A visual recognition-based power transmission tower deformation evaluation method

Abstract

The application provides a power transmission tower deformation evaluation method based on visual identification, and relates to the technical field of visual identification, and comprises the following steps: obtaining tower building information through interaction, and obtaining a power transmission tower model through restoration; performing structural deviation visual identification to obtain a tower deformation information set; positioning and connecting nodes to perform connection deviation visual identification to obtain a risk connection node set; presetting an array of flaw detection points to perform ultrasonic flaw detection to obtain an ultrasonic detection data set; performing internal defect identification to obtain an internal defect information set; fitting the internal defect information set, the risk connection node set and the tower deformation information set to the power transmission tower model to obtain a tower defect analysis model; and performing deformation risk quantitative evaluation to obtain a deformation risk coefficient. The application solves the technical problem that the existing power transmission tower deformation evaluation usually adopts single appearance detection, which leads to omission or misjudgment of complex deformation, and manual intervention is more, which leads to low accuracy and efficiency of evaluation.

Description

Technical Field

[0001] This invention relates to the field of visual recognition technology, and more specifically to a method for evaluating the deformation of power transmission towers based on visual recognition. Background Technology

[0002] Transmission towers are crucial structural components in power systems, and their structural safety directly affects the reliability and stability of power transmission. Over time and with changes in environmental conditions, transmission towers may experience deformation, damage to connection points, or internal defects, leading to reduced structural safety. Current technologies for assessing transmission tower deformation typically employ a single visual inspection method. On one hand, existing technologies struggle to effectively process and accurately analyze the large amounts of data generated by inspection equipment, and the lack of unified standards for deformation and defect assessment results in poor accuracy. On the other hand, relying on a single inspection method can lead to omissions or misjudgments of complex deformations, necessitating manual intervention when dealing with complex deformations, resulting in low levels of automation and inspection efficiency. Summary of the Invention

[0003] This application provides a visual recognition-based method for evaluating the deformation of transmission towers, aiming to solve the technical problems of existing technologies that typically rely on a single visual inspection for evaluating the deformation of transmission towers, leading to omissions or misjudgments of complex deformations, and requiring significant manual intervention, resulting in low accuracy and efficiency in the evaluation.

[0004] This application discloses a visual recognition-based method for assessing the deformation of transmission towers. The method includes: interactively obtaining the tower construction information of a target transmission tower, and reconstructing a transmission tower model based on the tower construction information; performing structural deviation visual recognition on the target transmission tower based on the transmission tower model to obtain a tower deformation information set; locating connection nodes on the target transmission tower according to the tower construction information, and performing connection deviation visual recognition on the obtained connection node set to obtain a risk connection node set; pre-setting a flaw detection point array, and performing ultrasonic flaw detection on the target transmission tower based on the flaw detection point array to obtain an ultrasonic testing dataset; identifying internal defects of the target transmission tower based on the ultrasonic testing dataset to obtain an internal defect information set; fitting the internal defect information set, the risk connection node set, and the tower deformation information set to the transmission tower model to obtain a tower defect analysis model; and performing a deformation risk quantification assessment on the target transmission tower based on the tower defect analysis model to obtain a deformation risk coefficient.

[0005] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0006] By interactively acquiring tower construction information and reconstructing transmission tower models, a digital model of the target transmission tower can be accurately constructed, providing a foundation for subsequent visual recognition and analysis. Based on the transmission tower model, structural deviation visual recognition of the target tower can comprehensively detect tower deformation, obtain detailed deformation information, accurately identify structural deviations, and promptly detect potential deformation problems. By locating connection nodes and performing connection deviation visual recognition, risky connection node sets can be accurately identified. This allows for prioritizing potentially risky connection points during maintenance and inspection, improving maintenance efficiency and accuracy. Pre-setting a flaw detection point array and performing ultrasonic flaw detection enables in-depth detection of internal defects in the transmission tower. This non-destructive testing method ensures accurate identification of internal defects without damaging the tower. Further damage was assessed, and detailed internal defect data was provided. Internal defect identification was performed based on the ultrasonic testing dataset, acquiring an internal defect information set, enabling a comprehensive understanding of internal structural defects and providing a reliable basis for subsequent risk assessment and maintenance decisions. The internal defect information set, risk connection node set, and tower deformation information set were fitted to the transmission tower model to obtain a tower defect analysis model. This comprehensive analysis model integrates information from different sources, providing a comprehensive assessment of tower defects and risks, ensuring the comprehensiveness and accuracy of the analysis results. Based on the tower defect analysis model, the deformation risk of the target transmission tower was quantitatively assessed, obtaining a deformation risk coefficient. This quantifies various test results and analysis data, comprehensively judging the overall risk level of the tower, providing a scientific basis for maintenance and management decisions, and reducing operational risks.

[0007] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0008] Figure 1 This application provides a schematic flowchart of a vision recognition-based method for evaluating the deformation of power transmission towers.

[0009] Figure 2 This application provides a schematic diagram of the process for obtaining the tower deformation information set in a vision-based tower deformation assessment method. Detailed Implementation

[0010] This application provides a vision-based method for evaluating the deformation of power transmission towers. This method addresses the technical problems of existing technologies that typically rely on a single visual inspection for evaluating the deformation of power transmission towers, leading to omissions or misjudgments of complex deformations. Furthermore, the method involves significant manual intervention, resulting in low accuracy and efficiency in the evaluation.

[0011] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0012] like Figure 1 As shown in the figure, this application provides a method for evaluating the deformation of transmission towers based on visual recognition. The method includes:

[0013] The tower construction information of the target transmission tower is obtained interactively, and the transmission tower model is reconstructed based on the tower construction information.

[0014] The tower construction information of the target transmission tower is collected from data sources such as design drawings and construction records. This includes the tower's geometry, material properties, geographical location, and environmental conditions. The collected tower construction information is cleaned and preprocessed to ensure data standardization. Based on the preprocessed construction information, key tower parameters are extracted, and a 3D modeling tool is used to create a model of the transmission tower.

[0015] Based on the transmission tower model, the structural deviation visual recognition is performed on the target transmission tower to obtain the tower deformation information set.

[0016] Images are captured from various angles of the target transmission tower according to a preset array of shooting points, ensuring coverage of the tower's key structures to obtain actual images. Simultaneously, virtual images are captured from the same points on the tower model, ensuring consistency with the actual images in terms of shooting points and viewing angles. The actual and virtual images are then spatially aligned and registered. Deviation detection is performed on the aligned images to identify areas of difference between the actual and virtual images, representing deformations or offsets in the tower structure.

[0017] The identified multiple differential regions are analyzed, and local deformation data of each region is extracted. This data includes, but is not limited to, the magnitude and direction of the deformation, as well as the spatial location of the deformation region. The multiple local deformation data are summarized to construct a complete tower deformation information set. This information set can reflect the structural deviation of the target transmission tower in the actual environment and provide data support for subsequent risk assessment.

[0018] Based on the tower construction information, the connection nodes are located on the target transmission tower. The obtained connection node set is visually identified for connection deviation to obtain the risk connection node set.

[0019] Based on the tower construction information, all key connection nodes on the target transmission tower are identified, and a set of connection nodes is obtained. These connection nodes are important support points of the tower structure, including the connection between the tower frame and the crossbeam, and the connection between the support rod and the main pole.

[0020] Based on the node locations in the connected node set, the UAV's flight path is planned. The planning considers the visibility and acquisition angle of each connected node, as well as the UAV's flight safety and the performance of the image acquisition equipment. Based on the planned path, a specific image acquisition path is determined to ensure coverage of all connected nodes, achieving optimal image coverage. The UAV is then controlled to acquire images of the connected nodes according to the image acquisition path.

[0021] Multiple sample connection deviation image sets of various sample connection nodes are obtained to train a deviation recognition model that can identify connection deviations. The collected connection node images are input into the model, and the model automatically identifies the connection nodes with deviations and locates the specific positions of these nodes on the tower, forming a risk connection node set.

[0022] A preset flaw detection point array is used, and ultrasonic flaw detection is performed on the target transmission tower based on the flaw detection point array to obtain an ultrasonic detection dataset.

[0023] Based on the structural characteristics and material properties of the tower, as well as the tower construction information obtained in the early stage, the distribution location of the flaw detection points is determined, and a flaw detection point array is generated. Flaw detection points are usually set in key stress parts and areas prone to defects, such as welds, connection nodes, support rods, etc.

[0024] Select a suitable ultrasonic flaw detector and configure multiple probe types to meet different testing needs. Place the probe at each preset flaw detection point, start the flaw detector, and the ultrasonic probe sends ultrasonic pulses into the tower. When the ultrasonic waves encounter materials or defects of different densities during propagation, they will generate echoes. The probe receives the echo signals and converts them into electrical signals, which are then transmitted to the flaw detector for processing. This generates ultrasonic waveform diagrams, time domain diagrams, spectrum diagrams, etc., which are then converted into specific testing parameters. Each flaw detection point is scanned, and the ultrasonic echo data of each point is recorded to obtain an ultrasonic testing dataset.

[0025] Based on the ultrasonic detection dataset, internal defects of the target transmission tower are identified to obtain an internal defect information set.

[0026] Based on the ultrasonic testing dataset, characteristic parameters related to internal defects are extracted, such as abnormal changes in echo amplitude, waveform distortion, slowdown in wave velocity, and changes in spectral characteristics. Internal defects in the target transmission tower are identified based on these characteristic parameters. For example, a threshold range for the characteristic parameters is set; signals exceeding or falling below this threshold are considered potentially defective. By analyzing the time and frequency domain characteristics of the ultrasonic signals, specific defect signal patterns are identified, such as cracks, holes, inclusions, and weld detachment. The ultrasonic signals at each flaw detection point are analyzed individually to identify existing internal defects and determine their specific location, size, depth, and other characteristics. The identified defect information is then integrated into an internal defect information set.

[0027] The internal defect information set, risk connection node set, and tower deformation information set are fitted to the transmission tower model to obtain the tower defect analysis model.

[0028] Based on a unified coordinate system, internal defects, risk connection nodes, and deformation information are aligned to the same spatial frame, so that they are all represented in the same coordinate system, which can be accurately located in the transmission tower model. The coordinate-aligned data is input into the transmission tower model and fitted to the structural frame of the model, so that each internal defect, risk connection node, and deformation information can be accurately reflected in the model. Based on the fitted data, a complete tower defect analysis model containing internal defects, risk connection nodes, and deformation information is generated.

[0029] Based on the tower defect analysis model, the deformation risk of the target transmission tower is quantitatively assessed to obtain the deformation risk coefficient.

[0030] Deformation-related data is extracted from the tower defect analysis model, including the magnitude, location, and direction of deformation, as well as other influencing factors. Based on the deformation data, the connection risk dispersion coefficient, internal defect risk coefficient, local deformation risk coefficient, and deformation risk dispersion coefficient are obtained. The above coefficients are integrated and calculated based on the tower deformation evaluation function to obtain the deformation risk coefficient. The obtained deformation risk coefficient provides maintenance personnel with a quantitative basis for risk assessment, enabling a more scientific evaluation of the impact of tower deformation on structural safety and the formulation of corresponding management and maintenance strategies.

[0031] Furthermore, such as Figure 2 As shown, based on the transmission tower model, structural deviation visual recognition is performed on the target transmission tower to obtain a tower deformation information set. The method further includes:

[0032] A preset array of shooting points is used to acquire images of the target transmission tower based on the array of shooting points, thereby obtaining an array of actual tower images. The array of shooting points is then fitted to the transmission tower model, and image acquisition and replication are performed on the transmission tower model to obtain a virtual tower image array. Based on the array of shooting points, visual recognition of image deviations is performed on the mapping between the actual tower image array and the virtual tower image array to obtain a set of tower deformation information.

[0033] Based on the structural characteristics of transmission towers, multiple shooting points are preset to cover all key parts of the transmission tower, including the tower base, tower body, tower head, and connection nodes. These points are integrated to obtain a shooting point array. According to the height and complexity of the tower, appropriate shooting equipment, such as a high-resolution camera, is selected. The shooting equipment is controlled to capture images of the target transmission tower according to the preset shooting point array, ensuring that the images can fully reflect the actual structure of the tower. The captured images are then organized according to the shooting points to generate a complete array of actual tower images.

[0034] The array of shooting points is fitted to the transmission tower model to ensure that the spatial position and angle of the shooting points are consistent with the actual acquisition situation. Based on the fitted array of shooting points, the actual image acquisition process is simulated on the transmission tower model to generate a virtual image array of the tower corresponding to the physical image array of the tower.

[0035] Based on a preset array of shooting points, corresponding points in the physical image array and the virtual image array are mapped, with each shooting point having a corresponding image pair. The physical and virtual images of the same shooting point are spatially aligned. Image deviation analysis methods, such as calculating pixel or shape differences, are used to identify deviations between the physical and virtual images. Based on the deviation identification results, deformed regions in the physical images are identified; these deformed regions are those that show significant differences in image comparison. The identified deformed regions are then refined to determine specific deformation data, and integrated to obtain the tower deformation information set.

[0036] Furthermore, the method further includes performing image deviation visual recognition based on the mapping of the tower physical image array and the tower virtual image array according to the shooting point array to obtain the tower deformation information set; the method also includes:

[0037] Color features are extracted from the array of physical tower images to obtain tower color features. Image contrast is optimized using these tower color features as constraints to obtain an optimized physical image array. A first physical tower image and a first virtual tower image mapped to a first shooting point are extracted from the optimized physical image array and the virtual tower image array, where the first shooting point is any shooting point in the shooting point array. Spatial alignment is performed on the first physical tower image and the first virtual tower image to obtain image registration results. Deformation region identification is performed based on the image registration results to locate the first deformation region. This process is repeated, traversing the optimized physical image array and the virtual tower image array to identify multiple deformation regions. Tower deformation is numerically quantified based on these multiple deformation regions to obtain a tower deformation information set, which includes multiple local deformation data of the multiple deformation regions.

[0038] Extract the color histogram from the image of the tower entity, and statistically analyze the distribution ratio of each color in the image, including hue distribution, saturation distribution, etc., to form global color features. Based on important regions in the image, such as key structural parts of the tower, extract local color features. Region segmentation algorithms can be used to divide the image into multiple regions, and color features are extracted for each region. The extracted global color features and local color features are integrated into the tower color features for subsequent image processing.

[0039] Using the tower's color characteristics as a constraint, the objectives for contrast optimization are set, including enhancing the visual clarity of key parts of the tower while maintaining color consistency in the image. Specifically, for the global image, image equalization techniques are applied to improve contrast, making details in both dark and bright areas more apparent. This method enhances the overall contrast of the image while preserving color characteristics. For local images, adaptive histogram equalization is applied to ensure a more uniform contrast optimization effect under different brightness conditions, avoiding excessive enhancement that could increase image noise. After contrast optimization, an optimized entity image array is obtained.

[0040] Select any shooting point from the shooting point array as the first shooting point and use it as the object of analysis. Extract the first tower entity image corresponding to the first shooting point from the optimized entity image array, and extract the first tower virtual image corresponding to the first shooting point from the tower virtual image array. The first tower entity image and the first tower virtual image both come from the same shooting point, namely the first shooting point.

[0041] Spatial alignment is the process of aligning two images—the physical image of the first tower and the virtual image of the first tower—in a spatial dimension, ensuring that corresponding structures in the two images are aligned in the same position. This lays the foundation for subsequent image registration. Specifically, key feature points are extracted from the physical and virtual images of the first tower. These feature points can be corner points, edge points, or other significant structures in the images. A feature matching algorithm is used to match the feature points in the two images, identifying corresponding feature point pairs. Based on the feature point matching results, a spatial transformation is performed between the two images to align them in space, thus obtaining the image registration result.

[0042] Deformation regions refer to areas where there are significant differences between the physical and virtual images of a transmission tower after alignment. These differences may indicate physical deformation or structural damage to the tower. Pixel-level or feature-level difference detection is performed on the spatially aligned first physical and virtual images of the tower. Difference detection methods include calculating image differences and using edge detection algorithms. Based on the detected difference regions, the location and extent of the first deformation region are determined and marked on the images. If the difference detection result shows that the two images are identical and no difference region exists, it indicates that there is no deformation region, and the image pair is skipped.

[0043] The algorithm iterates through the physical image array and the virtual tower image array, performing the same deformation region identification process on each pair of images. During identification, if two images are completely identical (i.e., there are no differences), the pair is skipped; otherwise, the deformation regions are marked. After the iteration is complete, the identified deformation regions are summarized to obtain multiple deformation regions, which represent the deformation of the entire transmission tower at different locations and from different perspectives.

[0044] Numerical quantification of tower deformation refers to the process of numerically analyzing multiple identified deformation regions to convert the degree and characteristics of deformation into specific numerical data. Specifically, each deformation region is analyzed to extract feature data including deformation displacement distance, deformation angle, and deformation area. The displacement distance of the deformation region is calculated, which is the offset of the deformation point from the normal structure within the region. The deformation direction of the deformation region is quantified to determine the directionality of the deformation. The area of ​​the deformation region is measured to represent the extent of the deformation. The quantified data of each deformation region are integrated to form a deformation information set containing multiple local deformation data.

[0045] Furthermore, the method further includes quantifying the tower deformation of the target transmission tower based on the multiple deformation regions to obtain the tower deformation information set;

[0046] Guided by the multiple deformation regions, laser point cloud data is collected on the target transmission tower to obtain multiple local point cloud data; geometric modeling is performed based on the multiple local point cloud data to obtain multiple local entity models; guided by the multiple deformation regions, the multiple local entity models are fitted to the transmission tower model to perform tower deformation numerical quantification to obtain multiple local deformation data, and the multiple local deformation data constitute the tower deformation information set.

[0047] Based on the identified multiple deformation regions, the laser scanning device is precisely positioned in these regions to ensure that the focus of data acquisition is on the deformation regions. The laser scanning device is used to scan each deformation region to obtain three-dimensional point cloud data of these regions. Multiple local point cloud data are integrated to obtain multiple local point cloud data, which include the surface shape, contour and any minor deformation of the region.

[0048] Local point cloud data is converted into triangular meshes to form a continuous surface mesh model. The generated mesh model is smoothed and optimized to ensure the continuity and accuracy of the surface, especially for the details of the deformed area. Based on the optimized mesh model, a three-dimensional solid model of the deformed area is generated as a local solid model. This model can accurately reflect the geometric features of the area. Multiple local solid models are obtained by traversing the multiple local point cloud data.

[0049] The location of each deformation region within the overall transmission tower model is determined. Based on the location results, the positional correspondence between each local entity model and the overall transmission tower model is established. This ensures that the local entity models are accurately aligned with the overall transmission tower model, guaranteeing precise alignment of the local models of the deformation regions. After model fitting, geometric data of the deformation regions is extracted from the local entity models, including indicators such as deformation, angle deviation, and translation distance. The extracted geometric data is quantified to calculate the specific deformation value of each deformation region. For example, displacement or deviation values ​​are calculated to determine the specific deformation of the tower in these regions. All local deformation data are then integrated to form a tower deformation information set, which covers detailed deformation data for each deformation region of the target transmission tower.

[0050] Furthermore, based on the tower construction information, the method locates connection nodes on the target transmission tower, performs visual identification of connection deviations on the obtained set of connection nodes, and obtains a set of risky connection nodes. The method also includes:

[0051] The connection nodes are located in the transmission tower model to obtain the connection node set; a flight path is planned based on the connection node set to obtain an image acquisition path; the UAV is guided to acquire images of the target transmission tower according to the image acquisition path to obtain a connection node image sequence; connection deviation is visually identified in the connection node image sequence to locate the risk connection node set.

[0052] Using the structural information in the transmission tower model, the locations of connection nodes are identified. These nodes include key structural parts such as tower intersections, support points, and bolt connections. The identified connection nodes are marked in the model to form a set of connection nodes.

[0053] The set of connected nodes is input into the path planning algorithm, which generates the optimal flight path connecting all nodes. The algorithm comprehensively considers factors such as the shortest path distance, energy consumption, and obstacle avoidance requirements to generate an image acquisition path that covers all connected nodes.

[0054] The planned image acquisition path is loaded into the UAV's navigation system, and the UAV is controlled to automatically navigate and fly along the predetermined image acquisition path, arriving at the location of each connection node in sequence. The onboard camera equipment is then activated to acquire images of the connection nodes, and the image acquisition results are integrated according to the order of the connection nodes to obtain a sequence of connection node images.

[0055] Multiple sample connection deviation image sets of various sample connection nodes are collected and used as training data to train the model and obtain a connection deviation identification model. The model is then used to analyze the image sequence of connection nodes to identify risky connection nodes. All risky connection nodes are then integrated to obtain a risky connection node set.

[0056] Furthermore, the method further includes: planning flight paths based on the set of connected nodes to obtain image acquisition paths; and the method also includes:

[0057] The process involves: interactively obtaining the acquisition vector set of the connection node set; pre-setting an image acquisition distance and using this distance as a constraint to locate the image acquisition node set based on the connection node set and the acquisition vector set; interactively obtaining the image acquisition device parameters of the UAV and performing field-of-view overlap analysis on the image acquisition node set based on the image acquisition device parameters and the image acquisition distance to obtain M sets of overlapping acquisition nodes; performing node aggregation on the M sets of overlapping acquisition nodes and updating the image acquisition node set based on the aggregation results to obtain an updated acquisition node set; and performing shortest flight path fitting based on the updated acquisition node set to obtain the image acquisition path.

[0058] Accessing past transmission tower maintenance logs reveals information such as image acquisition angles and directions used in different maintenance operations, especially effective acquisition vectors used to determine the status of connection nodes. Experience vectors are extracted from the maintenance logs to obtain a set of acquisition vectors for the connection node set. These vectors describe which directions of image acquisition can effectively determine the status of connection nodes. These experience vectors are based on actual situations and experience summaries from historical maintenance.

[0059] Based on the performance of the UAV and its camera equipment, as well as the size and location of the target connection nodes, a preset image acquisition distance is established to ensure that clear node images can be acquired during actual operation. According to the preset image acquisition distance, an acquisition vector set is applied to determine the image acquisition nodes for each connection node, forming an image acquisition node set. This ensures that all connection nodes are fully covered under the preset acquisition distance and vector constraints.

[0060] By interacting with the UAV control system, the parameters of the image acquisition equipment currently mounted on the UAV are obtained, including camera resolution, field of view, focal length, aperture, sensor size, and shooting mode. Based on the obtained image acquisition equipment parameters and the preset image acquisition distance, the field of view of each image acquisition node is calculated. The field of view determines the area that the camera can cover at a specified position and distance. By simulating the field of view of each image acquisition node, the field of view of adjacent nodes is compared and analyzed to identify the overlapping area of ​​the field of view between two or more image acquisition nodes. If the field of view of two or more nodes overlaps, the overlapping nodes are grouped together to form overlapping acquisition nodes. All nodes are traversed to obtain M groups of overlapping acquisition nodes, where M is a positive integer.

[0061] Based on image acquisition requirements and actual operation, a distance threshold is set, defined as the maximum allowable distance between image acquisition nodes. If the distance between connected nodes is less than this threshold, they can be aggregated into one node. A clustering algorithm is used to aggregate M groups of overlapping acquisition nodes. Based on the distance threshold and relative position, close acquisition nodes are merged into a new aggregate node, which represents the position of these original nodes. Based on the aggregation result, the original M groups of overlapping acquisition nodes are replaced with the new aggregate node to obtain an updated acquisition node set.

[0062] Using shortest path algorithms, such as Dijkstra's algorithm, path planning is performed on the updated collection node set. Taking into account factors such as obstacles, flight restrictions, and efficiency in the actual flight environment, the goal is to ensure that the UAV covers all collection nodes in the shortest possible time, and the obtained shortest flight path is used as the image acquisition path.

[0063] By aggregating nodes, the number of required acquisition nodes is reduced, thereby improving the efficiency of image acquisition. By planning the shortest flight path, the flight time of the UAV can be significantly shortened, energy consumption can be reduced, and the efficiency of mission completion can be improved.

[0064] Furthermore, the method further includes performing visual recognition of connection deviations on the image sequence of the connected nodes to locate the set of risky connected nodes.

[0065] Multiple sample connection deviation image sets of various sample connection nodes are obtained interactively; a connection deviation recognition model is constructed using the multiple sample connection deviation image sets; the connection node image sequence is synchronized to the connection deviation recognition model to perform connection deviation visual recognition and locate the risk connection node set, wherein the N risk connection nodes in the risk connection node set have N node spatial location identifiers.

[0066] Based on actual operational needs, different categories of sample connection nodes are defined, including normal connection nodes and various deviation connection nodes, such as loose, misaligned, and corroded nodes. Through an image acquisition system, sample image data of these connection nodes in different states are acquired. The images should cover all possible connection deviations. Each sample image is labeled to mark the type and degree of deviation of the connection node. This labeling information is used to train the recognition model. After the labeling is completed, multiple sample connection deviation image sets are integrated to obtain the results.

[0067] Based on deep learning algorithms, such as convolutional neural networks, a connection deviation recognition model is constructed. The model architecture is designed, including an input layer, convolutional layers, pooling layers, and fully connected layers. Multiple sample connection deviation image sets are divided into training, validation, and test sets. The model is trained using the training set, and its parameters are adjusted to accurately identify the deviation types of connection nodes. The training process includes techniques such as hyperparameter tuning and regularization to optimize model performance. The model's performance is evaluated on the validation set, and metrics such as recognition accuracy are checked. If the model performs poorly, adjustments and optimizations are made. The final model's performance is tested on the test set to ensure it has good generalization ability and practical application effect. Based on the evaluation results, the model is further optimized to finally obtain a connection deviation recognition model that meets the requirements for real-time identification and classification of connection node deviations.

[0068] The image sequence of the connection nodes is input into the connection deviation recognition model. Based on the features learned during the training process, the model identifies various types of deviations in the connection nodes, such as loosening, misalignment, and erosion. Each connection node in the image is marked as normal or has a deviation. All connection nodes marked as having deviations are extracted from the recognition results and their specific spatial locations are marked to form a risk connection node set. The risk connection node set will include N risk connection nodes with spatial locations.

[0069] Furthermore, based on the aforementioned tower defect analysis model, the deformation risk of the target transmission tower is quantitatively assessed to obtain a deformation risk coefficient. The method further includes:

[0070] In the tower defect analysis model, the average point spacing of N risk connection nodes in the risk connection node set is calculated to obtain the connection risk dispersion coefficient; based on the internal defect information set, defect extreme value calls are performed to obtain the internal defect extreme value array; the internal defect extreme value array is normalized to obtain the internal defect risk coefficient; the multiple local deformation data in the tower deformation information set are normalized, and the extreme value calls are performed on the normalization results to obtain the local deformation risk coefficient; the regional center coordinates of the multiple deformation areas are extracted from the tower defect analysis model to obtain multiple deformation risk nodes; the average point spacing of the multiple deformation risk nodes is calculated in the tower defect analysis model to obtain the deformation risk dispersion coefficient; a tower deformation evaluation function is introduced, and a comprehensive evaluation is performed on the connection risk dispersion coefficient, internal defect risk coefficient, local deformation risk coefficient, and deformation risk dispersion coefficient based on the tower deformation evaluation function to obtain the deformation risk coefficient.

[0071] Obtain the spatial coordinates of N risk connection nodes in the risk connection node set to represent their positions in three-dimensional space. Pair the N risk connection nodes together. For each pair of risk connection nodes, calculate the Euclidean distance between them based on the spatial coordinates to obtain the point spacing of the node pair. Traverse all node pairs and construct an N×N distance matrix based on the point spacing of all node pairs. Calculate the average value of all off-diagonal elements in the distance matrix to obtain the average point spacing between all risk connection nodes.

[0072] The connection risk dispersion coefficient is calculated based on the average point spacing. Specifically, the standard deviation of the point spacing of all node pairs is calculated, and then the standard deviation is divided by the average point spacing to obtain the connection risk dispersion coefficient. This coefficient reflects the degree of dispersion of risk connection nodes in spatial distribution. A higher dispersion coefficient indicates that the distribution of risk nodes in space is more uneven, indicating that the risk is concentrated in some areas.

[0073] The extreme parameters of each defect, such as the maximum length, maximum width, and depth of internal cracks, are extracted from the internal defect information set to obtain an internal defect extreme value array. This array is then normalized, for example, using a max-min normalization method to map the data to the [0,1] interval. For each defect, the normalized extreme parameters are integrated, for example, using a weighted average method, to calculate the overall internal defect risk coefficient. A high risk coefficient indicates a more severe defect.

[0074] Multiple local deformation data are quantified to obtain deformation angle, displacement distance, etc., and the maximum and minimum values ​​of all local deformation data are obtained. The deformation data are mapped to the [0,1] interval using the max-min normalization method to obtain normalized local deformation data. The extreme value is called on the normalized local deformation data, that is, the maximum value of all deformation data is retrieved, such as the maximum deformation angle, the maximum displacement distance, etc. The weighted average of the maximum values ​​of all deformation data is calculated to obtain the local deformation risk coefficient. A high risk coefficient indicates that the deformation is more severe.

[0075] Information on multiple deformation regions, including their spatial location, shape, and size, is extracted from the tower defect analysis model. Each deformation region is analyzed, and its center coordinates are calculated. For example, the average coordinates of all points within each deformation region are calculated to obtain the geometric center coordinates, which are then used as the region center coordinates. All region center coordinates are integrated to form multiple deformation risk nodes for subsequent risk analysis.

[0076] The Euclidean distance is used to calculate the distance between every two deformation risk nodes, and the mean of all distances is calculated. Based on the mean distance, the standard deviation of the distance between every two deformation risk nodes is calculated. The standard deviation is then divided by the mean distance to obtain the deformation risk dispersion coefficient, which is used to quantitatively analyze the degree of tower deformation risk.

[0077] A tower deformation evaluation function is introduced. The connection risk dispersion coefficient, internal defect risk coefficient, local deformation risk coefficient, and deformation risk dispersion coefficient are input into the tower deformation evaluation function to calculate the deformation risk coefficient. The calculated deformation risk coefficient reflects the overall risk level after considering connection deviation, internal defects, local deformation, and deformation dispersion. The higher the value, the more serious the use risk caused by deformation.

[0078] Furthermore, the tower deformation evaluation function is as follows:

[0079]

[0080] Where F is the deformation risk coefficient, and D c To connect the risk dispersion coefficients, R i R is the internal defect risk coefficient. l D represents the risk coefficient for local deformation. d w1 is the coefficient of variation of deformation risk, w2 is the weight of connection risk, w3 is the weight of deformation risk, and w1+w2+w3+w4=1.

[0081] The tower deformation evaluation function is as follows:

[0082]

[0083] For the tower deformation evaluation function, the first part By applying the square root to the connectivity risk dispersion coefficient, its nonlinear impact on the deformation risk coefficient can be reduced, and the resulting value reflects the contribution of the connectivity risk dispersion coefficient to the overall risk; Part Two Adjusting the influence of the internal defect risk coefficient relative to the local deformation risk coefficient ensures a reasonable assessment of internal defect risk considering local deformation; Part Three... By calculating the relative impact of local deformation and considering the influence of the deformation risk dispersion coefficient, this combined method can comprehensively evaluate the risk of local deformation and adjust the weights according to the deformation dispersion. In summary, this tower deformation evaluation function integrates various risk factors through calculation to conduct a scientific and reasonable risk assessment, providing a basis for subsequent maintenance and repair decisions.

[0084] In summary, the vision-based method for evaluating the deformation of transmission towers provided in this application has the following technical advantages:

[0085] By interactively acquiring tower construction information and reconstructing transmission tower models, a digital model of the target transmission tower can be accurately constructed, providing a foundation for subsequent visual recognition and analysis. Based on the transmission tower model, structural deviation visual recognition of the target tower can comprehensively detect tower deformation, obtain detailed deformation information, accurately identify structural deviations, and promptly detect potential deformation problems. By locating connection nodes and performing connection deviation visual recognition, risky connection node sets can be accurately identified. This allows for prioritizing potentially risky connection points during maintenance and inspection, improving maintenance efficiency and accuracy. Pre-setting a flaw detection point array and performing ultrasonic flaw detection enables in-depth detection of internal defects in the transmission tower. This non-destructive testing method ensures accurate identification of internal defects without damaging the tower. Further damage was assessed, and detailed internal defect data was provided. Internal defect identification was performed based on the ultrasonic testing dataset, acquiring an internal defect information set, enabling a comprehensive understanding of internal structural defects and providing a reliable basis for subsequent risk assessment and maintenance decisions. The internal defect information set, risk connection node set, and tower deformation information set were fitted to the transmission tower model to obtain a tower defect analysis model. This comprehensive analysis model integrates information from different sources, providing a comprehensive assessment of tower defects and risks, ensuring the comprehensiveness and accuracy of the analysis results. Based on the tower defect analysis model, the deformation risk of the target transmission tower was quantitatively assessed, obtaining a deformation risk coefficient. This quantifies various test results and analysis data, comprehensively judging the overall risk level of the tower, providing a scientific basis for maintenance and management decisions, and reducing operational risks.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating the deformation of transmission towers based on visual recognition, characterized in that, The method includes: Interactively obtain the tower construction information of the target transmission tower, and reconstruct the transmission tower model based on the tower construction information; Based on the transmission tower model, the target transmission tower is subjected to structural deviation visual recognition to obtain a tower deformation information set; Based on the tower construction information, locate the connection nodes on the target transmission tower, perform visual identification of connection deviations on the obtained connection node set, and obtain the risk connection node set. A preset flaw detection point array is used, and ultrasonic flaw detection is performed on the target transmission tower based on the flaw detection point array to obtain an ultrasonic detection dataset; Based on the ultrasonic detection dataset, internal defects of the target transmission tower are identified to obtain an internal defect information set. The internal defect information set, risk connection node set, and tower deformation information set are fitted to the transmission tower model to obtain the tower defect analysis model. Based on the tower defect analysis model, the deformation risk of the target transmission tower is quantitatively assessed to obtain the deformation risk coefficient. Based on the transmission tower model, the method performs structural deviation visual recognition on the target transmission tower to obtain a tower deformation information set, and further includes: A preset array of shooting points is used, and images of the target transmission tower are acquired based on the array of shooting points to obtain an array of actual tower images; The array of shooting points is fitted to the transmission tower model, and image acquisition and replication are performed on the transmission tower model to obtain a virtual image array of the tower; Based on the array of shooting points, perform image deviation visual recognition on the mapping of the physical image array and the virtual image array of the tower to obtain the tower deformation information set; Based on the array of shooting points, image deviation visual recognition is performed on the mapping between the physical image array and the virtual image array of the tower to obtain the tower deformation information set. The method further includes: Color features are extracted based on the array of images of the tower entities to obtain the color features of the towers. The image contrast of the tower entity image array is optimized by using the tower color features as a constraint to obtain an optimized entity image array. The first pole physical image and the first pole virtual image, mapped to the first shooting point, are extracted from the optimized physical image array and the pole virtual image array, wherein the first shooting point is any shooting point in the shooting point array; Spatial alignment is performed between the first physical tower image and the first virtual tower image to obtain image registration results; Based on the image registration results, deformation region identification is performed to locate the first deformation region; Similarly, the optimized physical image array and the tower virtual image array are traversed to identify deformation regions, thereby obtaining multiple deformation regions; The tower deformation is quantified numerically based on the multiple deformation regions to obtain the tower deformation information set, wherein the tower deformation information set includes multiple local deformation data of the multiple deformation regions; The method further includes quantifying the tower deformation of the target transmission tower based on the multiple deformation regions to obtain the tower deformation information set. Guided by the multiple deformation regions, laser point cloud data is collected on the target transmission tower to obtain multiple local point cloud data. Multiple local entity models are obtained by geometric modeling based on the aforementioned multiple local point cloud data; Guided by the multiple deformation regions, the multiple local entity models are fitted to the transmission tower model to perform tower deformation numerical quantification, thereby obtaining multiple local deformation data, which constitute the tower deformation information set. Based on the tower defect analysis model, the deformation risk of the target transmission tower is quantitatively assessed to obtain the deformation risk coefficient. The method further includes: In the tower defect analysis model, the average point spacing of N risk connection nodes in the risk connection node set is calculated to obtain the connection risk dispersion coefficient; Based on the internal defect information set, the internal defect extreme value call is performed to obtain the internal defect extreme value array; The internal defect extreme value array is normalized to obtain the internal defect risk coefficient; The multiple local deformation data in the tower deformation information set are normalized, and the extreme value is called on the normalization result to obtain the local deformation risk coefficient; The coordinates of the regional centers of the multiple deformation regions are extracted from the tower defect analysis model to obtain multiple deformation risk nodes. In the tower defect analysis model, the average point spacing of the multiple deformation risk nodes is calculated to obtain the deformation risk dispersion coefficient; A tower deformation evaluation function is introduced, and a comprehensive evaluation is performed on the connection risk dispersion coefficient, internal defect risk coefficient, local deformation risk coefficient, and deformation risk dispersion coefficient based on the tower deformation evaluation function to obtain the deformation risk coefficient.

2. The method for evaluating the deformation of transmission towers based on visual recognition as described in claim 1, characterized in that, Based on the tower construction information, connection nodes are located on the target transmission tower. The method further includes performing visual identification of connection deviations on the obtained set of connection nodes to obtain a set of risky connection nodes. The connection nodes are located in the transmission tower model to obtain the set of connection nodes; Flight path planning is performed based on the set of connection nodes to obtain the image acquisition path; The drone is guided to acquire images of the target power transmission tower according to the image acquisition path, thereby obtaining a sequence of connection node images. The connection deviation visual recognition is performed on the connection node image sequence to locate the risky connection node set.

3. The method for evaluating the deformation of transmission towers based on visual recognition as described in claim 2, characterized in that, The method further includes: planning a flight path based on the set of connected nodes to obtain an image acquisition path; and the method also includes: The collection vector set of the connection node set is obtained interactively; A preset image acquisition distance is used as a constraint, and the image acquisition node set is located according to the connection node set and the acquisition vector set. The image acquisition device parameters of the UAV are obtained interactively, and the field of view overlap analysis of the image acquisition node set is performed based on the image acquisition device parameters and the image acquisition distance to obtain M sets of overlapping acquisition nodes; Perform node aggregation on the M groups of overlapping acquisition nodes, and update the image acquisition node set based on the aggregation result to obtain the updated acquisition node set; The image acquisition path is obtained by fitting the shortest flight path based on the updated set of acquisition nodes.

4. The method for evaluating the deformation of transmission towers based on visual recognition as described in claim 2, characterized in that, The method further includes performing visual recognition of connection deviations on the image sequence of the connected nodes to locate the set of risky connected nodes. Interactively obtain a set of image data of multiple sample connection deviations for various sample connection nodes; A connection deviation recognition model is constructed using the multiple sample connection deviation image sets. The image sequence of the connection nodes is synchronized to the connection deviation identification model to perform connection deviation visual recognition and locate the risk connection node set, wherein the N risk connection nodes in the risk connection node set have N node spatial location identifiers.

5. The method for evaluating the deformation of transmission towers based on visual recognition as described in claim 1, characterized in that, The tower deformation evaluation function is as follows: Where F is the deformation risk coefficient, and D c To connect the risk dispersion coefficients, R i R is the internal defect risk coefficient. l D represents the risk coefficient for local deformation. d w1 is the coefficient of variation for deformation risk, w2 is the weight of connection risk, w3 is the weight of correlation risk, and w1+w2+w3=1.

Images