Method and System for Online Temperature Monitoring of Power Equipment Using Infrared Images

By combining deep learning image segmentation and environmental parameter compensation, the problems of fixed emissivity and atmospheric attenuation in infrared thermometry methods are solved, enabling high-precision automated online monitoring and fault location of power equipment, and reducing false alarm rate and missed alarm rate.

CN123111480APending Publication Date: 2026-11-17SHUOZHOU CITY PINGLU DISTRICT WOLONG WIND POWER CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611237368.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing infrared thermometry methods are difficult to achieve high-precision, automated online monitoring and early fault location of power equipment in complex outdoor environments, mainly due to the lack of systematic compensation for differences in emissivity of equipment components and atmospheric transmission attenuation, as well as the limitation of relying on manual interpretation.

Method used

By adopting a technical approach that combines deep learning image segmentation with dynamic compensation of environmental parameters, the system acquires infrared images and environmental parameters to perform noise suppression, equipment segmentation, component identification, radiative transfer compensation, and temperature inversion, thereby generating an accurate temperature field for equipment components and automatically calibrating abnormal areas.

Benefits of technology

It enables accurate temperature inversion of various components of power equipment and automatic identification of abnormal electrical conditions, improving the automation level of online monitoring and the pertinence and reliability of fault diagnosis, and reducing the false alarm rate and missed alarm rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111480A_ABST
    Figure CN123111480A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for online temperature monitoring of power equipment based on infrared images, belonging to the field of power equipment monitoring technology. It includes: acquiring infrared image sequences of the power equipment area and ambient temperature, relative humidity, and wind speed; suppressing and enhancing noise in the infrared images to obtain enhanced images; segmenting the enhanced images to obtain the main body of the equipment, identifying component categories and masks, and extracting the radiance of corresponding pixels in the masks to obtain the original radiance map; retrieving emissivity from an emissivity mapping library according to category; calculating atmospheric transmittance and process radiation using environmental parameters, and inversely deducing radiative transmission using emissivity to eliminate attenuation and reflection, obtaining a compensated radiation map; converting the compensated radiation and emissivity into temperature to construct a component temperature field; fusing the various temperature fields to obtain an overall distribution map, and labeling anomalies according to the tolerance threshold of each component. By employing a technical approach combining deep learning image segmentation and dynamic compensation of environmental parameters, it is possible to achieve automatic identification of temperature inversion and electrical anomalies of various components of power equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of electrical measurement and fault diagnosis technology, and in particular to a method and system for online condition monitoring and temperature analysis of power equipment using infrared thermal imaging technology. Background Technology

[0002] Electrical equipment forms the material basis for the stable operation of a power system, and its operational reliability directly affects the safety of the entire power grid. Infrared thermal imaging, as a non-contact temperature measurement technology, can visually present the temperature distribution on the surface of equipment in image form, and is therefore widely used in the inspection and maintenance of power equipment. Abnormal temperature rises on the surface of equipment are usually a direct external manifestation of electrical faults such as poor electrical connections, equipment overload, or insulation degradation. Therefore, accurate measurement of equipment temperature helps in the early warning and diagnosis of electrical faults.

[0003] In current technological practices, infrared inspection of power equipment typically relies on inspectors using handheld infrared thermal imagers to take on-site photos, or on fixed cameras to acquire infrared images. Operators then identify hotspots by observing color differences in the infrared images. Some detection systems allow users to manually input a uniform surface emissivity parameter and observation distance; the software then estimates the temperature of a specific point or selected area in the image based on this, serving as a basis for determining whether the equipment exhibits any abnormalities.

[0004] However, the aforementioned detection methods have inherent limitations in practical applications. Infrared radiation is affected by atmospheric absorption and scattering during transmission through the air, and its attenuation is closely related to ambient temperature, humidity, wind speed, and observation distance. Existing methods often neglect systematic compensation for these dynamic environmental factors. Secondly, electrical equipment is typically composed of various materials such as metals, ceramics, and painted surfaces, each with different surface emissivity. Using a single emissivity value to calculate the temperature of the entire equipment introduces inherent errors in the temperature measurement of different components. Furthermore, the analysis of thermal images mainly relies on manual interpretation, which is not only inefficient but also highly dependent on the operator's professional experience. For equipment with complex structures and numerous components, it is difficult to quickly and accurately locate specific faulty components. In summary, existing infrared thermometry methods, in complex outdoor environments, lack systematic compensation for differences in emissivity of equipment components and atmospheric transmission attenuation, and rely on manual interpretation, making it difficult to achieve high-precision, automated online equipment monitoring and early fault location. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method and system for online temperature monitoring of power equipment based on infrared images. It employs a technical approach that combines deep learning image segmentation with dynamic compensation of environmental parameters, enabling accurate temperature inversion of various components of power equipment and automatic identification of abnormal electrical conditions.

[0006] The above objectives can be achieved through the following approach:

[0007] This invention relates to an online temperature monitoring method and system for power equipment based on infrared images, belonging to the field of power equipment monitoring technology. It includes: acquiring infrared image sequences of the power equipment area and ambient temperature, relative humidity, and wind speed; suppressing and enhancing noise in the infrared images to obtain enhanced images; segmenting the enhanced images to obtain the main body of the equipment, identifying component categories and masks, and extracting the radiance of corresponding pixels in the masks to obtain the original radiance map; retrieving emissivity from an emissivity mapping library according to category; calculating atmospheric transmittance and path radiation using environmental parameters, and inversely extrapolating radiative transfer using emissivity to eliminate attenuation and reflection, obtaining a compensated radiation map; converting the compensated radiation and emissivity into temperature to construct a component temperature field; fusing the various temperature fields to obtain an overall distribution map, and labeling anomalies according to the tolerance thresholds of each component.

[0008] Optionally, a continuous sequence of infrared images of the area where the power equipment is located is acquired, along with synchronously collected environmental meteorological parameters, including at least ambient temperature, relative humidity, and wind speed.

[0009] The thermal radiation signals of power equipment are acquired by an infrared thermal imager at a preset frame rate to form an infrared image sequence and an infrared image of the infrared image sequence.

[0010] The environmental parameter acquisition unit synchronously collects air temperature, relative humidity, and wind speed, and records environmental meteorological parameters.

[0011] Optionally, noise suppression and detail enhancement processing are performed on the infrared images in the infrared image sequence to generate enhanced infrared images, including:

[0012] Non-uniformity correction is performed on single-frame infrared images in the infrared image sequence to eliminate fixed-pattern noise introduced by differences in detector response;

[0013] The corrected image is subjected to joint spatiotemporal filtering to suppress random noise and preserve device edges;

[0014] An enhanced infrared image is obtained by adaptive histogram stretching of the filtered image.

[0015] Optionally, the enhanced infrared image is segmented into a device region to obtain a device region image containing the main body of the power equipment, including:

[0016] The enhanced infrared image is input into a preset device segmentation model, and a probability map corresponding to the main body of the power equipment is output.

[0017] The probability map is binarized to obtain the device body mask;

[0018] The device area is extracted from the enhanced infrared image using the device body mask to obtain a device area image.

[0019] Optionally, in the device area image, component category labels and component mask areas for each component constituting the power equipment are determined, and the pixel radiance corresponding to the component mask areas is extracted from the enhanced infrared image to generate an original radiance map of the components, including:

[0020] The device area image is segmented into instances to distinguish different component instances, and the category label and component mask area of ​​each component instance are obtained.

[0021] For each component mask region, the original radiance value of each pixel located within the corresponding component mask region is extracted from the enhanced infrared image;

[0022] The original radiance values ​​of all pixels within the component mask area are combined into a matrix to form the original radiance map of the corresponding component.

[0023] Optionally, based on the component category label, the surface emissivity value matching the component category label is retrieved from a preset component emissivity mapping library to obtain the component emissivity, including:

[0024] The corresponding surface emissivity value is obtained by using a preset component emissivity mapping library and according to the correspondence between the component category labels;

[0025] The category labels are used to search the component emissivity mapping library and extract matching records.

[0026] The surface emissivity values ​​in the matching records are output as component emissivity.

[0027] Optionally, atmospheric transmittance and atmospheric path radiation are calculated using the environmental meteorological parameters. Combined with the component's emissivity, the infrared radiation transmission of each pixel in the original radiance map of the component is inversely extrapolated to eliminate the effects of atmospheric attenuation and environmental reflection, generating a compensated radiance map, including:

[0028] Atmospheric transmittance was calculated by fitting the atmospheric absorption coefficient with ambient temperature and correcting for wind speed.

[0029] Atmospheric path radiation is calculated based on ambient temperature, atmospheric transmittance, and the distance between the infrared thermal imager and the power equipment.

[0030] For each pixel in the original radiance map of the component, the compensated radiance pixel value is calculated using the radiance value, the atmospheric path radiation, the atmospheric transmittance, and the component emissivity.

[0031] The compensated radiance map is composed of all the compensated radiance pixel values.

[0032] Optionally, based on the compensated radiance map and the component emissivity, the radiance is converted into a temperature value to form a temperature field for the device component, including:

[0033] The temperature is deduced from the pixel value of the radiance by using the inverse function of Planck's blackbody radiation law.

[0034] The radiance values ​​of each pixel in the compensated radiance map are combined with the emissivity of the component and substituted into the mapping relationship to calculate the temperature value corresponding to each pixel.

[0035] The temperature values ​​of each pixel are arranged according to their original positions to form the temperature field of the device components.

[0036] Optionally, the temperature fields of various equipment components are spatially merged to generate an overall temperature distribution map of the power equipment, and abnormal temperature areas are identified based on preset temperature tolerance criteria for each component, including:

[0037] The temperature fields of all the equipment components belonging to the same power equipment are spatially aligned and stitched together to generate an overall temperature distribution map of the equipment.

[0038] Obtain the preset allowable temperature thresholds for each component of the power equipment, and search for pixel regions that exceed the corresponding maximum allowable temperature thresholds in the overall temperature distribution map of the equipment;

[0039] The retrieved pixel regions are highlighted visually on the overall temperature distribution map of the device as areas of temperature anomalies.

[0040] Based on the same inventive concept, this invention also provides an online temperature monitoring system for power equipment based on infrared images. The system includes: an infrared image acquisition module for acquiring a sequence of infrared images of the area where the power equipment is located; an environmental parameter acquisition module for simultaneously acquiring ambient temperature, relative humidity, and wind speed to obtain environmental meteorological parameters; an image preprocessing module for performing noise reduction and contrast enhancement processing on the infrared images in the infrared image sequence to generate enhanced infrared images; an equipment region segmentation module for extracting the main body region of the power equipment from the enhanced infrared images using an equipment segmentation model to generate an equipment region image; and a component identification and radiation extraction module for performing component-level semantic segmentation on the equipment region image, identifying the category labels of each component, generating component mask regions, and extracting the component from the enhanced infrared images. The pixel radiance within the component mask area forms the original radiance map of the component; the emissivity retrieval module retrieves the corresponding surface emissivity value based on the category label to obtain the component emissivity; the atmospheric compensation module calculates the atmospheric transmittance and atmospheric path radiation based on the environmental meteorological parameters, and uses the atmospheric transmittance, atmospheric path radiation, and the component emissivity to perform inverse atmospheric transmission compensation on the original radiance map of the component to generate a compensated radiance map; the temperature calculation module converts the radiance into a temperature value based on the compensated radiance map and the component emissivity to generate the equipment component temperature field; the fusion diagnosis module stitches the equipment component temperature fields of all components of the power equipment into an overall equipment temperature distribution map, and identifies abnormal temperature areas in the overall equipment temperature distribution map.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] This invention solves the temperature measurement error caused by fixed emissivity in traditional methods by combining accurate identification of equipment components in infrared images with automatic matching of component emissivity. Simultaneously, it dynamically compensates the infrared radiation transmission path using environmental meteorological parameters, eliminating the superposition interference of atmospheric attenuation and environmental reflection. These two steps work synergistically: adaptive emissivity matching ensures an accurate description of the target's own radiation characteristics, while atmospheric compensation guarantees the clean restoration of radiance attenuation and stray radiation along the transmission path. Together, they result in a final temperature field for the equipment components with higher accuracy than the summation of individual methods. Furthermore, this method eliminates the need for manual on-site emissivity calibration or the setting of reference bodies, improving the automation and environmental adaptability of online monitoring and reducing the false alarm and missed alarm rates for overheating faults in outdoor power equipment.

[0043] This invention, through the cascaded processing of a device region segmentation model and a component-level semantic recognition network, can decompose power equipment step by step to specific component units, and pinpoint abnormal areas on the final temperature distribution map with precision down to the component mask boundary. Compared to the coarse-grained analysis of traditional methods that can only identify hot spots in the overall equipment, this invention can clearly indicate which specific component is experiencing abnormal heating, thus providing maintenance personnel with precise troubleshooting directions. Furthermore, this invention uses independent temperature tolerance criteria for different components to determine anomalies, avoiding false alarms or missed alarms caused by using the same threshold for components with different heat dissipation conditions, thereby improving the specificity and reliability of fault diagnosis.

[0044] This invention forms a complete automated processing chain, from infrared image acquisition and synchronous acquisition of environmental parameters to image enhancement, equipment and component identification, radiative transfer compensation, temperature inversion, and automatic calibration and output of abnormal areas. Throughout the process, no manual emissivity setting, manual selection of the temperature measurement area, or manual interpretation of thermal images is required, eliminating the influence of operator subjective experience on the monitoring results. With this closed-loop architecture, the system can continuously process real-time input infrared image streams at a fixed frame rate, achieving continuous and uninterrupted monitoring of power equipment temperature. It is particularly suitable for long-term online operation requirements in scenarios such as unattended substations or transmission line towers, providing reliable data support for predictive maintenance of power equipment. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the online temperature monitoring method for power equipment based on infrared images, according to an embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram of the local pixel radiation brightness distribution and masking effect in an embodiment of the present invention.

[0047] Figure 3 This is a schematic diagram of the architecture of an online infrared temperature monitoring system for power equipment according to an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] refer to Figure 1One embodiment of the present invention proposes an online temperature monitoring method for power equipment based on infrared images. By combining component-level segmentation and recognition of infrared images with dynamic compensation of environmental meteorological parameters, high-precision online temperature measurement and fault calibration of equipment can be achieved.

[0050] S1. Obtain the infrared image sequence of the area where the power equipment is located and the environmental meteorological parameters collected simultaneously, including ambient temperature, relative humidity and wind speed.

[0051] In one embodiment of the present invention, step S1 includes the following steps:

[0052] The thermal radiation signals of power equipment are acquired by an infrared thermal imager at a preset frame rate to form an infrared image sequence and an infrared image of the infrared image sequence.

[0053] The environmental parameter acquisition unit synchronously collects air temperature, relative humidity, and wind speed, and records environmental meteorological parameters.

[0054] The basic data required for the synchronous acquisition of power equipment condition analysis mainly includes two parts: dynamic image data reflecting the thermal state of the equipment and environmental condition data that affects the accuracy of thermal imaging temperature measurement.

[0055] Specifically, an infrared thermal imager continuously captures images of the area containing the target electrical equipment to capture dynamic information about the changes in surface thermal radiation over time. An infrared thermal imager is a device that converts invisible thermal radiation emitted by an object into a visible image; the brightness value of each pixel in the image represents the intensity of the radiated energy at that corresponding point on the object. The imaging process is performed at a preset frame rate, which guides the number of times the infrared thermal imager acquires images per second. This preset frame rate is set based on the typical rate of change of the thermal characteristics of the electrical equipment. For example, based on the analysis of 200 sets of measured data from industrial sensors, it was found that large oil-immersed equipment such as transformers have high thermal inertia and their surface temperature changes slowly. Setting the preset frame rate to 1 Hz is sufficient to capture the gradual temperature change process without generating redundant data. The continuously acquired image sequence constitutes an infrared image sequence.

[0056] Using an environmental parameter acquisition unit, key environmental meteorological parameters are synchronously recorded at the power equipment site. Synchronous acquisition ensures that each frame of infrared image has corresponding environmental data at that moment, providing a time reference for subsequent accurate temperature calculations. The environmental parameter acquisition unit typically integrates multiple sensors; in this step, three core parameters need to be collected. The first is ambient temperature, i.e., the air temperature around the equipment, which directly affects the calculation of atmospheric path radiation; the second is relative humidity, i.e., the percentage of water vapor content in the air relative to the saturated water vapor content, a key factor affecting atmospheric absorption of infrared radiation; and the third is wind speed, i.e., the speed of airflow, which affects convective heat dissipation from the equipment surface and is used in subsequent steps to correct the atmospheric transport model.

[0057] After acquiring the raw environmental data, two derived parameters for radiometric correction need to be calculated: atmospheric transmittance and atmospheric path radiation. Atmospheric transmittance describes the proportion of thermal radiation emitted from the surface of the power equipment that penetrates the atmosphere before reaching the infrared thermal imager lens. Atmospheric path radiation refers to the thermal radiation emitted by the air column between the equipment and the thermal imager due to its own temperature. The calculation of these two parameters requires the collected ambient temperature and relative humidity, as well as an additional input parameter: the observation distance from the infrared thermal imager to the power equipment. The observation distance is the straight-line distance from the leading edge of the thermal imager objective lens to the surface of the equipment under test, and needs to be accurately measured using a laser rangefinder or on-site calibration.

[0058] Atmospheric path radiation The calculation is based on the emission of the atmosphere itself, which can be approximated as gray body radiation.

[0059] ,

[0060] in, This is atmospheric path radiation. The calculated atmospheric transmittance is dimensionless. The ambient temperature. This is an instrument constant related to Planck's law. This constant is determined by the response spectrum and calibration parameters of the infrared thermal imager, and its setting is based on the calibration data sheet provided when the thermal imager is shipped from the manufacturer.

[0061] After the calculations are completed, the obtained atmospheric transmittance and atmospheric path radiation, together with the previously collected ambient temperature, relative humidity and wind speed, constitute a complete set of environmental meteorological parameters for use in subsequent steps.

[0062] For example, suppose an outdoor box-type transformer needs to be inspected. The operator deploys an infrared thermal imager and an environmental parameter acquisition unit on-site. The preset frame rate of the infrared thermal imager is set to 1 Hz. The observation distance L from the infrared thermal imager to the power equipment is measured to be 20 meters using laser ranging. At a certain moment, the environmental parameter acquisition unit synchronously records the following data: ambient temperature 25 degrees Celsius, relative humidity 60%, and wind speed 2 m / s. The infrared thermal imager captures one frame of infrared image, forming a member of the infrared image sequence.

[0063] Convert the ambient temperature from Celsius to Kelvin:

[0064]

[0065] Relative humidity converted to decimal form:

[0066] ,

[0067] Using atmospheric model fitting coefficients , , Calculate atmospheric transmittance .

[0068] Calculate atmospheric path radiation Assuming that based on the calibration data of the thermal imager, the instrument constant... Set to 14388K.

[0069] ,

[0070] Atmospheric path radiation was calculated. .

[0071] Upon completion of this step, the obtained environmental meteorological parameters at that moment are: ambient temperature 25 degrees Celsius, relative humidity 60%, wind speed 2 m / s, atmospheric transmittance 0.937, and atmospheric path radiation. These data, along with the synchronously acquired infrared images, are stored for subsequent processing.

[0072] S2. Perform noise reduction and contrast enhancement processing on the infrared images in the infrared image sequence to generate enhanced infrared images;

[0073] In one embodiment of the present invention, step S2 includes the following steps:

[0074] Non-uniformity correction is performed on single-frame infrared images in the infrared image sequence to eliminate fixed-pattern noise introduced by differences in detector response;

[0075] The corrected image is subjected to joint spatiotemporal filtering to suppress random noise and preserve device edges;

[0076] An enhanced infrared image is obtained by adaptive histogram stretching of the filtered image.

[0077] Specifically, this step aims to preprocess infrared thermal imagers to address issues such as image blurring, high noise, and low contrast caused by factors such as hardware manufacturing processes and environmental interference. Through comprehensive image restoration and enhancement, it provides a high-quality input source for subsequent accurate equipment segmentation and component identification.

[0078] Non-uniformity correction is performed on single-frame infrared images in the infrared image sequence. Due to slight differences in the response characteristics of each detector element in the infrared focal plane array, even when facing a uniform radiation source, the output image will contain fixed pattern noise such as stripes or spots. To eliminate this hardware-introduced systematic error, this step employs a two-point correction method based on a standard blackbody radiation source. The gain correction coefficient for each pixel is calculated based on blackbody response images acquired at two different temperatures. and bias correction coefficient The correction formula is as follows:

[0079] ,

[0080] in, In order to be in At what moment, in the original output image of the infrared thermal imager, the first... Line 1 The original grayscale value of the column pixels. This is the corrected grayscale value corresponding to this pixel location after non-uniformity correction. For the first Line 1 The gain correction coefficient for column pixels is dimensionless. It is set based on the principle of normalizing the response slope of all detector elements to the same radiation increment. For the first Line 1 The offset correction coefficient for column pixels, expressed in grayscale values. Used to smooth out differences in dark current and background thermal noise among individual detectors.

[0081] A joint spatiotemporal filtering process is performed on the corrected image. Although non-uniformity correction eliminates fixed-pattern noise, random noise that dynamically changes over time still remains in the image due to environmental thermal fluctuations and stray capacitance in the circuit. Simply using spatial domain filtering would blur the edges of the power equipment, hindering subsequent component identification. Therefore, this step introduces a three-dimensional block-matching filter or a spatiotemporal three-dimensional bilateral filter. This process utilizes both the continuity of the infrared image sequence along the time axis and the structural similarity within a single frame. In the spatial domain, the spatial distance weights and gray-level difference weights of the bilateral filter are used to suppress noise while forcibly preserving the step edges of the power equipment outline. In the temporal domain, temporal averaging is performed on corresponding regions of multiple adjacent frames to smooth out random flickering particle noise.

[0082] Adaptive histogram stretching is applied to the filtered image. Infrared images typically have a wide dynamic range, but the energy difference between the power equipment and the background can cause details of the equipment to be concentrated in an extremely narrow grayscale range, resulting in a dark overall image with very low contrast. This step employs a contrast-limited adaptive histogram equalization algorithm. The entire image is divided into multiple non-overlapping sub-regions, and the histogram for each sub-region is calculated. To prevent excessive amplification of noise locally, a clipping threshold is set. The histogram portions exceeding the threshold are uniformly cropped and redistributed across the entire grayscale range. Bilinear interpolation is then used to eliminate boundary effects between sub-regions, resulting in an enhanced infrared image with improved contrast and clearer device texture.

[0083] For example, suppose that a frame of raw infrared image X at a certain moment was acquired in step S1. The detector resolution of the infrared thermal imager is... The image has a 14-bit bit depth and a grayscale value range of 0 to 16383. For a pixel at coordinates 100, 200, its original output grayscale value is... It is 8500.

[0084] Non-uniformity correction is performed. The parameters for this point, obtained from the infrared thermal imager's factory calibration or on-site pre-calibration, are known to be: gain coefficient. Bias coefficient Substitute the data into the correction formula:

[0085] ,

[0086] The corrected grayscale value becomes 8520, which eliminates the stripe error of the fixed pattern.

[0087] Initiate spatiotemporal joint filtering. Using the current frame as the center, extract the images from the two frames before and after it. For the pixel at coordinates (100, 200), its neighborhood spatial bilateral filtering weights and the temporal smoothing weights of the adjacent frames work together. Assuming that this point is located on the edge of a transformer bushing, traditional spatial filtering would cause its grayscale to slide towards the background, while spatiotemporal bilateral filtering identifies the grayscale step, and the calculated filtered output value stabilizes at 8515, successfully filtering out pixels with an amplitude of... Random white noise on both sides, while maintaining edge sharpness.

[0088] Perform adaptive histogram stretching. Divide the image into local blocks and set a shearing coefficient. Calculations show that, within the grayscale range of 8000 to 9000 where the transformer body is located, after contrast-limited local histogram remapping, the grayscale value of 8515 at this point is linearly and adaptively stretched and mapped to a grayscale level of 11200, which is more visually and for feature extraction. Thus, the value of this pixel in the enhanced infrared image is determined to be 11200, highlighting the outline and surface thermal details of the power equipment.

[0089] S3. Input the enhanced infrared image into the trained device region segmentation model to segment out the device region image containing the main body of the power equipment;

[0090] In one embodiment of the present invention, step S3 includes the following steps:

[0091] The enhanced infrared image is input into a preset device segmentation model, and a probability map corresponding to the main body of the power equipment is output.

[0092] The probability map is binarized to obtain the device body mask;

[0093] The device area is extracted from the enhanced infrared image using the device body mask to obtain a device area image.

[0094] Specifically, because complex outdoor environments typically contain irrelevant interference such as the sky, ground, buildings, and obstructions like weeds, directly measuring the temperature of the entire image would result in superimposed background radiation and severely deplete computational resources. Therefore, this step aims to filter out background interference in the image using a deep learning model, achieving accurate device-level coarse-grained localization.

[0095] The enhanced infrared image obtained after step S2 is input into a device segmentation model based on a convolutional neural network. In this embodiment, the device segmentation model uses a pre-trained lightweight deep network, whose backbone network has been fine-tuned for infrared features. The enhanced infrared image is processed through multiple convolutions and downsampling to extract high-dimensional spatial features, and then upsampling and skip connections are used to restore it to the original resolution. Finally, a softmax classification layer outputs a two-dimensional probability map reflecting the probability that each pixel belongs to the main body of the power equipment. Each pixel value in the probability map is between 0 and 1, representing the confidence level that the location belongs to the equipment target.

[0096] To obtain a clear device edge with well-defined pixel boundaries, the probability map needs to be binarized. This process sets a segmentation threshold. Based on a validation set of 1000 samples, after comprehensively weighing precision and recall, the results will be... The segmentation threshold for binarization is set to 0.65. This threshold is obtained through statistical optimization based on multi-scenario power infrared samples. Each pixel value in the probability map... Compare with this threshold. Its generated device body mask. The calculation process is described by the following physical determination formula:

[0097] ,

[0098] in, For the main body mask of the equipment in coordinate The binarization result at this point is either 0 or 1; This represents the device probability value at the corresponding coordinate in the probability graph. If the probability value is greater than or equal to the threshold... If the probability value is less than the threshold Th, the pixel is determined to be part of the electrical equipment, and its value is activated and set to 1 in the new image; otherwise, if the probability value is less than the threshold Th, it is determined to be the background and its value is set to 0. After traversing the entire image matrix, a mask consisting only of 0s and 1s for the main body of the equipment is obtained.

[0099] The device region image is obtained by using the device body mask to extract the device area from the enhanced infrared image. Specifically, the pixel values ​​corresponding to the enhanced infrared image matrix are compared with the device body mask. A pixel-by-pixel multiplication operation is performed. In the background area with a mask value of 0, the pixel radiance values ​​after the multiplication are all forcibly cleared to zero, appearing as pure black; while in the device body area with a mask value of 1, the corresponding original enhanced radiance pixel values ​​are completely and losslessly preserved. Through this point-by-point extraction, not only is all external background radiance interference eliminated, but also the precise extraction of the main body area of ​​the power equipment is achieved.

[0100] For example, suppose that an enhanced infrared image with a grayscale distribution between 0 and 16383 is generated in step S2. After inputting the device segmentation model, a pixel near the edge of the target transformer body is selected. The model predicts the probability that it belongs to a transformer. The value is 0.78. At this point, the system automatically retrieves the preset segmentation threshold. A determination is made. Since 0.78 > 0.65, according to the above binarization determination formula, the value of this point in the device body mask is assigned... In subsequent extraction operations, the original enhanced brightness value of this point in the enhanced infrared image, such as 11200, is multiplied by 1, and the value 11200 is completely preserved in the final device area image. As for another pixel belonging to the background of a distant building... The model predicts a probability value of only 0.12. Since 0.12 < 0.65, its mask value is... The enhanced brightness value of the background point, multiplied by 0, is forcibly reset to 0. After traversing the entire image, the output device area image successfully separates the background, retaining only the clean outline data of the transformer body.

[0101] S4. Based on the component-level semantic recognition network, determine the category labels and component mask areas of each component constituting the power equipment in the equipment area image, and extract the pixel radiance corresponding to the component mask area from the enhanced infrared image to generate the original radiance map of the component.

[0102] In one embodiment of the present invention, step S4 includes the following steps:

[0103] The enhanced infrared image is input into a preset device segmentation model, and a probability map corresponding to the main body of the power equipment is output.

[0104] The probability map is binarized to obtain the device body mask;

[0105] The device area is extracted from the enhanced infrared image using the device body mask to obtain a device area image.

[0106] Specifically, since power equipment is typically composed of a complex combination of core components made of different materials and with varying thermal properties, blindly measuring the temperature of a single unit will not accurately pinpoint the specific fault location. Therefore, this step aims to delve deeper from the equipment-level image to achieve fine-grained instance segmentation and raw radiant energy extraction at the component level.

[0107] The device region image obtained in step S3 is input into a component-level instance segmentation network. This network employs a two-stage or single-stage high-precision semantic instance segmentation architecture, with parallel classification, bounding box regression, and mask generation branches at its top. Within the device body after background removal, the network can simultaneously distinguish different component instances and predict their specific electrical component type, thereby outputting category labels for each independent component instance, such as high-voltage bushings, conductive contacts, heat sinks, or ceramic insulators, along with corresponding refined component mask regions. These component mask regions, in the form of point sets or binary matrices, pinpoint the precise pixel topological grid points occupied by the component in the image space.

[0108] For example, for each specific component mask region identified, the system performs pixel-level radiance extraction. Since the device region image in step S3 may have slight edge smoothing due to model convolution pooling, in order to ensure the absolute fidelity of the temperature measurement energy, the system returns to the enhanced infrared image generated in step S2, which has the purest radiation source characteristics, and precisely extracts the original radiance values ​​of all pixels located within the region according to the spatial coordinate boundaries of the component mask region.

[0109] The original radiance values ​​of all pixels extracted within the component's mask area are combined into a rectangular or square matrix according to their original spatial row and column geometric order, forming the original radiance map of the component. In this matrix, pixels not belonging to the component are uniformly filled with placeholders (0). Through this fine-grained mapping, the power equipment is successfully decomposed into several decoupled component radiance matrices with independent geometric distributions and energy characteristics. See the corresponding local pixel distribution diagram. Figure 2 . Figure 2 This image presents a two-dimensional thermal map of the radiance of a typical 4×4 local area of ​​a high-heat component after being masked by a device body mask and a component-level recognition network. Pixel values ​​on the left side of the image, belonging to the background or areas outside the component, are set to 0, representing pure black. Pixel radiance values ​​belonging to the target component, such as 95, 150, and 152, are preserved with high fidelity, representing a mid-gray to bright white area, clearly and quantitatively highlighting the spatial radiant energy distribution of the target component.

[0110] S5. Based on the component category label, retrieve the surface emissivity value that matches the category from the component emissivity mapping library to obtain the component emissivity;

[0111] In one embodiment of the present invention, step S5 includes the following steps:

[0112] The device area image is segmented into instances to distinguish different component instances, and the category label and component mask area of ​​each component instance are obtained.

[0113] For each component mask region, the original radiance value of each pixel located within the corresponding component mask region is extracted from the enhanced infrared image;

[0114] The original radiance values ​​of all pixels within the component mask area are combined into a matrix to form the original radiance map of the corresponding component.

[0115] Specifically, the materials used to construct electrical equipment in actual substations or transmission lines are extremely complex. For example, high-voltage contacts and conductive arms are typically made of silver-plated copper or aluminum alloys, which are prone to surface oxidation and dirt accumulation when exposed to the outdoors for extended periods. Insulators, on the other hand, are often made of porcelain or silicone rubber composite materials. The physical emissivity of different materials in the infrared band, i.e., their surface emissivity, exhibits inherent heterogeneity. Setting a uniform emissivity for the entire device, such as a fixed 0.90, as in traditional temperature measurement, would directly lead to significant systematic errors in the temperature measurement results at the copper-aluminum joints. This step aims to assign each component a unique, physically accurate emissivity parameter through automated tag retrieval.

[0116] The system pre-configures and retrieves a component emissivity mapping library associated with the model of the electrical equipment currently being tested in the back-end memory. This component emissivity mapping library, as a relational data mapping library, stores a strict correspondence between all known standard component categories and their corresponding surface emissivity values ​​for that equipment model. These surface emissivity values ​​are set based on: laboratory integral sphere reflectance measurements conducted before system deployment on various typical electrical materials, such as polished copper, alumina, insulating ceramics, and silicone rubber, under different aging cycles and pollution levels, or by consulting high-precision calibration constants determined from national standard tables of physical properties of electrical materials.

[0117] The system uses the category labels of each component instance automatically predicted and output by the deep learning network in step S4 as the core retrieval key, performs automatic high-speed retrieval in the component emissivity mapping table, and matches and extracts the matching record corresponding to the label record.

[0118] The specific surface emissivity values ​​retrieved from the matching records are output and bound as the component emissivity at that moment. This parameter will serve as a core physical weight, a key physical input for subsequent steps to eliminate the artifact of different brightness infrared radiation at the same temperature and to restore the true absolute temperature of the component.

[0119] For example, consider a specific component instance identified by the input instance segmentation network in step S4. The network outputs its bounding box and assigns it the category label "disconnector copper contact". Upon receiving this label, the system immediately retrieves the component emissivity mapping library associated with the disconnector model. In this mapping library, the calibrated surface emissivity value corresponding to the disconnector copper contact is 0.68. Through precise record retrieval, the system successfully matches the entry and assigns the component emissivity of this specific component instance. The output was determined to be 0.68. For another component instance in the same image frame identified as a ceramic post insulator, the system retrieved its matching surface emissivity value from the mapping library, which was 0.92. Through this adaptive matching, components of completely different materials obtained their own independent emissivity constants that conformed to their true physical characteristics: 0.68 for the contact and 0.92 for the insulator. This avoids the drawbacks of the one-size-fits-all parameter settings in traditional temperature measurement and provides an accurate physical parameter matrix for subsequent high-precision radiation transmission compensation.

[0120] S6. Calculate atmospheric transmittance and atmospheric path radiation using the environmental meteorological parameters, and combine them with the component emissivity to perform inverse infrared radiation transmission deduction on the radiance of each pixel in the original radiance map of the component, eliminate the influence of atmospheric attenuation and environmental reflection, and generate a compensated radiance map.

[0121] In one embodiment of the present invention, step S6 includes the following steps:

[0122] The corresponding surface emissivity value is obtained by using a preset component emissivity mapping library and according to the correspondence between the component category labels;

[0123] The category labels are used to search the component emissivity mapping library and extract matching records.

[0124] The surface emissivity values ​​in the matching records are output as component emissivity.

[0125] Specifically, the atmospheric transmittance calculation model used in this step, and the formula based on MODTRAN fitting exemplarily given in step S1 above, are both equivalent mathematical descriptions of atmospheric transport characteristics. The Beer-Lambert exponential decay model used in this step is easier to dynamically correct in conjunction with real-time wind speed and is the preferred implementation of this invention. Under the same environmental parameter input, the transmittance values ​​calculated by the two models deviate within an acceptable engineering range, typically less than 2%, and do not affect the final temperature measurement accuracy. Figure 2 As shown.

[0126] In infrared thermal imaging technology, the total radiance received by the infrared thermal imager detector is not entirely determined by the surface radiation of the target device. As the infrared radiation emitted by the target object travels through the atmosphere to the thermal imager, it is attenuated by atmospheric absorption and scattering; simultaneously, the atmospheric path between the target and the imager itself emits infrared energy due to its temperature. To eliminate these external environmental interferences, precise atmospheric transmission inverse compensation must be performed on each pixel of the infrared image.

[0127] For example, atmospheric absorption coefficients based on environmental meteorology are used to construct a model where atmospheric absorption of infrared radiation is primarily caused by gas molecules such as water vapor and carbon dioxide in the air. In specific infrared operating bands, such as... In the long-wave infrared band, water vapor content is the decisive factor causing attenuation fluctuations. The system collects ambient temperature data from the site. and relative humidity The atmospheric absorption coefficient under the current conditions is obtained by fitting an empirical formula:

[0128] ,

[0129] in, and These are empirical constants obtained through laboratory calibration or large-sample fitting for specific bands and regional meteorological backgrounds, used to accurately pinpoint the atmospheric response characteristics of cooled or uncooled infrared sensors. The typical value range is from 0.0001 to 0.0005. The typical value range is 0.01 to 0.05, and the specific value is determined through laboratory calibration. When calculating atmospheric transmittance using multiphysics correction, the energy of infrared radiation transmitted through the medium decreases exponentially with transmission distance. Furthermore, the wind speed in the field environment... This can cause uneven atmospheric turbulence between the target and the lens, resulting in weak scattering interference and refractive index disturbances in the optical path. To improve the inversion accuracy in complex industrial scenarios, this system introduces a wind speed correction factor based on the classic Beer-Lambert law. The final atmospheric transmittance calculation model is expressed as follows:

[0130] ,

[0131] in, Atmospheric absorption coefficient, unit: . The actual three-dimensional distance between the infrared thermal imager and the power equipment component is usually obtained in real time by the laser ranging module built into the thermal imager. The wind speed correction factor is a dimensionless coefficient with a very small value. It is determined by statistical regression of the influence of different turbulence intensities on the stability of infrared radiation transmission in wind tunnel simulation experiments. The wind speed is in real time, measured in meters per second.

[0132] A refined characterization of atmospheric path radiation: Atmospheric path radiation refers to the component of the signal received by an infrared sensor that originates from the thermal radiation of the atmosphere itself along its path. According to Kirchhoff's law, the emissivity of the atmosphere equals its absorptivity, i.e., ... Combining the integral expression of Planck's formula in the low-frequency range, the atmosphere along the entire path is equivalent to an environment at a specific temperature. The radiance per unit solid angle and per unit area of ​​a non-completely blackbody. The calculation is as follows:

[0133] ,

[0134] in, This is the instrument constant of the infrared thermal imager, expressed in K, and is determined by the factory calibration parameters of the thermal imager. Atmospheric transmittance; Let be the ambient temperature. The physical meaning of this formula is that the passing atmosphere is equivalent to a non-blackbody at the ambient temperature, whose emissivity equals its absorptivity. Then multiply by the blackbody radiance at ambient temperature.

[0135] Pixel-level atmospheric transmission inverse compensation calculation after obtaining atmospheric transmittance Atmospheric radiation and the surface emissivity of the component obtained through component category label mapping. Subsequently, the system outputs the original radiance map of the components from the infrared thermal imager, whose two-dimensional matrix form can be represented as follows: Perform pixel-by-pixel point operations.

[0136] Physical inverse compensation equation. For any pixel with coordinates (x, y) in the image, its acquired original radiance value... It includes the superposition of the residual target radiation after atmospheric attenuation and the atmospheric path radiation. Its physical forward degradation model is as follows:

[0137] ,

[0138] In order to reproduce the true blackbody equivalent radiance of the component surface after eliminating all external interference. By performing the inverse algebraic operation on the above equation, the following inverse atmospheric transport compensation equation is constructed:

[0139] ,

[0140] in, To extract the pure mixed radiation that originates from the device surface and penetrates the atmosphere from the total energy captured by the sensor, the noise energy that pours into the infrared lens due to the temperature of the atmosphere is subtracted. To compensate for the absorption and scattering attenuation of infrared energy during long-distance air transmission, the radiation intensity is equivalently amplified and restored to the radiation physical quantity level when it just leaves the object's surface and has not yet entered the atmosphere. Because actual electrical equipment components, such as aluminum connectors and porcelain insulators, are non-ideal blackbodies, their radiation capacity is discounted compared to a blackbody at the same temperature. This is calculated by dividing by the surface emissivity. This allows the actual radiance of a non-blackbody to be equivalently converted to the radiance value of an ideal blackbody at the same temperature. .

[0141] The compensated radiance map system traverses every grid point in the entire original image matrix, completing the aforementioned pipelined algebraic transformation, and then calculates the compensated radiance pixel values. The original two-dimensional spatial topology was rearranged, and a compensated radiance map was finally reconstructed and generated. This map eliminated multiple nonlinear coupling interferences from measurement distance, wind speed, humidity, air temperature, and material surface physical properties, achieving pixel-level energy normalization and purification. This lays a deterministic physical data foundation for subsequent precise and lossless mapping from radiance to temperature by directly calling the inverse function of Planck's law, thereby achieving high-precision reconstruction of the temperature field of equipment components.

[0142] S7. Based on the compensated radiance map and component emissivity, the radiance is converted into temperature values ​​through the inverse function of radiation thermometry, forming the temperature field of the equipment components.

[0143] In one embodiment of the present invention, step S7 includes the following steps:

[0144] The temperature is deduced from the pixel value of the radiance by using the inverse function of Planck's blackbody radiation law.

[0145] The radiance values ​​of each pixel in the compensated radiance map are combined with the emissivity of the component and substituted into the mapping relationship to calculate the temperature value corresponding to each pixel.

[0146] The temperature values ​​of each pixel are arranged according to their original positions to form the temperature field of the device components.

[0147] Specifically, after obtaining the compensated radiance map that eliminates the effects of atmospheric interference and material emissivity, the implementation process enters the core light-to-heat data conversion stage. This stage achieves a precise mapping from pure radiance to absolute physical temperature by invoking the inverse function of Planck's blackbody radiation law.

[0148] According to Planck's law of radiation in quantum physics, a blackbody at a temperature... The monochromatic radiance under certain conditions exhibits a definite nonlinear relationship with its absolute temperature. In the field of industrial infrared thermometry, the optical system and detector of an infrared thermal imager only respond to a specific wavelength range, such as... The integrated radiant energy within the body. To perform efficient and high-precision real-time temperature inversion in embedded systems or back-end processing, the experimentally calibrated Planck simplified formula is typically used to describe the relationship between the actual temperature of a non-blackbody and its compensated radiance:

[0149] ,

[0150] in, The target physical quantity is calculated and output, with the unit being K. The input source data consists of the brightness values ​​of the corresponding pixels in the compensated radiance map, in units of... . The surface emissivity value is obtained through the component category label mapping in the aforementioned steps. This is a dimensional parameter of temperature: its value is related to Planck's constant, the speed of light, and the center frequency of the band, and its dimensions are consistent with those of temperature. The radiation response coefficient, The linear correction intercept is used to fit the photoelectric conversion nonlinear curve of a specific sensor detector array. These calibration parameters are determined by the infrared thermal imager manufacturer before shipment using measurements from a multi-point blackbody radiation source in a controlled laboratory environment, combined with regression algorithms such as least squares, and are embedded in the device firmware or image metadata.

[0151] The pixel-level brightness to temperature conversion mechanism system performs a full matrix traversal operation on the two-dimensional compensated radiance map, and its execution mechanism is presented in the following three steps:

[0152] 1. Gray-body to black-body equivalent normalization. Due to the varying materials of electrical equipment, the calculated radiance of this pixel after compensation is... Combined with the corresponding component emissivity Performing blackbody equivalent transformation, i.e., calculation This brings the radiation energy levels of a non-ideal blackbody to the physical baseline of an ideal blackbody.

[0153] 2. Inverse Function Nonlinear Mapping. The normalized brightness data is substituted into the inverse function, and the nonlinearly suppressed temperature signal is demodulated through logarithmic operations. In this case, the logarithm of the denominator is inversely proportional to the logarithm of the radiation intensity. Finally, through proportional amplification of the numerator B, the true absolute temperature value corresponding to that pixel is directly calculated. .

[0154] 3. Temperature scale unit project conversion. Absolute temperature In electrical maintenance, this is not intuitive. After calculating the results, the system uniformly performs a temperature scale conversion, converting Kelvin to Celsius.

[0155] ,

[0156] For example, in the temperature field of a two-dimensional device component, after each pixel and grid point (x,y) in the image matrix has undergone the aforementioned algebraic mapping from radiance to Celsius temperature, all calculated temperature values ​​will be arranged in a matrix strictly according to their spatial geometric topological positions in the original infrared image. This two-dimensional data matrix consists of temperature scalars. This refers to the temperature field of the equipment components.

[0157] The temperature field of the equipment components has the following characteristics:

[0158] Spatial consistency means that every value in the temperature field corresponds one-to-one with the actual surface micro-area of ​​each component of the power equipment in the field of view of the thermal imager lens.

[0159] Physical independence means that the temperature matrix has broken free from the constraints of infrared photosensitive characteristics and transformed into a pure two-dimensional numerical field that reflects the true thermodynamic state of the device surface.

[0160] The temperature field of the equipment components eliminates infrared artifacts caused by different materials, such as the coexistence of copper conductive arms and porcelain insulators, resulting in different brightness levels at the same temperature or different temperatures at the same brightness. This provides a dataset with absolute physical significance for subsequent three-dimensional spatial stitching of multi-component images, as well as for comparing the maximum allowable temperature thresholds of components based on national operating procedures and identifying temperature anomalies.

[0161] S8. The temperature fields of various equipment components are spatially integrated to generate an overall temperature distribution map of the power equipment. Based on the temperature tolerance criteria of each component, abnormal temperature areas are identified to achieve high-precision online monitoring.

[0162] In one embodiment of the present invention, step S8 includes the following steps:

[0163] The temperature fields of all the equipment components belonging to the same power equipment are spatially aligned and stitched together to generate an overall temperature distribution map of the equipment.

[0164] Obtain the preset allowable temperature thresholds for each component of the power equipment, and search for pixel regions that exceed the corresponding maximum allowable temperature thresholds in the overall temperature distribution map of the equipment;

[0165] The retrieved pixel regions are highlighted visually on the overall temperature distribution map of the device as areas of temperature anomalies.

[0166] Specifically, after generating the two-dimensional temperature fields of each independent component, such as conductive arms, insulators, and fixed supports, the process enters the final global spatial fusion and intelligent diagnostic stage. This step achieves a leap from local isolated temperature measurement to global high-precision online monitoring through spatial geometric alignment of multi-physics fields and comparison with procedural criteria. The spatial fusion mechanism of heterogeneous multi-component temperature fields: The individual component temperature fields only reflect the two-dimensional thermal state of a specific local area within the thermal imager's field of view. To obtain a global thermal panorama of the equipment, these discrete temperature fields must be fused within a three-dimensional or unified two-dimensional spatial topology.

[0167] The system utilizes the 3D CAD mathematical model of the power equipment or the geometric relationships of images taken from multiple perspectives to rotate, scale, and translate the temperature field matrices of different components belonging to the same power equipment.

[0168] In areas where different components intersect or overlap, the system employs a weighted average or multi-scale grid fusion algorithm to eliminate temperature faults caused by differences in perspective or edge effects, thereby stitching together fragmented local matrices into a continuous and organic overall temperature distribution map of the power equipment.

[0169] The fused overall temperature distribution map is not only a visual image, but also a high-dimensional dataset. Each pixel in the image has dual attributes: spatial coordinates and absolute temperature.

[0170] The core of high-precision online monitoring based on intelligent calibration algorithms using temperature tolerance criteria lies in timely and accurate capture of early defects from massive amounts of thermal data. The system incorporates an automated expert system to intelligently review the temperature tolerance of the overall temperature distribution map pixel-by-pixel or region-by-region: dynamically acquiring the maximum allowable temperature threshold. Different power components exhibit significant differences in their materials, contact methods, insulation media, and aging resistance, resulting in vastly different safety limits stipulated in their operating procedures. Based on the spatial alignment results, the system automatically retrieves the corresponding national power operating procedures or equipment factory parameters.

[0171] For example, the maximum permissible temperature threshold for the copper conductive arm of a substation disconnector during long-term operation. Possibly set as However, the heat resistance thresholds of certain porcelain insulators or specific joints vary.

[0172] The system performs difference discrimination and anomaly identification by analyzing the actual calculated temperature value of each pixel in the overall temperature distribution map matrix. The threshold of its component Perform real-time difference comparison. The judgment logic is as follows:

[0173] ,

[0174] When a pixel or a locally connected region composed of multiple consecutive pixels satisfies At this point, the area is automatically defined by the system as a temperature anomaly zone, such as localized overheating, poor contact, or dielectric loss defects. After the calibration output and high-precision online monitoring closed loop complete the discrimination, the system transforms the anomalies at the data level into intuitive information visible to operation and maintenance, forming a complete closed loop of high-precision monitoring.

[0175] Visual identification enhancement: On the output overall temperature distribution map of the power equipment, the retrieved abnormal pixel areas will be forcibly rendered through specific highlighting, such as red highlights in pseudo-color synthesis, flashing outlines, or infrared hot spot selection, making them easy to identify in complex equipment backgrounds.

[0176] Quantitative alarm output: The system not only marks the location, but also automatically calculates and outputs the highest temperature, average temperature, overheated area and temperature rise rate of the abnormal area, prompting maintenance personnel to indicate the severity level of the defect, such as general overheating, severe overheating, and critical defect.

[0177] By employing an automated closed-loop process that encompasses the entire process from emissivity acquisition to atmospheric inverse compensation to inverse function temperature measurement to spatial fusion to criterion calibration, the system effectively addresses the problems of traditional infrared thermometry, such as high susceptibility to environmental interference, high material misjudgment rate, and inefficiency of manual inspection. This enables all-weather, high-precision, and quantitative online monitoring and anomaly early warning for power equipment.

[0178] Based on the same inventive concept, such as Figure 3 As shown, the present invention also provides an online temperature monitoring system for power equipment based on infrared images, the system comprising:

[0179] Infrared image acquisition module, used to acquire infrared image sequences of the area where the power equipment is located;

[0180] The environmental parameter acquisition module is used to simultaneously collect ambient temperature, relative humidity, and wind speed to obtain environmental meteorological parameters;

[0181] The image preprocessing module is used to perform noise reduction and contrast enhancement processing on the infrared images in the infrared image sequence to generate enhanced infrared images;

[0182] The equipment region segmentation module is used to extract the main area of ​​the power equipment from the enhanced infrared image using the equipment segmentation model, and generate an equipment region image.

[0183] The component identification and radiation extraction module is used to perform component-level semantic segmentation on the device area image, identify the category labels of each component, generate a component mask area, extract the pixel radiation brightness within the component mask area from the enhanced infrared image, and form the original radiation brightness map of the component.

[0184] The emissivity retrieval module is used to retrieve the corresponding surface emissivity value based on the category label to obtain the component emissivity;

[0185] The atmospheric compensation module is used to calculate the atmospheric transmittance and atmospheric path radiation based on the environmental meteorological parameters, and to perform atmospheric transmission inverse compensation on the original radiance map of the component using the atmospheric transmittance, atmospheric path radiation and the component emissivity to generate a compensated radiance map.

[0186] The temperature calculation module is used to convert the radiance into a temperature value based on the compensated radiance map and the emissivity of the component, and generate a temperature field for the equipment component.

[0187] The fusion diagnostic module is used to stitch together the temperature fields of all components of the power equipment into an overall temperature distribution map of the equipment, and to identify abnormal temperature areas in the overall temperature distribution map of the equipment.

[0188] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for online monitoring temperature of power equipment based on infrared image, characterized in that, include: Acquire a continuous sequence of infrared images of the area where the power equipment is located, as well as synchronously collected environmental meteorological parameters, which include at least ambient temperature, relative humidity, and wind speed. The infrared images in the infrared image sequence are subjected to noise suppression and detail enhancement processing to generate enhanced infrared images; The enhanced infrared image is segmented into equipment regions to obtain an equipment region image containing the main body of the power equipment; In the equipment area image, determine the component category labels and component mask areas of each component constituting the power equipment, and extract the pixel radiance corresponding to the component mask area from the enhanced infrared image to generate the original radiance map of the component. Based on the component category label, retrieve the surface emissivity value that matches the component category label from the preset component emissivity mapping library to obtain the component emissivity; Using the environmental meteorological parameters, atmospheric transmittance and atmospheric path radiation are calculated. Combined with the emissivity of the component, the infrared radiation transmission of each pixel in the original radiance map of the component is reversed to eliminate the influence of atmospheric attenuation and environmental reflection, and a compensated radiance map is generated. Based on the compensated radiance map and the emissivity of the component, the radiance is converted into a temperature value to form a temperature field for the equipment component. The temperature fields of each equipment component are spatially integrated to generate an overall temperature distribution map of the power equipment, and abnormal temperature areas are marked according to the preset temperature tolerance criteria of each component.

2. The method according to claim 1, wherein, Acquire continuous infrared image sequences of the area where the power equipment is located, along with synchronously collected environmental meteorological parameters, including at least ambient temperature, relative humidity, and wind speed. The thermal radiation signals of power equipment are acquired by an infrared thermal imager at a preset frame rate to form an infrared image sequence and an infrared image of the infrared image sequence. The environmental parameter acquisition unit synchronously collects air temperature, relative humidity, and wind speed, and records environmental meteorological parameters.

3. The method for online temperature monitoring of power equipment based on infrared images according to claim 2, characterized in that, The infrared images in the infrared image sequence are subjected to noise suppression and detail enhancement processing to generate enhanced infrared images, including: Non-uniformity correction is performed on single-frame infrared images in the infrared image sequence to eliminate fixed-pattern noise introduced by differences in detector response; The corrected image is subjected to joint spatiotemporal filtering to suppress random noise and preserve device edges; An enhanced infrared image is obtained by adaptive histogram stretching of the filtered image.

4. The method for online temperature monitoring of power equipment based on infrared images according to claim 3, characterized in that, The enhanced infrared image is segmented into equipment regions to obtain an equipment region image containing the main body of the power equipment, including: The enhanced infrared image is input into a preset device segmentation model, and a probability map corresponding to the main body of the power equipment is output. The probability map is binarized to obtain the device body mask; The device area is extracted from the enhanced infrared image using the device body mask to obtain a device area image.

5. The method for online temperature monitoring of power equipment based on infrared images according to claim 4, characterized in that, In the equipment area image, component category labels and component mask areas for each component constituting the power equipment are determined, and the pixel radiance corresponding to the component mask areas is extracted from the enhanced infrared image to generate the original radiance map of the components, including: The device area image is segmented into instances to distinguish different component instances, and the category label and component mask area of ​​each component instance are obtained. For each component mask region, the original radiance value of each pixel located within the corresponding component mask region is extracted from the enhanced infrared image; The original radiance values ​​of all pixels within the component mask area are combined into a matrix to form the original radiance map of the corresponding component.

6. The method for online temperature monitoring of power equipment based on infrared images according to claim 5, characterized in that, Based on the component category label, the surface emissivity value matching the component category label is retrieved from a preset component emissivity mapping library to obtain the component emissivity, including: The corresponding surface emissivity value is obtained by using a preset component emissivity mapping library and according to the correspondence between the component category labels; The category labels are used to search the component emissivity mapping library and extract matching records. The surface emissivity values ​​in the matching records are output as component emissivity.

7. The method for online temperature monitoring of power equipment based on infrared images according to claim 6, characterized in that, Using the aforementioned environmental meteorological parameters, atmospheric transmittance and atmospheric path radiation are calculated. Combined with the component's emissivity, the infrared radiation transmission of each pixel in the original radiance map of the component is inversely extrapolated to eliminate the effects of atmospheric attenuation and environmental reflection, generating a compensated radiance map, including: Atmospheric transmittance is obtained by fitting the atmospheric absorption coefficient with ambient temperature and correcting for wind speed. Atmospheric path radiation is calculated based on ambient temperature, atmospheric transmittance, and the distance between the infrared thermal imager and the power equipment. For each pixel in the original radiance map of the component, the compensated radiance pixel value is calculated using the radiance value, the atmospheric path radiation, the atmospheric transmittance, and the component emissivity. The compensated radiance map is composed of all the compensated radiance pixel values.

8. The method for online temperature monitoring of power equipment based on infrared images according to claim 7, characterized in that, Based on the compensated radiance map and the component emissivity, the radiance is converted into a temperature value to form a temperature field for the equipment component, including: The temperature is deduced from the pixel value of the radiance by using the inverse function of Planck's blackbody radiation law. The radiance values ​​of each pixel in the compensated radiance map are combined with the emissivity of the component and substituted into the mapping relationship to calculate the temperature value corresponding to each pixel. The temperature values ​​of each pixel are arranged according to their original positions to form the temperature field of the device components.

9. The method for online temperature monitoring of power equipment based on infrared images according to claim 8, characterized in that, The temperature fields of various equipment components are spatially merged to generate an overall temperature distribution map of the power equipment. Based on the preset temperature tolerance criteria of each component, abnormal temperature areas are identified, including: The temperature fields of all the equipment components belonging to the same power equipment are spatially aligned and stitched together to generate an overall temperature distribution map of the equipment. Obtain the preset allowable temperature thresholds for each component of the power equipment, and search for pixel regions that exceed the corresponding maximum allowable temperature thresholds in the overall temperature distribution map of the equipment; The retrieved pixel regions are highlighted visually on the overall temperature distribution map of the device as areas of temperature anomalies.

10. An online temperature monitoring system for power equipment based on infrared images, characterized in that, The system includes: Infrared image acquisition module, used to acquire infrared image sequences of the area where the power equipment is located; The environmental parameter acquisition module is used to simultaneously collect ambient temperature, relative humidity, and wind speed to obtain environmental meteorological parameters; The image preprocessing module is used to perform noise reduction and contrast enhancement processing on the infrared images in the infrared image sequence to generate enhanced infrared images; The equipment region segmentation module is used to extract the main area of ​​the power equipment from the enhanced infrared image using the equipment segmentation model, and generate an equipment region image. The component identification and radiation extraction module is used to perform component-level semantic segmentation on the device area image, identify the category labels of each component, generate a component mask area, extract the pixel radiation brightness within the component mask area from the enhanced infrared image, and form the original radiation brightness map of the component. The emissivity retrieval module is used to retrieve the corresponding surface emissivity value based on the category label to obtain the component emissivity; The atmospheric compensation module is used to calculate the atmospheric transmittance and atmospheric path radiation based on the environmental meteorological parameters, and to perform atmospheric transmission inverse compensation on the original radiance map of the component using the atmospheric transmittance, atmospheric path radiation and the component emissivity to generate a compensated radiance map. The temperature calculation module is used to convert the radiance into a temperature value based on the compensated radiance map and the emissivity of the component, and generate a temperature field for the equipment component. The fusion diagnostic module is used to stitch together the temperature fields of all components of the power equipment into an overall temperature distribution map of the equipment, and to identify abnormal temperature areas in the overall temperature distribution map of the equipment.