Small-target lightweight substation equipment defect detection method and system

Abstract

The invention belongs to the technical field of image processing, and provides a small-target lightweight substation equipment defect detection method and system, and the method comprises the steps: obtaining a substation equipment image; according to the acquired image and a preset small target lightweight defect detection model, completing defect detection of the substation equipment; wherein the preset small target lightweight defect detection model adopts a YOLOv8 target detection network which adopts an Officient Former V2 and a separation convolution attention mechanism as a backbone network, adopts scale time sequence feature fusion as a neck network and adopts a prediction head network of a shared convolution lightweight detection head, and image features are extracted based on the backbone network; obtaining feature maps of different scales; the neck network performs feature fusion on the feature maps with different scales to obtain a multi-scale feature map; and detecting the multi-scale feature map through the shared convolutional lightweight detection head, and performing defect positioning of the equipment image to complete defect detection of the substation equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a small-target lightweight substation equipment defect detection method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Substations are an important part of the power system. In daily operation and maintenance, substation equipment is complex and numerous, and manual operation and maintenance are stressful. The operating status of its equipment directly determines the effective operation of the power system. Traditional substation inspections rely on regular personnel inspections, which are inefficient and costly. For larger substations, personnel inspections may take more than two hours, and many traditional visual monitoring devices have problems with false detection and missed detection in terms of warning safety hazards and tracking sudden failures. With the rapid development of the power industry, traditional inspection methods can no longer meet the operation and maintenance and safety management needs of substations. Intelligent inspection methods are gradually becoming an inevitable trend in the development of substations and even the entire power system.

[0004] Conducting research on substation defect detection technology is an important topic in the construction of a new generation of smart substations. At present, defect detection mainly uses target detection algorithms to achieve functions such as substation equipment status monitoring, station-side meter defect identification, and personnel safety assessment. Therefore, substation defect detection technology based on artificial intelligence neural network algorithms has great research prospects.

[0005] However, according to existing research, most substation defect detection algorithms focus on improving accuracy without taking into account the characteristics of substation scenarios. There are still several challenges in using deep learning methods for substation defect detection:

[0006] 1) In the substation defect detection scenario, data is mostly collected through cameras, drones and other equipment. Due to the shooting angle, viewing angle and other factors, targets of different sizes will be produced, including smaller equipment targets. This will cause some defects to exist in such small target equipment. However, small target defect detection is not considered in existing algorithms.

[0007] 2) Considering the limitations of computing resources in actual substations, the model size should not be too large and the computing cost should not be too high. The existing model can be optimized and redesigned for better deployment on embeddable devices. Most existing studies only optimize detection speed or detection accuracy, without comprehensively considering the balance between accuracy and detection speed, and it is difficult to meet the engineering needs of actual substation inspections.

[0008] 3) Each model has a single target type for substation identification and positioning, and there is a lack of public data sets containing more types of substation equipment defects, making it difficult to detect multiple categories of defects on the same model. Summary of the invention

[0009] To solve the above problems, the present invention proposes a small target lightweight substation equipment defect detection method and system. Through the YOLOv8 target detection framework with Efficient Former V2 as the main network, the small target feature layer is integrated into the neck network as a separate branch, and a lightweight shared convolution detection head is designed to improve the small target positioning accuracy and lightweight of the model.

[0010] According to some embodiments, a first solution of the present invention provides a small target lightweight substation equipment defect detection method, which adopts the following technical solution:

[0011] A small target lightweight substation equipment defect detection method, comprising:

[0012] Obtain images of substation equipment;

[0013] Complete defect detection of substation equipment based on the acquired images and the preset small target lightweight defect detection model;

[0014] Among them, the preset small target lightweight defect detection model adopts the YOLOv8 target detection network, and the YOLOv8 target detection network includes a backbone network using Efficient Former V2 and a separated convolutional attention mechanism, a neck network using scale-time series feature fusion, and a prediction head network using a shared convolutional lightweight detection head; based on the backbone network, image features of the acquired image are extracted to obtain feature maps of different scales; the neck network performs feature fusion on the obtained feature maps of different scales to obtain a multi-scale feature map; the obtained multi-scale feature map is detected by a shared convolutional lightweight detection head to locate the defects of the equipment image and complete the defect detection of the substation equipment.

[0015] As a further technical limitation, the backbone network includes a local module, a local-global module, a downsampling module and a spatial pyramid pooling-large kernel separation convolution attention module; wherein the local-global module includes a lightweight local-global module mainly for lightweighting and a ViT module local-global module mainly for capturing global information; the local-global module obtains global information through a multi-head self-attention mechanism and outputs it in a fixed size through a linear layer.

[0016] Furthermore, the lightweight local-global module reduces the number of parameters of the pure ViT architecture by downsampling all linear layer output features to a uniform resolution size, and then inserting the original resolution into the next layer through the attention layer.

[0017] Furthermore, the downsampling module class captures dynamic and static feature information through pooling and convolution to expand the receptive field; the spatial pyramid pooling-large kernel convolution separation attention module uses 3×3 depth-separable convolution and [7 / 2]×[7 / 2] depth-separable convolution to replace 7×7 large kernel convolution to extract local and global information and reduce the number of parameters and computation of the model.

[0018] As a further technical limitation, the neck network includes a scale sequence feature fusion module and a zoom stitching module, which are used to fuse feature maps of different scales obtained by the backbone network to achieve information interaction of different scales.

[0019] As a further technical limitation, the prediction head network performs classification and regression tasks through multi-scale feature maps and shared convolution lightweight detection heads, completes target positioning tasks, and locates defects in device images.

[0020] According to some embodiments, a second solution of the present invention provides a small target lightweight substation equipment defect detection system, which adopts the following technical solution:

[0021] A small target lightweight substation equipment defect detection system, comprising:

[0022] An acquisition module, configured to acquire an image of substation equipment;

[0023] A detection module, which is configured to complete defect detection of substation equipment according to the acquired image and a preset small target lightweight defect detection model;

[0024] Among them, the preset small target lightweight defect detection model adopts the YOLOv8 target detection network, and the YOLOv8 target detection network includes a backbone network using Efficient Former V2 and a separated convolutional attention mechanism, a neck network using scale-time series feature fusion, and a prediction head network using a shared convolutional lightweight detection head; based on the backbone network, image features of the acquired image are extracted to obtain feature maps of different scales; the neck network performs feature fusion on the obtained feature maps of different scales to obtain a multi-scale feature map; the obtained multi-scale feature map is detected by a shared convolutional lightweight detection head to locate the defects of the equipment image and complete the defect detection of the substation equipment.

[0025] According to some embodiments, a third solution of the present invention provides a computer-readable storage medium, which adopts the following technical solution:

[0026] A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps in a small-target lightweight substation equipment defect detection method as described in the first solution of the present invention.

[0027] According to some embodiments, a fourth solution of the present invention provides an electronic device, which adopts the following technical solution:

[0028] An electronic device comprises a memory, a processor and a program stored in the memory and running on the processor, wherein when the processor executes the program, the steps in a small-target lightweight substation equipment defect detection method as described in the first scheme of the present invention are implemented.

[0029] According to some embodiments, a fifth solution of the present invention provides a computer program product, which adopts the following technical solution:

[0030] A computer program product includes software code, wherein the program in the software code executes the steps in a small-target lightweight substation equipment defect detection method as described in the first embodiment of the present invention.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention constructs a substation defect image dataset, takes YOLOv8 as the basic detection model, adopts EfficientFormer V2 as the backbone network, introduces a large-scale separation convolution attention mechanism on the basis of the spatial pooling pyramid layer, and improves the feature extraction ability of the model, which can reduce the amount of model calculation and improve the interaction of global features; for the detection problem of small target defect detection, the small target feature layer is integrated into the neck network as a separate branch to reduce the number of model parameters and the amount of calculation; for the problem of too many model parameters, a lightweight shared convolution detection head is proposed, the idea of ​​shared convolution is adopted, the feature output is predicted through the same detection head, and the shared convolution module is optimized by group normalization to reduce the complexity of the model; the problems of low small target positioning accuracy and excessive model parameters are solved, and the purpose of improving detection accuracy under the premise of meeting real-time detection is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings in the specification that constitute a part of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments of this embodiment and their descriptions are used to explain this embodiment and do not constitute improper limitations on this embodiment.

[0034] Figure 1 This is a flow chart of a small target lightweight substation equipment defect detection method in Embodiment 1 of the present invention;

[0035] Figure 2This is a schematic diagram of the structure of a small target lightweight defect detection model in Embodiment 1 of the present invention;

[0036] Figure 3 This is a schematic diagram of the network structure of the backbone network Efficient Former V2 in the first embodiment of the present invention;

[0037] Figure 4 Schematic diagram of the structure of the scale sequence feature fusion module in the first embodiment of the present invention;

[0038] Figure 5 Schematic diagram of the structure of the zoom splicing module in the first embodiment of the present invention;

[0039] Figure 6 Schematic diagram of batch normalization and group normalization structure in Embodiment 1 of the present invention;

[0040] Figure 7 This is a rendering of the substation defect detection in the first embodiment of the present invention;

[0041] Figure 8 This is a structural block diagram of a small-target lightweight substation equipment defect detection system in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0043] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0044] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0045] In the present invention, terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are relational words determined only for the convenience of describing the structural relationships of the various parts or elements of the present invention, and do not specifically refer to any part or element in the present invention and should not be understood as limitations on the present invention.

[0046] In the present invention, terms such as "fixed connection", "connected", "connection", etc. should be understood in a broad sense, indicating that it can be fixedly connected, integrally connected or detachably connected; it can be directly connected or indirectly connected through an intermediate medium. For relevant scientific research or technical personnel in this field, the specific meanings of the above terms in the present invention can be determined according to specific circumstances, and they cannot be understood as limitations on the present invention.

[0047] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0048] Embodiment 1

[0049] Embodiment 1 of the present invention introduces a small-target lightweight substation equipment defect detection method.

[0050] like Figure 1 A small target lightweight substation equipment defect detection method is shown, comprising:

[0051] Obtain images of substation equipment;

[0052] Complete defect detection of substation equipment based on the acquired images and the preset small target lightweight defect detection model;

[0053] Among them, the preset small target lightweight defect detection model adopts the YOLOv8 target detection network, and the YOLOv8 target detection network includes a backbone network using Efficient Former V2 and a separated convolutional attention mechanism, a neck network using scale-time series feature fusion, and a prediction head network using a shared convolutional lightweight detection head; based on the backbone network, image features of the acquired image are extracted to obtain feature maps of different scales; the neck network performs feature fusion on the obtained feature maps of different scales to obtain a multi-scale feature map; the obtained multi-scale feature map is detected by a shared convolutional lightweight detection head to locate the defects of the equipment image and complete the defect detection of the substation equipment.

[0054] This embodiment adopts Figure 2The small target lightweight defect detection model shown in the figure selects YOLOv8 as the basic architecture and adopts the Efficient Former V2 model as the backbone network; and on the basis of the spatial pooling pyramid layer, a large separation convolution attention mechanism is introduced to decompose the two-dimensional convolution kernel of the channel-by-channel convolution layer into cascaded horizontal and vertical one-dimensional convolution kernels, increase the size of the convolution kernel, thereby reducing the computational complexity and memory usage, while enhancing the long-distance dependency of the input image and improving the representation ability of the model; the neck network fuses the feature maps of different scales obtained by the backbone network to achieve information interaction of different scales; the multi-scale feature maps generated by the neck network perform classification and regression tasks by sharing the convolution lightweight detection head, thereby completing the target positioning task.

[0055] The YOLOv8 selected in this embodiment as the basic architecture includes three parts: Efficient FormerV2 is used as the backbone network for extracting image features; the scale-time feature fusion module and the zoom stitching module are used as the neck network for balancing the upper and lower layer information flow of the feature map extracted by the backbone network; the shared convolution and group normalization method are used as the prediction head network for regression coordinate position and classification.

[0056] This embodiment adopts Figure 3 In the backbone network shown in the figure, the local module introduces a 3×3 depth-separable convolution as an intermediate layer, which realizes the functions of pooling and convolution, while also capturing local information with a small performance overhead.

[0057] The local global module designed based on the ViT (neural network architecture based on self-attention mechanism, mainly used to process image data) architecture is divided into a lightweight local global (Light-weight Local Global) and a ViT module that mainly captures global information. The local global module obtains global information through the multi-head self-attention (Multi-Head Self-Attention) mechanism and then outputs it in a fixed size through a linear layer. The calculation formula for multi-head self-attention is:

[0058] MHSA(Q,K,V)=Softmax(Q·K T +ab)·V;

[0059] Among them, Q (Query), K (Key), V (Value) are obtained from the input features through the linear layer, and ab is a learnable attention bias for position encoding;

[0060] To overcome the problem that the architecture of inserting ViT only in the later layers of the network cannot take into account the feature outputs of different sizes, Lightweight Local Global downsamples all Q, K, and V to a uniform resolution size, and then inserts the original resolution into the next layer through the attention layer, reducing the number of parameters of the pure ViT architecture while improving model performance.

[0061] The downsampling module is similar to a lightweight local-global architecture. It adds the residual idea to the overall structure and captures dynamic and static feature information through pooling and convolution. It can bring a larger receptive field and effectively improve the model performance in the face of complex scene information in substations.

[0062] The spatial pyramid pooling-large kernel convolution separation attention module uses 3×3 depth-wise separable convolution and [7 / 2]×[7 / 2] depth-wise separable convolution to replace 7×7 large kernel convolution to extract local and global information, reducing the number of model parameters and computational complexity.

[0063] The neck network fuses the feature maps of different scales obtained by the backbone network to achieve information interaction of different scales, including the Scale Sequence Feature Fusion module and the Zoom Concatenate module.

[0064] This embodiment adopts Figure 4 The scale sequence feature fusion module shown in the figure aims to fuse the features of the backbone output. It first scales the features of different scales through Unsqueeze and Unsample operations, then stacks the features as the input of the 3D convolutional layer, and finally outputs them through the normalization layer.

[0065] This embodiment adopts Figure 5 The zoom stitching module structure shown is designed to learn the information of features of different scales in more detail. First, the larger scale features are processed by the convolution module, then downsampled by the pooling layer, and finally output by the convolution module. At this time, the large-scale features will be processed to be equivalent to the feature scale of the intermediate size. The feature scale of the intermediate size is only processed by the convolution module, while the small-scale features are upsampled by the nearest neighbor interpolation method after passing through the convolution module, so that their feature scale is equivalent to the feature scale of the intermediate size, and finally the three are spliced.

[0066] The neck network is based on the ASF-Neck network, and the P2 feature output of the backbone network is directly introduced into the Concat module. At the same time, the neck network including the scale sequence feature fusion module is used to further fuse the fused P3, P4, and P5 features with P2. Considering the characteristics of the zoom splicing module, if the P2 layer is directly introduced at the front end of the feature fusion, the super large feature map preprocessed by the Stem module will directly participate in the feature fusion, and the amount of parameters brought is huge. Therefore, P2 is placed at the back end of the feature fusion and only the scale sequence feature fusion module is used for fusion without considering zoom splicing. Such consideration is beneficial to the lightweight model.

[0067] The multi-scale feature map generated by the neck network performs classification and regression tasks through a shared convolutional lightweight detection head, thereby completing the target positioning task, including:

[0068] This embodiment adopts Figure 6 The batch normalization and group normalization structure, the shared convolution lightweight detection head adopts a very simple structural design to avoid the increase of model parameters due to complex architecture. At the same time, the batch normalization in the convolution module is replaced by the more performant group normalization. Batch normalization uses normalization along the batch size to enable various networks to be trained, but when the batch size is too small, its error will increase significantly. Group normalization can avoid the influence of different batch sizes on its accuracy by grouping channels.

[0069] The specific formula for group normalization is:

[0070]

[0071] Among them, x represents the feature input, for the input image, i is a four-dimensional vector, that is, i = (i N ,i C ,i H ,i W ), where N represents batch, C represents the number of channels, and H and W represent the height and width of the feature.

[0072] μ and σ represent the mean and standard deviation, namely:

[0073]

[0074] Here, ω is a small constant to prevent the denominator in (2) from being zero. i Represents a pixel set, and m represents the number of pixels.

[0075] For batch normalization, the pixel set depends on C, that is:

[0076] S i ={k|k C =i C};

[0077] Batch normalization normalizes each channel and calculates the mean and standard deviation along (N, H, W). For group normalization, the pixel set depends on C and is defined as:

[0078] S i ={k|k N =i N}.

[0079] Unlike batch normalization, group normalization calculates the mean and standard deviation along (C, H, W), which normalizes each batch and directly avoids the influence of batch size.

[0080] The detection effect in this embodiment is as follows Figure 7 As shown in the figure, YOLOv8 is used as the basic detection model, EfficientFormer V2 is used as the backbone network, and a large-scale separation convolution attention mechanism is introduced on the basis of the spatial pooling pyramid layer to improve the feature extraction ability of the model, which can reduce the amount of model calculation and improve the interaction of global features. A neck module for small target defect detection is proposed, and the small target feature layer is integrated into the neck network as a separate branch to reduce the number of model parameters and the amount of calculation. A lightweight shared convolution detection head is proposed, and the idea of ​​shared convolution is adopted to predict the feature output through the same detection head, and the shared convolution module is optimized by group normalization to reduce the complexity of the model. The detection performance of the YOLOv8 network is effectively improved, and the accurate identification of substation equipment defects is achieved.

[0081] This embodiment constructs a substation defect image dataset, uses YOLOv8 as the basic detection model, and adopts Efficient Former V2 as the backbone network. On the basis of the spatial pooling pyramid layer, a large-scale separation convolution attention mechanism is introduced to improve the feature extraction ability of the model, which can reduce the amount of model calculation and improve the interaction of global features. For the detection problem of small target defect detection, the small target feature layer is integrated into the neck network as a separate branch to reduce the number of model parameters and the amount of calculation. For the problem of too many model parameters, a lightweight shared convolution detection head is proposed. The idea of ​​shared convolution is adopted to predict the feature output through the same detection head, and the shared convolution module is optimized by group normalization to reduce the complexity of the model. The problems of low positioning accuracy of small targets and excessive model parameters are solved, and the purpose of improving detection accuracy under the premise of meeting real-time detection is achieved.

[0082] Embodiment 2

[0083] Embodiment 2 of the present invention introduces a small-target lightweight substation equipment defect detection system.

[0084] like Figure 8 A small target lightweight substation equipment defect detection system shown includes:

[0085] An acquisition module, configured to acquire an image of substation equipment;

[0086] A detection module, which is configured to complete defect detection of substation equipment according to the acquired image and a preset small target lightweight defect detection model;

[0087] Among them, the preset small target lightweight defect detection model adopts the YOLOv8 target detection network, and the YOLOv8 target detection network includes a backbone network using Efficient Former V2 and a separated convolutional attention mechanism, a neck network using scale-time series feature fusion, and a prediction head network using a shared convolutional lightweight detection head; based on the backbone network, image features of the acquired image are extracted to obtain feature maps of different scales; the neck network performs feature fusion on the obtained feature maps of different scales to obtain a multi-scale feature map; the obtained multi-scale feature map is detected by a shared convolutional lightweight detection head to locate the defects of the equipment image and complete the defect detection of the substation equipment.

[0088] The detailed steps are the same as those of the small-target lightweight substation equipment defect detection method provided in Example 1, and will not be repeated here.

[0089] Embodiment 3

[0090] Embodiment 3 of the present invention provides a computer-readable storage medium.

[0091] A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps in a small-target lightweight substation equipment defect detection method as described in Embodiment 1 of the present invention.

[0092] The detailed steps are the same as those of the small-target lightweight substation equipment defect detection method provided in Example 1, and will not be repeated here.

[0093] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0094] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0097] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0098] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0099] Embodiment 4

[0100] A fourth embodiment of the present invention provides an electronic device.

[0101] An electronic device comprises a memory, a processor and a program stored in the memory and running on the processor, wherein when the processor executes the program, the steps in a small-target lightweight substation equipment defect detection method as described in Embodiment 1 of the present invention are implemented.

[0102] The detailed steps are the same as those of the small-target lightweight substation equipment defect detection method provided in Example 1, and will not be repeated here.

[0103] Embodiment 5

[0104] Embodiment 5 of the present invention provides a computer program product.

[0105] A computer program product includes software code, wherein the program in the software code executes the steps in a small-target lightweight substation equipment defect detection method as described in Embodiment 1 of the present invention.

[0106] The detailed steps are the same as those of the small-target lightweight substation equipment defect detection method provided in Example 1, and will not be repeated here.

[0107] The detailed steps are the same as those of the small-target lightweight substation equipment defect detection method provided in Example 1, and will not be repeated here.

[0108] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0109] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0112] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0113] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0114] The above description is only a preferred embodiment of the present embodiment and is not intended to limit the present embodiment. For those skilled in the art, the present embodiment may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiment shall be included in the protection scope of the present embodiment.

Claims

1. A small target lightweight substation equipment defect detection method, characterized in that: include: Obtain images of substation equipment; Complete defect detection of substation equipment based on the acquired images and the preset small target lightweight defect detection model; Among them, the preset small target lightweight defect detection model adopts the YOLOv8 target detection network, and the YOLOv8 target detection network includes a backbone network using Efficient Former V2 and a separated convolutional attention mechanism, a neck network using scale-time series feature fusion, and a prediction head network using a shared convolutional lightweight detection head; based on the backbone network, the image features of the acquired image are extracted to obtain feature maps of different scales; the neck network performs feature fusion on the obtained feature maps of different scales to obtain a multi-scale feature map; the obtained multi-scale feature map is detected by a shared convolutional lightweight detection head to locate the defects of the equipment image and complete the defect detection of the substation equipment.

2. A small target lightweight substation equipment defect detection method as described in claim 1, characterized in that: The backbone network includes a local module, a local global module, a downsampling module and a spatial pyramid pooling-large kernel separation convolution attention module; wherein the local global module includes a lightweight local global module mainly for lightweight and a ViT module local global module mainly for capturing global information; the local global module obtains global information through a multi-head self-attention mechanism and outputs it in a fixed size through a linear layer.

3. A small target lightweight substation equipment defect detection method as described in claim 2, characterized in that: The lightweight local-global module reduces the number of parameters of the pure ViT architecture by downsampling all linear layer output features to a uniform resolution size, and then inserting the original resolution into the next layer through the attention layer.

4. A small target lightweight substation equipment defect detection method as described in claim 2, characterized in that: The downsampling module class captures dynamic and static feature information through pooling and convolution to expand the receptive field; the spatial pyramid pooling-large kernel convolution separation attention module uses 3×3 depth-separable convolution and [7 / 2]×[7 / 2] depth-separable convolution to replace 7×7 large kernel convolution to extract local and global information and reduce the number of parameters and computation of the model.

5. A small target lightweight substation equipment defect detection method as described in claim 1, characterized in that: The neck network includes a scale sequence feature fusion module and a zoom splicing module, which are used to fuse feature maps of different scales obtained by the backbone network to achieve information interaction of different scales.

6. A small target lightweight substation equipment defect detection method as described in claim 1, characterized in that: The prediction head network performs classification and regression tasks through multi-scale feature maps and shared convolution lightweight detection heads, completes target positioning tasks, and locates defects in device images.

7. A small target lightweight substation equipment defect detection system, characterized in that: include: An acquisition module, configured to acquire an image of substation equipment; A detection module, which is configured to complete defect detection of substation equipment according to the acquired image and a preset small target lightweight defect detection model; Among them, the preset small target lightweight defect detection model adopts the YOLOv8 target detection network, and the YOLOv8 target detection network includes a backbone network using Efficient Former V2 and a separated convolutional attention mechanism, a neck network using scale-time series feature fusion, and a prediction head network using a shared convolutional lightweight detection head; based on the backbone network, the image features of the acquired image are extracted to obtain feature maps of different scales; the neck network performs feature fusion on the obtained feature maps of different scales to obtain a multi-scale feature map; the obtained multi-scale feature map is detected by a shared convolutional lightweight detection head to locate the defects of the equipment image and complete the defect detection of the substation equipment.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of a small-target lightweight substation equipment defect detection method as described in any one of claims 1-6 are implemented.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps of a small-target lightweight substation equipment defect detection method as described in any one of claims 1-6 are implemented.

10. A computer program product comprising software code, characterized in that The program in the software code executes the steps of a small-target lightweight substation equipment defect detection method as described in any one of claims 1-6.

Images