.
[0092] Specifically, the least squares identification is implemented in the following manner:
[0093] ,
[0094] in, The set of coefficients for the mechanism model to be identified includes , To determine the number of data samples for blackbody calibration; Let be the true blackbody temperature of the i-th sample, in °C. This is the joint output function of the S1 and S2 mechanism models; The set of coefficients to be identified is given. The Levenberg-Marquardt algorithm is used iteratively to solve the problem. Initial values are set empirically, and the convergence condition is that the sum of squared residuals between two consecutive iterations changes by less than 10. -6 .
[0095] S5.3, Second Stage: Fix the coefficients of the mechanistic models described in S1 and S2 obtained in S5.2, and train the temperature measurement deviation correction network described in S3 and the resolution enhancement network described in S4 end-to-end based on the training set, updating the network weights. End-to-end training means treating S1 to S4 as a whole, using the loss function between the final output and the true label as the objective, and updating the network weights layer by layer through the backpropagation algorithm. This stage freezes the physical model coefficients, allowing the network learning focus to compensate for residual deviations not fully described by the physical model.
[0096] S5.4. Combine the mechanistic model coefficients obtained in S5.2 with the network weights obtained in S5.3 to output the optimal parameter set. The optimal parameter set refers to the parameter combination that minimizes the error of the overall model on the test set after two-stage identification. It includes all coefficients of the mechanistic models in S1 and S2 and all weights of the networks in S3 and S4, which are loaded for use in real-time correction in S6.
[0097] S6. Load the optimal parameter set, acquire infrared quantization values, focal plane temperature, temperature change rate, and meteorological elements in real time, and output corrected temperature values and high-resolution temperature distribution images after processing through S1 to S4. Load the pre-trained parameters, process the real-time acquired infrared data and meteorological information through the above steps in sequence, and finally output temperature values corrected by physical constraints and high-resolution temperature distribution images for engineering applications.
[0098] Specifically, in this embodiment, S6 includes:
[0099] S6.1 Load the optimal parameter set output by S5, and configure the coefficients of the mechanism models described in S1 and S2 and the weights of the networks described in S3 and S4 to the corresponding models and networks respectively. The optimal parameter set contains all parameters obtained from the two-stage identification. Loading refers to writing the above parameters into the storage unit of the real-time processing system, and enabling the S1 and S2 mechanism models and the S3 and S4 networks to read their respective coefficients and weights, completing the parameter initialization configuration before operation.
[0100] S6.2 Real-time acquisition of infrared quantization values, focal plane temperature, temperature change rate, and meteorological elements. Specifically, the infrared quantization values are output in real-time by the infrared detector; the focal plane temperature is read by a temperature sensing element integrated into the focal plane; the temperature change rate is calculated by the difference in focal plane temperatures between adjacent frames; and meteorological elements are synchronously collected by an external meteorological sensing unit. Real-time acquisition means that the above data are sampled synchronously under the same time reference, ensuring that all input quantities are aligned in time.
[0101] S6.3 Input the infrared quantization value, focal plane temperature, temperature change rate, and meteorological elements into the infrared quantization value-irradiation response mechanism model described in S1, and output the initial temperature estimation feature vector. The mechanism model in S1 has already loaded the coefficients identified in S5. Based on the input infrared quantization value and meteorological elements, the temperature estimate is calculated according to the response equations of each thermal inertia quantum region, and then concatenated into a feature vector based on the spatial position of the focal plane for output.
[0102] S6.4 Input the focal surface temperature, temperature change rate, and meteorological elements into the asymmetric focal surface temperature hysteresis response mechanism model described in S2, and output the actual working temperature of the focal surface. The mechanism model in S2 has already loaded the coefficients identified in S5. Based on the focal surface temperature change rate and meteorological elements, it determines whether the current path is one of heating or cooling. The hysteresis response equation for the corresponding path is then solved to output the actual working temperature of the focal surface considering thermal inertia hysteresis.
[0103] S6.5. Input the initial temperature estimation feature vector and the actual working temperature of the focal plane into the temperature measurement deviation correction network described in S3, and output the target temperature value. The S3 network has been loaded with the weights trained in S5, and is processed sequentially by the dimensionality reduction module, temporal fusion module, downsampling module, channel enhancement module, and regression output module. It is constrained by the Planck radiation residual term and the energy conservation regularization term, and outputs the deviation-corrected target temperature value.
[0104] S6.6 Input the target temperature value into the resolution enhancement network described in S4, and output the corrected temperature value and a high-resolution temperature distribution image. The S4 network has been loaded with the weights trained in S5, and is processed sequentially by the high-frequency component extraction module, the sub-pixel thermal diffusion prior module, the feature fusion module, and the pixel rearrangement module. An energy consistency constraint is applied, and the final output is the temperature value corrected by physical constraints, along with the local average energy and the original low-resolution temperature. Figure 1 High-resolution temperature distribution images.
[0105] Example 2
[0106] like Figure 2 As shown, this application provides an infrared thermometry compensation and correction system architecture diagram that integrates environmental meteorological data. This system is applied to the infrared thermometry compensation and correction method for integrating environmental meteorological data as described in Embodiment 1, specifically as follows:
[0107] Mechanism modeling module 210 is used to construct an infrared quantization value-irradiation response mechanism model that incorporates spatial heterogeneity of the focal surface temperature field and corrections by meteorological elements, and outputs an initial temperature estimation feature vector.
[0108] The hysteresis response module 220 is used to synchronously acquire the focal surface temperature, temperature change rate and meteorological elements, construct an asymmetric focal surface temperature hysteresis response mechanism model and output the true working temperature of the focal surface.
[0109] The temperature measurement deviation correction module 230 is used to input the initial temperature estimation feature vector and the actual working temperature of the focal surface into the temperature measurement deviation correction network and output the target temperature value.
[0110] The resolution enhancement module 240 is used to input the target temperature value into the resolution enhancement network, extract high-frequency components and rearrange pixels to output a high-definition temperature distribution image.
[0111] The parameter identification module 250 is used to identify the coefficients of the mechanism model in the mechanism modeling module and the hysteresis response module, as well as the weights of the network in the temperature measurement deviation correction module and the resolution enhancement module in stages, and output the optimal parameter set.
[0112] The real-time correction module 260 is used to load the optimal parameter set, acquire infrared quantization values, focal plane temperature, temperature change rate and meteorological elements in real time, and process them sequentially through the mechanism modeling module, hysteresis response module, temperature measurement deviation correction module and resolution enhancement module, and output the corrected temperature value and high-definition temperature distribution image.
[0113] Figure 3 This is an electronic device provided in one embodiment of this application. For example... Figure 3 As shown, the electronic device includes at least the following components: processor 301 and memory 300, communication interface 303, and bus 302.
[0114] In this embodiment of the application, memory 300 is used to store executable instructions of processor 301, which, when configured to execute instructions, implements the method as described in the first aspect.
[0115] In embodiments of this application, a computer-readable storage medium includes instructions that instruct a device to perform the method as described in the first aspect. For example, the instructions instruct the device to perform... Figure 1 The method is shown in the process steps.
[0116] In one embodiment of this application, the program operating in the electronic device may be a program that controls a central processing unit (CPU) or similar device to achieve the functions of the above-described embodiments of the present invention (a program that enables the computer to function). Information processed by these systems is then temporarily stored in random access memory (RAM) during processing, and subsequently stored in various ROMs such as read-only memory (FlashROM) and hard disk drives (HDDs), and read, corrected, and written by the CPU as needed.
[0117] It should be noted that a portion of the electronic device described above can also be implemented using a computer. In this case, the program for implementing the control function can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into the computer and executed.
[0118] It should be noted that the computer mentioned here refers to a computer built into an electronic device, employing hardware including an operating system and peripheral devices. Furthermore, computer-readable recording media refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage systems such as hard drives built into the computer.
[0119] Furthermore, computer-readable recording media can include: media that dynamically stores programs for short periods of time, such as communication lines used when transmitting programs via networks like the Internet or communication lines like telephone lines; and media that store programs for fixed periods of time, such as volatile memory inside a computer that serves as a server or client in this case. In addition, the aforementioned program can be a program used to implement the above-mentioned functions, or it can be a program that can implement the above-mentioned functions by combining them with programs already recorded in the computer.
[0120] Furthermore, the electronic device in the above embodiments can also be implemented as an assembly (system group) composed of multiple systems. Each system constituting the system group can possess some or all of the functions or functional blocks of the electronic device in the above embodiments. As a system group, it is sufficient to have all the functions or functional blocks of the electronic device.
[0121] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed in this application.
Claims
1. An infrared temperature measurement compensation correction method that fuses environmental meteorological data, characterized in that, Includes the following steps: S1. Construct an infrared quantization value-irradiance response mechanism model that incorporates spatial heterogeneity of the focal surface temperature field and meteorological element correction, and output the initial temperature estimation feature vector; S2. Simultaneously acquire the focal surface temperature, temperature change rate and meteorological elements, construct an asymmetric focal surface temperature hysteresis response mechanism model, and output the true working temperature of the focal surface. S3. Input the initial temperature estimation feature vector and the actual working temperature of the focal plane into the temperature measurement deviation correction network. After dimensionality reduction, temporal fusion, downsampling and channel enhancement, and the introduction of radiation residual and energy conservation regularization terms, the target temperature value is output. S4. Input the target temperature value into the resolution enhancement network, extract high-frequency components and rearrange pixels to output a high-definition temperature distribution image; S5. Identify the coefficients of the mechanism model described in S1 and S2 and the weights of the network described in S3 and S4 in stages, and output the optimal parameter set. S6. Load the optimal parameter set, acquire infrared quantization values, focal surface temperature, temperature change rate and meteorological elements in real time, and output the corrected temperature value and high-definition temperature distribution image after processing by S1 to S4 in sequence.
2. The method of claim 1, wherein, S1 includes: S1.1 Divide the detector focal plane into multiple thermal inertial quantum regions, and establish infrared quantization value-irradiation response equations for each sub-region; S1.
2. An atmospheric attenuation correction term is introduced into the response equation of each sub-region. The atmospheric attenuation correction term is calculated in real time based on ambient temperature, humidity, wind speed and solar irradiance. S1.
3. The outputs of the response equations of each sub-region are spliced together according to the focal plane spatial position to obtain the initial temperature estimation feature vector.
3. The method of claim 1, wherein, S2 includes: S2.
1. Synchronously collect focal surface temperature, focal surface temperature change rate, ambient wind speed and solar irradiance; S2.2 Using the rate of change of coke surface temperature and meteorological forcing term as inputs, construct an asymmetric coke surface temperature hysteresis response equation with asymmetric heating and cooling responses; S2.3 Solve the asymmetric focal surface temperature hysteresis response equation to obtain the true working temperature of the focal surface considering thermal inertia hysteresis and its confidence level.
4. The method of claim 1, wherein, The temperature measurement deviation correction network in S3 includes a dimension reduction module, a time-series fusion module, a downsampling module, a channel enhancement module, and a regression output module connected in sequence. The dimensionality reduction module performs principal component analysis on the input features consisting of the initial temperature estimation feature vector and the actual working temperature of the focal surface to extract the latent variables that dominate the temperature measurement error. The temporal fusion module employs a sequence convolution and dense connection structure to capture the dynamic evolution of focal thermal hysteresis. The downsampling module employs exponentially weighted downsampling to suppress transient temperature overshoot noise at the focal plane. The channel enhancement module strengthens the characterization capability of meteorological elements and focal surface temperature channels through adaptive channel weight enhancement. The regression output module maps the enhanced features to the target temperature space and outputs the target temperature value.
5. The method of claim 4, wherein, The S3 section introduces radiation residuals and energy conservation regularization terms, including: S3.1, Constructing Planck radiation residual term The target temperature value of the constraint network output is consistent with the radiation brightness corresponding to the measured infrared quantization value after the radiation brightness is converted by the Planck radiation law. S3.2 Constructing the energy conservation regularization term The energy balance between the target temperature value output by the constrained network and the actual working temperature of the focal surface and meteorological elements; S3.
3. Add the Planck radiation residual term and the energy conservation regularization term to the total loss function: , in, To calculate the mean square error between the predicted temperature and the actual temperature, These are the weighting coefficients.
6. The method according to claim 1, characterized in that, The resolution enhancement network in S4 includes a high-frequency component extraction module, a sub-pixel thermal diffusion prior module, a feature fusion module, and a pixel rearrangement module. The high-frequency component extraction module extracts high-frequency detail components of the temperature field by subtracting the upsampling from the downsampling of the low-resolution temperature map composed of the target temperature value. The subpixel thermal diffusion prior module constrains the spatial distribution of high-frequency detail components according to the thermal diffusion equation, suppressing non-physical temperature abrupt changes. The feature fusion module fuses high-frequency detail components with low-resolution temperature features through skip connections; The pixel rearrangement module rearranges the fused features by channel to generate a high-definition temperature distribution image.
7. The method according to claim 1, characterized in that, S5 includes: S5.1 Collect multi-condition samples covering different ambient temperatures, humidity, wind speeds, solar irradiance, and target emissivity to construct training and testing sets; S5.2, First stage: Fix the weights of the temperature measurement deviation correction network described in S3 and the resolution enhancement network described in S4, and use the least squares method to identify the coefficients of the infrared quantization value-irradiation response mechanism model described in S1 and the asymmetric focal surface temperature hysteresis response mechanism model described in S2 based on the blackbody calibration data. S5.3, Second stage: Fix the coefficients of the mechanism model described in S1 and S2 obtained in S5.2, train the temperature measurement deviation correction network described in S3 and the resolution enhancement network described in S4 end-to-end based on the training set, and update the network weights; S5.
4. Combine the mechanistic model coefficients obtained in S5.2 with the network weights obtained in S5.3 to output the optimal parameter set.
8. The method according to claim 1, characterized in that, S6 includes: S6.1 Load the optimal parameter set output by S5, and configure the coefficients of the mechanism model described in S1 and S2 and the weights of the network described in S3 and S4 to the corresponding model and network respectively. S6.
2. Real-time acquisition of infrared quantization values, focal surface temperature, temperature change rate, and meteorological elements; S6.3 Input the infrared quantization value, focal surface temperature, temperature change rate and meteorological elements into the infrared quantization value-irradiation response mechanism model in S1, and output the initial temperature estimation feature vector; S6.4 Input the focal surface temperature, temperature change rate and meteorological elements into the asymmetric focal surface temperature hysteresis response mechanism model described in S2, and output the actual working temperature of the focal surface. S6.5 Input the initial temperature estimation feature vector and the actual working temperature of the focal surface into the temperature measurement deviation correction network in S3, and output the target temperature value; S6.6 Input the target temperature value into the resolution enhancement network described in S4, and output the corrected temperature value and a high-definition temperature distribution image.
9. An infrared thermometry compensation and correction system integrating environmental meteorological data, used to implement the method as described in any one of claims 1 to 8, characterized in that, The system includes: The mechanism modeling module is used to construct an infrared quantized value-irradiance response mechanism model that incorporates spatial heterogeneity of the focal surface temperature field and corrections by meteorological elements, and outputs an initial temperature estimation feature vector. The hysteresis response module is used to synchronously acquire the focal surface temperature, temperature change rate and meteorological elements, construct an asymmetric focal surface temperature hysteresis response mechanism model, and output the actual working temperature of the focal surface. The temperature measurement deviation correction module is used to input the initial temperature estimation feature vector and the actual working temperature of the focal surface into the temperature measurement deviation correction network and output the target temperature value. The resolution enhancement module is used to input the target temperature value into the resolution enhancement network, and after high-frequency component extraction and pixel rearrangement, output a high-definition temperature distribution image. The parameter identification module is used to identify the coefficients of the mechanism model in the mechanism modeling module and the hysteresis response module, as well as the weights of the network in the temperature measurement deviation correction module and the resolution enhancement module in stages, and output the optimal parameter set. The real-time correction module is used to load the optimal parameter set and acquire infrared quantization values, focal plane temperature, temperature change rate and meteorological elements in real time. After being processed by the mechanism modeling module, hysteresis response module, temperature measurement deviation correction module and resolution enhancement module, the corrected temperature value and high-definition temperature distribution image are output.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, implement the infrared temperature measurement compensation and correction method for fusing environmental meteorological data as described in any one of claims 1 to 8.
