Equipment anomaly detection method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to an equipment anomaly detection method and device, computer equipment, a readable storage medium and a program product. The method comprises the following steps: carrying out feature extraction on an input substation equipment image to obtain initial image features, carrying out local information extraction on the initial image features through a feature enhancement module to obtain local features, carrying out global information extraction on the initial image features through the feature enhancement module to obtain global features, and splicing the local features and the global features to obtain a feature map, obtaining geometric structure information of the substation equipment according to the feature map, and carrying out anomaly detection on the substation equipment according to the geometric structure information to obtain a detection result. By adopting the method, substation equipment abnormity can be accurately detected.

Description

Technical Field

[0001] The present application relates to the field of target detection technology, and in particular to a method, apparatus, computer equipment, readable storage medium and program product for detecting device anomalies. Background Art

[0002] With the continuous advancement of smart grid and industrial automation technology, substations are the core components of the power system, and the stability and safety of their equipment operation are crucial to ensuring the normal operation of the entire power system. However, substations have a wide variety of equipment and complex structures, and the monitoring environment is often accompanied by a large number of complex backgrounds, equipment occlusions, and multi-directional target distribution. This brings great challenges to substation equipment defect detection.

[0003] Traditional substation monitoring methods mainly rely on manual observation or rule-based detection algorithms. However, due to the limitations of human subjective judgment and fatigue, manual observation detection efficiency is low and prone to errors; while rule-based detection algorithms are often difficult to adapt to complex and changeable monitoring scenarios, resulting in high false detection and missed detection rates. Summary of the invention

[0004] Based on this, it is necessary to provide an equipment anomaly detection method, device, computer equipment, readable storage medium and program product that can accurately detect abnormalities in substation equipment in response to the above technical problems.

[0005] In a first aspect, the present application provides a device anomaly detection method, which is applied to a target detection model; the target detection model includes a backbone network; the backbone network includes a feature enhancement module; including:

[0006] Extract features of the input substation equipment image to obtain initial image features;

[0007] The feature enhancement module is used to extract local information from the initial image features to obtain local features, and the feature enhancement module is used to extract global information from the initial image features to obtain global features;

[0008] The local features and global features are spliced ​​to obtain a feature map, and the geometric structure information of the substation equipment is obtained according to the feature map;

[0009] The substation equipment is detected for abnormalities according to the geometric structure information to obtain detection results; the detection results include abnormality location and abnormality type.

[0010] In one embodiment, the feature enhancement module includes a first convolution block and a second convolution block; the convolution kernels of the first convolution block and the second convolution block have different sizes; and the step of extracting local information of the initial image features through the feature enhancement module to obtain local features includes:

[0011] The initial image features are convolved by the first convolution block to obtain enhanced features; the first convolution block is used to increase the channel dimension of the initial image features;

[0012] The enhanced features are extracted through the second convolution block to obtain spatial features;

[0013] The spatial features are processed for channel restoration through the first convolution block, and the initial image features are residually connected with the spatial features after channel restoration;

[0014] The activation function is used to perform nonlinear processing on the features after residual connection to obtain local features.

[0015] In one embodiment, the feature enhancement module further includes a cross-covariance attention mechanism; and the step of extracting global information from the initial image features through the feature enhancement module to obtain global features includes:

[0016] The initial image features are processed through the cross-covariance attention mechanism, and the processed initial image features are convolved through the first convolution block to obtain the intermediate result features;

[0017] The intermediate result features are processed nonlinearly by the activation function, and the channel recovery processing is performed on the intermediate result features after the nonlinear processing by the first convolution block;

[0018] The intermediate result features after channel restoration are jump-connected with the initial image features to obtain the global features.

[0019] In one embodiment, the step of concatenating local features and global features to obtain a feature map includes:

[0020] Determine the splicing dimension, and perform dimension conversion processing on local features and global features respectively according to the splicing dimension;

[0021] The local features and global features after dimension conversion are concatenated to obtain a feature map.

[0022] In one embodiment, the step of obtaining geometric structure information of substation equipment according to the characteristic graph includes:

[0023] Perform pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain the corresponding horizontal direction features and vertical direction features;

[0024] According to the horizontal and vertical features, the local features and the global features are fused to obtain the fused features;

[0025] The geometric structure information of substation equipment is obtained based on the fused features.

[0026] In one embodiment, the step of fusing the local features and the global features to obtain the fused features according to the horizontal features and the vertical features includes:

[0027] Through the bidirectional gated recurrent unit, the horizontal and vertical features are processed respectively, and the processed horizontal and vertical features are fused to obtain geometric perception features;

[0028] The geometric perception features are subjected to horizontal and vertical convolution processing respectively to obtain fusion weights, and the local features and global features are fused according to the fusion weights to obtain fusion features.

[0029] In a second aspect, the present application also provides a device for detecting anomalies of a device, comprising:

[0030] A feature extraction module is used to extract features from the input substation equipment image to obtain initial image features;

[0031] An information extraction module is used to extract local information from the initial image features through the feature enhancement module to obtain local features, and to extract global information from the initial image features through the feature enhancement module to obtain global features;

[0032] The feature fusion module is used to combine local features and global features to obtain a feature map, and obtain the geometric structure information of the substation equipment based on the feature map;

[0033] The anomaly detection module is used to perform anomaly detection on substation equipment according to geometric structure information to obtain detection results; the detection results include anomaly location and anomaly type.

[0034] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method steps of any one of the first aspects are implemented.

[0035] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements any one of the method steps in the first aspect when the computer program is executed by a processor.

[0036] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which implements any one of the method steps in the first aspect when executed by a processor.

[0037] The above-mentioned equipment anomaly detection method, device, computer equipment, readable storage medium and program product avoid the limitations of a single perspective by adding a feature enhancement module to the target detection model. The feature enhancement module extracts features from both local and global levels and conducts comprehensive analysis. It can improve the target detection model's perception of feature geometric directions and effectively capture feature information in different directions, thereby more comprehensively and accurately capturing the characteristics of substation equipment and improving the accuracy of substation equipment anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 This is a diagram of an application environment based on a device anomaly detection method in an embodiment;

[0040] Figure 2 A schematic diagram of a process flow of a device abnormality detection method in one embodiment;

[0041] Figure 3 A schematic diagram of a flow chart of a device abnormality detection method in another embodiment;

[0042] Figure 4 A schematic diagram of the overall network structure of a target detection model in one embodiment;

[0043] Figure 5 is a schematic diagram of the structure of a feature enhancement module in an embodiment;

[0044] Figure 6 is a structural block diagram of a device abnormality detection device in one embodiment;

[0045] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] The device anomaly detection method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Among them, the terminal 102 is used to extract features from the input substation equipment image to obtain initial image features, extract local information from the initial image features through the feature enhancement module to obtain local features, extract global information from the initial image features through the feature enhancement module to obtain global features, splice the local features and the global features to obtain a feature map, obtain the geometric structure information of the substation equipment according to the feature map, and perform abnormal detection on the substation equipment according to the geometric structure information to obtain the detection result. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head mounted device may be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0048] In an exemplary embodiment, Figure 2 As shown, a device anomaly detection method is provided, which is applied to Figure 1 The terminal 102 in the example is used as an example to illustrate, including the following steps 202 to 208. Among them:

[0049] S202: Extract features from the input substation equipment image to obtain initial image features.

[0050] Optionally, the substation equipment image can be collected on-site by an imaging device, or can be acquired remotely through a network. The input substation equipment image is subjected to feature extraction by a target detection model to obtain initial image features, wherein the target detection model is a target detection network based on YOLOv5, wherein YOLOv5 is a convolutional neural network architecture based on deep learning, which can quickly and accurately locate and identify various target objects in the image.

[0051] Optionally, the main structure of the target detection model includes: backbone network (Backbone), neck (Neck) and head (Head), among which Backbone adopts efficient convolution operations such as deep separable convolution, which can effectively extract the feature information of the image while reducing the amount of calculation. For substation equipment images, it can well capture the key features such as the outline and texture of the equipment, providing a strong foundation for subsequent target detection. The Neck structure is responsible for fusing and optimizing the features extracted by the Backbone, integrating the features at different levels, so that it contains both rich detail information (i.e. local features) and high-level semantic information (i.e. global features). Head is used for the final target prediction, outputting the target category, location and other information. In the detection of substation equipment, it can accurately determine the type of equipment and its specific coordinate position in the image, so as to facilitate the judgment of whether it is abnormal and the related abnormal situation.

[0052] S204: extracting local information from the initial image features through a feature enhancement module to obtain local features, and extracting global information from the initial image features through a feature enhancement module to obtain global features.

[0053] Optionally, by introducing a feature enhancement module into the Backbone of the target detection model, the feature expression capability of the model can be improved. Specifically, the feature enhancement module usually adopts a Geometric Perception Feature Enhancement Module (GPFEM). GPFEM is an important module used to improve the feature extraction effect in the field of image processing and computer vision. It uses specific algorithms and convolution kernels to scan the input image or feature map to capture the geometric structure information, such as lines, angles, shapes, and the spatial position relationship of objects. For example, for substation equipment images, GPFEM can perceive the circular contour of the transformer, the straight line features of the transmission line, and the relative positions between different equipment.

[0054] Optionally, when extracting local information from the initial image features through the feature enhancement module, GPFEM will first divide the initial image features into multiple local areas. These areas can be divided according to the physical positions of different components in the substation equipment image or the pixel distribution law of the image itself, and then focus on these local areas through convolution kernels and algorithms that are sensitive to geometric structures to perceive subtle changes in geometric features and related texture, color and other features. After completing feature perception of each local area, GPFEM will integrate these scattered local features, and through appropriate feature fusion methods (such as splicing, weighted fusion, etc.), summarize the geometric perception features extracted from each local area and other related features to form a complete local feature set. This local feature set can comprehensively and meticulously reflect the status of each key local area in the substation equipment image.

[0055] Optionally, when extracting global information from the initial image features through the feature enhancement module, GPFEM will analyze the entire substation equipment image from a macro perspective, focusing on capturing large-scale features that can reflect the overall layout of the equipment, the relative position relationship between the main equipment, and the geometric structure of the entire scene. Afterwards, through a series of dimensionality reduction and abstraction operations, GPFEM integrates and refines the many perceived global geometric features and related image features to generate a representative global feature representation. This global feature covers the overall structural characteristics of the substation equipment image.

[0056] S206: local features and global features are combined to obtain a feature map, and geometric structure information of the substation equipment is obtained according to the feature map.

[0057] Optionally, the Neck part of the target detection model is responsible for fusing different levels of features from Backbone, and enhancing the network's detection capabilities for targets of different sizes by combining high-resolution (corresponding to detailed, more local features) and low-resolution (corresponding to more abstract, global features) features. By concatenating local features with global features, a more comprehensive, multi-dimensional feature description can be formed. In practical applications, local features and global features can be concatenated using a concatenation operation (Concat), which is a common feature concatenation operation in deep learning. Its main function is to merge multiple feature tensors along a specific dimension to form a new feature tensor containing more information. When processing substation equipment image features, features extracted from different angles (i.e., local and global) can be integrated together through the Concat operation.

[0058] Optionally, by performing horizontal and vertical feature enhancement on the feature graph, the directional perception capability of the model's geometric features can be improved, so that the model can accurately extract the geometric structure information contained in the feature graph from the horizontal and vertical spatial dimensions, and then obtain the geometric structure information of the substation equipment.

[0059] S208: Perform abnormality detection on the substation equipment according to the geometric structure information to obtain a detection result; the detection result includes the abnormality location and the abnormality type.

[0060] Optionally, target detection of substation equipment is performed based on geometric structure information, and abnormal parts in the equipment can be identified. For example, by comparing the geometric structure information with the normal geometric shape standard, the abnormality can be located on the specific target of the transformer according to the difference in geometric structure, and its specific coordinate position in the image can be determined, so as to achieve abnormal location. At the same time, based on the geometric structure information, the abnormality type can be further accurately determined and accurately classified. The final output detection result contains two key elements: abnormal location and abnormal type. Abnormal location clearly points out the specific location of the problem in the substation equipment, which is convenient for maintenance personnel to quickly find the fault point, while the abnormal type clarifies the nature of the fault, which helps maintenance personnel to prepare corresponding maintenance tools and formulate appropriate maintenance strategies in advance, so as to efficiently handle abnormal conditions of substation equipment and ensure the normal operation of the substation.

[0061] In the above-mentioned equipment anomaly detection method, a feature enhancement module is added to the target detection model. The feature enhancement module extracts features from both local and global levels and conducts comprehensive analysis, thereby avoiding the limitations of a single perspective. It can improve the target detection model's perception of feature geometric directions and effectively capture feature information in different directions, thereby more comprehensively and accurately capturing the characteristics of substation equipment and improving the accuracy of substation equipment anomaly detection.

[0062] In an exemplary embodiment, the feature enhancement module includes a first convolution block and a second convolution block; the sizes of the convolution kernels of the first convolution block and the second convolution block are different; the step of extracting local information from the initial image features through the feature enhancement module to obtain local features includes: performing convolution processing on the initial image features through the first convolution block to obtain enhanced features; the first convolution block is used to improve the channel dimension of the initial image features; performing feature extraction on the enhanced features through the second convolution block to obtain spatial features; performing channel restoration processing on the spatial features through the first convolution block, and performing residual connection between the initial image features and the spatial features after channel restoration; performing nonlinear processing on the features after residual connection through an activation function to obtain local features.

[0063] Optionally, the initial image features are convolved by the first convolution block, and the convolution kernel size is usually 1*1, which is used to increase the channel dimension of the initial image features. In image processing, a channel can be understood as a description of different aspects of image information. Improving the channel dimension means that the model can capture information in the image from more angles and methods, allowing the model to learn richer features, making the information carried by each pixel richer, and laying the foundation for subsequent more accurate feature extraction and processing.

[0064] Optionally, the enhanced features are extracted through a second convolution block with a different convolution kernel size from the first convolution block to obtain spatial features, wherein the convolution kernel size of the second convolution block may be 3*3, and a larger convolution kernel may capture a wider range of spatial information, while a smaller convolution kernel may focus on more detailed parts. The second convolution block can focus on extracting spatial structural information in the image, such as spatial features such as the positional relationship and shape of different components in the substation equipment, so that the model has a clearer understanding of the spatial layout of the image and can better understand the distribution and relationship of various parts of the substation equipment in space.

[0065] Furthermore, the spatial features are processed for channel restoration by using the first convolutional block again, and their channel dimensions are restored to a state that matches the initial image features. Then, the initial image features are residually connected with the spatial features after channel restoration. During the network training process, as the number of layers increases, problems such as gradient vanishing may occur, making the model difficult to train. Residual connection allows the model to directly learn the difference between input and output, that is, the residual, so that information can flow more smoothly in the network, avoiding performance degradation caused by too deep a network. At the same time, it also integrates the initial image features and the processed spatial features, retaining the basic information in the original image and the extracted spatial features.

[0066] Furthermore, the features after residual connection are processed nonlinearly through the activation function to obtain local features, where the role of the activation function is to introduce nonlinear factors into the model. Since the features in the substation equipment image are often nonlinear, it is difficult for the linear model to accurately describe them. The activation function can perform nonlinear transformation on the features after residual connection, so that the model can better fit the true distribution of the data and mine more complex feature relationships, thereby obtaining more representative local features. These local features can more accurately reflect the detailed information in the substation equipment image.

[0067] In this embodiment, by performing convolution processing on the initial image features through convolution blocks with different convolution kernel sizes, rich local features can be extracted from the substation equipment image, thereby improving the accuracy of substation equipment anomaly detection.

[0068] In an exemplary embodiment, the step of extracting global information from initial image features through a feature enhancement module to obtain global features includes: processing the initial image features through a cross-covariance attention mechanism, and convolving the processed initial image features through a first convolution block to obtain intermediate result features; performing nonlinear processing on the intermediate result features through an activation function, and performing channel restoration processing on the intermediate result features after nonlinear processing through the first convolution block; and performing jump connection between the intermediate result features after channel restoration and the initial image features to obtain global features.

[0069] Optionally, the cross-covariance attention mechanism is used to process the initial image features, so that the model can pay attention to the dependencies between different positions in the image and capture the global context information of the image. After that, the processed initial image features are convolved by the first convolution block to further extract and fuse features to obtain intermediate result features. The first convolution block can use the characteristics of its convolution kernel to perform more in-depth feature extraction on the processed features and mine more representative global feature information.

[0070] Furthermore, the intermediate result features are processed nonlinearly through activation functions, so that the model can better fit the complex feature relationships in the image. Since the global features in the substation equipment image often have complex nonlinear structures, the activation function can perform nonlinear transformations on the intermediate result features to enhance the model's ability to express these complex features. Afterwards, the first convolution block is used again to perform channel recovery processing on the intermediate result features after nonlinear processing, and the channel dimension is adjusted to a suitable state, so that the features can better fuse and interact with other features in subsequent processing. The intermediate result features after channel recovery are jump-connected with the initial image features to obtain global features, wherein the jump connection can directly introduce the original initial image features into the construction of global features, retaining the underlying basic information of the image and avoiding the loss of some key global information during the feature extraction process.

[0071] In this embodiment, the initial image features are processed through the cross-covariance attention mechanism, and the intermediate result features after channel recovery are jump-connected with the initial image features to obtain global features, so that the global features not only include the high-level semantic information extracted after cross-attention and convolution processing, but also integrate the underlying information of the initial image, making the global features richer and more comprehensive, and can more accurately reflect the overall structure and contextual relationship of the substation equipment image, thereby improving the accuracy of substation equipment anomaly detection.

[0072] In an exemplary embodiment, the step of splicing local features and global features to obtain a feature map includes: determining a splicing dimension, performing dimension conversion processing on the local features and the global features respectively according to the splicing dimension; and splicing the local features and the global features after the dimension conversion to obtain a feature map.

[0073] Optionally, before splicing local features and global features, it is necessary to first determine the splicing dimension, and perform dimension conversion processing on local features and global features respectively according to the determined splicing dimension to ensure that local features and global features have the same dimensional structure when splicing so that the splicing operation can be performed correctly. For example, if the channel dimension is selected for splicing, and the number of channels of local features and global features is different, it is necessary to adjust their channel dimensions to be consistent through convolution, reshaping and other processing to ensure that the features can still retain their original information and semantics after conversion. The local features and global features that have undergone dimension conversion processing are spliced ​​to obtain the final feature map. When splicing, the two features are merged along the determined splicing dimension, just like splicing two puzzle pieces together in a specific direction and position, forming a new feature map containing rich local and global information.

[0074] In this embodiment, by splicing local features and global features to obtain a feature map, the features of the substation equipment can be captured more comprehensively and accurately, thereby improving the accuracy of abnormality detection of the substation equipment.

[0075] In an exemplary embodiment, the step of obtaining geometric structure information of substation equipment according to a feature map includes: performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain corresponding horizontal direction features and vertical direction features; based on the horizontal direction features and the vertical direction features, fusing local features and global features to obtain fused features; and obtaining the geometric structure information of the substation equipment according to the fused features.

[0076] Optionally, by performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively, the corresponding horizontal features and vertical features are obtained, wherein the pooling operation can be understood as a way of information aggregation, which compresses the feature map in the corresponding direction by performing certain statistical operations on the rows or columns of the feature map, such as taking the maximum value, the average value, etc., to extract more representative features. When performing pooling processing in the horizontal direction, the features are aggregated along each row to obtain horizontal features that can represent the main features of the row; similarly, pooling processing in the vertical direction will obtain vertical features. Through pooling processing in the horizontal and vertical directions, the amount of data can be reduced while retaining key information, the computational complexity of the model can be reduced, and the main features of the feature map in different directions can be highlighted.

[0077] Optionally, local features and global features are fused based on the obtained horizontal and vertical features. The horizontal and vertical features provide the overall information of the feature map in the horizontal and vertical directions. Combining them with the local and global features can make the fused features contain both the local details and overall structural information of the equipment and the statistical characteristics of the feature map in different directions. The model can further analyze and process the fused features, such as using a specific convolutional layer or a fully connected layer to extract the geometric structure features, and map the fused features to the representation space of the geometric structure, thereby obtaining the geometric structure information of the substation equipment.

[0078] In this embodiment, by performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively, the corresponding horizontal direction features and vertical direction features are obtained, and the local features and the global features are fused to obtain the fused features. This can improve the model's perception of the geometric direction of the features, effectively capture the feature information in different directions, and thus accurately obtain the geometric structure information of the substation equipment.

[0079] In an exemplary embodiment, the step of fusing local features and global features to obtain fused features based on horizontal features and vertical features includes: processing the horizontal features and vertical features respectively through a bidirectional gated recurrent unit, and fusing the processed horizontal features and vertical features to obtain geometric perception features; performing horizontal convolution processing and vertical convolution processing on the geometric perception features respectively to obtain fusion weights, and fusing the local features and global features according to the fusion weights to obtain fused features.

[0080] Optionally, through the mechanism of gated recurrent unit (GRU), while considering the current feature information, the previously processed feature information is retained, and the gating mechanism is used to control the flow of information to determine which information needs to be retained and which needs to be updated, so as to effectively mine the long-term dependencies in the horizontal and vertical features and better capture the contextual information of the features. The horizontal and vertical features processed by the bidirectional gated recurrent unit are fused to obtain geometric perception features. By integrating the feature information in the horizontal and vertical directions, the geometric perception features contain both horizontal and vertical information, which can reflect the characteristics of the feature map in terms of geometric structure as a whole.

[0081] Furthermore, the geometric perception features are processed horizontally and vertically respectively, which is equivalent to further extracting and analyzing the geometric perception features from the horizontal and vertical directions. Through the convolution operation, the correlation information between the features at different positions can be extracted to obtain the fusion weight. The fusion weight will dynamically adjust the fusion ratio of local features and global features according to the features at different positions, and reasonably combine the two at each position to finally obtain the fusion feature, so that the fusion feature retains the detailed information of the local features and integrates the overall structural information of the global features.

[0082] In this embodiment, by fusing local features and global features according to horizontal features and vertical features to obtain fused features, the features of substation equipment can be described more comprehensively and accurately, thereby improving the comprehensiveness and accuracy of anomaly detection.

[0083] In an exemplary embodiment, Figure 3 As shown, a device anomaly detection method is provided, comprising:

[0084] (1) Build a target detection model for substation equipment anomalies based on YOLOv5.

[0085] Among them, the overall network structure of the target detection model is as follows Figure 4 As shown in Figure 1, the target detection model is built based on the YOLOv5 structure. By introducing the geometric perception feature enhancement module GPFEM into the C3 module of the Backbone part of the YOLOv5 structure, the feature expression ability of the model can be improved. Figure 4 As shown in the figure, the input image undergoes two standard convolution operations to perform preliminary extraction of image features. After that, each C3 module is replaced with a C3-GPFEM module to extract detail features and global context information respectively, and enhance the model's geometric structure perception ability. The SPPF module further compresses the feature map and aggregates multi-scale information to improve the network's expression of global features. Finally, after feature fusion and decoding operations in the Neck part, the multi-scale features are input into the detection head for target detection.

[0086] (2) The geometric perception feature enhancement module is used to extract local and global features in parallel, respectively obtaining detailed features and overall structural information, thereby enhancing the network's ability to perceive multi-scale features of targets in substation scenarios.

[0087] Among them, the structure of the feature enhancement module is as follows Figure 5 As shown, Figure 5 As shown, the input feature map F is divided into two branches to extract local and global information respectively.

[0088] For example, in the local feature extraction branch, the input feature F first uses 1*1 convolution to increase the channel dimension and increase the ability to express features. After increasing the number of channels, feature extraction can be more fully performed. After that, the feature map is subjected to 3*3 convolution for spatial feature extraction, and finally the number of channels is restored through 1*1 convolution. Each layer of convolution operation is followed by the ReLU activation function to maintain the ability of nonlinear feature extraction. After convolution, the feature will be updated and the local information will be retained, as shown in the following formula:

[0089]

[0090] in, , and They are all feature maps after convolution operation, representing the intermediate output results of each stage in the local feature extraction process. Conv1*1 means convolution with a convolution size of 1 and does not change the size of the feature map. Conv3*3 means convolution with a convolution size of 3 and does not change the size of the feature map.

[0091] The input feature F is connected to the feature output by the third convolution by residual connection to alleviate the gradient vanishing problem that may occur during deep network training. At the same time, the original input information can be retained. After that, the nonlinear representation ability of the feature is further improved through the ReLU activation function, and the local feature output is finally obtained. :

[0092]

[0093] in, ReLU represents the local features extracted after convolution and activation. ReLU is the rectified linear unit activation function, which performs nonlinear processing on the added feature map, removes the negative value part, and retains only the positive value, thereby enhancing the network's expressive ability.

[0094] For example, in the global feature extraction branch, the input feature F first passes through the cross-covariance attention mechanism XCA, and the feature map output by XCA is further extracted through a 1*1 convolution operation to extract the global feature information in space. After that, the feature map is normalized and nonlinearly processed through the ReLU activation function to ensure the stability of network training and increase the nonlinear feature expression ability. Finally, the feature is further processed through a 1*1 convolution operation to restore the number of channels and generate a global feature output. The specific process is shown in the following formula:

[0095]

[0096] Among them, F XCA is the feature map output by the cross-attention module, F G1 、F G2and F G3 It is the feature output after being processed by convolution and activation function in sequence.

[0097] Global Features It is added and fused with the input feature F through a skip connection to retain the input information and enhance the expression ability of the global feature. The formula is as follows:

[0098]

[0099] Furthermore, after obtaining the local features and global features, we use Concat to concatenate them to obtain .

[0100] (3) The feature map is pooled in the horizontal direction (i.e., X direction) and the vertical direction (i.e., Y direction) respectively to further extract the geometric structure information.

[0101] For example, the pooling operation compresses the information of the feature map in the spatial dimension, so as to better obtain the spatial features of each channel. The specific formula is as follows:

[0102]

[0103] in, From the feature map The features extracted by global average pooling in the X direction are Represents the value of the nth channel feature map at position (i, j), Represents the global average pooling function along the X axis, which compresses the feature map into a vector in the X direction. Finally, the shape of the feature map after pooling is compressed from C×H×W to C×H×1.

[0104] Similarly, for Y-direction pooling, the feature map Perform global average pooling along the Y direction. The formula is:

[0105]

[0106] in, From the feature map The features extracted by global average pooling in the Y direction are Represents the value of the nth channel feature map at position (i, j), Represents the global average pooling function along the Y axis, which compresses the feature map into a vector along the Y direction. Finally, the shape of the feature map after pooling is compressed from C×H×W to C×1×W.

[0107] Furthermore, for the pooled features in the X direction, the bidirectional gated recurrent unit (Bi-GRU) is used to capture the temporal dependency information in the X direction. Bi-GRU can use the forward and reverse GRU to simultaneously consider the temporal features in two directions. The forward and backward processing formulas of Bi-Gru are as follows:

[0108]

[0109] in, is the time step input of the X-direction pooling feature, represents the hidden state of the forward Gru at time step t, represents the hidden state of the backward Gru at time step t.

[0110] The final output is the sum of the forward and backward hidden states, as follows:

[0111]

[0112] Similarly, the features obtained by pooling in the Y direction , and Bi-Gru is also used to capture its timing dependency information. The formula is as follows:

[0113]

[0114] Finally, the features in the X and Y directions after Bi-Gru processing are fused to obtain the comprehensive geometric perception feature h: .

[0115] Furthermore, the feature h after Bi-Gru fusion is first subjected to two convolution operations in different directions. The first convolution uses a 1*11 convolution kernel to extract features in the horizontal direction, and the second convolution uses an 11*1 convolution kernel to extract features in the vertical direction. Through these two convolution operations in different directions, the model can capture different geometric structure information from the horizontal and vertical directions respectively, further enhancing the fusion expression of local and global information. The specific formula is:

[0116]

[0117] after, After the convolution and Sigmoid operations of the two branches, two weights are obtained to control the weighting of local and global features. Finally, the weighted local and global features are added together through the jump connection to obtain the final output. The specific formula is as follows:

[0118]

[0119] in, Represents the Sigmoid activation function, which is used to compress the output value into a specific range of [0, 1].

[0120] (4) Perform anomaly detection on substation equipment through target detection model.

[0121] Optionally, the image of the substation is input into a YOLOv5-based model. The model first extracts features through Backbone and improves the perception of geometric direction information through the geometric perception feature enhancement module GPFEM. After that, the feature map is passed to the Neck part, which is used to fuse the different levels of features from Backbone and combine high-resolution and low-resolution features to enhance the network's detection ability for targets of different sizes. Finally, the features processed by Neck are passed to the detection head, and the network performs anomaly detection and classification on the substation equipment in the image, and outputs the detection results, including anomaly location and anomaly category.

[0122] For example, 8705 original images containing substation anomalies were obtained by field shooting of substations and network acquisition, and the data set was divided into training set, validation set and test set with a ratio of 7:2:1. Due to the large number of substation defects, 14 common types were used as examples to make defect labels. In addition, the Stochastic Gradient Descent (SGD) optimizer was used to optimize the target detection model with a momentum coefficient of β = 0.937. Among them, the input image resolution is 640×640, the learning rate and batch size are set to 0.0001 and 16 respectively, the initialization learning rate is set to 0.01, the loss function is CIoU, and mAP50 and mAP50-95 are used as evaluation indicators. The target detection model is compared with the other five algorithms. Among them, the compared algorithms are mainly YOLOv5n, YOLOv5s, YOLOv7-tiny, YOLOv8n, and YOLOv10n. The experimental results are shown in Table 1:

[0123] Table 1

[0124]

[0125] In Table 1, mAP50 represents the average precision when the IoU threshold is 0.50, which is used to measure the detection accuracy. mAP50-95 combines the average precision under multiple IoU thresholds to more comprehensively evaluate the model performance. As can be seen from Table 1, the object detection model provided in the embodiment of the present application can effectively improve the detection accuracy and detection precision.

[0126] For example, in order to evaluate the effectiveness of the geometry-aware feature enhancement module, an ablation experiment was conducted on the YOLOv5-based framework and the geometry-aware feature enhancement module. The ablation experiment includes two experiments: (1) Base: Basic YOLOv5s framework. (2) Base+GPFEM: Based on the YOLOv5s basic framework, the geometry-aware feature enhancement module is added to the BottleNeck part of the C3 module of the backbone network. mAP50 and mAP50-95 are used as evaluation indicators. The experimental results are shown in Table 2:

[0127] Table 2

[0128]

[0129] It can be seen from Table 2 that in the YOLOv5s basic framework, after the BottleNeck part of the C3 module of the backbone network is added with the geometric perception feature enhancement module, the detection accuracy and detection precision of the model are improved.

[0130] In this embodiment, a feature enhancement module is added to the target detection model. The feature enhancement module extracts and comprehensively analyzes features from both local and global levels, thereby avoiding the limitations of a single perspective. The target detection model's perception of feature geometric directions can be improved, and feature information in different directions can be effectively captured, thereby more comprehensively and accurately capturing the characteristics of substation equipment and improving the accuracy of abnormal detection of substation equipment.

[0131] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0132] Based on the same inventive concept, the embodiment of the present application also provides a device anomaly detection device for implementing the device anomaly detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more device anomaly detection device embodiments provided below can refer to the limitations of the device anomaly detection method above, and will not be repeated here.

[0133] In an exemplary embodiment, Figure 6 As shown, a device anomaly detection device is provided, comprising: a feature extraction module 10, an information extraction module 20, a feature fusion module 30 and an anomaly detection module 40, wherein:

[0134] The feature extraction module 10 is used to extract features from the input substation equipment image to obtain initial image features.

[0135] The information extraction module 20 is used to extract local information from the initial image features through the feature enhancement module to obtain local features, and to extract global information from the initial image features through the feature enhancement module to obtain global features.

[0136] The feature fusion module 30 is used to combine local features and global features to obtain a feature map, and obtain geometric structure information of the substation equipment according to the feature map.

[0137] The anomaly detection module 40 is used to perform anomaly detection on the substation equipment according to the geometric structure information to obtain a detection result; the detection result includes anomaly location and anomaly type.

[0138] In an exemplary embodiment, the feature enhancement module includes a first convolution block and a second convolution block; the sizes of the convolution kernels of the first convolution block and the second convolution block are different; the information extraction module 20 is also used to perform convolution processing on the initial image features through the first convolution block to obtain enhanced features; the first convolution block is used to enhance the channel dimension of the initial image features; the enhanced features are extracted through the second convolution block to obtain spatial features; the spatial features are subjected to channel restoration processing through the first convolution block, and the initial image features are residually connected with the spatial features after channel restoration; the features after residual connection are nonlinearly processed through the activation function to obtain local features.

[0139] In an exemplary embodiment, the feature enhancement module also includes a cross-covariance attention mechanism; the information extraction module 20 is also used to process the initial image features through the cross-covariance attention mechanism, and convolve the processed initial image features through the first convolution block to obtain intermediate result features; perform nonlinear processing on the intermediate result features through an activation function, and perform channel restoration processing on the intermediate result features after nonlinear processing through the first convolution block; jump-connect the intermediate result features after channel restoration with the initial image features to obtain global features.

[0140] In an exemplary embodiment, the feature fusion module 30 is also used to determine the splicing dimension, and perform dimension conversion processing on the local features and the global features respectively according to the splicing dimension; the local features and the global features after the dimension conversion are spliced ​​to obtain a feature map.

[0141] In an exemplary embodiment, the feature fusion module 30 is also used to perform pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain corresponding horizontal direction features and vertical direction features; based on the horizontal direction features and the vertical direction features, the local features and the global features are fused to obtain fused features; and the geometric structure information of the substation equipment is obtained based on the fused features.

[0142] In an exemplary embodiment, the anomaly detection module 40 is also used to process horizontal directional features and vertical directional features respectively through a bidirectional gated recurrent unit, and fuse the processed horizontal directional features and vertical directional features to obtain geometric perception features; perform horizontal convolution processing and vertical convolution processing on the geometric perception features respectively to obtain fusion weights, and fuse local features and global features according to the fusion weights to obtain fusion features.

[0143] Each module in the above-mentioned device based on abnormality detection can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module above.

[0144] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 7As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a method for detecting device abnormalities is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0145] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0146] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the following steps when executing the computer program: extracting features of an input substation equipment image to obtain initial image features; extracting local information of the initial image features through a feature enhancement module to obtain local features, and extracting global information of the initial image features through the feature enhancement module to obtain global features; splicing the local features and the global features to obtain a feature map, and obtaining geometric structure information of the substation equipment based on the feature map; performing anomaly detection on the substation equipment based on the geometric structure information to obtain a detection result; the detection result includes anomaly location and anomaly type.

[0147] In one embodiment, the feature enhancement module includes a first convolution block and a second convolution block; the sizes of the convolution kernels of the first convolution block and the second convolution block are different; when the processor executes the computer program, the feature enhancement module is used to extract local information of the initial image features to obtain local features, including: convolution processing of the initial image features by the first convolution block to obtain enhanced features; the first convolution block is used to improve the channel dimension of the initial image features; feature extraction of the enhanced features by the second convolution block to obtain spatial features; channel restoration processing of the spatial features by the first convolution block, and residual connection of the initial image features with the spatial features after channel restoration; nonlinear processing of the residually connected features by an activation function to obtain local features.

[0148] In one embodiment, when a processor executes a computer program, it involves extracting global information from initial image features through a feature enhancement module to obtain global features, including: processing the initial image features through a cross-covariance attention mechanism, and convolving the processed initial image features through a first convolution block to obtain intermediate result features; performing nonlinear processing on the intermediate result features through an activation function, and performing channel recovery processing on the intermediate result features after nonlinear processing through the first convolution block; and jump-connecting the intermediate result features after channel recovery with the initial image features to obtain global features.

[0149] In one embodiment, the processor executes a computer program that involves splicing local features and global features to obtain a feature map, including: determining a splicing dimension, and performing dimension conversion processing on the local features and the global features respectively according to the splicing dimension; splicing the local features and the global features after the dimension conversion to obtain a feature map.

[0150] In one embodiment, the method of obtaining geometric structure information of substation equipment based on a feature map when a processor executes a computer program includes: performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain corresponding horizontal direction features and vertical direction features; based on the horizontal direction features and the vertical direction features, fusing local features and global features to obtain fused features; and obtaining the geometric structure information of the substation equipment based on the fused features.

[0151] In one embodiment, when a processor executes a computer program, the processor performs a fusion process on local features and global features according to horizontal features and vertical features to obtain fused features, including: processing horizontal features and vertical features respectively through a bidirectional gated recurrent unit, and fusing the processed horizontal features and vertical features to obtain geometric perception features; performing horizontal convolution processing and vertical convolution processing on the geometric perception features respectively to obtain fusion weights, and fusing local features and global features according to the fusion weights to obtain fused features.

[0152] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: feature extraction is performed on an input substation equipment image to obtain initial image features; local information extraction is performed on the initial image features through a feature enhancement module to obtain local features, and global information extraction is performed on the initial image features through the feature enhancement module to obtain global features; local features and global features are spliced ​​to obtain a feature map, and geometric structure information of the substation equipment is obtained according to the feature map; anomaly detection is performed on the substation equipment according to the geometric structure information to obtain a detection result; the detection result includes anomaly location and anomaly type.

[0153] In one embodiment, the feature enhancement module includes a first convolution block and a second convolution block; the sizes of the convolution kernels of the first convolution block and the second convolution block are different; when the computer program is executed by the processor, the feature enhancement module is used to extract local information of the initial image features to obtain local features, including: convolution processing of the initial image features by the first convolution block to obtain enhanced features; the first convolution block is used to improve the channel dimension of the initial image features; feature extraction of the enhanced features by the second convolution block to obtain spatial features; channel restoration processing of the spatial features by the first convolution block, and residual connection of the initial image features with the spatial features after channel restoration; nonlinear processing of the residually connected features by an activation function to obtain local features.

[0154] In one embodiment, the feature enhancement module also includes a cross-covariance attention mechanism; when the computer program is executed by the processor, the feature enhancement module is used to extract global information from the initial image features to obtain global features, including: processing the initial image features through the cross-covariance attention mechanism, and convolving the processed initial image features through the first convolution block to obtain intermediate result features; performing nonlinear processing on the intermediate result features through an activation function, and performing channel recovery processing on the intermediate result features after nonlinear processing through the first convolution block; and jump-connecting the intermediate result features after channel recovery with the initial image features to obtain global features.

[0155] In one embodiment, when a computer program is executed by a processor, the steps involved in splicing local features and global features to obtain a feature map include: determining a splicing dimension, and performing dimension conversion processing on the local features and the global features respectively according to the splicing dimension; and splicing the local features and the global features after the dimension conversion to obtain a feature map.

[0156] In one embodiment, the computer program executed by the processor involves obtaining geometric structure information of substation equipment based on the feature map, including: performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain corresponding horizontal direction features and vertical direction features; based on the horizontal direction features and the vertical direction features, fusing the local features and the global features to obtain the fused features; and obtaining the geometric structure information of the substation equipment based on the fused features.

[0157] In one embodiment, when a computer program is executed by a processor, the computer program involves fusing local features and global features according to horizontal features and vertical features to obtain fused features, including: processing horizontal features and vertical features respectively through a bidirectional gated recurrent unit, and fusing the processed horizontal features and vertical features to obtain geometric perception features; performing horizontal convolution processing and vertical convolution processing on the geometric perception features respectively to obtain fusion weights, and fusing local features and global features according to the fusion weights to obtain fused features.

[0158] In one embodiment, a computer program product is provided, including a computer program, which implements the following steps when executed by a processor: extracting features from an input substation equipment image to obtain initial image features; extracting local information from the initial image features through a feature enhancement module to obtain local features, and extracting global information from the initial image features through the feature enhancement module to obtain global features; splicing the local features and the global features to obtain a feature map, and acquiring geometric structure information of the substation equipment based on the feature map; performing anomaly detection on the substation equipment based on the geometric structure information to obtain a detection result; the detection result includes anomaly location and anomaly type.

[0159] In one embodiment, the feature enhancement module includes a first convolution block and a second convolution block; the sizes of the convolution kernels of the first convolution block and the second convolution block are different; when the computer program is executed by the processor, the feature enhancement module is used to extract local information of the initial image features to obtain local features, including: convolution processing of the initial image features by the first convolution block to obtain enhanced features; the first convolution block is used to improve the channel dimension of the initial image features; feature extraction of the enhanced features by the second convolution block to obtain spatial features; channel restoration processing of the spatial features by the first convolution block, and residual connection of the initial image features with the spatial features after channel restoration; nonlinear processing of the residually connected features by an activation function to obtain local features.

[0160] In one embodiment, the feature enhancement module also includes a cross-covariance attention mechanism; when the computer program is executed by the processor, the feature enhancement module is used to extract global information from the initial image features to obtain global features, including: processing the initial image features through the cross-covariance attention mechanism, and convolving the processed initial image features through the first convolution block to obtain intermediate result features; performing nonlinear processing on the intermediate result features through an activation function, and performing channel recovery processing on the intermediate result features after nonlinear processing through the first convolution block; and jump-connecting the intermediate result features after channel recovery with the initial image features to obtain global features.

[0161] In one embodiment, when a computer program is executed by a processor, the steps involved in splicing local features and global features to obtain a feature map include: determining a splicing dimension, and performing dimension conversion processing on the local features and the global features respectively according to the splicing dimension; and splicing the local features and the global features after the dimension conversion to obtain a feature map.

[0162] In one embodiment, the computer program executed by the processor involves obtaining geometric structure information of substation equipment based on the feature map, including: performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain corresponding horizontal direction features and vertical direction features; based on the horizontal direction features and the vertical direction features, fusing the local features and the global features to obtain the fused features; and obtaining the geometric structure information of the substation equipment based on the fused features.

[0163] In one embodiment, when a computer program is executed by a processor, the computer program involves fusing local features and global features according to horizontal features and vertical features to obtain fused features, including: processing horizontal features and vertical features respectively through a bidirectional gated recurrent unit, and fusing the processed horizontal features and vertical features to obtain geometric perception features; performing horizontal convolution processing and vertical convolution processing on the geometric perception features respectively to obtain fusion weights, and fusing local features and global features according to the fusion weights to obtain fused features.

[0164] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0165] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0166] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A device anomaly detection method, characterized in that: Applied to target detection models; The target detection model includes a backbone network; the backbone network includes a feature enhancement module; the method includes: Extract features of the input substation equipment image to obtain initial image features; Extracting local information from the initial image features through the feature enhancement module to obtain local features, and extracting global information from the initial image features through the feature enhancement module to obtain global features; The local features and the global features are combined to obtain a feature map, and the geometric structure information of the substation equipment is obtained according to the feature map; Anomaly detection is performed on the substation equipment according to the geometric structure information to obtain a detection result; the detection result includes anomaly location and anomaly type.

2. The method according to claim 1, characterized in that The feature enhancement module includes a first convolution block and a second convolution block; the convolution kernels of the first convolution block and the second convolution block have different sizes; the local information of the initial image features is extracted by the feature enhancement module to obtain the local features, including: The initial image feature is convolved by the first convolution block to obtain an enhanced feature; the first convolution block is used to increase the channel dimension of the initial image feature; Extracting the enhanced features through the second convolution block to obtain spatial features; Performing channel restoration processing on the spatial features through the first convolution block, and performing residual connection between the initial image features and the spatial features after channel restoration; The activation function is used to perform nonlinear processing on the features after residual connection to obtain local features.

3. The method according to claim 2, characterized in that The extracting global information of the initial image features by the feature enhancement module to obtain global features includes: Processing the initial image features through a cross-covariance attention mechanism, and performing convolution processing on the processed initial image features through the first convolution block to obtain intermediate result features; Performing nonlinear processing on the intermediate result features through an activation function, and performing channel recovery processing on the intermediate result features after the nonlinear processing through the first convolution block; The intermediate result features after channel restoration are jump-connected with the initial image features to obtain global features.

4. The method according to claim 1, characterized in that: The step of combining the local features and the global features to obtain a feature map includes: Determine a splicing dimension, and perform dimension conversion processing on the local features and the global features respectively according to the splicing dimension; The local features and global features after dimension conversion are concatenated to obtain a feature map.

5. The method according to claim 1, characterized in that The obtaining geometric structure information of the substation equipment according to the characteristic graph includes: Performing pooling processing on the feature map in the horizontal direction and the vertical direction respectively to obtain corresponding horizontal direction features and vertical direction features; According to the horizontal direction feature and the vertical direction feature, the local feature and the global feature are fused to obtain a fused feature; The geometric structure information of the substation equipment is obtained according to the fusion feature.

6. The method according to claim 5, characterized in that The fusing the local feature and the global feature to obtain a fused feature according to the horizontal direction feature and the vertical direction feature includes: The horizontal direction feature and the vertical direction feature are processed respectively by a bidirectional gated recurrent unit, and the processed horizontal direction feature and vertical direction feature are fused to obtain a geometric perception feature; The geometric perception features are respectively subjected to horizontal convolution processing and vertical convolution processing to obtain fusion weights, and the local features and the global features are fused according to the fusion weights to obtain fusion features.

7. A device for detecting abnormality of equipment, characterized in that: The device comprises: A feature extraction module is used to extract features from the input substation equipment image to obtain initial image features; An information extraction module, configured to extract local information from the initial image features through a feature enhancement module to obtain local features, and to extract global information from the initial image features through the feature enhancement module to obtain global features; A feature fusion module, used for splicing the local features and the global features to obtain a feature map, and obtaining geometric structure information of the substation equipment according to the feature map; The anomaly detection module is used to perform anomaly detection on the substation equipment according to the geometric structure information to obtain a detection result; the detection result includes anomaly location and anomaly type.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, 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 6 are implemented.

10. A computer program product, comprising 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 6 are implemented.

Images