Power transmission line galloping monitoring method, device, equipment, medium and product

Abstract

The invention discloses a power transmission line galloping monitoring method, device and equipment, a medium and a product. The method comprises the steps that continuous multi-frame power transmission line images are acquired, and at least one spacer is arranged on a power transmission line; the multiple frames of power transmission line images are sequentially input into a target detection model, spacer detection frames of multiple sizes in each frame of power transmission line image are obtained, and the target detection model sequentially comprises a shift window backbone network, multiple extrusion attention excitation networks, a neck network and a head network from input to output; based on the spacer detection frames of multiple sizes in each frame of the power transmission line image, determining a spacer motion track; according to the technical scheme of the invention, the power transmission line galloping monitoring method and device can improve the efficiency and precision of power transmission line galloping monitoring, and effectively avoid the occurrence of power grid safety accidents.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of power grids, and in particular, to a method, device, equipment, medium and product for monitoring the galloping of transmission lines. Background Art

[0002] When the transmission line is covered with ice and snow or encounters bad weather, it may cause the transmission line to gallop. The large-amplitude and low-frequency galloping of the transmission line will cause serious harm to the transmission line or its surrounding environment. In the lightest case, flashover or tripping may occur, and in the worst case, the transmission line may break or the transmission tower may collapse, seriously threatening the safe and stable operation of the power system. Therefore, how to monitor the galloping of transmission lines has become an urgent problem to be solved currently. Summary of the Invention

[0003] The embodiments of the present invention provide a method, device, equipment, medium and product for monitoring the galloping of transmission lines, so as to improve the efficiency and accuracy of monitoring the galloping of transmission lines, and thus effectively avoid the occurrence of power grid safety accidents.

[0004] According to one aspect of the present invention, there is provided a method for monitoring the galloping of a transmission line, including:

[0005] Obtaining a plurality of consecutive frames of transmission line images, where at least one spacer damper is arranged on the transmission line;

[0006] Sequentially inputting the plurality of frames of transmission line images into the target detection model to obtain spacer damper detection frames of multiple sizes in each frame of transmission line image, where the target detection model sequentially includes from input to output: a shifted window backbone network, a plurality of squeeze-and-excitation attention networks, a neck network, and a head network;

[0007] Determining the movement trajectory of the spacer damper based on the spacer damper detection frames of multiple sizes in each frame of transmission line image;

[0008] Determining the galloping trajectory of the transmission line according to the movement trajectory of the spacer damper.

[0009] According to another aspect of the present invention, there is provided a device for monitoring the galloping of a transmission line, and the device for monitoring the galloping of a transmission line includes:

[0010] An image acquisition module, configured to obtain a plurality of consecutive frames of transmission line images, where at least one spacer damper is arranged on the transmission line;

[0011] A spacer damper detection frame recognition module, configured to sequentially input the plurality of frames of transmission line images into the target detection model to obtain spacer damper detection frames of multiple sizes in each frame of transmission line image, where the target detection model sequentially includes from input to output: a shifted window backbone network, a plurality of squeeze-and-excitation attention networks, a neck network, and a head network;

[0012] The spacer movement trajectory determination module is used to determine the spacer movement trajectory based on the spacer detection frames of multiple sizes in each frame of the transmission line image;

[0013] The transmission line galloping trajectory determination module is used to determine the transmission line galloping trajectory according to the spacer movement trajectory.

[0014] According to another aspect of the present invention, there is provided an electronic device, which includes:

[0015] At least one processor; and

[0016] A memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the transmission line galloping monitoring method according to any embodiment of the present invention.

[0018] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the transmission line galloping monitoring method according to any embodiment of the present invention when executed.

[0019] According to another aspect of the present invention, there is provided a computer program product, and the computer program implements the transmission line galloping monitoring method according to any one of the embodiments of the present invention when executed by a processor.

[0020] In the embodiment of the present invention, by acquiring consecutive multiple frames of transmission line images, at least one spacer is arranged on the transmission line; the multiple frames of transmission line images are sequentially input into the target detection model to obtain spacer detection frames of multiple sizes in each frame of the transmission line image, wherein the target detection model sequentially includes, from input to output: a shifted window backbone network, multiple squeeze-and-excitation attention networks, a neck network, and a head network; determining the spacer movement trajectory based on the spacer detection frames of multiple sizes in each frame of the transmission line image; determining the transmission line galloping trajectory according to the spacer movement trajectory, solving the problem that the galloping of the transmission line seriously threatens the safe and stable operation of the power system, being able to accurately identify the spacer detection frames of multiple sizes through the target detection model, and then determining the spacer movement trajectory according to the spacer detection frames of multiple sizes, and further determining the transmission line galloping trajectory according to the spacer movement trajectory to realize the monitoring of the transmission line galloping, thereby achieving the effect of improving the efficiency and accuracy of the transmission line galloping monitoring, and further effectively avoiding the occurrence of power grid safety accidents.

[0021] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other relevant drawings can also be obtained based on these drawings.

[0023] Figure 1 is a flowchart of a method for monitoring the galloping of a transmission line in an embodiment of the present invention;

[0024] Figure 2 is a schematic structural diagram of a model to be trained in an embodiment of the present invention;

[0025] Figure 3 is a flowchart of another method for monitoring the galloping of a transmission line in an embodiment of the present invention;

[0026] Figure 4 is a schematic structural diagram of a shifted window backbone network in an embodiment of the present invention;

[0027] Figure 5 is a schematic structural diagram of a squeeze-and-excitation attention network in an embodiment of the present invention;

[0028] Figure 6 is a schematic structural diagram of a device for monitoring the galloping of a transmission line in an embodiment of the present invention;

[0029] Figure 7 is a schematic structural diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments 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 scope of protection of the present invention.

[0031] It should be noted that the terms "first", "second", etc. in the description, claims and the above drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0032] It can be understood that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, usage scopes, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to users and the authorization of users should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0033] Embodiment 1

[0034] Figure 1 The flowchart of a transmission line galloping monitoring method provided by an embodiment of the present invention. This embodiment is applicable to the situation of transmission line galloping monitoring. This method can be executed by the transmission line galloping monitoring device in the embodiments of the present invention. The device can be implemented in a software and / or hardware manner, such as Figure 1 As shown, the method specifically includes the following steps:

[0035] S110, obtain a plurality of consecutive frames of transmission line images, and at least one spacer damper is arranged on the transmission line.

[0036] In this embodiment, the computer device can collect video data containing the transmission line through a camera, process the video image to obtain a plurality of consecutive frames of transmission line images. The computer device includes but is not limited to various personal computers, laptop computers, smart phones, robots, unmanned aerial vehicles, tablet computers, and wearable devices, etc. It is also possible to use a video recording device to perform non-contact galloping video acquisition on the transmission line to obtain a plurality of consecutive frames of transmission line images. The video recording device then sends the collected transmission line images to the computer device so that the computer device can receive the transmission line images and perform subsequent processing based on the transmission line images.

[0037] In this embodiment, the spacer damper is a fitting installed on the bundled conductor to fix the distance between the bundled conductors, prevent the conductors from whipping each other, and suppress aeolian vibration and sub-span oscillation.

[0038] It should be noted that the spacer damper is more prominent than the sub-conductors of the bundled conductors, and the galloping condition of the spacer damper can well reflect the galloping condition of the line.

[0039] S120, input the multiple frames of transmission line images into the target detection model in sequence to obtain spacer damper detection frames of multiple sizes in each frame of the transmission line image.

[0040] In this embodiment, the target detection model can be deployed in a video recording device. For example, the target detection model can be deployed in a transmission tower monitoring camera. It should be noted that the target detection model can be a model pre-trained on a server side (such as a computer device). If the target detection model is deployed in a transmission tower monitoring camera, continuous multiple frames of transmission line images are obtained through the transmission tower monitoring camera. Obtaining continuous multiple frames of transmission line images through the transmission tower monitoring camera and processing the images based on the target detection model deployed in the transmission tower monitoring camera can reduce the delay caused by image transmission and further improve the efficiency of transmission line galloping monitoring.

[0041] In this embodiment, the target detection model sequentially includes, from input to output: a shifted window backbone network, multiple squeeze-and-excitation attention networks, a neck network, and a head network. The shifted window backbone network includes: a patch partitioning layer and multiple stage layers, and the neck network includes: multiple feature fusion layers, multiple convolutional layers, and multiple upsampling layers. The head network includes: a first convolutional module and a second convolutional module. The first convolutional module includes: a convolutional layer, and the second convolutional module includes: a convolutional layer, a normalization layer, and an activation function layer.

[0042] In this embodiment, the spacer damper detection frames of multiple sizes are respectively: a 4*4 spacer damper detection frame, an 8*8 spacer damper detection frame, a 16*16 spacer damper detection frame, and a 32*32 spacer damper detection frame.

[0043] In this embodiment, the method of inputting the multiple frames of transmission line images into the target detection model in sequence to obtain spacer damper detection frames of multiple sizes in each frame of the transmission line image can be: input the transmission line image into the shifted window backbone network to obtain multiple stage feature data; input the multiple stage feature data into the corresponding squeeze-and-excitation attention networks respectively to obtain multiple enhanced feature data; input the multiple enhanced feature data into the neck network to obtain multiple fused feature data; input the multiple fused feature data into the head network to obtain spacer damper detection frames of multiple sizes.

[0044] Optionally, inputting the multiple frames of transmission line images into the target detection model in sequence to obtain spacer damper detection frames of multiple sizes in each frame of the transmission line image includes:

[0045] Input the transmission line image into the shifted window backbone network to obtain multiple stages of feature data.

[0046] In this embodiment, the shifted window backbone network includes: a tile partitioning layer and multiple stage layers.

[0047] In this embodiment, the shifted window backbone network can solve the problem of excessively high computational complexity caused by the high pixel resolution of the transmission line image.

[0048] In this embodiment, the method of inputting the transmission line image into the shifted window backbone network to obtain multiple stages of feature data can be: input the transmission line image into the tile partitioning layer to obtain multiple sub-images; input the multiple sub-images into multiple stage layers to obtain multiple stages of feature data.

[0049] Optionally, the multiple stage layers are respectively: a first stage layer, a second stage layer, a third stage layer, and a fourth stage layer;

[0050] Inputting the transmission line image into the shifted window backbone network to obtain multiple stages of feature data includes:

[0051] Input the transmission line image into the tile partitioning layer to obtain multiple sub-images.

[0052] In this embodiment, the size of the transmission line image can be 640*640*3, and the size of the sub-image can be 160*160*128.

[0053] Input the multiple sub-images into the first stage layer to obtain first-stage feature data.

[0054] In this embodiment, the size of the first-stage feature data can be 160*160*128.

[0055] Input the first-stage feature data into the second stage layer to obtain second-stage feature data.

[0056] In this embodiment, the size of the second-stage feature data can be 80*80*256.

[0057] Input the second-stage feature data into the third stage layer to obtain third-stage feature data.

[0058] In this embodiment, the size of the third-stage feature data can be 40*40*512.

[0059] Input the third-stage feature data into the fourth stage layer to obtain fourth-stage feature data.

[0060] In this embodiment, the size of the feature data in the fourth stage can be 20*20*1024.

[0061] In this embodiment, the tile partitioning layer can be the Patch Partition layer. The tile partitioning layer is used to split the picture into non-overlapping patches. The first stage layer includes, in sequence from the input to the output: a linear embedding layer and a plurality of window sliding layers. The second stage layer includes, in sequence from the input to the output: a tile merging layer and a plurality of window sliding layers. The third stage layer includes, in sequence from the input to the output: a tile merging layer and a plurality of window sliding layers. The number of window sliding layers in the second stage layer is less than the number of window sliding layers in the third stage layer. The fourth stage layer includes, in sequence from the input to the output: a tile merging layer and a plurality of window sliding layers. The number of window sliding layers in the fourth stage layer is equal to the number of window sliding layers in the second stage layer. The tile merging layer can be the PatchMerging layer, and the window sliding layer can be the Swin Transformer Block. The linear embedding layer can be the LinearEmbedding layer.

[0062] In a specific example, the transmission line image is input into the Patch Partition layer, and the picture is segmented into non-overlapping patches. The first-stage layer includes: a Linear Embedding layer and a Swin Transformer Block. Feature mapping is performed through the Linear Embedding layer, projecting it to an arbitrary dimension C, and then the data after feature mapping is input into the Swin Transformer Block. 4x downsampling is adopted in this stage. The second-stage layer includes: a Patch Merging layer and a Swin Transformer Block. The input data first undergoes downsampling through the Patch Merging layer, the features of each group of 2×2 adjacent patches are concatenated, and a linear layer is applied to the concatenated 4C-dimensional features to obtain a feature map with the length and width becoming 1 / 2 of the original and the depth becoming 2 times the original, and then it passes through the Swin Transformer Block. The third-stage layer is similar to the second-stage layer. The third-stage layer includes: a Patch Merging layer and a Swin Transformer Block. First, downsampling is performed through the Patch Merging layer to further reduce the resolution of the feature map and increase the number of channels, and then it passes through the Swin Transformer Block to achieve further extraction and fusion of features. The fourth-stage layer includes: a Patch Merging layer and a Swin Transformer Block. First, it undergoes downsampling through the Patch Merging layer, and then it passes through the Swin Transformer Block.

[0063] The multiple-stage feature data are respectively input into the corresponding squeeze-and-excitation attention network to obtain multiple enhanced feature data.

[0064] In this embodiment, the number of the enhanced feature data is the same as the number of the squeeze-and-excitation attention networks. Each squeeze-and-excitation attention network outputs one enhanced feature data.

[0065] In this embodiment, the core idea of the squeeze-and-excitation attention network is to adaptively weight and adjust the features of different channels by learning the importance of each channel. Traditional convolution operations process the features of all channels equally, while the squeeze-and-excitation attention network can assign different weights to each channel according to the global information of the input features, thereby enhancing useful features and suppressing unimportant features.

[0066] In this embodiment, the squeeze-and-excitation attention network includes three stages: Squeeze stage: Global average pooling is performed on the feature map output by the convolutional layer to compress the information in the spatial dimensions (height and width) into a channel descriptor, thereby obtaining the global information of each channel and highlighting the global importance of each channel. Excitation stage: The channel descriptor obtained from the squeeze operation is input into a fully connected layer to learn the dependencies between channels, and then through the Sigmoid activation function, the importance weights of each channel are output. These weights represent the importance of different channels for the current task. Scale stage: Multiply the weights obtained from the excitation operation by the original feature map to achieve dynamic adjustment of the feature map, enabling the network to adaptively enhance the features of important channels and suppress the features of unimportant channels.

[0067] In this embodiment, the manner of inputting the feature data of multiple stages into the corresponding squeeze-and-excitation attention networks respectively to obtain multiple enhanced feature data can be: input the feature data of the first stage into the first squeeze-and-excitation attention network to obtain the first enhanced feature data; input the feature data of the second stage into the second squeeze-and-excitation attention network to obtain the second enhanced feature data; input the feature data of the third stage into the third squeeze-and-excitation attention network to obtain the third enhanced feature data; input the feature data of the fourth stage into the fourth squeeze-and-excitation attention network to obtain the fourth enhanced feature data.

[0068] Optionally, the shifted window backbone network includes: a tile partitioning layer and multiple stage layers, and the neck network includes: multiple feature fusion layers, multiple convolutional layers, and multiple upsampling layers.

[0069] In this embodiment, the number of stage layers can be 4, or the number of stage layers can be set as needed. The embodiments of the present invention do not limit this.

[0070] In this embodiment, the number of feature fusion layers can be 6, or it can be set as needed. The embodiments of the present invention do not limit this.

[0071] In this embodiment, the number of convolutional layers and upsampling layers can be 3, or it can be set as needed. The embodiments of the present invention do not limit this.

[0072] Optionally, the multiple squeeze-and-excitation attention networks are respectively: the first squeeze-and-excitation attention network, the second squeeze-and-excitation attention network, the third squeeze-and-excitation attention network, and the fourth squeeze-and-excitation attention network;

[0073] Inputting the multiple stage feature data into the corresponding squeeze-and-excitation attention networks respectively to obtain multiple enhanced feature data includes:

[0074] Input the first-stage feature data into the first squeeze-and-excitation attention network to obtain first enhanced feature data.

[0075] In this embodiment, the size of the first enhanced feature data can be: 160*160*128.

[0076] Input the second-stage feature data into the second squeeze-and-excitation attention network to obtain second enhanced feature data.

[0077] In this embodiment, the size of the second enhanced feature data can be: 80*80*256.

[0078] Input the third-stage feature data into the third squeeze-and-excitation attention network to obtain third enhanced feature data.

[0079] In this embodiment, the size of the third enhanced feature data can be: 40*840*512.

[0080] Input the fourth-stage feature data into the fourth squeeze-and-excitation attention network to obtain fourth enhanced feature data.

[0081] In this embodiment, the size of the fourth enhanced feature data can be: 20*20*1024.

[0082] Input the multiple enhanced feature data into the neck network to obtain multiple fused feature data.

[0083] In this embodiment, the multiple feature fusion layers are respectively: the first feature fusion layer, the second feature fusion layer, the third feature fusion layer, the fourth feature fusion layer, the fifth feature fusion layer, and the sixth feature fusion layer; the multiple convolutional layers are respectively: the first convolutional layer, the second convolutional layer, and the third convolutional layer; the multiple upsampling layers are respectively: the first upsampling layer, the second upsampling layer, and the third upsampling layer.

[0084] In this embodiment, the method of inputting the multiple enhanced feature data into the neck network to obtain multiple fused feature data may be as follows: input the fourth enhanced feature data into the first upsampling layer to obtain the upsampled fourth enhanced feature data; input the upsampled fourth enhanced feature data and the third enhanced feature data into the first feature fusion layer to obtain the first fused feature data; input the first fused feature data into the second upsampling layer to obtain the upsampled first fused feature data; input the upsampled first fused feature data and the second enhanced feature data into the second feature fusion layer to obtain the second fused feature data; input the second fused feature data into the third upsampling layer to obtain the upsampled second fused feature data; input the upsampled second fused feature data and the first enhanced feature data into the third feature fusion layer to obtain the third fused feature data; input the third fused feature data into the first convolutional layer to obtain the convolved third fused feature data; input the convolved third fused feature data and the second fused feature data into the fourth feature fusion layer to obtain the fourth fused feature data; input the fourth fused feature data into the second convolutional layer to obtain the convolved fourth fused feature data; input the convolved fourth fused feature data and the first fused feature data into the fifth feature fusion layer to obtain the fifth fused feature data; input the fifth fused feature data into the third convolutional layer to obtain the convolved fifth fused feature data; input the convolved fifth fused feature data and the fourth enhanced feature data into the sixth feature fusion layer to obtain the sixth fused feature data.

[0085] Optionally, the multiple feature fusion layers are respectively: the first feature fusion layer, the second feature fusion layer, the third feature fusion layer, the fourth feature fusion layer, the fifth feature fusion layer, and the sixth feature fusion layer; the multiple convolutional layers are respectively: the first convolutional layer, the second convolutional layer, and the third convolutional layer; the multiple upsampling layers are respectively: the first upsampling layer, the second upsampling layer, and the third upsampling layer.

[0086] In this embodiment, the feature fusion layer includes: a connection layer and a cross-stage local layer. The connection layer may be a Concat layer, and the cross-stage local layer may be a CSPLayer layer. The upsampling layer may be Upsample. The convolutional layer may be ConvModule.

[0087] Inputting the multiple enhanced feature data into the neck network to obtain multiple fused feature data includes:

[0088] Input the fourth enhanced feature data into the first upsampling layer to obtain the upsampled fourth enhanced feature data.

[0089] In this embodiment, the size of the fourth enhanced feature data after upsampling can be: 40*40*1024.

[0090] Input the fourth enhanced feature data after upsampling and the third enhanced feature data into the first feature fusion layer to obtain first fusion feature data.

[0091] In this embodiment, the size of the first fusion feature data can be: 40*40*512.

[0092] Input the first fusion feature data into the second upsampling layer to obtain the first fusion feature data after upsampling.

[0093] In this embodiment, the size of the first fusion feature data after upsampling can be: 80*80*512.

[0094] Input the first fusion feature data after upsampling and the second enhanced feature data into the second feature fusion layer to obtain second fusion feature data.

[0095] In this embodiment, the size of the second fusion feature data can be: 80*80*256.

[0096] Input the second fusion feature data into the third upsampling layer to obtain the second fusion feature data after upsampling.

[0097] In this embodiment, the size of the second fusion feature data after upsampling can be: 160*160*256.

[0098] Input the second fusion feature data after upsampling and the first enhanced feature data into the third feature fusion layer to obtain third fusion feature data.

[0099] In this embodiment, the size of the third fusion feature data can be: 160*160*128.

[0100] Input the third fusion feature data into the first convolutional layer to obtain the third fusion feature data after convolution.

[0101] In this embodiment, the size of the third fusion feature data after convolution can be: 80*80*128.

[0102] Input the third fusion feature data after convolution and the second fusion feature data into the fourth feature fusion layer to obtain fourth fusion feature data.

[0103] In this embodiment, the size of the fourth fusion feature data can be: 80*80*256.

[0104] Input the fourth fusion feature data into the second convolutional layer to obtain the fourth fusion feature data after convolution.

[0105] In this embodiment, the size of the fourth fusion feature data after convolution can be: 40*40*256.

[0106] Input the fourth fusion feature data after convolution and the first fusion feature data into the fifth feature fusion layer to obtain the fifth fusion feature data.

[0107] In this embodiment, the size of the fifth fusion feature data can be: 40*40*512.

[0108] Input the fifth fusion feature data into the third convolutional layer to obtain the fifth fusion feature data after convolution.

[0109] In this embodiment, the size of the fifth fusion feature data after convolution can be: 20*20*512.

[0110] Input the fifth fusion feature data after convolution and the fourth enhanced feature data into the sixth feature fusion layer to obtain the sixth fusion feature data.

[0111] In this embodiment, the size of the sixth fusion feature data can be: 20*20*1024.

[0112] Input the multiple fusion feature data into the head network to obtain spacer bar detection frames of multiple sizes.

[0113] In this embodiment, the head network includes: a first convolutional module and a second convolutional module. The first convolutional module includes: a convolutional layer. The second convolutional module includes: a convolutional layer, a normalization layer, and an activation function layer.

[0114] In this embodiment, the spacer bar detection frames of multiple sizes are respectively: a spacer bar detection frame of the first size, a spacer bar detection frame of the second size, a spacer bar detection frame of the third size, and a spacer bar detection frame of the fourth size.

[0115] Optionally, inputting the multiple fusion feature data into the head network to obtain spacer bar detection frames of multiple sizes includes:

[0116] Input the third fusion feature data, the fourth fusion feature data, the fifth fusion feature data, and the sixth fusion feature data into the head network to obtain a spacer bar detection frame of the first size, a spacer bar detection frame of the second size, a spacer bar detection frame of the third size, and a spacer bar detection frame of the fourth size in the transmission line image.

[0117] In this embodiment, the first dimension is smaller than the second dimension, the second dimension is smaller than the third dimension, and the third dimension is smaller than the fourth dimension. For example, the first dimension can be (4*4), the second dimension can be (8*8), the third dimension can be (16*16), and the fourth dimension can be (32*32).

[0118] Optionally, it further includes:

[0119] Build a model to be trained.

[0120] In this embodiment, the model to be trained includes: a shifted window backbone network, multiple squeeze-and-excitation attention networks, a neck network, and a head network. The shifted window backbone network includes: a tile partitioning layer and multiple stage layers, and the multiple stage layers are respectively: a first stage layer, a second stage layer, a third stage layer, and a fourth stage layer. The multiple squeeze-and-excitation attention networks are respectively: a first squeeze-and-excitation attention network, a second squeeze-and-excitation attention network, a third squeeze-and-excitation attention network, and a fourth squeeze-and-excitation attention network. The neck network includes: multiple feature fusion layers, multiple convolutional layers, and multiple upsampling layers. The multiple feature fusion layers are respectively: a first feature fusion layer, a second feature fusion layer, a third feature fusion layer, a fourth feature fusion layer, a fifth feature fusion layer, and a sixth feature fusion layer. The multiple convolutional layers are respectively: a first convolutional layer, a second convolutional layer, and a third convolutional layer. The multiple upsampling layers are respectively: a first upsampling layer, a second upsampling layer, and a third upsampling layer. The head network includes: a first convolutional module and a second convolutional module. The first convolutional module includes: a convolutional layer. The second convolutional module includes: a convolutional layer, a normalization layer, and an activation function layer.

[0121] Obtain a target sample set.

[0122] In this embodiment, the target sample set includes: image samples, and the image samples carry annotation data, and the annotation data includes: detection frames of spacer bars of multiple dimensions in the image samples.

[0123] In this embodiment, the detection frames of spacer bars of the multiple dimensions are respectively: a detection frame of a spacer bar of the first dimension, a detection frame of a spacer bar of the second dimension, a detection frame of a spacer bar of the third dimension, and a detection frame of a spacer bar of the fourth dimension. The first dimension is smaller than the second dimension, the second dimension is smaller than the third dimension, and the third dimension is smaller than the fourth dimension. For example, the first dimension can be (4*4), the second dimension can be (8*8), the third dimension can be (16*16), and the fourth dimension can be (32*32).

[0124] Input the image samples in the target sample set into the model to be trained to obtain prediction detection frames of multiple dimensions and class information of the target objects in the prediction detection frames.

[0125] In this embodiment, the target sample set may be pre-provided transmission line images. It should be noted that after model training on the server side, the trained model can be deployed in the transmission tower monitoring camera to improve the efficiency of transmission line galloping detection.

[0126] In this embodiment, if the processing capacity of the transmission tower monitoring camera is sufficient to support model training, model training can also be performed on the side of the transmission tower monitoring camera, and the embodiments of the present invention do not limit this.

[0127] In this embodiment, the method of inputting the image samples in the target sample set into the model to be trained to obtain prediction detection frames of multiple sizes may be: inputting the image samples in the target sample set into the shifted window backbone network to obtain multiple stage feature data; respectively inputting the multiple stage feature data into the corresponding squeeze-and-excitation attention network to obtain multiple enhanced feature data; inputting the multiple enhanced feature data into the neck network to obtain multiple fused feature data; and inputting the multiple fused feature data into the head network to obtain prediction detection frames of multiple sizes.

[0128] In this embodiment, the category information of the target object in the prediction detection frame may be spacer dampers.

[0129] In this embodiment, the prediction detection frames of multiple sizes are respectively: the prediction detection frame of the first size, the prediction detection frame of the second size, the prediction detection frame of the third size, and the prediction detection frame of the fourth size. The first size is smaller than the second size, the second size is smaller than the third size, and the third size is smaller than the fourth size. For example, the first size may be (4*4), the second size may be (8*8), the third size may be (16*16), and the fourth size may be (32*32).

[0130] Train the parameters of the model to be trained according to the target function formed by the spacer damper detection frames of multiple sizes, the prediction detection frames of multiple sizes, and the category information of the target object in the prediction detection frame in the image sample to obtain the target detection model, and the target function includes: a detection frame loss function and a classification loss function;

[0131] The detection frame loss function is:

[0132] where N is the number of detection branches, S2 is the total number of grids corresponding to each detection branch, B is the number of prediction detection frames generated by each grid at the resolution, is the exponential function. If the i-th detection branch, the j-th grid, and the k-th prediction detection frame contain the target object, then If the i-th detection branch, the j-th grid, and the k-th prediction detection frame do not contain the target object, then $h'w'$ is the area of the predicted detection box, $hw$ is the total area of the image, and $L$ WIoU is the weighted intersection over union loss function, and $L$ DFL is the distribution focal loss function;

[0133] The classification loss function is:

[0134] where is the true probability that the class information of the target object in the detection box is the spacer bar, is the predicted probability that the class information of the target object in the detection box is the spacer bar.

[0135] In this embodiment, the weighted intersection over union loss (WIoU) function assigns different weights to different detection boxes, so as to more carefully measure the difference between the predicted spacer bar detection box and the ground truth box.

[0136] In this embodiment, the distribution focal loss (DFL) function focuses on learning the distribution information of the spacer bar position, rather than just predicting a single value. It enables the model to better learn the uncertainty of the spacer bar position, thereby improving the accuracy of spacer bar detection.

[0137] In a specific example, Figure 2 is the structural schematic diagram of the model to be trained. As Figure 2As shown, the model to be trained includes: a shifted window backbone network, multiple squeeze-and-excitation attention networks, a neck network, and a head network. The shifted window backbone network includes: a Patch Partition layer, a first stage layer (a Liner Emdedding layer and 2 * Swin Transformer block layers), a second stage layer (a Patch Merging layer and 2 * Swin Transformer block layers), a third stage layer (a Patch Merging layer and 6 * Swin Transformer block layers), and a fourth stage layer (a Patch Merging layer and 2 * Swin Transformer block layers). The squeeze-and-excitation attention network is an SEAttention network. The neck network includes: an Upsample layer, a Concat layer, a CSPLayer_2Conv layer, an Upsample layer, a Concat layer, a CSPLayer_2Conv layer, an Upsample layer, a Concat layer, a CSPLayer_2Conv layer, a ConvModule layer, a Concat layer, a CSPLayer_2Conv layer, a ConvModule layer, a Concat layer, a CSPLayer_2Conv layer, a ConvModule layer, a Concat layer, and a CSPLayer_2Conv layer. The head network includes: multiple first convolution modules and multiple second convolution modules. The first convolution module includes: a convolution layer. The second convolution module includes: a convolution layer, a normalization layer, and an activation function layer. The first convolution module is Conv2D, and the second convolution module is ConvModule. Figure 2 The head network in [reference] includes: 8 * ConvModule and 8 * Conv2D.

[0138] The training process of the model to be trained is as follows: Input the power transmission line image sample of 640*640*3 into the PatchPartition layer to obtain a sub-image of 160*160*128, and input the sub-image into the first-stage layer to obtain the first-stage feature data of 160*160*128; Input the first-stage feature data into the second-stage layer to obtain the second-stage feature data of 80*80*256; Input the second-stage feature data into the third-stage layer to obtain the third-stage feature data of 40*40*512; Input the third-stage feature data into the fourth-stage layer to obtain the fourth-stage feature data of 20*20*1024; Input the first-stage feature data into the first squeeze-and-excitation attention module to obtain the first enhanced feature data of 160*160*128; Input the second-stage feature data into the second squeeze-and-excitation attention module to obtain the second enhanced feature data of 80*80*256; Input the third-stage feature data into the third squeeze-and-excitation attention module to obtain the third enhanced feature data of 40*40*512; Input the fourth-stage feature data into the fourth squeeze-and-excitation attention module to obtain the fourth enhanced feature data of 40*40*512; Input the fourth enhanced feature data into the first upsampling layer to obtain the upsampled fourth enhanced feature data of 40*40*1024; Input the upsampled fourth enhanced feature data and the third enhanced feature data into the first feature fusion layer to obtain the first fusion feature data of 40*40*512; Input the first fusion feature data into the second upsampling layer to obtain the upsampled first fusion feature data of 80*80*512; Input the upsampled first fusion feature data and the second enhanced feature data into the second feature fusion layer to obtain the second fusion feature data of 80*80*256; Input the second fusion feature data into the third upsampling layer to obtain the upsampled second fusion feature data of 160*160*256; Input the upsampled second fusion feature data and the first enhanced feature data into the third feature fusion layer to obtain the third fusion feature data of 160*160*128 (the first output of the neck network); Input the third fusion feature data into the first convolutional layer to obtain the convolved third fusion feature data of 80*80*128; Input the convolved third fusion feature data and the second fusion feature data into the fourth feature fusion layer to obtain the fourth fusion feature data of 80*80*256 (the second output of the neck network); Input the fourth fusion feature data into the second convolutional layer to obtain the convolved fourth fusion feature data of 40*40*256; Input the convolved fourth fusion feature data and the first fusion feature data into the fifth feature fusion layer to obtain the fifth fusion feature data of 40*40*512 (the third output of the neck network);Input the fifth fusion feature data into the third convolutional layer to obtain the fifth fusion feature data after convolution with a size of 20*20*512; input the fifth fusion feature data after convolution and the fourth enhanced feature data into the sixth feature fusion layer to obtain the sixth fusion feature data with a size of 20*20*1024 (the fourth output of the neck network); input the third fusion feature data into the ConvModule layer in the head network to obtain feature data with a size of 160*160*128, input the feature data of 160*160*128 into the Conv2D layer to obtain the predicted detection box (4*4) with the first size in the transmission line image and the predicted class information, generate a classification loss function according to the predicted class information and the class label carried by the image sample, and generate a detection box loss function according to the predicted detection box with the first size and the detection box with the first size carried by the image sample. Input the fourth fusion feature data into the ConvModule layer in the head network to obtain feature data with a size of 80*80*256, input the feature data of 80*80*256 into the Conv2D layer to obtain the predicted rod detection box with the second size in the transmission line image and the predicted class information, where the first size is smaller than the second size (8*8); input the fifth fusion feature data into the ConvModule layer in the head network to obtain feature data with a size of 40*40*512, input the feature data of 40*40*512 into the Conv2D layer to obtain the predicted rod detection box with the third size in the transmission line image and the predicted class information, where the second size is smaller than the third size (32*32); input the sixth fusion feature data into the ConvModule layer in the head network to obtain feature data with a size of 20*20*1024, input the feature data of 20*20*1024 into the Conv2D layer to obtain the predicted rod detection box with the fourth size in the transmission line image and the predicted class information, where the third size is smaller than the fourth size (32*32).;

[0139] S130. Based on the spacer bar detection boxes with multiple sizes in each frame of the transmission line image, determine the spacer bar movement trajectory.

[0140] In this embodiment, three-dimensional information reduction of the spacer bar is performed to calculate the movement trajectory of the spacer bar in three-dimensional space.

[0141] S140. Determine the transmission line galloping trajectory according to the spacer bar movement trajectory.

[0142] In this embodiment, by recording the spacer bar movement trajectory, the transmission line galloping trajectory is indirectly restored.

[0143] In a specific example, such as Figure 3As shown in the figure, obtain the dataset of spacer dampers on transmission lines captured by cameras, drones, etc. in the early stage from power grid-related enterprises. Use standard tools to label the spacer dampers in the dataset images to generate a standard file in XML format. To enhance the training effect, perform augmentation operations on the dataset, including random cropping and scaling, random grayscaling, random color transformation, left-right flipping, etc. To meet the input standards of the network, it is necessary to preprocess the images, including equal-proportion scaling and padding of the images. Aiming at the problem that the existing recognition methods for spacer dampers on transmission lines have insufficient recognition accuracy for spacer dampers far from the camera, the model to be trained in this embodiment can be an improved YOLOv8 model. By transforming the YOLOv8 deep learning network, the detection accuracy of small targets and blurred targets is improved. The specific measures include the following aspects:

[0144] (1) Backbone network: Replace the shifted window backbone network (for example, it can be Swin Transformer) with the backbone network of YOLOv8. With its self-attention mechanism and hierarchical design, Swin Transformer can better capture the detailed features of spacer dampers in the image; spacer dampers may present different sizes and dimensions in the image. Through the window splitting and shifting strategy, Swin Transformer realizes the transfer and fusion of feature maps at different scales, avoiding the information loss problem in traditional convolutional neural networks and improving the detection accuracy of small targets and occluded targets. As Figure 4 shown Figure 4 is the structural schematic diagram of the shifted window backbone network. The shifted window backbone network includes: Patch Partition layer, the first stage layer, the second stage layer, the third stage layer, and the fourth stage layer. The first stage layer includes: Liner Emdedding layer and 2 SwinTransformer block layers. The second stage layer includes: Patch Merging layer and 2 Swin Transformer block layers. The third stage layer includes: Patch Merging layer and 6 Swin Transformer block layers. The fourth stage layer includes: Patch Merging layer and 2 Swin Transformer block layers.

[0145] (2) Neck network. To improve the model's detection ability for small-sized spacer bars at a distance, multiple squeeze-and-excitation attention networks (such as SEAttention) are introduced at the connection between the shifted window backbone network and the neck network. In the Squeeze stage, global average pooling is used to obtain the global information of each channel, enabling the network to calculate the importance of each channel to the overall feature; in the Excitation stage, the weights of each channel are learned through fully connected layers and activation functions to achieve adaptive feature selection, which can strengthen useful features and suppress irrelevant features, enabling the network to capture the features of the spacer bar more accurately.

[0146] Figure 5 The structural schematic diagram of the squeeze-and-excitation attention network is as Figure 5 shown. The input feature X (X is the feature map output by the shifted window backbone network) is a three-dimensional tensor with a size of H′×W′×C′. The input feature X passes through the feature transformation F tr , and a new feature tensor U is output. F sq (.) is used to represent the operation in the Squeeze stage. The feature U is input into F sq (.), and a one-dimensional vector with a size of 1×1×C is output. In the Squeeze stage, global average pooling (such as Global AveragePooling) is usually used to reduce the feature values of each channel to a global vector, thereby capturing the global information of each channel. F ex (.,W) is used to represent the operation in the Excitation stage. The compressed feature is input into F ex (.,W), and an extended feature vector with a size still of 1×1×C is output. The Excitation stage usually consists of two fully connected layers, first reducing the dimension and then increasing the dimension, and finally generating a weight vector through the sigmoid function to ensure that their sum is 1. This process is used to learn the weight distribution between channels. F scale (.,.) is used to represent the operation in the Scaling stage. The extended feature is multiplied element-wise with the original feature U, and the enhanced feature X~ is output. This operation is used to adjust the feature values of each channel of the original feature, emphasizing the information of important channels and suppressing the information of unimportant channels. The finally output feature tensor X~ has a size of H×W×C, which is the same as the size of the input feature U. This feature tensor is usually passed to the subsequent network layers for further processing.

[0147] (3) Head network: To further enhance the detection ability for spacer bars, compared with the YOLOv8 model in the embodiments of the present invention, a detection branch dedicated to small target detection is added to further improve the detection accuracy of small targets. The feature map output by the squeeze-and-excitation attention network is input into the neck network.

[0148] (4) Loss function: The loss function consists of a classification loss function and a detection box loss function. To address the problem of unbalanced positive and negative samples in spacer images and improve the generalization ability and object detection accuracy of the model.

[0149] The detection box loss function is as follows:

[0150] where N is the number of detection branches, S2 is the total number of grids corresponding to each detection branch, B is the number of predicted detection boxes generated by each grid at the resolution, is the exponential function. If the i-th detection branch, the j-th grid, and the k-th predicted detection box contain the target object, then If the i-th detection branch, the j-th grid, and the k-th predicted detection box do not contain the target object, then h'w' is the area of the predicted detection box, hw is the total area of the image, and L WIoU is the weighted intersection over union loss function, and L DFL is the distribution focusing loss function;

[0151] The classification loss function is as follows:

[0152] where is the true probability that there is a spacer in the detection box, is the predicted probability that there is a spacer in the detection box.

[0153] Training the model to be trained on the server side: Divide the dataset of spacer dampers of transmission lines obtained in the early stage into a training set and a test set. Use the training set to train the model to be trained. Adopt the Warmup preheating mechanism, cosine annealing learning rate scheduling, SGD+Momentum optimizer, and exponential moving average weight smoothing technology to achieve the rapid convergence of the model and obtain the optimal network weights. At the same time, load the optimal network weights into the model to be trained, and use the test set to test the detection effect. If the test passes, load the optimal network weights into the model to be trained, and the obtained model is used as the object detection model. In this embodiment, the Warmup preheating mechanism is a learning rate adjustment strategy, which is usually used to optimize the model training process and avoid instability of the model in the initial stage of training. At the beginning of training, the parameters of the model are randomly initialized, and the distribution of the parameters may be quite different from the optimal solution at this time. If a large learning rate is used, the parameter update step of the model will be very large, which may cause the model to update the parameters too aggressively in the initial stage of training, making it difficult for the model to converge in the initial stage of training, and even may cause the problem of gradient explosion. The approach of the Warmup mechanism is to gradually increase the learning rate from a small value to the preset initial learning rate in the first few rounds or the first several iteration steps of training. As training progresses, the parameters of the model gradually approach the optimal solution. At this time, use the preset initial learning rate for training, which can enable the model to learn the features of the data more stably in the initial stage of training and avoid the negative impact of too large a learning rate on the model. Cosine annealing learning rate scheduling (Cosine LearningRate, CosineLR) adjusts the learning rate based on the change of the cosine function. It simulates the change of the learning rate as an annealing process. In the initial stage of training, the learning rate is large, and the model can quickly explore the parameter space. As training progresses, the learning rate gradually decreases according to the law of the cosine function, enabling the model to adjust the parameters more finely to approach the global optimal solution. The SGD+Momentum optimizer is developed on the basis of the Stochastic Gradient Descent (SGD) optimizer. Stochastic gradient descent calculates the gradient of the loss function with respect to the model parameters, and then updates the parameters along the opposite direction of the gradient to minimize the loss function. SGD has the problems of slow convergence speed and easy oscillation in the training process. To solve these problems, the concept of momentum is introduced. The core idea of momentum is to introduce "inertia" into the parameter update, that is, to refer to the previous gradient information to smooth the current parameter update process, avoid violent fluctuations when updating the parameters, and thus accelerate convergence. The exponential moving average weight smoothing technology (ExponentialMoving Average, EMA), also known as exponential weighted moving average, is a time series processing and optimization technology. EMA gives higher weights to recent data, and as the time of data points passes, the weights decrease exponentially.It can quickly respond to data changes while retaining a certain amount of historical data, thus achieving the purpose of smoothing data and reducing noise interference in the data.

[0154] The transmission tower monitoring camera collects on-site image data of the transmission line.

[0155] Perform pre-processing on the obtained transmission line pictures to make them adapt to the requirements of the network structure.

[0156] Deploy the object detection model to the transmission tower monitoring camera and use the object detection model to detect spacer damper targets.

[0157] Use the Distance Intersection over Union (DIoU) Non-Maximum Suppression (NMS) algorithm to remove redundant detection boxes. In traditional NMS, the detection box with the highest score is calculated with other detection boxes for the corresponding IOU value. When the IOU value exceeds the set threshold (commonly set to 0.5, often set to 0.7 in object detection), the boxes exceeding the threshold are suppressed. The IoU metric only considers the overlapping area and often causes incorrect suppression. Therefore, using DIoU instead of ordinary IoU as the standard for NMS, which not only considers the overlapping area but also the distance between the center points, can greatly improve the detection accuracy of overlapping targets.

[0158] Restore the three-dimensional information of the spacer damper detection result, extract the feature points of the spacer damper, and match them with the existing spacer damper model, thereby restoring the pose of the spacer damper relative to the camera.

[0159] Perform the above operations frame by frame on the galloping video to obtain a time series of the pose of the spacer damper relative to the camera. Perform operations such as filtering and fitting on this sequence to restore the motion trajectory of the spacer damper, indirectly restore the galloping information of the transmission line, and output the information to the cloud or other devices.

[0160] By using the object detection model to detect spacer damper targets on the transmission line, compared with using the YOLOv5 algorithm to detect spacer damper targets on the transmission line, the detection accuracy is improved by about 8.7%, and the detection speed is not significantly reduced.

[0161] For the technical solution provided by the embodiments of the present invention, in view of the challenges faced by existing object detection algorithms in spacer damper detection, such as scale, perspective change, complex background, lighting conditions, occlusion, and sample imbalance, which lead to a low recognition rate, the embodiments of the present invention propose a spacer damper target detection algorithm based on improved YOLOv8 for transmission line galloping monitoring. Without reducing the detection speed, the detection accuracy of the spacer damper is improved, thereby improving the fault detection efficiency.

[0162] The object detection model includes a shifted window backbone network, multiple squeeze-and-excitation attention networks, a neck network, and a head network. The shifted window backbone network utilizes its self-attention mechanism and hierarchical design, especially the role of its window splitting and shifting strategies in multi-scale feature transfer and fusion, to enhance the ability to capture the detailed features of spacer bars.

[0163] The shifted window backbone network realizes the transfer and fusion of feature maps at different scales, avoids the information loss problem in traditional convolutional neural networks, and improves the detection accuracy of small and occluded objects.

[0164] Multiple squeeze-and-excitation attention networks are introduced at the connection between the shifted window backbone network and the neck network. Adaptive feature selection is achieved through global average pooling and fully connected layers, strengthening useful features and suppressing irrelevant features, and enhancing the detection ability for small-sized spacer bars in the distance.

[0165] A detection branch dedicated to small object detection is added to the head network to further improve the detection accuracy of small objects.

[0166] Using the loss function provided in this embodiment can solve the problem of imbalance between positive and negative samples in spacer bar images, improve the generalization ability of the model and the object detection accuracy. A detection box weight factor is introduced into the loss function to further optimize the regression accuracy of the detection box.

[0167] The same convergence optimization algorithm as in this sample embodiment is adopted. The training process of the model is optimized to make it converge faster and achieve higher detection accuracy.

[0168] The head network in this embodiment adopts a decoupled head design, separating the classification and regression tasks for independent training and optimization.

[0169] The DIoU_NMS algorithm is adopted to replace the traditional NMS algorithm, which not only considers the overlapping area but also the distance between the center points, improving the detection accuracy of overlapping spacer bars.

[0170] For the spacer bar detection task, a specific data augmentation strategy is designed, including random cropping and scaling, random grayscaling, random color transformation, left-right flipping, etc.

[0171] Image pre - processing standardization. The input image is scaled and padded proportionally to adapt to the network structure of the object detection model. The technical solution of this embodiment is to obtain multiple consecutive frames of transmission line images, and at least one spacer damper is arranged on the transmission line; input the multiple frames of transmission line images into the object detection model in sequence to obtain spacer damper detection frames of multiple sizes in each frame of transmission line image. Among them, the object detection model sequentially includes, from input to output: a shifted window backbone network, multiple squeeze - and - excitation attention networks, a neck network, and a head network; based on the spacer damper detection frames of multiple sizes in each frame of transmission line image, determine the spacer damper movement trajectory; determine the transmission line galloping trajectory according to the spacer damper movement trajectory, which solves the problem that the galloping of the transmission line seriously threatens the safe and stable operation of the power system. It can accurately identify spacer damper detection frames of multiple sizes through the object detection model, and then determine the spacer damper movement trajectory according to the spacer damper detection frames of multiple sizes, and then determine the transmission line galloping trajectory according to the spacer damper movement trajectory to realize the monitoring of the transmission line galloping, thereby improving the efficiency and accuracy of the transmission line galloping monitoring, and effectively avoiding the occurrence of power grid safety accidents.

[0172] Embodiment 2

[0173] Figure 6 This is a schematic structural diagram of a transmission line galloping monitoring device provided by an embodiment of the present invention. This embodiment is applicable to the situation of transmission line galloping monitoring. The device can be implemented in software and / or hardware, and the device can be integrated in any device that provides the function of transmission line galloping monitoring, such as Figure 6 As shown, the transmission line galloping monitoring device specifically includes: an image acquisition module 610, a spacer damper detection frame recognition module 620, a spacer damper movement trajectory determination module 630, and a transmission line galloping trajectory determination module 640.

[0174] Among them, the image acquisition module is used to obtain multiple consecutive frames of transmission line images, and at least one spacer damper is arranged on the transmission line;

[0175] The spacer damper detection frame recognition module is used to input the multiple frames of transmission line images into the object detection model in sequence to obtain spacer damper detection frames of multiple sizes in each frame of transmission line image. Among them, the object detection model sequentially includes, from input to output: a shifted window backbone network, multiple squeeze - and - excitation attention networks, a neck network, and a head network;

[0176] The spacer damper movement trajectory determination module is used to determine the spacer damper movement trajectory based on the spacer damper detection frames of multiple sizes in each frame of transmission line image;

[0177] The transmission line galloping trajectory determination module is used to determine the transmission line galloping trajectory according to the spacer movement trajectory.

[0178] The above product can execute the method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0179] Embodiment III

[0180] Figure 7 The structural schematic diagram of the electronic device 10 that can be used to implement the embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0181] As Figure 7 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by at least one processor, and the processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0182] A plurality of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0183] The processor 11 may be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the transmission line galloping monitoring method.

[0184] In some embodiments, the transmission line galloping monitoring method may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the transmission line galloping monitoring method described above may be executed. Alternatively, in other embodiments, the processor 11 may be configured to execute the transmission line galloping monitoring method by any other suitable means (e.g., by means of firmware).

[0185] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), system-on-chip systems (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a dedicated or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0186] The computer programs for implementing the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processors of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the computer programs are executed by the processors, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0187] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0188] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0189] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0190] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is created by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0191] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0192] An embodiment of the present invention also provides a computer program product, including a computer program which, when executed by a processor, implements the transmission line galloping monitoring method according to any embodiment of the present invention.

[0193] In the process of implementing the computer program product, computer program code for performing the operations of the present invention can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer via any type of network - including a local area network (LAN) or a wide area network (WAN) - or can be connected to an external computer (for example, by using an Internet service provider to connect via the Internet).

[0194] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for monitoring the galloping of a transmission line, characterized in that, Including: Obtaining a plurality of consecutive frames of transmission line images, where at least one spacer damper is provided on the transmission line; Sequentially inputting the plurality of frames of transmission line images into a target detection model to obtain spacer damper detection frames of multiple sizes in each frame of transmission line image, where the target detection model sequentially includes, from input to output: a shifted window backbone network, a plurality of squeeze-and-excitation attention networks, a neck network, and a head network; Determining the movement trajectory of the spacer damper based on the spacer damper detection frames of multiple sizes in each frame of transmission line image; Determining the galloping trajectory of the transmission line according to the movement trajectory of the spacer damper.

2. The method according to claim 1, wherein Sequentially inputting the plurality of frames of transmission line images into the target detection model to obtain spacer damper detection frames of multiple sizes in each frame of transmission line image, including: Inputting the transmission line image into the shifted window backbone network to obtain a plurality of stage feature data; Respectively inputting the plurality of stage feature data into corresponding squeeze-and-excitation attention networks to obtain a plurality of enhanced feature data; Inputting the plurality of enhanced feature data into the neck network to obtain a plurality of fused feature data; Inputting the plurality of fused feature data into the head network to obtain spacer damper detection frames of multiple sizes.

3. The method according to claim 2, wherein The shifted window backbone network includes: a patch partition layer and a plurality of stage layers, and the neck network includes: a plurality of feature fusion layers, a plurality of convolutional layers, and a plurality of upsampling layers.

4. The method according to claim 3, wherein The plurality of stage layers are respectively: a first stage layer, a second stage layer, a third stage layer, and a fourth stage layer; Inputting the transmission line image into the shifted window backbone network to obtain a plurality of stage feature data, including: Inputting the transmission line image into the patch partition layer to obtain a plurality of sub-images; Inputting the plurality of sub-images into the first stage layer to obtain first stage feature data; Inputting the first stage feature data into the second stage layer to obtain second stage feature data; Inputting the second stage feature data into the third stage layer to obtain third stage feature data; Inputting the third stage feature data into the fourth stage layer to obtain fourth stage feature data.

5. The method according to claim 4, wherein The plurality of squeeze-and-excitation attention networks are respectively: a first squeeze-and-excitation attention network, a second squeeze-and-excitation attention network, a third squeeze-and-excitation attention network, and a fourth squeeze-and-excitation attention network; Respectively inputting the plurality of stage feature data into corresponding squeeze-and-excitation attention networks to obtain a plurality of enhanced feature data, including: Inputting the first stage feature data into the first squeeze-and-excitation attention network to obtain first enhanced feature data; Inputting the second stage feature data into the second squeeze-and-excitation attention network to obtain second enhanced feature data; Inputting the third stage feature data into the third squeeze-and-excitation attention network to obtain third enhanced feature data; Inputting the fourth stage feature data into the fourth squeeze-and-excitation attention network to obtain fourth enhanced feature data.

6. The method according to claim 5, wherein The multiple feature fusion layers are respectively: the first feature fusion layer, the second feature fusion layer, the third feature fusion layer, the fourth feature fusion layer, the fifth feature fusion layer, and the sixth feature fusion layer. The multiple convolutional layers are respectively: the first convolutional layer, the second convolutional layer, and the third convolutional layer. The multiple upsampling layers are respectively: the first upsampling layer, the second upsampling layer, and the third upsampling layer; Input the multiple enhanced feature data into the neck network to obtain multiple fused feature data, including: Input the fourth enhanced feature data into the first upsampling layer to obtain the upsampled fourth enhanced feature data; Input the upsampled fourth enhanced feature data and the third enhanced feature data into the first feature fusion layer to obtain the first fused feature data; Input the first fused feature data into the second upsampling layer to obtain the upsampled first fused feature data; Input the upsampled first fused feature data and the second enhanced feature data into the second feature fusion layer to obtain the second fused feature data; Input the second fused feature data into the third upsampling layer to obtain the upsampled second fused feature data; Input the upsampled second fused feature data and the first enhanced feature data into the third feature fusion layer to obtain the third fused feature data; Input the third fused feature data into the first convolutional layer to obtain the convolved third fused feature data; Input the convolved third fused feature data and the second fused feature data into the fourth feature fusion layer to obtain the fourth fused feature data; Input the fourth fused feature data into the second convolutional layer to obtain the convolved fourth fused feature data; Input the convolved fourth fused feature data and the first fused feature data into the fifth feature fusion layer to obtain the fifth fused feature data; Input the fifth fused feature data into the third convolutional layer to obtain the convolved fifth fused feature data; Input the convolved fifth fused feature data and the fourth enhanced feature data into the sixth feature fusion layer to obtain the sixth fused feature data.

7. The method according to claim 6, characterized in that, Input the multiple fused feature data into the head network to obtain spacer bar detection frames of multiple sizes, including: Input the third fused feature data, the fourth fused feature data, the fifth fused feature data, and the sixth fused feature data into the head network to obtain the spacer bar detection frames of the first size, the second size, the third size, and the fourth size in the transmission line image.

8. The method according to claim 1, wherein It also includes: Establish a model to be trained; Obtain a target sample set, where the target sample set includes: image samples, and the image samples carry annotation data, and the annotation data includes: spacer bar detection frames of multiple sizes in the image samples; Input the image samples in the target sample set into the model to be trained to obtain prediction detection frames of multiple sizes and the category information of the target objects in the prediction detection frames; Training the parameters of the model to be trained with an objective function formed based on spacer detection frames of multiple sizes in the image sample, prediction detection frames of the multiple sizes, and class information of the target object in the prediction detection frames to obtain an object detection model, where the objective function includes: a detection frame loss function and a classification loss function; The detection box loss function is as follows: where N is the number of detection branches, S2 is the total number of grids corresponding to each detection branch, and B is the number of predicted detection boxes generated by each grid at the resolution. is an exponential function. If the k-th predicted detection box in the j-th grid of the i-th detection branch contains the target object, then If the k-th predicted detection box in the j-th grid of the i-th detection branch does not contain the target object, then h'w' is the area of the predicted detection box, hw is the total area of the image, and L WIoU is the weighted intersection over union loss function, and L DFL is the distribution focal loss function. The classification loss function is as follows: Among them, is the true probability that the category information of the target object in the detection frame is a spacer bar, is the predicted probability that the category information of the target object in the detection frame is a spacer bar.

9. A transmission line galloping monitoring device, characterized in that, including: an image acquisition module configured to acquire multiple consecutive frames of transmission line images, where at least one spacer is provided on the transmission line; a spacer detection frame recognition module configured to sequentially input the multiple frames of transmission line images into the object detection model to obtain spacer detection frames of multiple sizes in each frame of transmission line image, where the object detection model sequentially includes, from input to output: a shifted window backbone network, multiple squeeze-and-excitation attention networks, a neck network, and a head network; a spacer movement trajectory determination module configured to determine a spacer movement trajectory based on the spacer detection frames of multiple sizes in each frame of transmission line image; a transmission line galloping trajectory determination module configured to determine a transmission line galloping trajectory according to the spacer movement trajectory.

10. An electronic device, characterized in that, The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; where the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the transmission line galloping monitoring method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a processor to implement the transmission line galloping monitoring method according to any one of claims 1-8 when executed.

12. A computer program product, characterized in that, The computer program product includes a computer program that implements the transmission line galloping monitoring method according to any one of claims 1-8 when executed by a processor.

Images