Real-time detection method for tree barrier hidden danger of power transmission line based on conductor vectorization fitting

Abstract

The application discloses a power transmission line tree barrier hidden danger real-time detection method based on conductor vectorization fitting, and relates to the technical field of intelligent inspection of power transmission lines. The method comprises the following steps: acquiring three-dimensional point cloud data and multispectral remote sensing data of the power transmission line, fitting to generate a vectorized conductor digital model, and classifying to obtain power line point clouds and vegetation point clouds; calculating a vegetation greenness index and a canopy sparsity, fusing to generate a growth activity coefficient, and calculating a risk correction clearance distance; when the risk correction clearance distance is lower than a preset safety threshold, taking a conductor direction projection distance as a one-dimensional feature and a danger degree as a height value, using a watershed algorithm to cluster and merge hidden danger detection points; calculating a tree barrier risk density index according to the merged hidden danger point set, and outputting a graded early warning signal. The application solves the technical problems in the prior art that a static clearance distance cannot reflect the dynamic risk of vegetation growth, and that repeated statistics of hidden danger points leads to distorted risk assessment.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for power transmission lines, specifically to a real-time detection method for tree obstruction hazards on power transmission lines based on conductor vectorization fitting. Background Technology

[0002] Tree growth within transmission line corridors is a major contributing factor to line tripping accidents. When the distance between trees and conductors is less than the safe clearance distance, accidents such as conductor-to-tree discharge or tree-to-conductor short circuits caused by tree contact with conductors may occur, seriously threatening the safe and stable operation of the power grid. Traditional methods for detecting tree obstructions mainly rely on manual line inspections or helicopter aerial photography, assessing tree risk through visual observation or subsequent data analysis. With the widespread adoption of drone technology, drone inspections equipped with lidar are gradually becoming the mainstream method, capable of quickly acquiring three-dimensional point cloud data of the transmission line corridor, classifying power lines and vegetation through point cloud data, and calculating the clearance distance between conductors and vegetation.

[0003] However, existing lidar-based tree obstacle detection methods still have many shortcomings. First, static clearance distance cannot reflect the dynamic risks of vegetation growth. Trees grow at significantly different rates under different seasons and conditions. Trees that currently meet safety clearance requirements may encroach on safe areas due to rapid growth within a few months. Existing methods do not incorporate vegetation growth activity factors to correct for the risk in clearance distance calculations. Second, lidar point cloud classification is affected by point cloud density and noise, easily leading to misclassification. For example, tree canopy points near power lines may be misclassified as power line points, or power line points may be misclassified as vegetation points, affecting the accuracy of clearance distance calculations. Third, the handling of hazard detection points is relatively crude. The same tree may correspond to multiple detection points. Directly accumulating these points can lead to duplicate counting of hazards, distorting risk assessment results. Fourth, existing methods lack a quantitative risk density index, making it impossible to perform overall risk ranking and graded early warning for line sections, hindering the orderly formulation of maintenance plans.

[0004] Therefore, there is an urgent need for a real-time detection method for tree barrier hazards that can integrate multispectral data to assess vegetation growth activity, accurately classify point clouds, intelligently merge hazard points, and quantify risk levels. Summary of the Invention

[0005] This invention addresses the technical problems in existing technologies, such as static clearance distance failing to reflect the dynamic risks of vegetation growth, point cloud classification being susceptible to noise interference, repeated statistics of potential hazard points leading to distorted risk assessments, and the lack of a quantitative risk density index and a graded early warning mechanism. It provides a real-time detection method for tree obstruction hazards on transmission lines based on conductor vectorization fitting.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a real-time detection method for tree obstruction hazards on transmission lines based on conductor vectorization fitting, including: Acquire 3D point cloud data and multispectral remote sensing data of transmission lines, fit and generate vectorized conductor digital models, and classify and obtain power line point clouds and vegetation point clouds. The vegetation greenness index is extracted based on the multispectral remote sensing data, and the canopy sparsity is calculated based on the local point cloud density of the vegetation point cloud. The greenness index and canopy sparsity are then fused to generate a growth activity coefficient. The risk-corrected clearance distance is obtained by correcting the clearance distance between the guide wire and the vegetation based on the growth activity coefficient. When the risk correction clearance distance is lower than the preset safety threshold, the watershed algorithm is used to cluster and merge the hidden danger detection points, taking the projection distance of the conductor direction as a one-dimensional feature and the degree of danger of the clearance distance as the height value, and merging the detection points in the same catchment basin into an independent hidden danger point. The tree obstacle risk density index is calculated based on the merged set of potential hazards, and an early warning signal is output.

[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention first integrates lidar point cloud and multispectral remote sensing data, generating a vectorized conductor model through catenary fitting. It then combines semantic segmentation and spatial consistency verification to achieve accurate classification of power lines and vegetation, improving the data reliability of clearance distance calculation. Secondly, it introduces a growth activity coefficient constructed by weighted fusion of normalized vegetation index and canopy sparsity, dynamically correcting the geometric clearance distance and quantifying differences in tree growth rates as risk factors, thus overcoming the deficiency of static clearance distance in ignoring vegetation growth trends. Thirdly, addressing the issue of multiple hazard detection points for the same tree, a one-dimensional watershed clustering and merging algorithm is used. With the conductor direction projection distance as the horizontal axis and the hazard level as the height, detection points within the same catchment basin are intelligently merged, achieving unique statistical analysis of hazard points. Finally, a tree obstacle risk density index is constructed, comprehensively considering the number of hazard points, clearance distance ratio, hazard scale, and historical risk level to generate a graded early warning system based on statistical thresholds. This system outputs four levels of signals: normal, attention, early warning, and alarm, along with corresponding work order suggestions, forming an intelligent operation and maintenance process from data collection to closed-loop handling. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the real-time detection method for tree obstruction hazards in transmission lines based on conductor vectorization fitting provided by the present invention. Figure 2 A schematic diagram illustrating the logic of the real-time detection method for tree obstruction hazards in transmission lines based on conductor vectorization fitting provided by the present invention; Figure 3This is a schematic diagram illustrating the process of clustering and merging hazard detection points using the watershed algorithm provided by the present invention. Detailed Implementation

[0009] Examples, such as Figure 1 , Figure 2 As shown, this embodiment of the invention provides a real-time detection method for tree obstruction hazards on transmission lines based on conductor vectorization fitting, including: S10: Acquire 3D point cloud data and multispectral remote sensing data of transmission lines, fit and generate vectorized conductor digital models, and classify and obtain power line point clouds and vegetation point clouds. Transmission lines are typically erected in complex outdoor environments, with extensive tree and vegetation along their corridors. The core task of detecting tree obstruction hazards is to accurately measure the spatial distance between the conductor and the vegetation. To accomplish this measurement, two key data points are required: first, three-dimensional point cloud data reflecting the spatial location of the conductor; and second, multispectral remote sensing data reflecting the vegetation's growth status.

[0010] The drone, equipped with a lidar scanner and camera, flies along the direction of the power line. The lidar emits laser pulses towards the ground and receives the echoes, recording the three-dimensional spatial coordinates of each reflection point to form a three-dimensional point cloud. This point cloud data contains spatial location information of various ground features such as power lines, towers, vegetation, and the ground. Simultaneously, the onboard camera collects multispectral remote sensing data, recording the reflectance of vegetation in multiple spectral bands to assess its growth activity and health status.

[0011] Next, after obtaining the 3D point cloud data, it is necessary to extract the spatial location of the conductors. Since the conductors of transmission lines form a catenary shape between two towers, rather than a straight line, simply connecting the two suspension points cannot accurately describe the actual spatial curve of the conductor. Therefore, it is necessary to fit and generate a vectorized conductor digital model, that is, to accurately describe the direction and shape of the conductor in 3D space in analytical form. This vectorized conductor digital model can be used to calculate the spatial coordinates of any point on the conductor, and can also serve as a spatial reference for verifying the power line point cloud in subsequent point cloud classification. Then, point cloud classification is performed, the purpose of which is to distinguish each point in the 3D point cloud data into power line points or vegetation points. In the raw point cloud data collected by the UAV, power line points, vegetation points, tower points, and ground points are mixed together. Through classification operations, power line points and vegetation points are extracted separately, providing corresponding point sets for subsequent clearance distance calculations.

[0012] Specifically, the two data acquisition steps mentioned above are used to provide a data foundation for subsequent clearance distance calculation and risk correction. The vectorized conductor digital model is used to provide the precise spatial location of the conductor, and the classified power line point cloud and vegetation point cloud are used to provide the objects for spatial distance calculation.

[0013] Specifically, the execution steps of this embodiment, including fitting and generating a vectorized conductor digital model, include: Obtain the tower coordinates of the transmission line and determine the suspension points of the conductor on the two towers within each span; Using the span as a unit, the catenary equation is fitted based on the hanging point position to generate the three-dimensional spatial curve equation of the conductor within each span. The curve equations of adjacent spans are spliced ​​together according to the tower coordinates to form a vectorized digital model of the entire continuous line.

[0014] First, obtain the coordinates of the transmission line towers to determine the suspension points of the conductors on the two towers within each span. A transmission line consists of multiple towers and conductors erected between them. The line between two adjacent towers is called a span. Each tower is equipped with an insulator string to suspend the conductor; the connection point between the conductor and the insulator string is called the suspension point. Through design drawings or on-site measurements, the three-dimensional spatial coordinates of each tower and the offset of the conductor's suspension point relative to the tower's base point can be obtained. Superimposing the tower coordinates with the suspension point offset yields the three-dimensional spatial coordinates of the conductor's suspension points on the two towers within each span.

[0015] Secondly, taking each span as a unit, the catenary equation is fitted based on the location of the suspension points to generate the three-dimensional spatial curve equation of the conductor within each span. Under the combined action of its own gravity and tension, the overhead conductor naturally suspends itself in a catenary shape, and its spatial curve conforms to the catenary equation. The basic form of the catenary equation is that in the vertical plane, y equals a multiplied by the hyperbolic cosine function x divided by a, where a is a parameter related to the conductor tension and weight per unit length; a larger value indicates a smoother conductor, and a smaller value indicates a larger sag. Given the three-dimensional spatial coordinates of two suspension points, it is necessary to determine the parameter 'a' in the catenary equation and the translation of the curve in the coordinate system to ensure that the catenary curve passes precisely through the two suspension points. The solution method involves substituting the coordinates of the two suspension points into the catenary equation to establish a system of equations containing the unknown parameter 'a', and then solving it using numerical iteration methods, such as Newton's iteration method or the bisection method, to search for parameter values ​​that make the equations true.

[0016] Simultaneously, the direction of the conductor on the horizontal projection plane must also be considered. The projections of the two suspension points on the horizontal plane form a straight line segment, the direction of which represents the direction of the conductor in the horizontal direction. The catenary curve fitted in the vertical plane is spatially rotated and stretched along the direction of the horizontal projection straight line segment, so that the catenary curve is mapped from a two-dimensional plane curve to a three-dimensional space. In an optional embodiment, the specific operation is as follows: using the horizontal distance between the two suspension points as the horizontal axis range, calculate the horizontal offset ratio of each point on the catenary relative to the left suspension point, then determine the projection position of the point on the horizontal plane according to the same ratio on the horizontal projection line, and finally combine the height value of the point to obtain the three-dimensional space coordinates.

[0017] Through the above synthesis, the three-dimensional spatial curve equation of the conductor within this span is obtained. This three-dimensional spatial curve equation can be expressed in parametric form, that is, given any arc length parameter or horizontal distance parameter, the three-dimensional coordinates of the corresponding point on the conductor can be calculated. This three-dimensional spatial curve equation can describe the precise position of any point on the conductor in three-dimensional space, providing a continuous geometric reference for subsequent clearance distance calculations.

[0018] Furthermore, the three-dimensional spatial curve equations of adjacent spans are spliced ​​together according to the tower coordinates to form a continuous vectorized digital model of the entire conductor line. The curve equation of each span describes the spatial shape of the conductor within its respective span. Since adjacent spans share the same tower coordinates, the curve equations of two adjacent spans naturally connect at the suspension point at the tower location. By sequentially combining the curve equations of all spans, the complete three-dimensional spatial curve equation of the conductor line from the start to the end of the line can be obtained. This vectorized digital model of the conductor line is stored in analytical form, allowing the coordinates of the conductor to be queried at any spatial location, providing a precise geometric reference for subsequent clearance distance calculations.

[0019] Through the above vectorized fitting, the traverse no longer exists in the form of discrete point clouds, but is described in the form of continuous and differentiable curve equations, forming a vectorized traverse digital model. Specifically, the advantages of this representation are: on the one hand, the shortest spatial distance from any point on the traverse to the vegetation point can be accurately calculated; on the other hand, in subsequent point cloud classification, the spatial position of the vectorized model can be used to verify the consistency of the point cloud classification results and correct misclassifications caused by point cloud density or noise.

[0020] Furthermore, point clouds of power lines and vegetation are obtained through classification, including: The three-dimensional point cloud data is input into a pre-trained point cloud semantic segmentation network, and the initial classification label of each point is output. The initial classification label includes power line label and vegetation label. Based on the spatial location of the vectorized conductor digital model, the initial classification labels are spatially consistent. Power line labels that deviate from the preset distance range of the vectorized conductor digital model are corrected to vegetation labels, and continuous vegetation labels within the preset distance range of the vectorized conductor digital model are corrected to power line labels. Based on the revised classification labels, the point cloud data is divided into power line point clouds and vegetation point clouds. The training steps of the point cloud semantic segmentation network include: Historical 3D point cloud data of the transmission line corridor was collected, and each point was labeled with its corresponding real classification label to form a training sample set; A point cloud semantic segmentation network is constructed based on deep learning. The training sample set is used as input and the real classification labels are used as the supervision target for supervised training until convergence, thus obtaining the trained point cloud semantic segmentation network.

[0021] First, the 3D point cloud data is input into a pre-trained point cloud semantic segmentation network, which outputs an initial classification label for each point. These initial labels include power line labels and vegetation labels. The point cloud semantic segmentation network is a deep learning-based neural network model capable of classifying each point in the 3D point cloud. It extracts local geometric features of each point and its neighborhood using a multilayer perceptron or convolutional operations to determine whether each point belongs to power lines or vegetation. Specifically, the input to this point cloud semantic segmentation network is a point cloud dataset containing N points, each with 3D spatial coordinates. The output is a classification label sequence of length N, with each location corresponding to a classification result for that point.

[0022] The training steps for the point cloud semantic segmentation network include: collecting historical 3D point cloud data of the power transmission line corridor and labeling each point with a corresponding ground truth classification label to form a training sample set. Ground truth classification labels are typically obtained manually, with labelers determining whether each point belongs to a power line, vegetation, tower, or ground based on the geometric shape and spatial location of the point cloud. The training sample set needs to cover multiple line types, various terrains, and different seasonal conditions to enhance the network's generalization ability.

[0023] Next, a point cloud semantic segmentation network is constructed based on deep learning. Supervised training is performed using the training sample set as input and the real classification labels as the supervised target until convergence, resulting in a trained point cloud semantic segmentation network. Optionally, during training, the cross-entropy loss function is used to measure the difference between the network's predicted labels and the real labels, and the network parameters are updated using the backpropagation algorithm. Training stops when the loss value on the validation set no longer decreases. The trained network can automatically output the initial classification label for each point on new point cloud data.

[0024] For example, in an optional embodiment, the point cloud semantic segmentation network may employ the RandLA-Net architecture. This network is designed to efficiently process large-scale point cloud data, progressively expanding the receptive field of each point while maintaining high sampling efficiency through random sampling and local feature aggregation modules.

[0025] The specific architecture parameters of the network are as follows: The input layer receives point cloud data containing three-dimensional spatial coordinates. The original feature dimension of each point is 3, i.e., x-coordinate, y-coordinate, and z-coordinate. After the first local feature aggregation module, the output feature dimension is 8, and the number of points is retained to one-quarter of the original number through random sampling. After the second local feature aggregation module, the output feature dimension is 16, and the number of points is sampled again to one-quarter of the previous stage. After the third local feature aggregation module, the output feature dimension is 32, and the number of points is sampled again to one-quarter of the previous stage. After the fourth local feature aggregation module, the output feature dimension is 64, and the number of points is sampled to one-quarter of the previous stage. After four encoding modules, the downsampled features are gradually restored to the original number of points through nearest neighbor interpolation. At the same time, the features of the corresponding layers in the encoding stage are spliced ​​by skip connections. Finally, the features of each point are mapped to a 2D output vector through a fully connected layer, corresponding to the predicted probabilities of power line labels and vegetation labels, respectively. After normalization by the Softmax activation function, the class with the larger probability value is taken as the initial classification label of the point.

[0026] During training, the cross-entropy loss function is used to calculate the difference between the network's predicted labels and the ground truth labels. The optimizer is Adam, with an initial learning rate of 0.001 and weight decay of 0.0001. The batch size is set based on the amount of point cloud data, with each batch containing two point cloud blocks, and each block containing approximately 20,000 points. The total number of training epochs is set to 100. Training is terminated when the loss function value on the validation set no longer decreases for 10 consecutive epochs, resulting in the trained point cloud semantic segmentation network.

[0027] Furthermore, based on the spatial location of the vectorized traverse digital model, a spatial consistency check is performed on the obtained initial classification labels. Power line labels deviating from the preset distance range of the vectorized traverse digital model are corrected to vegetation labels, while continuous vegetation labels within the preset distance range of the vectorized traverse digital model are corrected to power line labels. Specifically, purely data-driven point cloud semantic segmentation networks are susceptible to misclassification due to factors such as uneven point cloud density, local noise, or occlusion. For example, tree canopy points near the traverse may be misclassified as power line points, while power line points in sparse areas of the point cloud may be missed as vegetation points. The spatial consistency check uses the vectorized traverse digital model as a geometric benchmark to correct misclassification.

[0028] The preset distance range refers to a buffer zone around the conductor, which can be set to 3 to 5 times the conductor radius, or 0.5 meters depending on the actual conductor diameter and point cloud accuracy. For points classified as power lines, the shortest distance from that point to the vectorized conductor digital model is calculated. If this distance is greater than the preset distance range, it indicates that the point is far from the conductor and should not be classified as a power line; its label is then corrected to vegetation. For points classified as vegetation, if they are spatially continuous in a linear distribution and located within the preset distance range of the vectorized conductor digital model, they may be missed power line points; their labels are then corrected to power line. Through spatial consistency verification, the geometric constraints of the vectorized conductor digital model can be used to effectively reduce the false detection rate of purely data-driven classification.

[0029] Finally, based on the revised classification labels, the point cloud data is divided into power line point clouds and vegetation point clouds. Specifically, all points labeled as power lines are extracted to form a power line point cloud; all points labeled as vegetation are extracted to form a vegetation point cloud. The power line point cloud is used as a discrete representation of conductors in subsequent clearance distance calculations, while the vegetation point cloud is used as a discrete representation of obstacles in subsequent clearance distance calculations. Through the above classification steps, the original mixed point cloud data is divided into power line point clouds and vegetation point clouds, providing a semantically clear data foundation for subsequent clearance distance calculations and hazard detection.

[0030] S20: Extract the greenness index of vegetation based on the multispectral remote sensing data, calculate the canopy sparsity based on the local point cloud density of the vegetation point cloud, and fuse the greenness index and canopy sparsity to generate a growth activity coefficient. Furthermore, after completing point cloud classification and obtaining power line and vegetation point clouds, it is necessary to assess vegetation growth activity in order to adjust the clearance distance for risk. Different vegetation types exhibit significant differences in growth rate; for example, bamboo can grow tens of centimeters per day during its peak growing season, while conifers grow relatively slowly. Judging potential hazards solely based on the current static clearance distance may lead to a severe underestimation during periods of rapid vegetation growth—meaning the current clearance distance may appear safe, but vegetation could encroach on the safe zone months later. Therefore, it is essential to extract information reflecting the vegetation growth status from multispectral remote sensing data.

[0031] Multispectral remote sensing data records the reflectance of vegetation in different spectral bands. Healthy and active vegetation has higher reflectance in the near-infrared band and lower reflectance in the red band. This spectral characteristic can be used to calculate the Normalized Difference Vegetation Index (NDV). The NDC is calculated as: (Near-infrared reflectance minus red reflectance) divided by (Near-infrared reflectance plus red reflectance). The NDC ranges from -1 to 1; a higher value indicates more vigorous photosynthesis and stronger growth activity. The NDC can be used as a greenness index to quantify the level of vegetation growth activity.

[0032] Meanwhile, the local point cloud density of vegetation point clouds reflects the canopy closure and leaf area density. The denser the canopy, the more point clouds per unit volume, making it difficult for lidar pulses to penetrate to the lower layers after being blocked by the top of the canopy, resulting in sparse point clouds in the lower layers. For vegetation point clouds classified by point cloud semantic segmentation networks, voxelization is performed according to a preset voxel size, dividing the space into cubic grids with the same side length. The number of point clouds contained in each voxel is counted, and the local point cloud density of that voxel is calculated, i.e., the number of point clouds in the voxel divided by the voxel volume. The ratio of the local point cloud density of each voxel to the theoretical maximum density is used as the filling rate of that voxel, and the sparsity of that voxel is obtained by subtracting the filling rate from one. The sparsity of all voxels containing vegetation points is arithmetically averaged to obtain the canopy sparsity. The larger the canopy sparsity, the sparser the vegetation canopy, which may indicate areas that have not yet been fully covered by new shoots, or young vegetation in the early stages of growth, which usually has higher growth activity.

[0033] Finally, after obtaining the greenness index and canopy sparsity, the two are weighted and summed to obtain the growth activity coefficient. The greenness index reflects the photosynthetic intensity of the vegetation, while canopy sparsity reflects the structural development stage of the vegetation. The weighted fusion of the two can more comprehensively assess the growth activity of the vegetation. The weighting coefficients can be adjusted according to the growth characteristics of different vegetation types; for example, the weight of the greenness index can be appropriately increased for broad-leaved trees, and the weight of canopy sparsity can be appropriately increased for coniferous trees.

[0034] The resulting growth activity coefficient is a dimensionless positive number. A higher value indicates more active vegetation growth and a higher risk of encroaching on the safety zone of the guide wire in the near future. This growth activity coefficient allows the dynamic growth characteristics of vegetation to be quantified and incorporated into the risk assessment of clearance distance.

[0035] Specifically, the execution steps of this embodiment include: extracting the greenness index of vegetation based on the multispectral remote sensing data, calculating the canopy sparsity based on the local point cloud density of the vegetation point cloud, and fusing the greenness index and canopy sparsity to generate a growth activity coefficient, including: The reflectance of vegetation in the near-infrared and red bands is extracted from the multispectral remote sensing data, and the normalized vegetation index is calculated as the greenness index. The vegetation point cloud is voxelized according to the preset voxel size. The number of points in each voxel is counted. The ratio of the local point cloud density of each voxel to the theoretical maximum density is calculated. The ratio is subtracted from the result to obtain the sparsity. The arithmetic mean of the sparsity of all voxels covered by the vegetation point cloud is obtained as the canopy sparsity. The growth activity coefficient is obtained by weighting and summing the greenness index and the canopy sparsity.

[0036] In one optional embodiment, voxelization is the process of dividing a three-dimensional space into a regular cubic mesh, with each small cube called a voxel. The preset voxel size can be set according to the point cloud density of the lidar and the canopy structure characteristics, for example, it can be set to 1 meter. A voxel size that is too large will cause sparse areas within the canopy to be smoothly masked, while a size that is too small will result in excessive statistical noise due to the insufficient number of point clouds within a single voxel. After voxel division, the number of vegetation point clouds contained within each voxel is counted. For each voxel, its local point cloud density is calculated, which is the number of point clouds within the voxel divided by the voxel volume. The voxel volume is equal to the cube of the preset voxel size; for example, the volume of a 1-meter voxel is 1 cubic meter. The theoretical maximum density refers to the maximum number of point clouds that can be collected under ideal conditions when the lidar pulse completely covers the voxel space. This value is determined by the number of scan lines, angular resolution, flight altitude, and point cloud emission frequency of the lidar, and can be obtained by measurement in a standard calibration scenario or by calculation based on device parameters.

[0037] Dividing the local point cloud density by the theoretical maximum density yields the fill rate of that voxel. The closer this value is to 1, the denser the point cloud within the voxel and the denser the vegetation canopy. Subtracting the fill rate from 1 gives the sparsity. The closer this value is to 1, the sparser the point cloud within the voxel and the sparser the canopy, potentially indicating areas not yet fully covered by new shoots. The canopy sparsity is obtained by arithmetically averaging the sparsities of all voxels containing vegetation point clouds. This value represents the overall sparseness of the vegetation canopy throughout the entire corridor.

[0038] The growth activity coefficient is equal to the first weighting coefficient multiplied by the greenness index, plus the second weighting coefficient multiplied by the canopy sparseness. The sum of the first and second weighting coefficients is 1. The first weighting coefficient can be adjusted according to the vegetation type's dependence on photosynthetic activity, while the second weighting coefficient is adjusted based on the canopy structure's ability to characterize growth stages. For example, for broadleaf trees, due to their dense foliage and significant seasonal variations, the greenness index has a greater weight than canopy sparseness; for instance, the greenness index weight is set to 0.6, and the canopy sparseness weight is set to 0.4. For coniferous trees, due to their evergreen nature and relatively sparse canopy, the canopy sparseness weight is greater than the greenness index; for instance, the greenness index weight is set to 0.3, and the canopy sparseness weight is set to 0.7. By integrating the greenness index and canopy sparseness, the growth activity coefficient can comprehensively reflect the vegetation's photosynthetic activity and structural development stage, providing a quantitative basis for subsequent risk correction of net clearance.

[0039] S30: Based on the growth activity coefficient, correct the clearance distance between the guide wire and the vegetation to obtain the risk-corrected clearance distance; Furthermore, after obtaining the growth activity coefficient, it needs to be applied to the risk correction of the clearance distance between the conductor and vegetation. The clearance distance between the conductor and vegetation refers to the shortest spatial distance from a point on the conductor to the surrounding vegetation surface. This clearance distance directly reflects the proximity between the conductor and the tree. The smaller the clearance distance, the higher the risk of conductor-to-canopy discharge or canopy-to-conduct conductor contact. However, there are limitations to judging potential hazards solely based on the current geometric clearance distance. For example, a tree in its rapid growth phase may currently have a clearance distance of 2.5 meters, which may still be within the safe range. However, based on its growth rate, the clearance distance will shrink to 1.5 meters in a month, below the safe threshold. Without risk correction, the tree will not be marked as a potential hazard, causing maintenance personnel to miss the optimal pruning opportunity. Therefore, the growth activity coefficient needs to be used as a risk factor to downward correct the geometric clearance distance.

[0040] Specifically, the risk-corrected clearance distance is obtained by correcting the clearance distance between the guide wire and the vegetation based on the growth activity coefficient, including: Extract the set of traverse reference points from the vectorized traverse digital model, and calculate the shortest Euclidean distance from the vegetation point cloud to the set of traverse reference points as the geometric clearance distance. Dividing the geometric clearance distance by the growth activity coefficient yields the risk-corrected clearance distance.

[0041] First, a set of reference points for the traverse is extracted from the vectorized traverse digital model. The shortest Euclidean distance from the vegetation point cloud to the traverse reference point set is calculated as the geometric clearance distance. The vectorized traverse digital model accurately describes the direction and shape of the traverse in space in the form of a three-dimensional spatial curve equation. To facilitate the calculation of the distance from points to the traverse, the continuous curve needs to be discretized into a series of reference points. The sampling density of the reference points determines the accuracy of the clearance distance calculation. The smaller the sampling interval, the denser the reference points, and the closer the calculated shortest distance is to the true shortest distance, but the computational workload also increases accordingly. For example, the traverse reference point set can be generated by sampling at equal intervals on the traverse at a density of one reference point per meter.

[0042] Next, for each point in the vegetation point cloud, the Euclidean distance from that point to each reference point in the traverse reference point set is calculated. The Euclidean distance is calculated as the square root of the sum of the squares of the differences between the two points along the x-axis, y-axis, and z-axis. The minimum value among all calculated results is taken as the geometric clearance distance from the vegetation point to the traverse. This geometric clearance distance reflects the actual spatial separation between the vegetation and the traverse at the current moment. When this clearance distance is small, it indicates that the vegetation is very close to the traverse, posing a risk of immediate discharge or contact.

[0043] Furthermore, the risk-corrected net distance is obtained by dividing the geometric clearance distance by the growth activity coefficient. The growth activity coefficient is a dimensionless number between 0 and 1; a higher value indicates more active vegetation growth. When vegetation growth activity is high, the growth activity coefficient is close to 1. Dividing the geometric clearance distance by a number close to 1 results in a risk-corrected net distance that is not significantly different from the geometric clearance distance, but is usually smaller than the original value. When vegetation growth activity is low, the growth activity coefficient is small. Dividing the geometric clearance distance by a smaller number results in a risk-corrected net distance that is significantly larger than the geometric clearance distance. Specifically, the physical meaning of this calculation method is that vegetation with high growth activity will be evaluated using stricter standards; that is, under the same geometric clearance distance, vegetation with higher growth activity is judged to have a higher risk.

[0044] For example, for a plant with extremely high growth activity, even if the current geometric clearance is large, the risk-corrected clearance, after dividing by the growth activity coefficient, may still be below the safety threshold, thus prompting the system to issue an early warning. For a plant with extremely low growth activity, even if the current geometric clearance is close to the safety boundary, the risk-corrected clearance, after dividing by a smaller growth activity coefficient, may still be above the safety threshold, indicating that although the plant is nearby, it does not pose a growth threat and treatment can be appropriately delayed.

[0045] For example, assuming a geometric clearance of 2.4 meters and a growth activity coefficient of 0.8, the risk-corrected clearance is 2.4 divided by 0.8, which equals 3.0 meters. If the preset safety threshold is 2.5 meters, the risk-corrected clearance of 3.0 meters is greater than 2.5 meters, and no hazard alarm is triggered. If the geometric clearance remains unchanged, but the growth activity coefficient is 0.9, the risk-corrected clearance is 2.4 divided by 0.9, which is approximately 2.67 meters, still greater than 2.5 meters. If the growth activity coefficient is 0.95, the risk-corrected clearance is approximately 2.53 meters, slightly higher than the threshold. If the growth activity coefficient is 1.0, the risk-corrected clearance is 2.4 meters, less than 2.5 meters, triggering an alarm. This example illustrates that vegetation with higher growth activity requires a smaller geometric clearance to avoid triggering an alarm, meaning the system has stricter requirements for it.

[0046] It is important to note that the preset safety threshold represents the critical value of the minimum allowable clearance distance between the conductor and vegetation. This preset safety threshold is comprehensively set based on the voltage level of the transmission line, the regulations on the safety distance between the conductor and the ground, and relevant operation and maintenance standards. For example, it can be set to 2.5 meters for 110 kV lines, 3.0 meters for 220 kV lines, and 4.0 meters for 500 kV lines. By comparing the risk-corrected clearance distance with this preset safety threshold, it can be determined whether there is a potential safety risk from the current vegetation. When the risk-corrected clearance distance is lower than the threshold, a hazard alarm is triggered, thus issuing a warning before the vegetation actually touches the safety boundary of the conductor, allowing maintenance personnel sufficient time for pruning and handling.

[0047] In summary, through the above modifications, the assessment of clearance distance has been upgraded from static geometric measurement to dynamic risk prediction, enabling relevant personnel to identify potential hazards caused by rapid vegetation growth in advance and allowing sufficient time for handling in the operation and maintenance plan. The starting point of this step is to introduce the growth dynamics of vegetation into the detection of tree barrier hazards, making up for the problem of the lag in the assessment of static clearance distance during the rapid vegetation growth season.

[0048] S40: When the risk correction clearance distance is lower than the preset safety threshold, the watershed algorithm is used to cluster and merge the hidden danger detection points, taking the projection distance of the conductor direction as a one-dimensional feature and the degree of danger of the clearance distance as the height value, and merging the detection points in the same catchment basin into an independent hidden danger point. Furthermore, when the risk-corrected clearance distance is lower than a preset safety threshold, the vegetation point is marked as a hazard detection point. However, a tree typically corresponds to multiple vegetation points; for example, the canopy may generate dozens or even hundreds of point clouds. If each detection point is considered an independent hazard, the number of hazards will be severely overestimated, and maintenance personnel will be unable to accurately assess the actual number of trees that need pruning. Therefore, it is necessary to cluster and merge spatially adjacent hazard detection points belonging to the same tree, so that each tree corresponds to an independent hazard point.

[0049] Because the conductor extends along the route, the distribution of tree-related hazards naturally exhibits a one-dimensional characteristic, meaning that hazard points are mainly distributed at different locations along the route on both sides of the conductor. To simplify the clustering problem, this step projects the hazard detection points in three-dimensional space onto the coordinate axis of the conductor direction, reducing the three-dimensional clustering problem to a one-dimensional clustering problem. Specifically, for each hazard detection point, its projected distance along the conductor direction is calculated, and the three-dimensional coordinates of the hazard detection point are projected onto this direction vector to obtain a one-dimensional coordinate value. This projected distance reflects the position of the hazard detection point along the route direction. A one-dimensional terrain profile is constructed with this projected distance as the horizontal axis and the hazard level of each hazard detection point as the vertical axis. The calculation of the hazard level comprehensively considers the clearance distance and the growth activity coefficient; the smaller the clearance distance and the higher the growth activity, the greater the hazard level.

[0050] Furthermore, the watershed algorithm is used to cluster and segment the one-dimensional terrain profile. The watershed algorithm is a morphological image segmentation method whose basic principle is to treat changes in terrain elevation as watersheds. In this algorithm, the water level gradually rises from a local lowest point. When water overflows from one catchment basin to an adjacent catchment basin, a watershed is set at the boundary. All hazard detection points are traversed, and detection points belonging to the same catchment basin are merged into a single independent hazard point. Multiple detection points generated from the same tree canopy are continuous in the projection along the traverse direction and have similar levels of danger; they will fall into the same catchment basin during the water level spread. Detection points between adjacent trees have gaps in their projections or significant differences in their levels of danger; a watershed will form at the boundary during the water level spread, thus separating the detection points of the two trees.

[0051] By using watershed clustering, dozens of monitoring points generated from a single tree can be merged into a single independent hazard point. Each independent hazard point represents a tree that may pose a tree obstacle risk. The merged set of hazard points accurately reflects the actual number of trees with potential hazards and eliminates the interference of duplicate monitoring on subsequent risk statistics, making the risk assessment results more consistent with the actual situation on site.

[0052] Specifically, such as Figure 3As shown, the specific execution steps of this embodiment include: when the risk-corrected clearance distance is lower than a preset safety threshold, using the projected distance along the conductor direction as a one-dimensional feature and the degree of danger of the clearance distance as a height value, the watershed algorithm is used to cluster and merge the hazard detection points, including: Each hazard detection point is projected onto a one-dimensional axis along the direction of the guide wire, and a one-dimensional terrain profile is constructed based on the hazard level of the clearance distance and the height value of each hazard detection point. The detection point with the highest danger level value is selected from the one-dimensional terrain profile as the starting point to simulate the water level rise process; When the water level spreads from the current monitoring point to an adjacent monitoring point, calculate the height difference between the two monitoring points; If the height difference is less than or equal to the height difference threshold, the adjacent detection point will be assigned to the current catchment basin. If the height difference is greater than the height difference threshold, a watershed is set at the adjacent detection point, and a new catchment basin is started from the adjacent detection point. The simulation of the water level rise process is repeated until all monitoring points have been traversed, and all monitoring points in each catchment basin are grouped into an independent potential hazard point.

[0053] First, each hazard detection point is mapped onto a one-dimensional axis along the guide wire direction, using the hazard level of the clearance distance as the height value of each hazard detection point, thus constructing a one-dimensional terrain profile. Hazard detection points are vegetation point clouds identified as close to the guide wire after risk-corrected clearance distance screening. Each hazard detection point has precise coordinates in three-dimensional space. To simplify the three-dimensional clustering problem into a more manageable one-dimensional form, the projected distance of each hazard detection point along the guide wire direction is calculated. Specifically, the coordinates of the tower anchor points at both ends of the current span are obtained. The unit vector pointing from the left anchor point to the right anchor point is used as the guide wire direction vector. Then, the projection length of the difference between the hazard detection point coordinates and the left anchor point coordinates onto this direction vector is calculated, yielding the one-dimensional projected distance. All hazard detection points are sorted in ascending order of projected distance, with each point corresponding to a projected distance value and a hazard level value. A series of discrete points are plotted on a two-dimensional plane with the projected distance as the x-axis and the hazard level as the y-axis. Connecting all points in order of projected distance forms a broken line, i.e., the one-dimensional terrain profile. This one-dimensional topographic profile reflects the distribution of vegetation hazard to the conductor at different locations along the conductor direction. Areas with higher hazard levels correspond to higher peak values ​​on the topographic profile.

[0054] The calculation of the hazard level takes into account both the clearance distance and the growth activity coefficient at the location. The smaller the clearance distance, the higher the hazard level; the larger the growth activity coefficient, the higher the hazard level.

[0055] Specifically, the steps for calculating the degree of danger include: Obtain the preset safety threshold, and subtract the clearance distance from the preset safety threshold to obtain the basic hazard value; Obtain the growth activity coefficient at the location of the hazard detection point; The degree of danger is obtained by multiplying the base hazard value by the growth activity coefficient.

[0056] The first step is to obtain a preset safety threshold. Subtracting the clearance distance from this preset safety threshold yields the basic hazard value. The preset safety threshold is a safety benchmark value used to determine the proximity between the conductor and vegetation. It is determined based on the voltage level and maintenance standards of the transmission line. For example, for a 110 kV line, the preset safety threshold can be set to 2.5 meters. The clearance distance refers to the geometric clearance distance from the vegetation point to the conductor. Subtracting the clearance distance from the preset safety threshold calculates the basic hazard value. When the clearance distance is less than the preset safety threshold, the basic hazard value is positive, indicating that vegetation has encroached on the safe area; the larger the positive value, the higher the degree of danger. When the clearance distance equals the preset safety threshold, the basic hazard value is zero. When the clearance distance is greater than the preset safety threshold, the basic hazard value is negative, indicating that vegetation has not yet encroached on the safe area.

[0057] For example, if the preset safety threshold is 2.5 meters and the clearance distance is 2.0 meters, then the basic hazard value is 0.5 meters. If the clearance distance is 1.5 meters, then the basic hazard value is 1.0 meter.

[0058] The second step is to obtain the growth activity coefficient at the location of the hazard detection point. The growth activity coefficient is a fusion index extracted from multispectral remote sensing data and vegetation point cloud density, reflecting the degree of vegetation growth activity at that location. The growth activity coefficient ranges from 0 to 1, with higher values ​​indicating more active vegetation growth. Within the same corridor, vegetation at different locations may exhibit significant differences in growth activity coefficients due to variations in tree species and light and water conditions. Therefore, it is necessary to obtain the growth activity coefficient for each hazard detection point individually.

[0059] The third step is to multiply the basic hazard value by the growth activity coefficient to obtain the hazard level. The hazard level comprehensively reflects the current proximity of the vegetation to the guide wire and its future growth threat. When the clearance distance is less than the preset safety threshold, the basic hazard value is positive, and the hazard level obtained by multiplying it by the growth activity coefficient is also positive; the larger the value, the higher the risk. When the clearance distance is equal to the preset safety threshold, the basic hazard value is zero, and the hazard level is zero. When the clearance distance is greater than the preset safety threshold, the basic hazard value is negative, and the hazard level obtained by multiplying it by the growth activity coefficient is also negative, indicating that the vegetation does not currently pose a threat. However, its growth activity coefficient will affect the magnitude of the negative value; that is, vegetation with higher growth activity, even if the current clearance distance is larger, has a hazard level relatively closer to zero, meaning it is more likely to enter the danger zone as it grows.

[0060] For example, the preset safety threshold is 2.5 meters, the clearance distance is 2.0 meters, and the basic hazard value is 0.5 meters. If the growth activity coefficient is 0.8, then the hazard level is 0.5 multiplied by 0.8 equals 0.4. If the growth activity coefficient is 0.4, then the hazard level is 0.5 multiplied by 0.4 equals 0.2. The former indicates a higher hazard level, suggesting that faster-growing trees require higher priority treatment.

[0061] Furthermore, the detection point with the highest hazard value from the one-dimensional terrain profile is selected as the starting point to simulate the water level rise process. In the watershed algorithm, each local maximum is considered the source of a catchment basin. The highest global point is selected as the starting point, which corresponds to the location of the most dangerous hazard, usually the part of the tree canopy closest to the guide line. Simulating the water level rise means starting from the current height value and gradually lowering the threshold, causing water to spread to adjacent low-lying areas.

[0062] As the water level spreads from the current monitoring point to an adjacent monitoring point, the height difference between the two monitoring points is calculated. In the sorted one-dimensional sequence, each monitoring point has two adjacent points: the left adjacent point and the right adjacent point. When the water level drops from the current point to the water level of the adjacent point, the absolute value of the difference between the danger level of the current point and the danger level of the adjacent point is calculated as the height difference.

[0063] If the height difference is less than or equal to the height difference threshold, the adjacent detection points are assigned to the current catchment basin. The height difference threshold is a critical difference used in the watershed algorithm to determine whether adjacent detection points should be assigned to the same catchment basin. It is dynamically calculated based on the rate of change of clearance distance for the same potential hazard point during successive inspections and the growth activity coefficient of that point. Adjacent points with small height differences are considered to belong to different parts of the same tree as the current point, and therefore are assigned to the same catchment basin, i.e., treated as different detection points on the same tree.

[0064] Furthermore, if the height difference exceeds a height difference threshold, a watershed is established at adjacent monitoring points, and a new catchment basin begins with the adjacent monitoring point as the starting point. Adjacent points with large height differences indicate a significant difference in the danger level of the two trees, or a certain interval between the two trees, and should be considered as different trees. A watershed is established at the boundary as a dividing line, and a new catchment basin is opened with the adjacent monitoring point as the starting point.

[0065] The calculation steps for the height difference threshold include: Obtain records of the clearance distance for the same potential hazard point during each inspection, and calculate the rate of change of clearance distance. The height difference threshold is determined based on the rate of change of the clearance distance and the growth activity coefficient of the hazard detection point, wherein the height difference threshold is positively correlated with the rate of change of the clearance distance and also positively correlated with the growth activity coefficient.

[0066] The first step is to obtain the clearance distance records for the same potential hazard point during each inspection and calculate the rate of change of clearance distance. The same potential hazard point refers to an independent hazard point identified after clustering and merging, representing a specific tree. During multiple inspections, the system records the clearance distance measurement for this hazard point each time. By comparing the clearance distance between two adjacent inspections, the change in clearance distance for that tree within the inspection time interval can be calculated. Dividing the change in clearance distance by the inspection time interval yields the rate of change of clearance distance per unit time, such as meters per month or meters per quarter. A positive rate of change of clearance distance indicates that the distance between the tree and the guide line is increasing, possibly due to pruning or the tree's growth direction deviating from the guide line; a negative rate of change of clearance distance indicates that the clearance distance is decreasing, and the tree is growing towards the guide line. The larger the absolute value of the rate of change of clearance distance, the faster the tree grows or the more significant the pruning effect.

[0067] The second step involves determining the height difference threshold based on the rate of change of clearance distance and the growth activity coefficient of the hazard detection points. The height difference threshold is positively correlated with both the rate of change of clearance distance and the growth activity coefficient. This height difference threshold is the critical difference used in the watershed algorithm to determine whether adjacent detection points belong to the same catchment basin. For trees with a large rate of change of clearance distance and a high growth activity coefficient, it indicates that the tree is growing rapidly and its condition is changing quickly. The clearance distance may decrease significantly between two adjacent inspections. Therefore, a larger height difference threshold needs to be set to avoid incorrectly segmenting the same tree into multiple independent hazard points due to excessively large differences in the degree of danger between detection points at different times. Conversely, for trees with a small rate of change of clearance distance and a low growth activity coefficient, growth is slow, and the difference in the degree of danger between different detection points is relatively stable. A smaller height difference threshold can be set for these trees.

[0068] For example, a base height difference threshold can be set to 0.1. For potential hazards with a negative rate of change in clearance distance and a large absolute value, such as a monthly decrease in clearance distance of 0.3 meters and a growth activity coefficient of 0.8, the height difference threshold is calculated as follows: the height difference threshold equals the base height difference threshold multiplied by 1 plus the product of the absolute value of the rate of change in clearance distance and the growth activity coefficient. Specifically, the calculation is 0.1 × (1 + | - 0.3 | × 0.8) = 0.124. For slow-growing trees, such as trees with a negative rate of change in clearance distance of 0.05 meters per month and a growth activity coefficient of 0.2, the height difference threshold is 0.1 × (1 + 0.05 × 0.2) = 0.101. Through the above dynamic adjustment, the height difference threshold can adapt to the growth characteristics of different trees, ensuring that different detection points of the same tree can be correctly merged during the watershed clustering process, while also ensuring that detection points of adjacent trees are not incorrectly merged.

[0069] Finally, the simulated water level rise process is repeated until all monitoring points have been traversed. Specifically, the water level gradually rises from the initial height value. When the water level rises to the height of an adjacent monitoring point, a decision is made based on the height difference threshold to either merge the adjacent monitoring point into the current basin or establish a watershed to create a new basin. Traversal stops after all monitoring points have been assigned to a catchment basin. All monitoring points within each catchment basin are merged into an independent hazard point. Each catchment basin represents a tree, and all monitoring points within the basin are vegetation points on the tree canopy. After the merging operation, dozens of monitoring points generated by a tree are merged into an independent hazard point. Optionally, the representative location of the independent hazard point can be the point with the highest hazard level within the basin, or the center point of the weighted average of all points within the basin based on projected distance. The representative clearance distance of the independent hazard point can be the minimum geometric clearance distance of all points within the basin, or the weighted average based on hazard level.

[0070] Through the aforementioned watershed clustering and merging, intelligent mapping from point cloud-level detection to tree-level hazards is achieved, providing accurate hazard units for subsequent risk statistics. This step solves the problem of duplicate hazard statistics caused by multiple detection points corresponding to the same tree, enabling the number of hazards and risk assessment results to better reflect the actual situation on site.

[0071] S50: Calculate the tree obstacle risk density index based on the merged set of hidden danger points and output an early warning signal.

[0072] Finally, after clustering and merging the hazard detection points, each independent hazard point represents a tree with potential tree obstruction risk. To conduct an overall risk assessment of the transmission line span, information from multiple independent hazard points needs to be integrated into a quantitative comprehensive indicator, namely the tree obstruction risk density index. This index comprehensively reflects the severity of tree obstruction hazards within the current span, providing a quantitative basis for operation and maintenance decisions.

[0073] Specifically, the tree barrier risk density index is calculated based on the merged set of potential hazard points, including: The number of independent potential hazards after merging is used as the first component; Calculate the ratio of the representative clearance distance of each independent hidden danger point to the preset safety threshold, multiply it by the growth activity coefficient of each independent hidden danger point to obtain the weighted risk value, and take the arithmetic mean of the weighted risk values ​​of all independent hidden danger points as the second component; Calculate the vegetation point cloud count corresponding to each independent hazard point, and take the arithmetic mean of the vegetation point cloud counts of all independent hazard points as the third component; The tree barrier risk density index is obtained by weighted summation of the first, second, and third components.

[0074] First, the number of merged independent hazard points is counted and recorded as the first component. After watershed clustering, all monitoring points within each catchment basin are merged into one independent hazard point, each representing a tree that may pose a tree obstacle risk. The total number of all independent hazard points within the current section is counted and denoted as N. This value directly reflects the number of trees within the section that require attention and treatment. The more independent hazard points there are, the greater the maintenance workload and the higher the overall risk of the section. For example, if a section has 5 merged independent hazard points, the first component is recorded as 5.

[0075] Next, the ratio of the representative clearance distance to the preset safety threshold for each independent hazard point is calculated, multiplied by the growth activity coefficient of each independent hazard point, to obtain a weighted risk value. The arithmetic mean of the weighted risk values ​​of all independent hazard points is taken as the second component. For each independent hazard point, the smallest geometric clearance distance is selected from all detection points in the point set as the representative clearance distance. The representative clearance distance is divided by the preset safety threshold to obtain the clearance distance ratio. When the clearance distance ratio is less than 1, it indicates that the hazard point has currently encroached on the safe zone; when the clearance distance ratio is equal to 1, it indicates that the hazard point is exactly on the safe boundary; when the clearance distance ratio is greater than 1, it indicates that the hazard point is currently outside the safe zone. The smaller the clearance distance ratio, the closer the hazard point is to the guide wire, and the higher the risk. The clearance distance ratio is multiplied by the growth activity coefficient of the hazard point to obtain the weighted risk value. The higher the growth activity coefficient, the more active the tree growth, and the faster the future risk increases, thus the larger the weighted risk value. The weighted risk value ranges from 0 to positive infinity, but when the clearance ratio is no greater than 1 and the growth activity coefficient is no greater than 1, the weighted risk value is between 0 and 1. The weighted risk value is calculated for all independent potential hazard points within the section, and the arithmetic mean is taken as the second component. The larger the value of the second component, the closer the trees are to the guide rail on average or the more active their growth, and the higher the overall risk.

[0076] For example, within a certain section, there are three independent potential hazard points, representing clearance distances of 2.0 meters, 2.2 meters, and 2.8 meters, respectively. The preset safety threshold is 2.5 meters, and the growth activity coefficients are 0.8, 0.6, and 0.3, respectively. Then, the clearance distance ratios are 0.8, 0.88, and 1.12, respectively. The weighted risk values ​​are 0.64, 0.528, and 0.336, respectively. The arithmetic mean of the weighted risk values ​​is 0.501, and the second component is also 0.501.

[0077] Furthermore, the number of vegetation point clouds corresponding to each independent hazard point is calculated, and the arithmetic mean of the number of vegetation point clouds at all independent hazard points is taken as the third component. Each independent hazard point corresponds to a catchment basin, and the number of vegetation point clouds contained within the basin reflects the size and density of the tree canopy. A higher number of point clouds generally indicates a larger canopy, and thus a greater impact and damage range on the conductor should a tree fall or a branch break. The number of vegetation point clouds contained in each independent hazard point is counted, and the arithmetic mean of the number of vegetation point clouds at all independent hazard points is calculated and denoted as M. This value reflects the average size of the hazard-prone trees within the section. For example, in the above example, the number of vegetation point clouds at the three independent hazard points are 150, 80, and 30 respectively, and the arithmetic mean is 86.67, so the third component is denoted as 86.67.

[0078] Finally, the first, second, and third components are weighted and summed to obtain the tree barrier risk density index. Since the dimensions and numerical ranges of the three components differ significantly, dimensionless processing is required first. For example, each component can be divided by its possible maximum value or a min-max normalization method can be used to map each component to the interval between 0 and 1. After normalization, each component is multiplied by its corresponding weight coefficient and summed. The weight coefficient of the first component reflects the influence of the number of hazards on the overall risk; the weight coefficient of the second component reflects the influence of hazard proximity and growth activity; and the weight coefficient of the third component reflects the influence of hazard size. The sum of the three weight coefficients is 1. For example, the weight of the first component can be set to 0.3, the weight of the second component to 0.5, and the weight of the third component to 0.2, emphasizing the dominant role of clearance distance and growth activity on risk. The normalized first, second, and third components are multiplied by their respective weights and then summed to obtain the tree barrier risk density index. The tree barrier risk density index is a dimensionless number between 0 and 1. The closer the tree barrier risk density index is to 1, the higher the risk of tree barriers in the section, and the more priority should be given to handling them.

[0079] By weighting and integrating the above three components, the tree barrier risk density index comprehensively quantifies the overall risk of the section from three dimensions: the number of hidden dangers, the degree of proximity, and the scale of hidden dangers, providing a scientific numerical basis for subsequent graded early warning.

[0080] Finally, the execution steps of this embodiment include outputting a warning signal, including: Obtain the tree obstacle risk density index sequence from the historical inspection records within the current section, and calculate the arithmetic mean and standard deviation of the tree obstacle risk density index sequence; The sum of the arithmetic mean and one standard deviation is used as the first warning threshold, the sum of the arithmetic mean and two standard deviations is used as the second warning threshold, and the sum of the arithmetic mean and three standard deviations is used as the third warning threshold. When the tree obstacle risk density index is less than the first warning threshold, a normal signal is output. When the tree barrier risk density index is greater than or equal to the first warning threshold and less than the second warning threshold, an attention signal is output and a planned pruning suggestion is generated. When the tree obstacle risk density index is greater than or equal to the second warning threshold and less than the third warning threshold, an warning signal is output and a priority processing work order is generated. When the tree obstacle risk density index is greater than or equal to the third warning threshold, an alarm signal is output and an emergency response command is generated.

[0081] First, the tree obstacle risk density index sequence from historical inspection records within the current span is obtained, and its arithmetic mean and standard deviation are calculated. For each transmission line span, the system calculates and stores the tree obstacle risk density index during each inspection. The historical index values ​​are arranged in chronological order of inspection time, forming a tree obstacle risk density index sequence. Specifically, the arithmetic mean of this sequence, denoted as μ, is calculated, representing the long-term average level of tree obstacle risk for that span. The standard deviation, denoted as σ, is calculated, representing the degree of fluctuation in risk values; a larger standard deviation indicates a more significant change in risk level between different inspection cycles. Historical inspection records typically cover at least one complete growth cycle, such as the past year or two, to ensure the representativeness of the statistical results.

[0082] Next, the sum of the arithmetic mean and one standard deviation is used as the first warning threshold, i.e., μ + σ; the sum of the arithmetic mean and two standard deviations is used as the second warning threshold, i.e., μ + 2σ; and the sum of the arithmetic mean and three standard deviations is used as the third warning threshold, i.e., μ + 3σ. These three thresholds divide the risk level into four intervals: normal interval, attention interval, warning interval, and alert interval. Specifically, the principle of this threshold setting method is based on the normal distribution assumption. In a normal distribution, approximately 68% of the data falls within the range of the mean plus or minus one standard deviation, approximately 95% falls within the range of plus or minus two standard deviations, and approximately 99.7% falls within the range of plus or minus three standard deviations. Therefore, when the current index exceeds μ + σ, it indicates that it has exceeded the historical normal fluctuation range and deserves attention; when it exceeds μ + 2σ, it indicates that it is at a historically high level and requires a warning; and when it exceeds μ + 3σ, it indicates that it is at a historically extremely high level and requires emergency action.

[0083] Specifically, when the tree risk density index is less than the first warning threshold, a normal signal is output, indicating that the tree risk in the current section is within the historical normal fluctuation range and no additional maintenance measures are required. When the tree risk density index is greater than or equal to the first warning threshold and less than the second warning threshold, a concern signal is output, and a planned pruning recommendation is generated. The concern signal indicates that the tree risk in the section has exceeded the normal level, and it is recommended that the relevant trees be included in the next routine pruning plan, such as pruning within the next three months. When the tree risk density index is greater than or equal to the second warning threshold and less than the third warning threshold, a warning signal is output, and a priority handling work order is generated. The warning signal indicates that the tree risk in the section has reached a high level and needs to be prioritized for handling, such as completing pruning within the next month. When the tree risk density index is greater than or equal to the third warning threshold, an alarm signal is output, and an emergency response instruction is generated. The alarm signal indicates that the tree risk in the section has reached an extremely high level, posing an immediate safety hazard, and requires immediate handling, such as arranging on-site inspection and emergency pruning within 24 hours.

[0084] Through the aforementioned tiered early warning mechanism based on historical statistics, the early warning threshold can adapt to the historical risk level of each segment, avoiding frequent false alarms caused by setting overly strict thresholds in low-risk areas, and avoiding missed alarms caused by setting overly lenient thresholds in high-risk areas. Due to differences in terrain, vegetation type, and line parameters, the historical risk distribution varies across different segments. The dynamic threshold method provides each segment with independent judgment criteria, improving the targeting and accuracy of early warnings.

[0085] In summary, through the above-mentioned tree obstacle risk density index calculation and graded early warning output, this embodiment can transform a large amount of raw point cloud data into intuitive risk levels and operation and maintenance instructions, realize an intelligent operation and maintenance process from data collection to closed-loop handling, and generate structured early warning signals by comprehensively analyzing multi-dimensional hidden danger information, providing decision support for the orderly formulation of operation and maintenance plans.

[0086] In summary, the embodiments of this application have at least the following technical effects: This invention first acquires 3D point cloud data and multispectral remote sensing data, fits to generate a vectorized digital model of the power line, and classifies it into power line point clouds and vegetation point clouds, achieving precise spatial positioning of the power line and accurate classification of the point clouds, providing a reliable data foundation for clearance distance calculation. Second, it extracts the greenness index of vegetation from the multispectral remote sensing data and calculates canopy sparsity based on the local point cloud density of the vegetation point cloud. These two are then fused to generate a growth activity coefficient, which is used to correct the clearance distance between the power line and vegetation, resulting in a risk-corrected clearance distance. This ensures that the detection results reflect the dynamic risks of vegetation growth, avoiding misjudgments of static clearance distance during the rapid vegetation growth season. Third, when the risk-corrected clearance distance is lower than a preset safety threshold, using the projected distance along the power line direction as a one-dimensional feature and the degree of danger of the clearance distance as a height value, a watershed algorithm is used to cluster and merge hazard detection points. Detection points within the same catchment basin are merged into a single independent hazard point, solving the problem of duplicate hazard statistics caused by multiple detection points corresponding to the same tree.

[0087] Finally, the tree obstacle risk density index is calculated based on the merged set of hidden danger points, and a graded early warning signal is output based on the dynamic early warning threshold of historical inspection statistics. This enables a quantitative assessment of the overall risk of the line section and the differentiated formulation of operation and maintenance plans. This invention solves the technical problems in the prior art, such as the inability of static clearance distance to reflect the dynamic risk of vegetation growth, the distortion of risk assessment due to repeated statistics of hidden danger points, and the lack of a quantitative risk density index and a graded early warning mechanism.

Claims

1. A real-time detection method for tree obstruction hazards on transmission lines based on conductor vectorization fitting, characterized in that, include: Acquire 3D point cloud data and multispectral remote sensing data of transmission lines, fit and generate vectorized conductor digital models, and classify and obtain power line point clouds and vegetation point clouds. The vegetation greenness index is extracted based on the multispectral remote sensing data, and the canopy sparsity is calculated based on the local point cloud density of the vegetation point cloud. The greenness index and canopy sparsity are then fused to generate a growth activity coefficient. The risk-corrected clearance distance is obtained by correcting the clearance distance between the guide wire and the vegetation based on the growth activity coefficient. When the risk correction clearance distance is lower than the preset safety threshold, the watershed algorithm is used to cluster and merge the hidden danger detection points, taking the projection distance of the conductor direction as a one-dimensional feature and the degree of danger of the clearance distance as the height value, and merging the detection points in the same catchment basin into an independent hidden danger point. The tree obstacle risk density index is calculated based on the merged set of potential hazards, and an early warning signal is output.

2. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, Fitting and generating a vectorized conductor digital model, including: Obtain the tower coordinates of the transmission line and determine the suspension points of the conductor on the two towers within each span; Using the span as a unit, the catenary equation is fitted based on the hanging point position to generate the three-dimensional spatial curve equation of the conductor within each span. The curve equations of adjacent spans are spliced ​​together according to the tower coordinates to form a vectorized digital model of the entire continuous line.

3. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, The point clouds of power lines and vegetation were obtained through classification, including: The three-dimensional point cloud data is input into a pre-trained point cloud semantic segmentation network, and the initial classification label of each point is output. The initial classification label includes power line label and vegetation label. Based on the spatial location of the vectorized conductor digital model, the initial classification labels are spatially consistent. Power line labels that deviate from the preset distance range of the vectorized conductor digital model are corrected to vegetation labels, and continuous vegetation labels within the preset distance range of the vectorized conductor digital model are corrected to power line labels. Based on the revised classification labels, the point cloud data is divided into power line point clouds and vegetation point clouds. The training steps of the point cloud semantic segmentation network include: Historical 3D point cloud data of the transmission line corridor was collected, and each point was labeled with its corresponding real classification label to form a training sample set; A point cloud semantic segmentation network is constructed based on deep learning. The training sample set is used as input and the real classification labels are used as the supervision target for supervised training until convergence, resulting in a trained point cloud semantic segmentation network.

4. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, The vegetation greenness index is extracted from the multispectral remote sensing data, and the canopy sparsity is calculated based on the local point cloud density of the vegetation point cloud. The greenness index and canopy sparsity are then fused to generate a growth activity coefficient, including: The reflectance of vegetation in the near-infrared and red bands is extracted from the multispectral remote sensing data, and the normalized vegetation index is calculated as the greenness index. The vegetation point cloud is voxelized according to the preset voxel size. The number of points in each voxel is counted. The ratio of the local point cloud density of each voxel to the theoretical maximum density is calculated. The ratio is subtracted from the result to obtain the sparsity. The arithmetic mean of the sparsity of all voxels covered by the vegetation point cloud is obtained as the canopy sparsity. The growth activity coefficient is obtained by weighting and summing the greenness index and the canopy sparsity.

5. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, Based on the growth activity coefficient, the clearance distance between the guide wire and the vegetation is corrected to obtain the risk-corrected clearance distance, including: Extract the set of traverse reference points from the vectorized traverse digital model, and calculate the shortest Euclidean distance from the vegetation point cloud to the set of traverse reference points as the geometric clearance distance. Dividing the geometric clearance distance by the growth activity coefficient yields the risk-corrected clearance distance.

6. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, When the risk-corrected clearance distance is lower than a preset safety threshold, the watershed algorithm is used to cluster and merge the hazard detection points, using the projected distance along the conductor direction as a one-dimensional feature and the degree of danger of the clearance distance as a height value. This includes: Each hazard detection point is projected onto a one-dimensional axis along the direction of the guide wire, and a one-dimensional terrain profile is constructed based on the hazard level of the clearance distance and the height value of each hazard detection point. The detection point with the highest danger level value is selected from the one-dimensional terrain profile as the starting point to simulate the water level rise process; When the water level spreads from the current monitoring point to an adjacent monitoring point, calculate the height difference between the two monitoring points; If the height difference is less than or equal to the height difference threshold, the adjacent detection point will be assigned to the current catchment basin. If the height difference is greater than the height difference threshold, a watershed is set at the adjacent detection point, and a new catchment basin is started from the adjacent detection point. The simulation of the water level rise process is repeated until all monitoring points have been traversed, and all monitoring points in each catchment basin are grouped into an independent potential hazard point.

7. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 6, characterized in that, The steps for calculating the degree of danger include: Obtain the preset safety threshold, and subtract the clearance distance from the preset safety threshold to obtain the basic hazard value; Obtain the growth activity coefficient at the location of the hazard detection point; The degree of danger is obtained by multiplying the base hazard value by the growth activity coefficient.

8. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 6, characterized in that, The calculation steps for the height difference threshold include: Obtain records of the clearance distance for the same potential hazard point during each inspection, and calculate the rate of change of clearance distance. Based on the rate of change of the clearance distance and the growth activity coefficient of the hazard detection point, a height difference threshold is determined, wherein the height difference threshold is positively correlated with the rate of change of the clearance distance and also positively correlated with the growth activity coefficient.

9. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, The tree barrier risk density index is calculated based on the merged set of potential hazard points, including: The number of independent potential hazard points after merging is used as the first component; Calculate the ratio of the representative clearance distance of each independent hidden danger point to the preset safety threshold, multiply it by the growth activity coefficient of each independent hidden danger point to obtain the weighted risk value, and take the arithmetic mean of the weighted risk values ​​of all independent hidden danger points as the second component; Calculate the vegetation point cloud quantity corresponding to each independent hazard point, and take the arithmetic mean of the vegetation point cloud quantities of all independent hazard points as the third component; The tree barrier risk density index is obtained by weighted summation of the first, second, and third components.

10. The method for real-time detection of tree obstruction hazards on transmission lines based on conductor vectorization fitting according to claim 1, characterized in that, Output warning signals, including: Obtain the tree obstacle risk density index sequence from the historical inspection records within the current section, and calculate the arithmetic mean and standard deviation of the tree obstacle risk density index sequence; The sum of the arithmetic mean and one standard deviation is used as the first warning threshold, the sum of the arithmetic mean and two standard deviations is used as the second warning threshold, and the sum of the arithmetic mean and three standard deviations is used as the third warning threshold. When the tree obstacle risk density index is less than the first warning threshold, a normal signal is output. When the tree barrier risk density index is greater than or equal to the first warning threshold and less than the second warning threshold, an attention signal is output and a planned pruning suggestion is generated. When the tree obstacle risk density index is greater than or equal to the second warning threshold and less than the third warning threshold, an warning signal is output and a priority processing work order is generated. When the tree obstacle risk density index is greater than or equal to the third warning threshold, an alarm signal is output and an emergency response command is generated.

Images