Knowledge graph and improved YOLOv10 overhead conductor detection method and system

Abstract

The invention discloses an overhead conductor detection method and system based on a knowledge graph and improved YOLOv10, and relates to the technical field of overhead conductor detection, and the method comprises the steps: carrying out the sample expansion processing of each training overhead conductor data through employing a pre-trained generation network, and obtaining a conductor feature set; and training a preset YOLOv10 defect detection model by adopting the conductor feature set, generating a target defect detection model and target feature data, constructing a conductor knowledge graph according to the target feature data and acceptance standard information, and when an overhead conductor image to be detected is received, performing defect detection on the conductor. And performing defect detection on the to-be-detected overhead conductor image by adopting the target defect detection model and the conductor knowledge graph to obtain an overhead conductor detection result. The technical problems that an existing deep learning network only identifies a single hidden danger type, the requirement for detecting diversified hidden danger types on the overhead conductor is difficult to meet, and the operation reliability of the overhead conductor is reduced are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of overhead wire detection, and particularly to a method and system for overhead wire detection based on a knowledge graph and improved YOLOv10. Background Art

[0002] With the rapid development of the power system, overhead wires, as the core components of power transmission, their safe and stable operation is crucial for ensuring power supply. However, overhead wires are prone to various hidden dangers due to various environmental factors and operation problems during operation, such as unqualified conductor sag, pole tilt, insulator damage, etc. If these hidden dangers cannot be discovered and handled in a timely manner, it may lead to line failures, power outages, and even major accidents. Therefore, how to accurately and efficiently detect the hidden dangers existing in overhead wires has become a technical problem that needs to be solved urgently.

[0003] Currently, the existing technology mainly uses deep learning networks to automatically detect overhead wire images and perform hidden danger identification and positioning, but only identifies a single type of hidden danger, which is difficult to meet the detection requirements of diverse hidden danger types on overhead wires and reduces the reliability of the operation of overhead wires. Summary of the Invention

[0004] The present invention provides a method and system for overhead wire detection based on a knowledge graph and improved YOLOv10, which solves the technical problem that the existing technology mainly uses deep learning networks to automatically detect overhead wire images and perform hidden danger identification and positioning, but only identifies a single type of hidden danger, which is difficult to meet the detection requirements of diverse hidden danger types on overhead wires and reduces the reliability of the operation of overhead wires.

[0005] A method for overhead wire detection based on a knowledge graph and improved YOLOv10 provided by the first aspect of the present invention includes:

[0006] Obtain a plurality of training overhead wire data and acceptance standard information, and use a pre-trained generation network to perform sample augmentation processing on each of the training overhead wire data to obtain a wire feature set;

[0007] Use the wire feature set to input a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data;

[0008] Construct a wire knowledge graph using the target feature data and the acceptance standard information;

[0009] When a to-be-detected overhead wire image is received, input the to-be-detected overhead wire image into the target defect detection model to obtain to-be-detected feature data;

[0010] Input the to-be-tested feature data into the wire knowledge graph to obtain the overhead wire detection result.

[0011] Optionally, the step of performing sample augmentation processing on each of the training overhead wire data by using a pre-trained generation network to obtain a wire feature set includes:

[0012] Perform image preprocessing on the defective overhead wire images of each of the training overhead wire data to obtain a plurality of first sample images;

[0013] Input the normal overhead wire images of each of the training overhead wire data into the pre-trained generation network respectively to obtain a plurality of second sample images;

[0014] Construct a wire feature set by using each of the first sample images and each of the second sample images.

[0015] Optionally, the generation network includes a generator and a discriminator. The step of inputting the normal overhead wire images of each of the training overhead wire data into the pre-trained generation network respectively to obtain a plurality of second sample images includes:

[0016] Input the normal overhead wire images of each of the training overhead wire data and preset attribute control data into the generator respectively to obtain a plurality of wire potential hazard images, wherein an adaptive noise injection module is connected after the convolutional layer in the generator;

[0017] Fuse each of the wire potential hazard images and each of the normal overhead wire images to obtain a plurality of initial potential hazard sample images;

[0018] Input each of the initial potential hazard sample images into the discriminator for screening to obtain a plurality of second sample images.

[0019] Optionally, the step of inputting the wire feature set into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data includes:

[0020] Input the wire feature set into a preset YOLOv10 defect detection model for training until a preset stop condition is met to obtain a target defect detection model;

[0021] Input the wire feature set into the target defect detection model to obtain target feature data.

[0022] Optionally, the target defect detection model includes a backbone network, an intermediate module, and a detection head group. The step of inputting the to-be-tested overhead wire image into the target defect detection model to obtain to-be-tested feature data includes:

[0023] Feature extraction is performed on the overhead conductor image to be measured by using the backbone network, and a plurality of conductor feature maps are output group by group;

[0024] The plurality of conductor feature maps are continuously downsampled by using the intermediate module to obtain a variety of fused feature maps;

[0025] Defect detection is performed on various fused feature maps by using the detection head group to obtain the feature data to be measured.

[0026] Optionally, the backbone network includes a first convolutional group, a second convolutional group, and a third convolutional group. The step of performing feature extraction on the overhead conductor image to be measured by using the backbone network and outputting a plurality of conductor feature maps group by group includes:

[0027] Feature extraction is performed on the overhead conductor image to be measured by using the first convolutional group to obtain a first conductor feature map, where the first convolutional group includes a convolutional layer, a star operation module, a convolutional layer, a star operation module, and a convolutional layer connected in sequence;

[0028] Feature extraction is performed on the first conductor feature map by using the second convolutional group to obtain a second conductor feature map, where the second convolutional group includes a star operation module and a convolutional layer connected in sequence;

[0029] Feature extraction is performed on the second conductor feature map by using the third convolutional group to obtain a third conductor feature map, where the third convolutional group includes a star operation module, a convolutional layer, a spatial pyramid pooling module, and a partial attention module connected in sequence.

[0030] Optionally, the star operation module includes a depthwise separable convolutional layer, a first fully connected layer, a second fully connected layer, a feature fusion layer, a fully connected layer, and a depthwise separable convolutional layer. The step of performing feature extraction on the first conductor feature map by using the second convolutional group to obtain a second conductor feature map includes:

[0031] Feature extraction is performed on the first conductor feature map by using the depthwise separable convolutional layer to obtain a first depth conductor feature map;

[0032] Linear transformation is performed on the first depth conductor feature map by using the first fully connected layer to obtain a second depth conductor feature map;

[0033] Linear transformation is performed on the first depth conductor feature map by using the second fully connected layer to obtain a third depth conductor feature map;

[0034] Feature fusion is performed on the second depth conductor feature map and the third depth conductor feature map by using the feature fusion layer to obtain a fourth depth conductor feature map;

[0035] The fourth depth wire feature map is extracted by sequentially using a fully connected layer and a depthwise separable convolutional layer to obtain a second wire feature map.

[0036] Optionally, the intermediate module includes a first fusion module, a second fusion module, a downsampling module and a third fusion module, and the step of continuously downsampling the multiple wire feature maps through the intermediate module to obtain multiple fusion feature maps is characterized by comprising:

[0037] Performing feature fusion on the third wire feature map and the second wire feature map through the first fusion module to obtain a first wire fusion feature map, wherein the first fusion module includes an upsampling module and a feature fusion layer;

[0038] Performing feature fusion on the first wire fusion feature map and the first wire feature through the second fusion module to obtain a first fusion feature map, wherein the second fusion module includes a first cross-stage fusion module, an upsampling module, a feature fusion layer and a first cross-stage fusion module;

[0039] Performing a downsampling operation on the first wire fusion feature map through the downsampling module to obtain a second fusion feature map, wherein the downsampling module includes a first basic module, a splicing module and a first cross-stage fusion module connected in sequence;

[0040] The second fused feature map and the third wire feature map are feature-fused by the third fusion module to obtain a third fused feature map, wherein the third fusion module includes a channel downsampling module, a feature fusion layer and a second cross-stage fusion module.

[0041] Optionally, the step of using the detection head group to perform defect detection on various fused feature images to obtain feature data to be tested includes:

[0042] According to the resolution of the fused feature map, matching a corresponding detection head from the detection head group;

[0043] The detection head is used to perform defect detection on the fused feature map to obtain feature data to be tested, wherein the detection head includes a first group normalized convolution module, a second group normalized convolution module, a second group normalized convolution module and a prediction convolution module which are connected in sequence.

[0044] Optionally, a scaling module is connected after the prediction box convolution module in the prediction convolution module.

[0045] Optionally, the step of performing a downsampling operation on the first wire fusion feature map by the downsampling module to obtain a second fusion feature map includes:

[0046] Performing feature extraction on the first wire fusion feature map through the first basic module to obtain a first target fusion feature map, wherein the first basic module includes a two-dimensional convolution layer, a batch normalization layer, and a SiLU activation function layer connected in sequence;

[0047] The first target fusion feature map is sequentially subjected to feature extraction through a splicing module and a first cross-stage fusion module to obtain a second fusion feature map.

[0048] Optionally, the step of performing feature fusion on the second fused feature map and the third wire feature map by the third fusion module to obtain a third fused feature map includes:

[0049] A channel downsampling module is used to perform a sampling operation on the second fusion feature map to obtain a target sampling feature map;

[0050] Performing feature fusion on the target sampling feature map and the third wire feature map through a feature fusion layer to obtain an intermediate fusion feature map;

[0051] The intermediate fusion feature map is subjected to feature extraction through a second cross-stage fusion module to obtain a third fusion feature map.

[0052] The second aspect of the present invention provides an overhead wire detection system based on a knowledge graph and an improved YOLOv10, comprising:

[0053] An expansion module is used to obtain a plurality of training overhead conductor data and acceptance standard information, and use a pre-trained generation network to perform sample expansion processing on each of the training overhead conductor data to obtain a conductor feature set;

[0054] A training module, used to use the wire feature set to input a preset YOLOv10 defect detection model for training, and generate a target defect detection model and target feature data;

[0055] A construction module, used to construct a wire knowledge graph using the target feature data and the acceptance standard information;

[0056] A response module, for, when receiving an overhead wire image to be tested, inputting the overhead wire image to be tested into the target defect detection model to obtain feature data to be tested;

[0057] The detection module is used to input the feature data to be tested into the conductor knowledge graph to obtain the overhead conductor detection result.

[0058] An electronic device provided in the third aspect of the present invention includes a memory and a processor. A computer program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the overhead wire detection method based on the knowledge graph and improved YOLOv10 as described in any one of the above.

[0059] A computer-readable storage medium provided in the fourth aspect of the present invention has a computer program stored thereon. When the computer program is executed, it implements the overhead wire detection method based on the knowledge graph and improved YOLOv10 as described in any one of the above.

[0060] As can be seen from the above technical solutions, the present invention has the following advantages:

[0061] In the present invention, a pre-trained generation network is used to perform sample augmentation processing on each training overhead wire data to obtain a wire feature set. The wire feature set is used to train a preset YOLOv10 defect detection model to generate a target defect detection model and target feature data. According to the target feature data and acceptance standard information, a wire knowledge graph is constructed. When an overhead wire image to be measured is received, the target defect detection model and the wire knowledge graph are used to perform defect detection on the overhead wire image to be measured to obtain an overhead wire detection result. It overcomes the technical problem that the prior art mainly uses a deep learning network to automatically detect an overhead wire image, identify and locate potential hazards, but only identifies a single type of potential hazard, making it difficult to meet the detection requirements for diverse potential hazard types on overhead wires and reducing the reliability of overhead wire operation. Compared with traditional defect detection methods, in the present invention, a pre-trained generation network is used to perform sample augmentation processing on each training overhead wire data to obtain a wire feature set, enriching the training samples of the preset YOLOv10 defect detection model, enabling the target defect detection model to accurately identify various types of potential hazards, improving the detection accuracy of the target defect detection model, and improving the reliability of overhead wire operation. Description of the Drawings

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0063] Figure 1 It is a flowchart of the steps of an overhead wire detection method based on a knowledge graph and improved YOLOv10 provided in Embodiment 1 of the present invention;

[0064] Figure 2The flowchart of steps of an overhead conductor detection method based on a knowledge graph and improved YOLOv10 provided in the second embodiment of the present invention;

[0065] Figure 3 The structural schematic diagram of the generation network provided in the second embodiment of the present invention;

[0066] Figure 4 The structural schematic diagram of the target defect detection model provided in the second embodiment of the present invention;

[0067] Figure 5 The structural schematic diagram of the star operation module provided in the second embodiment of the present invention;

[0068] Figure 6 The structural schematic diagram of the detection head provided in the second embodiment of the present invention;

[0069] Figure 7 The structural block diagram of an overhead conductor detection system based on a knowledge graph and improved YOLOv10 provided in the third embodiment of the present invention;

[0070] Figure 8 The structural block diagram of an electronic device provided in the fourth embodiment of the present invention. Detailed implementation manners

[0071] The embodiments of the present invention provide an overhead conductor detection method and system based on a knowledge graph and improved YOLOv10, which are used to solve the technical problem that the existing technology mainly uses a deep learning network to automatically detect overhead conductor images, identify and locate potential hazards, but only identifies a single type of potential hazard, making it difficult to meet the detection requirements for diverse potential hazard types on overhead conductors and reducing the reliability of the operation of overhead conductors.

[0072] To make the objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0073] Please refer to Figure 1 , Figure 1 The flowchart of steps of an overhead conductor detection method based on a knowledge graph and improved YOLOv10 provided in the first embodiment of the present invention.

[0074] An overhead conductor detection method based on a knowledge graph and improved YOLOv10 provided by the present invention includes:

[0075] Step 101: Obtain multiple training overhead conductor data and acceptance standard information, and use a pre-trained generation network to perform sample augmentation processing on each training overhead conductor data to obtain a conductor feature set;

[0076] It should be noted that the training overhead conductor data includes normal overhead conductor images and defective overhead conductor images.

[0077] In the embodiment of the present invention, an unmanned aerial vehicle equipped with a high-definition camera is used to obtain overhead conductor images on the transmission line, and the overhead conductor images are screened and classified to obtain multiple normal overhead conductor images and multiple defective overhead conductor images, and the acceptance standard information is obtained. Image preprocessing is performed on each defective overhead conductor image to obtain multiple first sample images. Multiple normal overhead conductor images and preset attribute control data are respectively input into the pre-trained generation network to obtain multiple second sample images. A conductor feature set is constructed using each first sample image and each second sample image.

[0078] It should be noted that the attribute control data refers to predefined rules or parameters for automatically managing the physical or electrical characteristics of the conductor.

[0079] It should be noted that the acceptance standard information includes but is not limited to the defect type of the overhead conductor, equipment type, specifications, the solution corresponding to each defect type, and the evaluation criteria for the defect severity.

[0080] Step 102: Use the conductor feature set to input into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data;

[0081] In the embodiment of the present invention, the conductor feature set is used to train the preset YOLOv10 defect detection model until the preset stop condition is met, and the training of the preset YOLOv10 defect detection model is stopped to generate a target defect detection model. The conductor feature set is input into the target defect detection model for feature extraction to obtain target feature data.

[0082] Step 103: Construct a conductor knowledge graph using the target feature data and acceptance standard information;

[0083] In the embodiment of the present invention, knowledge extraction is performed on the acceptance standard information to obtain entities, relationships between entities, and entity attributes. According to the preset graph model and entities, relationships between entities, and entity attributes, a knowledge graph structure is constructed (the knowledge graph structure includes a structure hierarchy and semantic organization, where the structure layer includes an equipment layer, a parameter layer, and a standard layer, and the semantic organization includes equipment classification, parameter constraints, and standard applicability, etc.). A knowledge graph is generated in the data storage system. Each target feature in the target feature data is associated with the entities in the knowledge graph to obtain a conductor knowledge graph.

[0084] Step 104: When the image of the overhead conductor to be measured is received, input the image of the overhead conductor to be measured into the target defect detection model to obtain the feature data to be measured;

[0085] The feature data to be measured refers to the features of the overhead conductor detected from the image of the overhead conductor to be measured.

[0086] In the embodiment of the present invention, when the image of the overhead conductor to be measured is received, the image of the overhead conductor to be measured is used as the input of the target defect detection model to obtain the feature data to be measured.

[0087] Step 105: Input the feature data to be measured into the conductor knowledge graph to obtain the overhead conductor detection result.

[0088] In the embodiment of the present invention, the feature data to be measured is input into the conductor knowledge graph for retrieval to match the overhead conductor detection result corresponding to the image of the overhead conductor to be measured.

[0089] In the embodiment of the present invention, through the use of a pre-trained generation network to perform sample augmentation processing on each training overhead conductor data, a conductor feature set is obtained, and the conductor feature set is used to train a preset YOLOv10 defect detection model to generate a target defect detection model and target feature data. According to the target feature data and acceptance standard information, a conductor knowledge graph is constructed. When the image of the overhead conductor to be measured is received, the target defect detection model and the conductor knowledge graph are used to perform defect detection on the image of the overhead conductor to be measured to obtain the overhead conductor detection result. It overcomes the technical problem that the prior art mainly uses a deep learning network to automatically detect the image of the overhead conductor and perform hidden danger identification and positioning, but only identifies a single type of hidden danger, which is difficult to meet the detection requirements of diverse hidden danger types on the overhead conductor and reduces the reliability of the operation of the overhead conductor. Compared with the traditional defect detection method, in the present invention, through the use of a pre-trained generation network to perform sample augmentation processing on each training overhead conductor data, a conductor feature set is obtained, enriching the training samples of the preset YOLOv10 defect detection model, enabling the target defect detection model to accurately identify various types of hidden dangers, improving the detection accuracy of the target defect detection model, and improving the reliability of the operation of the overhead conductor.

[0090] Please refer to Figure 2 , Figure 2 which is the step flowchart of a method for detecting overhead conductors based on a knowledge graph and improved YOLOv10 provided in the second embodiment of the present invention.

[0091] A method for detecting overhead conductors based on a knowledge graph and improved YOLOv10 provided by the present invention includes:

[0092] Step 201: Obtain multiple pieces of training overhead conductor data and acceptance standard information, perform image preprocessing on the defective overhead conductor images of each piece of training overhead conductor data, and obtain multiple first sample images;

[0093] In the embodiment of the present invention, multiple pieces of training overhead conductor data and acceptance standard information on the overhead conductor are obtained by means of a drone, image preprocessing is performed on the defective overhead conductor images of each piece of training overhead conductor data, and multiple first sample images are obtained.

[0094] It is worth mentioning that the training overhead conductor data collected by the drone covers overhead conductor images under various weather conditions and different time periods, and positive and negative samples are obtained simultaneously to ensure that the training overhead conductor data can cover diverse lighting conditions, different scenarios, and background changes.

[0095] It should be noted that the image preprocessing process is specifically as follows: A1. Screen out the defective overhead conductor images that meet the usage requirements from the defective overhead conductor images of each piece of training overhead conductor data. A2. Perform image enhancement operations on each defective overhead conductor image respectively to obtain multiple first sample images. The image enhancement operations include horizontal flipping, vertical flipping, image blurring, and random transformation of hue and saturation, etc. The image enhancement operations can effectively increase the diversity of the data set, thereby reducing the dependence on large-scale labeled data, and significantly improving the robustness and accuracy of the target defect detection model under different environments and conditions to meet the requirements in actual applications.

[0096] It is worth mentioning that the horizontal and vertical flipping operations improve the model's recognition ability of the target from various perspectives by simulating the states of the transmission line in different directions. Introducing a certain degree of image blurring helps the model adapt to the low-clarity images that may appear in the transmission line and enhances its detection ability for the target in blurred images. Randomly adjusting the hue and saturation of the image to cope with the changes under different lighting conditions enables the model to maintain good detection performance in various lighting environments.

[0097] Step 202: Input the normal overhead conductor images of each piece of training overhead conductor data into a pre-trained generation network respectively to obtain multiple second sample images;

[0098] Furthermore, the generation network includes a generator and a discriminator, and step 202 includes the following sub-steps:

[0099] S11: Input the normal overhead conductor images of each piece of training overhead conductor data and preset attribute control data into the generator respectively to obtain multiple conductor hidden danger images, wherein an adaptive noise injection module is connected after the convolutional layer in the generator;

[0100] In the embodiment of the present invention, refer to Figure 3As shown, the normal overhead conductor images of each training overhead conductor data and the preset attribute control data (i.e., the attribute control data includes time series observation data, random noise vectors, and attribute control charts) are respectively input into the generator (the generator is a pre-trained MDN (Mozilla Developer Network) generator, i.e., a metadata network generator), and multiple conductor hidden danger images are obtained. Among them, an adaptive noise injection module is connected after the convolutional layer in the generator.

[0101] It is worth mentioning that the time series observation data: provides time-related information. For example, in an image generation task, if the image has a time dimension association (such as tasks related to consecutive frames in a video), this part of the data can provide relevant time-based feature information to help the generator generate content that matches the time information. Random noise vectors: introduce randomness, enabling the generator to generate diverse results. Just like when generating images, different noise vectors will cause the generator to produce images with different appearances, thus ensuring the richness of the generated samples. Attribute control charts: are used to control the attributes of the generated results. For example, in image generation, attributes such as the color, style, and object shape of the generated image can be controlled. Through this part of the input, users can specify certain specific attributes that they want the generated samples to have.

[0102] It is worth mentioning that the MDN generator can simulate the posterior distribution, thereby providing non-Gaussian posterior predictions, enabling the generation network to generate a predictive distribution with multiple modes. While traditional GAN models usually only generate single point estimates. The input of the traditional GAN model is (observations) and , where is a set of samples sampled from a normal distribution. Under and conditions, directly model the predictive likelihood of the input conditions to obtain the traditional GAN model.

[0103] The likelihood function of the traditional GAN model can be expressed as:

[0104]

[0105] Among them, is the mixing coefficient, is the mean, is the standard deviation, m is the number of mixing components, is the generated data, is the observation, is the set of samples sampled from the normal distribution, i is the sample index, t is the time, y t is the target value at time t, N i is the i-th normal distribution.

[0106] The discriminator also depends on , and the input of the discriminator model is designed as , where is the standard deviation of the observed set. For the true time series data value (i.e., the target value at time t), is maximized, and the generator tries to deceive the discriminator by generating (generated data), so that is maximized.

[0107] The loss function of the discriminator in the traditional GAN model is as follows. The loss function of the discriminator consists of two parts. One part is the loss for the real data (i.e., real images), and the other part is the loss for the generated data (i.e., forged images):

[0108]

[0109] where is the expected value of the time series set, is the expected value of the sample set sampled from the state distribution.

[0110] When the calculated loss function is greater than the preset loss threshold, it is determined that the generated data (i.e., forged images) is true, and the forged images are used as wire hidden danger images. When the calculated loss function is less than or equal to the preset loss threshold, it is determined that the generated data is false.

[0111] It should be noted that, as shown in Figure 3 , in the generator of the present invention, an adaptive noise injection module is added after each convolutional layer of the encoder-decoder architecture to explicitly inject Gaussian noise into the feature map after each convolutional layer. For each noise injection, the network learns an independent scalar to adjust the intensity of the injected noise. By explicitly reflecting the random variations of the defects, the generator network can generate more realistic and diverse defect samples.

[0112] It is worth mentioning that the adaptive noise injection module can generate a Gaussian noise matrix with the same size as each feature map. Assuming that the size of the feature map F is , we generate a noise matrix N from the Gaussian distribution:

[0113]

[0114] Introduce a learnable scalar parameter to control the intensity of the noise. The adjusted noise matrix is expressed as:

[0115]

[0116] Add the adjusted noise matrix N′ to the feature map F to obtain the noisy feature map F′:

[0117]

[0118] Among them, the formula for Gaussian noise is:

[0119]

[0120] Among them, x is a random variable, is the mean value, is the standard deviation, is the variance, is the Gaussian noise.

[0121] It should be noted that referring to Figure 3 shown, the Attribute Control Map of the attribute control data is used to provide better attribute control for the generator. The attribute control map adds specific types of defects to specific positions, where indicates the existence of defects at the corresponding positions, and C represents the number of defect categories. Incorporate the attribute control map A into the network through SPADE normalization (i.e., spatial adaptive normalization) and input it into each module in the decoder part. After introducing the spatial and attribute control map A, the prediction likelihood function of the generator is adjusted to:

[0122]

[0123] It is worth mentioning that by introducing spatial and category control, the generation network can more precisely control the generated prediction distribution, especially when it is necessary to generate predictions of specific categories at specific positions. This not only improves the flexibility of the generation network but also enhances its application potential in complex defect generation.

[0124] S12. Respectively fuse each wire hidden danger image and each normal overhead wire image to obtain multiple initial hidden danger sample images;

[0125] In the embodiment of the present invention, each wire hidden danger image and each normal overhead wire image are respectively subjected to image synthesis through the alpha blending technique to obtain multiple initial hidden danger sample images.

[0126] It should be noted that the alpha blending technique is a commonly used image fusion method. By introducing a transparency coefficient, a smooth transition of two images in space is achieved. Alpha blending adds two images and in a certain proportion to obtain the fused image. The formula is as follows:

[0127]

[0128] Among them, is the value of the fused image at position value, is the transparency coefficient, and its value range is [0, 1]. During the image synthesis process, an Alpha gradient can be defined for the edges of the hidden danger area, and the transparency gradually decreases from the center to the edge. According to the defined Alpha value, the hidden danger area is mixed with the normal image to achieve the natural superposition of the hidden danger area.

[0129] S13. Input each initial hidden danger sample image into the discriminator for screening to obtain multiple second sample images.

[0130] In the embodiment of the present invention, each initial hidden danger sample image is screened by the discriminator to obtain multiple second sample images.

[0131] Step 203. Construct a wire feature set using each first sample image and each second sample image.

[0132] In the embodiment of the present invention, the hidden danger areas of each first sample image and each second sample image are manually labeled by a standard system, and the weeds of each first sample image and each second sample image are labeled using the minimum bounding rectangle to obtain multiple labeled images (wherein, the category and coordinate information of the bounding box in each labeled image are recorded as the center point coordinates (x, y) of the bounding box and the width and height (w, h) of the bounding box to determine the relative position of the weeds in the image). Each labeled image is divided into a training set and a validation set according to a ratio of 8:2 to obtain a wire feature set.

[0133] Step 204. Input the wire feature set into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data;

[0134] Further, step 204 includes the following sub-steps:

[0135] S21. Input the wire feature set into a preset YOLOv10 defect detection model for training until a preset stop condition is met to obtain a target defect detection model;

[0136] In an embodiment of the present invention, a preset YOLOv10 defect detection model is trained using a training set of wire feature sets to obtain training wire feature data. The loss value of the training set is calculated based on the training wire feature data, and it is determined whether the loss value is less than a preset loss threshold. If the loss value is greater than or equal to the loss threshold, the network parameters of the preset YOLOv10 defect detection model are adjusted using a grid search method or a random search method, and the process jumps back to the step of training the preset YOLOv10 defect detection model using the training set of wire feature sets to obtain training wire feature data until the loss value is less than the loss threshold. If the loss value is less than the loss threshold, a target defect detection model is generated.

[0137] It is worth mentioning that after obtaining the target defect detection model, a validation set of wire feature sets can be used to evaluate the target defect detection model. By using the validation set to evaluate the target defect detection model, evaluation wire feature data is obtained. Based on the evaluation wire feature data, the accuracy (the accuracy is the ratio of the correctly predicted results among all the results predicted as positive samples), recall rate (the recall rate is the proportion of the positive examples correctly identified by the model among all the actual positive examples), and mean average precision value of the validation set are calculated. The network structure and network parameters of the target defect detection model are adjusted through the accuracy (P), recall rate (R), mean average precision value (mAP), frames per second (FPS) of the evaluation wire feature data, GFLOPS (giga floating-point operations per second), and the number of model parameters Pare(M).

[0138] The accuracy expression is:

[0139]

[0140] Where, is the accuracy, is the accurate number of weed images detected in the experimental detection of the capacitor bank fence, is the number of errors in the experimental detection.

[0141] The recall rate expression is:

[0142]

[0143] Where, is the recall rate, is the number of omissions of images that cannot be detected in the experiment.

[0144] The mean average precision expression is:

[0145]

[0146]

[0147] Where, is the average precision, i is the class number, is the total number of classes, is the average precision of the i-th class.

[0148] It is worth mentioning that the preset YOLOv10 defect detection model is an improved YOLOv10 model, which introduces large kernel convolution and a partial self-attention module (PSA). Large kernel depth convolution can effectively expand the receptive field of the model and enhance the model's ability to capture global information. However, simply using large kernel convolution in all stages may lead to contamination of shallow features (affecting the detection of small objects), so large kernel convolution is only used in small-scale models to reduce additional computational overhead. At the same time, the improved YOLOv10 model adopts a technique called "star operation", which implicitly realizes high-dimensional feature mapping in the low-dimensional feature space of the input. Similar to the traditional kernel function method, the star operation enriches the dimension of the feature space through non-linear transformation without increasing explicit computation, enabling the model to more effectively extract fine-grained features in the detection task. The improved YOLOv10 model recursively expands the dimension of implicit features through multiple layers of star operations, simulating an extremely high-dimensional feature space. This multi-level feature expansion method is equivalent to increasing the complexity and expression ability of features layer by layer in the model, ensuring that the model has sufficient feature capture ability in complex scenarios, especially the ability to extract subtle features.

[0149] The star operation usually refers to element-wise multiplication. In a neural network, it can be expressed as the fusion of two linearly transformed features. For example, there are two linearly transformed features and , the star operation can be expressed as:

[0150]

[0151] where, is the first weight matrix, is the second weight matrix, is the first bias term, is the second bias term, is the input feature.

[0152] Combining the weight matrix and the bias into one entity W and defining X as a vector containing the input feature and a constant term. In this way, the star operation can be simplified to:

[0153]

[0154] Different from traditional neural networks that increase the dimension by increasing the network width, the star operation brings an exponential increase in the complexity of implicit dimensions layer by layer through stacking.

[0155] S22. Input the wire feature set into the target defect detection model to obtain target feature data.

[0156] In the embodiment of the present invention, the wire feature set is used as the input of the target defect detection model to obtain target feature data.

[0157] Step 205. Construct a wire knowledge graph by using the target feature data and acceptance standard information;

[0158] In the embodiment of the present invention, knowledge extraction and structured representation are performed on the acceptance standard information to obtain entities, relationships between entities, and entity attributes. According to the preset graph model and entities, relationships between entities, and entity attributes (entity attributes include but are not limited to equipment models, voltage levels, temperature tolerance ranges, etc.), a knowledge graph structure is constructed (the knowledge graph structure includes a structure hierarchy and semantic organization. Among them, the structure layer includes an equipment layer, a parameter layer, and a standard layer, and the semantic organization includes equipment classification, parameter constraints, and standard applicability, etc.). A knowledge graph is generated in the data storage system. Associate each target feature in the target feature data (target features include but are not limited to poles and towers, wires, insulators, position coordinates, appearance features, target categories, detection confidence levels, etc.) with the entities in the knowledge graph to obtain a wire knowledge graph.

[0159] It is worth mentioning that after obtaining the wire knowledge graph, the wire knowledge graph can be verified and optimized to ensure its effectiveness and accuracy. The specific process is as follows: B1. Check the integrity of the graph of the wire knowledge graph to ensure that its structure is reasonable, the node relationships are accurate, and redundant and duplicate information is avoided. B2. Conduct inference tests through some typical acceptance cases to verify the matching accuracy and logical rationality of the knowledge graph for equipment parameters and acceptance standards. B3. According to the verification results, further optimize the graph structure and adjust unreasonable relationship definitions and information layouts.

[0160] Step 206. When a to-be-detected overhead wire image is received, input the to-be-detected overhead wire image into the target defect detection model to obtain to-be-detected feature data;

[0161] Further, refer to Figure 4 As shown, the target defect detection model includes a backbone network, an intermediate module, and a detection head group. Step 206 includes the following sub-steps:

[0162] S31. Use the backbone network to perform feature extraction on the to-be-detected overhead wire image and output multiple wire feature maps group by group;

[0163] Further, the backbone network includes a first convolutional group, a second convolutional group, and a third convolutional group. S31 includes the following sub-steps:

[0164] S311. Extract features from the image of the overhead wire to be measured through the first convolutional group to obtain the first wire feature map, where the first convolutional group includes a convolutional layer, a star operation module, a convolutional layer, a star operation module, and a convolutional layer connected in sequence;

[0165] In the embodiment of the present invention, features of the image of the overhead wire to be measured are extracted through a convolutional layer, a star operation module, a convolutional layer, a star operation module, and a convolutional layer in sequence to obtain the first wire feature map.

[0166] S312. Extract features from the first wire feature map through the second convolutional group to obtain the second wire feature map, where the second convolutional group includes a star operation module and a convolutional layer connected in sequence;

[0167] Further, referring to Figure 5 As shown, the star operation module includes a depthwise separable convolutional layer, a first fully connected layer, a second fully connected layer, a feature fusion layer, a fully connected layer, and a depthwise separable convolutional layer. S312 includes the following sub-steps:

[0168] S3121. Extract features from the first wire feature map through the depthwise separable convolutional layer to obtain the first depth wire feature map;

[0169] In the embodiment of the present invention, features of the first wire feature map are extracted through the depthwise separable convolutional layer to obtain the first depth wire feature map, where the convolutional kernel of the depthwise separable convolutional layer is 7 and the stride is 1.

[0170] It should be noted that the depthwise separable convolutional layer is an efficient convolutional operation. By decomposing the standard convolution into Depthwise convolution and Pointwise convolution, the amount of computation and the number of parameters are significantly reduced.

[0171] S3122. Perform a linear transformation on the first depth wire feature map through the first fully connected layer to obtain the second depth wire feature map;

[0172] S3123. Perform a linear transformation on the first depth wire feature map through the second fully connected layer to obtain the third depth wire feature map;

[0173] In the embodiment of the present invention, linear transformations are respectively performed on the first depth wire feature map through the first fully connected layer and the second fully connected layer to obtain the second depth wire feature map and the third depth wire feature map.

[0174] S3124. Use the feature fusion layer to fuse the features of the second depth wire feature map and the third depth wire feature map to obtain the fourth depth wire feature map;

[0175] In an embodiment of the present invention, the second depth wire feature map and the third depth wire feature map are feature-fused through a feature fusion layer to obtain a fourth depth wire feature map.

[0176] S3125. Feature extraction is sequentially performed on the fourth depth wire feature map through a fully connected layer and a depthwise separable convolution layer to obtain a second wire feature map.

[0177] In an embodiment of the present invention, feature extraction is sequentially performed on the fourth depth wire feature map through a fully connected layer and a depthwise separable convolution layer to obtain an initial second wire feature map, and the initial second wire feature map is feature-fused with the first wire feature map to obtain a second wire feature map.

[0178] S313. Feature extraction is performed on the second wire feature map through a third convolution group to obtain a third wire feature map, where the third convolution group includes a star operation module, a convolution layer, a spatial pyramid pooling module, and a partial attention module connected in sequence.

[0179] In an embodiment of the present invention, feature extraction is sequentially performed on the second wire feature map through a star operation module, a convolution layer, a spatial pyramid pooling module (SPPF), and a partial attention module (PSA) to obtain a third wire feature map.

[0180] It should be noted that the self-attention mechanism can enhance the global modeling ability of the model, but has a large computational complexity and memory occupancy. The partial attention module (PSA) divides the channel features into two parts and only applies self-attention to one part, thereby reducing the computational cost while retaining the global modeling ability.

[0181] It should be noted that the spatial pyramid pooling module (SPPF) reduces the computational amount by repeatedly using pooling operations while retaining the ability of multi-scale feature extraction.

[0182] S32. Multiple wire feature maps are continuously downsampled through an intermediate module to obtain multiple fusion feature maps;

[0183] Further, referring to Figure 4 As shown, the intermediate module includes a first fusion module, a second fusion module, a downsampling module, and a third fusion module, and S32 includes the following sub-steps:

[0184] S321. Feature fusion is performed on the third wire feature map and the second wire feature map through a first fusion module to obtain a first wire fusion feature map, where the first fusion module includes an upsampling module and a feature fusion layer.

[0185] In an embodiment of the present invention, the third wire feature map is upsampled by an upsampling module (Upsamle) to obtain a first upsampled feature map, and the first upsampled feature map and the second wire feature map are feature-fused by a feature fusion layer (Concat) to obtain a first wire fusion feature map.

[0186] S322. The first wire fusion feature map and the first wire feature are feature-fused by a second fusion module to obtain a first fusion feature map, where the second fusion module includes a first cross-stage fusion module, an upsampling module, a feature fusion layer, and a first cross-stage fusion module.

[0187] In an embodiment of the present invention, the first wire fusion feature map is feature-extracted by a first cross-stage fusion module (C2f) and an upsampling module in sequence to obtain a second upsampled feature map, the second upsampled feature map and the first wire feature are feature-fused by a feature fusion layer to obtain an initial first fusion feature map, and the initial first fusion feature map is feature-extracted by a first cross-stage fusion module (C2f) to obtain a first fusion feature map.

[0188] S323. The first wire fusion feature map is downsampled by a downsampling module to obtain a second fusion feature map, where the downsampling module includes a first basic module, a splicing module, and a first cross-stage fusion module connected in sequence.

[0189] Further, S323 includes the following sub-steps:

[0190] S3231. The first wire fusion feature map is feature-extracted by a first basic module to obtain a first target fusion feature map, where the first basic module includes a two-dimensional convolutional layer, a batch normalization layer, and a SiLU activation function layer connected in sequence.

[0191] In an embodiment of the present invention, the first wire fusion feature map is feature-extracted by a two-dimensional convolutional layer, a batch normalization layer, and a SiLU activation function layer in sequence to obtain a first target fusion feature map.

[0192] S3232. The first target fusion feature map is feature-extracted by a splicing module and a first cross-stage fusion module in sequence to obtain a second fusion feature map.

[0193] In an embodiment of the present invention, the first target fusion feature map is feature-extracted by a splicing module (Concat) and a first cross-stage fusion module (C2f) in sequence to obtain a second fusion feature map.

[0194] S324. Perform feature fusion on the second fused feature map and the third wire feature map through a third fusion module to obtain a third fused feature map, wherein the third fusion module includes a channel downsampling module, a feature fusion layer and a second cross-stage fusion module.

[0195] Further, S324 includes the following sub-steps:

[0196] S3241, using a channel downsampling module to perform a sampling operation on the second fusion feature map to obtain a target sampling feature map;

[0197] In the embodiment of the present invention, a sampling operation is performed on the second fused feature map through a channel downsampling module (ScDown) to obtain a target sampling feature map.

[0198] It should be noted that the channel downsampling module (ScDown) is a scale-aware downsampling module that retains multi-scale information through multi-scale convolution and feature fusion, thereby enhancing the model's perception of multi-scale targets.

[0199] S3242, performing feature fusion on the target sampling feature map and the third wire feature map through a feature fusion layer to obtain an intermediate fusion feature map;

[0200] In the embodiment of the present invention, the target sampling feature map and the third wire feature map are feature fused through a feature fusion layer (Concat) to obtain an intermediate fused feature map.

[0201] S3243. Perform feature extraction on the intermediate fusion feature map through the second cross-stage fusion module to obtain a third fusion feature map.

[0202] In the embodiment of the present invention, the feature extraction is performed on the intermediate fusion feature map through the second cross-stage fusion module (C2fC1B) to obtain a third fusion feature map.

[0203] It should be noted that the second cross-stage fusion module (C2fC1B) can be composed of a C2f module, a 1x1 convolution and a batch normalization layer to enhance feature extraction capabilities, adjust the number of channels and improve model stability.

[0204] S33. Use a detection head group to perform defect detection on various fused feature images to obtain feature data to be tested.

[0205] Further, S33 includes the following sub-steps:

[0206] S331, matching a corresponding detection head from the detection head group according to the resolution of the fused feature map;

[0207] In the embodiment of the present invention, a corresponding detection head is selected from the three detection heads of the detection head group according to the resolution of the fused feature map.

[0208] It is worth mentioning that the detection head group makes full use of the advantages of Group Normalization (GroupNorm) and shared convolution, effectively fusing feature information while minimizing the amount of calculation and computational complexity. Shared convolution is a core concept in convolutional neural networks (CNNs), allowing the same convolutional kernel to use the same weight parameters at different positions. This mechanism significantly reduces the number of parameters to be trained, improves computational efficiency, and reduces the risk of overfitting. Shared convolution can effectively extract local features while maintaining spatial structure information, enhancing the generalization ability of the model.

[0209] S332. Use a detection head to perform defect detection on the fused feature map to obtain the feature data to be measured. Among them, the detection head includes a first group normalization convolution module, a second group normalization convolution module, a second group normalization convolution module, and a prediction convolution module connected in sequence.

[0210] In the embodiment of the present invention, refer to Figure 6 As shown, it passes through a first group normalization convolution module (i.e., 1×1 depthwise separable convolution), a second group normalization convolution module (i.e., 3×3 depthwise separable convolution), a second group normalization convolution module (i.e., 3×3 depthwise separable convolution), and a prediction convolution module in sequence. Among them, the prediction convolution module includes a plurality of convolution regularization modules (i.e., Conv_Reg) and a plurality of convolution classification modules (i.e., Conv_Cls).

[0211] It should be noted that depthwise separable convolution divides the channels of an image with an input size of [N, C, H, W] into several groups (where H is the height and W is the width), calculates the variance and mean of each group, and then normalizes all data within the group. Since the calculation of GroupNorm in depthwise separable convolution depends on the number of channels C rather than the batch size N, accurate feature recognition can be performed when the computer memory is limited or the number of samples is small.

[0212] It should be noted that a scaling module is connected after the prediction box convolution module in the prediction convolution module.

[0213] It is worth mentioning that the scaling module (i.e., Scale module) is used to match the detection of targets at different scales.

[0214] Step 207. Input the feature data to be measured into the wire knowledge graph to obtain the overhead wire detection result.

[0215] In the embodiment of the present invention, the feature data to be measured is input into the wire knowledge graph for retrieval to obtain the overhead wire detection result.

[0216] In the embodiments of the present invention, by using a pre-trained generation network to perform sample augmentation processing on each piece of training overhead conductor data, a conductor feature set is obtained, and the conductor feature set is used to train a preset YOLOv10 defect detection model to generate a target defect detection model and target feature data. According to the target feature data and acceptance standard information, a conductor knowledge graph is constructed. When a to-be-detected overhead conductor image is received, the target defect detection model and the conductor knowledge graph are used to perform defect detection on the to-be-detected overhead conductor image to obtain an overhead conductor detection result. This overcomes the technical problem that the existing technology mainly uses a deep learning network to automatically detect overhead conductor images, identify and locate potential hazards, but only identifies a single type of potential hazard, making it difficult to meet the detection requirements for diverse potential hazard types on overhead conductors and reducing the reliability of the operation of overhead conductors. Compared with traditional defect detection methods, in the present invention, by using a pre-trained generation network to perform sample augmentation processing on each piece of training overhead conductor data, a conductor feature set is obtained, enriching the training samples of the preset YOLOv10 defect detection model, enabling the target defect detection model to accurately identify various types of potential hazards, improving the detection accuracy of the target defect detection model, and enhancing the reliability of the operation of overhead conductors.

[0217] Please refer to Figure 7 , Figure 7 which is a structural block diagram of an overhead conductor detection system based on a knowledge graph and improved YOLOv10 provided in Embodiment 3 of the present invention.

[0218] An overhead conductor detection system based on a knowledge graph and improved YOLOv10 provided by the present invention includes:

[0219] An augmentation module 301, configured to obtain multiple pieces of training overhead conductor data and acceptance standard information, and perform sample augmentation processing on each piece of training overhead conductor data by using a pre-trained generation network to obtain a conductor feature set;

[0220] A training module 302, configured to input the conductor feature set into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data;

[0221] A construction module 303, configured to construct a conductor knowledge graph by using the target feature data and acceptance standard information;

[0222] A response module 304, configured to, when receiving a to-be-detected overhead conductor image, input the to-be-detected overhead conductor image into the target defect detection model to obtain to-be-detected feature data;

[0223] A detection module 305, configured to input the to-be-detected feature data into the conductor knowledge graph to obtain an overhead conductor detection result.

[0224] Further, the augmentation module 301 includes:

[0225] A preprocessing sub-module for preprocessing defective overhead conductor images of each training overhead conductor data to obtain a plurality of first sample images;

[0226] A generation sub-module for respectively inputting normal overhead conductor images of each training overhead conductor data into a pre-trained generation network to obtain a plurality of second sample images;

[0227] A first construction sub-module for constructing a conductor feature set by using each first sample image and each second sample image.

[0228] Furthermore, the generation network includes a generator and a discriminator, and the generation sub-module includes:

[0229] A generation unit for respectively inputting normal overhead conductor images of each training overhead conductor data and preset attribute control data into the generator to obtain a plurality of conductor hidden danger images, wherein an adaptive noise injection module is connected after the convolutional layer in the generator;

[0230] A fusion unit for respectively fusing each conductor hidden danger image and each normal overhead conductor image to obtain a plurality of initial hidden danger sample images;

[0231] A screening unit for inputting each initial hidden danger sample image into the discriminator for screening to obtain a plurality of second sample images.

[0232] Furthermore, the training module 302 includes:

[0233] A training sub-module for training by inputting the conductor feature set into a preset YOLOv10 defect detection model until a preset stop condition is met to obtain a target defect detection model;

[0234] A detection sub-module for inputting the conductor feature set into the target defect detection model to obtain target feature data.

[0235] Furthermore, the target defect detection model includes a backbone network, an intermediate module and a detection head group, and the response module 304 includes:

[0236] A feature extraction sub-module for extracting features of the overhead conductor image to be measured by using the backbone network and outputting a plurality of conductor feature maps group by group;

[0237] A downsampling sub-module for continuously downsampling a plurality of conductor feature maps through the intermediate module to obtain a variety of fused feature maps;

[0238] A defect detection sub-module for detecting defects in various fused feature maps by using the detection head group to obtain the feature data to be measured.

[0239] Further, the backbone network includes a first convolutional group, a second convolutional group, and a third convolutional group. The feature extraction sub-module includes:

[0240] A first extraction unit for extracting features from the image of the overhead wire to be measured through the first convolutional group to obtain a first wire feature map, where the first convolutional group includes a convolutional layer, a star operation module, a convolutional layer, a star operation module, and a convolutional layer connected in sequence;

[0241] A second extraction unit for extracting features from the first wire feature map through the second convolutional group to obtain a second wire feature map, where the second convolutional group includes a star operation module and a convolutional layer connected in sequence;

[0242] A third extraction unit for extracting features from the second wire feature map through the third convolutional group to obtain a third wire feature map, where the third convolutional group includes a star operation module, a convolutional layer, a spatial pyramid pooling module, and a partial attention module connected in sequence.

[0243] Further, the star operation module includes a depthwise separable convolutional layer, a first fully connected layer, a second fully connected layer, a feature fusion layer, a fully connected layer, and a depthwise separable convolutional layer. The second extraction unit includes:

[0244] A first extraction subunit for extracting features from the first wire feature map through the depthwise separable convolutional layer to obtain a first depth wire feature map;

[0245] A first transformation subunit for linearly transforming the first depth wire feature map through the first fully connected layer to obtain a second depth wire feature map;

[0246] A second transformation subunit for linearly transforming the first depth wire feature map through the second fully connected layer to obtain a third depth wire feature map;

[0247] A first fusion subunit for fusing the features of the second depth wire feature map and the third depth wire feature map using the feature fusion layer to obtain a fourth depth wire feature map;

[0248] A second extraction subunit for extracting features from the fourth depth wire feature map through the fully connected layer and the depthwise separable convolutional layer in sequence to obtain a second wire feature map.

[0249] Further, the intermediate module includes a first fusion module, a second fusion module, a downsampling module, and a third fusion module. The downsampling sub-module includes:

[0250] A first fusion unit for fusing the features of the third wire feature map and the second wire feature map through the first fusion module to obtain a first wire fusion feature map, where the first fusion module includes an upsampling module and a feature fusion layer;

[0251] A second fusion unit, configured to perform feature fusion on the first wire fusion feature map and the first wire feature through a second fusion module to obtain a first fusion feature map, where the second fusion module includes a first cross-stage fusion module, an upsampling module, a feature fusion layer, and a first cross-stage fusion module;

[0252] A third fusion unit, configured to perform a downsampling operation on the first wire fusion feature map through a downsampling module to obtain a second fusion feature map, where the downsampling module includes a first basic module, a splicing module, and a first cross-stage fusion module connected in sequence;

[0253] A fourth fusion unit, configured to perform feature fusion on the second fusion feature map and the third wire feature map through a third fusion module to obtain a third fusion feature map, where the third fusion module includes a channel downsampling module, a feature fusion layer, and a second cross-stage fusion module.

[0254] Further, the third fusion unit includes:

[0255] A third extraction subunit, configured to perform feature extraction on the first wire fusion feature map through a first basic module to obtain a first target fusion feature map, where the first basic module includes a two-dimensional convolutional layer, a batch normalization layer, and a SiLU activation function layer connected in sequence;

[0256] A second fusion subunit, configured to perform feature extraction on the first target fusion feature map through a splicing module and a first cross-stage fusion module in sequence to obtain a second fusion feature map.

[0257] Further, the fourth fusion unit includes:

[0258] A sampling subunit, configured to perform a sampling operation on the second fusion feature map by using a channel downsampling module to obtain a target sampling feature map;

[0259] A third fusion subunit, configured to perform feature fusion on the target sampling feature map and the third wire feature map through a feature fusion layer to obtain an intermediate fusion feature map;

[0260] A fourth extraction subunit, configured to perform feature extraction on the intermediate fusion feature map through a second cross-stage fusion module to obtain a third fusion feature map.

[0261] Further, the defect detection sub-module includes:

[0262] A matching unit, configured to match a corresponding detection head from a detection head group according to the resolution of the fusion feature map;

[0263] The defect detection unit is used to detect defects in the fused feature map by using a detection head to obtain the to-be-detected feature data, where the detection head includes a first group normalization convolution module, a second group normalization convolution module, a second group normalization convolution module, and a prediction convolution module connected in sequence.

[0264] Further, a scaling module is connected after the prediction box convolution module in the prediction convolution module.

[0265] Please refer to Figure 8 , Figure 8 which is a structural block diagram of an electronic device provided in the fourth embodiment of the present invention.

[0266] An electronic device according to an embodiment of the present invention, the electronic device includes: a memory 401 and a processor 402, and a computer program is stored in the memory 401; when the computer program is executed by the processor 402, the processor 402 is caused to execute the knowledge graph-based and improved YOLOv10 overhead wire detection method according to any of the above embodiments.

[0267] The memory 401 can be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk, or a ROM. The memory 401 has a storage space 403 for program codes 413 for executing any method steps in the above methods. For example, the storage space 403 for program codes can include respective program codes 413 for implementing various steps in the above methods. These program codes can be read out from or written into one or more computer program products. These computer program products include program code carriers such as hard disks, compact discs (CDs), memory cards, or floppy disks. The program codes can be compressed in an appropriate form. When these codes are run by a computing processing device, the computing processing device is caused to execute the respective steps in the methods described above.

[0268] The fifth embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the knowledge graph-based and improved YOLOv10 overhead wire detection method according to any of the above embodiments is implemented.

[0269] The sixth embodiment of the present invention also provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer, the computer is caused to execute the knowledge graph-based and improved YOLOv10 overhead wire detection method according to any of the above embodiments.

[0270] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0271] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0272] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0273] In addition, each functional unit in various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0274] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0275] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting overhead conductors based on a knowledge graph and improved YOLOv10, characterized in that, Including: Obtain multiple pieces of training overhead conductor data and acceptance standard information, and use a pre-trained generation network to perform sample augmentation processing on each of the training overhead conductor data to obtain a conductor feature set; Use the conductor feature set to input into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data; Construct a conductor knowledge graph using the target feature data and the acceptance standard information; When receiving an image of a to-be-tested overhead conductor, input the image of the to-be-tested overhead conductor into the target defect detection model to obtain to-be-tested feature data; Input the to-be-tested feature data into the conductor knowledge graph to obtain an overhead conductor detection result.

2. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 1, wherein The step of using a pre-trained generation network to perform sample augmentation processing on each of the training overhead conductor data to obtain a conductor feature set includes: Perform image preprocessing on the defective overhead conductor images of each of the training overhead conductor data to obtain multiple first sample images; Respectively input the normal overhead conductor images of each of the training overhead conductor data into the pre-trained generation network to obtain multiple second sample images; Construct a conductor feature set using each of the first sample images and each of the second sample images.

3. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 2, wherein, The generation network includes a generator and a discriminator. The step of respectively inputting the normal overhead conductor images of each of the training overhead conductor data into the pre-trained generation network to obtain multiple second sample images includes: Respectively input the normal overhead conductor images of each of the training overhead conductor data and preset attribute control data into the generator to obtain multiple conductor hidden danger images, where an adaptive noise injection module is connected after the convolutional layer in the generator; Respectively fuse each of the conductor hidden danger images and each of the normal overhead conductor images to obtain multiple initial hidden danger sample images; Input each of the initial hidden danger sample images into the discriminator for screening to obtain multiple second sample images.

4. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 1, wherein The step of using the conductor feature set to input into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data includes: Use the conductor feature set to input into a preset YOLOv10 defect detection model for training until a preset stop condition is met to obtain a target defect detection model; Input the conductor feature set into the target defect detection model to obtain target feature data.

5. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 1, wherein The target defect detection model includes a backbone network, an intermediate module, and a detection head group. The step of inputting the image of the to-be-tested overhead conductor into the target defect detection model to obtain to-be-tested feature data includes: Use the backbone network to perform feature extraction on the image of the to-be-tested overhead conductor and output multiple conductor feature maps group by group; Perform continuous downsampling on the multiple conductor feature maps through the intermediate module to obtain multiple types of fused feature maps; Use the detection head group to perform defect detection on various fused feature maps to obtain to-be-tested feature data.

6. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 5, characterized in that The backbone network includes a first convolutional group, a second convolutional group, and a third convolutional group. The step of using the backbone network to extract features from the image of the overhead wire to be measured and outputting multiple wire feature maps group by group includes: Extracting features from the image of the overhead wire to be measured through the first convolutional group to obtain a first wire feature map, where the first convolutional group includes a convolutional layer, a star operation module, a convolutional layer, a star operation module, and a convolutional layer connected in sequence; Extracting features from the first wire feature map through the second convolutional group to obtain a second wire feature map, where the second convolutional group includes a star operation module and a convolutional layer connected in sequence; Extracting features from the second wire feature map through the third convolutional group to obtain a third wire feature map, where the third convolutional group includes a star operation module, a convolutional layer, a spatial pyramid pooling module, and a partial attention module connected in sequence.

7. The overhead conductor detection method based on the knowledge graph and improved YOLOv10 according to claim 6, wherein The star operation module includes a depthwise separable convolutional layer, a first fully connected layer, a second fully connected layer, a feature fusion layer, a fully connected layer, and a depthwise separable convolutional layer. The step of extracting features from the first wire feature map through the second convolutional group to obtain a second wire feature map includes: Extracting features from the first wire feature map through the depthwise separable convolutional layer to obtain a first depth wire feature map; Performing a linear transformation on the first depth wire feature map through the first fully connected layer to obtain a second depth wire feature map; Performing a linear transformation on the first depth wire feature map through the second fully connected layer to obtain a third depth wire feature map; Using the feature fusion layer to fuse the features of the second depth wire feature map and the third depth wire feature map to obtain a fourth depth wire feature map; Sequentially extracting features from the fourth depth wire feature map through the fully connected layer and the depthwise separable convolutional layer to obtain a second wire feature map.

8. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 6, wherein the intermediate module includes a first fusion module, a second fusion module, a downsampling module, and a third fusion module. The step of continuously downsampling the plurality of conductor feature maps through the intermediate module to obtain various fusion feature maps is characterized in that, Including: Fusing the features of the third wire feature map and the second wire feature map through the first fusion module to obtain a first wire fusion feature map, where the first fusion module includes an upsampling module and a feature fusion layer; Fusing the features of the first wire fusion feature map and the first wire feature through the second fusion module to obtain a first fusion feature map, where the second fusion module includes a first cross-stage fusion module, an upsampling module, a feature fusion layer, and a first cross-stage fusion module; Performing a downsampling operation on the first wire fusion feature map through the downsampling module to obtain a second fusion feature map, where the downsampling module includes a first basic module, a splicing module, and a first cross-stage fusion module connected in sequence; Fusing the features of the second fusion feature map and the third wire feature map through the third fusion module to obtain a third fusion feature map, where the third fusion module includes a channel downsampling module, a feature fusion layer, and a second cross-stage fusion module.

9. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 5, wherein The step of using the detection head group to detect defects in various fusion feature maps to obtain the feature data to be measured includes: Match the corresponding detection head from the detection head group according to the resolution of the fused feature map; Use the detection head to perform defect detection on the fused feature map to obtain the to-be-tested feature data, where the detection head includes a first group normalization convolution module, a second group normalization convolution module, a second group normalization convolution module, and a prediction convolution module connected in sequence.

10. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 9, characterized in that, A scaling module is connected after the prediction box convolution module in the prediction convolution module.

11. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 8, wherein, The step of obtaining the second fused feature map by performing downsampling operation on the first wire fused feature map through the downsampling module includes: Perform feature extraction on the first wire fused feature map through the first basic module to obtain a first target fused feature map, where the first basic module includes a two-dimensional convolution layer, a batch normalization layer, and a SiLU activation function layer connected in sequence; Perform feature extraction on the first target fused feature map through the splicing module and the first cross-stage fusion module in sequence to obtain the second fused feature map.

12. The overhead conductor detection method based on a knowledge graph and improved YOLOv10 according to claim 8, wherein The step of obtaining the third fused feature map by performing feature fusion on the second fused feature map and the third wire feature map through the third fusion module includes: Use the channel downsampling module to perform sampling operation on the second fused feature map to obtain the target sampling feature map; Perform feature fusion on the target sampling feature map and the third wire feature map through the feature fusion layer to obtain the intermediate fused feature map; Perform feature extraction on the intermediate fused feature map through the second cross-stage fusion module to obtain the third fused feature map.

13. An overhead conductor detection system based on a knowledge graph and improved YOLOv10, characterized in that, Include: An expansion module for obtaining a plurality of training overhead wire data and acceptance standard information, and performing sample expansion processing on each of the training overhead wire data by using a pre-trained generation network to obtain a wire feature set; A training module for inputting the wire feature set into a preset YOLOv10 defect detection model for training to generate a target defect detection model and target feature data; A construction module for constructing a wire knowledge graph by using the target feature data and the acceptance standard information; A response module for inputting the to-be-tested overhead wire image into the target defect detection model to obtain the to-be-tested feature data when receiving the to-be-tested overhead wire image; A detection module for inputting the to-be-tested feature data into the wire knowledge graph to obtain the overhead wire detection result.

14. An electronic device, characterized in that, It includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the method for detecting overhead wires based on a knowledge graph and improved YOLOv10 as described in any one of claims 1-12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the method for detecting overhead wires based on a knowledge graph and improved YOLOv10 as described in any one of claims 1-12.

Images