A method and system for identifying defects in multi-voltage power grids using infrared images

Abstract

This invention relates to the fields of image recognition and power grid inspection technology, and discloses a method and system for identifying defects in infrared images of power grids at multiple voltage levels. The method includes: obtaining a standardized infrared image set by adaptive non-uniform correction and contrast enhancement of raw infrared images acquired from multiple sources; obtaining equipment target boxes and voltage level labels by multi-scale feature fusion target detection of the standardized images; obtaining defect regions and initial confidence levels by cross-voltage level temperature rise defect segmentation of the equipment target regions; obtaining calibrated parameters and final defect levels by environmentally sensitive temperature calibration and classification of the defect regions; and obtaining a defect list and inspection report associated with the defect results through automatic matching and report generation in a ledger. This invention achieves high-precision automatic identification and closed-loop archiving of infrared defects in power grids at all voltage levels through cascaded processing of image preprocessing, equipment detection, defect segmentation, temperature calibration, and ledger archiving, thereby improving defect identification accuracy and inspection efficiency.

Description

Technical Field

[0001] This invention relates to the fields of image recognition and power line inspection technology, and in particular to a method and system for identifying defects in infrared images of power grids at multiple voltage levels. Background Technology

[0002] Currently, infrared temperature measurement inspection of power grids is a core means of detecting equipment heating defects and ensuring the safe operation of the power system. With the development of computer vision and deep learning technologies, AI-based infrared defect recognition methods are gradually being applied to power grid inspection scenarios. However, existing technologies still have many shortcomings and cannot meet the high-precision inspection needs of power grids across all voltage levels.

[0003] First, multi-source infrared images lack a unified and standardized preprocessing mechanism. Current inspection data comes from various devices such as handheld thermal imagers, fixed monitoring terminals, and inspection drones. Different detectors have significant differences in non-uniform noise, imaging contrast, and dynamic range. Existing methods often directly input the raw images into the recognition model without adaptive correction for differences in imaging quality. This results in large fluctuations in the recognition accuracy of the model under different acquisition sources, and a significant increase in the false negative rate of low-quality images.

[0004] Secondly, the defect identification model is not adapted to the differences in characteristics across multiple voltage levels. From 10kV distribution network switchgear to 1000kV UHV tension fittings, the physical dimensions, heating mechanisms, defect morphologies, and temperature distribution characteristics of equipment at different voltage levels are fundamentally different. Existing generalized identification models struggle to account for both the weak temperature rise defects of small fittings and the large-area heating of large substation equipment. Single models exhibit poor generalization across voltage level scenarios, resulting in a persistently high rate of missed detection for small defects at high altitudes in UHV systems.

[0005] Third, the temperature measurement results lack accurate calibration considering multiple factors. Existing methods mostly use fixed compensation coefficients for simple temperature correction, without comprehensively considering the coupled effects of multiple parameters such as ambient temperature, shooting distance, atmospheric humidity, and equipment emissivity. This results in temperature measurement errors of several degrees Celsius in scenarios such as long-distance ultra-high voltage inspections and high humidity weather, directly affecting the accuracy and reliability of defect classification.

[0006] Fourth, defect identification and record management are disconnected, failing to form a complete closed-loop process. After AI identifies and outputs defect results, manual matching of equipment records, entry of defect lists, and compilation of inspection reports are still required. The archiving workload is large and prone to equipment mismatch and information omission. It is impossible to achieve full automation from image acquisition to report archiving, which restricts further improvement of inspection efficiency.

[0007] Therefore, there is an urgent need for a power grid infrared image defect identification method that can adapt to all voltage levels, perform unified preprocessing of multi-source images, accurately calibrate multiple environmental factors, and automatically associate records, so as to improve the accuracy of defect identification and the efficiency of inspection work, and meet the actual needs of intelligent inspection of power grids at all levels. Summary of the Invention

[0008] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a multi-voltage-level power grid infrared image defect identification method. This method aims to solve the technical problems in existing technologies, such as poor generalization of infrared defect identification models, insufficient adaptation to multiple voltage levels, large temperature measurement errors, and disconnect between identification and record keeping. In particular, the inability of UHV high-altitude fittings and low-voltage distribution network equipment to share a unified identification framework leads to a high rate of false negatives.

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for identifying defects in infrared images of power grids at multiple voltage levels.

[0010] The method for identifying defects in multi-voltage level power grid infrared images includes: Step S10: Acquire the original infrared images of the power grid equipment from multiple sources and the corresponding acquisition parameters, perform image preprocessing tasks using an adaptive non-uniformity correction and contrast enhancement mechanism, and output a standardized infrared image set; Step S20: Based on the standardized infrared image set, a multi-scale feature fusion target detection mechanism is used to perform the device localization and classification task, and output the device target box and corresponding device type and voltage level labels; Step S30: Based on the device target box and voltage level label, a cross-voltage level adapted temperature rise defect segmentation mechanism is used to perform the defect identification task, and the defect region mask and initial defect confidence score are output. Step S40: Based on the defect area mask and the collected environmental parameters, perform the defect classification task using an environmentally aware temperature calibration and classification mechanism, and output the calibrated defect temperature parameters and the final defect level; Step S50: Based on the final defect level and equipment label, use the ledger automatic matching and report generation mechanism to perform the archiving output task, and output the defect list and standardized inspection report after associating with the ledger.

[0011] Preferably, step S10, which involves acquiring the original infrared images of the power grid equipment from multiple sources and their corresponding acquisition parameters, performing image preprocessing tasks using an adaptive non-uniformity correction and contrast enhancement mechanism, and outputting a standardized infrared image set, specifically includes: Step S101: Collect raw infrared images from three sources: handheld thermal imager, fixed monitoring point, and inspection drone, and simultaneously acquire corresponding acquisition parameters, including ambient temperature, relative humidity, shooting distance, device emissivity, and acquisition device calibration parameters. Step S102: Correct the non-uniformity of the infrared detector based on the two-point correction method to eliminate fixed-mode noise caused by pixel response differences; Step S103: The grayscale enhancement of the corrected image is performed using the limited contrast adaptive histogram equalization algorithm. The enhancement amplitude is adaptively adjusted according to the average grayscale value of the image to avoid false defects caused by over-enhancement. Step S104: Perform structured annotation on the images according to the acquisition source and device scenario, and output a standardized infrared image set containing image data and acquisition metadata.

[0012] Preferably, step S20, which involves performing a device localization and classification task based on the standardized infrared image set using a multi-scale feature fusion target detection mechanism, and outputting the device target box along with corresponding device type and voltage level labels, specifically includes: Step S201: Construct a multi-scale feature fusion detection network, which fuses shallow detail features and deep semantic features through a feature pyramid structure to adapt to the detection needs of power grid equipment of different sizes; Step S202: Input the standardized infrared image into the detection network and output the target detection box, equipment type classification result and corresponding voltage level label for each device. The equipment types include tension pipe, current drain plate, current transformer, switch cabinet and transformer. Step S203: Use non-maximum suppression to remove redundant detection frames, retain the detection results with the highest confidence, and complete the accurate positioning and classification of the equipment.

[0013] Preferably, step S30, which involves performing a defect identification task based on the device target frame and voltage level label using a cross-voltage level adapted temperature rise defect segmentation mechanism, and outputting a defect region mask and initial defect confidence score, specifically includes: Step S301: Based on the identified equipment voltage level, adaptively call the defect segmentation weight model corresponding to the voltage level to form a cross-voltage level model adaptation mechanism; Step S302: Input the image region corresponding to the target box of the cropping device into the segmentation network, perform pixel-level segmentation on the abnormal temperature rise region, and generate a defect region mask; Step S303: Calculate the initial defect confidence level of the defect region:

[0014] in The average temperature of the defect area. The average background temperature in the normal operating area of ​​the same equipment. Assuming a weighted average based on temperature, To score the classification of the segmentation network, Weights for classification scores; Step S304: Output the defect region mask and the corresponding initial defect confidence level.

[0015] Preferably, in step S301, the defect segmentation weight model is trained separately for four voltage level ranges: 10kV distribution network, 35kV-220kV high voltage, 500kV-1000kV ultra-high voltage, and ±500kV-±800kV DC. The model weights are automatically switched according to the equipment voltage level to adapt to the differences in heating characteristics of equipment at different voltage levels.

[0016] Preferably, step S40, which involves performing a defect classification task based on the defect region mask and collected environmental parameters using an environmentally aware temperature calibration and classification mechanism, and outputting the calibrated defect temperature parameters and the final defect level, specifically includes: Step S401: Extract the ambient temperature, relative humidity, shooting distance and atmospheric attenuation coefficient from the collected parameters, and construct a temperature compensation calibration model; Step S402: Calculate the true temperature correction value of the defect area using the calibration model. :

[0017] in The measured temperature is from an infrared image. For ambient temperature, To calibrate the reference temperature, For shooting distance, Relative humidity, , and These are the calibration coefficients corresponding to ambient temperature, shooting distance, and relative humidity, respectively. Step S403: Calculate the true temperature of all pixels in the defect area after calibration, extract the highest temperature of the defect, and calculate the difference between the average true temperature of the defect area and the average true temperature of the normal background area of ​​the same device to obtain the relative temperature rise. Step S404: Based on the infrared defect classification standard for power grid equipment, and combined with the highest temperature of the defect after calibration, the relative temperature rise and the equipment voltage level, the defect is classified into three levels: general defect, major defect and emergency defect. Step S405: Output the calibrated defect temperature parameters and the final defect level, wherein the calibrated defect temperature parameters include the highest defect temperature and the relative temperature rise.

[0018] Preferably, step S50, which involves using an automatic ledger matching and report generation mechanism to perform an archiving output task based on the final defect level and equipment label, and outputting a defect list and standardized inspection report linked to the ledger, specifically includes: Step S501: Based on the equipment type, voltage level, and collection location information, automatically match it with the pre-stored power grid equipment ledger, and associate the equipment number, the substation / line to which it belongs, and the operation and maintenance team information; Step S502: Bind the defect temperature, defect level, and defect location to the corresponding equipment ledger information to generate a structured defect list; Step S503: Automatically generate a standardized test report that conforms to power grid operation and maintenance specifications. The report includes embedded metrological calibration supporting documents and defect infrared spectra, and supports batch export and classified archiving.

[0019] The present invention also provides a multi-voltage level power grid infrared image defect identification system, comprising: Image preprocessing module: used to acquire the original infrared images of power grid equipment from multiple sources and the corresponding acquisition parameters, and to perform image preprocessing tasks using an adaptive non-uniformity correction and contrast enhancement mechanism, and output a standardized infrared image set; Equipment detection and classification module: used to perform equipment positioning and classification tasks based on the standardized infrared image set using a multi-scale feature fusion target detection mechanism, and output equipment target boxes and corresponding equipment type and voltage level labels; Defect segmentation and identification module: used to perform defect identification tasks based on the device target box and voltage level label using a cross-voltage level adapted temperature rise defect segmentation mechanism, and output defect region mask and initial defect confidence score; Temperature calibration and grading module: used to perform defect grading tasks based on the defect area mask and collected environmental parameters using an environmentally aware temperature calibration and grading mechanism, and output the calibrated defect temperature parameters and the final defect grade; Ledger archiving output module: This module is used to perform archiving output tasks based on the final defect level and equipment label using an automatic ledger matching and report generation mechanism, and outputs a defect list and standardized inspection report associated with the ledger.

[0020] The present invention also provides a multi-voltage level power grid infrared image defect identification device, the multi-voltage level power grid infrared image defect identification device comprising: The system includes a memory, a processor, and a multi-voltage-level power grid infrared image defect identification program stored in the memory and executable on the processor. When the multi-voltage-level power grid infrared image defect identification program is executed by the processor, it implements the above method.

[0021] The present invention also provides a computer program product, the computer program product including a multi-voltage level power grid infrared image defect identification program, which implements the above method when executed by a processor.

[0022] The beneficial effects of this invention are as follows: This invention uses an adaptive non-uniformity correction and contrast enhancement image preprocessing mechanism to standardize infrared images acquired from multiple sources, eliminating inherent noise from the detector and differences in imaging quality between different acquisition sources. This provides high-quality, consistent input data for subsequent recognition, significantly improving the stability and overall accuracy of the recognition algorithm. This invention employs a multi-scale feature fusion device target detection mechanism that integrates shallow details and deep semantic features, while simultaneously addressing the detection needs of millimeter-level hardware and meter-level power equipment. It accurately locates the equipment area and simultaneously identifies the equipment type and voltage level, providing precise area trimming and labeling basis for subsequent voltage level defect identification. This invention employs a cross-voltage-level adaptive temperature rise defect segmentation mechanism, training dedicated models for four voltage level ranges and achieving adaptive weight switching. This accurately matches the differences in heating mechanisms and defect morphologies of equipment at different voltage levels, solving the problem of poor generalization of a single general model and significantly reducing the missed detection rate of minor defects in ultra-high voltage equipment and complex defects in low-voltage equipment. This invention utilizes a multi-factor temperature calibration mechanism based on environmental perception to construct a compensation model that integrates multiple parameters such as ambient temperature, shooting distance, and atmospheric humidity. This model corrects temperature measurement errors under different scenarios, improves the accuracy of temperature measurement, ensures the scientific and reliable nature of defect classification, and meets the precise diagnostic requirements of power grid operation and maintenance. This invention achieves intelligent association between defect results and equipment ledgers and one-click generation of standardized reports through an automatic ledger matching and report generation mechanism, forming a closed loop of the entire process of "image acquisition - defect identification - classification - archiving", which greatly reduces the workload of manual archiving and improves the overall efficiency and standardization of inspection work. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the first embodiment of a multi-voltage-level power grid infrared image defect identification method according to the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0025] 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.

[0026] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the method for identifying defects in infrared images of multi-voltage power grids according to the present invention. The first embodiment of the method for identifying defects in infrared images of multi-voltage power grids according to the present invention is presented.

[0027] In the first embodiment, the method for identifying defects in multi-voltage level power grid infrared images includes: Step S10: Acquire the original infrared images of the power grid equipment from multiple sources and the corresponding acquisition parameters, perform image preprocessing tasks using an adaptive non-uniform correction and contrast enhancement mechanism, and output a standardized infrared image set.

[0028] This step receives raw infrared images from three types of acquisition devices: handheld high-precision thermal imagers, station-based fixed monitoring terminals, and inspection drones. It simultaneously acquires metadata for each frame, including ambient temperature and humidity, shooting distance, device emissivity, acquisition device model, and calibration parameters. First, the non-uniformity of the infrared detector is corrected using a two-point correction method. The correction coefficient for each pixel is calculated using response data from two reference points at high and low temperatures, eliminating fixed-pattern stripe noise caused by pixel response differences. Then, a contrast-limited adaptive histogram equalization algorithm is used for grayscale enhancement. The algorithm divides the image into multiple local blocks and adaptively adjusts the contrast enhancement based on the average grayscale value of each block: increasing the enhancement for overall dark, long-distance aerial images and decreasing the enhancement for uniformly distributed, close-range images within the station to avoid over-enhancement and false hotspots. Finally, the images and metadata are structured and encapsulated according to the acquisition source and device scenario, outputting a standardized infrared image set in a unified format.

[0029] The core of this preprocessing is to unify infrared images from different sources and of different qualities under a consistent imaging standard, eliminating the interference of differences in front-end acquisition equipment on subsequent recognition algorithms and solving the problem of large accuracy fluctuations when recognizing mixed multi-source data. After correction and enhancement, the contrast of weak temperature rise defects in the image is significantly improved, and detector noise is effectively suppressed, providing high-quality input for subsequent target detection and defect segmentation.

[0030] Traditional infrared recognition often uses raw images directly without targeted preprocessing, resulting in low recognition rates for low-contrast images taken from long distances by drones. Furthermore, detector noise on handheld devices is easily misidentified as minor defects. This step uses adaptive preprocessing to improve image quality without sacrificing accurate defect information, ensuring that the subsequent recognition model maintains stable high accuracy across various acquisition scenarios.

[0031] For example, during the annual inspection of the 220kV transmission line in Zhenjiang, the infrared images of hardware taken by drones at high altitudes had low overall contrast and blurred details, while the images of equipment inside the station taken by handheld devices showed slight detector stripe noise. After this preprocessing step, the contrast of the weak temperature rise area of ​​the hardware in the drone images was improved by about 40%, and the stripe noise in the handheld images was completely eliminated. Both types of images achieved consistent standardized quality, laying the foundation for subsequent unified identification.

[0032] Step S20: Based on the standardized infrared image set, a multi-scale feature fusion target detection mechanism is used to perform the device localization and classification task, and output the device target box and corresponding device type and voltage level labels.

[0033] This step uses a standardized infrared image set as input to construct a multi-scale target detection network based on a feature pyramid. Through bottom-up feature extraction and top-down feature fusion, it outputs feature maps at different scales: shallow feature maps retain high-resolution details for detecting small fittings such as tension pipes and current drain plates; deep feature maps contain strong semantic information for detecting large equipment such as switchgear and transformers. After inputting the standardized images into the network, the network outputs the bounding box coordinates, equipment type classification results, and corresponding voltage level labels for each device, covering common power grid equipment such as tension pipes, current drain plates, current transformers, voltage transformers, switchgear, and main transformers, with voltage levels covering the full range from 10kV to 1000kV. Finally, a non-maximum suppression algorithm is used to remove overlapping redundant detection boxes, retaining the detection result with the highest confidence for each device, completing the accurate localization and multi-label classification of the equipment, and outputting the device target box and type and voltage level labels, which are then passed to step S30 for defect segmentation.

[0034] By introducing multi-scale feature fusion detection, the algorithm can accurately detect various devices of vastly different sizes simultaneously in the same image, solving the dilemma of traditional single-scale detection networks where "large devices miss small targets, and large devices are inaccurately located in small target scenes." The synchronously output voltage level labels provide direct basis for the adaptive switching of models in subsequent steps, achieving seamless integration of detection and recognition.

[0035] Traditional equipment inspection often employs single-scale networks, optimized for specific scenarios, making it difficult to simultaneously adapt to the inspection needs of large equipment within a station and small line fittings. When applied across different scenarios, the inspection accuracy drops significantly. This step utilizes a multi-scale fusion architecture, enabling a single network to cover the inspection of all types and voltage levels of equipment, significantly improving generalization capabilities.

[0036] For example, in the same aerial infrared image containing a 220kV current-draining plate and surrounding tower structures, the detection network accurately located the current-draining plate area, which is only 2 cm in diameter. It also identified the equipment type as a current-draining plate and the voltage level as 220kV. The cross-sectional area ratio (CLI) between the detection frame and the actual area reached 0.92, with no missed or false detections. In site images containing multi-faceted switchgear, all switchgear was accurately located and classified, achieving a detection accuracy rate of 98.5%.

[0037] Step S30: Based on the device target box and voltage level label, a cross-voltage level adapted temperature rise defect segmentation mechanism is used to perform the defect identification task, and the defect region mask and initial defect confidence are output.

[0038] After receiving the equipment target bounding box and voltage level label, this step first automatically matches the corresponding defect segmentation model weights according to the voltage level: 10kV distribution network equipment uses low-voltage specific weights, 35kV-220kV uses high-voltage weights, 500kV and above uses ultra-high-voltage weights, and DC equipment uses DC specific weights, achieving adaptive model switching. Then, the equipment area image is cropped based on the target bounding box coordinates and input into the segmentation network with the corresponding weights. The encoder-decoder structure performs pixel-level segmentation of the temperature rise anomaly area, outputting a binarized defect area mask. Next, the initial defect confidence of the defect area is calculated.

[0039] in The average temperature of the defect area. The average background temperature in the normal operating area of ​​the same equipment. Assuming a weighted average based on temperature, To score the classification of the segmentation network, As a weight for the classification score, the reliability of the defect is evaluated by combining the relative temperature rise and the network classification score. Finally, the defect region mask and the initial defect confidence level are output for temperature calibration and grading in step S40.

[0040] By introducing a cross-voltage level model adaptation mechanism, defect segmentation no longer relies on a single general model. Instead, it is specifically optimized for the heating characteristics of each voltage level, accurately capturing the differences in defect morphology under different voltage levels. For example, defects in UHV fittings are mostly localized point-like high temperatures with large temperature rise gradients, while defects in low-voltage switchgear are mostly planar heating with gradual temperature rises. The model trained according to voltage levels can extract corresponding features in a targeted manner, significantly improving segmentation accuracy and recall.

[0041] Traditional single-model defect identification is prone to missing minor defects with slight temperature rises in ultra-high voltage scenarios, and misidentifying normal heating as defects in low-voltage scenarios, resulting in a persistently high overall false negative and false positive rate. This step, through voltage-level modeling and adaptive switching, significantly improves the identification accuracy across the entire voltage range without increasing computational complexity.

[0042] For example, when analyzing images of the current-draining plate of a 220kV power line in Zhenjiang, the system automatically applied high-voltage level segmentation weights to accurately segment three localized heating defects with an area of ​​less than 10 square millimeters. The initial confidence levels were 0.89, 0.82, and 0.76, respectively, all higher than the judgment threshold of 0.6. On-site verification confirmed the existence of all three defects, achieving a 100% recall rate with no false positives. The recognition performance was significantly better than that of general models.

[0043] Step S40: Based on the defect area mask and the collected environmental parameters, perform the defect classification task using an environmentally aware temperature calibration and classification mechanism, and output the calibrated defect temperature parameters and the final defect level.

[0044] After receiving the defect area mask, this step extracts the original temperature measurement data for the corresponding area. Simultaneously, it retrieves parameters such as ambient temperature, shooting distance, and relative humidity from the acquired metadata, and substitutes these parameters into a multi-factor temperature calibration model to calculate the true temperature. The calibration model then calculates the corrected true temperature value for the defect area. :

[0045] in The measured temperature is from an infrared image. For ambient temperature, For shooting distance, Relative humidity, , and These are the calibration coefficients corresponding to ambient temperature, distance, and humidity, respectively. After obtaining the calibrated true temperature, the temperature rise difference between the defect and the normal area of ​​the same equipment is calculated. Combining the equipment voltage level and defect type, the defects are classified into three levels: general defects, major defects, and emergency defects. Finally, the highest temperature, average temperature rise, and final defect level of the calibrated defect are output and passed to step S50 for archiving.

[0046] After introducing multi-factor temperature calibration, the temperature measurement results are no longer significantly affected by environmental and shooting conditions. The temperature measurement error under different scenarios can be controlled within ±1℃, ensuring the consistency and accuracy of defect classification. The classification process strictly adheres to power grid industry standards and combines voltage level-differentiated judgment thresholds to ensure that the classification results meet the actual operation and maintenance requirements.

[0047] Traditional temperature measurements often rely directly on raw equipment readings without refined environmental compensation. This leads to significant temperature deviations in scenarios such as high summer temperatures and long-distance aerial photography, resulting in issues like over-classifying minor defects and under-classifying major defects in defect grading. This step, through an environmentally aware calibration model, significantly improves the accuracy of temperature measurements and the reliability of grading.

[0048] For example, a 220kV drain plate defect was observed by a drone at a distance of 50 meters, with an ambient temperature of 30℃ and relative humidity of 70%. The initial temperature measurement was 52℃. After calibration and model correction, the actual temperature was 56.8℃, with a relative temperature rise of 26.8K, which, according to standards, constitutes a major defect. If the original temperature were used directly for classification, it would be misclassified as a minor defect. The calibration result deviates from the on-site close-range measurement by only 0.5℃, demonstrating accurate and reliable classification.

[0049] Step S50: Based on the final defect level and equipment label, use the ledger automatic matching and report generation mechanism to perform the archiving output task, and output the defect list and standardized inspection report after associating with the ledger.

[0050] After receiving the defect classification results and equipment tags, this step automatically matches them with the pre-stored power grid equipment ledger database in the cloud, based on equipment type, voltage level, collected GPS coordinates or station interval information. It associates the corresponding equipment number, its associated line / substation, maintenance team, last inspection time, and other ledger information. Once matching is complete, the defect location, temperature parameters, defect level, and ledger information are bound together to generate a structured defect list. Finally, the system automatically generates a standardized inspection report conforming to power grid operation and maintenance specifications. The report includes an inspection overview, detailed defect statistics, defect infrared images, metering calibration supporting documents, and rectification suggestions. It supports single defect export and batch export of the entire station, automatically archiving the reports to the corresponding equipment ledger, completing a closed loop from identification to archiving.

[0051] The introduction of an automatic ledger matching and report generation mechanism completely replaces the tedious work of manually entering ledger data and compiling reports, avoiding manual mismatches and omissions, and significantly improving archiving efficiency and data accuracy. The standardized report format also ensures consistent output across different inspection projects, meeting the requirements of project acceptance and operation and maintenance management.

[0052] In traditional inspection work, report preparation and record archiving account for more than 40% of the total workload, and manual processing is prone to information errors and omissions. This step, through automated archiving, reduces the report preparation time for a single station from several hours to minutes, while ensuring data accuracy and traceability.

[0053] For example, during the inspection of the 220kV line in Zhenjiang, eight overheating defects were identified. After the system automatically matched the records, all defects were associated with the corresponding tower numbers and maintenance teams, generating a complete annual temperature measurement report with one click. The entire process took less than 10 minutes, which is significantly more efficient than the 3 days of manual compilation. Moreover, the report is standardized and complete, and can be directly used for maintenance acceptance.

[0054] Example 2: Furthermore, the present invention provides a multi-voltage-level power grid infrared image defect identification system, employing a multi-voltage-level power grid infrared image defect identification method from the above embodiments. This system addresses the technical problems in the prior art, such as poor generalization of infrared defect identification models, insufficient adaptation to multiple voltage levels, large temperature measurement errors, and disconnect between identification and record keeping. In particular, it addresses the high false negative rate caused by the inability to share a unified identification framework between ultra-high voltage high-altitude fittings and low-voltage distribution network equipment. The beneficial effects of the multi-voltage-level power grid infrared image defect identification system provided by the present invention are the same as those of the multi-voltage-level power grid infrared image defect identification method provided in the above embodiments. Other technical features of the multi-voltage-level power grid infrared image defect identification system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0055] Example 3: This invention provides a multi-voltage-level power grid infrared image defect identification device. The device includes: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enable the at least one processor to perform the multi-voltage-level power grid infrared image defect identification method described in Example 1. The multi-voltage-level power grid infrared image defect identification device in this invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. This multi-voltage-level power grid infrared image defect identification device is merely an example and should not limit the functionality or scope of the invention. The device may also include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes based on a program stored in a read-only memory or a program loaded from a storage device into a random access memory. The random access memory also stores various programs and data required for the operation of a multi-voltage-level power grid infrared image defect identification device. The processing unit, read-only memory, and random access memory are interconnected via a bus. The I / O interface is also connected to the bus. Typically, the following systems can be connected to the I / O interface: input devices including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices including, for example, magnetic tapes, hard disks, etc.; and communication devices. The communication device allows the multi-voltage-level power grid infrared image defect identification device to communicate wirelessly or wiredly with other devices to exchange data. Although a multi-voltage-level power grid infrared image defect identification device with various systems has been described, it should be understood that it is not required to implement or possess all the systems described. Alternatively, more or fewer systems can be implemented.

[0056] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the multi-voltage-level power grid infrared image defect identification method described above. The computer program product provided by this invention can solve the technical problems in the prior art, such as poor generalization of infrared defect identification models, insufficient adaptation to multiple voltage levels, large temperature measurement errors, and disconnect between identification and record keeping, especially the high false negative rate caused by the inability to share a unified identification framework between ultra-high voltage high-altitude fittings and low-voltage distribution network equipment. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the multi-voltage-level power grid infrared image defect identification method provided in the above embodiments, and will not be repeated here.

[0057] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a read-only memory. When the computer program is executed by a processing device, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0058] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0059] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the present invention and its equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A method for identifying defects in multi-voltage level power grid infrared images, characterized in that, The methods include: Step S10: Acquire the original infrared images of the power grid equipment from multiple sources and the corresponding acquisition parameters, perform image preprocessing tasks using an adaptive non-uniformity correction and contrast enhancement mechanism, and output a standardized infrared image set; Step S20: Based on the standardized infrared image set, a multi-scale feature fusion target detection mechanism is used to perform the device localization and classification task, and output the device target box and corresponding device type and voltage level labels; Step S30: Based on the device target box and voltage level label, a cross-voltage level adapted temperature rise defect segmentation mechanism is used to perform the defect identification task, and the defect region mask and initial defect confidence score are output. Step S40: Based on the defect area mask and the collected environmental parameters, perform the defect classification task using an environmentally aware temperature calibration and classification mechanism, and output the calibrated defect temperature parameters and the final defect level; Step S50: Based on the final defect level and equipment label, use the ledger automatic matching and report generation mechanism to perform the archiving output task, and output the defect list and standardized inspection report after associating with the ledger.

2. The method for identifying defects in multi-voltage level power grid infrared images according to claim 1, characterized in that, Step S10, which involves acquiring the original infrared images of the power grid equipment from multiple sources and their corresponding acquisition parameters, performing image preprocessing tasks using an adaptive non-uniformity correction and contrast enhancement mechanism, and outputting a standardized infrared image set, specifically includes: Step S101: Collect raw infrared images from three sources: handheld thermal imager, fixed monitoring point, and inspection drone, and simultaneously acquire corresponding acquisition parameters, including ambient temperature, relative humidity, shooting distance, device emissivity, and acquisition device calibration parameters. Step S102: Correct the non-uniformity of the infrared detector based on the two-point correction method to eliminate fixed-mode noise caused by pixel response differences; Step S103: The grayscale enhancement of the corrected image is performed using the limited contrast adaptive histogram equalization algorithm. The enhancement amplitude is adaptively adjusted according to the average grayscale value of the image to avoid false defects caused by over-enhancement. Step S104: Perform structured annotation on the images according to the acquisition source and device scenario, and output a standardized infrared image set containing image data and acquisition metadata.

3. The method for identifying defects in multi-voltage level power grid infrared images according to claim 1, characterized in that, In step S20, the step of performing equipment localization and classification tasks based on the standardized infrared image set using a multi-scale feature fusion target detection mechanism, and outputting the equipment target box and corresponding equipment type and voltage level labels, specifically includes: Step S201: Construct a multi-scale feature fusion detection network, which fuses shallow detail features and deep semantic features through a feature pyramid structure to adapt to the detection needs of power grid equipment of different sizes; Step S202: Input the standardized infrared image into the detection network and output the target detection box, equipment type classification result and corresponding voltage level label for each device. The equipment types include tension pipe, current drain plate, current transformer, switch cabinet and transformer. Step S203: Use non-maximum suppression to remove redundant detection frames, retain the detection results with the highest confidence, and complete the accurate positioning and classification of the equipment.

4. The method for identifying defects in multi-voltage level power grid infrared images according to claim 1, characterized in that, Step S30, which involves performing a defect identification task based on the device target frame and voltage level label using a cross-voltage level adapted temperature rise defect segmentation mechanism, and outputting a defect region mask and initial defect confidence score, specifically includes: Step S301: Based on the identified equipment voltage level, adaptively call the defect segmentation weight model corresponding to the voltage level to form a cross-voltage level model adaptation mechanism; Step S302: Input the image region corresponding to the target box of the cropping device into the segmentation network, perform pixel-level segmentation on the abnormal temperature rise region, and generate a defect region mask; Step S303: Calculate the initial defect confidence level of the defect region: in The average temperature of the defect area. The average background temperature in the normal operating area of ​​the same equipment. Assuming a weighted average based on temperature, To score the classification of the segmentation network, Weights for classification scores; Step S304: Output the defect region mask and the corresponding initial defect confidence level.

5. The method for identifying defects in multi-voltage level power grid infrared images according to claim 4, characterized in that, In step S301, the defect segmentation weight model is trained separately for four voltage level ranges: 10kV distribution network, 35kV-220kV high voltage, 500kV-1000kV ultra-high voltage, and ±500kV-±800kV DC. The model weights are automatically switched according to the equipment voltage level to adapt to the differences in heating characteristics of equipment at different voltage levels.

6. The method for identifying defects in multi-voltage level power grid infrared images according to claim 1, characterized in that, In step S40, the step of performing defect classification based on the defect region mask and the collected environmental parameters using an environmentally aware temperature calibration and classification mechanism, and outputting the calibrated defect temperature parameters and the final defect level, specifically includes: Step S401: Extract the ambient temperature, relative humidity and shooting distance from the collected parameters, and construct a temperature compensation calibration model; Step S402: Calculate the true temperature correction value of the defect area using the calibration model. : in The measured temperature is from an infrared image. For ambient temperature, To calibrate the reference temperature, For shooting distance, Relative humidity, , and These are the calibration coefficients corresponding to ambient temperature, shooting distance, and relative humidity, respectively. Step S403: Calculate the true temperature of all pixels in the defect area after calibration, extract the highest temperature of the defect, and calculate the difference between the average true temperature of the defect area and the average true temperature of the normal background area of ​​the same device to obtain the relative temperature rise. Step S404: Based on the infrared defect classification standard for power grid equipment, and in combination with the highest temperature of the defect after calibration, the relative temperature rise and the equipment voltage level, the defect level is classified. Step S405: Output the calibrated defect temperature parameters and the final defect level, wherein the calibrated defect temperature parameters include the highest defect temperature and the relative temperature rise.

7. The method for identifying defects in multi-voltage level power grid infrared images according to claim 1, characterized in that, In step S50, the step of executing the archiving output task based on the final defect level and equipment label using an automatic ledger matching and report generation mechanism, and outputting the defect list and standardized inspection report after association with the ledger, specifically includes: Step S501: Based on the equipment type, voltage level, and collection location information, automatically match it with the pre-stored power grid equipment ledger, and associate the equipment number, the substation / line to which it belongs, and the operation and maintenance team information; Step S502: Bind the defect temperature, defect level, and defect location to the corresponding equipment ledger information to generate a structured defect list; Step S503: Automatically generate a standardized test report that conforms to power grid operation and maintenance specifications. The report includes embedded metrological calibration supporting documents and defect infrared spectra, and supports batch export and classified archiving.

8. A multi-voltage level power grid infrared image defect identification system, applied to the multi-voltage level power grid infrared image defect identification method according to any one of claims 1 to 7, characterized in that, include: Image preprocessing module: used to acquire the original infrared images of power grid equipment from multiple sources and the corresponding acquisition parameters, and to perform image preprocessing tasks using an adaptive non-uniformity correction and contrast enhancement mechanism, and output a standardized infrared image set; Equipment detection and classification module: used to perform equipment positioning and classification tasks based on the standardized infrared image set using a multi-scale feature fusion target detection mechanism, and output equipment target boxes and corresponding equipment type and voltage level labels; Defect segmentation and identification module: used to perform defect identification tasks based on the device target box and voltage level label using a cross-voltage level adapted temperature rise defect segmentation mechanism, and output defect region mask and initial defect confidence score; Temperature calibration and grading module: used to perform defect grading tasks based on the defect area mask and collected environmental parameters using an environmentally aware temperature calibration and grading mechanism, and output the calibrated defect temperature parameters and the final defect grade; Ledger archiving output module: This module is used to perform archiving output tasks based on the final defect level and equipment label using an automatic ledger matching and report generation mechanism, and outputs a defect list and standardized inspection report associated with the ledger.

9. A multi-voltage level power grid infrared image defect identification device, characterized in that, The multi-voltage level power grid infrared image defect identification device includes: The system includes a memory, a processor, and a multi-voltage-level power grid infrared image defect identification program stored in the memory and executable on the processor. When the multi-voltage-level power grid infrared image defect identification program is executed by the processor, it implements a multi-voltage-level power grid infrared image defect identification method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a multi-voltage-level power grid infrared image defect identification program, which, when executed by a processor, implements a multi-voltage-level power grid infrared image defect identification method according to any one of claims 1 to 7.

Images