An infrared abnormal heating image detection method and system for distribution network inspection

Abstract

The present application relates to the technical field of power equipment inspection, in particular to an infrared abnormal heating image detection method and system for distribution network inspection. It aims to solve the problems of low efficiency, high false detection rate and weak anti-interference of existing infrared inspection. The method includes: obtaining the equipment target box coordinates and quantity through the target detection model of the power equipment infrared image; obtaining the temperature distribution map, the highest temperature, the lowest temperature, the average temperature and the temperature standard deviation through the DJI SDK; merging the equipment or element class with the target box quantity < 5 into a larger target box, and drawing a rectangular box and labeling the temperature parameters for the part exceeding the 3σ range when there is no target box; calculating the dynamic threshold for the image with the target box, combining the local maximum temperature in the target box to mark the suspected abnormal area; screening the abnormal heating points according to the temperature difference; drawing a rectangular box for the abnormal points and labeling the temperature parameters and the equipment or element class. The present application realizes full-process automation, improves the inspection efficiency and accuracy, and is suitable for large-scale distribution network operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power equipment inspection technology, specifically to an infrared abnormal heating image detection method and system for distribution network inspection. Background Technology

[0002] In power system operation and maintenance, the operating status of critical equipment such as transformers, line joints, and switches directly affects power grid safety. These devices generate heat during operation, and the heating characteristics under fault conditions (such as localized high temperatures and abnormal temperature distribution) differ significantly from those under normal conditions. Therefore, infrared thermal imaging technology has become a core inspection method. By capturing "thermal images" of the equipment using infrared thermal imagers (image colors correspond to temperature levels), staff can analyze these thermal images to identify potential faults, enabling non-contact "health checks" of power equipment.

[0003] With the large-scale deployment of drone inspection technology, a single power distribution network inspection can collect thousands to tens of thousands of infrared images. Massive data processing and accurate fault identification have become core industry demands. However, existing infrared abnormal heating detection technology still has significant shortcomings: the traditional manual interpretation method is inefficient, with each image analysis taking 2-3 minutes, resulting in lengthy inspection cycles and excessive workload for personnel, making it difficult to meet the needs of large-scale inspections; at the same time, infrared imaging is easily affected by environmental factors, such as sunlight, temperature and humidity changes, and thermal radiation reflection between equipment, all of which generate background noise. These factors, combined with differences in shooting angles and insufficient image contrast, further increase the difficulty of fault identification. Summary of the Invention

[0004] To address the technical problems of existing manual-led analysis methods, such as insufficient efficiency in processing massive amounts of images, high false alarm and false negative rates in anomaly detection, and weak resistance to background noise, this invention provides, in a first aspect, an infrared anomaly heating image detection method for distribution network inspection. This method can identify potential faults in power equipment and simultaneously label the type of faulty equipment or component and its heating temperature parameters. The method includes the following steps: Acquire infrared images of power equipment; The location coordinates and quantity of target boxes for each type of device or component in the infrared image are obtained through a target detection model. Obtain the temperature distribution map and the highest temperature (K) of the corresponding infrared image. max Minimum temperature K min Mean temperature μ and standard deviation σ; Classification based on the number of bounding boxes: When there are no bounding boxes in the image, calculate the temperature calibration range of the image, draw rectangular boxes for areas in the image where the temperature exceeds the calibration range, and label the highest temperature in K. max Minimum temperature K minThe average temperature μ and standard deviation σ; when there are target boxes in the image, determine whether the number of target boxes for each type of equipment or component is less than the calibrated number. If so, merge the target boxes of that type of equipment into one for processing; otherwise, process them one by one. For images containing target boxes, calculate a dynamic threshold, extract the maximum temperature, temperature point values, average value and standard deviation within each target box, and determine whether there are temperature points within the target box that exceed the dynamic threshold. If so, mark the temperature points exceeding the threshold as suspected abnormal areas; otherwise, discard the target box. Calculate the temperature difference between each suspected abnormal area; retain areas with a temperature difference greater than 15℃ as abnormal heat points. When the abnormal heat point is obtained by merging the target boxes with a number less than the calibration number and then processing them, draw a rectangle around it and label its highest temperature value, lowest temperature Kmin, average temperature value, temperature difference and the specific category of the corresponding equipment or component. When the number of abnormal heat points is greater than or equal to the number of target boxes obtained through subsequent processing, the target boxes are labeled with their highest temperature value, lowest temperature in Kmin, average temperature value, temperature difference, and the specific category of the corresponding equipment or component.

[0005] As a further improvement to the above solution, the infrared abnormal heating image detection method for power distribution network inspection involves reading the temperature value of each pixel in the infrared image using the DJI SDK, constructing a temperature matrix, forming a temperature distribution map, and then reading the highest temperature K in the infrared image from the temperature distribution map. max Minimum temperature K min The average temperature μ, the standard deviation σ of the temperature distribution, and the local temperature maximum, average, and standard deviation within the target box.

[0006] As a further improvement to the above scheme, the target detection model adopts a large visual model of the power distribution network.

[0007] As a further improvement to the above scheme, before the infrared image is detected by the target detection model, the infrared image is first subjected to noise reduction and image quality enhancement: And / or, preprocessing includes removing image noise using noise reduction methods such as Gaussian filtering and median filtering, and improving image quality through enhancement techniques such as contrast stretching and histogram equalization, in order to optimize the detection effect of subsequent target detection models.

[0008] As a further improvement to the above scheme, for device or component categories with fewer than 5 target boxes, the minimum x1 and y1 (x1, y1) values ​​among the top-left corner coordinates (x1, y1) of all target boxes of this type of device are extracted. min y min ), and the coordinates of the lower right corner (x) 2,The largest x2 and y2(x) in y2) max y max ), with (x min y min (x) is the top left corner, (x) max y max Generate a larger bounding box that can encompass all the original bounding boxes for the lower right corner coordinates.

[0009] As a further improvement to the above scheme, the calculation formula for the calibration range is as follows: The temperature distribution in infrared images follows a Gaussian distribution. Where μ is the average temperature, σ is the standard deviation, and x is the temperature value.

[0010] The calibration range is set to a 3σ range (μ-3σ, μ+3σ) that conforms to a Gaussian distribution.

[0011] As a further improvement to the above scheme, the dynamic threshold calculation method is as follows: Where α is the experimental parameter, usually taken as 3, corresponding to the 3σ principle, μ is the average temperature, and σ is the standard deviation.

[0012] As a further improvement to the above scheme, when calculating the temperature difference (TemperatureDifference) for suspected abnormal regions, Where Max Temperature is the highest temperature in the suspected abnormal area within the target box, and AverageTemperature is the category average temperature, which is the average temperature obtained by summing and dividing the average temperatures of all target boxes in that category.

[0013] As a further improvement to the above scheme, the value of the calibration quantity is 5.

[0014] Secondly, the present invention also provides an infrared abnormal heating image detection system for distribution network inspection, used to implement the above-mentioned infrared abnormal heating image detection method, including: Infrared image preprocessing module: performs noise reduction and image quality enhancement on the acquired infrared images of power equipment; adopts a large-scale visual model of power distribution network that has been specifically trained and optimized for power equipment. This model can run efficiently on edge devices, outputs the position coordinates of the target boxes of the equipment and the category of the equipment or component, and counts the number of target boxes of each category of equipment or component in each image. Infrared temperature data extraction module: Utilizes the DJI SDK to read temperature values ​​from infrared images, constructs a temperature matrix to form a temperature distribution map, and calculates temperature statistical parameters, including the maximum temperature K.max Minimum temperature K min , average temperature μ and standard deviation of temperature σ; The target bounding box-based filtering module: determines whether target bounding boxes exist in the infrared image; if so, it counts the number of target bounding boxes in each category. If the number of target bounding boxes in a certain category is ≥5, it proceeds directly to the next step; if the number of target bounding boxes in a certain category is <5, it extracts the minimum top-left corner coordinate (x, y) of all target bounding boxes in that category. min y min ) and the coordinates of the largest bottom right corner (x max y max After generating a unified larger target box that surrounds all the original target boxes, proceed to the next process; if no target box exists, calculate the 3σ range and mark the area outside the range in the temperature distribution map as a suspected abnormal area; The calculation and annotation module calculates a dynamic threshold for images with bounding boxes, obtains the local maximum, local average, and local standard deviation of temperature within each bounding box, and marks the bounding box as a suspected abnormal region if the local maximum exceeds the dynamic threshold. It calculates the temperature difference (maximum temperature of the suspected abnormal region - average temperature of the category) for each suspected abnormal region. When the temperature difference is greater than 15℃, the region is retained as an abnormal heat point. When an abnormal heat point is obtained by merging bounding boxes with a number less than the calibration number and then processing them, a rectangle is drawn around it, and its highest temperature value, lowest temperature in Kmin, average temperature value, temperature difference, and the specific category of the corresponding device or component are annotated. When an abnormal heat point is obtained by merging bounding boxes with a number greater than or equal to the calibration number and then processing them, its highest temperature value, lowest temperature in Kmin, average temperature value, temperature difference, and the specific category of the corresponding device or component are annotated around it.

[0015] 1. Significantly Improved Screening Efficiency: By combining visible light model pre-screening with statistical analysis, the number of infrared images requiring manual analysis can be reduced by more than 80%, greatly improving inspection efficiency. For example, in overhead line inspections, traditional methods require analyzing 590,000 infrared images, while this method only requires analyzing about 80,000.

[0016] 2. Improve anomaly location accuracy: By using visible light model-assisted location and temperature difference threshold judgment, the location accuracy of abnormal heating points can be improved by more than 30%.

[0017] 3. Enhanced System Adaptability: The dynamic temperature threshold adjustment method enables the system to adapt to different ambient temperatures and different equipment or component types, improving the robustness of screening. For example, in high-temperature environments during summer, the traditional fixed threshold method may misjudge normal operating temperatures as abnormal, while this method, by dynamically adjusting the threshold, can still accurately identify truly abnormal heat points. Attached Figure Description

[0018] Figure 1 This is a flowchart of an infrared abnormal heating image detection method for power distribution network inspection. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that when a component is said to be "installed on" another component, it can be directly on the other component or it may be in a component that is centered on it. When a component is said to be "set on" another component, it can be directly set on the other component or it may also be in a component that is centered on it. When a component is said to be "fixed to" another component, it can be directly fixed to the other component or it may also be in a component that is centered on it.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Please see Figure 1 , Figure 1 This invention provides a method for detecting abnormal heating images in infrared images for distribution network inspection. The method can achieve full automation, improve inspection efficiency and accuracy, and is suitable for large-scale distribution network operation and maintenance. It includes the following steps: S1: Visible Light Model Prediction and Target Box Acquisition 1.1 Infrared Image Preprocessing The infrared image of the power equipment is acquired. The infrared image is then subjected to noise reduction and enhancement processing. The noise reduction processing uses noise reduction methods such as Gaussian filtering and median filtering. The enhancement processing uses enhancement techniques such as contrast stretching and histogram equalization to enhance image quality, thereby improving image quality and facilitating subsequent processing.

[0023] 1.2 Object Detection Model Selection and Object Box Classification and Quantity Statistics An object detection model is used to detect power equipment in infrared images. This model can employ a large-scale power distribution network vision model, which has been trained and optimized for power equipment (transformers, insulators, conductors, etc.) to achieve high accuracy (mAP>90%) and high efficiency (inference speed>30 FPS). The infrared images are fed into the large-scale power distribution network vision model, and the model output includes the location coordinates of the bounding boxes and the equipment or component category (e.g., transformers, disconnect switches, etc.). The number of bounding boxes for each equipment or component category in each image is then counted.

[0024] For example, for a certain image, there are 3 bounding boxes for the transformer category and 8 bounding boxes for the insulator category. Each bounding box contains its location coordinate information.

[0025] S2: Infrared Temperature Data Extraction and Statistical Analysis 2.1 DJI SDK Temperature Reading By using the DJI SDK to read the temperature value of each pixel in the infrared image, a temperature matrix is ​​constructed to form a temperature distribution map.

[0026] For example, for an infrared image with a resolution of 320×240, the SDK can extract 320×240 temperature values ​​to form a temperature distribution map.

[0027] 2.2 Calculation of Statistical Parameters Using the DJI SDK described above, after obtaining the temperature distribution map of the entire infrared image, the highest temperature (K) in the infrared image can be read from the temperature distribution map. max Minimum temperature K min Average temperature μ, standard deviation of temperature distribution σ: Maximum value K max The highest temperature value in the image.

[0028] Minimum temperature K min : The lowest temperature value in the image.

[0029] Average value μ: The average temperature value of all pixels in the image.

[0030] Standard deviation σ: Standard deviation of temperature distribution.

[0031] For example, the maximum temperature of an infrared image is 85℃, the average temperature is 35℃, and the standard deviation is 5℃.

[0032] 2.3 Gaussian Distribution Analysis As can be seen from the temperature distribution map above, the temperature distribution of the infrared image follows a Gaussian distribution, and its probability density function is: Where μ is the average temperature, σ is the standard deviation, and x is the temperature value.

[0033] S3: Filtering logic based on the number of target boxes Based on the above statistics on the number of bounding boxes for each device or component category in each image, two scenarios are identified: "bounding boxes appear in the image" and "bounding boxes are not present in the image". 3.1 When a target bounding box appears in the image, there are two possible scenarios: A. When the number of target boxes for a certain type of device is greater than or equal to 5, it is considered that there are enough devices of that type, and the process proceeds directly to the next step of screening.

[0034] B. When the number of bounding boxes for a certain type of device is less than 5, it is considered that there may be few such devices or occlusion, requiring further processing: For device or component categories with fewer than 5 bounding boxes, extract the smallest x1 and y1 (i.e., x1, y1) among the top-left corner coordinates (x1, y1) of all bounding boxes for that type of device. min y min ), and the largest x2 and y2 (i.e., x) in the lower right corner coordinates (x2, y2). max y max ), with (x min y min (x) is the top left corner, (x) max y max Generate a larger bounding box that can encompass all the original bounding boxes for the lower right corner coordinates; For example, when the number of target boxes for the transformer class is 3 and the number of target boxes for the insulator class is 8, the 3 target boxes for the transformer class are merged into 1 larger target box to cover a more comprehensive area. The insulator class then proceeds directly to the next step.

[0035] 3.2 When there is no target bounding box in an image, The selection is based on the Gaussian distribution of the entire image: the temperature range of 3σ (μ-3σ, μ+3σ) that conforms to the Gaussian distribution is calculated, and rectangular boxes are drawn for temperature regions in the infrared image that exceed the 3σ range, with the highest temperature in K marked. max Minimum temperature K min The average temperature μ and the standard deviation of the temperature distribution σ.

[0036] For example, an infrared image has a maximum temperature of 85℃, an average temperature of 35℃, and a standard deviation of 5℃. If no target bounding box is detected, the calculated 3σ range is 20℃ to 50℃ (35℃ - 15℃ to 35℃ + 15℃). Based on this 3σ range, rectangular bounding boxes are drawn around all pixels in the infrared image whose temperatures fall outside this range, and the highest temperature (K) is labeled accordingly. max The average temperature μ and the standard deviation of the temperature distribution σ.

[0037] S4: Dynamic threshold adjustment based on Gaussian distribution 4.1 Calculate the dynamic threshold for the infrared images containing target bounding boxes. Calculate the dynamic temperature threshold based on the Gaussian distribution parameters of the entire image: Where α is the experimental parameter (usually taken as 3, corresponding to the 3σ principle), μ is the average temperature, and σ is the standard deviation.

[0038] For example, if a certain image has μ=35℃ and σ=5℃, then Threshold=35+3×5=50℃.

[0039] 4.2 Preliminary Screening of Abnormal Hot Spots The target bounding boxes in the infrared image are combined with a dynamic threshold for preliminary screening: a. The temperature distribution map obtained through the DJI SDK can be used to read the local maximum temperature value, temperature value at each point, average value, and standard deviation within each target box; b. If any temperature point within the target box exceeds the dynamic threshold, and if it does, mark the temperature point exceeding the threshold as a suspected abnormal area; otherwise, discard the target box. The above filtering steps are performed on all target boxes of each of the above devices to obtain the suspected abnormal areas of the infrared image.

[0040] By following the steps above, we can obtain suspected abnormal areas from all possible scenarios in infrared images.

[0041] S5: Final screening based on temperature difference 5.1 Temperature Difference Calculation When calculating the temperature difference for suspected abnormal regions Where Max Temperature is the highest temperature in the suspected abnormal area within the target box, and AverageTemperature is the category average temperature, which is the average temperature obtained by summing and dividing the average temperatures of all target boxes in that category.

[0042] 5.2 Temperature Difference Threshold Judgment Set a temperature difference threshold (usually 15℃) for final screening: if the temperature difference exceeds 15℃, retain the area as an abnormal heat source; otherwise, remove the area. For example, if the highest temperature of a target box is 85℃ and the average temperature is 35℃, then the temperature difference (TemperatureDifference) is 50℃ > 15℃, and it is retained as an abnormal region.

[0043] 5.3 Marking of Abnormal Heating Points When an abnormal heat point is obtained by merging the above target boxes (where the number of target boxes is less than the calibrated number) and then processing them, draw a rectangle around it and label its highest temperature value, lowest temperature value, average temperature value, temperature difference, and the specific category of the corresponding equipment or component. When the number of abnormal heat points is greater than or equal to the number of target boxes specified above and obtained through subsequent processing, the highest temperature value, lowest temperature value, average temperature value, temperature difference, and specific category of the corresponding equipment or component are marked on the target boxes.

[0044] In summary, this invention achieves end-to-end closed-loop processing from power equipment image acquisition and target localization to temperature analysis and anomaly confirmation through a fully automated design of "detection-optimization-screening-annotation." Specifically, the "detection" stage relies on a target detection model to accurately identify the type and location of equipment or components, replacing the tedious traditional manual image searching; the "optimization" stage solves the problems of ambiguous positioning and misjudgment of areas without targets in small batches through target box merging and 3σ range analysis; the "screening" stage uses a dual standard of dynamic threshold and temperature difference to avoid the incompatibility of fixed thresholds with complex environments; and the "annotation" stage intuitively presents the core fault parameters, providing maintenance personnel with clear decision-making basis.

[0045] This automated solution not only reduces the traditional 2-3 minute analysis time per image by manual means to seconds, significantly alleviating the workload of inspection personnel in processing massive amounts of data, but also significantly reduces the false alarm rate through precise verification at multiple stages, effectively avoiding missed detection of real faults and misjudgment of normal areas. Ultimately, this method can adapt to the large-scale inspection needs of thousands to tens of thousands of images per inspection in scenarios such as overhead distribution lines, providing key technical support for the transformation of power grid operation and maintenance from "manual-driven" to "intelligent and efficient," and helping to improve the timeliness of power equipment fault early warning and the stability of power grid operation.

[0046] In other embodiments, an infrared abnormal heating image detection system for distribution network inspection is also provided to implement the method, including: Infrared image preprocessing module: The infrared images are denoised and the image quality is enhanced. The large visual model of the power distribution network is adopted. After training and optimization for power equipment, it can run efficiently on edge devices to obtain the location coordinates of the target box and the category of the equipment or component (such as transformer, disconnect switch, etc.). The number of target boxes of each equipment or component category in each image is counted.

[0047] Infrared temperature data extraction module: Utilizes the DJI SDK to read temperature values ​​from infrared images, constructs a temperature matrix, generates a temperature distribution map, and calculates and statistically analyzes various temperature-related parameters (maximum temperature K). maxMinimum temperature K min (mean μ and standard deviation σ).

[0048] The target bounding box-based filtering module determines whether a target bounding box exists in the infrared image based on the above information. If a target bounding box exists, it counts the number of target bounding boxes in each category. If the number of target bounding boxes in a certain category is ≥5, it proceeds directly to the next process. If the number of target bounding boxes is <5, it merges all target bounding boxes of that category into a unified larger target bounding box and proceeds to the next process.

[0049] If there is no target bounding box predicted by the target detection model in the image, calculate the 3σ range (μ±3σ). Draw a rectangular bounding box for the region outside this range and label it with the highest temperature K. max Minimum temperature K min Average temperature μ and temperature standard deviation σ.

[0050] The calculation and annotation module calculates a dynamic threshold for images with target boxes, obtains the local maximum, average, and standard deviation of temperature within each target box, and determines whether any temperature points within the target box exceed the dynamic threshold. If so, the temperature points exceeding the threshold are marked as suspected abnormal areas; otherwise, the target box is discarded. The temperature difference (maximum temperature of suspected abnormal areas – average temperature of the category) is calculated. If the temperature difference is greater than 15℃, the area is retained as an abnormal heat point. When an abnormal heat point is obtained by merging target boxes with a number less than the calibration number and then processing them, a rectangle is drawn around it, and its highest temperature value, lowest temperature in Kmin, average temperature value, temperature difference, and the specific category of the corresponding device or component are annotated. When an abnormal heat point is obtained by merging target boxes with a number greater than or equal to the calibration number and then processing them, its target box is annotated with its highest temperature value, lowest temperature in Kmin, average temperature value, temperature difference, and the specific category of the corresponding device or component.

[0051] The infrared abnormal heating image detection system for power distribution network inspection provided in this embodiment achieves full-process processing from image optimization and data acquisition to anomaly detection and annotation through four major modules: infrared image preprocessing, infrared temperature data extraction, target box screening, and calculation and annotation. It provides differentiated processing for scenarios with and without target boxes, accurately identifies abnormal heating points and annotates key parameters, efficiently adapting to the needs of power distribution network inspection.

[0052] The advantages of this invention are: 1. Significantly Improved Screening Efficiency: By combining visible light model pre-screening with statistical analysis, the number of infrared images requiring manual analysis can be reduced by more than 80%, greatly improving inspection efficiency. For example, in overhead line inspections, traditional methods require analyzing 590,000 infrared images, while this method only requires analyzing about 80,000.

[0053] 2. Improve anomaly location accuracy: By using visible light model-assisted location and temperature difference threshold judgment, the location accuracy of abnormal heating points can be improved by more than 30%.

[0054] 3. Enhanced System Adaptability: The dynamic temperature threshold adjustment method enables the system to adapt to different ambient temperatures and different equipment or component types, improving the robustness of screening. For example, in high-temperature environments during summer, the traditional fixed threshold method may misjudge normal operating temperatures as abnormal, while this method, by dynamically adjusting the threshold, can still accurately identify truly abnormal heat points.

[0055] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0056] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for detecting abnormal heating in infrared images for distribution network inspection, used to identify potential faults in power equipment, and to label the type of faulty equipment or component and its temperature parameters, characterized in that: Infrared images of power equipment are acquired, and the location coordinates and quantity of target boxes for each equipment or component category are extracted using a target detection model; simultaneously, the corresponding temperature distribution map and the highest temperature (K) are acquired. max Minimum temperature K min Average temperature and temperature distribution standard deviation ; Processing is categorized by the number of target boxes: When there are no target boxes, calculate the temperature calibration range, draw a bounding box for areas outside the range, and label them with K. max K min , and When there are bounding boxes, if the number of bounding boxes of a certain type is less than the number of calibration boxes, they are merged into one process; if the number is greater than or equal to the number of calibration boxes, they are processed independently. For images with bounding boxes, calculate the dynamic threshold, extract the maximum temperature, temperature values, average value and standard deviation within each bounding box, and mark temperature points that exceed the dynamic threshold as suspected abnormal areas; otherwise, discard them. Calculate the temperature difference between each suspected abnormal area; retain areas with a difference greater than 15℃ as abnormal heat points. If the abnormal heat source comes from a merged box, draw a rectangle and label it with the highest temperature value, lowest temperature value, average temperature value, temperature difference, and the specific category of the equipment or component; if it comes from a single box, directly label the above temperature parameters and category in the original target box. Before the infrared image is detected by the target detection model, it is first subjected to noise reduction and image quality enhancement: And / or, preprocessing includes removing image noise using Gaussian filtering and median filtering, and improving image quality using enhancement techniques including contrast stretching and histogram equalization, in order to optimize the detection performance of subsequent target detection models; The method for obtaining infrared image temperature parameters involves reading the temperature value of each pixel in the infrared image using the DJI SDK, constructing a temperature matrix, forming a temperature distribution map, and then reading the highest temperature (K) in the infrared image from the temperature distribution map. max Minimum temperature K min Average temperature and the standard deviation of temperature distribution The local maximum, average, and standard deviation of temperature within the target box.

2. The infrared abnormal heating image detection method for distribution network inspection according to claim 1, characterized in that, The target detection model adopts a large-scale visual model of the power distribution network.

3. The infrared abnormal heating image detection method for distribution network inspection according to claim 1, characterized in that, For device or component categories with fewer than 5 bounding boxes, extract the smallest x1 and y1 (x1, y1) among the top-left corner coordinates (x1, y1) of all bounding boxes in that category. min y min ), and the coordinates of the lower right corner (x) 2, The largest x2 and y2(x) in y2) max y max ), with (x min y min (x) is the top left corner, (x) max y max Generate a larger bounding box that can encompass all original bounding boxes for the lower right corner coordinates.

4. The infrared abnormal heating image detection method for distribution network inspection according to claim 1, characterized in that, The formula for calculating the calibration range is as follows: The temperature distribution in infrared images follows a Gaussian distribution. in, The average temperature. Here, x represents the standard deviation, and x represents the temperature value. The calibration range is set to 3 ohms that conform to a Gaussian distribution. scope( , ).

5. The infrared abnormal heating image detection method for distribution network inspection according to claim 1, characterized in that, The above dynamic threshold (Threshold) is calculated as follows: ； in, For experimental parameters, it is usually set to 3, corresponding to 3 in principle, The average temperature. The standard deviation is denoted as .

6. The infrared abnormal heating image detection method for distribution network inspection according to claim 1, characterized in that, When calculating the temperature difference for suspected abnormal regions Temperature Difference=Max Temperature-Average Temperature Where Max Temperature is the highest temperature in the suspected abnormal area within the target box, and Average Temperature is the category average temperature, which is the average temperature obtained by summing and dividing the average temperatures of all target boxes in that category.

7. The infrared abnormal heating image detection method for distribution network inspection according to claim 1, characterized in that, The value of the calibration quantity is 5.

8. An infrared abnormal heating image detection system for power distribution network inspection, used to implement the infrared abnormal heating image detection method of claim 1, characterized in that, include: Infrared image preprocessing module: performs noise reduction and image quality enhancement on the acquired infrared images of power equipment; A large visual model of the power distribution network, which has been specifically trained and optimized for power equipment, is adopted. This model can run efficiently on edge devices, outputting the location coordinates of the target boxes of the devices and the categories of the devices or components, and counting the number of target boxes of each category of device or component in each image. Infrared temperature data extraction module: Utilizes the DJI SDK to read temperature values ​​from infrared images, constructs a temperature matrix to form a temperature distribution map, and calculates temperature statistical parameters, including the maximum temperature K. max Minimum temperature K min Average temperature and temperature standard deviation ; Target bounding box-based filtering module: Determines whether a target bounding box exists in the infrared image; If they exist, count the number of bounding boxes for each category. If the number of bounding boxes for a certain type of device or component is ≥5, proceed directly to the next step. If the number of bounding boxes for a certain type of device or component is <5, extract the minimum top-left corner coordinate (x, y) of all bounding boxes for that type of device or component. min y min ) and the coordinates of the largest bottom right corner (x max y max After generating a unified, larger bounding box that surrounds all original bounding boxes, proceed to the next step; if no bounding box exists, calculate 3 The range is used to mark areas on the temperature distribution map that exceed this range as suspected abnormal areas. The calculation and annotation module calculates a dynamic threshold for images with target boxes, obtains the local maximum temperature, temperature point values, local average temperature, and local standard deviation within each target box, and determines whether there are any temperature points within the target box that exceed the dynamic threshold. If so, the temperature points exceeding the threshold are marked as suspected abnormal areas; otherwise, the target box is discarded. For each suspected abnormal area, calculate the temperature difference: Temperature difference = Maximum temperature of suspected abnormal area - Average temperature of category. When the temperature difference is greater than 15℃, the area is retained as an abnormal heat point. When an abnormal heat point is obtained by merging the above target boxes (the number of target boxes is less than the calibration number) and then processing them, draw a rectangle around it and label it with its highest temperature value, lowest temperature (Kmin), average temperature value, temperature difference, and the specific category of the corresponding equipment or component. When an abnormal heat point is obtained by merging the above target boxes (the number of target boxes is greater than or equal to the calibration number) and then processing them, label its target box with its highest temperature value, lowest temperature (Kmin), average temperature value, temperature difference, and the specific category of the corresponding equipment or component.

Images