Defect multi-label determination method and device for power equipment and electronic equipment

Abstract

The invention provides a defect multi-label determination method and device for power equipment and electronic equipment, and relates to the technical field of power. The method comprises the following steps: acquiring an equipment image of target power equipment with defects and a target text label for identifying the defects of the target power equipment; and performing multi-view feature extraction on the equipment image of the target power equipment to obtain image features under multiple views. And then, based on the target text tag and the image features under the multiple views, performing modeling through a preset method to generate a multi-tag classification model. And then, randomly acquired electrical equipment images are directly input into the trained multi-label classification model, and a plurality of defect labels possibly existing in the electrical equipment images can be predicted and output by the multi-label classification model. According to the invention, the defects of the power equipment can be comprehensively and accurately determined.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a method, device and electronic device for determining defects of electric power equipment using multiple labels. Background Art

[0002] Power equipment refers to various equipment and devices used for power generation, transmission, distribution and consumption. It plays a vital role in the power system. The reliability of the power system depends on the normal operation of power equipment. Therefore, it is necessary to conduct regular inspections on power equipment to identify possible defects in power equipment so that it can be repaired or replaced in time when power equipment fails.

[0003] In traditional defect identification methods for power equipment, defect analysis of power equipment usually relies on data from a single perspective or a single data source, such as single sensor data, a single image or video, a single physical quantity measurement, or data at a single time point. The analysis results are usually presented in the form of a label, which is based on the analysis results of a single perspective or a single data source and is used to identify a specific defect in the power equipment.

[0004] However, the existing technology has the technical problem that the defect identification results of power equipment are not comprehensive and accurate enough. Summary of the invention

[0005] The present application provides a method, device and electronic device for determining defects of electric power equipment by multiple labels, so as to solve the technical problem that the defect identification results of electric power equipment in the prior art are not comprehensive and accurate enough.

[0006] In a first aspect, the present application provides a multi-label defect determination method for electric power equipment, comprising:

[0007] Acquire a device image of a target power device and a corresponding target text label; wherein the target power device represents a power device with defects, and the target text label is used to identify the defects of the target power device;

[0008] Perform multi-view feature extraction on the device image to obtain image features under the multi-views; perform modeling based on the target text label and the image features under the multi-views by a preset method to generate a multi-label classification model;

[0009] Acquire an image of an electric power device waiting to be labeled, input the image of the device waiting to be labeled into the multi-label classification model, and determine multiple defect labels output by the multi-label classification model; wherein the defect labels indicate defects existing in the electric power device waiting to be labeled.

[0010] In a possible design, the number of target power devices acquired is a, and the a target power devices include n device images; wherein a and n are both positive integers greater than or equal to 1, and n is greater than or equal to a;

[0011] The extracting multi-view features of the device image to obtain image features under multiple views includes:

[0012] Determine m pre-specified preset feature extraction methods; wherein the m pre-specified feature extraction methods correspond to m views, and m is a positive integer greater than or equal to 1;

[0013] Building an image database based on the n device images, and performing multi-view feature extraction on the n device images in the image database respectively according to the m preset feature extraction methods;

[0014] Determine the extracted image features under m views, wherein each view includes n image features.

[0015] In a possible design, the image features under multiple views include image features under m views, where m is a positive integer greater than or equal to 1;

[0016] The method of modeling based on the target text label and the image features under multiple views by a preset method to generate a multi-label classification model includes:

[0017] Based on the image features in the m views, m image feature matrices are respectively constructed; wherein each image feature matrix includes n image features from n device images in the same view, and n is a positive integer greater than or equal to 1;

[0018] Constructing a text label matrix based on the target text label; replicating the text label matrix m times according to the m image feature matrices to obtain m text label matrices;

[0019] The m image feature matrices correspond to the m text label matrices one by one, and the corresponding text label matrices are set based on the m image feature matrices;

[0020] The m image feature matrices and the m text label matrices are modeled by a preset method to generate a multi-label classification model.

[0021] In a possible design, the m image feature matrices and the m text label matrices are modeled by a preset method to generate a multi-label classification model, including:

[0022] By a preset method, the m image feature matrices are respectively mapped to corresponding text label matrices to generate an initial model; wherein the initial model indicates a mapping relationship between the m image feature matrices and the corresponding text label matrices;

[0023] The initial model is optimized and the optimized initial model is determined as a multi-label classification model.

[0024] In a possible design, the m image feature matrices are respectively mapped to corresponding text label matrices by a preset method, including:

[0025] Constructing corresponding m adjacency graphs for the m image feature matrices respectively; wherein the nodes in the adjacency graph represent the image feature matrix, and the edges in the adjacency graph represent the similarity or proximity relationship between the nodes;

[0026] Based on the adjacency graph, the local manifold structure of the m image feature matrices is captured; based on the local manifold structure, the m image feature matrices are respectively mapped to the corresponding text label matrices through the preset method to retain the local manifold structure of the corresponding image feature matrix in the text label matrix; wherein the local manifold structure characterizes the intrinsic geometric structure of the image features in the image feature matrix.

[0027] In a possible design, the optimizing the initial model and determining the optimized initial model as a multi-label classification model includes:

[0028] Adding a Laplace regularization term to the initial model to modify the initial model;

[0029] The modified initial model is optimized based on the preset optimization strategy until the model converges to obtain a multi-label classification model.

[0030] In a possible design, the text labels in the text label matrix are in the form of binary vectors.

[0031] The setting of the corresponding text label matrix based on the m image feature matrices includes:

[0032] Analyze the image features included in each image feature matrix and the corresponding text label matrix;

[0033] Determine the electric power equipment defect indicated by the image features in the target image feature matrix, and determine the target defect label in the text label matrix corresponding to the target image feature matrix; wherein the target image feature matrix is ​​one of the m image feature matrices, and the target defect label indicates the electric power equipment defect;

[0034] The binary vector corresponding to the target defect label is set to 1, and the binary vectors corresponding to other text labels except the target defect label in the text label matrix corresponding to the target image matrix are set to 0.

[0035] In a second aspect, the present application provides a defect multi-label determination device for electric power equipment, comprising:

[0036] An acquisition module, used to acquire an equipment image of a target electric equipment and a corresponding target text label; wherein the target electric equipment represents an electric equipment with defects, and the target text label is used to identify the defects of the target electric equipment;

[0037] An extraction module, used for performing multi-view feature extraction on the device image to obtain image features under multiple views;

[0038] A construction module is used to perform modeling based on the target text label and the image features under multiple views through a preset method to generate a multi-label classification model;

[0039] The acquisition module is further used to acquire the image of the power equipment to be labeled;

[0040] An input module is used to input the image of the device to be labeled into the multi-label classification model and determine multiple defect labels output by the multi-label classification model; wherein the defect label indicates the defects of the power equipment waiting to be labeled.

[0041] In a possible design, the number of target power devices acquired is a, and the a target power devices include n device images; wherein a and n are both positive integers greater than or equal to 1, and n is greater than or equal to a;

[0042] The extraction module further includes: a determination module, configured to determine m pre-specified preset feature extraction methods; wherein the m pre-specified feature extraction methods correspond to m views, and m is a positive integer greater than or equal to 1;

[0043] The construction module is further used to construct an image database based on the n device images;

[0044] The extraction module is further used to perform multi-view feature extraction on the n device images in the image database according to the m preset feature extraction methods;

[0045] The determination module is further used to determine the extracted image features under m views; wherein each view includes n image features.

[0046] In a possible design, the image features under multiple views include image features under m views, where m is a positive integer greater than or equal to 1;

[0047] The building blocks are also used to:

[0048] Based on the image features in the m views, m image feature matrices are respectively constructed; wherein each image feature matrix includes n image features from n device images in the same view, and n is a positive integer greater than or equal to 1;

[0049] Constructing a text label matrix based on the target text label; replicating the text label matrix m times according to the m image feature matrices to obtain m text label matrices;

[0050] The m image feature matrices correspond to the m text label matrices one by one, and the corresponding text label matrices are set based on the m image feature matrices;

[0051] The m image feature matrices and the m text label matrices are modeled by a preset method to generate a multi-label classification model.

[0052] In a possible design, the construction module further includes: a mapping module, which is used to map the m image feature matrices to corresponding text label matrices respectively through a preset method to generate an initial model; wherein the initial model indicates a mapping relationship between the m image feature matrices and the corresponding text label matrices;

[0053] The construction module also includes: a processing module, which is used to optimize the initial model and determine the optimized initial model as a multi-label classification model.

[0054] In a possible design, the construction module is further used to construct corresponding m adjacency graphs for the m image feature matrices respectively; wherein the nodes in the adjacency graph represent the image feature matrix, and the edges in the adjacency graph represent the similarity or proximity relationship between the nodes;

[0055] The mapping module further comprises: a capturing module for capturing the local manifold structure of the m image feature matrices based on the adjacency graph;

[0056] The mapping module is also used to map the m image feature matrices to corresponding text label matrices respectively based on the local manifold structure through the preset method, so as to retain the local manifold structure of the corresponding image feature matrix in the text label matrix; wherein the local manifold structure represents the intrinsic geometric structure of the image features in the image feature matrix.

[0057] In a possible design, the processing module further includes: a correction module, configured to add a Laplace regularization term to the initial model to correct the initial model;

[0058] The processing module is also used to optimize the corrected initial model based on a preset optimization strategy until the model converges to obtain a multi-label classification model.

[0059] In a possible design, the text labels in the text label matrix are in the form of binary vectors.

[0060] The construction module further includes: an analysis module for analyzing the image features included in each image feature matrix and the corresponding text label matrix;

[0061] The determination module is further used to determine the power equipment defect indicated by the image features in the target image feature matrix, and determine the target defect label in the text label matrix corresponding to the target image feature matrix; wherein the target image feature matrix is ​​one of the m image feature matrices, and the target defect label indicates the power equipment defect;

[0062] The construction module also includes: a setting module, which is used to set the binary vector corresponding to the target defect label to 1, and set the binary vectors corresponding to other text labels in the text label matrix corresponding to the target image matrix, except for the target defect label, to 0.

[0063] In a third aspect, an embodiment of the present application provides an electronic device, comprising: at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method described in the first aspect above and various possible designs.

[0064] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the method described in the first aspect and various possible designs is implemented.

[0065] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect and various possible designs of the first aspect.

[0066] The multi-label determination method, device and electronic device for defects of electric power equipment provided in the present application obtain equipment images of target electric power equipment with defects, and target text labels that identify defects of target electric power equipment. Multi-view feature extraction is performed on the equipment image of the target electric power equipment to obtain image features under multiple views. Afterwards, modeling is performed by a preset method based on the target text labels and the image features under multiple views to generate a multi-label classification model. It can be seen that multi-view learning and multi-label learning are integrated in the process of building a multi-label classification model. Multi-view learning enables the multi-label classification model to extract image features under different views, increasing the information diversity of the equipment image, and multi-label learning enables the multi-label classification model to process multiple text labels at the same time, improving the ability to identify defects that may exist in complex equipment images. Furthermore, the acquired equipment image to be labeled of the electric power equipment waiting to be labeled is input into the multi-label classification model, and multiple defect labels output by the multi-label classification model are obtained. These defect labels can comprehensively and accurately identify the defects in the equipment image to be labeled. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0068] Figure 1 Schematic diagram of the process of the multi-label determination method for defects of power equipment provided in the embodiment of the present application Figure 1 ;

[0069] Figure 2 A schematic diagram of the structure of a method for determining defects of electric power equipment by multiple labels provided in an embodiment of the present application;

[0070] Figure 3 Schematic diagram of the process of the multi-label determination method for defects of power equipment provided in the embodiment of the present application Figure 2 ;

[0071] Figure 4 A schematic diagram of the structure of a device for determining defects of electric power equipment with multiple labels provided in an embodiment of the present application;

[0072] Figure 5 A hardware structure diagram of an electronic device provided in an embodiment of the present application.

[0073] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0074] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0075] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein, for example.

[0076] In the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0077] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0078] Power equipment is the basis for the normal operation of the power system, and the reliability of the power system depends to a large extent on the normal operation of power equipment. Any failure or abnormality of power equipment may lead to interruption of the power system or reduced efficiency.

[0079] Since power equipment directly affects the production, transmission, distribution and use of electricity, it is necessary to conduct regular inspections of power equipment to identify possible defects in the power equipment. Once defects in power equipment are detected, timely measures can be taken to repair or replace them to ensure the continuity and reliability of the power system and the stability of power supply.

[0080] Traditional power equipment defect identification methods usually rely on data from a single perspective or a single data source to perform defect analysis on power equipment, such as single sensor data, a single image or video, a single physical quantity measurement, or data from a single time point.

[0081] This means that only specific types of operating data of the power equipment can be captured. For example, a single sensor can only capture specific types of data related to the sensor, a single image or video can only capture defects visible from that angle, and a single physical quantity can only reflect the state of the power equipment related to that physical quantity. Data at a single point in time is only based on data at a specific point in time and cannot capture dynamic changes and trends in the operating status of the power equipment.

[0082] After completing the defect analysis of the power equipment, the analysis results are usually presented in the form of a label. This label is a simplified identifier used to describe a specific defect in the power equipment, such as overheating, abnormal vibration, etc.

[0083] However, because traditional methods rely on a single perspective or a single data source for analysis, that is, the labels corresponding to the analysis results are derived based on limited information. For example, if only the data from the temperature sensor is used, the label may only reflect temperature-related faults and ignore other important factors.

[0084] It can be seen that relying only on data from a single perspective or a single data source to analyze defects in power equipment may not capture all relevant information about the operation of the power equipment, which may result in some defects not being detected or misjudging the normal state as an abnormal state. In addition, power equipment may involve changes in multiple parameters due to complex defect modes, and defect identification through data from a single perspective or a single data source may easily miss some defects that are not easy to detect.

[0085] Therefore, existing methods for identifying defects in power equipment cannot comprehensively and accurately identify possible defects in power equipment.

[0086] In view of the above technical problems, in order to avoid the defect identification errors of power equipment caused by information from a single perspective, the inventors thought of collecting the equipment images and text labels of the power equipment at the same time, and increasing the information diversity of the power equipment through multimodal information. At the same time, in order to further obtain more comprehensive information about the power equipment, the inventors considered performing multi-view feature extraction on the collected equipment images to obtain image features under multiple views. Then, based on the extracted image features under multiple views and the collected text labels, a multi-label classification model that allows simultaneous identification of multiple defect categories is constructed. Through the multi-label classification model, the defects of the power equipment can be comprehensively and accurately identified.

[0087] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0088] An embodiment of the present application provides a multi-label defect determination method for electric power equipment. Figure 1 Schematic diagram of the process of the multi-label determination method for defects of power equipment provided in the embodiment of the present application Figure 1 ,like Figure 1 As shown, the multi-label defect determination method of the electric power equipment includes:

[0089] S101, obtaining a device image of a target power device and a corresponding target text label.

[0090] The target power equipment refers to the power equipment with defects, and the target text label is used to identify the specific defects of the target power equipment. For example, the equipment image can be a visible light image, an infrared image, an ultraviolet image, a hyperspectral image, etc., and the target text label can be cracks, breakage, missing screws, missing R pins, oil leakage, discoloration, etc.

[0091] There are many ways to obtain device images and target text labels. You can directly obtain the device image of the target power equipment and the corresponding target text label from the Internet; if the acquired device image lacks the corresponding target text label, you can manually annotate it. You can also manually use different types of handheld sensors to obtain the device image of the target power equipment, and manually annotate the device image to obtain the target text label. The method of obtaining the device image and target text label is determined by actual needs and is not specifically limited here.

[0092] It is understandable that the original equipment image obtained may have noise, blur, uneven brightness and other problems, and the target text label obtained may have spelling errors, inconsistent formats or redundant information. Therefore, after obtaining the equipment image of the target power equipment and the corresponding target text label, the equipment image and the target text label are preprocessed to make them more suitable for subsequent analysis and processing.

[0093] Furthermore, the preprocessed device image and target text label are stored separately and managed in association. That is, the device image and target text label are stored separately, and the device image and the corresponding target text label are associated to ensure the correct correspondence between the device image and the target text label. The separate storage and association management method not only improves the efficiency and flexibility of data management, but also provides greater convenience and scalability for subsequent analysis, training and application.

[0094] S102, performing multi-view feature extraction on the device image to obtain image features under the multi-views; and performing modeling by a preset method based on the target text label and the image features under the multi-views to generate a multi-label classification model.

[0095] Assume that the number of target power devices acquired in step S101 is a, and the a target power devices include n device images, where a and n are both positive integers greater than or equal to 1, and n is greater than or equal to a. This means that each target power device includes at least one device image, and it should be noted that each device image corresponds to at least one target text label.

[0096] In a possible implementation, a specific method of performing multi-view feature extraction on a device image is as follows: m pre-specified preset feature extraction methods are determined, where the m preset feature extraction methods correspond to m views, and m is a positive integer greater than or equal to 1. An image database is constructed based on n device images, and multi-view feature extraction is performed on the n device images in the image database according to the m preset feature extraction methods, so as to obtain image features under the m views.

[0097] Explanatory, by performing feature extraction on n device images according to any one of the m preset feature extraction methods, a corresponding image feature can be extracted from each device image, and therefore, each view includes n image features.

[0098] It should be noted that the preset feature extraction methods may include global image (GIST) feature extraction, image color (RGB) feature extraction, Histograms of oriented gradients (HOG) feature extraction, local binary patterns (LBP) feature extraction, scale invariant feature transform (SIFT) feature extraction, color code (HSV) feature extraction, and machine learning or deep learning feature extraction, etc., and the specific preset feature extraction method needs to be selected according to the actual situation.

[0099] Among them, the GIST feature extraction method corresponds to the GIST image view, the RGB feature extraction method corresponds to the RGB image view, the HOG feature extraction method corresponds to the HOG image view, the LBP corresponds to the LBP image view, the SIFT feature extraction method corresponds to the SIFT image view, and the HSV feature extraction method corresponds to the HSV image view. The machine learning or deep learning feature extraction method is used to extract diversified features of the device image through machine learning or deep learning, and corresponds to the machine learning or deep learning image view.

[0100] After obtaining the image features under m views, m image feature matrices are constructed based on the image features under m views, that is, each view corresponds to an image feature matrix. Each image feature matrix includes n image features from n device images under the same view.

[0101] Next, a text label matrix is ​​constructed based on the target text label, and the text label matrix includes all the acquired target text labels. According to the m image feature matrices, the text label matrix is ​​copied m times to obtain m text label matrices. After that, the m image feature matrices are mapped one-to-one with the m text label matrices, that is, each image feature matrix has a corresponding text label matrix.

[0102] It should be explained that the text labels in the text label matrix are in the form of binary vectors, and the corresponding text label matrix needs to be set based on m image feature matrices to associate the image features in the image feature matrix with the text labels in the corresponding text label matrix.

[0103] Specifically, the image features included in each image feature matrix are analyzed with the corresponding text label matrix to determine the defects of the power equipment indicated by the image features in the target image feature matrix. The target image feature matrix is ​​one of the m image feature matrices. Next, in the text label matrix corresponding to the target image feature matrix, the target defect label indicating the defect of the power equipment is determined. The binary vector corresponding to the target defect label is set to 1, and the binary vectors corresponding to the other text labels except the target defect label in the text label matrix corresponding to the target image matrix are set to 0.

[0104] For the convenience of description, the mathematical expression of the m image feature matrix is ​​defined as: χ={X 1 ,X 2 ,…,X m}. Among them, X v ∈R n×dv represents the image features under the vth view, R represents a real number set, n is the number of samples, that is, the number of device images, dv represents the image feature dimension of the vth view, m represents the number of views, 1≤v≤m.

[0105] If sample i is related to text label j, the text label matrix Y∈R n×l The element in the i-th row and j-th column of is set to 1, otherwise it is set to 0, where l is the number of text labels. Define a multi-view multi-label training dataset D with n samples = {(X i ,Y i )|1≤i≤n}. Among them, X i ∈χ represents the i-th sample, which contains m feature vectors, expressed as {X i 1 ,X i 2 ,…,X i m}, each Yi ∈{0,1} l×1 Yes X i The corresponding label vector.

[0106] Explanatory, image features under different views are complementary and can be used to predict text labels. Moreover, each text label should only be associated with image features from different image feature matrices. Therefore, m image feature matrices and m text label matrices can be modeled by a preset method to generate a multi-label classification model.

[0107] In a possible implementation, the m image feature matrices are respectively mapped to corresponding text label matrices by a preset method to generate an initial model, wherein the initial model indicates a mapping relationship between the m image feature matrices and the corresponding text label matrices.

[0108] Specifically, the m image feature matrices are mapped to the corresponding text label matrices as follows: corresponding m adjacency graphs are constructed for the m image feature matrices, wherein the nodes in the adjacency graph represent the image feature matrices and the edges represent the similarity or proximity relationship between the nodes. Next, the local manifold structure of the m image feature matrices is captured based on the adjacency graph, and based on the local manifold structure, the m image feature matrices are mapped to the corresponding text label matrices by a preset method, so as to retain the local manifold structure of the corresponding image feature matrix in the text label matrix. The local manifold structure represents the intrinsic geometric structure of the image features in the image feature matrix.

[0109] After the initial model is generated, it is necessary to optimize the initial model and determine the optimized initial model as a multi-label classification model. Specifically, a Laplace regularization term can be added to the initial model to correct the initial model. Afterwards, the corrected initial model is optimized based on a preset optimization strategy until the model converges to obtain a multi-label classification model.

[0110] Next, the construction process of the multi-label classification model is explained in detail through a specific example. In a specific example, the least squares method is used to model the image feature matrix and the label feature matrix of each view, and the mathematical formula is expressed as follows:

[0111]

[0112] in,‖·‖ F is the Frobenius norm, ‖·‖ 1 is the L1-norm, and α is a balancing hyperparameter greater than 0. W v ∈R dv×l Represents the regression coefficient matrix of the vth image feature matrix, Z∈R n×lRepresents the predicted text label matrix, containing continuous values. In the true text label matrix Y, the elements of positive labels are set to 1, and the elements of negative or unobserved text labels are set to 0. Z is similar to but not exactly the same as Y, because Y may contain some missing values ​​and noise values. In addition, the continuous elements of Z can be binarized, for example, elements greater than a 0.5 threshold are set to 1 (representing a positive label) and elements less than a 0.5 threshold are set to 0 (representing a negative label).

[0113] Furthermore, according to the smoothness hypothesis, device images with similar image features should have similar text labels. Therefore, the local manifold structure of the corresponding image feature matrix can be preserved in the text label matrix through graph learning technology. That is, the similarity matrix of each image feature matrix is ​​calculated, and then the k-adjacent graph S is constructed. v , the mathematical formula is as follows:

[0114]

[0115] Among them, X i v , X j v is the image feature matrix X v ∈R n×dv The vectors of the i-th and j-th rows of X ij v is an adjacency graph S v The present invention uses the Gaussian kernel function

[0116]

[0117] As the distance metric of the data, σ controls the width of the adjacency. Then, the local manifold structure S v Combined with the predicted text label matrix Z, we have the following formula:

[0118]

[0119] Among them, Z i ∈R 1×l is the i-th row of the predicted text label matrix Z, Z j ∈R 1×l is the j-th row of the predicted text label matrix Z.

[0120] This means that in the vth image feature matrix, device images with similar image features should have similar predicted text labels. Because the local manifold structure of the image feature matrix is ​​preserved in the predicted text label matrix, formula (1-3) can be rewritten as:

[0121]

[0122] Among them, S v The Laplace matrix L v =D v -S v , S v The degree matrix

[0123] However, formula (1-4) only mines low-order information in the local manifold structure, but does not mine high-order information, which may result in insufficient information mining. Therefore, high-order information will be introduced based on the above process to enhance the expressiveness of the model.

[0124] D 1 v ∈R n×n represents the first-order degree matrix, and the first-order Laplace matrix is ​​represented by L 1 v =D 1 v -S 1 v Specifically, in graph theory, the similarity between two points can be represented by a first-order relationship. At the same time, if many vertices with first-order connections to two vertices coincide, then the two vertices have a second-order relationship. That is, S :,i v Represents the first-order similarity matrix S 1 v The second-order similarity matrix can be defined as: S 2 v =(S :,i v ) T S :,j v Therefore, the second-order Laplacian matrix can be defined as L 2 v =D 2 v -S 2 v .

[0125] Considering the complementarity of low-order information and high-order information, the optimal Laplacian matrix can achieve the optimal combination of low-order Laplacian matrix and high-order Laplacian matrix, so as to fully capture the local manifold structure of the image feature matrix. The optimal Laplacian matrix is ​​defined as follows:

[0126] L 12 =λ 1 L 1 +λ 2 L 2 (1-5)

[0127] Among them, λ 1 and λ 2 is a balance parameter greater than 0.

[0128] In addition, the predicted text label matrix Z must not only be smooth in the feature space, but also be consistent with the true text label matrix Y. Therefore, there is the following formula:

[0129]

[0130] Where Tr(·) represents the trace function, β is a balanced hyperparameter greater than 0, (·) T Represents the transpose of a matrix.

[0131] It should be noted that the regression coefficient matrix W of formula (1-1) v The column vectors in represent the image features related to the text labels. For highly related text labels, W v The corresponding column vectors in have great similarity, which is similar to the relationship between image feature similarity and text label similarity. Furthermore, the Laplacian graph can be used to represent the relationship between text label association and image feature sharing, that is, the following formula:

[0132]

[0133] It should be noted that the above formula can be further written as:

[0134]

[0135] Where R∈R n×n Represents the regression coefficient relationship diagram, the Laplace matrix of R is L R =D R -R, the degree matrix of R is

[0136] Based on the correlation of text labels, using Gaussian kernel function

[0137]

[0138] Construct a known graph R 0 ∈R n×n Since the known true text label matrix Y is usually incomplete or noisy, the correlation graph can be refined while learning the image feature coefficients. The graph regularization of the correlation of text labels can be expressed as:

[0139]

[0140] Among them, γ is a balancing hyperparameter greater than 0.

[0141] Combining formula (1-1), formula (1-6) and formula (1-9) together, we can get the revised initial model:

[0142]

[0143] Furthermore, an alternating optimization strategy can be used to optimize formula (1-10) until the model converges to obtain the final multi-label classification model.

[0144] S103: Acquire a device image of the power equipment waiting to be labeled, input the device image to the multi-label classification model, and determine a plurality of defect labels output by the multi-label classification model.

[0145] Among them, the defect label indicates the defects existing in the power equipment waiting to be tagged.

[0146] This step involves using a multi-label classification model to process the image of the device to be labeled. The multi-label classification model can be regarded as a machine learning model that can simultaneously identify multiple defect types in the image of the device to be labeled. The purpose of this step is to automatically detect defects in power equipment. The multi-label classification model allows multiple defect types to be identified simultaneously, which is particularly important for complex defect identification tasks.

[0147] The multi-label defect determination method of the power equipment provided in the present application obtains n equipment images of a target power equipment with defects, and a target text label identifying the defects of the target power equipment. According to the pre-specified m preset feature extraction methods, multi-view feature extraction is performed on the n equipment images to obtain image features under m views. Afterwards, m image feature matrices are respectively constructed based on the image features under the m views, and m text label matrices corresponding to the m image feature matrices are constructed based on the target text labels. Furthermore, m corresponding adjacency graphs are respectively constructed for the m image feature matrices, and the local manifold structure of the m image feature matrices is captured based on the m adjacency graphs. The m image feature matrices are respectively mapped to the corresponding text label matrices by a preset method to generate an initial model. By optimizing the initial model, the optimized initial model is determined as a multi-label classification model. It can be seen that in the process of building a multi-label classification model, multi-view learning and multi-label learning are integrated. Multi-view learning enables the multi-label classification model to extract image features under different views, increasing the information diversity of the equipment image, and multi-label learning enables the multi-label classification model to process multiple text labels at the same time, improving the ability to identify defects that may exist in complex equipment images. Furthermore, the image of the power equipment to be labeled is input into the multi-label classification model, and multiple defect labels output by the multi-label classification model are obtained. These defect labels can comprehensively and accurately identify the defects in the image of the equipment to be labeled.

[0148] Next, a specific embodiment is used to Figure 1 The specific process of the multi-label defect determination method for the power equipment shown is summarized. Figure 2 A schematic diagram of the structure of the multi-label defect determination method for power equipment provided in an embodiment of the present application, such as Figure 2 As shown, the multi-label defect determination method of the power equipment includes: a data acquisition and processing unit, a feature extraction unit, a multi-view and multi-label construction unit, a mapping unit, a manifold structure learning unit, a label correlation constraint unit, a model optimization unit and a multi-label prediction output unit.

[0149] The data acquisition and processing unit includes: a device image acquisition subunit, a text label acquisition subunit, a device image screening subunit, a text label screening subunit, a device image cleaning subunit, a text label cleaning subunit and a marking subunit.

[0150] Specifically, the device image acquisition subunit is used to acquire historical device images, and the text label acquisition subunit is used to acquire multi-label text data corresponding to the historical device images. The device image screening subunit is used to screen out defective images from historical device images, and the text label screening subunit is used to screen out text labels indicating defects of historical power equipment from multi-label text data. The device image cleaning subunit is used to remove low-quality and damaged historical device images, and the text label cleaning subunit and the annotation subunit are used to remove erroneous text labels, correct text labels, or manually supplement missing text labels.

[0151] The feature extraction unit includes: a GIST feature extraction subunit, an RGB feature extraction subunit, a HOG feature extraction subunit, an LBP feature extraction subunit, a SIFT feature extraction subunit, an HSV feature extraction subunit, and a deep learning or machine learning feature extraction subunit.

[0152] The mapping unit is used to model the image feature matrix and the corresponding text label matrix under each view. The manifold structure learning unit is used to retain the local manifold structure of the corresponding image feature matrix in the text label matrix. The label correlation constraint unit is used to capture the correlation of text labels to further improve the accuracy of label prediction. The model optimization unit is used to solve the model. The multi-label prediction output unit is used to output multiple defect labels corresponding to the device image.

[0153] It should be noted that the data acquisition and processing unit is connected to the feature extraction unit, the feature extraction unit is connected to the multi-view and multi-label construction unit, the multi-view and multi-label construction unit is connected to the mapping unit, the mapping unit is connected to the manifold structure learning unit, the manifold structure learning unit is connected to the model optimization unit, and the model optimization unit is connected to the multi-label prediction output unit.

[0154] based on Figure 2 The structure shown, Figure 3 Schematic diagram of the process of the multi-label determination method for defects of power equipment provided in the embodiment of the present application Figure 2 ,like Figure 3 As shown, the multi-label defect determination method for electric power equipment specifically includes the following steps:

[0155] S301, acquiring a device image and a corresponding text label through a data acquisition and processing unit;

[0156] S302, performing multi-view feature extraction on the device image by a feature extraction unit;

[0157] S303, constructing an image feature matrix and a text label matrix through a multi-view and multi-label construction unit;

[0158] S304, modeling is performed through a mapping module based on the image feature matrix and the text label matrix;

[0159] S305, retaining the local manifold structure of each image feature matrix through a manifold structure learning unit;

[0160] S306, capturing the relevance of text labels through a label relevance constraint unit;

[0161] S307, optimizing the model through the model optimization unit to obtain a multi-label classification model;

[0162] S308. Output multiple defect labels predicted by the multi-label classification model through a multi-label prediction output unit.

[0163] Through multi-view feature extraction, the multi-label classification model's understanding of the multi-dimensional features of the device image is comprehensively enhanced. At the same time, the local manifold structure of the corresponding image feature matrix is ​​retained in the text label matrix, so that device images with similar image features have similar text labels, which improves the prediction accuracy of the multi-label classification model. Because the label correlation is taken into account, the relationship between text labels can be effectively identified, thereby optimizing the label prediction process of the multi-label classification model, so that related text labels can reinforce each other, improving the overall classification performance of the multi-label classification model. It can be seen that the defect labels output by the multi-label classification model can comprehensively and accurately identify the defects in the device image.

[0164] Figure 4 A schematic diagram of the structure of a multi-label defect determination device for electric power equipment provided in an embodiment of the present application, such as Figure 4 As shown, the defect multi-label determination device 400 of the power equipment includes: an acquisition module 401, an extraction module 402, a construction module 403, and an input module 404;

[0165] The acquisition module 401 is used to acquire the device image of the target power device and the corresponding target text label; wherein the target power device represents the power device with defects, and the target text label is used to identify the defects of the target power device;

[0166] An extraction module 402 is used to extract multi-view features of the device image to obtain image features under multiple views;

[0167] A construction module 403 is used to perform modeling based on target text labels and image features under multiple views by a preset method to generate a multi-label classification model;

[0168] The acquisition module 401 is also used to acquire the image of the device to be labeled of the power device waiting to be labeled;

[0169] The input module 404 is used to input the image of the device to be labeled into the multi-label classification model, and determine multiple defect labels output by the multi-label classification model; wherein the defect label indicates the defects of the power equipment to be labeled.

[0170] In a possible design, the number of target power devices acquired is a, and the a target power devices include n device images; wherein a and n are both positive integers greater than or equal to 1, and n is greater than or equal to a;

[0171] The extraction module 402 further includes: a determination module 405, configured to determine m pre-specified preset feature extraction methods; wherein the m pre-specified feature extraction methods correspond to m views, and m is a positive integer greater than or equal to 1;

[0172] The construction module 403 is further used to construct an image database based on the n device images;

[0173] The extraction module 402 is further used to perform multi-view feature extraction on the n device images in the image database according to m preset feature extraction methods;

[0174] The determination module 405 is further configured to determine the extracted image features under the m views, wherein each view includes n image features.

[0175] In a possible design, the image features under multiple views include image features under m views, where m is a positive integer greater than or equal to 1;

[0176] The building block 403 is further used for:

[0177] Based on the image features in the m views, m image feature matrices are respectively constructed; wherein each image feature matrix includes n image features from n device images in the same view, and n is a positive integer greater than or equal to 1;

[0178] Construct a text label matrix based on the target text label; according to the m image feature matrices, copy the text label matrix m times to obtain m text label matrices;

[0179] The m image feature matrices correspond to the m text label matrices one by one, and the corresponding text label matrices are set based on the m image feature matrices;

[0180] Through a preset method, m image feature matrices and m text label matrices are modeled to generate a multi-label classification model.

[0181] In a possible design, the construction module 403 further includes: a mapping module 406, which is used to map the m image feature matrices to corresponding text label matrices respectively through a preset method to generate an initial model; wherein the initial model indicates a mapping relationship between the m image feature matrices and the corresponding text label matrices;

[0182] The construction module 403 also includes: a processing module 407, which is used to optimize the initial model and determine the optimized initial model as a multi-label classification model.

[0183] In a possible design, the construction module 403 is further used to construct corresponding m adjacency graphs for the m image feature matrices respectively; wherein the nodes in the adjacency graph represent the image feature matrix, and the edges in the adjacency graph represent the similarity or proximity relationship between the nodes;

[0184] The mapping module 406 further includes: a capturing module 408 for capturing the local manifold structure of the m image feature matrices based on the adjacency graph;

[0185] The mapping module 406 is also used to map the m image feature matrices to the corresponding text label matrices respectively based on the local manifold structure through a preset method, so as to retain the local manifold structure of the corresponding image feature matrix in the text label matrix; wherein the local manifold structure represents the intrinsic geometric structure of the image features in the image feature matrix.

[0186] In a possible design, the processing module 407 further includes: a correction module 409, configured to add a Laplace regularization term to the initial model to correct the initial model;

[0187] The processing module 407 is further used to optimize the modified initial model based on a preset optimization strategy until the model converges to obtain a multi-label classification model.

[0188] In one possible design, the text labels in the text label matrix are in the form of binary vectors.

[0189] The construction module 403 further includes: an analysis module 410 for analyzing the image features included in each image feature matrix and the corresponding text label matrix;

[0190] The determination module 405 is further used to determine the power equipment defect indicated by the image features in the target image feature matrix, and determine the target defect label in the text label matrix corresponding to the target image feature matrix; wherein the target image feature matrix is ​​one of the m image feature matrices, and the target defect label indicates the power equipment defect;

[0191] The construction module 403 also includes: a setting module 411, which is used to set the binary vector corresponding to the target defect label to 1, and set the binary vectors corresponding to other text labels except the target defect label in the text label matrix corresponding to the target image matrix to 0.

[0192] The defect multi-label determination device for electric power equipment provided in the embodiment of the present application can be used to execute the defect multi-label determination method for electric power equipment in any of the above embodiments. Its implementation principle and technical effect are similar and will not be repeated here.

[0193] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software calling through processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software calling through processing elements, and some modules can be implemented in the form of hardware. In addition, all or part of these modules can be integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0194] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown, the electronic device may include: a transceiver 51 , a processor 52 , and a memory 53 .

[0195] The processor 52 executes the computer execution instructions stored in the memory, so that the processor 52 executes the scheme in the above embodiment. The processor 52 can be a general-purpose processor, including a central processing unit CPU, a network processor (NP), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0196] The memory 53 is connected to the processor 52 via a system bus and completes communication between them. The memory 53 is used to store computer program instructions.

[0197] The transceiver 51 may be used to communicate with other devices.

[0198] The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory.

[0199] The electronic device provided in the embodiments of the present application can be used to execute the method provided in any of the above embodiments. The implementation principles and technical effects are similar and will not be repeated here.

[0200] An embodiment of the present application further provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on a computer, the computer executes the method provided in any of the above embodiments.

[0201] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, the method provided in any of the above embodiments can be implemented.

[0202] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules 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 an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0203] The modules described as separate components may or may not be physically separated, and the components shown as modules 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 modules may be selected according to actual needs to implement the solution of this embodiment.

[0204] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The above-mentioned module-composed unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0205] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of various embodiments of the present application.

[0206] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0207] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0208] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0209] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0210] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic control unit or a main control device.

[0211] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-label defect determination method for power equipment, characterized in that: include: Acquire a device image of a target power device and a corresponding target text label; wherein the target power device represents a power device with defects, and the target text label is used to identify the defects of the target power device; Perform multi-view feature extraction on the device image to obtain image features under the multi-views; perform modeling based on the target text label and the image features under the multi-views by a preset method to generate a multi-label classification model; Acquire an image of an electric power device waiting to be labeled, input the image of the device waiting to be labeled into the multi-label classification model, and determine multiple defect labels output by the multi-label classification model; wherein the defect labels indicate defects existing in the electric power device waiting to be labeled.

2. The method according to claim 1, characterized in that The number of target power devices acquired is a, and the a target power devices include n device images; wherein a and n are both positive integers greater than or equal to 1, and n is greater than or equal to a; The extracting multi-view features of the device image to obtain image features under multiple views includes: Determine m pre-specified preset feature extraction methods; wherein the m pre-specified feature extraction methods correspond to m views, and m is a positive integer greater than or equal to 1; Building an image database based on the n device images, and performing multi-view feature extraction on the n device images in the image database respectively according to the m preset feature extraction methods; Determine the extracted image features under m views, wherein each view includes n image features.

3. The method according to claim 1, characterized in that The multi-view image features include image features under m views, where m is a positive integer greater than or equal to 1; The method of modeling based on the target text label and the image features under multiple views by a preset method to generate a multi-label classification model includes: Based on the image features in the m views, m image feature matrices are respectively constructed; wherein each image feature matrix includes n image features from n device images in the same view, and n is a positive integer greater than or equal to 1; Constructing a text label matrix based on the target text label; replicating the text label matrix m times according to the m image feature matrices to obtain m text label matrices; The m image feature matrices correspond to the m text label matrices one by one, and the corresponding text label matrices are set based on the m image feature matrices; The m image feature matrices and the m text label matrices are modeled by a preset method to generate a multi-label classification model.

4. The method according to claim 3, characterized in that The method of modeling the m image feature matrices and the m text label matrices by a preset method to generate a multi-label classification model includes: By a preset method, the m image feature matrices are respectively mapped to corresponding text label matrices to generate an initial model; wherein the initial model indicates a mapping relationship between the m image feature matrices and the corresponding text label matrices; The initial model is optimized and the optimized initial model is determined as a multi-label classification model.

5. The method according to claim 4, characterized in that The m image feature matrices are mapped to corresponding text label matrices respectively by a preset method, including: Constructing corresponding m adjacency graphs for the m image feature matrices respectively; wherein the nodes in the adjacency graph represent the image feature matrix, and the edges in the adjacency graph represent the similarity or proximity relationship between the nodes; Based on the adjacency graph, the local manifold structure of the m image feature matrices is captured; based on the local manifold structure, the m image feature matrices are respectively mapped to the corresponding text label matrices through the preset method to retain the local manifold structure of the corresponding image feature matrix in the text label matrix; wherein the local manifold structure characterizes the intrinsic geometric structure of the image features in the image feature matrix.

6. The method according to claim 4, characterized in that The step of optimizing the initial model and determining the optimized initial model as a multi-label classification model includes: Adding a Laplace regularization term to the initial model to modify the initial model; The modified initial model is optimized based on the preset optimization strategy until the model converges to obtain a multi-label classification model.

7. The method according to any one of claims 3 to 6, characterized in that The text labels in the text label matrix are in the form of binary vectors. The setting of the corresponding text label matrix based on the m image feature matrices includes: Analyze the image features included in each image feature matrix and the corresponding text label matrix; Determine the electric power equipment defect indicated by the image features in the target image feature matrix, and determine the target defect label in the text label matrix corresponding to the target image feature matrix; wherein the target image feature matrix is ​​one of the m image feature matrices, and the target defect label is a text label indicating the electric power equipment defect; The binary vector corresponding to the target defect label is set to 1, and the binary vectors corresponding to other text labels except the target defect label in the text label matrix corresponding to the target image matrix are set to 0.

8. A multi-label defect determination device for electric power equipment, characterized in that: include: An acquisition module, used to acquire an equipment image of a target electric power device and a corresponding target text label; wherein the target electric power device represents an electric power device with defects, and the target text label is used to identify the defects of the target electric power device; An extraction module, used for performing multi-view feature extraction on the device image to obtain image features under multiple views; A construction module is used to perform modeling based on the target text label and the image features under multiple views through a preset method to generate a multi-label classification model; The acquisition module is further used to acquire the image of the power equipment to be labeled; An input module is used to input the image of the device to be labeled into the multi-label classification model and determine multiple defect labels output by the multi-label classification model; wherein the defect label indicates the defects of the power equipment waiting to be labeled.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the defect multi-tag determination method for electric power equipment according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the defect multi-tag determination method for electric power equipment according to any one of claims 1 to 7.

Images