Fault detection method, device and equipment for transformer bushing and storage medium

Abstract

The invention provides a fault detection method and device for a transformer bushing, equipment and a storage medium. The method comprises the following steps: acquiring an infrared image of a bushing to be detected; inputting the infrared image into a pre-trained casing pipe fault detection model to identify the fault type of the to-be-detected casing pipe; wherein the casing pipe fault detection model comprises an image segmentation module and a fault detection module, the fault detection module is constructed based on a visual Transform module, the image segmentation module is obtained by improving the uncertainty and gradient of pixels in a Point Rend module, the image segmentation module is used for performing contour segmentation processing on an infrared image of a casing pipe to be detected, and the fault detection module is used for detecting the casing pipe to be detected. And the fault detection module is used for detecting the contour image. According to the invention, the detection accuracy of the sleeve can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer detection, and in particular to a transformer bushing fault detection method, device, equipment and storage medium. Background Art

[0002] The transformer is one of the most important components in the power system, which plays the role of voltage conversion and power transmission. The transformer bushing is the outlet device that leads the high-voltage wire inside the transformer to the outside of the oil tank. It plays the role of fixing the lead wire and insulating it from the outside. There may be more than ten bushings on a transformer, and the failure caused by the bushing is one of the most common failures of the transformer. These failures will directly affect the safety of the power grid and even cause chain accidents, eventually causing explosions, causing serious harm to equipment and personnel.

[0003] At present, there are three main methods for bushing detection. The first method is to study the pulse characteristics and spectrum characteristics of partial discharge based on UHF detection means and bushing defect models. The second method is to analyze the reasons for excessive dielectric loss through oil chromatography test combined with on-site frequency domain dielectric spectrum test. The third method is to diagnose the status of bushing infrared images through neural networks.

[0004] However, the first method requires a preventive test to be carried out when the transformer is out of service, which lacks reliability and timeliness. The second method is not easy to locate the fault. The inventors of the third method found that it is difficult to clearly segment the bushing boundary for infrared images of bushings with complex backgrounds, resulting in the inability to accurately detect. Therefore, a method for accurately detecting transformer bushings is urgently needed. Summary of the invention

[0005] The embodiments of the present invention provide a transformer bushing fault detection method, device, equipment and storage medium to solve the problem of inaccurate detection in current detection methods.

[0006] In a first aspect, an embodiment of the present invention provides a transformer bushing fault detection method, comprising:

[0007] Acquire an infrared image of the casing to be inspected;

[0008] Input the infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected;

[0009] Among them, the casing fault detection model includes an image segmentation module and a fault detection module. The fault detection module is built based on the visual Transformer module. The image segmentation module is obtained by improving the uncertainty and gradient of pixels in the PointRend module. The image segmentation module is used to perform contour segmentation on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected. The fault detection module is used to detect the contour image.

[0010] In a possible implementation, the processing process of the image segmentation module is:

[0011] Performing an initial prediction on the infrared image and generating a first prediction point set; wherein the first prediction point set includes a plurality of pixel points;

[0012] Determine a gradient map of the infrared image based on a first pixel gradient threshold, a second pixel gradient threshold, and an edge detection algorithm;

[0013] Determine a first coordinate of each pixel point in the first prediction point set in the image coordinate space, convert the first coordinate of each pixel point in the first prediction point set into a second coordinate in the feature space, and determine a gradient of each pixel point in the first prediction point set in a gradient map of the infrared image;

[0014] The gradients of all pixel points in the first prediction point set are combined into a first gradient set;

[0015] Multiply the first prediction point set and the negative first gradient set element by element to obtain a first uncertainty set, and select the top N pixels with the largest uncertainty from the first uncertainty set;

[0016] Based on the difference between the predicted category and the true category of the first N pixels with the largest uncertainty, the PointRend module is back-propagated to determine the contour image of the casing to be detected.

[0017] In a possible implementation, determining a gradient map of an infrared image based on a first pixel gradient threshold, a second pixel gradient threshold, and an edge detection algorithm includes:

[0018] Calculate the gradient of all pixels in the infrared image;

[0019] Based on the Canny algorithm, the first pixel gradient threshold and the second pixel gradient threshold, it is determined that the infrared image only contains a gradient map of 0 and 1.

[0020] In a possible implementation, converting the first coordinate of each pixel point in the first prediction point set into the second coordinate in the feature space includes:

[0021] Based on the bilinear interpolation method, the first coordinate of each pixel point in the first prediction point set is converted into a second coordinate in the feature space.

[0022] In one possible implementation, backpropagation is performed on the PointRend module based on the difference between the predicted category and the true category of the top N pixels with the largest uncertainty, including:

[0023] The difference between the predicted category and the true category of all target pixels is determined as the point loss; where the target pixel is any pixel among the first N pixels with the largest uncertainty;

[0024] The point losses of all target pixels are back-propagated to the entire network of the PointRend module and the network parameters are updated.

[0025] In one possible implementation, the fault detection module includes a local attention mechanism and a global attention mechanism.

[0026] In a possible implementation, the fault detection module is a Swin Transformer module, and the Swin Transformer module includes a Patch Partition layer, a Linear Embedding layer, a SwinT Block layer, and a PatchMerging layer;

[0027] Among them, the Patch Partition layer is used for block partitioning and dimensionality reduction operations, the Linear Embedding layer is used for linear transformation, the Patch Merging layer is used for downsampling, and the SwinT Block layer includes a multi-head attention module for moving windows.

[0028] In a second aspect, an embodiment of the present invention provides a transformer bushing fault detection device, comprising:

[0029] An acquisition module, used for acquiring an infrared image of the casing to be inspected;

[0030] A recognition module, used for inputting the infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected;

[0031] Among them, the casing fault detection model includes an image segmentation module and a fault detection module. The fault detection module is built based on the visual Transformer module. The image segmentation module is obtained by improving the uncertainty and gradient of pixels in the PointRend module. The image segmentation module is used to perform contour segmentation on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected. The fault detection module is used to detect the contour image.

[0032] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements the steps of the method described in the first aspect or any possible implementation manner of the first aspect.

[0033] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect or any possible implementation method of the first aspect are implemented.

[0034] The embodiment of the present invention provides a transformer bushing fault detection method, device, equipment and storage medium. In order to detect the bushing to be detected, it is necessary to first obtain an infrared image of the bushing to be detected, and then input the obtained infrared image into a pre-trained bushing fault detection model to identify the fault type of the bushing to be detected. The bushing fault detection model includes an image segmentation module and a fault detection module. The fault detection module is constructed based on a visual Transformer module. The image segmentation module is obtained by improving the uncertainty and gradient of pixels in the PointRend module. The image segmentation module is used to perform contour segmentation processing on the infrared image of the bushing to be detected to obtain a contour image of the bushing to be detected. By improving the uncertainty and gradient of pixels in the PointRend module, the accuracy and clarity of the segmentation boundary of the image segmentation module under complex backgrounds can be further improved, and the boundary of the bushing can be more accurately segmented, which is beneficial to subsequent fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 is a flow chart of an implementation method of a transformer bushing fault detection method provided by an embodiment of the present invention;

[0037] Figure 2 is a block diagram of a casing fault detection model provided by an embodiment of the present invention;

[0038] Figure 3 is a block diagram of another casing fault detection model provided by an embodiment of the present invention;

[0039] Figure 4It is a structural schematic diagram of a transformer bushing fault detection device provided by an embodiment of the present invention;

[0040] Figure 5 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0043] The transformer has a complex structure, with numerous and interrelated internal components. The infrared image of the transformer bushing contains a complex background, which makes it more difficult to extract the contour of the transformer bushing. The inventors found that the current image segmentation technology is only applicable to images with relatively simple backgrounds, and has strong limitations for infrared images with complex backgrounds, which makes the detection accuracy of existing bushing detection low.

[0044] In order to solve the problems in the prior art, the embodiments of the present invention provide a transformer bushing fault detection method, device, equipment and storage medium. The transformer bushing fault detection method provided by the embodiments of the present invention is first introduced below.

[0045] See also Figure 1 , which shows a flowchart of the implementation of the transformer bushing fault detection method provided by an embodiment of the present invention, and is described in detail as follows:

[0046] S110, obtaining an infrared image of the casing to be inspected.

[0047] Infrared images are formed by using infrared remote sensors to receive infrared rays reflected by objects or emitted by themselves.

[0048] An infrared image of the casing to be inspected is obtained by an infrared remote sensor. The infrared image includes not only the casing to be inspected but also the complex background where the casing to be inspected is located.

[0049] S120: Input the infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected.

[0050] like Figure 2 and 3As shown in the figure, the casing fault detection model includes an image segmentation module and a fault detection module. The image segmentation module is obtained by improving the uncertainty and gradient of the pixels in the PointRend module, and is used to perform contour segmentation processing on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected. The fault detection module is built based on the visual Transformer module and is used to detect the contour image.

[0051] Since the infrared image includes not only the casing to be detected but also the complex background where the casing to be detected is located, in order to accurately detect the casing, the infrared image needs to be segmented.

[0052] In addition, most infrared images have low contrast, blurred images, and low signal-to-noise ratio. This is extremely unfavorable for subsequent feature extraction, recognition, or tracking. Therefore, it is necessary to suppress noise, improve image signal-to-noise ratio, adjust infrared image contrast, and enhance infrared image edges and lines.

[0053] For transformers, the structure is complex, and the internal components are numerous and interrelated. The infrared image of the transformer bushing contains a complex background, which also increases the difficulty of extracting the contour of the transformer bushing. Previous image segmentation technology is only applicable to images with relatively simple backgrounds. For infrared images with complex backgrounds, it has strong limitations. Therefore, in order to improve the certainty of detection, the accuracy of the image segmentation module must be improved first.

[0054] In this embodiment, PointRend is a deep learning module that treats segmentation as an image rendering problem, and improves the accuracy and clarity of boundaries by performing more detailed analysis and weighting of features at specific points. Therefore, for objects such as transformer bushings that require precise contours, PointRend can effectively capture subtle boundary details and avoid blurring or over-smoothing problems.

[0055] In order to understand the improvement of the uncertainty and gradient of pixels in PointRend by the present invention, several basic concepts need to be clarified first.

[0056] Uncertainty means that for each pixel in the feature map, it may belong to the background or to the object. Uncertainty measures the confidence of the prediction result, and a prediction point with higher uncertainty is equivalent to lower confidence. For example, for a segmentation task with two classes, the goal is to separate the foreground from the background.

[0057] For segmentation tasks with more than two categories, the uncertainty can be defined as:

[0058] uncertainty=-1×(mask[:,0]-mask[:,1]);

[0059] The uncertainty is estimated as the difference between the first and second predictions interpolated as mask[:,0] and mask[:,1].

[0060] As can be seen from the above formula, the value of uncertainty is always less than 0, so a value close to 0 means greater uncertainty.

[0061] Point selection means: selecting some pixels from multiple pixels based on the maximum uncertainty.

[0062] One of the key concepts of the PointRend algorithm is to refine point predictions based on uncertainty and predictions. The overall framework of PointRend can be summarized as follows: First, a rough prediction is generated based on the mask coarse predictor. Second, the points with the largest uncertainty are selected. Third, the category information of these points is predicted and compared with the true value to calculate the difference. Finally, the rough prediction result is refined based on these points and interpolated, and the second to fourth steps are repeated.

[0063] In order to further improve the segmentation effect of the PointRend module, the image segmentation module of the present invention improves the uncertainty and gradient in PointRend. The specific process is as follows:

[0064] First, the infrared image is initially predicted and a first prediction point set is generated. The first prediction point set includes multiple pixel points, which is also an uncertainty point set. The method of generating rough predictions and uncertainty points in this step is the same as that in the PointRend algorithm. Exemplarily, the first prediction point set, which is also the uncertainty point set, is U(x,y).

[0065] Secondly, based on the first pixel gradient threshold, the second pixel gradient threshold and the edge detection algorithm, the gradient map of the infrared image is determined. In this embodiment, the gradients of all pixels in the infrared image can be calculated first. Then, based on the Canny algorithm, the first pixel gradient threshold and the second pixel gradient threshold, the gradient map GM(x,y) of the infrared image containing only 0 and 1 is determined. Among them, the first pixel gradient threshold is less than the second pixel gradient threshold.

[0066] Then, the first coordinate of each pixel in the first prediction point set in the image coordinate space is determined, and the first coordinate of each pixel in the first prediction point set is converted into a second coordinate in the feature space, and the gradient of each pixel in the first prediction point set in the gradient map of the infrared image is determined. In this embodiment, since the first coordinate in the image coordinate space is the coordinate in the image space, it is necessary to convert the coordinate in the image space into the coordinate in the feature space. The specific conversion method can be: based on the bilinear interpolation method, the first coordinate of each pixel in the first prediction point set is converted into the second coordinate in the feature space. And the gradient of each pixel in the first prediction point set U(x,y) in the gradient map of the infrared image is determined.

[0067] Next, the gradients of all pixels in the first prediction point set are combined into a first gradient set. Exemplarily, the first gradient set can be represented as GM′(x, y).

[0068] Next, the first prediction point set and the negative first gradient set are multiplied element by element to obtain a first uncertainty set, and the first N pixels with the largest uncertainty are selected from the first uncertainty set. Exemplarily, the first prediction point set U(x, y) and -GM′(x, y) are multiplied element by element to obtain the first uncertainty set uncertainty′(x, y). The first N non-pixel points with the largest uncertainty are selected from uncertainty′(x, y).

[0069] It should be noted that element-by-element multiplication refers to the operation of multiplying the elements at corresponding positions in two matrices or vectors of the same dimension. After the elements at corresponding positions are multiplied, a new matrix or vector is obtained, each element of which is the product of the elements at corresponding positions in the original matrix or vector.

[0070] The reason for using the negative first gradient set to multiply element by element is that the original uncertainty value is less than 0, and the points around the boundary need to be assigned a larger uncertainty. After multiplication, the value of the uncertainty point will be greater than 0, while the points not around the boundary will be assigned a value close to 0.

[0071] Finally, based on the difference between the predicted category and the true category of the first N pixels with the largest uncertainty, the PointRend module is back-propagated to determine the contour image of the casing to be detected. Exemplarily, first, the difference between the predicted category and the true category of all target pixels is determined as the point loss. Among them, the target pixel is any pixel among the first N pixels with the largest uncertainty. Then, the point loss of all target pixels is back-propagated to the entire network of the PointRend module, and the network parameters are updated.

[0072] For infrared images of casings with complex backgrounds, the accuracy of the boundary determines the accuracy of the segmentation results. However, for traditional methods that are highly dependent on pixel values, the segmentation performance is not ideal due to the presence of noise. Therefore, the present invention allocates greater uncertainty to the pixels on the boundary and pays attention to the point loss of the pixels, so that the segmentation effect of the segmentation network is better. By improving the PointRend algorithm from the two aspects of gradient and uncertainty, the algorithm combines gradient and uncertainty, pays more attention to the boundary of the area of ​​interest, and obtains more accurate segmentation results than other methods.

[0073] The image segmentation module composed of the improved PointRend algorithm improves the accuracy and clarity of the segmentation boundary of the red jacket image under complex background.

[0074] In some embodiments, after obtaining the bushing image segmentation boundary, it is also necessary to perform fault detection on the image obtained by the bushing image segmentation boundary. However, since there are a variety of insulating materials and electrical components inside the transformer bushing, these components may have uneven temperature distribution due to different working conditions or local faults, so fault diagnosis is difficult.

[0075] In order to improve the accuracy of casing fault detection, the present invention uses a visual Transformer module to construct a fault detection module for detecting contour images.

[0076] In some embodiments, the fault detection module includes a local attention mechanism and a global attention mechanism, which can effectively capture the complex structure and details of the transformer bushing. This capability helps to extract useful features from a large amount of image data for subsequent fault diagnosis analysis.

[0077] In this embodiment, if Figure 3 As shown, the fault detection module is a Swin Transformer module, which is a new visual Transformer that uses a shift window to calculate self-attention.

[0078] The Swin Transformer module includes Patch Partition layer, Linear Embedding layer, SwinTBlock layer and Patch Merging layer. Among them, the Patch Partition layer is used for block partitioning and dimensionality reduction operations, the LinearEmbedding layer is used for linear transformation, the Patch Merging layer is used for downsampling, and the SwinT Block layer includes a multi-head attention module for moving windows.

[0079] As part of the multi-head attention, each attention head is calculated as follows:

[0080]

[0081] Where Q, K, and V represent the query matrix, key matrix, and value matrix respectively, d is the vector dimension, B is the bias matrix, and SoftMax is the multi-class activation function.

[0082] The casing fault detection model provided by the present invention can improve the accuracy of fault diagnosis by using the SwinTransformer module through the use of local attention mechanism and global attention mechanism.

[0083] The fault detection method of the present invention, in order to detect the casing to be detected, needs to first obtain an infrared image of the casing to be detected, and then input the obtained infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected. Among them, the casing fault detection model includes an image segmentation module and a fault detection module, the fault detection module is constructed based on the visual Transformer module, the image segmentation module is obtained by improving the uncertainty and gradient of the pixels in the PointRend module, and the image segmentation module is used to perform contour segmentation processing on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected. By improving the uncertainty and gradient of the pixels in the PointRend module, the accuracy and clarity of the segmentation boundary of the image segmentation module under complex backgrounds can be further improved, and the boundary of the casing can be more accurately segmented, which is beneficial to subsequent fault detection.

[0084] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0085] Based on the transformer bushing fault detection method provided in the above embodiment, the present invention also provides a specific implementation of a transformer bushing fault detection device applied to the transformer bushing fault detection method. Please refer to the following embodiment.

[0086] like Figure 4 As shown, a transformer bushing fault detection device 400 is provided, the device comprising:

[0087] An acquisition module 410 is used to acquire an infrared image of the casing to be inspected;

[0088] The identification module 420 is used to input the infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected;

[0089] Among them, the casing fault detection model includes an image segmentation module and a fault detection module. The fault detection module is built based on the visual Transformer module. The image segmentation module is obtained by improving the uncertainty and gradient of pixels in the PointRend module. The image segmentation module is used to perform contour segmentation on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected. The fault detection module is used to detect the contour image.

[0090] In a possible implementation, the recognition module 420 is used to perform an initial prediction on the infrared image and generate a first prediction point set; wherein the first prediction point set includes a plurality of pixel points;

[0091] Determine a gradient map of the infrared image based on a first pixel gradient threshold, a second pixel gradient threshold, and an edge detection algorithm;

[0092] Determine a first coordinate of each pixel point in the first prediction point set in the image coordinate space, convert the first coordinate of each pixel point in the first prediction point set into a second coordinate in the feature space, and determine a gradient of each pixel point in the first prediction point set in a gradient map of the infrared image;

[0093] The gradients of all pixel points in the first prediction point set are combined into a first gradient set;

[0094] Multiply the first prediction point set and the negative first gradient set element by element to obtain a first uncertainty set, and select the top N pixels with the largest uncertainty from the first uncertainty set;

[0095] Based on the difference between the predicted category and the true category of the first N pixels with the largest uncertainty, the PointRend module is back-propagated to determine the contour image of the casing to be detected.

[0096] In a possible implementation, the identification module 420 is used to calculate the gradient of all pixels in the infrared image;

[0097] Based on the Canny algorithm, the first pixel gradient threshold and the second pixel gradient threshold, it is determined that the infrared image only contains a gradient map of 0 and 1.

[0098] In a possible implementation, the identification module 420 is configured to transform the first coordinate of each pixel point in the first prediction point set into a second coordinate in the feature space based on a bilinear interpolation method.

[0099] In a possible implementation, the identification module 420 is used to determine the difference between the predicted category and the true category of all target pixels as a point loss; wherein the target pixel is any pixel among the first N pixels with the largest uncertainty;

[0100] The point losses of all target pixels are back-propagated to the entire network of the PointRend module and the network parameters are updated.

[0101] In one possible implementation, the fault detection module includes a local attention mechanism and a global attention mechanism.

[0102] In a possible implementation, the fault detection module is a Swin Transformer module, and the Swin Transformer module includes a Patch Partition layer, a Linear Embedding layer, a SwinT Block layer, and a PatchMerging layer;

[0103] Among them, the Patch Partition layer is used for block partitioning and dimensionality reduction operations, the Linear Embedding layer is used for linear transformation, the Patch Merging layer is used for downsampling, and the SwinT Block layer includes a multi-head attention module for moving windows.

[0104] Figure 5 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 5 As shown, the electronic device 5 of this embodiment includes: a processor 50, a memory 51, and a computer program 52 stored in the memory 51 and executable on the processor 50. When the processor 50 executes the computer program 52, the steps in the above-mentioned transformer bushing fault detection method embodiments are implemented, for example Figure 1 Alternatively, when the processor 50 executes the computer program 52, the functions of each module in the above-mentioned device embodiments are realized, for example, Figure 4 Functions of modules 410 to 420 are shown.

[0105] Exemplarily, the computer program 52 may be divided into one or more modules, which are stored in the memory 51 and executed by the processor 50 to implement the present invention. The one or more modules may be a series of computer program instruction segments that can implement specific functions, and the instruction segments are used to describe the execution process of the computer program 52 in the electronic device 5. For example, the computer program 52 may be divided into Figure 4 Modules 410 to 420 are shown.

[0106] The electronic device 5 may include, but is not limited to, a processor 50 and a memory 51. Those skilled in the art will appreciate that Figure 5It is only an example of the electronic device 5 and does not constitute a limitation of the electronic device 5. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0107] The processor 50 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0108] The memory 51 may be an internal storage unit of the electronic device 5, such as a hard disk or memory of the electronic device 5. The memory 51 may also be an external storage device of the electronic device 5, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 5. Further, the memory 51 may also include both an internal storage unit of the electronic device 5 and an external storage device. The memory 51 is used to store the computer program and other programs and data required by the electronic device. The memory 51 may also be used to temporarily store data that has been output or is to be output.

[0109] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0110] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0111] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0112] In the embodiments provided by the present invention, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0113] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0114] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0115] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned transformer bushing fault detection method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium.

[0116] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A transformer bushing fault detection method, characterized in that: include: Acquire an infrared image of the casing to be inspected; Inputting the infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected; Among them, the casing fault detection model includes an image segmentation module and a fault detection module. The fault detection module is constructed based on the visual Transformer module. The image segmentation module is obtained by improving the uncertainty and gradient of pixels in the PointRend module. The image segmentation module is used to perform contour segmentation processing on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected, and the fault detection module is used to detect the contour image.

2. The fault detection method according to claim 1, characterized in that: The processing process of the image segmentation module is: Performing an initial prediction on the infrared image and generating a first prediction point set; wherein the first prediction point set includes a plurality of pixel points; Determining a gradient map of the infrared image based on a first pixel gradient threshold, a second pixel gradient threshold, and an edge detection algorithm; Determine a first coordinate of each pixel point in the first prediction point set in the image coordinate space, convert the first coordinate of each pixel point in the first prediction point set into a second coordinate in the feature space, and determine a gradient of each pixel point in the first prediction point set in the gradient map of the infrared image; The gradients of all pixels in the first prediction point set are combined into a first gradient set; Multiplying the first prediction point set and the negative first gradient set element by element to obtain a first uncertainty set, and selecting top N pixels with maximum uncertainty from the first uncertainty set; Based on the difference between the predicted category and the true category of the first N pixels with the largest uncertainties, the PointRend module is back-propagated to determine the contour image of the casing to be detected.

3. The fault detection method according to claim 2, characterized in that: The step of determining the gradient map of the infrared image based on the first pixel gradient threshold, the second pixel gradient threshold and the edge detection algorithm comprises: Calculating the gradients of all pixels in the infrared image; Based on the Canny algorithm, the first pixel gradient threshold and the second pixel gradient threshold, it is determined that the infrared image only includes a gradient map of 0 and 1.

4. The fault detection method according to claim 2, characterized in that: The converting the first coordinate of each pixel point in the first prediction point set into a second coordinate in the feature space includes: Based on a bilinear interpolation method, the first coordinates of each pixel point in the first prediction point set are converted into second coordinates in a feature space.

5. The fault detection method according to claim 2, characterized in that: The back-propagation of the PointRend module based on the difference between the predicted category and the true category of the first N pixels with the largest uncertainties includes: The difference between the predicted category and the true category of all target pixels is determined as the point loss; wherein the target pixel is any pixel among the first N pixels with the largest uncertainty; The point losses of all the target pixels are back-propagated to the entire network of the PointRend module, and the network parameters are updated.

6. The fault detection method according to claim 1, characterized in that: The fault detection module includes a local attention mechanism and a global attention mechanism.

7. The fault detection method according to claim 6, characterized in that: The fault detection module is a Swin Transformer module, and the Swin Transformer module includes a Patch Partition layer, a Linear Embedding layer, a SwinT Block layer and a Patch Merging layer; Among them, the Patch Partition layer is used for block partitioning and dimensionality reduction operations, the Linear Embedding layer is used for linear transformation, the Patch Merging layer is used for downsampling, and the SwinT Block layer includes a multi-head attention module for moving windows.

8. A transformer bushing fault detection device, characterized in that: include: An acquisition module, used for acquiring an infrared image of the casing to be inspected; An identification module, used for inputting the infrared image into a pre-trained casing fault detection model to identify the fault type of the casing to be detected; Among them, the casing fault detection model includes an image segmentation module and a fault detection module. The fault detection module is constructed based on the visual Transformer module. The image segmentation module is obtained by improving the uncertainty and gradient of pixels in the PointRend module. The image segmentation module is used to perform contour segmentation processing on the infrared image of the casing to be detected to obtain the contour image of the casing to be detected, and the fault detection module is used to detect the contour image.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images