Power construction risk assessment method and system

Abstract

The invention provides an electric power construction risk assessment method and system, and relates to the technical field of electric power construction, and the method comprises the steps: determining a defect region of interest through obtaining a defect recognition image of a to-be-assessed electric power structure and corresponding infrared thermal imaging data, and analyzing the brightness information in the defect region of interest; judging whether the defect recognition image is qualified or not based on the analysis result; based on the judgment result, trust components of the defect recognition image analysis result and the infrared thermal imaging data analysis result in power construction risk assessment are adjusted, and a defect risk assessment conclusion of the to-be-assessed power structure is output. According to the method, the unreliability of visible light visual data caused by a novel high-reflective material can be intelligently identified, and decision components of different sensor information are adaptively adjusted, so that wrong underestimation of potential structure defect risks is avoided, the assessment accuracy and reliability in a complex inspection environment are remarkably improved, and the method is suitable for popularization and application. And the potential safety hazard caused by misjudgment is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of electric power construction, and particularly relates to an electric power construction risk assessment method and system. BACKGROUND

[0002] In the daily maintenance and management of modern electric power infrastructure, periodic inspection and risk assessment of those parts of the electric power structure that have been completed construction is a key link to ensure that the entire power grid system can operate safely and stably. In the past, we mainly relied on manual on-site visual inspection. This way is not only inefficient, but also easily affected by the subjective judgment of the inspector, and there are also great safety hazards when facing high altitude or complex environments. In order to solve these problems, the industry has generally begun to use robot inspection systems equipped with various sensors, hoping to collect data and assess risks through automation and refinement.

[0003] However, in actual application, the internal parameters and rules involved in the judgment method for identifying defects of electric power structures are mostly established and optimized according to a large amount of past collected inspection data. The past collected inspection data mainly comes from the electric power structures of the electric power towers or substations that have been running for a long time. The surface coating of these old structures naturally ages, and some slight rust or dust appears. The surface of the electric power structure tends to "diffuse reflection" in terms of the reflection characteristics of visible light. After the light is shone, it will spread evenly in all directions. This diffuse reflection characteristic allows the texture, color, and possible visual features of the electric power structure surface, such as small cracks and rust caused by loose bolts, to be clearly captured by high-definition visible light cameras. This provides stable and easily predictable input information for the defect identification method. By learning the relationship between these image features and actual defects, various electric power structure defect problems can be effectively identified and distinguished.

[0004] However, as the electric power infrastructure is continuously updated and expanded, newly completed electric power structures such as electric power towers or substations use the latest hot-dip galvanizing process for their key fastening bolts and connecting plates. This process provides very good corrosion resistance, but also makes the surface of these components present a very strong "mirror reflection" characteristic when they are first put into use. Compared with old structures, these new electric power structure surfaces are like mirrors, which can reflect incident light in a single direction, forming a very strong highlight area. The obvious difference in optical characteristics brings new challenges to defect identification based on visible light images.

[0005] When the inspection robot approaches to take pictures of the fastening bolts and connecting plates of the new power structure with new galvanized process, the shooting process will be strongly disturbed by the mirror reflection. The built-in program of the inspection robot will also automatically adjust to such situations. For example, the automatic exposure program adjusts the exposure parameters by analyzing the brightness distribution in the image to avoid large areas of overexposure or underexposure in the image. When there is a very bright mirror reflection point in the picture, in order to prevent the highlight area from becoming completely "dead white", the automatic exposure program will greatly reduce the overall exposure. Although the above compensation mechanism protects the highlight area from complete distortion, it causes other non-reflection areas in the image, especially the key details between the fastening bolts and the connecting plates, to become severely underexposed. At the same time, the working principle of the automatic focusing system is usually to find the edge or texture with the highest contrast in the image to determine the focus, which may be misled by strong and clear highlight reflection points, and incorrectly lock the focus on these reflection spots instead of the fastening bolt body or its joint surface with the connecting piece, resulting in the structure features that need to be detected in the image becoming unclear.

[0006] When dealing with low-quality images caused by the above reasons, the defect recognition method established according to the collected inspection data in the past cannot effectively extract the key visual features to judge whether the fastening bolt is loose. When facing the new power structure, even if there is potential looseness or defects, it often gives false judgments.

[0007] However, in addition to the high-definition visible light camera, the inspection robot is also equipped with an infrared thermal imager to assist in detecting the abnormal temperature rise of the connecting point caused by excessive resistance due to poor contact. The infrared thermal imager measures temperature by detecting the infrared radiation emitted by the surface of the object, which is less affected by the surface reflection characteristics. When detecting the same high-reflective connecting point, the infrared thermal imager reading shows that the temperature of this connecting point is different from other normal connecting points, and the temperature difference has exceeded the temperature fluctuation range in the normal operating state, indicating that there may be an increase in contact resistance due to poor fastening, which in turn causes local heating. This forms a logical contradiction in judgment: the main visual system associated with the high-definition visible light camera reports no defects due to image quality problems, while the thermal imaging system associated with the infrared thermal imager sends an abnormal signal.

[0008] The current risk assessment method usually takes the visual recognition result as the main basis for judgment and gives the highest trust degree when dealing with such information from different sources, and only takes the thermal imaging reading as secondary reference or auxiliary verification information. When the visual system clearly outputs the conclusion of no defect, the attention to abnormal signals in the thermal imaging data is greatly reduced, or even directly ignored. The core problem of this processing mechanism is the lack of an intelligent judgment method to understand the specific situation that high reflectivity may lead to unreliable visual information. Without being able to flexibly adjust the components or trust degree of different sensor data according to the specific situation, the system may still give a wrong conclusion after comprehensive evaluation, ignoring the potential safety hazards revealed by the thermal imaging data, thereby delaying the discovery and processing of actual defects and burying safety hazards. SUMMARY

[0009] To this end, the present application proposes a power construction risk assessment method and system, which aims to intelligently identify the unreliability of visible light visual data caused by new high-reflectivity materials, and adaptively adjust the decision components of different sensor information, thereby avoiding the false underestimation of potential structural defect risks, significantly improving the evaluation accuracy and reliability in complex inspection environments, and effectively reducing the safety hazards caused by misjudgment.

[0010] The first object of the present application is to propose a power construction risk assessment method, comprising: obtaining a defect recognition image and corresponding infrared thermal imaging data of the power structure to be evaluated, determining a defect attention area in the defect recognition image, and analyzing the brightness information in the defect attention area; judging whether the defect recognition image is qualified based on the analysis result of the brightness information of the defect attention area; setting a trust component range of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment, adjusting the trust component of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment based on the judgment result of the defect recognition image; determining the risk score of the power structure to be evaluated based on the adjusted trust component of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment, and outputting the defect risk assessment conclusion of the power structure to be evaluated.

[0011] Among them, obtaining a defect recognition image and corresponding infrared thermal imaging data of the power structure to be evaluated, determining a defect attention area in the defect recognition image, and analyzing the brightness information in the defect attention area, comprising: collecting the defect recognition image of the power structure to be evaluated by the high-definition visible light camera carried by the inspection robot, and collecting the infrared thermal imaging data of the power structure to be evaluated by the thermal imager carried by the inspection robot; The defect recognition image is analyzed to determine the proportion of the pixel points with the highest brightness and the lowest brightness. The defect recognition image is analyzed to determine the proportion of the pixel points with the highest brightness and the lowest brightness.

[0012] The defect recognition image is analyzed to determine the proportion of the pixel points with the highest brightness and the lowest brightness. The proportion of the pixel points with the highest brightness = (the number of the pixel points with the highest brightness / the total number of the pixel points in the defect focus area) x 100% The proportion of the pixel points with the lowest brightness = (the number of the pixel points with the lowest brightness / the total number of the pixel points in the defect focus area) x 100% The pixel point with the highest brightness is a pixel point with a pixel value of 255, and the pixel point with the lowest brightness is a pixel point with a pixel value of 0.

[0013] Based on the analysis result of the brightness information of the defect focus area, it is determined whether the defect recognition image is qualified, including: A qualified judgment threshold is set, wherein the qualified judgment threshold includes a highest brightness pixel point qualified judgment threshold and a lowest brightness pixel point qualified judgment threshold; It is determined whether the proportion of the pixel points with the highest brightness is greater than the highest brightness pixel point qualified judgment threshold, or whether the proportion of the pixel points with the lowest brightness is greater than the lowest brightness pixel point qualified judgment threshold; If it is determined that the proportion of the pixel points with the highest brightness is greater than the highest brightness pixel point qualified judgment threshold, or that the proportion of the pixel points with the lowest brightness is greater than the lowest brightness pixel point qualified judgment threshold, then the defect recognition image is determined to be an unqualified image.

[0014] The trust component range of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment is set, and the trust component of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment is adjusted based on the judgment result of the defect recognition image, including: The analysis result of the defect recognition image in the power construction risk assessment is set as a first trust component, and the value range of the first trust component is set; The analysis result of the infrared thermal imaging data in the power construction risk assessment is set as a second trust component, and the value range of the second trust component is set; If the defect recognition image is determined to be qualified, the first trust component is taken as the maximum value of the first trust component value range, and the second trust component is taken as the minimum value of the second trust component value range; If the defect recognition image is determined to be unqualified, the first trust component is taken as the minimum value of the first trust component value range, and the second trust component is taken as the maximum value of the second trust component value range.

[0015] wherein, based on the adjusted defect identification image and the infrared thermal imaging data in the trust component of the power construction risk assessment, the risk score of the power structure to be evaluated is determined, and the defect risk assessment conclusion of the power structure to be evaluated is output, including: a preset defect identification risk score interval and an infrared thermal imaging risk score interval; based on the defect identification image of the power structure to be evaluated, the defect identification risk score value of the defect identification image is determined; based on the infrared thermal imaging data of the power structure to be evaluated, the infrared thermal imaging risk score value of the infrared thermal imaging data is determined; based on the determined defect identification risk score value and infrared thermal imaging risk score value, combined with the first trust component value and the second trust component value, the total risk score of the power construction risk assessment of the power structure to be evaluated is determined; based on the total risk score, the defect risk assessment conclusion of the power structure to be evaluated is determined and output.

[0016] wherein, based on the determined defect identification risk score value and infrared thermal imaging risk score value, combined with the first trust component value and the second trust component value, the total risk score of the power construction risk assessment of the power structure to be evaluated is determined, and the total risk score formula is expressed as: total risk score=(infrared thermal imaging risk score value x second trust component value)+(defect identification risk score value x first trust component value) wherein, the infrared thermal imaging risk score value is determined according to the temperature rise deviation detected by the thermal imager; the defect identification risk score value is determined according to the defect possibility identified by the high-definition visible light camera in the limited clear area.

[0017] wherein, before the steps of obtaining the defect identification image and the corresponding infrared thermal imaging data of the power structure to be evaluated, including: moving along the preset inspection path, collecting surface temperature data of all key detection points in the power structure to be evaluated, and calculating the operating temperature reference of each key detection point under normal operating state; comparing the surface temperature data of each key detection point with the corresponding operating temperature reference; if the temperature difference between the surface temperature data of one key detection point and the operating temperature reference is greater than the preset temperature difference threshold, the key detection point is marked as a potential attention point; otherwise, the key detection point is marked as a low risk point, and the defect attention area in each defect identification image containing the potential attention point is analyzed.

[0018] wherein, for each defect focus area in the defect recognition image containing a potential focus point, analysis is performed, including: By adjusting the shooting angle of the high-definition visible light camera of the inspection robot, the defect recognition image of each key detection point marked as a potential focus point is collected and image processed; The defect recognition image after image processing is subjected to information quality evaluation index calculation; the information quality evaluation index includes edge sharpness score and texture richness score; wherein, The edge sharpness score is obtained by calculating the average gradient amplitude of all pixel points contained in the potential focus point in the defect recognition image; the gradient amplitude formula is represented as: Gradient amplitude = sqrt(Gx 2 + Gy 2 ) Edge sharpness score = average gradient amplitude = (sum of all pixel gradient amplitudes) / (total number of pixels) The texture richness score is obtained by analyzing the entropy value of the local binary pattern (LBP) feature of the potential focus point in the defect recognition image; the higher the entropy value, the richer the detailed texture information contained in the image; the texture richness score formula is represented as: Texture richness score = -sum(p(i) * log2(p(i))) Where p(i) is the probability of occurrence of LBP mode i; Based on the calculated information quality evaluation index, it is judged whether the defect recognition image containing the potential focus point is qualified.

[0019] A second object of the present application is to provide a power construction risk assessment system, comprising: An analysis module is configured to obtain defect recognition images and corresponding infrared thermal imaging data of a power structure to be evaluated, determine defect focus areas in the defect recognition images, and analyze brightness information in the defect focus areas; A judgment module is configured to judge whether the defect recognition images are qualified based on the analysis results of the brightness information in the defect focus areas; An adjustment module is configured to set a trust component range of the defect recognition images and the infrared thermal imaging data in power construction risk assessment, adjust the trust component of the defect recognition images and the infrared thermal imaging data in power construction risk assessment based on the judgment results of the defect recognition images; An output module is configured to determine a risk score of the power structure to be evaluated based on the trust component of the adjusted defect recognition images and the infrared thermal imaging data in power construction risk assessment, and output a defect risk assessment conclusion of the power structure to be evaluated.

[0020] Different from the prior art, the power construction risk assessment method of the present application, by acquiring the defect identification image and the corresponding infrared thermal imaging data of the power structure to be evaluated, determining the defect attention area in the defect identification image, and statistically analyzing the brightness information in the defect attention area; based on the statistical analysis result of the brightness information of the defect attention area, judging whether the defect identification image is qualified; based on the judgment result of the defect identification image, adjusting the trust component of the analysis result of the defect identification image and the infrared thermal imaging data analysis result in the power construction risk assessment; based on the adjusted trust component, outputting the defect risk assessment conclusion of the power structure to be evaluated. The purpose is to intelligently identify the unreliability of visible light visual data caused by new high-reflective materials, and in this specific situation, adaptively adjust the decision component of different sensor information, so as to avoid the false underestimate of the potential structural defect risk, significantly improve the evaluation accuracy and reliability in complex inspection environment, and effectively reduce the safety hidden danger caused by misjudgment.

[0021] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0022] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which: Figure 1 is a flow diagram of a power construction risk assessment method provided by the present application.

[0023] Figure 2 is a structural diagram of a power construction risk assessment system provided by the present application. DETAILED DESCRIPTION

[0024] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0025] As shown in Figure 1 , the present application provides a power construction risk assessment method, comprising: S110: acquiring the defect identification image and the corresponding infrared thermal imaging data of the power structure to be evaluated, determining the defect attention area in the defect identification image, and analyzing the brightness information in the defect attention area.

[0026] In the actual power structure construction inspection process, the construction risk of the power structure needs to be evaluated. The evaluation method is usually to take the power structure image as the evaluation object for evaluation, and the focus is on the fastening bolts and their connecting plates in the power structure. The power structure mentioned in this embodiment refers to power towers and substations.

[0027] In the embodiment of the present application, the inspection robot collects inspection data of the completed power structure, specifically, the high-definition photographing device carried by the inspection robot collects images of the completed power structure, and the collected images are used as defect recognition images. In addition, the infrared thermal imager carried by the inspection robot collects infrared thermal imaging data of the completed power structure.

[0028] After the defect recognition image is collected, the inspection robot processes the defect recognition image according to a preset program through the image processing module built-in. The purpose of image processing is to identify the defect focus area in the defect recognition image. In this application, the defect focus area is the area containing the fastening bolts and their surrounding connecting plates in the defect recognition image. After identifying the defect focus area in the defect recognition image, the image processing module marks the defect focus area. The marking methods include but are not limited to line frame marking and different color marking.

[0029] After marking the defect focus area, the pixel points in the defect focus area are analyzed to determine the proportion of the pixel points with the highest brightness and the lowest brightness. Specifically, the proportion of the pixel points with the highest brightness and the lowest brightness is calculated by the following formula: Proportion of pixel points with the highest brightness = (number of pixel points with the highest brightness / total number of pixel points in the defect focus area) x 100% Proportion of pixel points with the lowest brightness = (number of pixel points with the lowest brightness / total number of pixel points in the defect focus area) x 100% Wherein, the pixel point with the highest brightness is the pixel point with a pixel value equal to 255, and the pixel point with the lowest brightness is the pixel point with a pixel value equal to 0.

[0030] In the embodiment of the present application, the pixel point with the highest brightness is defined as the pixel point with a pixel value equal to 255. In digital image processing, especially in common 8-bit grayscale images or color images with 8-bit per color channel, 255 is the maximum brightness value that a pixel can represent, representing pure white or fully saturated brightness. Therefore, setting the pixel point with the highest brightness to have a pixel value equal to 255 means identifying the areas in the recognition image that have the highest brightness.

[0031] Setting the pixel value of the pixel point with the highest brightness to be equal to 255 has the following significant benefits: Accurate identification of overexposed or saturated areas in images: As mentioned earlier, the newly constructed power structures employ a hot-dip galvanizing process, which results in a strong "mirror reflection" characteristic on their surfaces. When the high-definition visible light camera mounted on the inspection robot captures these highly reflective components, the automatic exposure program may significantly reduce the overall exposure to avoid the high-light areas becoming completely "dead white" (i.e., the pixel value reaches the maximum, losing all details). However, even after exposure adjustment, strong mirror reflection can still cause the pixel value of local areas to reach the saturation state, i.e., the pixel value equals 255. By defining the highest brightness pixel point as equal to 255, the overexposed or saturated areas caused by mirror reflection can be accurately captured.

[0032] Effective judgment of the loss of key detail information: When the pixel value in the image reaches 255, the image detail information in that area is completely lost, forming a dead white area. In high-light areas, all details are smoothed out due to pixel value saturation, and the method mistakenly considers it as a smooth surface without any features. These lost details may include small gaps between bolt edges and gaskets, or specific rust lines caused by loosening, which are key visual features for determining whether the bolt is loose. By identifying areas where the pixel value equals 255, it can be directly determined whether these key details can be effectively identified due to overexposure, thereby providing direct evidence for subsequent image eligibility judgment.

[0033] Provide objective and quantitative image quality evaluation criteria: Defining the highest brightness pixel point as equal to 255 provides a clear, objective, and quantifiable standard for evaluating image quality. This value is a fixed upper limit in digital image processing and is not affected by subjective judgment, making the image quality evaluation process standardized and automated. By counting the proportion of these pixel points in the defect focus area, a quantitative index can be obtained for comparison with the pre-set eligibility threshold, thereby avoiding the subjectivity and inconsistency of manual visual inspection.

[0034] Lay the foundation for subsequent trust component adjustment: Identifying areas in the image with a large number of pixel values equal to 255 is one of the key criteria for determining whether the defect recognition image is eligible. Once the image is determined to be unqualified, the trust component of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment needs to be adjusted, i.e., reducing the trust component of the visible light image and increasing the trust component of the infrared thermal imaging data. This dynamic adjustment mechanism of the trust component based on image quality can effectively solve the problem mentioned in the background technology that "there is a lack of an intelligent judgment method to understand that high reflectivity can lead to unreliable visual information in this specific situation", ensuring that when the visible light image is unreliable, the infrared thermal imaging data can be relied upon for risk assessment, thereby avoiding the false underestimation of potential safety hazards.

[0035] S120: Determine whether the defect recognition image is qualified based on the analysis result of the defect attention region brightness information.

[0036] In the step of determining whether the image is qualified, a qualified judgment threshold is first set; wherein the qualified judgment threshold includes a highest brightness pixel point qualified judgment threshold and a lowest brightness pixel point qualified judgment threshold. It is determined whether the highest brightness pixel point proportion is greater than the highest brightness pixel point qualified judgment threshold, or whether the lowest brightness pixel point proportion is greater than the lowest brightness pixel point qualified judgment threshold. If it is determined that the highest brightness pixel point proportion is greater than the highest brightness pixel point qualified judgment threshold, or that the lowest brightness pixel point proportion is greater than the lowest brightness pixel point qualified judgment threshold, it is determined that the defect recognition image is an unqualified image.

[0037] In an embodiment of the present application, the internal preset qualified judgment logic of the inspection robot system is set, that is, the qualified judgment threshold is set. The qualified judgment threshold in this embodiment is determined according to a large number of tests on new high-reflective materials. For example, if the calculated highest brightness pixel point proportion exceeds 10%, or the lowest brightness pixel point proportion exceeds 30%, it is determined that the current defect recognition image is unqualified. Such unqualification is usually due to the mirror reflection characteristics of new hot-dip galvanizing materials, resulting in local area overexposure or large area underexposure. At this time, the control system inside the robot will immediately generate an explicit internal marker, for example, set a Boolean state variable named “VisualDataCompromised” to “true”, indicating that the current visual data is unreliable and is not suitable as the main basis for judgment.

[0038] When it is determined that the defect recognition image is unqualified, a marker is set for the unqualified defect recognition image to identify the defect recognition image as an unqualified image.

[0039] As mentioned above, if the calculated pure white pixel proportion exceeds 10% or the pure black pixel proportion exceeds 30%, the system will determine that the current visible light image is unqualified and consider that the visual information is polluted, which embodies the specific standard and result of quality judgment of the visible light image. If the proportion of pure white pixels exceeds 10% or the proportion of pure black pixels exceeds 30%, it is determined that the image quality does not meet the requirements and is marked as visual information pollution. Such marking indicates that the current visual data is unreliable and is not suitable as the main basis for defect judgment.

[0040] The upper limit of the proportion of pure white pixels is set to 10% and the upper limit of the proportion of pure black pixels is set to 30%, which is based on the observation and test of the optical characteristics of the new hot-dip galvanizing material in actual application. The new hot-dip galvanizing process makes the surface of the power structure component have strong mirror reflection characteristics. When the high-definition camera carried by the inspection robot shoots these high-reflective components, mirror reflection will cause two main distortion phenomena in the image: First, very strong highlight areas will appear in local areas, causing the pixel value to reach the maximum saturation state, i.e. pure white pixels, and the details of these areas will be lost.

[0041] Second, to avoid complete distortion of these highlight areas, the automatic exposure program of the camera will reduce the overall exposure, which causes other non-reflective areas in the image, especially the key details between the bolts and the connecting plates, to become severely underexposed, appearing as pure black pixels, and the details of these areas are also difficult to identify.

[0042] The traditional defect recognition method is established according to the data of old structures (surface diffuse reflection characteristics) and cannot effectively handle such image distortion caused by mirror reflection. In the highlight saturated area, the defect features are smoothed; in the dark underexposed area, the defect features are submerged in the signal. Therefore, when the proportion of pure white or pure black pixels in the image reaches a certain degree, the visual data loses its value as a reliable basis for defect recognition.

[0043] The two specific percentage thresholds of 10% and 30% are determined after a large number of tests on new high-reflective materials. These tests aim to find a critical point, i.e. when the proportion of brightness saturated areas (pure white or pure black pixels) in the image exceeds these thresholds, the existing defect recognition method based on visible light images cannot effectively extract key visual features such as loose bolts, thus giving incorrect judgments of no defects or low risk. Therefore, these thresholds represent a quantitative standard for the quality of visual data in a specific scenario, which is sufficient to cause the failure of the defect recognition method.

[0044] S130: Set the trust component range of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment, based on the judgment result of the defect recognition image, adjust the trust component of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment.

[0045] In this embodiment, the trust component of the analysis result of the defect recognition image in the power construction risk assessment is set as the first trust component, and the value range of the first trust component is set. The trust component of the analysis result of the infrared thermal imaging data in the power construction risk assessment is set as the second trust component, and the value range of the second trust component is set. If the defect recognition image is determined to be qualified, the first trust component is taken as the maximum value of the first trust component value range, and the second trust component is taken as the minimum value of the second trust component value range; if the defect recognition image is determined to be unqualified, the first trust component is taken as the minimum value of the first trust component value range, and the second trust component is taken as the maximum value of the second trust component value range.

[0046] As the minimum and maximum values of the preset values, they are not directly set, but are indirectly determined by first setting the "value range" of the first trust component and the second trust component. Once the value range of a certain trust component is determined, the minimum and maximum values in the range are also determined.

[0047] When setting the value range of the first trust component and the second trust component, the following aspects are usually considered comprehensively: First, based on experience and domain knowledge. In the field of power construction risk assessment, visible light images and infrared thermal imaging data each have their advantages and limitations. For example, visible light images have intuitive nature in identifying surface defects (such as cracks, corrosion, loose bolts), while infrared thermal imaging data have unique advantages in detecting temperature anomalies (such as poor contact, overload heating). According to the historical performance of these data sources in different scenarios and expert experience, the relative importance of their evaluation can be preliminarily determined, and a reasonable value range can be set. For example, if it is believed that the reliability of visible light images is higher under ideal conditions, the value range of their trust component can be set to [0.5, 1.0], while the value range of infrared thermal imaging data is [0, 0.5].

[0048] Second, based on data characteristics and reliability. For the first trust component, the setting of its value range will consider the information quality of the defect recognition image under different lighting, distance, clarity, etc. If the image quality fluctuates greatly, a wider range may need to be set to accommodate such changes. For the second trust component, the setting of its value range will consider the degree of influence of environmental temperature, emissivity, etc. on infrared thermal imaging data. Generally, infrared thermal imaging data are less affected by surface reflection characteristics, and may be more stable than visible light images in some cases, so their trust component will be given a higher weight when the image quality is poor.

[0049] Third, based on system design goals and risk preferences. System designers can adjust the value range of the trust component according to the sensitivity requirements of the risk assessment results. For example, if the system is very sensitive to any potential risk, even if the visible light image quality is poor, it is hoped that the infrared thermal imaging data can play a sufficient role, then the maximum value of the second trust component can be set relatively high, even equal to the maximum value of the first trust component.

[0050] Fourth, standardization and quantification. In order to facilitate calculation and understanding, the value range of the trust component is usually standardized to a specific interval, such as [0, 1] or [0, 100]. In this standardized interval, the minimum value is usually 0 (indicating complete distrust or zero weight), and the maximum value is usually 1 or 100 (indicating complete trust or the highest weight).

[0051] In summary, the minimum and maximum values as pre-set values are determined according to the pre-set value ranges of the first and second trust components. The setting of these value ranges is a comprehensive decision-making process that needs to consider field experience, data characteristics, system goals, and standardization requirements.

[0052] In embodiments of the present application, the first trust component aims to quantify the credibility and importance of visible light images in risk assessment. Since visible light images can provide rich detailed information under ideal conditions, they play an irreplaceable role in identifying visual defects such as loose bolts, structural cracks, and surface corrosion. However, as described in the background art, when facing high-reflective materials or complex lighting environments, the quality of visible light images may be severely degraded, resulting in reduced accuracy of defect identification. Therefore, setting the value range of the first trust component is to define the maximum and minimum influence that visible light images may have in risk assessment.

[0053] The value range of the first trust component is usually set as a continuous interval, such as [0, 1] or [0%, 100%]. Among them, the upper limit of the range (maximum value) represents the dominant position that visible light images should occupy in total risk assessment when the quality of visible light images is excellent and the information is highly reliable; the lower limit of the range (minimum value) represents the minimum influence that visible light images should be given in total risk assessment when the quality of visible light images is poor and the information is almost unusable. By setting this range, the system can flexibly adjust the weight of visible light images in the final risk score calculation according to the actual quality of the image, thereby avoiding misjudgment caused by image quality problems.

[0054] The second trust component aims to quantify the credibility and importance of infrared thermal imaging data in risk assessment. Infrared thermal imaging data is primarily used to detect temperature anomalies, which are often early indicators of potential issues such as poor electrical connections, overloads, or mechanical friction, and its imaging principle is less affected by visible light environments. Therefore, infrared thermal imaging data can serve as an effective supplement to visible light images, especially when the quality of visible light images is limited. The value range of the second trust component is set to define the maximum and minimum influence that infrared thermal imaging data can play in risk assessment.

[0055] Similar to the first trust component, this value range is also generally set as a continuous interval, such as [0, 1] or [0%, 100%]. The upper limit (maximum value) of the range represents the highest weight that infrared thermal imaging data should be given when visible light images are unreliable, to make up for the lack of visible light information and ensure that potential thermal hazards are fully considered; the lower limit (minimum value) represents that when visible light images are highly reliable, the weight of infrared thermal imaging data as auxiliary information can be appropriately reduced. By setting this range, the system can ensure that infrared thermal imaging data can play a key role when visible light images fail, thereby improving the robustness of overall risk assessment.

[0056] The setting of the value range of the two trust components forms the basis of the intelligent fusion evaluation mechanism of the present application. They allow the system to dynamically and complementarily adjust the weights according to the pass / fail judgment results of the defect recognition image: When the defect recognition image is judged to be qualified, the first trust component takes the maximum value, and the second trust component takes the minimum value. This reflects that when the quality of visible light images is reliable, the system prefers to trust and rely on the detailed visual defect information provided by them.

[0057] When the defect recognition image is judged to be unqualified, the first trust component takes the minimum value, and the second trust component takes the maximum value. This indicates that when the quality of visible light images is damaged, the system shifts the focus of evaluation to infrared thermal imaging data, and uses its characteristics of being unaffected by light to capture potential temperature anomaly risks.

[0058] This dynamic adjustment mechanism, through the pre-set value range, ensures that the adjustment of the trust component has clear boundaries and sufficient flexibility, which can adapt to the characteristics of the inspection data of the power structure under different environments and states, thereby significantly improving the accuracy and reliability of the risk assessment of power construction. The specific value range (such as [0, 1]) and the specific values of the maximum and minimum values (such as 0.8 and 0.2) can be optimized and calibrated according to a large amount of historical data analysis, expert experience, sensor performance, and the expected sensitivity of the risk assessment results.

[0059] When the defect recognition image is judged as qualified, it indicates that the image quality is good and can provide reliable defect recognition information. At this time, in order to make full use of the high-quality image data, the first trust component is set to its maximum value. This means that the defect recognition image will dominate in the final risk assessment. Correspondingly, the second trust component is set to its minimum value to reduce the weight of the infrared thermal imaging data in the assessment, avoiding interference due to data redundancy or potential inconsistency.

[0060] When the defect recognition image is judged as unqualified, it indicates that the image quality has problems and may not be able to provide accurate defect information (for example, details are blurred due to high reflection, overexposure or underexposure). In this case, in order to avoid the negative impact of unqualified images on the assessment results, the first trust component is set to its minimum value, thereby greatly reducing its role in the assessment. At the same time, the second trust component is set to its maximum value, so that the infrared thermal imaging data plays a more important role in the assessment, because the infrared thermal imaging data is usually not affected by the quality problems of visible light images (such as lighting, blur, etc.), and can provide independent and reliable temperature anomaly information.

[0061] This setting method combines the quality judgment result of the defect recognition image with the dynamic adjustment mechanism of the trust component, realizing the adaptive management of the weight of different data sources in risk assessment. The system can intelligently decide whether to focus more on the detail recognition ability of visible light images or rely more on the temperature anomaly detection ability of infrared thermal imaging data.

[0062] When the visible light image quality is high, its rich detail information is crucial for identifying bolt loosening, component deformation and other defects, so it is given a high weight. When the visible light image quality is poor (for example, due to insufficient lighting, blur or mirror reflection, it cannot clearly identify defects), the system shifts the focus of the assessment to the infrared thermal imaging data, and uses its sensitivity to temperature anomalies to discover potential electrical overheating, poor contact and other problems. This dynamic adjustment mechanism ensures that in any case, risk assessment can be based on the most reliable data source, effectively avoiding the limitations that may exist in a single data source.

[0063] This adaptive weight adjustment mechanism makes the entire evaluation process more flexible and intelligent, and can better adapt to complex and variable field environments, such as in the inspection of new high-reflective material power structures, it can provide more accurate risk warning and decision support.

[0064] Specifically, in the subsequent step of structural defect risk assessment, if the "VisualDataCompromised" flag generated in the previous step is detected to be "true", this flag will be used as a mandatory trigger condition. The logic of risk assessment is temporarily and completely adjusted to the trust component of different information sources. Specifically, the system sets the second trust component of the infrared thermal imaging data analysis result to the highest, for example, the value can be 0.9, and the first trust component of the defect recognition image analysis result of the high-definition visible light camera is reduced to the lowest, for example, the value can be 0.1. By adjusting, it is ensured that when the visual image data quality is questionable, more reliance can be placed on thermal imaging data for judgment.

[0065] S140: Based on the adjusted trust components of the defect recognition image and the infrared thermal imaging data in the power construction risk assessment, determine the risk score of the power structure to be evaluated, and output the defect risk assessment conclusion of the power structure to be evaluated.

[0066] In this step, the defect recognition risk score interval and the infrared thermal imaging risk score interval are preset; based on the defect recognition image of the power structure to be evaluated, the defect recognition risk score value of the defect recognition image is determined; based on the infrared thermal imaging data of the power structure to be evaluated, the infrared thermal imaging risk score value of the infrared thermal imaging data is determined; based on the determined defect recognition risk score value and the infrared thermal imaging risk score value, combined with the first trust component value and the second trust component value, the total risk score of the power construction risk assessment of the power structure to be evaluated is determined; based on the total risk score, the defect risk assessment conclusion of the power structure to be evaluated is determined and output.

[0067] In this embodiment, the defect recognition risk score interval and the infrared thermal imaging risk score interval are the key link for realizing risk quantization and standardized assessment in the power construction risk assessment method. The setting is based on multiple considerations, aiming to provide a unified quantitative scale for risk information of different sources, and to provide a basis for subsequent intelligent fusion and decision-making.

[0068] The defect identification risk score interval and the infrared thermal imaging risk score interval are set to standardize and quantify heterogeneous risk information from high-definition visible light cameras (for defect identification images) and thermal imagers (for infrared thermal imaging data). In power construction risk assessment, different types of defects (such as structural damage and abnormal electrical connections) have different manifestations and potential hazards. By presetting a risk score interval for each data source, these heterogeneous information can be mapped to a unified numerical scale, for example, 0 to 100 points, where 0 represents no risk and 100 represents the highest risk. This standardization makes the risk assessment results of different data sources comparable, lays the foundation for subsequent weighted fusion to calculate the total risk score, and avoids evaluation difficulties or biases due to data type differences.

[0069] Specifically, the defect identification risk score interval is set for the risk assessed by the defect identification images collected by the high-definition visible light camera. The setting is based on the following considerations: Quantification of defect type and severity: visible light images can identify defects including but not limited to loose bolts, deformed connecting plates, surface corrosion, cracks, etc. The type, size, location and development trend of these defects all affect their risk level. For example, a small surface scratch and a deep crack represent completely different risks. Therefore, when setting the score interval, different defect types and their severity need to be carefully divided and quantified and mapped to specific numerical values within the score interval.

[0070] Defect likelihood and frequency: defect identification risk score is determined based on the likelihood of defects identified by high-definition visible light cameras in a limited clear area, which indicates that the score interval needs to reflect the probability of defect occurrence. For example, some structural parts have a higher defect likelihood due to design or material characteristics, which should be reflected in the score interval setting; in addition, the occurrence frequency of various defects in historical data is also an important reference.

[0071] Impact on power system operation: Different defects have different impacts on the safe and stable operation of the power system. For example, a slight surface corrosion may only affect aesthetics and long-term life, while a serious bolt loosening may lead to structural failure. The setting of the score interval should be able to distinguish these impacts and map defects with greater impact on the system to higher risk scores.

[0072] Performance of image recognition algorithm: Although this method solves the image quality problem through the trust component mechanism, when setting the defect identification risk score interval, the inherent capabilities and limitations of the visible light image recognition algorithm still need to be considered. For example, for some subtle defects that are difficult to identify, their risk scores may need to be set more carefully.

[0073] The infrared thermal imaging risk score interval is set for the risk assessed from the infrared thermal imaging data collected by the thermal imager. Its setting is based on the following considerations: Temperature rise deviation and fault correlation: The infrared thermal imaging risk score is determined according to the temperature rise deviation detected by the thermal imager, which is a key indicator of abnormal heating of power equipment and is usually closely related to electrical or mechanical faults such as poor contact, overload, and insulation aging. The setting of the score interval needs to establish a mapping relationship between the temperature rise deviation and the actual fault severity. For example, a slight temperature rise (such as 5°C) may represent early warning, while a significant temperature rise (such as more than 15°C) may indicate a serious fault.

[0074] Hazard of temperature anomaly: Different degrees of temperature anomaly cause different hazards to power equipment and systems. A slight temperature rise may only cause increased energy consumption, while severe overheating may cause equipment to burn, fire, or even widespread power outages. The score interval should reflect this level of damage, mapping a large temperature rise deviation to a higher risk score.

[0075] Environmental factors and operating benchmarks: Thermal imaging data is affected by environmental temperature, load conditions, and other factors. Therefore, when setting the score interval, the temperature rise deviation needs to be evaluated in combination with the operating temperature benchmark to ensure the objectivity of the risk score. The score interval should be able to adapt to temperature fluctuations under different operating conditions.

[0076] Analysis of historical thermal imaging data: By analyzing the frequency, duration, and final fault type and loss of temperature anomalies in historical thermal imaging data, empirical evidence and statistical support can be provided for the setting of the risk score interval.

[0077] The setting of the defect recognition risk score interval and the infrared thermal imaging risk score interval is to combine with the overall risk management strategy of power construction. For example, 0-30 can be set as low risk, 31-60 as medium risk, and 61-100 as high risk. These interval divisions should correspond to the actual maintenance response levels (such as regular inspection, immediate repair, and emergency shutdown), thereby providing clear and actionable guidance for decision-makers. The pre-set defect recognition risk score interval and infrared thermal imaging risk score interval enable the system to directly output a clear defect risk assessment conclusion based on the calculated total risk score, thereby improving the practicality and efficiency of risk assessment.

[0078] The final risk score calculation formula can be expressed as: Total risk score = (infrared thermal imaging risk score value x second trust component value) + (defect recognition risk score value x first trust component value) The infrared thermal imaging risk score is determined according to the temperature rise deviation detected by the infrared thermal imaging system, and the defect identification risk score is determined according to the possibility of defects identified by the high-definition visible light camera in the limited clear area.

[0079] In the power construction risk assessment method, the determination process of the infrared thermal imaging risk score and the defect identification risk score of the power structure to be evaluated is based on different data sources and evaluation standards, aiming to quantify the two main risk types that may exist in the power structure. Among them, The determination process of the infrared thermal imaging risk score is as follows: The infrared thermal imaging risk score is determined according to the temperature rise deviation detected by the thermal imager, and the temperature rise deviation refers to the difference between the surface temperature of the key detection point in the power structure and the operating temperature reference of the detection point in the normal operating state. When determining, Collect the surface temperature data of the power structure to be evaluated and determine the operating temperature reference: The inspection robot moves along the preset path and collects the surface temperature data of all key detection points in the power structure to be evaluated through the thermal imager or temperature sensor mounted. At the same time, according to historical data, design parameters or environmental conditions and other information, the operating temperature reference of each key detection point in the normal operating state is calculated or obtained.

[0080] Calculate the temperature deviation: Compare the surface temperature data collected by each key detection point with the corresponding operating temperature reference to calculate the temperature difference, i.e. the temperature rise deviation.

[0081] Map to risk score: According to the preset rules or model, the calculated temperature rise deviation is mapped to the preset infrared thermal imaging risk score interval. Generally, the larger the temperature rise deviation, the more serious the thermal anomaly, and the higher the corresponding infrared thermal imaging risk score. For example, a threshold can be set, and when the temperature rise deviation exceeds the threshold, the risk score will increase significantly.

[0082] The following is an embodiment of the infrared thermal imaging risk score setting: Suppose a fastening bolt connection point of a certain power structure, the surface temperature detected by the thermal imager is 70℃. According to the historical operating data, the operating temperature reference of this type of connection point in the normal operating state is 45℃. Then, the temperature rise deviation is 70℃ - 45℃ = 25℃.

[0083] The preset risk assessment rule is: the temperature rise deviation degree corresponds to the risk score 0-30 in 0-10℃, 31-60 in 10-20℃, and 61-100 above 20℃. The temperature rise deviation degree of 25℃ will result in a higher value of the infrared thermal imaging risk score of the connection point, for example, 85. By setting the infrared thermal imaging risk score, the quantitative processing of the infrared imaging risk is realized.

[0084] The determination process of the defect identification risk score value is as follows: The defect identification risk score is determined according to the possibility of defects identified by the high-definition visible light camera in the limited clear area. This involves analyzing the visible light image to identify and evaluate structural defects. Specifically, Obtain and process defect identification images: The inspection robot collects defect identification images of the power structure to be evaluated through the high-definition visible light camera. These images may be processed (such as denoising, enhancement) to improve the quality.

[0085] Determine the defect attention area: In the defect identification image, the defect attention area containing the key components such as the fastening bolt and the surrounding connecting plate is identified.

[0086] Evaluate image quality: The brightness information of the defect attention area is analyzed, and whether the image is qualified is judged according to indicators such as the proportion of the highest and lowest pixel points. If the image is not qualified, it may need to be re-collected or the subsequent evaluation strategy needs to be adjusted.

[0087] Identify defects and evaluate the possibility: In the defect attention area of the qualified defect identification image, the image recognition algorithm or manual interpretation is used to identify the possible defect types (such as corrosion, crack, loose bolt, deformation, etc.) and their severity.

[0088] Map to risk score: According to the identified defect type, severity and possibility of defects, combined with the preset evaluation model or rule, the defect information is mapped into the preset defect identification risk score interval. Generally, the more serious and the higher the possibility of defects, the higher the defect identification risk score value.

[0089] The following is an embodiment of the defect identification risk score value setting: Take the example of the above-mentioned fastening bolt connection point. The defect recognition image obtained by the high-definition visible light camera is determined to be qualified after image quality evaluation. By analyzing the defect attention area of the image, it is identified that there is slight rust marks on the surface of the bolt, but no obvious cracks or looseness is found. According to the preset defect evaluation standard: no obvious defect corresponds to risk score 0-20, slight rust corresponds to risk score 21-40, moderate rust or small cracks correspond to risk score 41-70, and serious defect corresponds to risk score 71-100. Then the defect recognition risk score of this connection point is 35 due to slight rust.

[0090] Through the above process, the infrared thermal imaging risk score and the defect recognition risk score are quantified from the thermal and visual dimensions respectively, which provides basic data for subsequent calculation of the total risk score combined with the trust component.

[0091] Specifically, based on the adjusted information trust component, the weak abnormal signal that may exist in the thermal imaging data is preferentially adopted. For example, even if the visible light image is blurred due to high reflection, but if the thermal imaging data shows that the connection point has an abnormal temperature rise of 15 degrees Celsius, and since the thermal imaging data is assigned a high trust component of 1.0, this temperature rise signal will dominate in the final risk calculation. Finally, the final structure defect risk assessment conclusion is given, for example, the connection point is marked as "potential high risk", and the corresponding warning is issued, suggesting that manual review or further detailed detection be performed.

[0092] Through the power construction risk assessment method of the present application, the "smart eyes" are installed for the inspection robot. When the inspection robot inspects the newly completed power structure, the surface of the power structure will reflect light like a mirror under the sun due to the use of new hot-dip galvanizing process. This reflection will cause the image taken by the robot camera to have local very bright (overexposure) and most of the area very dark (underexposure). The conventional judgment method, because its internal judgment rule is established according to the old structure image with rough surface and uniformly scattered light, it cannot understand the special image distortion caused by mirror reflection, and often mistakenly thinks that there is no defect in the image, thus giving a false conclusion of safety.

[0093] The scheme of the present application first checks the quality of the photo taken by the high-definition visible light camera. Calculate what percentage of the highest and lowest brightness areas in the photo. It is precisely because of the mirror reflection characteristics of the new material that this extreme pixel distribution becomes the key basis for judging the quality of the image. If the proportion of extreme areas exceeds the preset limit, it is determined that the main sensor (high-definition visible light camera) sensing data is invalid, and the data invalidation of the main sensor is responded to.

[0094] Once the defect recognition image quality is determined to be problematic, the judgment strategy is immediately adjusted to minimize the trust in the defect recognition image and maximize the trust in the infrared thermal imager data. This is because the working principle of the infrared thermal imager is different from that of visible light. It judges temperature by sensing the heat emitted by the object, so it is less affected by the surface reflection characteristics of the object. In this special scene, its data is more reliable. This processing method solves the problem of conflict between the two sources of information and avoids excessive reliance on failed visual information.

[0095] In this way, even if the defect recognition image is unclear, if the infrared thermal imager detects even a slight temperature rise (such as 5 degrees Celsius) at the connection point, the system will tend to judge that there is a potential risk here due to the high trust in the thermal imaging data. This overcomes the rigidity of traditional evaluation logic, allowing it to adjust the judgment basis flexibly according to the actual situation, thereby avoiding the false underestimation of potential safety hazards.

[0096] In contrast, the conventional scheme will make a wrong judgment directly when it encounters a "non-standard" photo with high reflectivity because it cannot identify the defect features on the photo. This scheme is more suitable for old power structures with uniform surface light reflection and stable image quality, because in that case, the visible light image is usually clear and reliable and can be directly used as the main basis for judgment. However, in the special high-reflectivity scene caused by new materials, the conventional scheme cannot identify data anomalies, leading to misjudgment.

[0097] The following is a specific embodiment of the present application, which sets up to check a newly completed power tower by a patrol robot. The bolts and connecting plates on the tower are made of the latest hot-dip galvanized process. The patrol robot is equipped with a high-definition visible light camera and an infrared thermal imager.

[0098] Without the use of the present application scheme: The patrol robot follows the conventional process and first takes a picture of the bolt with a high-definition visible light camera. Due to the strong specular reflection on the surface of the bolt, the camera's automatic exposure system will overall reduce the exposure to avoid overexposure in the high light area, resulting in the bolt gap and other key details becoming completely dark, while the high light area is bright and has no details.

[0099] Then, the patrol robot performs visual analysis on the collected image. The analysis model is trained based on a large number of old and rough surface bolt images and is used to identify defects from clear textures, edges, and rust. Faced with the image taken this time, which is full of highlights and shadows, the model cannot extract any effective defect features, and finally reports: "No bolt loosening or structural defect detected, risk level: low." At the same time, the infrared thermal imager detected that the temperature of this bolt was 5 degrees Celsius higher than the ambient temperature, which was a weak signal but had exceeded the normal fluctuation range, implying possible poor contact. However, since the traditional risk assessment logic defaults to visual information as the primary and most reliable, it will take the visual module's "low risk" conclusion as the primary basis and regard the weak abnormal signal of thermal imaging as secondary information, or even directly ignore it. Ultimately, the inspection robot will report to the control center: "The bolt is in good condition, no abnormalities." In this way, a potential safety hazard is mistakenly overlooked.

[0100] After adopting the scheme of the present application: When the inspection robot arrives at the bolt position and is ready to take a picture, the process of the present scheme is started: 1. Image health degree rapid diagnosis: The inspection robot first takes a picture of the bolt with a high-definition visible light camera. This image will also have local overexposure and underexposure due to high reflectivity.

[0101] The inspection robot immediately analyzes this image. The rectangular area where the bolt is located is identified, assuming that this area is 100 pixels wide, 100 pixels high, and there are a total of 10,000 pixels.

[0102] All pixels are traversed, and the number of pixels with a brightness value (0-255) of 255 (pure white) and the number of pixels with a brightness value of 0 (pure black) are counted.

[0103] Assuming the statistics are: there are 1800 pure white pixels and 3500 pure black pixels.

[0104] Then, the pure white pixel ratio = (1800 / 10000) × 100% = 18%.

[0105] The pure black pixel ratio = (3500 / 10000) × 100% = 35%.

[0106] 2. Visual information pollution label generation: The preset failure judgment threshold is: when the pure white pixel ratio exceeds 10% or the pure black pixel ratio exceeds 30%, the image health degree is unqualified.

[0107] In this example, the 18% pure white pixel ratio exceeds the 10% threshold, and the 35% pure black pixel ratio also exceeds the 30% threshold. Therefore, it is immediately determined that this image is unqualified, and an internal label "VisualDataCompromised" is generated, set to "true". This label is like a warning sign, telling the subsequent evaluation module: "This visual data is problematic and cannot be fully trusted." 3. Adaptive adjustment of trust components in risk assessment: The inspection robot acquires data from the infrared thermal imager. Suppose the thermal imager detects that the temperature of the bolt is 5 degrees Celsius higher than the surrounding environment.

[0108] When entering the risk assessment phase, first check the "VisualDataCompromised" flag. Since the flag is "True", immediately adjust the trust components for the two information sources: The trust component for thermal imaging data is raised from the regular 0.3 to 0.9.

[0109] The trust component for visible light visual data is reduced from the regular 0.7 to 0.1.

[0110] Suppose during visual analysis, the "defect likelihood score" given on the blurred image is 0.1 (very low), while the "abnormal score" given by the thermal imaging module based on the 5-degree temperature rise is 0.8 (relatively high).

[0111] Then the final total risk score is calculated as: Total risk score = (0.8 × 0.9) + (0.1 × 0.1) = 0.72 + 0.01 = 0.73.

[0112] This 0.73 score is much higher than the low risk score that might be obtained under regular circumstances (for example, if the visual weight is high, the risk score might be 0.1 × 0.7 + 0.8 × 0.3 = 0.07 + 0.24 = 0.31).

[0113] 4. Final risk judgment output: Since the calculated total risk score of 0.73 reaches the preset high risk threshold (for example, 0.6), immediately send an alarm to the control center: "There is a potential high risk in the bolt, please review immediately."

[0114] In this embodiment, the high risk threshold is set according to empirical data. Specifically, based on the collected historical data, the above calculation process is performed multiple times, and the lowest total risk score value among the multiple total risk scores is taken as the high risk threshold in this embodiment.

[0115] In this way, even if the visual data is distorted due to high reflection, the system can intelligently identify this distortion and adjust the judgment strategy in time, shifting the focus of decision-making to more reliable thermal imaging data. This ensures that even weak temperature rise signals are fully valued in the final risk assessment, thereby avoiding the false underestimation of potential defects and ensuring the safe operation of power facilities.

[0116] The creativity of the scheme is that the focus of traditional risk assessment is extended from only judging whether the target object has defects to judging whether the collected data itself has defects first. This processing method does not try to repair a high-reflectivity damaged image, nor forces the defect identification logic to understand a blurred image, but admits the limitations of the data in a specific scene. Based on this, the method of the application establishes a new decision-making process: when the data quality problem is identified, it can intelligently switch to a backup and more reliable judgment scheme. This idea cleverly transforms a seemingly complex image processing problem into an information integrity management and decision strategy adjustment problem, providing a new perspective for solving such multi-sensor information conflicts. This method is designed for the extremely strong mirror reflection characteristics of new hot-dip galvanizing materials in the early stage of operation. It is this material characteristic that causes a unique local overexposure and underexposure pattern in visible light images, making the image quality judgment method through pixel statistics extremely effective and reliable. Therefore, this method is tailored for solving the specific interference of specific materials under specific lighting to the visual system, and has strong pertinence.

[0117] The method can intelligently identify the unreliability of visible light visual data caused by new high-reflectivity materials, and adaptively adjust the decision components of different sensor information in this specific situation, thereby avoiding false underestimation of the risk of potential structural defects, significantly improving the evaluation accuracy and reliability in complex inspection environments, and effectively reducing the safety hazards caused by misjudgment.

[0118] Further, in the daily maintenance and management of modern power infrastructure, regular inspection and risk assessment of the power structure part that has completed the construction is a key link to ensure the safe and stable operation of the entire power grid system. In the past, manual on-site visual inspection was mainly relied on. This method is not only inefficient and easily affected by the subjective judgment of the inspector, but also has great safety hazards when facing high-altitude or complex environments. In order to solve these problems, the industry has generally begun to use robot inspection systems equipped with various sensors, hoping to collect data and assess risks through automation and refinement.

[0119] The parameters and rules inside those judgment methods, which are currently used to identify defects in power structures, are mostly established and optimized based on a large amount of past inspection data. These old data mainly come from power towers or substations that have been in operation for many years. The surface coatings of these old structures, which have been exposed to wind and sun for a long time, often naturally age and develop some minor rust or dust. These conditions make their surfaces more inclined to "diffuse reflection" of visible light, that is, the light spreads evenly in all directions after being reflected. This diffuse reflection characteristic allows the texture, color, and possible visual features such as small cracks, rust caused by loose bolts, etc. on the surface of the structure to be clearly captured by high-definition cameras. This provides stable and easily predictable input information for defect identification methods.

[0120] However, as the power infrastructure is continuously updated and expanded, a batch of newly completed power towers or substation structures use the latest hot-dip galvanizing process for key fastening bolts and connecting plates. This process provides excellent corrosion resistance, but also makes the surface of these components exhibit very strong "mirror reflection" characteristics when they are first put into use. Compared to the old components used to establish judgment methods, the surface of these new components is like a mirror, which can reflect incident light in a single direction, forming a very strong highlight area. This significant difference in optical characteristics presents new challenges for defect identification based on visible light images. When the inspection robot approaches to take pictures of these high-reflective bolts and connecting plates using the new galvanizing process, the automatic exposure and autofocus programs carried by the high-definition camera on the robot will be strongly disturbed by the mirror reflection. For example, the automatic exposure program usually adjusts the exposure parameters by analyzing the brightness distribution in the image to avoid large areas of overexposure or underexposure. However, when there are local very bright mirror reflection points in the picture, to prevent these highlight areas from becoming completely "dead white", the automatic exposure program tends to significantly reduce the overall exposure. This compensation mechanism, while protecting the highlight areas from complete distortion, causes other non-reflective areas in the image, especially the key details between the bolts and connecting plates, to become severely underexposed, appearing dark and difficult to see clearly. At the same time, the automatic focusing system may also be misled by these strong and clear highlight reflection points, causing the structure features that need to be detected in the image to become blurred and unclear.

[0121] In the foregoing embodiments, the proportion of pure white and pure black pixels in a specific region of interest in the visible light image is calculated and compared with a preset failure judgment threshold to intelligently identify whether the visible light data is polluted by high light reflection and generate a corresponding pollution label for adjusting the component or trust degree of different sensor data in the subsequent risk assessment link. However, in the actual power facility operating environment, the situation is often more complex. Imagine a situation, after these newly completed power structures are put into operation for a period of time, due to the industrial environment in which they are located (for example, power transmission towers near chemical plants, or substations around coal-fired power plants), the air is filled with fine dust or industrial aerosols that are difficult to detect with the naked eye. These pollutants will form a very thin, semi-transparent adhesion layer on the surface of these high-reflectivity bolts and connecting plates in a non-uniform manner. Although this adhesion is not enough to completely obscure the metal luster or make the surface completely diffuse reflection, it will partially scatter the incident specular reflection light and slightly absorb the light. As a result, the extremely strong highlight area that would have caused the pixel to reach pure white (255) may now have its brightness value slightly attenuated, for example, to between 240-250, no longer fully saturated. Similarly, the deep shadow area that would have been completely black (0) may also have its brightness value slightly increased due to the slight scattering of ambient light, for example, to between 5-15, no longer pure black. This means that although the overall contrast and detail clarity of the image have still been severely reduced due to this "mixed effect of diffuse reflection and absorption", the proportion of pure white and pure black pixels in the image may have fallen below the preset failure judgment threshold. At this time, the "image health degree rapid diagnosis" module in the current solution, since its judgment logic strictly depends on the proportion of extreme pixels, may incorrectly judge that the image is "healthy" and thus will not generate a "visual information pollution" label.

[0122] Further, if the inspection robot happens to perform the inspection during the daytime, when there are fast-moving clouds in the sky, or when there are tall buildings, trees, etc. around the power facility, these factors will cause the intensity and angle of the natural light shining on the bolt surface to change rapidly and locally. This makes the camera auto-exposure and auto-focus system frequently adjust the parameters within a very short time (e.g. a few hundred milliseconds). Under such dynamic lighting conditions, even the “non-pure white / black” brightness decay areas caused by non-uniform attachments will have their brightness values fluctuate dramatically within a very short time, sometimes reaching or approaching the critical value of pure white / black, but then quickly falling back. This leads to a phenomenon that at a certain moment, the image may satisfy the condition of “pure white / black pixel ratio exceeding the threshold”, but because the duration of this state is extremely short, it is missed in the slight delay of image acquisition and processing, or it is repeatedly switched within a short time, making it difficult for the system to stably capture this transient “pollution state”. This transient, non-continuous image quality degradation will still seriously affect the accuracy of subsequent defect recognition, but the current health diagnosis mechanism based on “steady-state” pixel ratio may not be able to reliably identify it.

[0123] In addition, in some complex structures, multiple high-reflective components (e.g. multiple closely arranged bolt heads, connecting plates and brackets) that use new galvanizing processes will be closely gathered together. When external light shines on one of the components, the light will be specularly reflected, and then this reflected light may be reflected again onto an adjacent another high-reflective component, and then reflected into the camera lens of the inspection robot from this component. This “multiple reflection” phenomenon will produce a special visual interference: it is no longer a simple local overexposure or underexposure, but forms a complex pattern of halos, shadows and secondary highlights. These patterns are not directly caused by defects in the detected components themselves, but by the repeated refraction and reflection of light between multiple high-reflective surfaces. These secondary spots and shadows, whose brightness values are usually between pure white and pure black, greatly distort the real texture and edge information of the detected components. For example, a bolt head may have an unnatural bright edge on its edge due to the reflection of the adjacent connecting plate, which covers the actual bolt loosening gap. The current “pure white / black pixel ratio” method, which focuses on extreme saturated pixels, is completely powerless for this kind of image distortion pattern caused by complex optical paths, which is not extreme in brightness value but is also disturbing. It will judge this image as “healthy”, although the key defect features in the image have been blurred or covered by these optical interferences.

[0124] Further, in the part of the power construction that has been completed, especially for the power structure that uses new high-reflective materials, when the visible light image of the inspection robot is disturbed by various factors such as surface attachments, dynamic light changes, or complex multiple reflections, resulting in a significant decrease in the overall clarity, details, or edge information of the image, but the proportion of extremely bright or extremely dark pixels in the image does not reach the preset judgment standard, how to accurately identify the unreliability of the visual information in this non-extreme brightness saturation state and avoid false judgments of potential structural problem risks.

[0125] The scheme of the embodiment no longer attempts to make the visible light camera see every detail under all complex lighting conditions, but uses infrared thermal imaging as the first, fast and reliable screening line. Only when the thermal imaging system finds a potential temperature anomaly will it guide the robot to invest valuable visual detection resources to conduct more detailed and targeted visual inspection of the suspected point. This strategy can effectively deal with the unreliability of visual information in the non-extreme brightness saturation mode caused by surface attachments, dynamic light changes, or complex multiple reflections, and avoid false judgments of potential structural defect risks.

[0126] To solve the above problems, the embodiment of the present application sets the following steps before the steps of acquiring the defect identification image of the power structure to be evaluated and the corresponding infrared thermal imaging data: S101: Move along the preset inspection path and collect surface temperature data of all key detection points in the power structure to be evaluated, and calculate the operating temperature reference of each key detection point in the normal operating state.

[0127] Specifically, when the inspection robot (for example, equipped with a high-precision infrared thermal imager such as FLIR T1020 series) performs daily inspection in the part of the power construction that has been completed, it first performs a large-scale infrared thermal imaging scan. The inspection robot moves along the preset inspection path, and the infrared thermal imager mounted thereon continuously collects surface temperature data of all key detection points (for example, bolt connection points, connecting plates, etc.).

[0128] The industrial-grade edge computing device (for example, Advantech UNO-2271G) inside the inspection robot processes these thermal imaging data in real time. For each detection point, the device dynamically calculates the expected temperature reference of the detection point in the normal operating state according to the current environmental conditions (for example, the real-time environmental temperature and humidity obtained through the environmental sensors built into the robot) and the operating state of the power facility (for example, the current line load current obtained through the interface with the power SCADA system).

[0129] The simplified formula for calculating the expected temperature reference can be expressed as: T 预期 = T 环境+ K x I 负载 2 where T 预期 is the real-time ambient temperature, I 负载 is the current load current of the power line, and K is an empirical coefficient related to material properties and connection resistance, which can be obtained by calibrating the temperature performance of similar healthy connection points under different loads and environments in advance.

[0130] In this way, the system can quickly and universally obtain temperature information of all detection points, which is almost not affected by optical interference such as specular reflection, surface attachments or dynamic lighting in visible light images.

[0131] S102: Compare the surface temperature data of each key detection point with the corresponding operating temperature reference.

[0132] After completing the thermal imaging data acquisition and expected temperature reference calculation, the industrial-grade edge computing device inside the robot will immediately compare the actual measured temperature of each detection point with the corresponding expected temperature reference.

[0133] S103: If the temperature difference between the surface temperature data of one key detection point and the operating temperature reference is greater than a preset temperature difference threshold, mark the key detection point as a potential attention point; otherwise, mark the key detection point as a low-risk point, and analyze the defect attention area in each defect recognition image containing the potential attention point.

[0134] If the actual measured temperature (T 测量 ) of a detection point and the expected temperature reference (T 预期 ) exceed a preset small threshold (for example, ΔT 阈值 is set to 2 degrees Celsius), that is, the condition T 测量 -T 预期 > ΔT 阈值 is met, the detection point will be marked as a “potential attention point” by the system.

[0135] For the vast majority of detection points with a temperature difference within the normal range, the system will directly determine them as “low risk” and skip the subsequent visual detection link, thereby greatly saving the inspection time and computing resources. Only a few points marked as “potential attention points” will trigger the next step of targeted high-precision visual detailed inspection procedure.

[0136] Specifically, the defect attention area in each defect recognition image containing the potential attention point is analyzed, including the following steps: S1031: Collect and process defect recognition images for each key detection point marked as a potential point of interest by adjusting the shooting angle of the high-definition visible light camera of the inspection robot.

[0137] Once a "potential point of interest" is identified, the robot control system (e.g., ROS-based navigation and control stack) will immediately adjust the robot's pose and position to accurately move to the optimal shooting angle and distance of the "potential point of interest".

[0138] The robot vision module (e.g., equipped with NVIDIA Jetson AGX Orin module) will start its advanced visual information collection and analysis protocol to overcome the unreliability of visual information in the presence of non-extreme brightness saturation at this point: The robot will collect images of the point from 3 to 5 slightly different angles (through slight robot movement or fine adjustment of the camera gimbal). At the same time, the system will try to optimize local exposure, for example, by analyzing the brightness histogram of key detail areas such as bolt gaps and edges in the image, and adjusting exposure parameters accordingly to ensure moderate brightness in these areas, avoiding detail loss due to overall overexposure or underexposure.

[0139] S1032: Calculate information quality evaluation indicators for defect recognition images after image processing; information quality evaluation indicators include edge sharpness score and texture richness score.

[0140] For the collected images, the vision module will perform multi-dimensional information quality evaluation, not just relying on the proportion of pure white or pure black pixels. It will calculate at least two key indicators: i. Edge sharpness score: Calculate the gradient amplitude of key structures (such as bolt heads and connecting plate edges) in the image. For example, use the Sobel operator to calculate the horizontal and vertical gradients of the image, then calculate the average of the gradient amplitude. The higher the score, the clearer the edge.

[0141] Gradient amplitude = sqrt(Gx 2 + Gy 2 ) Edge sharpness score = average pixel gradient amplitude = (sum of all pixel gradient amplitudes) / (total number of pixels) Gx and Gy represent the horizontal and vertical gradients of the image, respectively.

[0142] ii. Texture richness score: Analyze the entropy value of the local binary pattern (LBP) features in key areas of the image. The higher the entropy value, the richer the detail texture information contained in the image.

[0143] Texture richness score = -sum(p(i) x log2(p(i))) where p(i) is the probability of LBP pattern i occurring; According to the comprehensive performance of the multi-dimensional indicators, the reliability of the current visual information is judged. For example, if the edge sharpness score is lower than 0.6 or the texture richness score is lower than 0.7, even if there is no large area of pure white or pure black pixels, the system will determine that the visual data of this point is still unreliable, and generate an internal label of "poor visual data quality".

[0144] For the detection points marked as "potential attention points", the system decision unit (for example, an industrial PC running Linux operating system) will comprehensively consider the results of thermal imaging abnormal signals and high-precision visual detailed inspection.

[0145] The judgment logic is as follows: a. If the point has "temperature rise abnormality" and "high-precision visual detailed inspection confirms the existence of defects" (for example, identifying loose bolts, cracks, etc.), it is determined as "high risk" and the highest level of warning is immediately issued.

[0146] b. If the point has "temperature rise abnormality" but "high-precision visual detailed inspection result shows poor visual data quality" (i.e. unable to clearly identify defects), it is determined as "medium-high risk" and it is suggested to "immediately perform manual review or further professional detection".

[0147] c. If the point has "temperature rise abnormality" but "high-precision visual detailed inspection does not find obvious defects" (and the visual data quality is good), it is determined as "medium risk" and it is suggested to "include in the list of key observation, and perform regular review".

[0148] Finally, through the wireless communication module (for example, a communication unit supporting 5G or Wi-Fi), the alarm information with specific risk level, location information and suggested operation is sent in real time to the power control center or the mobile terminal of the maintenance personnel.

[0149] This scheme is like installing an intelligent triage system for the inspection robot. This system no longer makes the robot perform the most detailed visual inspection on all power structures, but first performs a quick screening.

[0150] The conventional inspection method is to check all structures by a high-definition visible light camera. When encountering power structures made of new materials with mirror-like shiny surfaces, due to surface attachments, dynamic light changes or complex reflections between components, the photos will become blurred, with reduced contrast, and even with strange light spots and shadows, but not in the extreme case of complete white or black. In this case, the high-definition visible light camera will mistakenly think that "there is no problem" and miss potential risks because it cannot see the true texture and edge of the bolt. This method is more suitable for old power structures with rough surfaces and uniform light reflections, because in such an environment, the high-definition visible light camera can usually stably obtain clear images.

[0151] The present solution is different. First, an infrared thermal imager is used to make a rapid temperature measurement of all power structures. The infrared thermal imager is not affected by light and can quickly find out whether any connection point has a slight temperature rise. Such a temperature rise is often an early signal of poor contact caused by loose bolts.

[0152] It is precisely because of the possible thin layer of attachments on the surface of new materials, or the rapid changes of environmental light, and the complex light reflections between multiple highly reflective components, that the high-definition visible light camera may not be able to see clearly in non-extreme brightness saturation. The infrared thermal imager can penetrate optical interference and directly sense the internal heat changes of the object.

[0153] Only when the infrared thermal imager finds that a connection point has an abnormal temperature (even if it is only a slight temperature rise), the system will determine that this point is suspicious, and then conduct a deep inspection by a high-definition visible light camera. At this time, the high-definition visible light camera adjusts the shooting position, takes pictures from multiple angles, even tries to optimize local exposure, and carefully evaluates the clarity and detail richness of the image. If after these efforts, the high-definition visible light camera still cannot obtain reliable image information, the system will not ignore the warning of the infrared thermal imager, but will make a comprehensive judgment and issue an abnormal alarm.

[0154] This working method avoids the "confusion" of the high-definition visible light camera in most cases, and can focus on the points that really have potential problems. It ensures that even in the most complex visual interference, the potential risks revealed by temperature abnormalities will not be ignored, thereby greatly improving the accuracy and efficiency of inspection.

[0155] In addition, in other embodiments of the present invention, the following problems also exist: In the scheme of the present application, there is also the case that when the inspection robot conducts more detailed visual inspection on a certain bolted joint point screened out by thermal imaging, although the robot has accurately adjusted the posture and collected images, at this time, a more hidden visual information misleading may occur. The bolts and connecting plates of the new hot-dip galvanizing process, in the early stage of operation, will form a layer of uneven, semi-transparent oxide or sulfide film on the micro level due to long-term contact with specific trace aerosols (e.g. sulfur compounds discharged by chemical plants) in the industrial environment. This layer of film is not uniformly covered, but presents a micron-level mottled distribution, changing the optical properties of the local area. This causes the original pure mirror reflection to now produce slight scattering and interference effects in these micro-uniform areas. When external light shines on these areas, it is no longer a single-directional reflection, but forms countless extremely small, varying brightness "sparkling points" or "weak halos". The brightness values of these "sparkling points" and "weak halos" are usually between pure white and pure black, but their density and distribution are extremely high. The high-definition visible light camera carried by the inspection robot has very high capture sensitivity on this high-density, weak "sparkling point" and "weak halo" at the pixel level. At the same time, the image signal processor (ISP) inside the camera, when performing image noise reduction processing, its default noise reduction algorithm often tends to retain the high-frequency details in the image to ensure the sharpness of the image. When these high-density, weak "sparkling points" and "weak halos" enter the camera and are processed by the ISP, they are misjudged as real high-frequency details or tiny textures in the image. The noise reduction algorithm not only fails to effectively filter out these optical artifacts, but may even enhance them to some extent, making them appear as a large number of tiny, high-contrast bright spots or local brightness fluctuations in the image. This "pseudo-high-frequency feature" (i.e. a large number of tiny, non-structural high-contrast bright spots or local brightness fluctuations) produced by the combined action of micro-surface changes and camera ISP characteristics misleads the image quality indicators calculated by the visual processing module. For example, the real edges of the bolt edges or the connecting plate joints may be slightly blurred due to uneven environmental light or slight attachments. However, these "pseudo-high-frequency features" introduce a large number of additional, high-gradient tiny brightness changes in the image. When calculating the gradient amplitude of the overall or local area, these pseudo-features are counted, causing the calculated edge sharpness score to be systematically raised, even exceeding the level of the actual clear image. A bolt edge that should be blurred has a sharpness score that is misjudged as "good". Similarly, these high-density "sparkling points" and "weak halos" present a high degree of complexity and randomness in the local pixel pattern. When the visual module analyzes the entropy value of the local binary pattern (LBP) feature, these non-structural, high-frequency brightness fluctuations are mistakenly identified as rich texture information, resulting in a significant exaggeration of the texture richness score.A surface that should be smooth or have only slight wear texture may be misclassified as "rich in detail" in terms of texture richness score. As a result, even though the true structural features of a bolt (such as tiny gaps or rust lines caused by bolt loosening) become more difficult to discern, or are even completely obscured, by the superposition of high reflectivity and microscopic optical artifacts, the system, based on its multi-dimensional image information quality assessment method (edge ​​sharpness, texture richness), incorrectly judges the current visual data quality as "reliable." This misjudgment causes the system to still heavily rely on this actually "contaminated" visual information for defect identification, potentially leading to another erroneous conclusion of "no defects" or "low risk," ignoring the subtle temperature rise warning previously issued by the thermal imaging system, and creating deeper safety hazards.

[0156] In completed power construction sections, especially for structures using new high-reflectivity materials, when the visible light images from inspection robots exhibit a large number of unstructured, high-frequency pseudo-features due to the interaction between the microscopic unevenness of the material surface and ambient light, as well as the superposition of camera image signal processor characteristics, how can we prevent these pseudo-features from enhancing edge sharpness scores and texture, thereby accurately identifying the unreliable state of visual information and preventing the underestimation of potential structural defects?

[0157] This example's solution changes the traditional approach to visual information evaluation. Instead of relying solely on the spatial features of static images to judge image quality, it introduces temporal analysis. By acquiring high-frame-rate video streams of target points and analyzing the "pulse characteristics" of the brightness changes of each pixel over time, the system can accurately identify unstructured pseudo-features generated by the interaction of microscopic inhomogeneities on the material surface, ambient light, and the superposition of camera image signal processor characteristics. This avoids these pseudo-features from misleadingly improving image quality evaluation metrics, ensuring the accurate and reliable identification of true visual information even under complex visual interference.

[0158] The method steps in this embodiment are as follows: 1. High frame rate video stream capture: When a patrol robot (e.g., equipped with a high-precision infrared thermal imager and a high-definition visible light camera, such as a Basler ace2 series industrial camera) is performing routine patrol in a section of power construction that has been completed, and through preliminary screening by thermal imaging, it identifies that a certain bolt connection point has a slight temperature anomaly and is marked as a "potential point of interest", the robot control system (e.g., a ROS-based navigation and control stack) will immediately adjust the robot's pose and position to accurately move to the optimal shooting angle and distance of the "potential point of interest". At this time, the robot vision module (e.g., equipped with a NVIDIA Jetson AGX Orin module) will start its advanced visual information acquisition protocol and perform a short high-frame-rate video stream acquisition on the target point. For example, the camera acquires a 1-second video at a frame rate of 60 frames per second (60 fps), a total of 60 frames of images, with a resolution of 1920x1080 pixels.

[0159] 2. Target region pixel sequence extraction: After collecting the high-frame-rate video stream, the image processing unit inside the robot (e.g., a NVIDIA Jetson AGX Orin module) will first identify and frame the area of interest in each frame of the video, such as a rectangular area containing the bolt and its surrounding connection plate. This area can be determined by pre-set geometric position information or by simple image processing methods (such as preliminary identification based on color and shape). Then, for each pixel point (x, y) in the area of interest, the processing unit will extract its brightness value in all collected video frames. If the image is a color image, the brightness value can be converted to a grayscale value (e.g., by weighting the average R, G, B channel pixel values, such as brightness = 0.299R + 0.587G + 0.114B). In this way, each pixel point will correspond to a brightness sequence over time: B(x, y, t), where t represents the frame index (from 1 to N, N is the total number of frames).

[0160] 3. Quantification of pixel brightness fluctuation characteristics: For each pixel point (x, y) brightness sequence B(x, y, t) extracted in step 2, the image processing unit will calculate the degree of brightness fluctuation in the entire video acquisition period. Here, variance is used as a quantitative indicator because it can effectively reflect the dispersion of data points from their mean value, thus capturing the "pulse" or "flicker" characteristics of brightness values.

[0161] The specific calculation formula is as follows: First, calculate the mean value Mean_B(x, y) of the pixel brightness sequence: Mean_B(x, y) = (1 / N) sum_{t=1 to N} B(x, y, t) Then the variance of the pixel brightness sequence, Variance_pixel(x, y), is calculated: Variance_pixel(x, y) = (1 / N) sum_{t=1 to N} (B(x, y, t) - Mean_B(x,y)) 2 Where N represents the total number of frames of the video stream collected for the pixel point (x, y). The larger the variance value, the more intense the brightness fluctuation of the pixel point in the video stream, and the more likely it is a "glint point" caused by optical artifacts.

[0162] B(x, y, t) represents the brightness value of the pixel point (x, y) in the t-th frame of the video stream. If the original image is a color image, this brightness value is usually converted to a gray value for processing.

[0163] sum_{t=1 to N} B(x, y, t) represents the sum of the brightness values of the pixel point (x, y) in all N frames.

[0164] (1 / N) represents dividing the total brightness value by the total number of frames to obtain the average brightness.

[0165] The purpose of Mean_B(x, y) is to provide a reference value for subsequent calculation of the degree of pixel brightness fluctuation. In the pixel brightness fluctuation characteristic quantization step, Mean_B(x, y) is an intermediate result for calculating the pixel brightness sequence variance Variance_pixel(x, y). The larger the variance value, the more intense the brightness fluctuation of the pixel point in the video stream. This brightness fluctuation characteristic is a key basis for identifying non-structural artifact pixels generated by material surface microscopic uneven changes, environmental light interaction or camera image signal processor characteristics superposition. Therefore, Mean_B(x, y) as a necessary component of variance calculation indirectly supports the judgment of the reliability of the main visual image, thereby avoiding the false low estimate of the risk of potential structural defects due to artifact misdirection.

[0166] 4. Artifact pixel identification and contamination index calculation: After calculating the variance of brightness of all pixels in the region of interest, a threshold of brightness fluctuation (e.g., Threshold_Fluctuation, which can be determined by a large number of tests on healthy structures and structures with false features) is preset. If the Variance_pixel(x, y) of a certain pixel (x, y) exceeds the Threshold_Fluctuation, the pixel is identified as a false feature pixel because it shows non-structural, high-frequency brightness fluctuations.

[0167] Then, the system calculates the total number of false feature pixels in the region of interest, and takes the percentage of the total number of pixels in the region of interest as the optical pollution index: Optical Pollution Index = (Number of False Feature Pixels / Total Number of Pixels in the Region of Interest) x 100% This index directly quantifies the degree of pollution of visual information by optical artifacts.

[0168] 5. Visual information reliability judgment and decision adjustment: The system will preset a pollution judgment threshold (e.g., Threshold_Pollution_Index, which is also determined by experiments). If the calculated optical pollution index exceeds the Threshold_Pollution_Index, the system will judge that the current visible light visual data is unreliable due to false feature interference, and generate a clear internal marker, for example, set a Boolean state variable named "VisualDataUnreliable" to "true".

[0169] In the subsequent step of structural defect risk assessment, if the "VisualDataUnreliable" marker is detected to be "true", this marker will be used as a mandatory trigger condition. It will make the risk assessment logic temporarily and completely adjust the trust component of different information sources. Specifically, the system will set the trust component of infrared thermal imaging data analysis results to the highest (e.g., 0.9), and reduce the trust component of visible light visual analysis results to the lowest (e.g., 0.1). This adjustment ensures that when the quality of visual data is questionable, the system can rely more on thermal imaging data for judgment, thereby avoiding false underestimation of potential structural defect risks. If the "optical pollution index" does not exceed the threshold, it means that the visual data is not severely polluted by false features, and the system can continue to rely on the results of its multi-dimensional image information quality evaluation (such as edge sharpness, texture richness) for defect identification.

[0170] The scheme shoots a video through the inspection robot, and then checks each pixel point in the video. The intensity of the brightness change of each pixel point in the video playing process is calculated. If the brightness of a pixel point jumps sharply, it is marked as a suspicious point. By counting the proportion of suspicious points in the picture, it can be accurately judged whether the visual data is qualified. Figure 1

[0171] The method of the present application overcomes the limitations of the traditional scheme. The traditional scheme is to judge the quality of the photo according to the content displayed on the photo. It may be misled by the false high-frequency information in the picture and mistakenly think that the photo quality is good, so that the real structural problem may be overlooked. This traditional method is more suitable for old power structures with uniform surface light reflection and stable image quality, because in that case, the details on the photo are usually true and reliable. But for this special optical interference brought by new materials, the traditional method is not up to the task.

[0172] The present scheme is different, which identifies the authenticity of information by going deep into the time dimension. It ensures that even in the most complex visual interference, the potential risks revealed by temperature anomalies will not be ignored, thereby greatly improving the accuracy and reliability of the inspection.

[0173] As shown in Figure 2 The present application also provides a power construction risk assessment system 200, comprising: An analysis module 210 is configured to obtain a defect identification image and corresponding infrared thermal imaging data of a power structure to be evaluated, determine a defect attention area in the defect identification image, and analyze the brightness information in the defect attention area. A judgment module 220 is configured to judge whether the defect identification image is qualified based on the statistical analysis result of the brightness information of the defect attention area. An adjustment module 230 is configured to set a trust component range of the defect identification image and the infrared thermal imaging data in power construction risk assessment, and adjust the trust component of the defect identification image and the infrared thermal imaging data in power construction risk assessment based on the judgment result of the defect identification image. An output module 240 is configured to determine a risk score of the power structure to be evaluated based on the trust component of the adjusted defect identification image and infrared thermal imaging data in power construction risk assessment, and output a defect risk assessment conclusion of the power structure to be evaluated.

[0174] ​In the description of the application, reference to "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that a particular feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. The illustrative appearances of the described terms in various places in the specification are not intended to exclude that the terms are used in other embodiments or examples of the application. Furthermore, the described specific features, structures, materials or characteristics can be combined in any suitable manner in one or more embodiments or examples of the application. In addition, the described embodiments or examples of the application are not to be taken in a limiting sense but are understood to be merely representative at least some embodiments of the application. Any combination of one or more described embodiments or examples of the application, or variations and / or modifications thereof, can be possible.

[0175] Furthermore, the terms "first", "second", etc. are used herein only to describe the names of particular features and do not imply or suggest relative importance or a number of the indicated features. Thus, features defined with "first", "second" etc. can explicitly or implicitly include at least one of the features. In the description of the application, the meaning of "plurality" is at least two, for example two, three, etc. unless explicitly specified otherwise.

[0176] Any process or method descriptions or blocks in flow charts or otherwise described herein represent embodiments which can be managed as one or more modules, segments, or portions of code which include one or more executable instructions for implementing specific logic functions or steps, and the alternate implementations can be possible where in the steps or order of steps can be performed in an order as discussed or in reverse order, depending on the functionality involved, or can be performed in other orders, unless explicitly specified otherwise.

[0177] Although embodiments of the application have been shown and described above, it is to be understood that the application is not limited to the embodiments described, and that various modifications, changes and substitutions can be introduced without departing from the scope of the application.

Claims

1. A method of electrical construction risk assessment, characterized by, The method comprises the following steps: Obtain defect identification images and corresponding infrared thermal imaging data of the power structure to be evaluated, determine defect attention areas in the defect identification images, and analyze the brightness information in the defect attention areas. Determine whether the defect identification images are qualified based on the analysis results of the brightness information in the defect attention areas. Set the trust component range of the defect identification images and the infrared thermal imaging data in the power construction risk assessment, and adjust the trust component of the defect identification images and the infrared thermal imaging data in the power construction risk assessment based on the judgment result of the defect identification images. Determine the risk score of the power structure to be evaluated based on the adjusted trust component of the defect identification images and the infrared thermal imaging data in the power construction risk assessment, and output the defect risk assessment conclusion of the power structure to be evaluated.

2. The method of claim 1, wherein, Obtain defect identification images and corresponding infrared thermal imaging data of the power structure to be evaluated, determine defect attention areas in the defect identification images, and analyze the brightness information in the defect attention areas, comprising: Collect the defect identification images of the power structure to be evaluated by a high-definition visible light camera carried by a patrol robot, and collect the infrared thermal imaging data of the power structure to be evaluated by a thermal imager carried by the patrol robot. Identify the defect attention areas in the defect identification images, wherein the defect attention areas are areas in the defect identification images that contain fastening bolts and their surrounding connecting plates. Statistically analyze the pixel points in the defect attention areas to determine the proportion of pixel points with the highest and lowest brightness.

3. The method of claim 2, wherein, In the step of determining the proportion of pixel points with the highest and lowest brightness, The proportion of pixel points with the highest brightness = (the number of pixel points with the highest brightness / the total number of pixel points in the defect attention area) × 100% The proportion of pixel points with the lowest brightness = (the number of pixel points with the lowest brightness / the total number of pixel points in the defect attention area) × 100% Wherein, the pixel point with the highest brightness is a pixel point with a pixel value equal to 255, and the pixel point with the lowest brightness is a pixel point with a pixel value equal to 0.

4. The method of claim 3, wherein, Determine whether the defect identification images are qualified based on the analysis results of the brightness information in the defect attention areas, comprising: Set a qualification judgment threshold; wherein the qualification judgment threshold includes a highest brightness pixel point qualification judgment threshold and a lowest brightness pixel point qualification judgment threshold. Determine whether the proportion of pixel points with the highest brightness is greater than the highest brightness pixel point qualification judgment threshold, or determine whether the proportion of pixel points with the lowest brightness is greater than the lowest brightness pixel point qualification judgment threshold. If it is determined that the proportion of pixel points with the highest brightness is greater than the highest brightness pixel point qualification judgment threshold, or it is determined that the proportion of pixel points with the lowest brightness is greater than the lowest brightness pixel point qualification judgment threshold, then the defect identification images are determined to be unqualified images.

5. The method of claim 1, wherein, Set the trust component range of the defect identification images and the infrared thermal imaging data in the power construction risk assessment, and adjust the trust component of the defect identification images and the infrared thermal imaging data in the power construction risk assessment based on the judgment result of the defect identification images, comprising: The analysis result of the defect identification image in the power construction risk assessment is set as a first trust component, and a value range of the first trust component is set; The analysis result of the infrared thermal imaging data in the power construction risk assessment is set as a second trust component, and a value range of the second trust component is set; If the defect identification image is determined to be qualified, the first trust component is taken as a maximum value of the value range of the first trust component, and the second trust component is taken as a minimum value of the value range of the second trust component; If the defect identification image is determined to be unqualified, the first trust component is taken as a minimum value of the value range of the first trust component, and the second trust component is taken as a maximum value of the value range of the second trust component.

6. The method of claim 5, wherein, Based on the adjusted trust components of the defect identification image and the infrared thermal imaging data in the power construction risk assessment, a risk score of the power structure to be evaluated is determined, and a defect risk assessment conclusion of the power structure to be evaluated is output, including: a preset defect identification risk score interval and an infrared thermal imaging risk score interval; based on a defect identification image of the power structure to be evaluated, a defect identification risk score value of the defect identification image is determined; based on infrared thermal imaging data of the power structure to be evaluated, an infrared thermal imaging risk score value of the infrared thermal imaging data is determined; based on the determined defect identification risk score value and the infrared thermal imaging risk score value, in combination with the first trust component value and the second trust component value, a total risk score of the power construction risk assessment of the power structure to be evaluated is determined; based on the total risk score, a defect risk assessment conclusion of the power structure to be evaluated is determined and output.

7. A method of electrical construction risk assessment according to claim 6, wherein, In the step of determining a total risk score of the power construction risk assessment of the power structure to be evaluated based on the determined defect identification risk score value and the infrared thermal imaging risk score value, in combination with the first trust component value and the second trust component value, the total risk score formula is represented as: Total risk score=(infrared thermal imaging risk score value x second trust component value)+(defect identification risk score value x first trust component value) wherein the infrared thermal imaging risk score value is determined according to a temperature rise deviation detected by a thermal imager; and the defect identification risk score value is determined according to a defect possibility identified by a high-definition visible light camera in a limited clear area.

8. The method of claim 1, wherein, Before the steps of obtaining a defect identification image and corresponding infrared thermal imaging data of the power structure to be evaluated, including: moving along a preset inspection path to collect surface temperature data of all key detection points in the power structure to be evaluated, and calculating an operating temperature reference of each key detection point in a normal operating state; comparing the surface temperature data of each key detection point with the corresponding operating temperature reference; If the surface temperature data of one of the key detection points is greater than the preset temperature difference threshold value from the temperature of the operation temperature reference, the key detection point is marked as a potential attention point; otherwise, the key detection point is marked as a low-risk point, and a defect attention area in the defect identification image containing the potential attention point is analyzed.

9. The method of claim 8, wherein, The analysis of the defect attention area in the defect identification image containing the potential attention point includes: Collecting and processing the defect identification image of each key detection point marked as a potential attention point by adjusting the shooting angle of the high-definition visible light camera of the inspection robot; Calculating the information quality evaluation index of the defect identification image after image processing; the information quality evaluation index includes an edge sharpness score and a texture richness score; wherein, The edge sharpness score is obtained by calculating the average gradient amplitude of all pixel points contained in the potential attention point in the defect identification image; the gradient amplitude formula is represented as: Gradient magnitude = sqrt(Gx 2 + Gy 2 ) Edge sharpness score = average gradient amplitude = (sum of all pixel gradient amplitudes) / (total number of pixels) Gx and Gy represent the horizontal gradient and the vertical gradient of the defect identification image, respectively; The texture richness score is obtained by analyzing the entropy value of the local binary pattern (LBP) feature of the potential attention point in the defect identification image; the higher the entropy value, the richer the detailed texture information contained in the image; the texture richness score formula is represented as: Texture richness score = -sum (p(i) * log2 (p(i))) Where p(i) is the probability of the occurrence of LBP mode i; Based on the calculated information quality evaluation index, it is determined whether the defect identification image containing the potential attention point is qualified.

10. A power construction risk assessment system, characterized by, Including: The analysis module is configured to obtain a defect identification image and corresponding infrared thermal imaging data of an electric power structure to be evaluated, determine a defect attention area in the defect identification image, and analyze brightness information in the defect attention area; The judgment module is configured to determine whether the defect identification image is qualified based on the analysis result of the brightness information in the defect attention area; The adjustment module is configured to set a trust component range of the defect identification image and the infrared thermal imaging data in electric power construction risk assessment, and adjust the trust component of the defect identification image and the infrared thermal imaging data in electric power construction risk assessment based on the judgment result of the defect identification image; The output module is configured to determine a risk score of the electric power structure to be evaluated based on the trust component of the defect identification image and the infrared thermal imaging data in electric power construction risk assessment after adjustment, and output a defect risk assessment conclusion of the electric power structure to be evaluated.

Images