A Feature Analysis Method and System for Infrared Thermal Imagers

Abstract

The present invention relates to the field of infrared thermal imaging technology, and discloses a feature analysis method and system for an infrared thermal imager. The method includes: obtaining infrared thermal images of multiple focal planes, and using an adaptive filtering and enhancement algorithm to remove noise and improve contrast. Using an edge detection algorithm to determine the heat source area and obtain precise positioning data. Then, based on an improved optical flow algorithm, the heat source influence area is extracted, a preliminary thermal flow vector field is calculated, and abnormal vectors are removed through a preset noise filtering mechanism. Subsequently, the filtered thermal flow vector field is input into a thermodynamic model for correction to improve the accuracy of the thermal flow trend. Finally, a complete thermal distribution image is generated based on the corrected thermal flow vector field, and thermal flow feature reconstruction is performed to obtain a high-precision thermal flow vector field. The method can dynamically reflect the change of heat quantity.

Description

Technical Field

[0001] The present invention relates to the field of infrared thermal imaging technology, and particularly to a feature analysis method and system for an infrared thermal imager. Background Art

[0002] Infrared thermal imaging technology is an imaging technology based on thermal radiation detection. By detecting the infrared radiation intensity on the surface of a target object, it is converted into temperature information and a visual thermal image is generated. This technology is widely used in industrial inspection, medical diagnosis, building heat loss detection, environmental monitoring, safety protection and other fields. In recent years, with the improvement of the performance of infrared detectors and the enhancement of data processing capabilities, infrared thermal imaging technology has gradually developed towards high resolution, intelligent analysis and real-time monitoring, aiming to provide more accurate and rich information to meet the application requirements in complex environments.

[0003] Currently, infrared thermal imaging technology is mainly used to measure the temperature distribution on the surface of an object and reflect the thermal radiation conditions in different regions by generating a thermal image. Common thermal imaging devices rely on highly sensitive detectors and image processing algorithms to ensure the accuracy of temperature data and the visualization effect. Some advanced systems combine image enhancement, target recognition and intelligent analysis methods to improve the clarity of thermal images and the ability to extract information.

[0004] Although the current infrared thermal imaging technology has been widely applied in static temperature measurement, its main limitation is that it can only provide the temperature distribution at a certain moment and lacks an accurate description of the dynamic changes of heat. Traditional infrared thermal imaging systems can only output static thermal images and are difficult to capture the flow direction, transmission path and diffusion trend of heat, which limits their application in dynamic thermal monitoring, thermal energy optimization management and fault prediction. Summary of the Invention

[0005] The present invention provides a feature analysis method and system for an infrared thermal imager to dynamically reflect heat changes.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a feature analysis method for an infrared thermal imager, including:

[0007] Obtaining infrared thermal images of multiple focal planes;

[0008] Based on an adaptive filtering and enhancement algorithm, performing noise removal and improving the contrast on the infrared thermal images of the multiple focal planes to obtain enhanced infrared thermal images;

[0009] Based on an edge detection algorithm, determining heat sources on the enhanced infrared thermal images to obtain precise positioning data of heat source regions;

[0010] Based on the improved optical flow algorithm, extract the heat source influence area from the accurate positioning data of the heat source area to obtain a preliminary thermal flow vector field;

[0011] Based on a preset noise filtering mechanism, remove abnormal vectors from the preliminary thermal flow vector field to obtain a filtered thermal flow vector field;

[0012] Input the filtered thermal flow vector field into a thermodynamic model to output a corrected thermal flow vector field;

[0013] Generate a thermal distribution image based on the corrected thermal flow vector field to obtain a complete thermal distribution image;

[0014] Reconstruct the thermal flow characteristics based on the complete thermal distribution image to obtain a high-precision thermal flow vector field.

[0015] Preferably, based on an adaptive filtering and enhancement algorithm, remove noise from the infrared thermal image of the multi-focal plane and improve the contrast to obtain an enhanced infrared thermal image, including:

[0016] Perform normalization processing on the infrared thermal image of the multi-focal plane to obtain a multi-focal plane infrared thermal image with equalized contrast;

[0017] Based on an adaptive filtering algorithm, remove noise from the multi-focal plane infrared thermal image with equalized contrast to obtain a noise-reduced infrared thermal image;

[0018] Based on an enhancement algorithm, optimize the local contrast of the noise-reduced infrared thermal image to obtain an enhanced infrared thermal image.

[0019] Preferably, based on an edge detection algorithm, determine the heat source from the enhanced infrared thermal image to obtain accurate positioning data of the heat source area, including:

[0020] Perform thermal radiation screening on the enhanced infrared thermal image to obtain a preliminary heat source distribution map;

[0021] Based on an edge detection algorithm, refine the contour of the preliminary heat source distribution map to obtain a complete heat source distribution area;

[0022] Based on morphological analysis and region growing algorithm, screen and optimize the complete heat source distribution area to obtain accurate positioning data of the heat source area;

[0023] The accurate positioning data of the heat source area includes: heat source center coordinates, heat source contour shape, and thermal distribution feature map.

[0024] Preferably, based on an improved optical flow algorithm, precise positioning data of the heat source region is used to extract the heat source influence region, obtaining a preliminary thermal flow vector field, including:

[0025] Taking the heat source center coordinates as the basis, the key region of heat change is determined to obtain the heat source influence region;

[0026] Based on the morphological analysis method combined with the shape of the heat source contour, the heat propagation boundary is defined for the heat source influence region, obtaining the heat propagation boundary region;

[0027] Based on the improved optical flow algorithm combined with the heat distribution feature map, temporal analysis is performed on the heat propagation boundary region, and the pixel-level temperature gradient change between adjacent frames is calculated to obtain a preliminary thermal flow vector field.

[0028] Preferably, based on a preset noise filtering mechanism, abnormal vectors are removed from the preliminary thermal flow vector field to obtain a filtered thermal flow vector field, including:

[0029] Based on the gradient anomaly detection method, the preliminary thermal flow vector field is subjected to anomaly recognition to obtain an abnormal vector classification list;

[0030] Based on the adaptive smoothing filtering method, the abnormal vectors in the abnormal vector classification list are removed to obtain a denoised thermal flow vector field;

[0031] Based on the interpolation reconstruction method, data filling is performed on the denoised thermal flow vector field to obtain a filtered thermal flow vector field.

[0032] Preferably, the filtered thermal flow vector field is input into a thermodynamic model to output a corrected thermal flow vector field, including:

[0033] The thermodynamic model is trained by a PINN neural network model;

[0034] Through the input layer of the thermodynamic model, vector features are generated for the filtered thermal flow vector field to obtain high-dimensional features of the thermal flow vector;

[0035] Through the hidden layer of the thermodynamic model, physical constraints are imposed on the high-dimensional features of the thermal flow vector to obtain physically constrained thermal flow vector features;

[0036] Through the output layer of the thermodynamic model, feature mapping is performed on the physically constrained thermal flow vector features to obtain a corrected thermal flow vector field.

[0037] Preferably, based on the corrected thermal flow vector field, a heat distribution image is generated to obtain a complete heat distribution image, including:

[0038] Based on the bilinear interpolation method, perform spatial reconstruction on the corrected heat flow vector field, and use the Fourier transform method for smoothing to obtain the reconstructed heat flow vector field;

[0039] Based on the Gaussian weight mapping method, generate a temperature distribution matrix for the reconstructed heat flow vector field to obtain a temperature distribution matrix;

[0040] According to the temperature distribution matrix, perform histogram equalization to optimize the contrast and obtain a complete heat distribution image.

[0041] Preferably, based on the complete heat distribution image, perform heat flow feature reconstruction to obtain a high-precision heat flow vector field, including:

[0042] Based on the time series analysis method, extract the pixel-level temperature change trend from the complete heat distribution image to obtain preliminary heat flow trajectory data;

[0043] Based on the Kalman filter method, smooth the preliminary heat flow trajectory data and generate features to obtain high-precision heat flow vector features;

[0044] Based on the Farneback optical flow method combined with the B-spline interpolation method, perform optical flow optimization and interpolation correction on the high-precision heat flow vector features to obtain a high-precision heat flow vector field.

[0045] In a second aspect, the present invention provides a feature analysis system for an infrared thermal imager, including:

[0046] A data acquisition module for acquiring infrared thermal images of multiple focal planes;

[0047] An infrared thermal image module for removing noise and enhancing the contrast of the infrared thermal images of multiple focal planes based on an adaptive filtering and enhancement algorithm to obtain enhanced infrared thermal images;

[0048] A heat source determination module for determining the heat source of the enhanced infrared thermal image using an edge detection algorithm to obtain precise positioning data of the heat source area;

[0049] A preliminary heat flow vector field module for extracting the heat source influence area from the precise positioning data of the heat source area based on an improved optical flow algorithm to obtain a preliminary heat flow vector field;

[0050] A noise filtering module for removing abnormal vectors from the preliminary heat flow vector field based on a preset noise filtering mechanism to obtain a filtered heat flow vector field;

[0051] A thermodynamics module for inputting the filtered heat flow vector field into a thermodynamics model and outputting a corrected heat flow vector field;

[0052] A heat distribution image module for generating a heat distribution image based on the corrected heat flow vector field to obtain a complete heat distribution image;

[0053] A heat flow vector field module for reconstructing heat flow characteristics based on the complete heat distribution image to obtain a high-precision heat flow vector field.

[0054] In a third aspect, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for feature analysis of an infrared thermal imager described in any one of the above is implemented.

[0055] In a fourth aspect, the present invention also provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program. Wherein, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method for feature analysis of an infrared thermal imager described in any one of the above.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) This method optimizes multi-focal plane infrared thermal images through an adaptive filtering and enhancement algorithm, improves the contrast by normalization processing, and combines adaptive filtering to remove noise, thereby generating high-quality infrared images. This method effectively improves the recognition ability of the heat source area and provides clear and stable data input for subsequent heat flow analysis.

[0058] (2) This method refines the contour through an edge detection algorithm, combines morphological analysis and region growth to optimize the heat source area, and finally obtains the heat source center coordinates, contour shape, and heat distribution feature map. This process ensures the accurate positioning of the heat source area and improves the reliability of subsequent heat flow analysis.

[0059] (3) This method is based on an improved optical flow algorithm, which acts on accurately positioned data to calculate the movement trend of heat in space. The heat propagation boundary is defined through morphological analysis, and then the temperature gradient change between adjacent frames is calculated in combination with time series analysis. Finally, a preliminary heat flow vector field is generated, providing basic data for the dynamic modeling of heat flow.

[0060] (4) This method introduces a physics-informed neural network (PINN) to optimize the heat flow vector field by combining a neural network with physical constraints. High-dimensional features are generated through the input layer, physical constraints are imposed on the high-dimensional features through the hidden layer, and feature mapping is performed using the output layer, ultimately generating a corrected heat flow vector field that conforms to the laws of thermodynamics. This method improves the accuracy and physical consistency of heat flow data, making it closer to the real heat conduction process.

[0061] In summary, the present invention combines technical means such as multi-focal plane thermal imaging, optical flow calculation, morphological analysis, and PINN neural network, breaking through the limitation of traditional infrared thermal imaging that only provides static thermal images and realizing the dynamic reflection of heat changes. Description of the Drawings

[0062] Figure 1 is a schematic flowchart of the feature analysis method for an infrared thermal imager provided by the first embodiment of the present invention;

[0063] Figure 2 is a schematic diagram of the feature analysis system for an infrared thermal imager provided by the second embodiment of the present invention. Detailed Embodiments

[0064] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0065] Referring to Figure 1 , the first embodiment of the present invention provides a feature analysis method for an infrared thermal imager, including the following steps:

[0066] S11, obtaining infrared thermal images of multiple focal planes;

[0067] S12, based on an adaptive filtering and enhancement algorithm, removing noise from the infrared thermal images of the multiple focal planes and improving the contrast to obtain enhanced infrared thermal images;

[0068] S13, based on an edge detection algorithm, determining heat sources for the enhanced infrared thermal images to obtain precise positioning data of the heat source regions;

[0069] S14, based on an improved optical flow algorithm, extracting the heat source influence regions from the precise positioning data of the heat source regions to obtain a preliminary heat flow vector field;

[0070] S15. Based on a preset noise filtering mechanism, remove abnormal vectors from the preliminary heat flow vector field to obtain a filtered heat flow vector field;

[0071] S16. Input the filtered heat flow vector field into a thermodynamic model to output a corrected heat flow vector field;

[0072] S17. Generate a heat distribution image based on the corrected heat flow vector field to obtain a complete heat distribution image;

[0073] S18. Reconstruct heat flow characteristics based on the complete heat distribution image to obtain a high-precision heat flow vector field.

[0074] In step S11, obtain infrared thermal images of multiple focal planes

[0075] It should be noted that in step S11, it is first necessary to obtain infrared thermal images of multiple focal planes to ensure higher accuracy and integrity in subsequent heat flow analysis. Infrared thermal imagers have different focal length adjustment functions. By changing the focal length, infrared thermal images on multiple focal planes can be obtained. This process can be completed by an automatic focusing system or, in an experimental environment, by gradually adjusting the focal length and collecting thermal images. In practical applications, for example, during the thermal diagnosis of equipment, the thermal distribution of some equipment may exhibit thermal gradient characteristics at different depths due to complex structures. Thermal images of a single focal plane will miss some key thermal information. Therefore, it is necessary to collect thermal images of multiple focal planes to ensure that complete thermal field information is captured.

[0076] During the data acquisition process, it is first necessary to set the parameters of the infrared thermal imager, including resolution, frame rate, temperature measurement range, etc. Resolution determines the detail performance of thermal images, frame rate affects the time resolution of heat changes, and temperature measurement range ensures that the instrument can cover the temperature range of the object to be measured. In actual industrial inspections, for example, during the inspection of power equipment, the temperature of high-voltage equipment can reach several hundred degrees Celsius, while the operating temperature of some precision electronic equipment is only several tens of degrees Celsius. Therefore, it is necessary to adjust the temperature measurement range according to the specific application scenario to obtain clear and accurate thermal images.

[0077] The key to obtaining multi - focal - plane infrared thermal images lies in the adjustment of the focal length. In an automated system, a stepper motor or a servo motor can be used to control the focal - length adjustment of the infrared lens, enabling the thermal imager to capture images at different focal planes and store the data in the system. In the manual operation mode, the focal length can be adjusted manually and the thermal images corresponding to each focal length can be recorded. Taking the thermal imaging detection of a building's exterior wall as an example, wall structures with different depths have different thermal conductivities. Using infrared thermal images with different focal planes can effectively distinguish the thermal distributions inside and outside the wall, thus more accurately analyzing the building's energy - saving situation or internal wall damage.

[0078] After the thermal images are collected, preliminary data sorting and storage are required. Since the data volume of multi - focal - plane thermal images is large, an efficient data storage and management system is needed. The system needs to record parameters such as the focal length, shooting time, and ambient temperature corresponding to each thermal image for subsequent data processing and comparative analysis. In this way, it can be ensured that the multi - focal - plane thermal image data used in subsequent steps is complete, traceable, and highly reliable.

[0079] In summary, obtaining multi - focal - plane infrared thermal images is a key step in ensuring the accuracy of heat - flow analysis. Through reasonable focal - length adjustment, parameter setting, and data management, clear and complete thermal - image data can be obtained, providing a reliable data basis for subsequent thermal - feature extraction, heat - source identification, and heat - flow analysis. In practical applications, whether it is industrial equipment monitoring, building heat - loss detection, or medical infrared diagnosis, this method can effectively improve the accuracy of thermal - imaging analysis, making subsequent analysis more accurate and reliable.

[0080] In step S12, based on the adaptive filtering and enhancement algorithm, for the multi - focal - plane infrared thermal image, noise removal is performed and the contrast is improved to obtain an enhanced infrared thermal image, including:

[0081] Perform normalization processing on the multi - focal - plane infrared thermal image to obtain a multi - focal - plane infrared thermal image with balanced contrast;

[0082] Based on the adaptive filtering algorithm, perform noise removal on the multi - focal - plane infrared thermal image with balanced contrast to obtain a denoised infrared thermal image;

[0083] Based on the enhancement algorithm, perform local contrast optimization on the denoised infrared thermal image to obtain an enhanced infrared thermal image.

[0084] It should be noted that in the process of optimizing multi-focal plane infrared thermal images based on adaptive filtering and enhancement algorithms, it mainly includes noise removal and contrast enhancement to improve the clarity and analyzability of the images. This process is crucial for subsequent heat source identification and thermal flow vector field calculation. Infrared thermal imaging data is easily affected by environmental noise, sensor errors, and equipment thermal drift during acquisition, resulting in a decline in image quality. To ensure the reliability of the data, a series of signal processing methods are required to optimize the images.

[0085] First, normalize the acquired multi-focal plane infrared thermal images to balance the image contrast. Since there are uneven brightness and temperature distributions in the images collected by the infrared thermal imager at different focal lengths, directly using the original data for analysis may lead to incorrect heat source identification results. Therefore, it is necessary to normalize the thermal images of all focal planes to make the temperature distributions of the images at different focal lengths consistent. The normalization process is generally achieved through linear transformation or histogram equalization techniques, so that the contrast and brightness of the images at different focal lengths reach a unified standard. For example, in the inspection of industrial equipment, the thermal radiation intensities of different equipment are different, and normalization can eliminate the influence of equipment materials and background temperatures, making heat sources in the same temperature range appear the same in the image. In building detection, infrared thermal imaging technology is used to evaluate the heat loss of walls or pipes. Without normalization, areas with small internal temperature differences in the wall cannot be clearly shown, thus affecting the analysis of building energy consumption. Through normalization, the contrast of these small temperature differences can be enhanced, making heat leakage areas easier to identify.

[0086] After the normalization process, an adaptive filtering method based on image local features is used to remove noise from the infrared thermal images. The infrared thermal imager is easily affected by environmental interference during operation, such as air flow, humidity changes, electromagnetic interference, etc., resulting in random noise in the image. In addition, the response characteristics of the sensor will also introduce fixed pattern noise, which will affect the boundary identification of heat sources and thermal flow calculation. Therefore, it is necessary to adjust the filtering parameters according to the local features of the image to optimize the noise reduction effect. Specifically, first calculate the local gradient of the infrared image to detect the temperature change amplitude in different regions. For regions with large gradient changes, reduce the smoothing intensity of the filtering to maintain the clarity of the heat source boundary; for regions with small gradient changes, increase the filtering smoothing intensity to reduce noise interference. For example, in the thermal imaging detection of high-voltage equipment, the temperature gradient on the surface of insulators is large, while the temperature gradient at the metal connection parts is relatively small. If a fixed filtering parameter is used to process the entire image, the high-temperature characteristics in the insulator area may be weakened, affecting the accuracy of defect detection. By adaptively adjusting the filtering parameters, the clarity of the heat source contour can be ensured while reducing noise, improving the accuracy of analysis.

[0087] After noise removal, local contrast optimization needs to be performed on the denoised infrared thermal image to enhance the boundary features of heat sources and improve the visualization effect of heat flow features. The dynamic range of infrared thermal images is affected by sensor performance and data processing algorithms, resulting in insufficient brightness contrast in some areas with small temperature differences in the original image, making the heat source boundaries blurred and difficult to accurately identify. To solve this problem, local contrast optimization technology is required to make areas with small temperature differences more clearly presented in the thermal image and improve the accuracy of subsequent heat source identification and heat flow analysis.

[0088] The core objective of local contrast optimization is to highlight areas with small temperature gradients, making the boundaries of heat sources and heat flow features clearer. Specific optimization methods include histogram equalization, adaptive gamma correction, and local region enhancement. First, in terms of contrast equalization, the local histogram equalization method is used to independently process different regions of the image instead of performing global equalization on the entire image. This method calculates the brightness distribution within local regions and performs normalization adjustments, making the brightness changes in areas with small temperature differences more obvious and enhancing the distinguishability of heat source boundaries. For example, in building heat loss detection, the thermal bridge effect in the wall causes small temperature differences in some areas inside the wall. If global histogram equalization is used, the contrast of some high-temperature areas may be reduced, while the local equalization method can ensure that these subtle temperature differences are highlighted, making hidden energy leakage areas more visible.

[0089] Secondly, in terms of adaptive gamma correction, the gamma value is adjusted according to the local brightness information of the image to enhance the brightness contrast in low-contrast areas. The specific operation is to block-process the image and analyze the brightness distribution of each block. For areas with low brightness and small temperature differences, the gamma value is increased to make the temperature gradients in these areas more obvious; for areas with high brightness, the gamma value is decreased to prevent overexposure. For example, in power transmission line inspection, thermal damage on the surface of insulators may manifest as tiny temperature anomalies. If a fixed gamma value is directly used to process the entire image, the temperature anomalies in low-temperature areas may be difficult to detect. The adaptive gamma correction method can adjust the gamma value according to the temperature distribution in different regions, making the temperature anomalies on the surface of insulators more prominent and improving the accuracy of fault detection.

[0090] Finally, in terms of local region enhancement, a Laplacian-enhanced-based method is used to strengthen the high-frequency components of the thermal image to enhance edge details and make the heat flow path and heat source boundaries more obvious. This method can improve the detail level of the thermal image while keeping the original temperature information unchanged.

[0091] In addition, in automotive night vision systems, infrared thermal imaging is used to detect pedestrians or animals in low-light environments. However, due to the small temperature difference between pedestrians and the background, it is difficult to clearly distinguish the pedestrian contours in ordinary infrared images. After local contrast optimization, the thermal radiation characteristics of pedestrians are enhanced, enabling the night vision system to more clearly identify pedestrian contours and improve driving safety. In the food industry, infrared thermal imaging can be used to detect the temperature uniformity during food processing. For example, in the baking industry, it is necessary to ensure uniform temperature distribution of bread or cakes, and unoptimized thermal images cannot accurately reflect temperature details. After contrast optimization, the differences in temperature distribution become more obvious, facilitating the control of temperature consistency during the baking process and improving food quality.

[0092] In practical engineering applications, such as the inspection of power lines, overheating of cable joints or locally abnormal heating areas of transformers can be detected through high-altitude infrared thermal imaging. However, due to complex external weather conditions, such as solar radiation and wind speed changes, the directly obtained infrared thermal images often have problems of noise interference and insufficient contrast. Through the above processing steps, irrelevant noise can be removed while maintaining key heat source information, improving the reliability of the data and providing high-quality input data for subsequent heat flow analysis.

[0093] In summary, the optimization of multi-focal plane infrared thermal images based on adaptive filtering and enhancement algorithms, including normalization processing, noise removal, and local contrast optimization, ensures the quality and stability of thermal images. By adjusting the filtering parameters according to the local features of the image to optimize the noise reduction effect, and performing local contrast optimization after noise reduction to enhance the temperature gradient features, the processed infrared thermal images have higher clarity and analyzability.

[0094] In step S13, based on the edge detection algorithm, for the enhanced infrared thermal image, heat source determination is performed to obtain precise positioning data of the heat source area, including:

[0095] For the enhanced infrared thermal image, thermal radiation screening is performed to obtain a preliminary heat source distribution map;

[0096] Based on the edge detection algorithm, the contour of the preliminary heat source distribution map is refined to obtain a complete heat source distribution area;

[0097] Based on morphological analysis and region growing algorithms, the complete heat source distribution area is screened and optimized to obtain precise positioning data of the heat source area;

[0098] The precise positioning data of the heat source area includes: heat source center coordinates, heat source contour shape, and heat distribution feature map.

[0099] It should be noted that in this step, based on the edge detection algorithm, the enhanced infrared thermal image is used to determine the heat source to obtain the precise positioning data of the heat source area. This process includes thermal radiation screening, edge detection, morphological analysis, and region growing algorithm to ensure that the final heat source area can accurately reflect the actual thermal distribution of the target. This process is crucial for subsequent heat flow analysis and heat transfer calculation. Because if the heat source area is inaccurately located, it will lead to a large deviation in the heat flow vector field calculated subsequently, thereby affecting the accuracy and reliability of the entire system.

[0100] First, the enhanced infrared thermal image is subjected to thermal radiation screening to obtain a preliminary heat source distribution map. The original data collected by the infrared thermal imager contains multiple heat source information, and affected by the ambient background temperature, the temperature gradient in the image is relatively complex. To accurately screen out the target heat source, it is necessary to set a threshold according to the thermal radiation intensity to eliminate the influence of ambient temperature changes and ensure that only the areas with significant thermal radiation characteristics are retained. For example, during the inspection of power equipment, the fault points of transformers appear as local overheating areas, while insulators show false hot spots due to ambient temperature changes. Through thermal radiation screening, the low-temperature background information can be effectively removed, and only the heating parts are retained, improving the accuracy of heat source identification. In building heat loss analysis, the temperature at the wall cracks or pipe leakage points is slightly higher than the surrounding area. Screening the parts with higher thermal radiation intensity can quickly lock the areas with energy leakage and improve the detection efficiency.

[0101] Next, based on the edge detection algorithm, the preliminary heat source distribution map is refined in contour to obtain a complete heat source distribution area. Since the temperature change in the infrared thermal image is continuous, the boundary of the heat source is often ambiguous. Therefore, it is necessary to use the edge detection algorithm to clarify the boundary of the heat source and make the heat source contour clearer. Canny edge detection is used for gradient calculation. Canny edge detection is an algorithm for extracting image edges, and its calculation process involves multiple steps to ensure that clear boundary information is extracted. In the gradient calculation stage, the algorithm first performs Gaussian filtering on the input infrared thermal image to reduce noise interference. Since the infrared thermal imaging image usually contains areas with continuous temperature gradient changes rather than clear boundaries, directly performing gradient calculation may lead to edge detection errors. Therefore, Gaussian filtering can smooth the image and reduce unnecessary high-frequency noise. In practical applications, for example, in the analysis of infrared thermal images of industrial equipment, the surface of the equipment may have noise points in the thermal imaging data due to dust or oxide layers. The role of Gaussian filtering is to reduce these interference factors to ensure more accurate subsequent gradient calculation.

[0102] After Gaussian filtering, the Canny algorithm uses the Sobel operator to calculate the gradient of the image. The Sobel operator is an edge detection method for calculating the gradient changes of an image in the x and y directions. In an infrared thermal image, the change in temperature gradient often coincides with the boundary of the heat source. Therefore, the purpose of calculating the gradient is to find the location where the temperature changes most drastically. The process of gradient calculation involves convolution operations in two directions, that is, calculating the gradients of the image in the horizontal and vertical directions respectively, and obtaining the gradient magnitude by taking the square root of the sum of squares, while calculating the gradient direction. Specifically, when detecting heat leakage in a building wall, there are areas on the wall surface with relatively small local temperature gradients, while the temperature change at the pipe leakage is more obvious. Through gradient calculation, the areas with the largest temperature changes can be identified, and these areas are usually the locations of pipe leakage or wall cracks.

[0103] After obtaining the gradient information, the Canny algorithm performs non-maximum suppression to remove non-boundary pixels. Since the gradient calculation results in an information image with multiple direction changes, there may be many non-edge pixels in it. Therefore, non-maximum suppression is needed to accurately locate the edges. In this process, the gradient value of each pixel is compared with the adjacent pixels along its gradient direction. If the gradient value of this pixel is not the local maximum, it will be suppressed to zero. In the thermal imaging detection of high-voltage cables, the temperature of the cable changes along the wire direction, and non-maximum suppression can ensure that only the main boundaries of the thermal distribution on the wire surface are retained, while removing background noise, thereby improving the accuracy of detection.

[0104] Finally, in the last step of edge detection, the Canny algorithm uses double-threshold processing to further refine the boundaries. By setting a high threshold and a low threshold, it ensures that only reliable boundaries are retained, and at the same time, the edge points above the low threshold but below the high threshold are used for connection. This can effectively reduce false boundaries and improve the robustness of edge detection. For example, during the infrared detection of an automobile engine, some high-temperature components may produce uneven temperature distributions, and double-threshold processing can ensure that only the thermal boundaries of the main engine components are extracted, without being affected by small-scale temperature changes, thereby providing more accurate fault diagnosis results.

[0105] Subsequently, based on morphological analysis and region growing algorithm, the complete heat source distribution region is screened and optimized to obtain the precise positioning data of the heat source region. Morphological analysis is mainly used to remove noise points and fill small gaps, making the contour of the heat source region more complete. The region growing algorithm merges pixel regions with similar temperature characteristics by setting growth conditions to form a complete heat source target. For example, during the thermal monitoring of new energy batteries, the heat generation region of the battery pack presents an irregular shape due to the influence of the heat sink. Through morphological analysis, the interference of the heat sink on the thermal image can be removed, and the actual heat generation region of the battery core can be completely identified by the region growing algorithm, providing accurate temperature data for the battery management system. In the fault detection of aero-engine blades, due to the different thermal conductivity characteristics of the blade materials, there is an obvious temperature gradient in their temperature distribution. Using the region growing algorithm, adjacent high-temperature regions can be accurately aggregated to obtain the actual temperature distribution of the blade to evaluate its working state.

[0106] Finally, the precise positioning data of the heat source region obtained includes the heat source center coordinates, the heat source contour shape, and the heat distribution characteristic map. Among them, the heat source center coordinates are used to determine the core heat generation point position of the heat source, which is of great significance for fault location and energy transfer analysis. For example, in the fault detection of high-voltage cable joints, the overheating point of the joint is often the core part of the fault. By extracting the heat source center coordinates, the part that needs to be repaired can be accurately marked. The heat source contour shape is used to analyze the diffusion mode of the heat source and help judge the influence range of the heat source. For example, in fire rescue, the spread trend of the fire source can be evaluated through the heat source contour shape to formulate a reasonable fire extinguishing strategy. The heat distribution characteristic map is used for further heat flow calculation to ensure that the calculation of the heat flow vector field can be carried out based on accurate heat source data. For example, in industrial production, the temperature control requirements of some workpieces are accurate to a certain range. The heat distribution characteristic map can help monitor the temperature distribution in real time to adjust the heating or cooling strategy to ensure the stability of the production process.

[0107] In step S14, based on the improved optical flow algorithm, the precise positioning data of the heat source region is used to extract the heat source influence region, and a preliminary heat flow vector field is obtained, including:

[0108] Taking the heat source center coordinates as the basis, the key region of heat change is determined to obtain the heat source influence region;

[0109] Based on the morphological analysis method combined with the heat source contour shape, the heat source influence region is used to define the heat propagation boundary, and a heat propagation boundary region is obtained;

[0110] Based on the improved optical flow algorithm combined with the heat distribution characteristic map, the heat propagation boundary region is subjected to temporal analysis, and the pixel-level temperature gradient change between adjacent frames is calculated to obtain a preliminary heat flow vector field.

[0111] It should be noted that in this step, based on the improved optical flow algorithm, the heat source influence area is extracted from the accurate positioning data of the heat source area, and the preliminary distribution of the heat flow vector field is calculated. This process mainly includes determining the key area of heat change, defining the heat propagation boundary, and calculating the heat flow vector field based on the optical flow algorithm. The goal of this series of operations is to provide high-precision input data for subsequent heat transfer analysis to ensure the accurate calculation of the heat flow direction and intensity.

[0112] First, based on the heat source center coordinates, the key area of heat change is determined to obtain the heat source influence area. During the infrared thermal imaging process, the target heat source does not exist in isolation but will affect the temperature of the surrounding area. Therefore, it is necessary to further expand on the basis of the initial heat source area to determine its influence range. For example, in the detection of power equipment, the overheating point of a transformer is not limited to a single point but will spread to the surrounding wires and insulating materials. By analyzing the temperature gradient change of the heat source center coordinates, the size of the influence area can be reasonably determined to ensure that the subsequent calculated heat flow vector field can cover the actual heat propagation range. In addition, in the thermal processing technology of industrial production, some workpieces will exhibit uneven heat diffusion during processing. If only the heat source itself is concerned and the temperature changes of the surrounding materials are ignored, it will affect the precise adjustment of the temperature control system. Therefore, the area expansion based on the heat source center coordinates can provide more valuable heat distribution information.

[0113] Next, based on the morphological analysis method combined with the heat source contour shape, the heat propagation boundary of the heat source influence area is defined to obtain the heat propagation boundary area. By analyzing the heat source contour shape, the basic contour information of the heat source is extracted. The heat source in the infrared thermal image usually shows a gradient diffusion, and the boundary is not a clear line but a gradually changing transition area. To more accurately define the heat propagation boundary, it is necessary to determine a reasonable boundary range based on the heat gradient analysis. For example, in the inspection of power equipment, the overheating point of a transformer often transfers heat to the surrounding air and wires, and the direction of heat diffusion depends on the thermal conductivity of the equipment materials and the convection of the surrounding air. By analyzing the contour shape of the heat source, the main direction of heat diffusion can be reasonably estimated and constrained in the calculation of the heat propagation area to avoid interference from irrelevant areas.

[0114] Secondly, the morphological analysis method is used to further refine the heat propagation boundary. The basic operations of morphological analysis include dilation, erosion, opening operation, and closing operation, etc. These operations can optimize the boundary of the heat source area, make it smoother, and eliminate isolated noise points. For example, in the analysis of heat loss in building walls, the boundary of the heat leakage area may have an irregular shape due to factors such as wind speed and material thickness. The erosion operation can remove misdetected isolated high-temperature points, while the dilation operation can fill in the missing parts of the boundary caused by the resolution limitation of the detection equipment. In addition, the opening operation can eliminate smaller pseudo-boundaries, and the closing operation can fill in small holes inside the heat source boundary, making the heat propagation area more in line with the actual situation.

[0115] After the preliminary morphological optimization is completed, it is necessary to further adjust the boundary in combination with the heat gradient information. Since the propagation path of heat diffusion is affected by the thermal conductivity, the propagation boundary cannot be simply delimited based on the temperature threshold, but should be calculated in combination with the thermal conductivity characteristics of the material. For example, in the thermal management of semiconductor chips, the thermal conductivities of different material layers inside the chip are different, and the heat diffusion paths are not evenly distributed. By analyzing the change of the heat gradient, it is possible to determine which areas have a faster temperature decay rate and delimit the heat propagation boundary accordingly, so as to ensure that the calculation area only covers the range actually affected by heat.

[0116] Finally, after obtaining the heat propagation boundary area, dynamic adjustment is still needed to adapt to the heat diffusion characteristics under different working conditions. In some cases, heat diffusion may be affected by changes in the external environment. For example, a change in wind speed will change the path of heat flow. In this case, a dynamic boundary definition strategy can be introduced, that is, combining real-time environmental parameters to adaptively adjust the heat propagation boundary. For example, in the monitoring of a fire scene, the heat diffusion boundary of the fire will change with the direction of air flow. Using the dynamic boundary definition method can more accurately predict the spread trend of the fire, thus providing a reliable basis for fire fighting decisions.

[0117] The improved optical flow algorithm mainly combines the heat distribution characteristics and temporal information on the basis of the traditional optical flow method to improve the calculation accuracy of the heat flow vector field. When calculating the pixel motion, the traditional optical flow algorithm usually relies on the brightness consistency assumption, that is, it is assumed that the pixel gray value of the image changes little in the time series. However, in infrared thermal imaging, due to the influence of heat convection, conduction, and environmental interference, the temperature gradient is often unstable, resulting in errors in the vector field calculated by the traditional optical flow method. Therefore, in the improvement process, a multi-scale feature extraction and adaptive weight adjustment mechanism are introduced to improve the calculation accuracy.

[0118] First, to adapt to the dynamic changes of infrared thermal images, the improved optical flow algorithm adopts a multi-scale Gaussian pyramid structure to reduce the influence of local temperature fluctuations. In thermal flow analysis, heat transfer usually occurs at multiple spatial scales, such as large-scale thermal convection phenomena and small-scale material conduction. Therefore, when calculating the temperature gradient change between adjacent frames, the infrared thermal image is first decomposed by the Gaussian pyramid, and the optical flow vectors are calculated layer by layer from low resolution to high resolution. At the low-resolution layer, the large-scale heat diffusion trend is mainly extracted, while at the high-resolution layer, it is used to capture local temperature details. This can ensure that the calculated thermal flow vector field can reflect the overall trend and does not lose key local thermal change information.

[0119] Secondly, when calculating the pixel-level temperature gradient change between adjacent frames, an adaptive weight adjustment mechanism based on thermal distribution characteristics is introduced. In standard optical flow calculation, the contribution weights of all pixel points are equal, but in infrared thermography, regions with larger temperature changes usually reflect thermal flow characteristics better than regions with stable temperatures. Therefore, in the improved algorithm, higher weights are assigned to high-gradient regions when calculating the optical flow vector, while for regions with smaller temperature changes, their influence on the overall calculation result is reduced. For example, in the thermal monitoring of industrial equipment, the temperature gradients of the heat dissipation parts and the core heating parts of the equipment are different. The traditional method treats all pixels equally, resulting in errors in the vector field. Through adaptive weight adjustment, the calculation result can be ensured to be more in line with the physical actual situation and the accuracy of the thermal flow vector field can be improved.

[0120] In the actual calculation process, the pixel-level temperature gradient change between adjacent frames is solved by the optical flow constraint equation. First, the temperature values at the same pixel position of two frames of images are obtained, and the temporal gradient, that is, the temperature change rate of the pixel point, is calculated. Then, combined with the spatial gradient information, that is, the change rates of temperature in the horizontal and vertical directions, a partial differential equation system is established. Since the applicability of the standard optical flow constraint equation in infrared thermography is poor, the improved algorithm uses the Farneback optical flow method combined with B-spline interpolation to optimize the temperature gradient change. The advantage of the Farneback optical flow method is that it can directly calculate dense optical flow vectors without edge detection, making the calculation of temperature gradient change smoother. And B-spline interpolation is used to enhance the continuity of the thermal flow vector field and avoid vector mutations caused by local temperature fluctuations.

[0121] In summary, this step ensures the accuracy of the thermal flow vector field calculation through a multi-level data processing method. First, determine the key regions of heat change to ensure that the selected regions cover all important parts affected by heat; secondly, use morphological analysis to limit the thermal propagation boundary so that the heat diffusion path conforms to the actual physical laws; finally, use the improved optical flow algorithm to calculate the spatio-temporal changes during the heat propagation process to form a preliminary thermal flow vector field.

[0122] In step S15, based on a preset noise filtering mechanism, abnormal vectors are removed from the preliminary heat flow vector field to obtain a filtered heat flow vector field, including:

[0123] Based on a gradient anomaly detection method, the preliminary heat flow vector field is subjected to anomaly identification to obtain an abnormal vector classification list;

[0124] Based on an adaptive smoothing filtering method, the abnormal vector classification list is subjected to abnormal vector rejection to obtain a denoised heat flow vector field;

[0125] Based on an interpolation reconstruction method, the denoised heat flow vector field is subjected to data filling to obtain a filtered heat flow vector field.

[0126] It should be noted that in this step, based on a preset noise filtering mechanism, abnormal vectors are removed from the preliminary heat flow vector field to obtain more reliable and stable heat flow vector field data. Since during the calculation of the heat flow vector, factors such as sensor noise, environmental interference, and calculation errors cause some vectors to deviate abnormally from the normal distribution, a systematic method needs to be adopted for identification, rejection, and filling to improve the stability and accuracy of the final vector field.

[0127] First, based on a gradient anomaly detection method, the preliminary heat flow vector field is subjected to anomaly identification to obtain an abnormal vector classification list. When calculating the heat flow vector field, the direction or magnitude of some vectors will deviate significantly from the heat flow conditions in their surrounding areas. For example, in areas with relatively uniform heat transfer, there are abnormal high or low temperature flow situations. These anomalies are caused by sensor noise, optical flow calculation errors, or local environmental changes and need to be screened by the gradient detection method. The specific operation method is to calculate the gradient change rate of each vector and compare it with the gradients of neighboring vectors. If the gradient change of a certain vector exceeds the set threshold, it is marked as abnormal. For example, when detecting building heat leakage using infrared thermal imaging, if the heat flow direction in a certain wall area shows a significant inconsistency with the adjacent area, it indicates that the measurement data at that location is abnormal and needs further processing. In addition, in industrial equipment monitoring, the heat flow vector field should present a stable flow pattern along the pipeline or material surface. If there is a sudden change somewhere, it is a data error and also indicates that there is an abnormality in the equipment.

[0128] Next, based on the adaptive smoothing filtering method, the abnormal vector classification list is used to eliminate abnormal vectors, resulting in a denoised heat flow vector field. When eliminating abnormal vectors, a smoothing filtering method is adopted to maintain the overall continuity of the vector field and avoid data mutations or incoherence caused by directly eliminating vectors. The core idea of adaptive smoothing filtering is to dynamically adjust abnormal vectors according to the distribution of surrounding vectors to reduce noise interference. For example, in a heat flow monitoring system, if most vectors in a region point in the same direction while the direction of a certain vector deviates significantly, the adaptive smoothing filtering method can be used to adjust the vector to a direction more consistent with the overall trend. This method can be applied to the monitoring of high-temperature equipment, such as the analysis of the heat distribution in furnaces in the metallurgical industry, to ensure that the calculated heat flow vector field can truly reflect the heat conduction inside the equipment.

[0129] Finally, based on the interpolation reconstruction method, data filling is performed on the denoised heat flow vector field to obtain the final filtered heat flow vector field. After eliminating abnormal vectors, there will be void regions in the vector field. To ensure data integrity, interpolation processing is required. Interpolation reconstruction methods mainly include linear interpolation, B-spline interpolation, etc. Among them, the B-spline interpolation method can fill the data missing regions more smoothly, making the reconstructed vector field more consistent with the true heat flow characteristics. For example, in medical diagnosis using thermal imaging, if there are breaks in the heat flow vector field due to abnormal data in some regions, the interpolation reconstruction method can be used to restore the heat flow information in these regions, enabling doctors to accurately judge the heat distribution in the lesion area. In addition, in the monitoring of semiconductor manufacturing processes, accurate calculation of heat flow is crucial for process optimization. If there are discontinuity problems in the vector field data, the interpolation reconstruction method can help fill in the key heat transfer paths to ensure the stability of the calculation results.

[0130] Generally speaking, this step ensures the stability and reliability of the heat flow vector field data through methods such as gradient anomaly detection, adaptive smoothing filtering, and interpolation reconstruction. First, through gradient anomaly detection, abnormal vectors are screened out and their characteristics are classified and recorded. Secondly, through the adaptive smoothing filtering method, abnormal data is eliminated while maintaining the smoothness of the vector field. Finally, through interpolation reconstruction, the data missing regions are filled to ensure the integrity of the heat flow vector field.

[0131] In step S16, the filtered heat flow vector field is input into a thermodynamic model to output a corrected heat flow vector field, including:

[0132] The input into the thermodynamic model is trained by a PINN neural network model;

[0133] Through the input layer of the thermodynamic model, vector feature generation is performed on the filtered heat flow vector field to obtain high-dimensional features of the heat flow vector.

[0134] Through the hidden layer of the thermodynamic model, physical constraints are imposed on the high-dimensional features of the heat flow vector to obtain the physically constrained heat flow vector features;

[0135] Through the output layer of the thermodynamic model, feature mapping is performed on the physically constrained heat flow vector features to obtain the corrected heat flow vector field.

[0136] It should be noted that in this step, the filtered heat flow vector field is input into the thermodynamic model to output the corrected heat flow vector field. This thermodynamic model uses a physics-guided neural network model, namely the PINN neural network model, to optimize the calculation accuracy of the vector field through physical constraints, so that the finally output heat flow vector field conforms to the laws of thermodynamics. The PINN neural network model combines data-driven methods and physical equation constraints during the training process, so that it can accurately predict the heat flow vector distribution under the condition of limited observation data. The main purpose of this step is to perform feature extraction, physical constraint, and mapping optimization in sequence through the input layer, hidden layer, and output layer of the model, so as to realize the correction of the preliminary heat flow vector field.

[0137] In the model training stage, it is first necessary to construct the structure of the PINN neural network and prepare the dataset for training. The training data mainly includes two parts. One part is the real heat flow vector samples obtained from the simulation data, and the other part is the physical constraint data calculated based on the thermodynamic equations. During the training process, the model needs to minimize both the data error loss and the physical equation constraint loss to ensure that the trained network can not only match the observed data but also conform to the physical laws. For example, in the heat flow direction prediction task, the data error loss is the deviation between the heat flow vector output by the network and the real heat flow vector, while the physical equation loss is the difference between the flow trend calculated based on the Fourier heat conduction equation and the model prediction result. By optimizing these two parts of the loss, the model can learn a heat flow vector field that conforms to both the observed data and the laws of thermodynamics.

[0138] In the model application stage, first, vector feature generation is performed on the filtered heat flow vector field through the input layer of the thermodynamic model. The role of the input layer is to preprocess the input data and map it to a high-dimensional feature space for subsequent calculations. Specifically, the input layer will extract the local temperature gradient, velocity field features, and spatial distribution patterns of the input vector field, and normalize these features to improve the stability of training. For example, in the high-temperature environment monitoring task, the input layer can extract the temperature gradient between different measurement points and estimate the heat diffusion direction and speed based on this gradient information.

[0139] Next, through the hidden layer of the thermodynamic model, physical constraint calculations are performed on the high-dimensional features of the heat flow vector. The main task of the hidden layer is to optimize the heat flow vector field through physical equations to make it conform to the laws of thermodynamics. This process uses the physical loss calculation method in PINN. That is, during network training, the hidden layer not only calculates the non-linear mapping relationship of neurons but also combines physical constraints such as the Fourier heat conduction equation and the energy conservation equation to adjust the heat flow vector. For example, in the task of optimizing the heat dissipation of semiconductor chips, the hidden layer can predict the heat dissipation path based on the surface temperature distribution of the chip and ensure that the heat dissipation flow direction conforms to the heat conduction law, avoiding unreasonable heat accumulation areas.

[0140] Finally, through the output layer of the thermodynamic model, feature mapping is performed on the physically constrained heat flow vector features to generate a corrected heat flow vector field. The role of the output layer is to adjust the vector field after physical optimization to make it closer to the real heat flow distribution and improve the robustness of the calculation. Specifically, the output layer uses non-linear transformation to map the input features, so that the finally output vector field is smoother in spatial distribution and conforms to the observed data. For example, in the intelligent building temperature control system, the output layer can automatically correct the heat flow path based on the temperature difference inside and outside the building to optimize energy consumption and improve the uniformity of indoor temperature.

[0141] Generally speaking, in this step, by inputting the filtered heat flow vector field into the PINN neural network model and through feature extraction in the input layer, physical constraint calculations in the hidden layer, and mapping optimization in the output layer, the correction of the heat flow vector field is realized, making it more conform to the laws of thermodynamics, and improving the calculation accuracy and stability. This method is applicable to a variety of application scenarios, such as temperature control optimization of industrial equipment, energy-saving analysis of intelligent buildings, and high-precision thermal imaging data processing, providing a reliable heat flow calculation method for related fields.

[0142] In step S17, according to the corrected heat flow vector field, heat distribution images are generated to obtain complete heat distribution images, including:

[0143] Based on the bilinear interpolation method, spatial reconstruction is performed on the corrected heat flow vector field, and smoothing processing is performed using the Fourier transform method to obtain a reconstructed heat flow vector field;

[0144] Based on the Gaussian weight mapping method, a temperature distribution matrix is generated for the reconstructed heat flow vector field to obtain a temperature distribution matrix;

[0145] According to the temperature distribution matrix, histogram equalization is performed to optimize the contrast to obtain a complete heat distribution image.

[0146] It should be noted that in this step, a complete thermal distribution image is generated based on the corrected heat flow vector field. This process includes spatial reconstruction, calculation of the temperature distribution matrix, and histogram equalization optimization to ensure that the finally obtained thermal distribution image can clearly and accurately reflect the spatial distribution of heat, improve the visualization effect of heat flow characteristics, and optimize the feature analysis ability of the infrared thermal imager.

[0147] First, spatial reconstruction is performed on the corrected heat flow vector field based on the bilinear interpolation method, and smoothing processing is carried out using the Fourier transform method to ensure the continuity of heat flow data in space. In the infrared thermal imaging system, the heat flow vector field is calculated from the thermal distribution data at multiple time points. However, due to the limited resolution of the acquisition device or data calculation errors, there are problems of missing data or sudden changes in temperature gradients in some areas. The bilinear interpolation method can effectively solve this problem, making the transition of the vector field smoother in space. Specifically, when the thermal imager monitors complex heat sources (such as heat transfer on the surface of industrial equipment), there are phenomena of data breaks or large temperature differences in some areas. Through the bilinear interpolation method, the missing areas can be compensated and calculated based on the temperature values of adjacent pixels, thereby obtaining smooth and continuous thermal distribution information. In addition, to further optimize the data after spatial reconstruction, the Fourier transform is used for smoothing processing to eliminate spatial high-frequency noise and improve the stability of heat distribution. When detecting the heat flow characteristics of precision instruments, this method can effectively reduce local anomalies caused by measurement noise, making the heat transfer path clearer.

[0148] Secondly, based on the Gaussian weight mapping method, a temperature distribution matrix is generated for the reconstructed heat flow vector field. The temperature distribution matrix is an important part of the thermal distribution image, which can directly reflect the spatial distribution of heat. Since the transfer of heat in space is affected by the thermal conductivity of different materials, the change of the temperature gradient is not uniform. Therefore, when constructing the temperature distribution matrix, it is necessary to comprehensively consider the heat flow direction and the temperature change of adjacent areas. The Gaussian weight mapping method is a weighted calculation method based on the Gaussian distribution function. It is mainly used to smooth the data distribution, emphasize the influence of local areas, and retain global information at the same time. In the process of generating the temperature distribution matrix, the core idea of the Gaussian weight mapping method is to calculate a weighted average based on the temperature value of the heat source point and the temperature information of adjacent areas to more reasonably reflect the spatial diffusion of heat. The characteristic of the Gaussian distribution is that the weight of the central area is higher and gradually decreases towards the periphery, making the influence degree of the high-temperature area on its surrounding pixels conform to the physical conduction law.

[0149] In specific applications, it is first necessary to determine the initial temperature values of each pixel point in the heat flow vector field. Since the diffusion of heat is affected by factors such as the thermal conductivity of the material, the ambient temperature, and the boundary conditions, there may be significant differences in the temperature gradients of adjacent pixel points. To reasonably construct the temperature distribution matrix, it is necessary to establish a weighted window around each pixel point and use the Gaussian distribution function to calculate the contribution degree of each pixel point. Usually, the weight is calculated based on the Euclidean distance between the pixel point and the center point. The closer the pixel point is to the center point, the greater the weight, and vice versa. For example, in the thermal imaging detection of building walls, the heat diffusion on the wall surface is not uniform, and the thermal conductivity of some materials is low, resulting in a slower change in the local temperature gradient. If the linear interpolation method is directly used to generate the temperature matrix, it may lead to distortion of the temperature distribution. After adopting the Gaussian weight mapping method, the heat can be more reasonably distributed, making the heat diffusion pattern of the wall conform to the actual situation and improving the accuracy of building energy conservation analysis.

[0150] Secondly, the Gaussian weight mapping method can also be used to reduce measurement errors and improve the smoothness of the heat distribution image. In infrared thermal imaging applications, due to equipment noise, environmental interference, and sensor accuracy limitations, the thermal images collected may have locally unstable temperature points. Directly using the original temperature data may result in mutation points in the temperature distribution matrix, making the heat flow calculation results unstable. Through the Gaussian weight mapping method, these mutation points can be smoothed, making the temperature distribution more uniform. For example, in the inspection of power equipment, the thermal imaging detection of high-voltage cables usually requires analyzing the diffusion of temperature hot spots. If the temperature data in some areas are abnormal due to external interference, the Gaussian weight mapping can smooth these abnormal points, making the diffusion pattern of the temperature hot spots more in line with the actual heat dissipation characteristics of the cable.

[0151] After the temperature distribution matrix is generated, the Gaussian weight mapping method can also be used to optimize the temperature gradient calculation to ensure the rationality of the heat diffusion trend. Traditional temperature gradient calculation methods usually use differential operations, but this method is easily affected by noise, resulting in discontinuous temperature gradients and affecting the calculation of heat flow vectors. After adopting the Gaussian weight mapping, more neighborhood information can be considered when calculating the temperature gradient, making the gradient calculation results smoother and more continuous.

[0152] Finally, based on the calculated temperature distribution matrix, histogram equalization is performed to optimize the contrast. Histogram equalization is a technique used to enhance the contrast of an image. Its basic principle is to redistribute the gray levels in the image so that the pixel value distribution is more uniform, thereby enhancing the detail information in low-contrast images. In the application of infrared thermal imagers, the original thermal distribution image may have a concentrated gray level distribution in a relatively narrow range due to a small temperature difference, making it difficult to clearly distinguish areas with small thermal gradients. For example, during the thermal monitoring of industrial equipment, the temperature anomaly of some components may only be a few degrees higher than during normal operation. However, due to the narrow overall temperature range, it is difficult to detect these small differences directly by observing the image. Through histogram equalization, these small temperature gradient changes can be stretched to a wider gray range, making the temperature differences more obvious and improving the identifiability of abnormal points.

[0153] The specific steps for optimizing the contrast by histogram equalization include calculating the histogram of the original thermal distribution image, constructing the cumulative distribution function, and using the cumulative distribution function to map the pixel gray values. First, it is necessary to count the number of pixels at each gray level in the thermal distribution image to obtain the gray histogram of the image. For example, in an infrared thermal image, if most of the pixel gray levels are concentrated in a lower temperature range and there are fewer pixels in the higher temperature region, then the overall contrast of this image is low, and the difference between the high and low temperature regions is not obvious. Next, by calculating the cumulative distribution function, the new gray value corresponding to each gray level is determined, so that the pixel values are evenly distributed across the entire gray range. This conversion process can enhance the originally low-contrast regions, making the temperature differences more prominent and improving the visualization effect of the thermal distribution image. For example, in building heat loss detection, the heat bridge effect inside the wall will cause subtle changes in the temperature distribution. The untreated infrared thermal image may not be able to clearly show these temperature gradients. After optimization by histogram equalization, the regions with small temperature differences are enhanced, making the energy leakage points inside the wall more obvious, thereby improving the accuracy of building energy efficiency analysis.

[0154] During the optimization process of the thermal distribution image, histogram equalization not only improves the contrast but also reduces background interference, making the heat source region more prominent. For some scenarios with a relatively uniform background temperature, such as the thermal inspection of power equipment, the temperature change in the background region is small, and the abnormally heated points may only account for a small part of the image, resulting in a limited gray distribution in the overall thermal image and making it difficult to detect the abnormal points. Through histogram equalization, the contrast of these local high-temperature regions can be enhanced.

[0155] In addition, histogram equalization can also play an optimization role when processing infrared thermal images with low signal-to-noise ratio. Since signal noise may occur in the infrared imager in a low temperature difference environment, these noises may affect the overall quality of the image, making the temperature gradient change not smooth enough. Through equalization processing, the visual error caused by uneven gray distribution can be reduced, making the heat flow characteristics clearer. For example, in medical infrared thermal imaging diagnosis, certain skin diseases or vascular abnormalities may cause slightly higher local temperatures, but due to the small overall temperature difference, the lesion area may not be detected by directly observing the original image. Through histogram equalization, these areas with small temperature differences can be highlighted, improving the accuracy of diagnosis and providing a more intuitive basis for lesion analysis for doctors.

[0156] In summary, in this step, spatial reconstruction is performed by the bilinear interpolation method, smoothing is performed by the Fourier transform method, a temperature distribution matrix is generated by the Gaussian weight mapping method, and the contrast is optimized by histogram equalization, realizing the generation of a complete thermal distribution image.

[0157] In step S18, based on the complete thermal distribution image, heat flow feature reconstruction is performed to obtain a high-precision heat flow vector field, including:

[0158] Based on the time series analysis method, for the complete thermal distribution image, the pixel-level temperature change trend is extracted to obtain preliminary heat flow trajectory data;

[0159] Based on the Kalman filtering method, for the preliminary heat flow trajectory data, data smoothing is performed and features are generated to obtain high-precision heat flow vector features;

[0160] Based on the Farneback optical flow method combined with the B-spline interpolation method, for the high-precision heat flow vector features, optical flow optimization and interpolation correction are performed to obtain a high-precision heat flow vector field.

[0161] It should be noted that in step S18, based on the complete thermal distribution image, heat flow feature reconstruction is performed, and finally a high-precision heat flow vector field is obtained. The core of this process is to extract heat flow trajectory data based on the time series analysis method and use the Kalman filtering method to smooth and optimize the data to reduce the noise interference of the heat flow vector. Subsequently, the Farneback optical flow method is combined to calculate the heat flow direction and speed, and finally the continuity of the vector field is optimized by the B-spline interpolation method to improve the accuracy of the calculation results.

[0162] First, based on the time series analysis method, the temperature change trend of each pixel in the complete thermal distribution image is extracted. By analyzing the temperature changes between different time frames, this method constructs preliminary thermal flow trajectory data. In the application of infrared thermal imagers, the thermal distribution image consists of multiple time series images, and each frame records the temperature distribution information at a specific time point. By tracking the temperature change trend of each pixel at different time points, the preliminary direction of thermal flow can be deduced. For example, in the detection of motor winding heating, by continuously collecting infrared thermal images of the motor during operation, it can be found how the heat gradually spreads along the winding, thereby determining whether there are abnormal heating areas in the motor. During the reconstruction of thermal flow characteristics, the time series analysis method can effectively extract these change trends and convert them into preliminary thermal flow trajectory data.

[0163] Second, based on the Kalman filter method, the preliminary thermal flow trajectory data is smoothed and feature generation is performed. Since the data collected by the infrared thermal imager is affected by factors such as ambient temperature fluctuations and sensor accuracy limitations, there will be local abnormal data points, resulting in jitter or deviation in the thermal flow trajectory. The Kalman filter is a method that can dynamically predict and update state estimates and is suitable for processing time series data with large noise interference. In this step, the Kalman filter is used to smooth the thermal flow trajectory data, removing unreasonable data points with drastic temperature changes in a short period of time, and improving the stability of the thermal flow vector calculation. For example, in the temperature rise monitoring of large transformers, by recording the temperature changes of different parts of the transformer with an infrared thermal imager, if the temperature of a certain point fluctuates drastically in a short period of time, the Kalman filter method can, through the analysis of historical temperature data, reasonably predict the true temperature change trend of this point, remove abnormal noise, and ensure the smoothness and accuracy of the thermal flow trajectory.

[0164] Next, based on the Farneback optical flow method, the optical flow optimization of the high-precision thermal flow vector features is carried out. The Farneback optical flow method is a dense optical flow calculation method that can calculate the motion direction and speed of each pixel point based on the pixel changes between adjacent frames. In the application scenario of infrared thermal imagers, the Farneback optical flow method can be used to analyze how heat spreads on the surface of an object and calculate the direction of thermal flow. For example, in the thermal analysis of high-power electronic components, the infrared thermal imager can collect the temperature distribution on the surface of the components, and calculate the thermal flow vector through the Farneback optical flow, thereby determining whether the heat dissipation design is reasonable. If heat accumulates in a local area, it indicates that the heat dissipation efficiency of this area is low, and it is necessary to optimize the heat dissipation material or increase the heat dissipation channels. Through optical flow calculation, the thermal flow path can be accurately obtained, providing data support for the thermal management of electronic components.

[0165] Finally, based on the B-spline interpolation method, the thermo-flow vector field optimized by optical flow is interpolated and corrected to improve the smoothness and accuracy of the final result. The B-spline interpolation method is a mathematical method for smooth curve fitting and is widely used in computer graphics, signal processing, and data compensation. In the processing of the thermo-flow vector field in infrared thermal imaging, B-spline interpolation can effectively solve the problem of discontinuity of the thermo-flow vector field caused by discrete pixel calculations, making the trend of heat flow smoother and conforming to the actual heat propagation characteristics. B-spline interpolation constructs multiple low-order polynomial segments and introduces control points between each segment, making the overall curve or data fitting result more natural and having good differentiability and local control characteristics.

[0166] In the specific calculation process, B-spline interpolation first defines a set of discrete control points, which are determined by the thermo-flow vector data obtained from the preliminary calculation. Then, based on these control points, B-spline basis functions are constructed. The basis functions are piecewise polynomials, usually cubic B-spline functions, that is, each interpolation segment is jointly determined by four adjacent control points, making the entire curve have continuous first and second derivatives. In the processing of the thermo-flow vector field, this means that the heat propagation direction and speed changes will be smoother, avoiding unreasonable vector mutations. For example, in the detection of the thermal insulation layer of a spacecraft, there may be mutation points in the thermo-flow vector field in some areas due to material properties, which may be caused by sensor errors or local anomalies. In this case, the B-spline interpolation method can smooth these mutation areas, making the change of the heat flow vector field in space more natural and conforming to physical laws.

[0167] In summary, in step S18, the preliminary thermo-flow trajectory data is extracted by the time series analysis method, the data is smoothed by the Kalman filtering method, then the thermo-flow vector is calculated by the Farneback optical flow method, and finally the B-spline interpolation method is used for optimization, so that the finally generated high-precision thermo-flow vector field has good continuity and accuracy.

[0168] The working process of the present invention is described below with a relatively common scenario as an example. Please also refer to Figure 2 which is Figure 1 a schematic diagram of the working scenario of the method.

[0169] In actual application scenarios, such as the temperature monitoring of industrial equipment, some equipment may experience local overheating due to uneven heat dissipation or internal component failures during long-term operation, which will affect the normal operation of the equipment and even cause safety problems. The present invention provides a high-precision infrared thermal feature analysis method. Through multi-step image processing and calculation, heat flow information is obtained, thereby realizing the accurate monitoring of the temperature state of the equipment and the analysis of the heat flow trend. The following is combined with Figure 1The method steps shown and Figure 2 the system modules shown are used to illustrate the actual working process of the present invention.

[0170] In step S11, the data acquisition module is used to collect infrared thermal images of multiple focal planes. The infrared thermal imager is installed in the monitoring area and captures the target device at different focal lengths to obtain infrared thermal images of multiple focal planes, ensuring the integrity of image clarity and depth information. For example, in the application of industrial equipment monitoring, for a running motor or transformer, multi-focal plane imaging can capture the subtle changes in the surface temperature of the device and the surrounding air temperature, providing rich raw data for subsequent analysis.

[0171] In step S12, the infrared thermal image module preprocesses the collected multi-focal plane infrared thermal images. This module applies an adaptive filtering algorithm to remove noise from the images and uses an enhancement algorithm to improve the contrast, highlighting the temperature gradient change characteristics. For example, when monitoring high-temperature equipment, affected by environmental light, smoke, or external heat sources, noise may appear in the infrared thermal images, affecting temperature recognition. Through adaptive filtering, interference factors can be eliminated, and the enhancement algorithm can improve the image clarity, making the temperature difference more obvious and facilitating subsequent processing.

[0172] In step S13, the heat source determination module identifies the heat source area in the enhanced infrared thermal image. This step combines heat radiation intensity analysis, gradient calculation, and edge detection algorithms to accurately locate the temperature anomaly area. For example, when monitoring industrial boilers or heating pipelines, heat source detection can identify surface hot spots, clarify the temperature distribution range, and determine the specific shape of the heat source, such as an ellipse or an area with an irregular contour, ensuring accurate positioning of the target area and providing basic data for subsequent heat flow analysis.

[0173] In step S14, the preliminary heat flow vector field module calculates the heat flow trend based on the accurate positioning data of the heat source area. This step uses an improved optical flow algorithm to analyze the heat flow situation by calculating the pixel-level temperature gradient change between adjacent frames. For example, when monitoring high-power electronic devices, this method can identify the changes in the surface temperature gradient of the device, reveal the uneven heat dissipation phenomenon, and calculate the path of heat diffusion from the hot spot area to the cooling area, thus providing key data on the heat dissipation performance of the device.

[0174] In step S15, the noise filtering module is used to remove abnormal vectors from the preliminary heat flow vector field. This step adopts a preset noise filtering mechanism, including gradient anomaly detection, adaptive smoothing filtering, and interpolation reconstruction methods, to improve the reliability of the heat flow vector. For example, in an open production environment, due to factors such as air convection and fan disturbance, short-term temperature fluctuations will occur, interfering with the calculation of the heat flow trend. Through the gradient anomaly detection method, abnormal vectors that do not conform to physical laws can be identified and removed, thus ensuring the stability of the data.

[0175] In step S16, the heat flow vector field optimized by noise is input into the thermodynamic model for correction. The present invention adopts a physics-informed neural network (PINN) model, which combines physical equations and data-driven learning, enabling the corrected heat flow vector field to more accurately reflect the actual heat flow state. For example, in the heat dissipation analysis of complex industrial equipment, the PINN model can correct the heat flow vector based on the heat conduction equation and adjust the weights in combination with historical data, making the output heat flow vector field more in line with the thermal characteristics of the equipment.

[0176] In step S17, the heat distribution image module generates a complete heat distribution image based on the corrected heat flow vector field. This step adopts the bilinear interpolation method for spatial reconstruction and combines Fourier transform for smoothing processing to ensure the continuity of the heat map. For example, when monitoring the surface temperature of a metallurgical furnace, this step can generate a smoother and artifact-free temperature distribution map, visualizing the temperature field changes inside the furnace and facilitating operators to judge the furnace temperature state.

[0177] In step S18, the heat flow vector field module optimizes the complete heat distribution image to obtain a high-precision heat flow vector field. This step adopts time series analysis methods to extract the pixel-level temperature change trend, combines Kalman filtering for data smoothing, and then applies the Farneback optical flow method combined with B-spline interpolation technology to optimize and correct the heat flow vector. For example, in the thermal monitoring of aeroengine turbine blades, this method can identify the heat diffusion caused by air flow and improve the monitoring accuracy of the blade heat dissipation condition through the optimized high-precision heat flow vector field.

[0178] In summary, the present invention realizes a complete process from data acquisition to the generation of a high-precision heat flow vector field through multiple steps. Compared with traditional static heat maps, this method can provide more accurate temperature flow information, contributing to multiple application scenarios such as industrial equipment monitoring and the optimization of electronic component heat dissipation, and improving the safety and stability of equipment operation.

[0179] Referring to Figure 2 , the second embodiment of the present invention provides a feature analysis system for an infrared thermal imager, including:

[0180] A data acquisition module for acquiring infrared thermal images of multiple focal planes;

[0181] An infrared thermal image module for removing noise and enhancing the contrast of the infrared thermal images of the multiple focal planes based on an adaptive filtering and enhancement algorithm to obtain enhanced infrared thermal images;

[0182] A heat source determination module for determining heat sources in the enhanced infrared thermal images based on an edge detection algorithm to obtain precise positioning data of the heat source regions;

[0183] A preliminary thermal flow vector field module for extracting the heat source influence regions from the precise positioning data of the heat source regions based on an improved optical flow algorithm to obtain a preliminary thermal flow vector field;

[0184] A noise filtering module for removing abnormal vectors from the preliminary thermal flow vector field based on a preset noise filtering mechanism to obtain a filtered thermal flow vector field;

[0185] A thermodynamics module for inputting the filtered thermal flow vector field into a thermodynamics model and outputting a corrected thermal flow vector field;

[0186] A thermal distribution image module for generating a thermal distribution image based on the corrected thermal flow vector field to obtain a complete thermal distribution image;

[0187] A thermal flow vector field module for reconstructing thermal flow characteristics based on the complete thermal distribution image to obtain a high-precision thermal flow vector field.

[0188] It should be noted that a feature analysis system for an infrared thermal imager provided in an embodiment of the present invention is used to execute all the process steps of a feature analysis method for an infrared thermal imager in the above embodiment. The working principles and beneficial effects of the two correspond one by one, so they will not be elaborated here.

[0189] An embodiment of the present invention also provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a feature analysis program for an infrared thermal imager. When the processor executes the computer program, it implements the steps in the above-mentioned embodiments of the feature analysis method for an infrared thermal imager, such as Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-mentioned device embodiments, such as the thermal flow vector field module.

[0190] Exemplarily, the computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0191] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0192] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects various parts of the entire electronic device through various interfaces and lines.

[0193] The memory can be used to store the computer program and / or module. By running or executing the computer program and / or module stored in the memory, and invoking the data stored in the memory, the processor realizes various functions of the electronic device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, the image playback function, etc.); the data storage area can store the data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0194] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device that can carry the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0195] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationships between the modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0196] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. In particular, for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A feature analysis method for an infrared thermal imager, characterized in that, Including: Obtaining infrared thermal images of multiple focal planes; Based on an adaptive filtering and enhancement algorithm, performing noise removal on the infrared thermal images of the multiple focal planes and enhancing the contrast to obtain enhanced infrared thermal images; Based on an edge detection algorithm, determining heat sources for the enhanced infrared thermal images to obtain precise positioning data of the heat source regions; Based on an improved optical flow algorithm, extracting the heat source influence regions from the precise positioning data of the heat source regions to obtain a preliminary thermal flow vector field; Wherein, the improved optical flow algorithm includes introducing a multi-scale feature extraction mechanism and an adaptive weight adjustment mechanism; the multi-scale feature extraction mechanism includes using a multi-scale Gaussian pyramid structure to perform Gaussian pyramid decomposition on the infrared thermal images, calculating optical flow vectors layer by layer from low resolution to high resolution, extracting large-range heat diffusion trends at the low resolution layer, and capturing local temperature details at the high resolution layer; the adaptive weight adjustment mechanism includes assigning higher weights to high-gradient regions during the calculation of optical flow vectors; the algorithm further combines the Farneback optical flow method with B-spline interpolation to optimize the calculation results of temperature gradient changes; Based on a preset noise filtering mechanism, removing abnormal vectors from the preliminary thermal flow vector field to obtain a filtered thermal flow vector field; Inputting the filtered thermal flow vector field into a preset thermodynamic model to output a corrected thermal flow vector field, including: The thermodynamic model is obtained by training a PINN neural network model; Through the input layer of the thermodynamic model, generating vector features for the filtered thermal flow vector field to obtain high-dimensional features of the thermal flow vector; Through the hidden layer of the thermodynamic model, performing physical constraints on the high-dimensional features of the thermal flow vector to obtain physically constrained thermal flow vector features; Through the output layer of the thermodynamic model, performing feature mapping on the physically constrained thermal flow vector features to obtain a corrected thermal flow vector field; Wherein, in the training stage, the PINN neural network model includes constructing a PINN neural network structure and preparing a dataset for training, and the dataset includes real thermal flow vector samples obtained from simulation data and physical constraint data calculated based on thermodynamic equations; During the training process, the model simultaneously minimizes the data error loss and the physical equation constraint loss so that the model output not only matches the observed data but also conforms to the thermodynamic laws; The input layer extracts local temperature gradients, velocity field features, and spatial distribution patterns in the filtered thermal flow vector field and performs normalization processing to generate high-dimensional features; The hidden layer combines the Fourier heat conduction equation and the energy conservation equation to perform non-linear physical constraint optimization on the high-dimensional features to generate physically consistent vector features; The output layer performs non-linear mapping on the physically constrained features and outputs a corrected thermal flow vector field; Generating a heat distribution image based on the corrected thermal flow vector field to obtain a complete heat distribution image, including: Based on the bilinear interpolation method, perform spatial reconstruction on the corrected heat flow vector field, and use the Fourier transform method for smoothing to obtain the reconstructed heat flow vector field; Based on the Gaussian weight mapping method, generate a temperature distribution matrix for the reconstructed heat flow vector field to obtain a temperature distribution matrix; According to the temperature distribution matrix, perform histogram equalization to optimize the contrast and obtain a complete heat distribution image; According to the complete heat distribution image, perform heat flow feature reconstruction to obtain a high-precision heat flow vector field.

2. The feature analysis method for an infrared thermal imager according to claim 1, wherein The method for removing noise and enhancing the contrast of the multi-focal plane infrared thermal image based on the adaptive filtering and enhancement algorithm includes: Perform normalization processing on the multi-focal plane infrared thermal image to obtain a multi-focal plane infrared thermal image with equalized contrast; Based on the adaptive filtering algorithm, remove noise from the multi-focal plane infrared thermal image with equalized contrast to obtain a denoised infrared thermal image; Based on the enhancement algorithm, perform local contrast optimization on the denoised infrared thermal image to obtain an enhanced infrared thermal image.

3. The feature analysis method for an infrared thermal imager according to claim 1, characterized in that The method for determining the heat source of the enhanced infrared thermal image based on the edge detection algorithm to obtain precise positioning data of the heat source area includes: Perform heat radiation screening on the enhanced infrared thermal image to obtain a preliminary heat source distribution map; Based on the edge detection algorithm, refine the contour of the preliminary heat source distribution map to obtain a complete heat source distribution area; Based on the morphological analysis and region growing algorithm, screen and optimize the complete heat source distribution area to obtain precise positioning data of the heat source area; The precise positioning data of the heat source area includes: heat source center coordinates, heat source contour shape, and heat distribution feature map.

4. The feature analysis method for an infrared thermal imager according to claim 3, characterized in that, The method for extracting the heat source influence area from the precise positioning data of the heat source area based on the improved optical flow algorithm to obtain a preliminary heat flow vector field includes: Based on the heat source center coordinates, determine the key area of heat change to obtain the heat source influence area; Based on the morphological analysis method combined with the heat source contour shape, define the heat propagation boundary for the heat source influence area to obtain a heat propagation boundary area; Based on the improved optical flow algorithm combined with the heat distribution feature map, perform temporal analysis on the heat propagation boundary area and calculate the pixel-level temperature gradient change between adjacent frames to obtain a preliminary heat flow vector field.

5. The feature analysis method for an infrared thermal imager according to claim 1, characterized in that, The method for removing abnormal vectors from the preliminary heat flow vector field based on a preset noise filtering mechanism to obtain a filtered heat flow vector field includes: Based on the gradient anomaly detection method, perform anomaly recognition on the preliminary heat flow vector field to obtain a list of anomaly vector classifications; Based on the adaptive smoothing filtering method, remove the abnormal vectors from the list of anomaly vector classifications to obtain a denoised heat flow vector field; Based on the interpolation reconstruction method, perform data filling on the denoised heat flow vector field to obtain a filtered heat flow vector field.

6. The characteristic analysis method for an infrared thermal imager according to claim 1, wherein The method for performing heat flow feature reconstruction according to the complete heat distribution image to obtain a high-precision heat flow vector field includes: Based on the time series analysis method, extract the pixel-level temperature change trend from the complete thermal distribution image to obtain preliminary thermal flow trajectory data; Based on the Kalman filtering method, smooth the preliminary thermal flow trajectory data and generate features to obtain high-precision thermal flow vector features; Based on the Farneback optical flow method combined with the B-spline interpolation method, optimize the optical flow and perform interpolation correction on the high-precision thermal flow vector features to obtain a high-precision thermal flow vector field.

7. A feature analysis system for an infrared thermal imager, characterized in that, Used to implement the feature analysis method for an infrared thermal imager according to any one of claims 1 to 6, including: A data acquisition module for acquiring infrared thermal images of multiple focal planes; An infrared thermal image module for removing noise and enhancing the contrast of the infrared thermal images of multiple focal planes based on an adaptive filtering and enhancement algorithm to obtain enhanced infrared thermal images; A heat source determination module for determining the heat source based on an edge detection algorithm for the enhanced infrared thermal images to obtain precise positioning data of the heat source area; A preliminary thermal flow vector field module for extracting the heat source influence area from the precise positioning data of the heat source area based on an improved optical flow algorithm to obtain a preliminary thermal flow vector field; A noise filtering module for removing abnormal vectors from the preliminary thermal flow vector field based on a preset noise filtering mechanism to obtain a filtered thermal flow vector field; A thermodynamics module for inputting the filtered thermal flow vector field into a thermodynamics model and outputting a corrected thermal flow vector field; A thermal distribution image module for generating a thermal distribution image based on the corrected thermal flow vector field to obtain a complete thermal distribution image; A thermal flow vector field module for reconstructing the thermal flow characteristics based on the complete thermal distribution image to obtain a high-precision thermal flow vector field.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the feature analysis method for an infrared thermal imager according to any one of claims 1 to 6.

Images