Foreign matter identification method, device and equipment for power transmission line, medium and product

Abstract

The invention discloses a foreign matter identification method and device for a power transmission line, equipment, a medium and a product, and belongs to the technical field of power transmission line inspection, and the foreign matter identification method comprises the steps: obtaining an image of a target region power transmission line, and carrying out the enhancement processing of the image, and obtaining an enhanced image; a YOLOv8 model is adopted as a basic model, and the YOLOv8 model is improved by fusing Mobile-VIT-GCL to obtain a foreign matter recognition model; adding a foreign matter label to the enhanced image to generate a training sample, and training a foreign matter recognition model through the training sample; and performing foreign matter identification on the power transmission line through the trained foreign matter identification model. According to the method, the YOLOv8 model is improved by fusing Mobile-VIT-GCL to obtain the foreign matter recognition model, the anti-interference capability of the model under the complex weather condition is enhanced, and the real-time performance, stability and precision of foreign matter recognition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transmission line inspection, and in particular to a method, device, equipment, medium and product for identifying foreign objects in a transmission line. Background Art

[0002] High-voltage power lines are exposed to the elements year-round, and foreign objects often become lodged in overhead transmission lines. This damages the lines and, in severe cases, disrupts the normal operations of residents and businesses. Traditional methods for detecting foreign objects on transmission lines suffer from low accuracy and efficiency in complex weather conditions, making them unable to meet the power system's requirements for high real-time performance, stability, and precision.

[0003] For example, patent publication number CN119625577A discloses a method for detecting foreign objects in power transmission lines, which belongs to the field of image recognition technology. By constructing a power transmission line foreign object detection model based on YOLOv8, the model includes a backbone network, a neck network, and a head network connected in sequence, and replacing the C2F module in the backbone network and the neck network with a DMSA module, the DMSA module accumulates the feature mapping of multiple receptive fields, so that the detection results retain more delicate multi-scale features, and reduce the amount of calculation while ensuring that feature information is not lost, significantly improving the efficiency of power transmission line foreign object detection. However, the invention makes too few changes to the adaptability of the main model itself. Since power transmission line foreign objects have characteristics such as diverse morphologies and complex distribution environments, there is still much room for improvement in anti-interference, efficiency, and accuracy for power transmission line foreign object detection. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, device, equipment, medium and product for identifying foreign objects in transmission lines, so as to solve technical problems such as low efficiency of traditional transmission line detection methods and difficulty in detection under complex weather conditions.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In a first aspect, the present invention provides a method for identifying foreign objects in a power transmission line, comprising:

[0007] Acquiring an image of the power transmission line in the target area, and performing enhancement processing on the image to obtain an enhanced image;

[0008] The YOLOv8 model is used as the basic model, and the foreign object recognition model is obtained by integrating the Mobile-VIT-GCL to improve the YOLOv8 model;

[0009] adding a foreign body label to the enhanced image to generate a training sample, and training the foreign body recognition model using the training sample;

[0010] Foreign objects on the power transmission line are identified using the trained foreign object identification model.

[0011] Optionally, performing enhancement processing on the image to obtain an enhanced image includes:

[0012] Using the Cutout method to randomly crop regions in the image;

[0013] The pixels in the cropped area are replaced by black pixels or by the pixels in the patch block using the CutMix method.

[0014] Optionally, the foreign object recognition model obtained by improving the YOLOv8 model by fusing Mobile-VIT-GCL includes:

[0015] The Conv module in the fifth layer of the backbone network of the YOLOv8 model is replaced with the Mobile-VIT module, and the CAFM module is added after the SPPF module of the backbone network;

[0016] Replace the Bottleneck in the C2f module of the Neck network of the YOLOv8 model with the StarBlocks module in the StarNet network;

[0017] The Head network of the YOLOv8 model is modified into three 1×1 convolutional layers of different scales and one 3×3 shared convolutional layer.

[0018] Optionally, modifying the YOLOv8 model further includes:

[0019] The MV2 module and the inverted residual layer in the Mobile-VIT module are streamlined and removed, and a GCL layer is introduced after the Mobile-VIT module.

[0020] In a second aspect, the present invention provides a foreign object identification device for a power transmission line, comprising:

[0021] an image processing module configured to acquire an image of the power transmission line in the target area and perform enhancement processing on the image to obtain an enhanced image;

[0022] The model building module is configured to use the YOLOv8 model as the base model and improve the YOLOv8 model by integrating Mobile-VIT-GCL to obtain the foreign object recognition model;

[0023] A model training module is configured to add foreign body labels to the enhanced image to generate training samples, and train the foreign body recognition model using the training samples;

[0024] The model application module is configured to perform foreign object recognition on the power transmission line through the trained foreign object recognition model.

[0025] Optionally, the foreign object recognition model obtained by improving the YOLOv8 model by fusing Mobile-VIT-GCL includes:

[0026] The Conv module in the fifth layer of the backbone network of the YOLOv8 model is replaced with the Mobile-VIT module, and the CAFM module is added after the SPPF module of the backbone network;

[0027] Replace the Bottleneck in the C2f module of the Neck network of the YOLOv8 model with the StarBlocks module in the StarNet network;

[0028] The Head network of the YOLOv8 model is modified into three 1×1 convolutional layers of different scales and one 3×3 shared convolutional layer.

[0029] Optionally, modifying the YOLOv8 model further includes:

[0030] The MV2 module and the inverted residual layer in the Mobile-VIT module are streamlined and removed, and a GCL layer is introduced after the Mobile-VIT module.

[0031] In a third aspect, the present invention provides an electronic device, including a processor and a storage medium;

[0032] The storage medium is used to store instructions;

[0033] The processor is configured to operate according to the instructions to execute the steps of the above method.

[0034] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0035] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] This invention provides a method, device, equipment, medium, and product for identifying foreign objects on power transmission lines. This model uses the YOLOv8 model as its base model and improves it by integrating Mobile-VIT-GCL. This model enhances its anti-interference capabilities in complex weather conditions and improves its understanding of contextual information. While maintaining a lightweight design, it enables the model to detect more small objects. This method meets the power system's requirements for high real-time performance, stability, and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 1 is a flow chart of a method for identifying foreign objects in a power transmission line according to an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of the structure of the YOLOv8 model provided by an embodiment of the present invention;

[0040] Figure 3 Schematic diagram of the pruning structure of the Mobile-VIT module provided in an embodiment of the present invention;

[0041] Figure 4 is a schematic diagram of the structure of a gated convolutional layer provided by an embodiment of the present invention;

[0042] Figure 5 1 is a schematic diagram of the structure of the StarNet network and the Neck network improvements provided by an embodiment of the present invention;

[0043] Figure 6 Schematic diagram of the structure of a CAFM module provided by an embodiment of the present invention;

[0044] Figure 7 3 is a schematic structural diagram of a convolution-based lightweight detection head provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0046] Example 1:

[0047] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying foreign objects in a power transmission line, comprising the following steps:

[0048] Step S1: Acquire an image of the power transmission line in the target area, and perform enhancement processing on the image to obtain an enhanced image.

[0049] The means of acquiring images include first installing image acquisition devices around the transmission lines to capture line status information in real time; then, using the 5G communication wireless network, the collected image data is transmitted to the ground processing center to achieve remote monitoring, avoiding the potential dangers of manual tower inspections; since the amount of collected data is relatively limited, the image data is enhanced for efficient processing to improve the accuracy of subsequent target detection tasks.

[0050] Image enhancement methods are mainly divided into single image enhancement and multiple image enhancement. Image enhancement methods such as Cutout, Mixup, and Cutmix are widely used in current image enhancement methods due to their advantages such as simple operation and strong practicality. Specifically in this embodiment, the image enhancement process to obtain the enhanced image includes:

[0051] Cutout method is used to randomly crop the area in the image;

[0052] The pixels in the cropped area are replaced by black pixels or by the pixels in the patch block using the CutMix method.

[0053] Step S2: Use the YOLOv8 model as the basic model, and improve the YOLOv8 model by integrating Mobile-VIT-GCL to obtain a foreign object recognition model.

[0054] The detection and identification of foreign objects on transmission lines must meet the requirements of high accuracy, high real-time performance, and good stability. By comparing the performance of various existing target detection algorithms and comprehensively considering indicators such as effectiveness and accuracy, the YOLOv8 model was adopted.

[0055] like Figure 2 As shown, YOLOv8 is an object detection model released open source by Ultralytics on January 10, 2023. As the latest version of the YOLO (You Only Look Once) series, it integrates multiple innovative technologies based on YOLOv5, making it an efficient, flexible, and versatile state-of-the-art (SOTA) model in the field of computer vision. YOLOv8 continues the YOLO series' single-stage detector design. Its overall architecture consists of three parts: Backbone (backbone network), Neck (feature fusion network), and Head (detection head). It directly predicts object categories and bounding boxes in an end-to-end manner.

[0056] The backbone network generally consists of a convolutional module, a SPPF module, and a C2f module to extract features from input transmission line inspection images. The C2f module reduces computational complexity while ensuring global feature acquisition and integrating information at different scales. The SPPF module stitches feature maps of different scales to improve the detection of foreign objects of varying sizes on transmission lines.

[0057] The Neck network is based on the PANet (Path Aggregation Network) structure, which optimizes the top-down and bottom-up feature fusion paths, removes redundant convolutional layers, and further enhances the information transmission efficiency of multi-scale features. It introduces the SPPF module to pool and splice feature maps of different scales, enhancing the model's detection sensitivity for small targets.

[0058] The Head network is mainly used to detect and classify foreign objects on transmission lines.

[0059] Although the YOLOv8 model has undergone extensive optimizations based on previous generations of YOLO models, it still suffers from low accuracy and efficiency in the complex environments studied by the present invention for detecting foreign objects on power lines. This is primarily due to the fact that foreign objects in the power lines studied by the present invention may be small in size and have complex backgrounds. This results in the YOLOv8 model's insufficient global information perception and slow reasoning speed. To address these issues, the present invention proposes to improve the YOLOv8 model by integrating Mobile-VIT-GCL to obtain a foreign object recognition model.

[0060] Specifically in this embodiment, the foreign object recognition model obtained by improving the YOLOv8 model by integrating Mobile-VIT-GCL includes:

[0061] (1) The Conv module of the fifth layer of the backbone network of the YOLOv8 model is replaced with the Mobile-VIT module. The Transformer layer in the Mobile-VIT module is combined with its own multi-head self-attention mechanism and feedforward structure to extract the features that can be used as distinguishing features in the transmission line foreign object detection image.

[0062] like Figure 3As shown in the figure, compared to the Conv module before the improvement, the Mobile-VIT module, due to the self-attention mechanism contained in its own Transformer layer, can better capture the global information of the input foreign object detection image, and can also achieve weighted features of different regions, directly improving the feature expression ability of the entire YOLOv8 model. Therefore, the introduction of the Mobile-VIT module can improve the YOLOv8 model's ability to detect foreign objects on power transmission lines in complex environments, and at the same time improve the YOLOv8 model's ability to resist interference in complex environments, and can effectively identify small targets in complex backgrounds. Although the introduction of the Mobile-VIT module can improve the detection accuracy of the YOLOv8 model, it still has deficiencies in indicators such as computational cost and real-time performance. Therefore, it is necessary to further lightweight the YOLOv8 model. Currently, model scaling is one of the commonly used lightweight design methods. However, it usually only simply increases the width and depth of the network, often resulting in gradient vanishing and overfitting problems. The present invention decides to prune the Mobile-VIT module, removing and streamlining some of the MV2 modules and inverted residual layers. By pruning the Mobile-VIT module, the number of parameters and computational complexity of the overall model can be reduced, thereby solving the computational cost problem of the model.

[0063] Although streamlining the MV2 and inverted residual layers in the Mobile-VIT module improves the efficiency of the model, the MV2 and mainly play the role of image feature extraction. After streamlining, the feature extraction ability of the overall model will decrease. The GCL (gated convolution) layer can adjust the weights of the input image features without the need to introduce additional model parameters. Therefore, introducing the gated convolution layer after the Mobile-VIT module has basically no effect on the overall size of the model and can improve the representation ability of the YOLOv8 model. Figure 4 As shown in the figure, dynamic adjustment of the gated convolutional layer can dynamically adjust the information flow in the graph, obtain more contextual correlation information, and improve the model's ability to understand features. Ultimately, this achieves a lightweight improvement to the YOLOv8 model, improving the model's detection accuracy while reducing the model's computational cost.

[0064] (2) Improve the Neck network. The Neck network in the YOLOv8 model mainly constructs its own attention matrix through point multiplication. This method will increase the computational complexity as the feature extraction capability increases, thereby reducing the complexity of the entire model. The introduction of the Mobile-VIT module and the gated convolutional layer greatly increases the feature extraction capability of the YOLOv8 model. Therefore, it is necessary to improve the Neck network to adapt to the overall YOLOv8 model. The StarNet network is currently a widely used network in computer vision and artificial intelligence. It has an ultra-lightweight network structure that reduces the complexity of the model while improving the feature extraction capability of the model.

[0065] The StarNet network is mainly based on star operations, which maps the features of the model input to a high-dimensional feature space. Its main principle is to operate the features of the two linear transformations obtained by element-wise multiplication:

[0066]

[0067] Where, are the weight matrices of two linear transformations, For the input and output of the star operation, are the biases of the two linear transformations, This is a star operation.

[0068] like Figure 5 As shown in the figure, the StarNet network mainly includes a four-level structure, and each level relies on a combination of convolutional layer downsampling and StarBlocks feature extraction; it can be seen that StarBlocks mainly uses DWConv to perform deep convolution to obtain the preliminary features of the image; on this basis, batch normalization operations are performed to finally obtain feature maps with expanded width and channels, and then the feature maps are transferred to the second DWConv through the ReLU function for further feature extraction; the DWConv convolution operation can improve the training and reasoning capabilities of the model while reducing the amount of calculation; however, the overall StarNet network is relatively complex, and its structure is also relatively bloated, which is not suitable for introducing the entire StarNet network into the YOLOv8 model to improve the Neck network; therefore, the present invention introduces the StarBlocks module in the StarNet network into the Neck network, and uses StarBlocks to replace the Bottleneck in the original C2f module, thereby forming the following Figure 5 The C2f_Star module shown.

[0069] Optimizing the Neck network through the constructed C2f_Star module not only improves the model's feature extraction capabilities and object detection accuracy, but also introduces StarBlocks, which have fewer parameters, significantly reducing the computational complexity of the original C2f module. Ultimately, this significantly increases the model's real-time performance for foreign object detection on power lines and reduces the overall computational cost of the model.

[0070] (3) The feature extraction capability of the backbone network of the traditional YOLOv8 model is still relatively weak, so the present invention decided to introduce an attention module to increase the model's attention to some key feature information, especially the attention to small target information in complex backgrounds. At present, the more common attention modules include CBAM, GAM and SENet, etc. These modules often have very limited ability to focus on small targets in complex environments, and the introduction of these modules tends to increase the computational complexity of the model, which is inconsistent with the previous model lightweight design. In order to solve this problem, the present invention introduces a convolutional attention fusion module (CAFM) to increase the local and global feature modeling capabilities of the YOLOv8 model, thereby improving the overall feature extraction capability of the backbone network while ensuring the lightweight design of the network.

[0071] like Figure 6 As shown in the figure, the CAFM module mainly includes global and local branches. The channel dimension of the local branch is adjusted through 1×1 convolution. This can greatly enhance the model's cross-channel interaction ability and improve the ability to integrate all input power line foreign object detection images. After the channel cleaning operation, the information between different channels is mixed. On this basis, the output tensors obtained in each group are cascaded according to the actual channel dimension to obtain a new tensor. Finally, the new tensor is extracted with the help of 3×3 convolution operation. The calculation of the entire local branch can be expressed as:

[0072]

[0073] Where, and They are the feature maps output in the local branch and the 3×3×3 convolution operation, For the mixed cleaning operation of the channel, is a 1×1 convolution, It represents the corresponding feature map of the input.

[0074] In the global branch of the CAFM module, advanced 1×1 and 3×3 deep convolution operations are used to generate query (Q), key (K), and value (V). Based on the reshaping of Q, K, and V, the attention matrix between Q and K is obtained, and the output feature map of the global branch is derived:

[0075]

[0076] Where, is the final feature map output by the global branch, where Q, K, and V represent the query, key, and value obtained by the deep convolution, respectively. Represent the Q, K, and V reshaping results respectively, is a scalable parameter that mainly controls The size of the matrix multiplication between them;

[0077] Based on the output feature maps of the global branch and the local branch, the output of the CAFM module can be further converted to Expressed as:

[0078]

[0079] In light of the characteristics of the YOLOv8 model's backbone network, the present invention decided to add a CAFM module after the SPPF module in the backbone network. This allows for further global and local feature extraction of the feature maps output by the SPPF module, thereby improving the backbone network's feature extraction capabilities for transmission line foreign object detection images.

[0080] (4) The detection head of the traditional YOLOv8 model plays an important role in the entire model, but it takes up a lot of computing resources. Therefore, in order to improve the detection efficiency of the model, the detection head must be designed to be lightweight. Figure 7 As shown, the detection heads in the YOLOv8 model are distributed across different feature layers. Each feature layer has a separate detection head, which results in a huge amount of computation in actual calculations. Independent detection heads are also relatively inefficient. Therefore, this invention uses a shared detection head approach as the basis for a lightweight detection head design.

[0081] Currently, the common FCOS and RetinaNet basically adopt the idea of shared convolution, which can not only improve the ability of foreign body detection, but also reduce the number of actual parameters. Based on the idea of convolution weight sharing, this paper merges each group of independent convolutions of the YOLOv8 model. The specific structure of the improved convolution-based lightweight detection head (LSC-Detect) is as follows: Figure 6As shown in the figure, the improved detection head primarily consists of three 1×1 convolution operations on feature layers at different scales. After each convolution operation, a subsequent 3×3 shared convolution operation is performed. This shared convolution approach significantly reduces parameters, enabling the detection model to achieve high detection accuracy while keeping the model lightweight. Furthermore, in the design of a shared detection head for object detection, ConvCls generally refers to the ClassificationConvolution module, a convolutional layer used to generate class predictions. It classifies the features extracted by the detection head and outputs the class probability distribution for each anchor box or candidate box. Specifically, ConvCls is a submodule in the detection head that sits alongside ConvReg (regression convolution, responsible for bounding box coordinate adjustment). It takes a shared feature map as input and outputs the class scores for each anchor box at each spatial location (e.g., (H, w, num_anchors x num_classes)). It typically consists of a 1×1 or 3×3 convolutional layer followed by a classifier (such as Softmax or Sigmoid). In single-stage detectors (such as YOLO and RetinaNet), the classification and regression branches often share underlying features but decouple the tasks through different convolutional layers.

[0082] In summary, the improved YOLOv8 model replaces the fifth Conv layer in the model's backbone network with a pruned Mobile-ViT module, and adds a gated convolutional layer (GCL) after the Mobile-ViT module to adjust the weights of the input image features. This ensures improved detection accuracy while maintaining a lightweight design. Furthermore, the CAFM module is added after the SPPF module to further improve the backbone network, enhancing its feature extraction capabilities for power line foreign object detection images. The C2f_Star module then optimizes the Neck network, improving the model's feature extraction capabilities and object detection accuracy while reducing the overall computational cost. Finally, the model's detection head is improved, with a convolution-based lightweight detection head (LSC-Detct) proposed. This significantly reduces parameters, enabling the detection model to detect small objects in complex backgrounds while remaining as lightweight as possible.

[0083] Step S3: adding foreign body labels to the enhanced image to generate training samples, and training the foreign body recognition model through the training samples.

[0084] Step S4: Identify foreign objects on the power transmission line using the trained foreign object identification model.

[0085] After improving the YOLOv8 model by integrating Mobile-VIT-GCL to obtain a foreign object recognition model, the corresponding algorithm is used to analyze the image, train data, and test to determine whether there are foreign objects in the image and accurately locate their position. This process can achieve real-time monitoring and early warning of foreign objects on the transmission line, thereby improving inspection efficiency, reducing the risk of tower inspections, and ensuring the safe and stable operation of the transmission line.

[0086] Example 2:

[0087] An embodiment of the present invention provides a foreign object identification device for a power transmission line, comprising:

[0088] An image processing module is configured to acquire an image of the power transmission line in the target area and perform enhancement processing on the image to obtain an enhanced image;

[0089] The model building module is configured to use the YOLOv8 model as the base model and improve the YOLOv8 model by integrating Mobile-VIT-GCL to obtain the foreign object recognition model;

[0090] A model training module is configured to add foreign body labels to the enhanced images to generate training samples, and train the foreign body recognition model through the training samples;

[0091] The model application module is configured to identify foreign objects on the transmission line through the trained foreign object recognition model.

[0092] Specifically, the foreign object recognition model obtained by integrating Mobile-VIT-GCL to improve the YOLOv8 model includes:

[0093] The Conv module in the fifth layer of the YOLOv8 model's backbone network is replaced with the Mobile-VIT module. The MV2 module and inverted residual layer in the Mobile-VIT module are streamlined and removed. The GCL layer is introduced after the Mobile-VIT module, and the CAFM module is added after the SPPF module in the backbone network.

[0094] Replace the Bottleneck in the C2f module of the Neck network of the YOLOv8 model with the StarBlocks module in the StarNet network;

[0095] The Head network of the YOLOv8 model is modified into three 1×1 convolutional layers of different scales and one 3×3 shared convolutional layer.

[0096] Example 3:

[0097] Based on the foreign object identification method for a power transmission line provided in the first embodiment, an embodiment of the present invention provides an electronic device including a processor and a storage medium;

[0098] The storage medium is used to store instructions;

[0099] The processor is configured to operate according to the instructions to execute the steps of the above method.

[0100] Example 4:

[0101] Based on the foreign object identification method for a power transmission line provided in the first embodiment, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0102] Embodiment 5:

[0103] Based on the foreign object identification method for a power transmission line provided in the first embodiment, an embodiment of the present invention provides a computer program product, including a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0104] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, 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 magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] 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 flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, 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 flowcharts and / or block diagrams. 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.

[0106] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work 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 The function specified in one or more boxes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0108] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying foreign objects in a power transmission line, characterized in that: include: Acquiring an image of the power transmission line in the target area, and performing enhancement processing on the image to obtain an enhanced image; The YOLOv8 model is used as the basic model, and the foreign object recognition model is obtained by integrating the Mobile-VIT-GCL to improve the YOLOv8 model; adding a foreign body label to the enhanced image to generate a training sample, and training the foreign body recognition model using the training sample; Foreign objects on the power transmission line are identified using the trained foreign object identification model.

2. The method for identifying foreign objects in a power transmission line according to claim 1, wherein: The performing enhancement processing on the image to obtain an enhanced image comprises: Using the Cutout method to randomly crop regions in the image; The pixels in the cropped area are replaced by black pixels or by the pixels in the patch block using the CutMix method.

3. The method for identifying foreign objects in a power transmission line according to claim 1, wherein: The foreign object recognition model obtained by improving the YOLOv8 model by integrating Mobile-VIT-GCL includes: The Conv module in the fifth layer of the backbone network of the YOLOv8 model is replaced with the Mobile-VIT module, and the CAFM module is added after the SPPF module of the backbone network; Replace the Bottleneck in the C2f module of the Neck network of the YOLOv8 model with the StarBlocks module in the StarNet network; The Head network of the YOLOv8 model is modified into three 1×1 convolutional layers of different scales and one 3×3 shared convolutional layer.

4. The method for identifying foreign objects in a power transmission line according to claim 3, wherein: The modification of the YOLOv8 model also includes: The MV2 module and the inverted residual layer in the Mobile-VIT module are streamlined and removed, and a GCL layer is introduced after the Mobile-VIT module.

5. A foreign object identification device for a power transmission line, characterized in that: include: an image processing module configured to acquire an image of the power transmission line in the target area and perform enhancement processing on the image to obtain an enhanced image; The model building module is configured to use the YOLOv8 model as the base model and improve the YOLOv8 model by integrating Mobile-VIT-GCL to obtain the foreign object recognition model; A model training module is configured to add foreign body labels to the enhanced image to generate training samples, and train the foreign body recognition model using the training samples; The model application module is configured to perform foreign object recognition on the power transmission line through the trained foreign object recognition model.

6. The foreign object identification device for a power transmission line according to claim 5, characterized in that: The foreign object recognition model obtained by improving the YOLOv8 model by integrating Mobile-VIT-GCL includes: The Conv module in the fifth layer of the backbone network of the YOLOv8 model is replaced with the Mobile-VIT module, and the CAFM module is added after the SPPF module of the backbone network; Replace the Bottleneck in the C2f module of the Neck network of the YOLOv8 model with the StarBlocks module in the StarNet network; The Head network of the YOLOv8 model is modified into three 1×1 convolutional layers of different scales and one 3×3 shared convolutional layer.

7. The foreign object identification device for a power transmission line according to claim 6, characterized in that: The modification of the YOLOv8 model also includes: The MV2 module and the inverted residual layer in the Mobile-VIT module are streamlined and removed, and a GCL layer is introduced after the Mobile-VIT module.

8. An electronic device, characterized in that: including processors and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 4.

9. 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 the method according to any one of claims 1 to 4 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images