A transmission line hidden danger identification method based on wire width correction and differential shooting

Abstract

The application discloses a power transmission line hidden danger identification method based on wire width correction and differential shooting, comprising the following steps: constructing a power transmission line hidden danger identification model based on an improved Transform network structure, the improvement including introducing a width modulation mechanism for embedding power transmission line width information into model feature expression; training the power transmission line hidden danger identification model; in a wind disaster scene, a UAV performs differential shooting on a power transmission line to be detected according to a preset inspection route, obtains a set of images to be detected, and performs differential processing to eliminate background interference and highlight the power transmission line region, thereby obtaining a set of input images; based on an optical imaging relationship and UAV spatial position information, correcting the line imaging width of each input image to obtain a power transmission line width correction value corresponding to each input image; inputting the input image and the power transmission line width correction value into the trained power transmission line hidden danger identification model to output a hidden danger identification result.

Description

Technical Field

[0001] This invention belongs to the field of power transmission line detection technology, specifically relating to a method for identifying potential hazards in power transmission lines based on conductor width correction and differential imaging. Background Technology

[0002] As a crucial component of the power system, the operational status of transmission lines directly impacts the safety and stability of the power grid. Due to long-term operation, environmental corrosion, and extreme weather conditions, transmission lines are prone to hazards such as broken conductor strands and foreign object entanglement. Failure to detect and address these hazards promptly can lead to line tripping or even widespread power outages. Therefore, efficient and accurate hazard identification of transmission lines is of significant engineering importance.

[0003] With the development of drone technology and computer vision technology, drone-based methods for identifying potential hazards in power transmission lines have gradually become the mainstream solution. In the prior art, Chinese patent application CN118608990A discloses a method, apparatus, and equipment for foreign object identification in power transmission lines. The method includes acquiring an initial identification model and images of the power transmission line and foreign objects collected by a drone, constructing a line dataset and a foreign object dataset; adjusting the structure of the initial identification model using a full-dimensional dynamic convolutional layer to generate an intermediate identification model; training the intermediate identification model using the line dataset and the foreign object dataset respectively to generate a target line identification model and a target foreign object identification model; when receiving an image to be detected, calling the target line identification model and the target foreign object identification model to identify foreign object information from the image to achieve automatic identification of potential hazards in power transmission lines. Compared with manual inspection, this type of method has significant advantages in terms of inspection efficiency, inspection range, and personnel safety.

[0004] However, existing image-based methods for identifying potential hazards in power transmission lines still have certain shortcomings in practical applications. Firstly, under complex weather conditions such as windstorms, power transmission lines sway significantly due to wind force, while the background environment (such as towers, ground, and vegetation) remains relatively static. This makes it difficult to distinguish the power transmission lines from the background in a single image, easily introducing a large amount of background interference information and affecting the accuracy of hazard identification. Existing methods often rely on complex target detection or segmentation models to suppress background interference. These models have complex structures, high computational costs, and insufficient robustness in strong wind scenarios.

[0005] Secondly, due to the disorderly swaying of power lines and cameras during windstorms, the width of the power line image fluctuates irregularly and becomes blurred. The image width of the same transmission line varies significantly across different images. Most existing hazard identification methods do not effectively model these variations in transmission line image width, assuming a consistent line width across different images. This assumption is difficult to apply in real-world inspection scenarios, especially when side-shot, oblique-shot, or wind-induced displacement occurs. Changes in the line image width directly affect the model's ability to identify fine-grained hazard features such as broken strands and foreign objects.

[0006] In addition, although some existing technologies have introduced deep learning models to extract features and classify transmission line images, most of them only use the image as the sole input and fail to fully integrate the geometric prior information of the transmission line. This results in poor adaptability of the model to transmission lines of different specifications and wire diameters, and limited generalization performance.

[0007] Therefore, how to effectively suppress background interference and highlight the characteristics of the transmission line itself in complex environments such as wind disasters, and how to reasonably model the image features in combination with the actual width information of the transmission line, so as to improve the accuracy and robustness of the identification of hidden dangers of transmission lines, remains a technical problem that needs to be solved in the existing technology. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for identifying potential hazards in power transmission lines based on conductor width correction and differential imaging.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] This invention provides a method for identifying potential hazards in power transmission lines based on conductor width correction and differential imaging, comprising the following steps:

[0011] A power transmission line hazard identification model is constructed. The power transmission line hazard identification model is based on an improved Transformer network structure. The improved Transformer network structure includes an improved Transformer encoder and a multilayer perceptron. The improved Transformer encoder is provided with a modulation layer. The modulation layer is used to perform width modulation on the input features using the modulation parameters generated by the multilayer perceptron.

[0012] A training dataset including images of transmission lines and corresponding widths of transmission lines is constructed, and the hidden danger identification model of the transmission lines is trained based on the training dataset;

[0013] In the event of a windstorm, drones are used to perform differential photography of the power transmission lines to be inspected according to a preset inspection route to obtain a set of images to be inspected.

[0014] The image set to be detected is subjected to differential processing to eliminate background interference and highlight the transmission line area, thereby obtaining an input image set for hazard identification;

[0015] Based on the optical imaging relationship and the UAV spatial position information, the line imaging width of each input image in the input image set is corrected to obtain the transmission line width correction value corresponding to each input image.

[0016] The input image set and the corresponding transmission line width correction value are input into the trained transmission line hazard identification model, and the hazard identification result is output.

[0017] Furthermore, the improvement to the Transformer encoder includes setting a modulation layer for width modulation of intermediate features after the normalization layer and the multi-head self-attention layer of the Transformer encoder.

[0018] The multilayer perceptron is used to perform feature mapping on the transmission line width correction value and generate modulation parameters for controlling the modulation layer to perform width modulation on the intermediate features.

[0019] Furthermore, the modulation layer is used to perform width modulation on the output features of the normalized layer and the output features of the multi-head self-attention layer based on the modulation parameters output by the multilayer perceptron, thereby embedding the transmission line width information into the feature representation of the Transformer encoder, as shown below:

[0020]

[0021]

[0022] in, , , These are the modulation parameters output by the multilayer sensor; , These are the output features of the normalization layer and the multi-head self-attention layer, respectively. The output features of the normalized layer after modulation; This represents the modulated output features of the multi-head self-attention layer.

[0023] Furthermore, the construction of a training dataset including transmission line images and corresponding transmission line widths, and the training of the transmission line hazard identification model based on the training dataset, specifically includes:

[0024] Images of power transmission lines with varying widths and different hazard labels were acquired using drones, and the actual width of each transmission line was labeled for each image. and actual hazard category labels ;

[0025] Image of the power transmission line Corresponding to the actual width of the transmission line The combined input features are used to label real-world hazard categories. As the training objective, construct the training dataset. :

[0026]

[0027] in, This represents the total number of images in the training dataset; Indicates the first Images of power transmission lines; Indicates the first The actual width of the transmission line, i.e., the actual diameter of the corresponding transmission line; Indicates the first The actual hazard category labels corresponding to the images of the power transmission lines; the hazard category labels include normal, broken conductor strands, and foreign object entanglement;

[0028] Transmission line images from the training dataset Corresponding to the actual width of the transmission line The improved Transformer encoder and multilayer perceptron of the power transmission line hazard identification model are input respectively, and the predicted hazard category label is output.

[0029] Based on the predicted hazard category labels and the actual hazard category labels, the cross-entropy loss function is calculated. By minimizing the cross-entropy loss function, the parameters of the transmission line hazard identification model are optimized, thereby enabling the training of the transmission line hazard identification model.

[0030] Furthermore, in the scenario of a windstorm, the method of using a drone to perform differential imaging of the power transmission line to be inspected according to a preset inspection route to obtain an image set for inspection specifically includes:

[0031] In the scenario of a wind disaster, the drone flies along a preset inspection route. When it reaches each collection point in the preset inspection route, it collects the current three-dimensional position coordinates of the drone through the drone's GPS.

[0032] At each acquisition point, images were taken at preset time intervals to obtain two images of the transmission line taken at different times, which were then recorded as follows: and ; Indicates the first The previous image taken at each collection point; Indicates the first The last image taken at each collection point;

[0033] Based on the transmission line images acquired from each collection point, a set of images to be detected is constructed. ; This indicates the total number of data collection points in the preset inspection route.

[0034] Furthermore, the set of images to be detected undergoes differential processing to eliminate background interference and highlight the transmission line area, thereby obtaining an input image set for hazard identification, specifically including:

[0035] Image set to be detected Pixel-level difference operations are performed on each pair of preceding and following transmission line images to obtain the corresponding difference image. The formula is:

[0036]

[0037] in, For the first i Differential images of each acquisition point;

[0038] All the difference images obtained As input images for hazard identification, an input image set for hazard identification is obtained. .

[0039] Furthermore, the formula for the transmission line width correction value is as follows:

[0040]

[0041]

[0042] in, This represents the power line width correction value for the corresponding input image in a windstorm scenario; This represents the original imaging width of the transmission line in the input image, that is, the pixel width of the transmission line in the input image after differential processing; This indicates the three-dimensional position coordinates collected by GPS when the drone takes images in a windstorm scene; This indicates the three-dimensional position coordinates of the drone when it takes a reference image at the corresponding collection point in a windless scene, which is the reference coordinate of the collection point corresponding to the preset inspection route; This indicates the coordinates of the orthogonal imaging position of the UAV along the normal direction of the power transmission line in the reference shooting state; The focal length for a camera mounted on a drone; This represents the direction vector component of the actual width of the transmission line in the three-dimensional spatial coordinate system.

[0043] Furthermore, the direction vector components are determined based on the actual width of the transmission line. , This refers to the actual width of the transmission line.

[0044] Furthermore, the step of inputting the input image set and the corresponding transmission line width correction value into the trained transmission line hazard identification model and outputting the hazard identification result specifically includes:

[0045] The input images in the input image set are sequentially fed into the improved Transformer encoder of the trained power transmission line hazard identification model to extract features from the input images;

[0046] Correct the width of the transmission line corresponding to the input image. The feature vector is constructed by feature mapping and then synchronously input into the multilayer perceptron in the power transmission line hidden danger identification model.

[0047] The width feature vector is nonlinearly transformed by the multilayer perceptron to generate modulation parameters for width modulation of intermediate features of the improved Transformer encoder.

[0048] In the improved Transformer encoder, the image features are subjected to width modulation processing based on the modulation parameters to obtain a feature representation that integrates transmission line width correction information.

[0049] Based on the feature representation of the fused transmission line width correction information, the transmission line hazard identification result corresponding to the input image is obtained through the output layer of the transmission line hazard identification model.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] (1) In view of the problem that the imaging width of the same transmission line in different images in the prior art is significantly different, but most hidden danger identification methods do not model the changes in imaging width, the present invention calculates the transmission line width correction value based on the three-dimensional spatial position of the UAV and the optical imaging relationship, thereby realizing the numerical correction of the line width fluctuation under wind disaster, making the line width information input to the model more accurate and consistent, thereby improving the model's ability to distinguish fine-grained hidden dangers and robustness, and avoiding misidentification caused by changes in imaging width.

[0052] (2) In view of the problem that the images captured by inspection drones in wind disaster scenarios are easily affected by background interference and cannot accurately extract the conductor target, the present invention can effectively eliminate the static background area by performing pixel-level differential processing on the two images of the same collection point during the inspection process, and only retain the characteristics of the transmission line and its possible dynamic hidden dangers. This enables the conductor area to be highlighted under wind-induced swaying conditions, improves the quality of the input image for hidden danger identification, and enables the subsequent hidden danger identification model to accurately capture fine-grained features such as broken strands and foreign object entanglement.

[0053] (3) In view of the problem that the existing transmission line hazard identification model does not fully integrate the line width information with the image features, the present invention improves the Transformer network structure by setting a width modulation layer after the encoder normalization layer and the multi-head self-attention layer, and maps the line width correction value to the modulation parameter through a multilayer perceptron, thereby embedding the line width information into the Transformer feature expression, so that the model can dynamically adjust the attention weight and feature intensity during the feature extraction process, thereby enhancing the sensitivity to minor abnormalities of the conductor and improving the accuracy of identifying hazards such as broken strands and foreign object entanglement.

[0054] (4) To address the problem that existing models cannot perform real-time adaptive width modulation in wind disaster scenarios, this invention inputs the input image and the corresponding width correction value into the improved Transformer encoder and the multilayer perceptron respectively during the model inference stage. It then generates modulation parameters using width feature vectors to modulate the image features, enabling the model to dynamically correct the features of each image under wind disaster conditions. This makes the hazard identification results more accurate and stable, improving the reliability of identification in complex environments. By combining the improved Transformer encoder with the multilayer perceptron to fuse and modulate the line width information, and simultaneously using differential images to enhance the dynamic features of the conductors, the model achieves joint identification of multiple types of hazards. This allows the model to accurately identify different types of fine-grained hazards such as broken strands and foreign object entanglement, improving inspection efficiency and safety. Attached Figure Description

[0055] Figure 1 This is a flowchart of the power transmission line hazard identification method according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of a power transmission line hazard identification model according to an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of a method for correcting the width of a conductor image during a windstorm, according to an embodiment of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0059] Example 1:

[0060] This embodiment provides a method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging, such as... Figure 1 As shown, it includes the following steps:

[0061] Step S1: Construct a hidden danger identification model for transmission lines;

[0062] like Figure 2 As shown, in this embodiment, the transmission line hazard identification model is based on an improved Transformer network structure. Specifically, the improved Transformer network structure includes two main parts: an improved Transformer encoder and a multilayer perceptron (MLP).

[0063] The key feature of the improved Transformer encoder lies in its introduction of a width modulation mechanism. Traditional Transformer encoders typically rely on standard normalization layers and multi-head self-attention layers. In this embodiment, a modulation layer is added after these layers. The modulation layer adjusts intermediate features based on the input width correction information, enabling the Transformer encoder to effectively handle input features under different width conditions. Through this modulation layer, the width information of the transmission line can be embedded into the feature representation, thereby improving the robustness and accuracy of the hazard identification model under various conditions.

[0064] MLP is used for feature mapping of transmission line width correction values. Specifically, during image processing, the width correction value is calculated based on optical imaging relationships and UAV position corrections, and then modulation parameters are generated through a multilayer perceptron. These modulation parameters are used to control the modulation layer to modulate the width of intermediate features, enabling the Transformer encoder to better adapt to input images with different imaging widths.

[0065] The function of the modulation layer is to perform width modulation on the output features of the normalized layer and the multi-head self-attention layer based on the modulation parameters output by the multilayer perceptron. The modulated features are represented as follows:

[0066]

[0067]

[0068] in, , , These are the modulation parameters output by the multilayer sensor; , These are the output features of the normalization layer and the multi-head self-attention layer, respectively. The output features of the normalized layer after modulation; This represents the modulated output features of the multi-head self-attention layer. In this way, the Transformer encoder can combine different imaging width information to achieve more accurate hazard identification.

[0069] By introducing a modulation layer, the model can adjust its feature representation according to different width information, thereby maintaining high recognition accuracy in various complex scenarios. Especially in extreme environments such as windstorms, its ability to identify fine-grained hazards such as broken wire strands and entangled foreign objects is significantly improved. A multilayer perceptron is used to generate modulation parameters. Through this process, the model can dynamically and adaptively adjust according to the width information of the input image, rather than relying on fixed parameter settings. This allows the hazard identification model to exhibit high adaptability in various inspection situations.

[0070] Step S2: Construct a training dataset including images of transmission lines and corresponding transmission line widths, and train the transmission line hazard identification model based on the training dataset, specifically including:

[0071] Images of power transmission lines with varying widths and different hazard labels were acquired using drones, and the actual width of each transmission line was labeled for each image. and actual hazard category labels ;

[0072] Image of transmission lines Corresponding to the actual width of the transmission line The combined input features are used to label real-world hazard categories. As the training objective, construct the training dataset. :

[0073]

[0074] in, This represents the total number of images in the training dataset; Indicates the first Images of power transmission lines; Indicates the first The actual width of the transmission line, i.e., the actual diameter of the corresponding transmission line; Indicates the first The actual hazard category labels corresponding to the images of the power transmission lines; the hazard category labels include normal, broken conductor strands, and foreign object entanglement;

[0075] Transmission line images from the training dataset Corresponding to the actual width of the transmission line The improved Transformer encoder and multilayer perceptron of the power transmission line hazard identification model are input respectively, and the predicted hazard category label is output.

[0076] Based on the predicted and actual hazard category labels, a cross-entropy loss function is calculated. The parameters of the transmission line hazard identification model are optimized by minimizing this cross-entropy loss function, thus training the model. In step S2, the training dataset effectively optimizes the model's parameters, enabling it to accurately identify hazard categories in different transmission line images and effectively handle images with varying widths. The trained model exhibits strong robustness and can be applied to practical inspection scenarios for accurate hazard detection and risk assessment. By constructing a training dataset containing different transmission line widths and hazard labels, and through the fusion of a multilayer perceptron and an improved Transformer encoder, the model can effectively identify hazards under diverse inspection conditions, particularly handling the impact of width variations on hazard identification accuracy. Training with the cross-entropy loss function ensures that the model's predictions closely approximate the actual hazard categories, guaranteeing accurate hazard identification and providing solutions for complex situations encountered in actual inspections.

[0077] Step S3: In a windstorm scenario, a drone is used to perform differential imaging of the power transmission line to be inspected along a preset inspection route to obtain a set of images to be inspected, specifically including:

[0078] In typhoon scenarios, power transmission lines often sway continuously due to wind, causing significant changes in the shape and position of the conductors in images taken at different times. This variation makes it difficult for traditional image comparison methods to directly and effectively identify potential hazards in power transmission lines. Therefore, differential imaging becomes an effective method in this environment because it can highlight the dynamically changing parts of the image—namely, the power transmission lines affected by the typhoon—while the background remains unchanged, reducing interference from irrelevant backgrounds.

[0079] In typhoon scenarios, drones fly along pre-set inspection routes and accurately record the three-dimensional coordinates of each data collection point using their GPS system. Because the attitude and position of power transmission lines constantly change during typhoons, accurate spatial location recording ensures that each image precisely corresponds to a specific power transmission line area.

[0080] At each acquisition point, two images are taken at preset time intervals to acquire the previous image. And the next image The core of this method lies in the fact that, due to the dynamic changes of transmission lines under windy conditions, the difference between two images mainly lies in the changes in the conductors, while the background area remains almost unchanged due to the relative stability of the environment. Therefore, the changes in the differential image can effectively reflect the dynamic changes of the transmission lines.

[0081] Based on the transmission line images acquired from each collection point, a set of images to be detected is constructed. ; This indicates the total number of data collection points in the preset inspection route.

[0082] This method of image acquisition effectively eliminates background interference in windstorm scenes, focusing attention on power transmission lines affected by wind. Differential imaging highlights the characteristics of potential hazards in dynamically changing sections of power transmission lines, playing a crucial role in detecting fine-grained hazards such as broken conductor strands and foreign object entanglement. Furthermore, utilizing the drone's GPS information and timed image capture ensures that each acquired image has a relatively consistent viewpoint and spatial positioning, thereby improving the accuracy and reliability of subsequent image differential analysis and hazard identification.

[0083] Step S4: Perform differential processing on the image set to be detected to eliminate background interference and highlight the transmission line area, obtaining the input image set for hazard identification, specifically including:

[0084] In step S3, two images with different time intervals have been obtained through differential imaging by the drone. However, these images still contain background parts and irrelevant static elements. In windstorm scenarios, the dynamic changes of power transmission lines are key to identifying potential hazards. Differential processing can effectively eliminate background interference and focus attention on the dynamically changing parts of the power transmission lines, thereby improving the accuracy of hazard identification.

[0085] Image set to be detected Pixel-level difference operations are performed on each pair of preceding and following transmission line images to obtain the corresponding difference image. The formula is:

[0086]

[0087] in, For the first i Differential images of each acquisition point;

[0088] All the difference images obtained As input images for hazard identification, an input image set for hazard identification is obtained. .

[0089] By performing differential processing on the image set to be detected, background interference in the windstorm environment can be effectively eliminated, highlighting the dynamic changes in the transmission line area. In this way, the differential images become more targeted, focusing on the changes in the transmission lines at different points in time, enabling subsequent hazard identification models to more accurately identify fine-grained hazard features, such as broken conductor strands and foreign object entanglement. This differential processing not only improves the accuracy of hazard identification but also effectively reduces the interference of background noise on the identification results, thereby enhancing the robustness and reliability of the entire hazard identification system.

[0090] Step S5: Based on the optical imaging relationship and the UAV spatial position information, correct the line imaging width of each input image in the input image set to obtain the transmission line width correction value corresponding to each input image;

[0091] In windstorm scenarios, power transmission lines may sway or deform due to wind force, causing variations in their image width at different shooting angles. To accurately identify potential hazards in power transmission lines, the line width in the image must be corrected. This correction process needs to take into account optical imaging relationships and the spatial position of the drone.

[0092] like Figure 3 As shown, in a windstorm scenario, the width of the power transmission line in the image is affected by multiple factors, such as the shooting angle and imaging distance, due to the dynamic shooting of the drone and the influence of wind. These factors cause changes in the image width, affecting the accuracy of the hazard identification model. Therefore, it is necessary to correct the image width of each input image so that a standardized width value can be obtained regardless of the change in the imaging angle, thereby ensuring that the model can accurately identify fine-grained hazard features. Furthermore, the drone's shooting position may shift due to wind and other reasons, leading to instability in the shooting angle and position. Therefore, it is necessary to spatially correct the width of the power transmission line based on the difference between the actual shooting position of the drone and the reference position. The calculation of the correction value needs to take into account the change in the drone's position during the windstorm.

[0093] The formula for the correction value of transmission line width is:

[0094]

[0095]

[0096] in, This represents the power line width correction value for the corresponding input image in a windstorm scenario; This represents the original imaging width of the transmission line in the input image, that is, the pixel width of the transmission line in the input image after differential processing; This indicates the three-dimensional position coordinates collected by GPS when the drone takes images in a windstorm scene; This indicates the three-dimensional position coordinates of the drone when it takes a reference image at the corresponding collection point in a windless scene, which is the reference coordinate of the collection point corresponding to the preset inspection route; This indicates the coordinates of the orthogonal imaging position of the UAV along the normal direction of the power transmission line in the reference shooting state; The focal length for a camera mounted on a drone; This represents the direction vector components of the actual width of the transmission line in a three-dimensional coordinate system. The direction vector components are determined based on the actual width of the transmission line. , This refers to the actual width of the transmission line.

[0097] Step S5 ensures accurate correction of the transmission line imaging width, eliminating errors caused by drone position offset and shooting angle changes in windy environments. By combining optical imaging relationships and drone spatial position information, the corrected width accurately reflects the actual width of the transmission line, thus providing more precise input for subsequent hazard identification. This method effectively improves the robustness of hazard identification, especially in complex windy environments, maintaining high identification accuracy and ensuring timely detection and handling of transmission line hazards.

[0098] Step S6: Input the input image set and the corresponding transmission line width correction value into the trained transmission line hazard identification model, and output the hazard identification result, specifically including:

[0099] The input images in the input image set are sequentially fed into the improved Transformer encoder of the trained power transmission line hazard identification model to extract features from the input images;

[0100] Correct the width of the transmission line corresponding to the input image. The feature vector is constructed by feature mapping and then synchronously input into the multilayer perceptron in the power transmission line hidden danger identification model.

[0101] The width feature vector is nonlinearly transformed by a multilayer perceptron to generate modulation parameters for width modulation of intermediate features of the improved Transformer encoder.

[0102] In the improved Transformer encoder, the image features are width-modulated based on the modulation parameters to obtain a feature representation that incorporates the transmission line width correction information.

[0103] Based on the feature representation that integrates transmission line width correction information, the output layer of the transmission line hazard identification model is used to obtain the transmission line hazard identification result corresponding to the input image.

[0104] Example 2:

[0105] This embodiment provides a power transmission line hazard identification system based on width correction and differential imaging, including:

[0106] Image Acquisition Module: This module consists of a drone that flies along a pre-set inspection route in typhoon-affected areas, capturing images of power transmission lines using an onboard high-definition camera. In typhoon conditions, the power lines sway violently due to wind, causing the drone's position to shift. The image acquisition module periodically takes two images at set time intervals and records the three-dimensional coordinates of each acquisition point using a GPS system to ensure accurate recording of spatial information during image acquisition.

[0107] Differential Processing Module: This module performs pixel-level differential processing on two consecutive images of the transmission line. By comparing two images of the same location, background interference is eliminated and the transmission line area is highlighted. Differential processing generates a differential image, eliminating background changes in windstorm scenes and making changes in the transmission line area more apparent. The differential processing module can automatically generate an input image set based on the input image set for subsequent training and prediction of the hazard identification model.

[0108] Transmission Line Width Correction Module: This module corrects the width of the transmission line in the input image based on optical imaging relationships and the spatial position information of the UAV. Especially in windstorm scenarios, the width of the transmission line can change in images taken at different times due to UAV position shifts and the influence of wind. The width correction module uses the UAV's shooting position and direction vector to correct the width of the line in the image, ensuring that the image width matches the actual width of the transmission line, thereby improving the accuracy of hazard identification.

[0109] Hazard Identification Model: This model is based on an improved Transformer network structure, including an improved Transformer encoder and a multilayer perceptron. The input corrected width and difference images serve as training data. After processing by the hazard identification model, the output hazard category label is determined. The improved Transformer encoder introduces a width modulation mechanism during feature extraction, which can embed the width information of the transmission line into the feature representation, thereby improving the model's ability to identify fine-grained hazard features such as broken strands and foreign object entanglement.

[0110] Training and Optimization Module: This module is used to train the transmission line hazard identification model and optimize its parameters. By constructing a training dataset that includes transmission line images and their corresponding widths, as well as hazard labels, the model is trained using the cross-entropy loss function, ultimately enabling the hazard identification model to accurately identify different hazard types.

[0111] Output module: Based on the output of the hazard identification model, the system will provide the identified hazard category (such as normal, broken wire strand, foreign object entanglement, etc.). The identification results can be displayed on the interface for relevant personnel to further analyze and process.

[0112] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging, characterized in that, Includes the following steps: A power transmission line hazard identification model is constructed. The power transmission line hazard identification model is based on an improved Transformer network structure. The improved Transformer network structure includes an improved Transformer encoder and a multilayer perceptron. The improved Transformer encoder is provided with a modulation layer. The modulation layer is used to perform width modulation on the input features using the modulation parameters generated by the multilayer perceptron. A training dataset including images of transmission lines and corresponding widths of transmission lines is constructed, and the hidden danger identification model of the transmission lines is trained based on the training dataset; In the event of a windstorm, drones are used to perform differential photography of the power transmission lines to be inspected according to a preset inspection route to obtain a set of images to be inspected. The image set to be detected is subjected to differential processing to eliminate background interference and highlight the transmission line area, thereby obtaining an input image set for hazard identification; Based on the optical imaging relationship and the UAV's spatial position information, the line imaging width of each input image in the input image set is corrected to obtain the transmission line width correction value corresponding to each input image. The input image set and the corresponding transmission line width correction value are input into the trained transmission line hazard identification model, and the hazard identification result is output.

2. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 1, characterized in that, The improvements to the Transformer encoder include setting a modulation layer for width modulation of intermediate features after the normalization layer and the multi-head self-attention layer of the Transformer encoder. The multilayer perceptron is used to perform feature mapping on the transmission line width correction value and generate modulation parameters for controlling the modulation layer to perform width modulation on the intermediate features.

3. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 2, characterized in that, The modulation layer is used to perform width modulation on the output features of the normalized layer and the output features of the multi-head self-attention layer based on the modulation parameters output by the multilayer perceptron, thereby embedding the transmission line width information into the feature representation of the Transformer encoder, as shown below: ， ， in, , , These are the modulation parameters output by the multilayer sensor; , These are the output features of the normalization layer and the multi-head self-attention layer, respectively. The output features of the normalized layer after modulation; This represents the modulated output features of the multi-head self-attention layer.

4. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 1, characterized in that, The construction of the training dataset, which includes images of transmission lines and corresponding widths of transmission lines, and the training of the transmission line hazard identification model based on the training dataset, specifically includes: Images of power transmission lines with varying widths and different hazard labels were acquired using drones, and the actual width of each transmission line was labeled for each image. and actual hazard category labels ; Image of the transmission line Corresponding to the actual width of the transmission line The combined input features are used to label real-world hazard categories. As the training objective, construct the training dataset. : ， in, This represents the total number of images in the training dataset; Indicates the first Images of power transmission lines; Indicates the first The actual width of the transmission line, i.e., the actual diameter of the corresponding transmission line; Indicates the first The actual hazard category labels corresponding to the images of the power transmission lines; the hazard category labels include normal, broken conductor strands, and foreign object entanglement; Transmission line images from the training dataset Corresponding to the actual width of the transmission line The improved Transformer encoder and multilayer perceptron of the power transmission line hazard identification model are input respectively, and the predicted hazard category label is output. Based on the predicted hazard category labels and the actual hazard category labels, the cross-entropy loss function is calculated. By minimizing the cross-entropy loss function, the parameters of the transmission line hazard identification model are optimized, thereby enabling the training of the transmission line hazard identification model.

5. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 1, characterized in that, In the context of a windstorm, the method involves using drones to differentially photograph the power transmission lines to be inspected along a pre-set inspection route to acquire a set of images. Specifically, this includes: In the scenario of a wind disaster, the drone flies along a preset inspection route. When it reaches each collection point in the preset inspection route, it collects the current three-dimensional position coordinates of the drone through the drone's GPS. At each acquisition point, images were taken at preset time intervals to obtain two images of the transmission line taken at different times, which were then recorded as follows: and ; Indicates the first The previous image taken at each collection point; Indicates the first The last image taken at each collection point; Based on the transmission line images acquired from each collection point, a set of images to be detected is constructed. ; This indicates the total number of data collection points in the preset inspection route.

6. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 5, characterized in that, The set of images to be detected is subjected to differential processing to eliminate background interference and highlight the transmission line area, thereby obtaining an input image set for hazard identification, specifically including: Image set to be detected Pixel-level difference operations are performed on each pair of preceding and following transmission line images to obtain the corresponding difference image. The formula is: ， in, For the first i Differential images of each acquisition point; All the difference images obtained As input images for hazard identification, an input image set for hazard identification is obtained. .

7. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 1, characterized in that, The formula for the transmission line width correction value is as follows: ， ， in, This represents the power line width correction value for the corresponding input image in a windstorm scenario; This represents the original imaging width of the transmission line in the input image, that is, the pixel width of the transmission line in the input image after differential processing; This indicates the three-dimensional position coordinates collected by GPS when the drone takes images in a windstorm scene; This indicates the three-dimensional position coordinates of the drone when it takes a reference image at the corresponding collection point in a windless scene, which is the reference coordinate of the collection point corresponding to the preset inspection route; This indicates the coordinates of the orthogonal imaging position of the UAV along the normal direction of the power transmission line in the reference shooting state; The focal length for a camera mounted on a drone; This represents the direction vector component of the actual width of the transmission line in the three-dimensional spatial coordinate system.

8. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 7, characterized in that, The direction vector components are determined based on the actual width of the transmission line. , This refers to the actual width of the transmission line.

9. The method for identifying potential hazards in transmission lines based on conductor width correction and differential imaging according to claim 1, characterized in that, The step of inputting the input image set and the corresponding transmission line width correction value into the trained transmission line hazard identification model and outputting the hazard identification result specifically includes: The input images in the input image set are sequentially fed into the improved Transformer encoder of the trained power transmission line hazard identification model to extract features from the input images; Correct the width of the transmission line corresponding to the input image. The feature vector is constructed by feature mapping and then synchronously input into the multilayer perceptron in the power transmission line hidden danger identification model. The width feature vector is nonlinearly transformed by the multilayer perceptron to generate modulation parameters for width modulation of intermediate features of the improved Transformer encoder. In the improved Transformer encoder, the image features are subjected to width modulation processing based on the modulation parameters to obtain a feature representation that incorporates transmission line width correction information. Based on the feature representation of the fused transmission line width correction information, the transmission line hazard identification result corresponding to the input image is obtained through the output layer of the transmission line hazard identification model.

Images