A Method and System for Cable Fault Target Detection Based on Improved YOLO10 Infrared Image

Abstract

This invention provides a method and system for cable fault infrared image target detection based on an improved YOLO10, belonging to the field of cable fault detection technology. The method includes: acquiring historical infrared and visible light images and fault records of the cable under test; generating an RGB-T image fused with multimodal information; layer-by-layer labeling of fault areas, types, and temperature characteristics; constructing and training an improved YOLO10 model; inputting the current RGB-T image to obtain the labeled content; determining the highest-risk fault point and redrawing the bounding box; simulating the temperature field diffusion over the next 12 hours, calculating the proportion of areas with excessive sustained expansion rate to determine the insulation breakdown risk level. This invention solves single-modal defects through RGB-T image fusion, improves positioning accuracy through layered labeling and an improved model, and achieves dynamic risk assessment by combining dynamic bounding boxes and temperature field simulation, significantly improving the accuracy of cable fault detection and risk determination.

Description

Technical Field

[0001] This invention relates to the field of cable fault detection technology, specifically to a cable fault infrared image target detection method and system based on an improved YOLO10. Background Technology

[0002] In power system operation and maintenance, cables serve as a critical carrier for power transmission, and their operating status directly affects the safety and stability of the power grid. Currently, fault detection technology based on infrared imaging is widely used in cable operation and maintenance because it can intuitively reflect the temperature distribution of objects, and it achieves fault identification by capturing the thermal radiation characteristics of the fault area; at the same time, ultraviolet imaging technology is also often used for auxiliary detection.

[0003] In the prior art, CN109784336A provides a method for identifying fault points in infrared images based on the YOLO target detection algorithm, which includes the following steps: (1) collecting infrared or ultraviolet image data of cable joints as a dataset using an infrared or ultraviolet imager; (2) enhancing the collected dataset based on limited contrast adaptive histogram equalization, labeling the original images using image annotation software, and randomly dividing the training set and test set at a ratio of 3:1; (3) inputting the training set into the improved YOLOv7 model for fault diagnosis training to obtain the trained model; (4) using the trained improved YOLOv7 model to perform effect testing and fault diagnosis on infrared or ultraviolet images in the test sample library, and judging the severity of the fault by comparing the fault overlap areas of the two types of images.

[0004] However, the following shortcomings still exist. As mentioned above, single-modal images (infrared or ultraviolet) have inherent limitations. Infrared images lack cable surface texture, making it difficult to distinguish between "insulation layer damage" and "surface stains." Ultraviolet images have low sensitivity to low-temperature faults, making it easy to miss early faults. This lack of information directly leads to the inability to refine fault types and supplement key parameters such as temperature gradients during annotation, resulting in only simple region annotation. Simple annotation prevents the model from learning the unique characteristics of different faults during training. When multiple faults occur concurrently (such as "overheating" and "physical damage" coexisting), the model struggles to extract distinguishing criteria from single-modal images. The final output bounding box is prone to containing redundant background or missing some fault areas. The model cannot accurately segment the boundaries of different faults, resulting in large positioning errors and reducing the accuracy of judging the degree of cable faults.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for detecting cable faults using infrared images based on an improved YOLO10, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for detecting cable faults using infrared images based on an improved YOLO10, comprising the following steps:

[0009] S1. Obtain infrared and visible light images and fault records of the cable under test over a historical period. After aligning the coordinates of the two types of images, extract the infrared pixel temperature values ​​and embed them into the RGB three channels to generate an RGB-T image.

[0010] S2. Based on the fault record, the RGB-T image is labeled in layers. The coordinates and area of ​​the fault region bounding box are labeled once, and the fault type in the region is labeled twice. The temperature gradient distribution characteristics and temperature thresholds of each type are supplemented.

[0011] S3. Construct an improved YOLO10 model, using RGB-T images of the cable under test at various historical time periods as input and the labeled content as output labels, and train the model;

[0012] S4. Input the RGB-T image of the cable under test for the current time period into the model, and output the labeled content;

[0013] S5. Based on the fault area boundary box, RGB-T image and fault type of the cable under test in the current time period, determine the type and number of fault points within the boundary box. Combine the boundary box coordinates, temperature gradient distribution characteristics, area and temperature threshold of each fault type, screen the highest risk fault point, use the fault area boundary box as the initial boundary box, determine the boundary box type according to the positional relationship between the highest risk fault point and the initial boundary box, and redraw the fault area boundary box with the highest risk fault point as the center.

[0014] S6. Perform finite element model analysis according to the scope of the redefined fault area boundary box, simulate the temperature field diffusion process in the next 12 hours in combination with environmental parameters, dynamically output the temperature field distribution at different times, obtain the continuous expansion rate of the region at two adjacent times, calculate the proportion of groups exceeding the preset threshold, and determine the insulation breakdown risk level of the cable under test.

[0015] Furthermore, environmental parameters include temperature, wind speed, and relative humidity.

[0016] Furthermore, after aligning the coordinates of the two types of images, the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate an RGB-T image. The specific steps are as follows:

[0017] From infrared and visible light images, the corner of the cable joint is selected as a common feature point, and the pixel coordinates of the feature point in the two types of images are recorded respectively.

[0018] The feature points are calculated using a feature point matching algorithm to obtain the coordinate deviation between the two types of images, and a coordinate transformation matrix is ​​established based on this coordinate deviation.

[0019] Geometric transformations are performed on the infrared image based on the coordinate transformation matrix, including correcting positional deviations by translation, unifying angles by rotation, and matching sizes by scaling, so that the pixel coordinates at the same position in the infrared image and the visible light image completely coincide, thereby completing coordinate alignment;

[0020] Based on the coordinate-aligned infrared image, the infrared pixel temperature corresponding to each pixel is read;

[0021] For noisy pixels in the image, a neighborhood mean filtering method is used to replace the outlier value with the average temperature of the normal pixels around the pixel. For pixels with missing temperature data, the temperature value is supplemented by interpolation, so as to obtain continuous and complete infrared pixel temperature.

[0022] Establish a mapping table between pixel position and processed infrared pixel temperature, and clarify the temperature information of each coordinate point;

[0023] The R, G, and B channel pixel values ​​of the visible light image are retained and used as the basic information of the RGB-T image;

[0024] Based on the determined coordinate point temperature information, the acquired continuous and complete infrared pixel temperature is standardized and converted into a numerical form compatible with RGB channels through grayscale mapping.

[0025] A four-channel fusion method is adopted, in which the processed temperature value is used as the T channel and combined with the RGB three-channel data to form an RGB-T image containing spatial texture and temperature information.

[0026] Furthermore, the specific steps of S2 are as follows:

[0027] Extract the fault time, location, type, and temperature characteristics of the cable under test from historical fault records;

[0028] Based on the time and location of the fault, the corresponding image is matched in the historical RGB-T image, and the suspected fault area is identified by its spatial texture features and temperature features.

[0029] Based on the suspected fault area, a labeling is made. Referring to the cable structure characteristics of the RGB channel and the temperature anomaly range of the T channel, the boundary box of the fault area is defined and the coordinates are recorded. At the same time, the area is calculated and labeled.

[0030] Secondary labeling is performed based on the recorded fault type. The specific fault type is confirmed by combining the spatial texture features and temperature features of the suspected fault area. Then, the temperature gradient distribution features of this type are extracted within the bounding box to determine and label the temperature threshold range.

[0031] Furthermore, based on the fault area boundary box, RGB-T image, and fault type of the cable under test for the current time period, the type and number of fault points within the boundary box are determined. The specific steps are as follows:

[0032] Retrieve the coordinates of the fault area bounding box, the corresponding RGB-T image, and the labeled fault type of the cable under test for the current time period output by S4;

[0033] Locate the bounding box region in the RGB-T image, and extract the spatial texture features of the RGB channels and the temperature distribution features of the T channel within the region;

[0034] By combining the spatial texture features and temperature gradient distribution features corresponding to each fault type in the secondary annotation, the region within the bounding box is analyzed.

[0035] If there is only one region within the bounding box that matches the characteristics of a certain fault type, then that region is determined to be a fault point of the corresponding type. If there are multiple suspected regions, their spatial texture features and temperature distribution features are compared according to the sub-regions to determine whether they are multiple fault points of the same type or fault points of different types.

[0036] Record the location coordinates of each fault point and the corresponding fault type, including a single fault point, multiple fault points of the same type, or multiple fault points of different types.

[0037] Furthermore, by combining the bounding box coordinates, temperature gradient distribution characteristics of each fault type, area, and temperature threshold, the highest-risk fault points are screened. The specific steps are as follows:

[0038] For each fault point, the temperature data of that fault point is extracted from channel T, and the difference between the temperature data and the corresponding fault type temperature threshold is calculated using the following formula:

[0039] ;

[0040] in, For the first Temperature difference at each fault point The first one extracted from channel T The temperature at each fault point For the first Temperature thresholds for each fault point corresponding to its fault type For the index of the fault point, , The number of fault points;

[0041] Based on the temperature gradient distribution characteristics corresponding to different fault types, a diffusion risk coefficient is set:

[0042] If the fault type is joint overheating and the temperature gradient distribution is a steep gradient, then the diffusion risk coefficient is assigned a value of [value missing]. ;

[0043] If the fault type is insulation aging and the temperature gradient distribution is a gentle gradient, then the diffusion risk coefficient is assigned a value of [value missing]. ;

[0044] in, ,and All are constants greater than 0;

[0045] The ratio of the area of ​​the sub-region where the fault point is located to the area of ​​the bounding box is calculated to generate the area proportion, based on the following formula:

[0046] ;

[0047] in, For the first The area percentage of each fault point For the first The area of ​​the sub-region where each fault point is located. For the first The area of ​​the bounding box of the fault region where each fault point is located;

[0048] The first Temperature difference at each fault point After normalization, based on the normalized temperature difference, diffusion risk coefficient, and area proportion, the comprehensive risk coefficient of each fault point is calculated using the following formula:

[0049] ;

[0050] in, For the first The overall risk coefficient of each failure point For the first The risk coefficient of the spread of each fault point;

[0051] If the fault type is joint overheating, the first Risk coefficient of diffusion at each fault point If the fault type is insulation aging, the first Risk coefficient of diffusion at each fault point ;

[0052] The comprehensive risk coefficients of all failure points are sorted in descending order, and the failure point corresponding to the comprehensive risk coefficient ranked first is the highest risk failure point.

[0053] When there are at least two comprehensive risk factors that rank first;

[0054] Compare the temperature differences between the fault points and select the fault point with the largest temperature difference as the highest risk fault point.

[0055] If the temperature difference is still the same, then further compare their diffusion risk coefficients and select the fault point with the largest diffusion risk coefficient as the highest risk fault point.

[0056] If both of the above are the same, compare the area proportion of the fault point and select the fault point with the largest area proportion as the highest risk fault point.

[0057] When there is only one fault point within the bounding box, there is no need to sort; the fault point is directly identified as the highest-risk fault point.

[0058] Furthermore, using the fault region boundary box as the initial boundary box, the boundary box type is determined based on the positional relationship between the highest-risk fault point and the initial boundary box. The fault region boundary box is then redefined with the highest-risk fault point as the center. The specific steps are as follows:

[0059] If the highest risk fault point is completely inside the initial bounding box, that is, all parts of the fault point are within the initial bounding box, then the bounding box type is internal.

[0060] If the highest-risk failure point is partially inside the initial bounding box and partially outside the initial bounding box, then the bounding box type is partially extended.

[0061] If the highest risk fault point is completely outside the initial bounding box, that is, all parts of the fault point are not within the initial bounding box, then the bounding box type is external.

[0062] Centered on the highest-risk fault point, the fault area boundary box is redefined based on the determined boundary box type:

[0063] For internal fault types, the newly defined fault area boundary box is limited to the initial boundary box, centered on the highest risk fault point, covering the fault point and the affected area around it;

[0064] For some extended types, the newly defined fault area bounding box must include both the portion of the highest-risk fault point within the initial bounding box and the portion beyond it;

[0065] For external fault types, the newly defined fault area boundary box is centered on the highest-risk fault point and is set according to the influence range of the fault point, without being limited by the initial boundary box.

[0066] Furthermore, finite element model analysis is performed according to the newly defined boundary of the fault area. Combined with environmental parameters, the temperature field diffusion process over the next 12 hours is simulated, dynamically outputting the temperature field distribution at different times, and obtaining the regional expansion rate between two adjacent times. The specific steps are as follows:

[0067] The spatial scope of finite element calculation is defined by using the newly defined fault area boundary box as the boundary.

[0068] Based on the bounding box and cable configuration, the region is divided into small finite element units, which serve as the basic units for temperature field calculation.

[0069] Temperature, wind speed, and relative humidity are converted into boundary conditions of the finite element model. The current temperature of the highest risk fault point is set as the initial temperature and assigned to its element. The initial temperatures of other elements are set with reference to historical data.

[0070] Based on Fourier's law of heat conduction, a three-dimensional unsteady-state heat conduction equation is established. The equation is discretized to each element using finite element software, and solved using the time-stepping method.

[0071] Set the time step and calculate the temperature change of all units in each step;

[0072] By combining boundary conditions, the temperature field distribution at each time step is iteratively solved until convergence. After each time step, the temperature values ​​of all cells at that time step are saved, forming temperature field distribution data at 12 time steps, denoted as . ,in, For the first The temperature at that moment For indexing time;

[0073] Set the preset temperature threshold For the temperature data at each moment, the elements in the finite element unit whose temperature exceeds the preset threshold are identified one by one and identified as the over-threshold region;

[0074] The number of finite element elements in the region exceeding the threshold at each time step is counted, and the region expansion rate between two adjacent time steps is calculated using the following formula:

[0075] ;

[0076] in, For the first The moment to the The rate of sustained regional expansion at any given time. For the first The number of finite element elements in the region exceeding the threshold at each time point. For the first The number of finite element elements in the region exceeding the threshold at each time point. , This represents the time interval between two adjacent moments.

[0077] Furthermore, the percentage of groups exceeding a preset threshold is calculated to determine the insulation breakdown risk level of the cable under test. The specific steps are as follows:

[0078] From the The moment to the There are 11 time points, with 11 adjacent time points. The statistics are as follows: The number of groups is denoted as , A preset threshold for the rate of continued regional expansion;

[0079] The percentage of groups exceeding a preset threshold across all adjacent time points is calculated using the following formula:

[0080] ;

[0081] in, The percentage of groups exceeding a preset threshold across all adjacent time points;

[0082] like If the insulation of the cable under test is low, then the risk of insulation breakdown is low.

[0083] like If the insulation breakdown risk of the cable under test is moderate, then the cable under test is of medium risk.

[0084] like If the insulation of the cable under test is damaged, the risk of insulation breakdown is high.

[0085] in, To classify the critical values ​​for low and medium risk of insulation breakdown, The critical values ​​are used to classify the risk of insulation breakdown as medium or high.

[0086] To achieve the above objectives, the present invention also provides the following technical solution:

[0087] A cable fault infrared image target detection system based on an improved YOLO10, the system being used to execute any of the above-described cable fault infrared image target detection methods based on an improved YOLO10, comprising:

[0088] The image acquisition module is used to acquire infrared and visible light images and fault records of the cable under test over a historical period. After aligning the coordinates of the two types of images, the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate an RGB-T image.

[0089] The annotation module is used to perform layered annotation of RGB-T images based on fault records. It first annotates the coordinates and area of ​​the fault region bounding box, then annotates the fault type within the region, and supplements the temperature gradient distribution features and temperature thresholds for each type.

[0090] The model building module is used to build an improved YOLO10 model. It takes the RGB-T images of the cable under test in various historical time periods as input and the labeled content as output labels to train the model.

[0091] The simulation module is used to input the RGB-T image of the cable under test for the current time period into the model and output the labeled content;

[0092] The filtering module is used to determine the type and number of fault points within the boundary box based on the fault area boundary box, RGB-T image and fault type of the cable under test in the current time period. It combines the boundary box coordinates, temperature gradient distribution characteristics of each fault type, area and temperature threshold to filter the highest risk fault point. The fault area boundary box is used as the initial boundary box. The boundary box type is determined according to the positional relationship between the highest risk fault point and the initial boundary box. The fault area boundary box is redefined with the highest risk fault point as the center.

[0093] The analysis module is used to perform finite element model analysis according to the range of the redefined fault area boundary box, simulate the temperature field diffusion process in the next 12 hours in combination with environmental parameters, dynamically output the temperature field distribution at different times, obtain the continuous expansion rate of the region between two adjacent times, calculate the proportion of groups exceeding the preset threshold, and determine the insulation breakdown risk level of the cable under test.

[0094] Compared with the prior art, the beneficial effects of the present invention are:

[0095] This invention achieves multimodal information fusion by generating RGB-T images, laying the foundation for accurate detection. The image integrates infrared temperature information, visible light texture, and temperature value, directly making up for the shortcomings of single-modal images in the prior art. It solves the problem that infrared images are difficult to distinguish between insulation layer damage and surface stains due to the lack of texture, and overcomes the problem of early fault missed detection caused by the low sensitivity of ultraviolet images to low-temperature faults. It provides comprehensive and accurate feature support for subsequent fault identification.

[0096] By relying on hierarchical annotation and improved models to optimize fault location accuracy, hierarchical annotation is used to supplement fault types and temperature gradients. Combined with an improved YOLO10 model trained with the annotation content as the output label, the model can learn the unique features of different faults and achieve accurate segmentation in multi-fault concurrent scenarios. The output bounding box no longer contains redundant background or omits some fault areas. Combined with the redefined bounding box that accurately corresponds to the influence range of the fault point, the problem of existing technologies being unable to accurately segment different fault boundaries and having large positioning deviations is completely solved.

[0097] The risk assessment is upgraded by using dynamic boundary boxes and temperature field simulation. The redefined boundary boxes focus on the highest risk fault points, ensuring the accuracy of the subsequent analysis scope. On this basis, the temperature field diffusion is simulated by combining environmental parameters, and the risk is dynamically assessed by calculating the proportion of the region's continuous expansion rate. This breaks through the limitations of the existing static assessment technology and significantly improves the accuracy of the insulation breakdown risk level determination. Attached Figure Description

[0098] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0099] Figure 2 This is a block diagram of the modules of the present invention;

[0100] Figure 3 This is a schematic diagram showing the fitting of the comprehensive risk coefficient and the normalized temperature difference in this invention.

[0101] Figure 4 This is a schematic diagram showing the fitting of the comprehensive risk coefficient and the diffusion risk coefficient of the present invention;

[0102] Figure 5 This is a schematic diagram showing the fitting of the comprehensive risk coefficient and area ratio of the present invention. Detailed Implementation

[0103] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0104] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0105] Example 1:

[0106] Please see Figures 1-5 The present invention provides a technical solution:

[0107] A method for detecting cable faults using infrared images based on an improved YOLO10, comprising the following steps:

[0108] S1. Obtain infrared and visible light images and fault records of the cable under test over a historical period. After aligning the coordinates of the two types of images, extract the infrared pixel temperature values ​​and embed them into the RGB three channels to generate an RGB-T image.

[0109] Based on the above embodiments, infrared and visible light images and fault records of the cable under test over a historical period are obtained. The specific steps are as follows:

[0110] Clearly define the cable model, laying environment, and service life, and select historical data from the past 3-5 years to ensure that the samples include different seasons, load conditions, and fault states.

[0111] The cable joints are photographed periodically using an infrared thermal imager, and the image acquisition time, ambient temperature, humidity and real-time load current of the cable are recorded.

[0112] Simultaneously use a high-definition visible light camera to capture the same area, ensuring that the shooting angles and distances of the infrared and visible light images are consistent, thus providing a basis for subsequent coordinate alignment;

[0113] Extract historical fault records from the cable maintenance log, including the fault occurrence time, specific location, fault type, fault handling measures, and post-fault detection data;

[0114] Associate fault records with infrared / visible light images of the corresponding time period, and label the image number, load parameters and environmental conditions when the fault occurred;

[0115] Remove blurry, overexposed, or invalid images caused by equipment malfunction, and retain clear infrared and visible light images;

[0116] Historical data is classified into normal state, minor anomaly, and severe fault to ensure a balanced number of samples at each level, providing a comprehensive label distribution for model training.

[0117] Based on the above embodiments, after aligning the coordinates of the two types of images, the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate an RGB-T image. The specific steps are as follows:

[0118] From infrared and visible light images, the corner of the cable joint is selected as a common feature point, and the pixel coordinates of the feature point in the two types of images are recorded respectively.

[0119] The feature points are calculated using a feature point matching algorithm to obtain the coordinate deviation between the two types of images, and a coordinate transformation matrix is ​​established based on this coordinate deviation.

[0120] Geometric transformations are performed on the infrared image based on the coordinate transformation matrix, including correcting positional deviations by translation, unifying angles by rotation, and matching sizes by scaling, so that the pixel coordinates at the same position in the infrared image and the visible light image completely coincide, thereby completing coordinate alignment;

[0121] Based on the coordinate-aligned infrared image, the infrared pixel temperature corresponding to each pixel is read;

[0122] For noisy pixels in the image, a neighborhood mean filtering method is used to replace the outlier value with the average temperature of the normal pixels around the pixel. For pixels with missing temperature data, the temperature value is supplemented by interpolation, so as to obtain continuous and complete infrared pixel temperature.

[0123] Establish a mapping table between pixel position and processed infrared pixel temperature, and clarify the temperature information of each coordinate point;

[0124] The R, G, and B channel pixel values ​​of the visible light image are retained and used as the basic information of the RGB-T image;

[0125] Based on the determined coordinate point temperature information, the acquired continuous and complete infrared pixel temperature is standardized and converted into a numerical form compatible with RGB channels through grayscale mapping.

[0126] A four-channel fusion method is adopted, in which the processed temperature value is used as the T channel and combined with the RGB three-channel data to form an RGB-T image containing spatial texture and temperature information.

[0127] S2. Based on the fault record, the RGB-T image is labeled in layers. The coordinates and area of ​​the fault region bounding box are labeled once, and the fault type in the region is labeled twice. The temperature gradient distribution characteristics and temperature thresholds of each type are supplemented.

[0128] Based on the above embodiments, the specific steps of S2 are as follows:

[0129] Extract the fault time, location, type, and temperature characteristics of the cable under test from historical fault records;

[0130] Based on the time and location of the fault, the corresponding image is matched in the historical RGB-T image, and the suspected fault area is identified by its spatial texture features and temperature features.

[0131] Based on the suspected fault area, a labeling is made. Referring to the cable structure characteristics of the RGB channel and the temperature anomaly range of the T channel, the boundary box of the fault area is defined and the coordinates are recorded. At the same time, the area is calculated and labeled.

[0132] Secondary labeling is performed based on the recorded fault type. The specific fault type is confirmed by combining the spatial texture features and temperature features of the suspected fault area. The temperature gradient distribution features of this type are extracted within the bounding box, and the temperature threshold range is determined and labeled.

[0133] S3. Construct an improved YOLO10 model, using RGB-T images of the cable under test at various historical time periods as input and the labeled content as output labels, and train the model;

[0134] Based on the above embodiments, the improved YOLO10 model is a multimodal fusion target detection model based on the original YOLO10 target detection framework. It focuses on optimizing the temperature feature fusion capability and small target fault detection accuracy for the characteristics of infrared images of cable faults. It is suitable for fault area localization and type identification of RGB-T images.

[0135] Components of the improved YOLO10 model:

[0136] (1) Multimodal feature fusion

[0137] The input layer receives 4 channels of data, including three RGB channels and one temperature channel. The feature dimensions of the temperature channel and the RGB channel are unified by 1×1 convolution, and then fused by residual connection to retain the temperature gradient information.

[0138] A new temperature attention mechanism is added to weight the fused features, highlighting the feature weights of high-temperature anomaly regions and enhancing the distinction between fault and normal regions.

[0139] (2) Feature extraction and enhancement

[0140] The backbone network follows CSPDarknet (cross-stage partial network - darknet), and the parameters are reduced by using lightweight depthwise separable convolutions to avoid overfitting on small samples;

[0141] Combined with SPP-FPN (Spatial Pyramid Pooling-Feature Pyramid Network) to enhance multi-scale feature extraction capabilities and adapt to fault regions of different sizes;

[0142] The PAN (Path Aggregation Network) structure is improved by adding a cross-layer temperature feature interaction module. High-resolution detail features and low-resolution temperature features are fused through skip connections to improve the localization accuracy of small targets.

[0143] (3) Output layer extension

[0144] An improved decoupling head is adopted, which adds temperature feature auxiliary output on the basis of the original bounding box + category output, so as to meet the needs of fault location, type identification and temperature characteristic analysis.

[0145] Improve the input-output logic of the YOLO10 model

[0146] Input: 4-channel RGB-T image, where RGB provides physical structure information and the temperature channel provides abnormal thermal information, realizing visual + temperature dual-dimensional fault feature input;

[0147] Output: For each fault region, output bounding box coordinates, area, fault type, and temperature characteristics, forming a complete output chain from detection to analysis;

[0148] Improved training logic of YOLO10 model

[0149] Using historical RGB-T images as training data, three types of labels (boundary box coordinates, fault type one-heat code, and measured temperature features) are matched. The model output is constrained to be consistent with the real temperature features by the temperature feature loss function, so as to ensure the accuracy of temperature-related predictions.

[0150] Ultimately, this enables the model to simultaneously learn visual structure and temperature anomaly patterns from RGB-T images, achieving accurate detection and feature extraction of cable faults.

[0151] S4. Input the RGB-T image of the cable under test for the current time period into the model, and output the labeled content;

[0152] Based on the above embodiments, the specific steps of S4 are as follows:

[0153] Infrared and visible light images of the cable under test are acquired during the current time period. The coordinates are aligned with the corners of the cable joints, and the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate a 4-channel RGB-T image. The temperature channels are standardized to match the temperature range during training to ensure that the input feature distribution is consistent with the model during training.

[0154] The preprocessed RGB-T image is input into the trained improved YOLO10 model. The model outputs results through a hierarchical logic of boundary localization → type recognition → temperature feature extraction.

[0155] Output: Outputs the coordinates and area of ​​the bounding boxes of all detected fault areas;

[0156] Secondary output: For each region within a bounding box, output: fault type, temperature gradient distribution characteristics, and temperature threshold;

[0157] The "boundary information" and "type and temperature characteristics" of the same fault area are correlated and matched to form a structured annotation result, which includes the bounding box coordinates, area, fault type, temperature gradient distribution characteristics, and temperature threshold.

[0158] S5. Based on the fault area boundary box, RGB-T image and fault type of the cable under test in the current time period, determine the type and number of fault points within the boundary box. Combine the boundary box coordinates, temperature gradient distribution characteristics, area and temperature threshold of each fault type, screen the highest risk fault point, use the fault area boundary box as the initial boundary box, determine the boundary box type according to the positional relationship between the highest risk fault point and the initial boundary box, and redraw the fault area boundary box with the highest risk fault point as the center.

[0159] Based on the above embodiments, the fault area boundary box, RGB-T image, and fault type of the cable under test in the current time period are used to determine the type and number of fault points within the boundary box. The specific steps are as follows:

[0160] Retrieve the coordinates of the fault area bounding box, the corresponding RGB-T image, and the labeled fault type of the cable under test for the current time period output by S4;

[0161] Locate the bounding box region in the RGB-T image, and extract the spatial texture features of the RGB channels and the temperature distribution features of the T channel within the region;

[0162] By combining the spatial texture features and temperature gradient distribution features corresponding to each fault type in the secondary annotation, the region within the bounding box is analyzed.

[0163] If there is only one region within the bounding box that matches the characteristics of a certain fault type, then that region is determined to be a fault point of the corresponding type. If there are multiple suspected regions, their spatial texture features and temperature distribution features are compared according to the sub-regions to determine whether they are multiple fault points of the same type or fault points of different types.

[0164] Record the location coordinates of each fault point and the corresponding fault type, including a single fault point, multiple fault points of the same type, or multiple fault points of different types.

[0165] Based on the above embodiments, and combining the bounding box coordinates, temperature gradient distribution characteristics of each fault type, area, and temperature threshold, the highest-risk fault point is screened. The specific steps are as follows:

[0166] For each fault point, the temperature data of that fault point is extracted from channel T, and the difference between the temperature data and the corresponding fault type temperature threshold is calculated using the following formula:

[0167] ;

[0168] in, For the first The temperature difference between the fault points is the largest possible value. The greater the temperature difference, the more severe the temperature exceedance, indicating that the temperature at the fault point exceeds the safety threshold to a greater extent.

[0169] In the formula, The first one extracted from channel T The temperature at each fault point For the first Temperature thresholds for each fault point corresponding to its fault type For the index of the fault point, , The number of fault points;

[0170] Based on the temperature gradient distribution characteristics corresponding to different fault types, a diffusion risk coefficient is set:

[0171] When the fault type is joint overheating, the temperature gradient distribution is characterized by a steep gradient (i.e., a large temperature difference between the fault point and the surrounding area, and drastic temperature changes). This characteristic means that heat is concentrated and transferred quickly, and the high-temperature area can easily diffuse rapidly to the surrounding insulation layer, potentially causing insulation layer breakdown in a short period of time. Therefore, its diffusion risk is higher, and a higher risk weighting coefficient is needed to amplify the risk.

[0172] When the fault type is insulation aging, the temperature gradient distribution is characterized by a gentle gradient (the temperature difference between the fault area and the surrounding area is small, and the temperature changes slowly). This characteristic indicates that the heat distribution is relatively uniform, the diffusion rate is slow, and the urgency of risk diffusion in the short term is low. Therefore, the coefficient is set to a low value.

[0173] Furthermore, overheating of the joint is often a sudden failure. Once the temperature exceeds the threshold, it can easily lead to serious accidents such as electric arcs and short circuits. The risk is sudden and highly destructive, and its potential for spread needs to be closely monitored, hence the coefficient is even higher.

[0174] Insulation aging is a progressive fault. After the temperature exceeds the threshold, it is more likely to manifest as a gradual decline in insulation performance. It will not directly lead to breakdown in the short term. The risk is slow and predictable, so the coefficient is relatively low.

[0175] Therefore, if the fault type is joint overheating and the temperature gradient distribution is a steep gradient, then the diffusion risk coefficient is assigned a value of [value missing]. If the fault type is insulation aging and the temperature gradient distribution is a gentle gradient, then the diffusion risk coefficient is assigned a value of [value missing]. ;

[0176] in, ,and All are constants greater than 0;

[0177] The ratio of the area of ​​the sub-region where the fault point is located to the area of ​​the bounding box is calculated to generate the area proportion, based on the following formula:

[0178] ;

[0179] in, For the first The larger the area percentage of each fault point, the wider the impact range, the wider the overall area of ​​insulation layer damage, and the higher the risk level of the fault point.

[0180] In the formula, For the first The area of ​​the sub-region where each fault point is located. For the first The area of ​​the bounding box of the fault region where each fault point is located;

[0181] Table 1. Changes in the overall risk coefficient of the fault point with normalized temperature difference, diffusion risk coefficient, and area proportion.

[0182]

[0183] As shown in Table 1, the normalized temperature difference gradually increases from 0 to 0.96, and this continuous increase provides the fundamental impetus for the rise in the comprehensive risk coefficient. With the other two parameters changing simultaneously, each increase in the temperature difference leads to a stepwise increase in the comprehensive risk coefficient, demonstrating the direct impact of temperature differences on failure risk.

[0184] The diffusion risk coefficient starts at 0.50 and increases continuously to 0.98, maintaining a monotonically increasing trend. As a key indicator for measuring the probability of fault diffusion, its increasing characteristic amplifies the overall risk coefficient beyond just temperature difference and area proportion, especially in the higher value range, where its contribution to the overall risk coefficient is more significant.

[0185] Area percentage: Starting from 0.10, increasing to 0.82, reflecting the gradual expansion of the fault's impact range. As the area percentage increases, the area affected by the fault expands, and together with temperature difference and diffusion risk coefficient, this leads to a continuous increase in the overall risk coefficient.

[0186] according to Figures 3-5 It can be seen that the comprehensive risk coefficient is linearly positively correlated with the normalized temperature difference, diffusion risk coefficient, and area proportion.

[0187] Normalized temperature difference: The fitted line is a straight line. The comprehensive risk coefficient increases steadily with the increase of temperature difference, indicating that the driving effect of temperature difference on risk is linear and continuous. The larger the temperature difference, the more obvious the risk increase.

[0188] Diffusion risk coefficient: The fitted line is linear, and the risk coefficient increases synchronously with the diffusion coefficient, reflecting the direct linear amplification of risk by diffusion. The higher the diffusion coefficient, the faster the risk accumulates.

[0189] Area proportion: The fitted line is a straight line, and the risk coefficient increases linearly with the increase of the area proportion, reflecting the linear relationship between the scope of the fault's impact and the degree of risk. The larger the proportion, the wider the risk coverage.

[0190] The first Temperature difference at each fault point After normalization, based on the normalized temperature difference, diffusion risk coefficient, and area proportion, the comprehensive risk coefficient of each fault point is calculated using the following formula:

[0191] ;

[0192] in, For the first The comprehensive risk coefficient of each fault point is used to assess the risk level of the fault point by combining three indicators: normalized temperature difference, diffusion risk coefficient, and area proportion. The larger the comprehensive risk coefficient, the higher the risk level of the fault point.

[0193] In this process, multiple temperature data sets need to be collected for each fault point, and the temperature differences between these multiple sets need to be calculated. Using min-max normalization Mapped to the [0,1] interval;

[0194] If the fault type is joint overheating, the first Risk coefficient of diffusion at each fault point If the fault type is insulation aging, the first Risk coefficient of diffusion at each fault point ;

[0195] Based on the above, it should be noted that:

[0196] Temperature difference refers to the difference between the temperature at the fault point and the temperature threshold corresponding to that fault type. When the temperature does not exceed the threshold, This indicates no risk of temperature exceeding the limit; when the threshold is exceeded, the larger the temperature difference, the more severe the temperature exceedance. An increase indicates that the higher the temperature at the fault point exceeds the safety threshold, the greater the direct risk of insulation material aging or breaking down due to high temperature. Therefore, the risk level of the fault point is higher, ultimately leading to a higher overall risk coefficient. The larger.

[0197] The diffusion risk coefficient is a coefficient set based on the temperature gradient distribution characteristics of the fault type. It is used to quantify the diffusion rate and urgency of the fault. An increase in the risk level indicates that the heat from the fault spreads faster and the fault expands more rapidly. This means the fault may worsen in a short period and threaten a larger area of ​​the insulation layer, thus increasing the risk level at the fault point and ultimately leading to a higher overall risk coefficient. The larger.

[0198] Area percentage refers to the ratio of the area of ​​the sub-region where the fault point is located to the area of ​​the boundary box of the fault region to which the fault point belongs. An increase in area percentage indicates that the fault point occupies a larger space in the fault region, has a wider impact range, and a wider overall area of ​​insulation damage. Therefore, the risk level of the fault point is higher, ultimately leading to a higher comprehensive risk coefficient. The larger.

[0199] Therefore, the overall risk coefficient is positively correlated with the temperature difference at the fault point, the diffusion risk coefficient, and the area ratio.

[0200] The three indicators (temperature difference, diffusion risk coefficient, and area ratio) do not have independent impacts on risk, but rather exhibit a synergistic effect that is cumulative: a large temperature difference ( High) fault points, if they also have fast diffusion characteristics ( High) and large range of influence ( The actual risk of the three indicators (high) is much higher than the simple sum of the individual effects of the three indicators. The multiplicative form can reflect this "risk superposition and amplification" characteristic. The increase of any one indicator will significantly increase the overall risk. However, if the temperature difference indicator is 0, the overall risk is 0, which is consistent with the logic of "no direct risk if the temperature exceeds the standard".

[0201] In summary, the above-mentioned functional form is used to express the functional relationship between the risk coefficient, the temperature difference at the fault point, the diffusion risk coefficient, and the area ratio.

[0202] Based on the above embodiments, the comprehensive risk coefficients of all fault points are sorted in descending order, and the fault point corresponding to the first comprehensive risk coefficient is the highest risk fault point.

[0203] When there are at least two comprehensive risk factors that rank first;

[0204] Compare the temperature differences between the fault points and select the fault point with the largest temperature difference as the highest risk fault point.

[0205] If the temperature difference is still the same, then further compare their diffusion risk coefficients and select the fault point with the largest diffusion risk coefficient as the highest risk fault point.

[0206] If both of the above are the same, compare the area proportion of the fault point and select the fault point with the largest area proportion as the highest risk fault point.

[0207] When there is only one fault point within the bounding box, there is no need to sort; the fault point is directly identified as the highest-risk fault point.

[0208] Based on the above embodiments, the initial bounding box is the fault region detected by the model through the RGB-T image, but the following problems may exist:

[0209] The scope is too large: it includes non-core risk areas, causing noise areas to interfere with the results in subsequent analysis;

[0210] Range deviation: Due to detection errors or fault propagation, the highest risk point may be partially or completely outside the initial bounding box. If the initial box is still used as the analysis range, the real high-risk area will be missed.

[0211] By redefining the boundary box according to the location relationship of the highest risk points, irrelevant areas can be eliminated, range deviations can be corrected, and the analysis object can be focused on the core area where the risk is most concentrated.

[0212] Based on the above embodiments, the fault area boundary box is used as the initial boundary box. The boundary box type is determined according to the positional relationship between the highest-risk fault point and the initial boundary box. The fault area boundary box is then redefined with the highest-risk fault point as the center. The specific steps are as follows:

[0213] If the highest risk fault point is completely inside the initial bounding box, that is, all parts of the fault point are within the initial bounding box, then the bounding box type is internal.

[0214] If the highest-risk failure point is partially inside the initial bounding box and partially outside the initial bounding box, then the bounding box type is partially extended.

[0215] If the highest risk fault point is completely outside the initial bounding box, that is, all parts of the fault point are not within the initial bounding box, then the bounding box type is external.

[0216] Centered on the highest-risk fault point, the fault area boundary box is redefined based on the determined boundary box type:

[0217] For internal fault types, the newly defined fault area boundary box is limited to the initial boundary box, centered on the highest risk fault point, covering the fault point and the affected area around it;

[0218] For some extended types, the newly defined fault area bounding box must include both the portion of the highest-risk fault point within the initial bounding box and the portion beyond it;

[0219] For external fault types, the newly defined fault area boundary box is centered on the highest-risk fault point and is set according to the influence range of the fault point, without being limited by the initial boundary box.

[0220] By dynamically adjusting the scope of analysis, risk can be focused and the scope can be precise.

[0221] S6. Perform finite element model analysis according to the scope of the redefined fault area boundary box, simulate the temperature field diffusion process in the next 12 hours in combination with environmental parameters, dynamically output the temperature field distribution at different times, obtain the continuous expansion rate of the region at two adjacent times, calculate the proportion of groups exceeding the preset threshold, and determine the insulation breakdown risk level of the cable under test.

[0222] Based on the above embodiments, environmental parameters include temperature, wind speed, and relative humidity; these three parameters are key external factors affecting the diffusion of the cable temperature field, and their specific effects are as follows:

[0223] Ambient temperature directly affects the direction and rate of heat exchange between the cable and the outside world. For example, when the ambient temperature is lower than the temperature of the cable fault area, heat will dissipate from the cable to the environment, slowing down the expansion of the temperature field; conversely, a high-temperature environment will inhibit heat dissipation and accelerate the temperature diffusion of the fault area.

[0224] Wind speed affects the rate of temperature field diffusion through convective heat dissipation. The higher the wind speed, the faster the airflow, and the more efficiently the heat on the cable surface is carried away, which can significantly reduce the rate of temperature field expansion. Low wind speed or no wind environment will weaken convective heat dissipation, causing heat to accumulate around the fault area and accelerate its diffusion.

[0225] Relative humidity primarily affects the thermal conductivity and surface heat dissipation efficiency of insulation materials. In high humidity environments, condensation may form on the cable surface, increasing thermal conductivity. At the same time, the convective heat dissipation capacity of humid air decreases, which may promote the expansion of the temperature field. In low humidity environments, heat is dissipated through air convection, slowing down diffusion.

[0226] The methods for collecting temperature, wind speed, and relative humidity are as follows:

[0227] Using the newly defined fault area boundary as the core, collection points are reasonably deployed around it to ensure coverage of the fault area and the surrounding environment that may be affected by its temperature diffusion.

[0228] A contact temperature sensor is used to directly contact the cable surface or the ambient medium near the fault area to collect the ambient temperature in real time.

[0229] Select a wind speed sensor and install it in a representative location with good ventilation conditions around the fault area, such as the air outlet of a cable trench or an open area under an overhead cable, to ensure that the height of the sensor is consistent with the height of the airflow in the fault area.

[0230] A humidity sensor is used and installed in the same location as the temperature sensor to synchronously collect the relative humidity of the air around the fault area.

[0231] Based on the above embodiments, finite element analysis is performed according to the range of the redefined fault area boundary box. The temperature field diffusion process over the next 12 hours is simulated in combination with environmental parameters. The temperature field distribution at different times is dynamically output, and the continuous expansion rate of the region between two adjacent times is obtained. The specific steps are as follows:

[0232] The spatial scope of finite element calculation is defined by using the newly defined fault area boundary box as the boundary.

[0233] Based on the bounding box and cable shape, the region is divided into small finite element units as the basic units for temperature field calculation. The units in the fault core region are densified, while the units in the far regions can be appropriately enlarged.

[0234] Assigning corresponding thermophysical properties to different units, including:

[0235] Thermal conductivity, specific heat capacity, and density of each layer of cable material; thermal conductivity and convective heat transfer coefficient of the surrounding medium;

[0236] The initial temperature field is set such that the initial temperature of the unit where the highest risk fault point is located is set to the current measured temperature; the initial temperature of other units is referenced from historical data: the normal cable area is set to the ambient temperature, and the area around the fault area is set to a gradient decrease.

[0237] Temperature, wind speed, and relative humidity are transformed into boundary conditions for a finite element model, quantifying the influence of the external environment on the temperature field.

[0238] Ambient temperature is used as the far-field boundary condition of the model, meaning that the temperature of the elements far from the fault region is always maintained at the ambient temperature, thus defining the benchmark for heat exchange.

[0239] The convective heat transfer coefficient is calculated using a formula based on wind speed. The higher the wind speed, the higher the convective heat transfer coefficient, the faster the heat dissipates from the fault area to the air, and the slower the temperature field expands.

[0240] Relative humidity corrects the thermal conductivity and convective heat transfer coefficient of insulating materials, quantifying the influence of external factors on heat diffusion;

[0241] Based on Fourier's law of heat conduction, a three-dimensional unsteady-state heat conduction equation was established. The equation was discretized into each element using the finite element software (ANSYS), and solved using the time-stepping method.

[0242] Set the time step (1 hour), and calculate the temperature change of all units in each step;

[0243] Combined with boundary conditions, the temperature field distribution at each time step is solved iteratively until convergence (temperature difference between adjacent iterations ≤ 0.1℃).

[0244] After each time step of calculation is completed, the temperature values ​​of all cells at that moment are saved, forming temperature field distribution data at 12 time steps, denoted as . ,in, For the first The temperature at that moment For indexing time;

[0245] Set the preset temperature threshold For each moment's temperature data, the elements in the finite element set whose temperature exceeds a preset threshold are identified one by one and designated as the threshold-exceeding region. This refers to the critical temperature for heat resistance of insulating materials.

[0246] The number of finite element elements in the region exceeding the threshold at each time step is counted, and the region expansion rate between two adjacent time steps is calculated using the following formula:

[0247] ;

[0248] in, For the first The moment to the The larger the area expansion rate at a given time, the more rapid the expansion of the over-threshold area per unit time, the faster the fault deteriorates, and the higher the urgency of insulation breakdown.

[0249] In the formula, For the first The number of finite element elements in the region exceeding the threshold at each time point. For the first The number of finite element elements in the region exceeding the threshold at each time point. , This represents the time interval between two adjacent moments.

[0250] Based on the above embodiments, the percentage of groups exceeding a preset threshold is calculated to determine the insulation breakdown risk level of the cable under test. The specific steps are as follows:

[0251] From the The moment to the There are 11 time points, with 11 adjacent time points. The statistics are as follows: The number of groups is denoted as , A preset threshold for the rate of continued regional expansion;

[0252] Among them, the preset threshold for the rate of continued regional expansion. The setup method is as follows:

[0253] Using the expansion rate data of the 12 hours before insulation breakdown as a sample, we screened out the critical expansion rate values ​​that are strongly correlated with the final breakdown, collected historical cases of breakdown of similar cables, and extracted the expansion rate data of all adjacent moments within the 12 hours. The minimum expansion rate that occurs before breakdown is denoted as... Collect early warning cases that did not break down but were processed, extract their expansion rate data, and calculate the maximum expansion rate that can be stabilized without further processing, denoted as... ; Pick and The mean;

[0254] The percentage of groups exceeding a preset threshold across all adjacent time points is calculated using the following formula:

[0255] ;

[0256] in, The percentage of groups exceeding a preset threshold across all adjacent time points;

[0257] like If the insulation of the cable under test is low, then the risk of insulation breakdown is low.

[0258] like If the insulation breakdown risk of the cable under test is moderate, then the cable under test is of medium risk.

[0259] like If the insulation of the cable under test is damaged, the risk of insulation breakdown is high.

[0260] in, To classify the critical values ​​for low and medium risk of insulation breakdown, The critical values ​​are used to classify the risk of insulation breakdown as medium or high.

[0261] in, , The specific setup method is as follows:

[0262] Collect temperature field diffusion data for the 12 hours preceding insulation breakdown in similar cables throughout history, and extract the regional sustained expansion rate at adjacent time points. Statistics show that "exceeding the preset threshold" percentage of groups The correspondence between "and whether it is ultimately penetrated";

[0263] Analysis of 100 insulation breakdown cases revealed that in 90% of the cases, within 12 hours prior to breakdown... Analysis of 100 cases where penetration was not achieved but warnings were still required revealed that... At that time, breakdown could be avoided through proper handling; analysis of 100 normal operating cases revealed... At this time, the temperature field expands stably, and there is no risk of breakdown;

[0264] Take the largest value in the normal case "Value" and "Minimum in Medium-Risk Cases" The mean of the "values", as in the case above. (The threshold between normal and requiring early warning); Take the largest of the medium-risk cases "Value" and "Minimum in the breakdown case" The mean of the "values", as in the case above. (The threshold between warning and high risk).

[0265] Please see Figure 2 The present invention also provides a technical solution:

[0266] A cable fault infrared image target detection system based on an improved YOLO10, the system being used to execute any of the above-described cable fault infrared image target detection methods based on an improved YOLO10, comprising:

[0267] The image acquisition module is used to acquire infrared and visible light images and fault records of the cable under test over a historical period. After aligning the coordinates of the two types of images, the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate an RGB-T image.

[0268] The annotation module is used to perform layered annotation of RGB-T images based on fault records. It first annotates the coordinates and area of ​​the fault region bounding box, then annotates the fault type within the region, and supplements the temperature gradient distribution features and temperature thresholds for each type.

[0269] The model building module is used to build an improved YOLO10 model. It takes the RGB-T images of the cable under test in various historical time periods as input and the labeled content as output labels to train the model.

[0270] The simulation module is used to input the RGB-T image of the cable under test for the current time period into the model and output the labeled content;

[0271] The filtering module is used to determine the type and number of fault points within the boundary box based on the fault area boundary box, RGB-T image and fault type of the cable under test in the current time period. It combines the boundary box coordinates, temperature gradient distribution characteristics of each fault type, area and temperature threshold to filter the highest risk fault point. The fault area boundary box is used as the initial boundary box. The boundary box type is determined according to the positional relationship between the highest risk fault point and the initial boundary box. The fault area boundary box is redefined with the highest risk fault point as the center.

[0272] The analysis module is used to perform finite element model analysis according to the range of the redefined fault area boundary box, simulate the temperature field diffusion process in the next 12 hours in combination with environmental parameters, dynamically output the temperature field distribution at different times, obtain the continuous expansion rate of the region between two adjacent times, calculate the proportion of groups exceeding the preset threshold, and determine the insulation breakdown risk level of the cable under test.

[0273] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0274] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by software, electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0275] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0276] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for detecting cable faults using infrared images based on an improved YOLOv10, characterized in that, The specific steps include: S1. Obtain infrared and visible light images and fault records of the cable under test over a historical period. After aligning the coordinates of the two types of images, extract the infrared pixel temperature values ​​and embed them into the RGB three channels to generate an RGB-T image. S2. Based on the fault record, the RGB-T image is labeled in layers. The coordinates and area of ​​the fault region bounding box are labeled once, and the fault type in the region is labeled twice. The temperature gradient distribution characteristics and temperature thresholds of each type are supplemented. S3. Construct an improved YOLO10 model, using RGB-T images of the cable under test at various historical time periods as input and the labeled content as output labels, and train the model; S4. Input the RGB-T image of the cable under test for the current time period into the model, and output the labeled content; S5. Based on the fault area boundary box, RGB-T image and fault type of the cable under test in the current time period, determine the type and number of fault points within the boundary box. Combine the boundary box coordinates, temperature gradient distribution characteristics, area and temperature threshold of each fault type, screen the highest risk fault point, use the fault area boundary box as the initial boundary box, determine the boundary box type according to the positional relationship between the highest risk fault point and the initial boundary box, and redraw the fault area boundary box with the highest risk fault point as the center. S6. Perform finite element model analysis according to the scope of the redefined fault area boundary box, simulate the temperature field diffusion process in the next 12 hours in combination with environmental parameters, dynamically output the temperature field distribution at different times, obtain the continuous expansion rate of the region at two adjacent times, calculate the proportion of groups exceeding the preset threshold, and determine the insulation breakdown risk level of the cable under test.

2. The cable fault infrared image target detection method based on improved YOLO10 according to claim 1, characterized in that, Environmental parameters include temperature, wind speed, and relative humidity.

3. The cable fault infrared image target detection method based on improved YOLO10 according to claim 1, characterized in that, After aligning the coordinates of the two types of images, the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate an RGB-T image. The specific steps are as follows: From infrared and visible light images, the corner of the cable joint is selected as a common feature point, and the pixel coordinates of the feature point in the two types of images are recorded respectively. The feature points are calculated using a feature point matching algorithm to obtain the coordinate deviation between the two types of images, and a coordinate transformation matrix is ​​established based on this coordinate deviation. Geometric transformations are performed on the infrared image based on the coordinate transformation matrix, including correcting positional deviations by translation, unifying angles by rotation, and matching sizes by scaling, so that the pixel coordinates at the same position in the infrared image and the visible light image completely coincide, thereby completing coordinate alignment; Based on the coordinate-aligned infrared image, the infrared pixel temperature corresponding to each pixel is read; For noisy pixels in the image, a neighborhood mean filtering method is used to replace the outlier value with the average temperature of the normal pixels around the pixel. For pixels with missing temperature data, the temperature value is supplemented by interpolation, so as to obtain continuous and complete infrared pixel temperature. Establish a mapping table between pixel position and processed infrared pixel temperature, and clarify the temperature information of each coordinate point; The R, G, and B channel pixel values ​​of the visible light image are retained and used as the basic information of the RGB-T image; Based on the temperature information of the determined coordinate points, the temperature of continuous and complete infrared pixels is standardized and converted into a numerical form compatible with RGB channels through grayscale mapping. A four-channel fusion method is adopted, in which the processed temperature value is used as the T channel and combined with the RGB three-channel data to form an RGB-T image containing spatial texture and temperature information.

4. The cable fault infrared image target detection method based on improved YOLO10 according to claim 1, characterized in that, The specific steps of S2 are as follows: Extract the fault time, location, type, and temperature characteristics of the cable under test from historical fault records; Based on the time and location of the fault, the corresponding image is matched in the historical RGB-T image, and the suspected fault area is identified by its spatial texture features and temperature features. Based on the suspected fault area, a labeling is made. Referring to the cable structure characteristics of the RGB channel and the temperature anomaly range of the T channel, the boundary box of the fault area is defined and the coordinates are recorded. At the same time, the area is calculated and labeled. Secondary labeling is performed based on the recorded fault type. The specific fault type is confirmed by combining the spatial texture features and temperature features of the suspected fault area. The temperature gradient distribution features of this type are extracted within the bounding box, and the temperature threshold range is determined and labeled.

5. The cable fault infrared image target detection method based on improved YOLO10 according to claim 4, characterized in that, Based on the fault area bounding box, RGB-T image, and fault type of the cable under test for the current time period, determine the type and number of fault points within the bounding box. The specific steps are as follows: Retrieve the coordinates of the fault area bounding box, the corresponding RGB-T image, and the labeled fault type of the cable under test for the current time period output by S4; Locate the bounding box region in the RGB-T image, and extract the spatial texture features of the RGB channels and the temperature distribution features of the T channel within the region; By combining the spatial texture features and temperature gradient distribution features corresponding to each fault type in the secondary annotation, the region within the bounding box is analyzed. If there is only one region within the bounding box that matches the characteristics of a certain fault type, then that region is determined to be a fault point of the corresponding type. If there are multiple suspected regions, their spatial texture features and temperature distribution features are compared according to the sub-regions to determine whether they are multiple fault points of the same type or fault points of different types. Record the location coordinates of each fault point and the corresponding fault type, including a single fault point, multiple fault points of the same type, or multiple fault points of different types.

6. The cable fault infrared image target detection method based on improved YOLO10 according to claim 5, characterized in that, By combining the bounding box coordinates, temperature gradient distribution characteristics of each fault type, area, and temperature threshold, the highest-risk fault point is selected. The specific steps are as follows: For each fault point, the temperature data of that fault point is extracted from channel T, and the difference between the temperature data and the corresponding fault type temperature threshold is calculated using the following formula: ； in, For the first Temperature difference at each fault point The first one extracted from channel T The temperature at each fault point For the first The temperature threshold corresponding to each fault point and fault type For the index of the fault point, , The number of fault points; Based on the temperature gradient distribution characteristics corresponding to different fault types, a diffusion risk coefficient is set: If the fault type is joint overheating and the temperature gradient distribution is a steep gradient, then the diffusion risk coefficient is assigned a value of [value missing]. ; If the fault type is insulation aging and the temperature gradient distribution is a gentle gradient, then the diffusion risk coefficient is assigned a value of [value missing]. ; in, ,and All are constants greater than 0; The ratio of the area of ​​the sub-region where the fault point is located to the area of ​​the bounding box is calculated to generate the area proportion, based on the following formula: ； in, For the first The area percentage of each fault point For the first The area of ​​the sub-region where each fault point is located. For the first The area of ​​the bounding box of the fault region where each fault point is located; The first Temperature difference at each fault point After normalization, based on the normalized temperature difference, diffusion risk coefficient, and area proportion, the comprehensive risk coefficient of each fault point is calculated using the following formula: ； in, For the first The overall risk coefficient of each failure point For the first The risk coefficient of the spread of each fault point; If the fault type is joint overheating, the first Risk coefficient of diffusion at each fault point If the fault type is insulation aging, the first Risk coefficient of diffusion at each fault point ; The comprehensive risk coefficients of all failure points are sorted in descending order, and the failure point corresponding to the comprehensive risk coefficient ranked first is the highest risk failure point. When there are at least two comprehensive risk factors that rank first; Compare the temperature differences between the fault points and select the fault point with the largest temperature difference as the highest risk fault point. If the temperature difference is still the same, then further compare their diffusion risk coefficients and select the fault point with the largest diffusion risk coefficient as the highest risk fault point. If both of the above are the same, compare the area proportion of the fault point and select the fault point with the largest area proportion as the highest risk fault point. When there is only one fault point within the bounding box, there is no need to sort; the fault point is directly identified as the highest-risk fault point.

7. The cable fault infrared image target detection method based on improved YOLO10 according to claim 6, characterized in that, Using the fault region boundary box as the initial boundary box, the boundary box type is determined based on the positional relationship between the highest-risk fault point and the initial boundary box. The fault region boundary box is then redefined with the highest-risk fault point as the center. The specific steps are as follows: If the highest risk fault point is completely inside the initial bounding box, that is, all parts of the fault point are within the initial bounding box, then the bounding box type is internal. If the highest-risk failure point is partially inside the initial bounding box and partially outside the initial bounding box, then the bounding box type is partially extended. If the highest risk fault point is completely outside the initial bounding box, that is, all parts of the fault point are not within the initial bounding box, then the bounding box type is external. Centered on the highest-risk fault point, the fault area boundary box is redefined based on the determined boundary box type: For internal fault types, the newly defined fault area boundary box is limited to the initial boundary box, centered on the highest risk fault point, covering the fault point and the affected area around it; For some extended types, the newly defined fault area bounding box must include both the portion of the highest-risk fault point within the initial bounding box and the portion beyond it; For external fault types, the newly defined fault area boundary box is centered on the highest-risk fault point and is set according to the influence range of the fault point, without being limited by the initial boundary box.

8. The cable fault infrared image target detection method based on improved YOLO10 according to claim 7, characterized in that, Finite element model analysis was performed based on the newly defined boundary of the fault area. Environmental parameters were used to simulate the temperature field diffusion process over the next 12 hours, dynamically outputting the temperature field distribution at different times and obtaining the regional expansion rate between adjacent time points. The specific steps are as follows: The spatial scope of finite element calculation is defined by using the newly defined fault area boundary box as the boundary. Based on the bounding box and cable configuration, the region is divided into small finite element units, which serve as the basic units for temperature field calculation. Temperature, wind speed, and relative humidity are converted into boundary conditions of the finite element model. The current temperature of the highest risk fault point is set as the initial temperature and assigned to its element. The initial temperatures of other elements are set with reference to historical data. Based on Fourier's law of heat conduction, a three-dimensional unsteady-state heat conduction equation is established. The equation is discretized to each element using finite element software, and solved using the time-stepping method. Set the time step and calculate the temperature change of all units in each step; By combining boundary conditions, the temperature field distribution at each time step is iteratively solved until convergence. After each time step, the temperature values ​​of all cells at that time step are saved, forming temperature field distribution data at 12 time steps, denoted as . ,in, For the first The temperature at that moment For indexing time; Set the preset temperature threshold For the temperature data at each moment, the elements in the finite element unit whose temperature exceeds the preset threshold are identified one by one and identified as the over-threshold region; The number of finite element elements in the region exceeding the threshold at each time step is counted, and the region expansion rate between two adjacent time steps is calculated using the following formula: ； in, For the first The moment to the The rate of sustained regional expansion at any given time. For the first The number of finite element elements in the region exceeding the threshold at each time point. For the first The number of finite element elements in the region exceeding the threshold at each time point. , This represents the time interval between two adjacent moments.

9. The cable fault infrared image target detection method based on improved YOLO10 according to claim 8, characterized in that, The percentage of groups exceeding a preset threshold is calculated to determine the insulation breakdown risk level of the cable under test. The specific steps are as follows: From the The moment to the There are 11 time points, with 11 adjacent time points. The statistics are as follows: The number of groups is denoted as , A preset threshold for the rate of continued regional expansion; The percentage of groups exceeding a preset threshold across all adjacent time points is calculated using the following formula: ； in, The percentage of groups exceeding a preset threshold across all adjacent time points; like If the insulation of the cable under test is low, then the risk of insulation breakdown is low. like If the insulation breakdown risk of the cable under test is moderate, then the cable under test is of medium risk. like If the insulation of the cable under test is damaged, the risk of insulation breakdown is high. in, To classify the critical values ​​for low and medium risk of insulation breakdown, The critical values ​​are used to classify the risk of insulation breakdown as medium or high.

10. A cable fault infrared image target detection system based on improved YOLO10, the system being used to execute the cable fault infrared image target detection method based on improved YOLO10 as described in any one of claims 1-9, characterized in that, include: The image acquisition module is used to acquire infrared and visible light images and fault records of the cable under test over a historical period. After aligning the coordinates of the two types of images, the infrared pixel temperature values ​​are extracted and embedded into the RGB three channels to generate an RGB-T image. The annotation module is used to perform layered annotation of RGB-T images based on fault records. It first annotates the coordinates and area of ​​the fault region bounding box, then annotates the fault type within the region, and supplements the temperature gradient distribution features and temperature thresholds for each type. The model building module is used to build an improved YOLO10 model. It takes the RGB-T images of the cable under test in various historical time periods as input and the labeled content as output labels to train the model. The simulation module is used to input the RGB-T image of the cable under test for the current time period into the model and output the labeled content; The filtering module is used to determine the type and number of fault points within the boundary box based on the fault area boundary box, RGB-T image and fault type of the cable under test in the current time period. It combines the boundary box coordinates, temperature gradient distribution characteristics of each fault type, area and temperature threshold to filter the highest risk fault point. The fault area boundary box is used as the initial boundary box. The boundary box type is determined according to the positional relationship between the highest risk fault point and the initial boundary box. The fault area boundary box is redefined with the highest risk fault point as the center. The analysis module is used to perform finite element model analysis according to the range of the redefined fault area boundary box, simulate the temperature field diffusion process in the next 12 hours in combination with environmental parameters, dynamically output the temperature field distribution at different times, obtain the continuous expansion rate of the region between two adjacent times, calculate the proportion of groups exceeding the preset threshold, and determine the insulation breakdown risk level of the cable under test.

Images