Multi-model transmission line channel three-dimensional fusion method, system and device based on BDF Bayesian data fusion and medium

Abstract

The invention relates to the technical field of charging pile intelligent scheduling, and discloses a multi-model power transmission line channel three-dimensional fusion method, system and device based on BDF Bayesian data fusion and a medium, and the method comprises the steps: obtaining multi-source power transmission line data; performing data fusion on the registered multi-source data through a BDF Bayesian fusion strategy; generating a three-dimensional fusion model of the power transmission line channel; constructing a three-dimensional grid model, and carrying out object recognition through a deep learning algorithm to obtain various obstacles in the power transmission line channel; and constructing a multi-factor weighted scoring model in combination with the geometric features of the obstacles and meteorological data to obtain risk scores of the obstacles and perform hidden danger analysis. When data in a complex environment is processed, data inconsistency and noise interference are overcome, and the precision and robustness of the model are improved; the method effectively processes data changes in a dynamic environment, and adapts to various change factors in a complex scene.

Description

Technical Field

[0001] This invention relates to the field of transmission line safety prediction technology, and in particular to a three-dimensional fusion method, system, equipment and medium for multi-model transmission line channels based on BDF Bayesian data fusion. Background Technology

[0002] With the rapid development of power systems, the safety monitoring and maintenance of transmission lines are receiving increasing attention. Traditional transmission line inspection methods mainly rely on manual inspections or periodic ground checks. This approach is not only inefficient but also prone to overlooking potential safety hazards, especially in complex and dynamic natural environments where the blind spots and limitations of manual inspections are particularly evident. With the continuous maturation of new monitoring technologies such as drones, LiDAR, high-precision cameras, and infrared sensors, intelligent monitoring of transmission lines is gradually becoming an important means to improve detection efficiency and accuracy. These advanced sensors can acquire a large amount of high-precision data, including images, point clouds, and thermal imaging, providing rich information for transmission line monitoring. However, effectively integrating data from different sensors to achieve high-precision, comprehensive 3D modeling and risk assessment of transmission lines remains a significant technological challenge.

[0003] In existing technologies, 3D modeling and monitoring of power transmission lines typically rely on data from multiple sensors, such as LiDAR point cloud data, high-resolution image data, and infrared thermal imaging data, to acquire information. These data often exhibit varying degrees of accuracy, resolution, noise characteristics, and viewing angles across different sensors. Traditional fusion methods (such as weighted averaging and Kalman filtering) struggle to effectively handle these differences, resulting in low data fusion accuracy, particularly in complex environments. Furthermore, current multi-source data fusion methods are mostly designed for static scenes and are ill-suited to handling dynamic environmental changes, such as tree growth and weather variations. This significantly limits the effectiveness of existing methods in dynamic monitoring.

[0004] Meanwhile, Bayesian data fusion technology has been widely applied in data science and artificial intelligence, demonstrating its unique advantages, especially in handling uncertainty, noise, and prior information in multi-source data. However, existing Bayesian data fusion methods have not been fully utilized in the fusion of multi-model 3D data for power transmission lines. Most existing technologies focus primarily on the fusion of single-type data, failing to fully leverage the potential of Bayesian methods in multi-source, multi-model data fusion. Therefore, how to leverage the advantages of Bayesian data fusion, combined with data from different sensors, especially in fields like power transmission lines with unique structural and environmental characteristics, remains a pressing issue. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a multi-model three-dimensional fusion method and system for transmission line channels based on BDF Bayesian data fusion to solve the problems of current methods being unable to provide efficient and accurate monitoring and early warning under real-time or dynamic conditions, being unable to effectively handle uncertainties and noise in multi-source data, being unable to fully utilize prior knowledge from different data sources for prediction, and having insufficient model fusion.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion, comprising:

[0009] Acquire 3D data, image data, and meteorological data of the environment surrounding multi-source transmission lines;

[0010] The registered multi-source data are fused using the BDF Bayesian fusion strategy.

[0011] A three-dimensional fusion model of the transmission line corridor is generated based on the fused data;

[0012] Based on the 3D fusion model, a 3D mesh model is constructed, and object recognition is performed through deep learning algorithms to obtain various obstacles in the power transmission line channel.

[0013] By combining the geometric features of obstacles with meteorological data, a multi-factor weighted scoring model is constructed to obtain the risk score of obstacles and conduct hazard analysis.

[0014] As a preferred embodiment of the multi-model transmission line channel three-dimensional fusion method based on BDF Bayesian data fusion described in this invention, the method involves fusing registered multi-source data using a BDF Bayesian fusion strategy, including:

[0015] Define a state vector that describes the state of the environment;

[0016] Based on the defined state vector, the prior probabilities of each type of data are constructed. Based on the spatial location, category, size and ambient temperature of each obstacle, a probability distribution is defined for each obstacle type to obtain the prior probability model of the corresponding obstacle.

[0017] Acquire observation data from various sensors of the corresponding obstacles, including the spatial coordinates and reflection intensity information of each point in the three-dimensional data, the object appearance information in the image data, thermal imaging data, and meteorological data;

[0018] Each sensor's observation data for the corresponding obstacle is defined as its own likelihood function.

[0019] The likelihood functions of each obstacle are weighted to obtain a joint likelihood function. The joint likelihood function is then combined with the prior probability model of the obstacle, and the posterior probability is calculated using Bayes' theorem.

[0020] The state vector of the corresponding obstacle is updated according to the posterior probability, thereby realizing the fusion of multi-source data of the corresponding obstacle.

[0021] The beneficial effects of this preferred technical solution are as follows: by using the BDF Bayesian data fusion method, information from different sensors is accurately fused, making full use of the complementarity of various types of data, thereby improving the accuracy of obstacle recognition.

[0022] As a preferred embodiment of the multi-model three-dimensional fusion method for transmission line channels based on BDF Bayesian data fusion described in this invention, the method includes: generating a three-dimensional fusion model of the transmission line channel based on the fused data, comprising:

[0023] By integrating data from multiple sources, integrated point cloud data is obtained.

[0024] By combining the updated state vectors of the corresponding obstacles with the integrated point cloud data, a three-dimensional model of all obstacles is obtained, generating a three-dimensional fusion model of the power line channel.

[0025] As a preferred embodiment of the multi-model three-dimensional fusion method for transmission line channels based on BDF Bayesian data fusion described in this invention, the method includes: constructing a three-dimensional mesh model based on the three-dimensional fusion model, comprising:

[0026] Based on the point cloud data in the 3D fusion model, a normal vector is calculated for each point in the point cloud for surface reconstruction;

[0027] The Poisson reconstruction algorithm is used to transform the point cloud into a continuous surface of obstacles;

[0028] Map the attributes in the fusion model to the obstacle reconstruction surface;

[0029] The reconstructed continuous surface model is transformed into a three-dimensional mesh model by applying Delaunay triangulation, generating a polygonal mesh.

[0030] The attributes in the fusion model are assigned to the mesh vertices to obtain a 3D mesh model.

[0031] As a preferred embodiment of the multi-model three-dimensional fusion method for transmission line channels based on BDF Bayesian data fusion described in this invention, the method involves: using deep learning algorithms to identify various obstacles in the transmission line channel, including:

[0032] Extract the vertices of the 3D mesh model into point cloud form and retain the point cloud information;

[0033] Analyze the relationships between points using local feature analysis of point cloud computing to extract geometric features; extract color and edge features from image data; extract temperature features from thermal imaging data;

[0034] Based on the extracted features, multimodal feature vectors are constructed to form a point cloud dataset;

[0035] The PointNet++ algorithm is used to classify and segment feature point clouds to identify obstacle types;

[0036] By using RGB images and thermal imaging data and fusing multimodal information, the obstacle type and probability are output.

[0037] As a preferred embodiment of the multi-model three-dimensional fusion method for transmission line channels based on BDF Bayesian data fusion described in this invention, the method involves: combining obstacle geometric features with meteorological data to construct a multi-factor weighted scoring model to obtain the obstacle risk score, including:

[0038] Obtain the type of obstacle;

[0039] Calculate the shortest distance between the obstacle and the power transmission line;

[0040] Calculate the height and dimensions of the obstacle;

[0041] Acquire meteorological factors to assess the impact of meteorological conditions on obstacle stability;

[0042] By using a weighted scoring method, the overall risk of obstacles in each area is scored, resulting in an obstacle risk score.

[0043] As a preferred embodiment of the multi-model three-dimensional fusion method for transmission line channels based on BDF Bayesian data fusion described in this invention, it further includes: performing accuracy analysis on the multi-factor weighted scoring model, specifically including:

[0044] All types of obstacles are evaluated using four metrics: accuracy, recall, F1 score, and precision.

[0045] The accuracy of the model is evaluated based on the numerical range of each indicator on different types of obstacles.

[0046] Secondly, this invention provides a three-dimensional fusion system for multi-model transmission line channels based on BDF Bayesian data fusion, comprising:

[0047] The acquisition module is used to acquire 3D data, image data, and meteorological data of the environment surrounding multi-source transmission lines;

[0048] The fusion module is used to fuse registered multi-source data using the BDF Bayesian fusion strategy.

[0049] The fusion model construction module is used to generate a three-dimensional fusion model of the transmission line channel based on the fused data;

[0050] The recognition module is used to construct a three-dimensional mesh model based on the three-dimensional fusion model and to identify objects through deep learning algorithms to obtain various obstacles in the transmission line channel;

[0051] The scoring module is used to combine the geometric features of obstacles with meteorological data to construct a multi-factor weighted scoring model, obtain the risk score of the obstacle, and conduct hazard analysis.

[0052] Thirdly, the present invention provides a computer device, comprising:

[0053] Memory and processor;

[0054] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion.

[0055] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the multi-model transmission line channel three-dimensional fusion method based on BDF Bayesian data fusion.

[0056] Compared with existing technologies, the beneficial effects of this invention are as follows: By introducing Bayesian data fusion theory, this invention integrates the data characteristics and noise properties of different sensors, thereby improving the accuracy and stability of multi-source data fusion. Bayesian data fusion, by utilizing prior information, noise models, and uncertainties from various data sources, can overcome data inconsistencies and noise interference in existing technologies when processing data in complex environments, improving the accuracy and robustness of the model. Unlike traditional data fusion methods, this method can effectively handle data changes in dynamic environments and adapt to various changing factors in complex scenarios, such as weather changes and tree growth. It not only provides more accurate 3D models but also better adapts to changes in complex dynamic environments, providing more reliable technical support for intelligent monitoring, risk assessment, and maintenance management of transmission lines. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a schematic diagram of the overall process of a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion according to an embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of the overall architecture of a multi-model transmission line channel three-dimensional fusion system based on BDF Bayesian data fusion, according to an embodiment of the present invention.

[0060] Figure 3 This is a bar chart of evaluation indexes for an evaluation model in a multi-model three-dimensional fusion system for transmission line channels based on BDF Bayesian data fusion, as described in one embodiment of the present invention. Detailed Implementation

[0061] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0062] Example 1, referring to Figure 1 As an embodiment of the present invention, a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion is provided, comprising:

[0063] S100: Acquire 3D data, image data, and meteorological data of the environment surrounding multi-source transmission lines;

[0064] S200: The registered multi-source data is fused using the BDF Bayesian fusion strategy;

[0065] S300: A three-dimensional fusion model of the transmission line corridor is generated based on the fused data;

[0066] S400: Based on a 3D fusion model, a 3D mesh model is constructed, and object recognition is performed through deep learning algorithms to obtain various obstacles in the transmission line channel;

[0067] S500: Combining obstacle geometric features with meteorological data, a multi-factor weighted scoring model is constructed to obtain the obstacle risk score and conduct hazard analysis.

[0068] It should be noted that existing multi-source data fusion methods, especially in the fusion of point cloud data and image data, often struggle to balance the accuracy differences and noise characteristics of data from different sensors. Different types of sensors, such as LiDAR, cameras, and infrared sensors, typically acquire data with varying resolutions, viewpoints, and noise characteristics. Traditional data fusion methods, such as Kalman filtering and weighted averaging, struggle to effectively handle these differences, resulting in poor accuracy and stability of the fusion results. Furthermore, traditional fusion algorithms, such as weighted averaging and Kalman filtering, lack the advantages of Bayesian data fusion, including uncertainty modeling, noise suppression, and information gain. Transmission lines possess unique structural and environmental characteristics, such as high-altitude structures, tree growth, and climate change. Existing technologies have not specifically optimized for these characteristics, leading to poor application performance in specific scenarios and failing to fully leverage the unique advantages of sensor data in this field, thereby reducing the accuracy and reliability of models in specialized scenarios.

[0069] Therefore, steps S100-S500 combine data from various types of sensors using the BDF Bayesian data fusion method to efficiently fuse these data, overcoming the limitations of a single data source and improving the accuracy and robustness of obstacle identification and risk assessment. Furthermore, through 3D modeling and refined obstacle classification, accurate identification and spatial positioning of various obstacles in the transmission line corridor are achieved, with detailed classification of different obstacles. Simultaneously, considering dynamic environmental parameters, comprehensive prediction and real-time assessment of potential transmission line risks are performed. This allows for automatic monitoring of environmental changes around the transmission line without human intervention, and obstacle identification, risk assessment, and early warning based on real-time data.

[0070] Example 2, refer to Figures 1-3 Tables 1 and 2 are examples of one embodiment of the present invention. Based on the above embodiment, a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion is provided.

[0071] In this embodiment of the application, step S100 involves acquiring three-dimensional data, image data, and meteorological data of the environment surrounding the multi-source transmission line;

[0072] Specifically, this can be achieved through various types of sensors, including: LiDAR sensors, which provide high-precision 3D point cloud data to capture the spatial location and structural information of each object in the environment; high-definition RGB imaging cameras, which can capture images of power lines and their surrounding environment, providing auxiliary visual information to facilitate subsequent object recognition and classification; infrared thermal imaging cameras, which can be used to capture heat sources in the environment, especially to identify temperature anomalies in power equipment or temperature changes in trees and power lines; and meteorological sensors, which can collect real-time meteorological data such as temperature, humidity, and wind speed to assist in analyzing data changes, especially adaptability in dynamic environments.

[0073] In one alternative implementation, the LiDAR data acquisition location can be selected from the typical environment surrounding a power transmission line corridor, including nine types of trees (such as eucalyptus, camphor, and pine), ground, water, and buildings. By mounting the LiDAR sensor on a drone, the power transmission line corridor and its surrounding environment can be scanned from different angles and altitudes. The acquisition frequency and range are set to approximately 50,000 points per second (assuming a point cloud density of 50 kHz) during flight, with the scanning range ranging from a few meters to several hundred meters from the sensor. The data format is a point cloud dataset output by the LiDAR sensor, where each point has spatial coordinates (x, y, z), and the point cloud data can cover the power transmission line and an area within 100 meters of it.

[0074] For example, for eucalyptus trees, the point cloud data collected by the LiDAR sensor shows the tree's height, the outline of its crown, and its relative position to power lines. If a eucalyptus tree is close to a power line, its height may be over 15 meters and its crown diameter may reach 10 meters.

[0075] In another alternative implementation, based on the above implementation, the Voxel Grid filtering algorithm can also be used to downsample the point cloud data, reducing the data volume while retaining key feature information; and the spatial location and corresponding reflection intensity of each point can be calculated, the calculation formula of which can be expressed as:

[0076]

[0077] in, This represents the spatial coordinates of each point in the point cloud. For a single point in a point cloud, This represents the number of data collection points.

[0078] For RGB image data acquisition, high-definition cameras can be mounted on flight platforms or vehicles to capture images of power transmission lines and their surrounding environment. RGB image data will primarily be used to assist in object recognition and classification, such as trees and buildings, and will be combined with point cloud data for 3D modeling. The data format can be set to a resolution of 1920*1080, with each frame including pixel data from the red, green, and blue channels.

[0079] It should be noted that when identifying nine common tree species, the images captured by the RGB camera show the external features of the trees, such as the color of the bark and the shape of the leaves, which helps the subsequent deep learning algorithm to identify and classify them.

[0080] Specifically, thermal imaging cameras are used to detect temperature information in areas such as trees, power lines, and buildings; they are usually output in the form of a temperature field, with each pixel representing the temperature value of an area. This numerical information is very important for detecting whether trees and power lines are overheating or abnormal.

[0081] For example, taking pine trees as an example, if the temperature near the crown of a pine tree is high, it may indicate that it is close to a high-voltage power transmission line and there may be heat accumulation caused by electromagnetic radiation, which requires further analysis.

[0082] Specifically, meteorological sensors collect data such as ambient temperature, humidity, wind speed, and precipitation. This data helps to assess changes in the environment around power transmission lines, especially the impact of meteorological conditions on tree growth and tree obstruction.

[0083] For example, assuming that the weather sensor records a temperature of 30°C, a wind speed of 10 m / s, and a humidity of 60%, it can be applied to the subsequent Bayesian data fusion process.

[0084] It should be noted that traditional transmission line monitoring methods typically rely on a single data source, such as visual images or point cloud data alone. This single-data-source approach is easily affected by environmental factors, resulting in low identification accuracy. In contrast, this patented technology comprehensively utilizes multiple types of information, including point clouds, RGB images, thermal imaging images, and meteorological sensor data, fully leveraging the advantages of each data type and improving data utilization efficiency. By fusing data with a risk assessment model, specific maintenance and inspection plans can be generated based on different risk scores, prioritizing high-risk areas and ensuring the safe operation of the power system.

[0085] In this embodiment of the application, step S200 involves fusing the registered multi-source data using a BDF Bayesian fusion strategy, including the following steps A1-A6:

[0086] A1: Defines a state vector that describes the state of the environment;

[0087] Specifically, Bayes' theorem updates the state estimate using prior knowledge and observed data. For multi-sensor fusion, it is assumed that a state vector needs to be estimated. Such as the three-dimensional coordinates and temperature of each point in the environment, while the observation data of each sensor are... Bayes' theorem formula is as follows:

[0088]

[0089] in, It is the prior probability, representing the probability of a state variable without any observed data. The estimate; It is the likelihood function, representing the likelihood under a given state. In this case, the sensor observes the data The probability of; This is the evidence term, representing the probability of all possible observations, and is usually a normalization constant.

[0090] It should be noted that, using Bayes' theorem, the posterior probability can be obtained by combining the sensor's observation data with the prior probability. This updates the state estimate.

[0091] In this embodiment, state variables It can be multi-dimensional, including point cloud data, the object's position, shape, and thermal imaging temperature, among other information.

[0092] For example, obstacles (such as trees, buildings, water, etc.) in the transmission line channel are described as three-dimensional point cloud data in space;

[0093] State variables ={location, type, size, temperature};

[0094] The spatial location of each obstacle (such as the three-dimensional coordinates of trees, buildings, etc.);

[0095] Categories of obstacles (such as the classification of trees like eucalyptus and pine);

[0096] The dimensions of each obstacle (such as the height of trees, the diameter of tree crowns, etc.);

[0097] Environmental temperature information (such as the temperature of tree or building surfaces).

[0098] A2: Based on the defined state vector, construct the prior probability of each type of data. Based on the spatial location of each obstacle, the category of each obstacle, the size of each obstacle, and the ambient temperature information, define the probability distribution for each obstacle type to obtain the prior probability model of the corresponding obstacle.

[0099] For example, taking eucalyptus trees as an example, prior information can be collected first:

[0100] Eucalyptus tree location: 30-50 meters from the line, assuming a Gaussian distribution with a mean of 40 meters and a standard deviation of 5 meters.

[0101] Type: Eucalyptus, probability 0.2 (based on regional tree species distribution).

[0102] Height: 10-20 meters, mean 15 meters, standard deviation 2 meters.

[0103] Crown diameter: 5-15 meters, mean 10 meters, standard deviation 2 meters.

[0104] Temperature: 20-30°C, mean 25°C, standard deviation 3°C.

[0105] Reconstruct the prior probabilities:

[0106]

[0107] It should be noted that prior probability This represents a preliminary estimate of these state variables before any observational data is available. Prior information typically comes from historical data or environmental models. Historical environmental data can be used to estimate tree distribution, such as the fact that eucalyptus trees typically grow in certain areas; existing topographical information about power transmission line corridors can be used to infer the location of buildings or water bodies.

[0108] A3: Acquire observation data from various sensors of the corresponding obstacles, including the spatial coordinates and reflection intensity information of each point in the 3D data, the appearance information of the object in the image data, thermal imaging data, and meteorological data;

[0109] Specifically, LiDAR point cloud data provides the spatial coordinates and reflectance intensity information for each point, describing the geometric shape of objects such as trees and buildings; RGB image data includes the color appearance information of objects such as trees and buildings, which helps in target identification and classification; thermal imaging data records the surface temperature of objects in the environment, such as the temperature difference between trees, buildings and power transmission lines; and meteorological sensor data records environmental information such as temperature and humidity.

[0110] A4: Define the observation data of each sensor for the corresponding obstacle as its own likelihood function;

[0111] Specifically, the observation data of each sensor All of these carry a certain degree of error, therefore it is necessary to define a likelihood function. That is, given state variables At that time, the observation data The probability distribution of the likelihood function represents the relationship between the observed data and the true state. It is usually assumed that the data error follows a Gaussian distribution.

[0112] For LiDAR point cloud data, the likelihood function can be expressed as the distance between the spatial coordinates of each point in the point cloud and the location of an obstacle. For RGB image data, the likelihood function can be the degree to which each pixel in the image matches the appearance of the target object. For thermal imaging data, the likelihood function can represent the difference between the temperature value and the surface temperature of the target object.

[0113] A5: Weight the likelihood functions of each obstacle to obtain the joint likelihood function. Combine the joint likelihood function with the prior probability model of the obstacle and calculate the posterior probability using Bayes' theorem.

[0114] Specifically, in practice, since data from multiple sensors is involved, a weighted average method is typically used to fuse the observation information from each sensor. Posterior probability. The calculation can be performed through the following process:

[0115] The observation data from each sensor are weighted, with each sensor's data assigned a different weight based on its reliability and accuracy. The larger the weight, the greater the proportion of the sensor's data in the posterior estimation. LiDAR data has high spatial accuracy, so it has a larger weight, while meteorological sensor data has a smaller impact and therefore a lower weight.

[0116]

[0117] in, For the first The weight of each sensor, For the first The likelihood function of each sensor. This represents the prior probability.

[0118] A6: Update the state vector of the corresponding obstacle according to the posterior probability to achieve the fusion of multi-source data of the corresponding obstacle.

[0119] Specifically, by combining data from all sensors, the spatial location, category, and size of each obstacle (such as trees and buildings) can be updated. By fusing LiDAR and thermal imaging data, it may be possible to obtain the precise location, height, canopy size of a eucalyptus tree, as well as information such as the temperature difference between it and power transmission lines.

[0120] For example, in practical applications, data from multiple sensors are typically fused simultaneously;

[0121] LiDAR point cloud data for eucalyptus trees provides their 3D coordinates, RGB image data provides their appearance features, and thermal imaging data describes their surface temperature. The Bayesian data fusion process integrates data from these three sensors to update the tree's spatial location, appearance features, size, and temperature. After Bayesian fusion of all sensor data, a 3D model of the eucalyptus tree is obtained, and the following can be estimated using the fused posterior probability:

[0122] Tree location: 30 meters from the power transmission line;

[0123] Tree height: 15 meters;

[0124] Tree crown diameter: 10 meters;

[0125] Tree temperature: 24°C.

[0126] It should be noted that in step S200, the BDF Bayesian data fusion method is used to accurately fuse information from different sensors (such as LiDAR point clouds, RGB images, thermal imaging data, meteorological sensor data, etc.), making full use of the complementarity of various types of data and improving the accuracy of obstacle recognition. Whether in complex terrain conditions or in environments with changing weather, this method can effectively extract the spatial features and attributes of obstacles, avoiding the limitations of a single data source in harsh environments.

[0127] In one alternative implementation, a preprocessing step is included before registration in step A1;

[0128] LiDAR point cloud data may contain a large number of noisy points, especially in complex environments such as areas with obstacles like trees, buildings, and water. To remove this noise, the Voxel Grid filter can be used to downsample the point cloud. The Voxel Grid filter formula is as follows:

[0129]

[0130] in, This represents each point in the collected point cloud. The number of points in each grid, This represents the mean of the grid. This process effectively reduces the density of point cloud data, improves processing speed, and reduces the burden of subsequent calculations.

[0131] Meanwhile, since the ground often contains a large number of low points, and these low points do not need to be considered during the modeling process, ground point removal algorithms, such as height-based region segmentation algorithms, are used to remove ground points, thereby retaining only the points related to obstacles such as trees and buildings.

[0132] For example, assuming the height of the ground point cloud is 0-1 meter, all points below 1 meter can be filtered out by setting a height threshold.

[0133] For RGB image data, distortion may occur during camera capture, causing straight lines in the image to appear curved. Camera calibration is used to correct image distortion. By using the intrinsic and extrinsic parameters obtained from the calibration, geometric distortion in the image can be eliminated. The formula can be expressed as:

[0134]

[0135] in, These are the pixel coordinates in the image. Let be a rotation matrix. It is a translation vector. This is the intrinsic parameter matrix of the camera.

[0136] Meanwhile, for RGB images, histogram equalization and Gaussian blur can be used to enhance image contrast and remove background noise, which helps in subsequent target recognition and classification tasks and improves the recognition accuracy of targets such as trees and buildings.

[0137] For thermal imaging data, the output of a thermal imaging camera is the temperature value for each pixel. Due to the influence of the external environment, the acquired temperature data may exhibit varying measurement ranges. A normalization algorithm is used to standardize the temperature data, mapping all temperature values ​​to a range of 0 to 1, thus eliminating differences between different devices.

[0138] Meteorological data (such as temperature and humidity) may contain missing or outlier values ​​during the data collection process. A moving average algorithm is used to smooth the data, and interpolation is used to fill in the missing data, ensuring that complete meteorological data is available for subsequent analysis at each time point.

[0139] In another optional implementation, based on the above preprocessing method, the registration method may include:

[0140] ① Registering LiDAR data with RGB images: Registration of LiDAR point cloud data with RGB image data can be achieved through geometric transformation, typically using the Iterative Closest Point (ICP) algorithm to align corresponding points in the point cloud and the image. Specifically, the ICP algorithm process includes:

[0141] Preliminary registration is performed based on the relative positions of the camera and the LiDAR sensor (given by the extrinsic parameter matrix);

[0142] Find the points in the point cloud that correspond to feature points in the RGB image, such as tree trunks or building corners; calculate the Euclidean distance between each pair of corresponding points and minimize these distances; iterate until the registration error converges.

[0143] The ICP registration formula can be expressed as:

[0144]

[0145] in, For points in a LiDAR point cloud, For the corresponding points in the RGB image, Let be a rotation matrix. It is a translation vector. The number of matched point pairs.

[0146] ② Register the LiDAR data with the thermal imaging data. Thermal imaging data is typically acquired based on the sensor's orientation and angle. Therefore, during registration, the spatial coordinates in the LiDAR point cloud need to be mapped to the temperature values ​​in the thermal imaging image. This process is similar to the registration of LiDAR and RGB images. However, since the scale of thermal imaging data differs from that of image data, the key to the registration process is aligning the spatial distribution of temperature values ​​with the geometric information of the LiDAR point cloud data. In this process, a similar ICP algorithm can be used, adjusting the rotation and transformation matrices to ensure that the temperature field in the thermal imaging image is correctly mapped to the corresponding three-dimensional spatial coordinates.

[0147] ③ Joint registration of multi-sensor data: In multi-sensor registration, the ultimate goal is to fuse data from all sensors (LiDAR, RGB, thermal imaging, meteorology, etc.) into a unified 3D space. This allows for subsequent adjustment of the weights of each sensor's data using a Bayesian data fusion algorithm, based on prior probabilities and the probability distribution of observed data, to ensure that the final generated 3D model is spatially precisely aligned.

[0148] It should be noted that data preprocessing and registration are crucial steps in successfully aligning data from different sensors and providing accurate data support for subsequent 3D modeling and risk assessment, ensuring that sensor data can be fused and accurately reflect the spatial relationships of the environment. Data preprocessing mainly includes processes such as noise reduction, downsampling, and standardization, while data registration aligns data from different sensors to the same coordinate system, thereby constructing an accurate 3D environment model.

[0149] In this embodiment of the application, step S300, which generates a three-dimensional fusion model of the transmission line corridor based on the fused data, includes the following steps B1-B2:

[0150] B1: Integrate multi-source data to obtain integrated point cloud data;

[0151] For example, LiDAR can provide the geometry of eucalyptus point clouds, such as location and size; RGB mapping textures to point clouds to confirm types, such as green leaves and gray bark of eucalyptus; thermal imaging to assign temperature values ​​(24.1°C) to point cloud points; and meteorology to add environmental background to the model, such as the impact of wind speed of 10 m / s on the risk of lodging.

[0152] B2: Combine the updated state vectors of the corresponding obstacles with the integrated point cloud data to obtain a three-dimensional model of all obstacles, and generate a three-dimensional fusion model of the power line channel.

[0153] Specifically, the updated state vector is combined with the point cloud data to form a unified representation. Each point contains coordinates, obstacle type labels, and temperature values. This process is repeated for all obstacles (9 types of trees, buildings, ground, water, etc.) to generate a complete 3D fusion model.

[0154] It should be noted that the 3D fusion model contains a point cloud dataset of all obstacles, with each point having attributes (location, type, size, temperature).

[0155] In this embodiment of the application, step S400, based on the three-dimensional fusion model, constructs a three-dimensional mesh model, including the following steps C1-C5:

[0156] C1: Based on the point cloud data in the 3D fusion model, calculate the normal vector for each point in the point cloud for surface reconstruction;

[0157] In one alternative implementation, the minimum eigenvector can be calculated as the normal vector based on principal component analysis (PCA) of the neighborhood points (k=30).

[0158] For example, the normal vector of the eucalyptus point (30.05, 0.1, 15) is (0, 0, 1), representing the surface of the canopy facing upwards.

[0159] C2: Use the Poisson reconstruction algorithm to transform the point cloud into a continuous surface of obstacles;

[0160] Specifically, the Poisson reconstruction algorithm is expressed as:

[0161]

[0162] in, It is the objective function. It is the normal field, and ∇ is the divergence operation. These are spatial coordinates.

[0163] C3: Maps the attributes in the fusion model to the obstacle reconstruction surface;

[0164] It should be noted that through steps C1-C3, a high-precision three-dimensional point cloud model can be obtained, which can accurately represent various obstacles such as trees, buildings, and ground in the transmission line channel.

[0165] C4: Apply Delaunay triangulation to the reconstructed continuous surface model to convert it into a 3D mesh model, generating a polygonal mesh;

[0166] It should be noted that after completing the point cloud reconstruction, the next step is to perform mesh modeling, converting the reconstructed point cloud into a polygonal mesh model, and using the Delaunay triangulation method to generate a 3D mesh. This method can generate triangular meshes without interior angles less than 90 degrees, making it suitable for modeling complex 3D objects such as trees and buildings.

[0167] Specifically, the Delaunay triangulation formula is expressed as:

[0168]

[0169] in, Represents a triangular mesh. It is a point set.

[0170] C5: Assign the attributes in the fusion model to the mesh vertices to obtain a 3D mesh model.

[0171] It should be noted that C4-C5 mesh modeling can produce a complete 3D model containing trees, buildings, ground, water, etc., and the geometry of these obstacles can be used for subsequent identification and analysis.

[0172] In this embodiment of the application, step S400 uses a deep learning algorithm to identify objects and obtain various obstacles in the transmission line channel, including the following steps D1-D5:

[0173] D1: Extract the vertices of the 3D mesh model into point cloud form and retain the point cloud information;

[0174] Specifically, point cloud information can include coordinates, normal vectors, and attributes (type label, temperature, color, etc.).

[0175] D2: Analyze the relationships between points from local features of point cloud computing to extract geometric features; extract color and edge features from image data; extract temperature features from thermal imaging data;

[0176] D3: Based on the extracted features, construct multimodal feature vectors to form a point cloud dataset;

[0177] D4: Use the PointNet++ algorithm to classify and segment the feature point cloud and identify obstacle types;

[0178] Specifically, the PointNet++ algorithm formula is as follows:

[0179]

[0180] in, For the first Feature representation of each point, For the first point cloud There are several points, and MLP stands for Multilayer Perceptron Network.

[0181] It should be noted that PointNet++ is a point cloud-based deep learning algorithm capable of handling irregular 3D point cloud data. Through multi-level point cloud feature learning, PointNet++ can effectively identify and classify different categories of obstacles. It hierarchically aggregates local features of the point cloud and then combines these local features with global features for classification and segmentation tasks.

[0182] In another alternative implementation, the PointNet++ model can be trained using a labeled training dataset (including trees such as eucalyptus and pine, and categories such as buildings and water) to learn the features of each point. This network can then classify each point in the point cloud and identify different categories of obstacles.

[0183] Based on the above implementation method, the trained PointNet++ model can accurately identify nine common types of trees (such as eucalyptus, pine, and tung trees), buildings, ground, and water obstacles. Specific data experiments demonstrate that after processing the input LiDAR point cloud data with PointNet++, the recognition accuracy for eucalyptus trees can reach 98%, pine trees 96%, and buildings 95%.

[0184] It should also be noted that obstacle recognition refers to the automatic identification of potential obstacles related to power transmission lines from a 3D mesh model. Because the environment surrounding power transmission lines is complex and contains various obstacles, intelligent algorithms are needed for effective obstacle recognition. Through precise 3D modeling and obstacle classification, different types of obstacles can be identified, such as different kinds of trees, buildings, and water. Particularly in tree recognition, by combining the geometric features of trees (such as height and canopy size) and meteorological information (such as wind speed), the risk level of various types of trees can be accurately assessed, ensuring the safety of power transmission lines.

[0185] D5: Utilizes RGB images and thermal imaging data, fuses multimodal information, and outputs obstacle types and probabilities.

[0186] It should be noted that, in addition to using PointNet++ for point cloud classification, data from RGB images and thermal imaging images can also be used for auxiliary recognition.

[0187] In one alternative implementation, tree appearance classification is performed using texture features from RGB images combined with a convolutional neural network (CNN). For example, eucalyptus leaves and pine needles have unique texture features. Different tree species can be distinguished using image classification networks such as ResNet.

[0188] Thermal imaging provides information about the surface temperature of trees and buildings. By analyzing temperature changes, it can help identify high-temperature areas, which are usually associated with buildings, fire sources, or other hot objects. In actual experiments, combining thermal imaging with observations revealed that when the air temperature is high, the temperature of certain trees (such as pine trees) and buildings is significantly higher than that of the surrounding environment, thus allowing for the assessment of whether these objects pose a risk.

[0189] It should be noted that existing technologies often use two-dimensional images or point cloud data to identify obstacles. This invention, however, utilizes three-dimensional modeling technology to provide a more precise description and analysis of obstacles in three-dimensional space. By accurately modeling the three-dimensional features of obstacles such as trees, buildings, and water bodies (e.g., tree height, canopy size, building volume), the risk posed by various obstacles to power transmission lines can be assessed more precisely. The combination of three-dimensional modeling and fine-grained obstacle classification not only improves identification accuracy but also provides richer and more accurate data support for subsequent risk assessment. Especially in complex environments, three-dimensional data can more accurately describe the shape and spatial distribution of obstacles, helping the system identify different types of risk sources. Through the fused three-dimensional model and image data, precise obstacle classification can be performed, allowing each tree category to be classified and labeled according to its height, width, and distance from the power line.

[0190] In this embodiment of the application, step S500 combines the geometric features of the obstacle with meteorological data to construct a multi-factor weighted scoring model, obtaining the obstacle's risk score, including E1-E5:

[0191] E1: Obtain obstacle type;

[0192] It should be noted that different types of obstacles (such as trees, buildings, water, etc.) pose different levels of threat to power transmission lines. For example, trees pose a greater risk of falling over, while buildings pose a lesser risk of collapsing.

[0193] E2: Calculate the shortest distance between obstacles and power transmission lines;

[0194] It should be noted that the distance between obstacles such as trees, buildings, ground, and water and power transmission lines is an important factor in assessing potential risks. The closer the distance, the greater the risk.

[0195] Specifically, based on 3D point cloud data and obstacle information acquired by sensors, the spatial distance between each obstacle (such as trees, buildings, etc.) and the power transmission line is calculated. Specifically, the minimum distance for each obstacle (i.e., the shortest distance from the obstacle surface to the power transmission line) is calculated and used as an important risk assessment factor, which can be expressed as:

[0196]

[0197] in, Indicates the shortest distance between the obstacle and the power transmission line. , , and These are the three-dimensional coordinates of the obstacle and the power transmission line, respectively.

[0198] E3: Calculate the height and dimensions of obstacles;

[0199] It should be noted that the height of trees, the size of their canopies, and the height of buildings are key indicators for assessing potential risks. The greater the height, the greater the potential impact of the obstacle on the power transmission line.

[0200] Specifically, for trees, a risk scoring function is defined based on the tree's height. and crown diameter Calculate its risk score. For buildings, this can be determined by their height. To estimate the risk.

[0201] Example of a tree risk scoring function:

[0202]

[0203] in, and The maximum height and crown diameter of the tree. and It is a weighting coefficient, which represents the degree of influence of different factors on risk assessment.

[0204] Building risk scoring function:

[0205]

[0206] in, The maximum height of the building. These are the weighting coefficients.

[0207] E4: Acquire meteorological factors to assess the impact of meteorological conditions on obstacle stability;

[0208] It should be noted that meteorological conditions such as wind speed and humidity may exacerbate the collapse or movement of obstacles, increasing the risk to power transmission lines.

[0209] Specifically, real-time data obtained through weather sensors can be used to assess the impact of weather conditions on obstacles. In strong winds, the risk of trees falling increases significantly.

[0210] The effect of wind speed on trees can be calculated using the following formula:

[0211]

[0212] in, This is the current wind speed. That is the maximum safe wind speed.

[0213] E5: By using a weighted scoring method, the overall risk of obstacles in each area is scored to obtain the obstacle risk score.

[0214] Specifically, a comprehensive risk score is obtained by weighting the above factors. The formula for calculating the comprehensive risk score Rtotal is as follows:

[0215]

[0216] in, The weighting coefficients for each factor reflect their contribution to the final risk assessment.

[0217] It should be noted that this assessment method can not only identify the location, type, and geometric features of obstacles, but also conduct a comprehensive risk assessment by combining environmental data (such as wind speed, temperature, and humidity). By comprehensively considering factors such as the spatial distance between obstacles (trees, buildings, water, etc.) and transmission lines, the size of the obstacles, and meteorological conditions, it can accurately assess the potential threat of various obstacles to transmission lines, providing a scientific basis for power sector decision-making.

[0218] Furthermore, based on real-time sensor data updates, it can dynamically track changes in obstacles in the transmission line corridor and adjust risk assessment results in real time. In practical applications, if new obstacles appear or weather conditions change, the risk score can be automatically recalculated and an alarm can be triggered, which helps to carry out timely inspections or cleanup work, thereby reducing safety hazards.

[0219] In an alternative implementation, the hazard analysis in step S500 can be performed using a risk assessment table.

[0220] For example, taking nine common trees (eucalyptus, camphor, paper mulberry, oak, tung tree, cypress, pine, fir, and bamboo), ground, water, and buildings along power transmission line corridors as examples, risk scores are calculated based on factors such as the distance, height, and size of each obstacle, as well as weather conditions, and a comprehensive risk assessment is derived.

[0221] Distance calculation: The shortest distance between each obstacle (tree, building, etc.) and the power line is 30 meters, 50 meters, 100 meters, etc.

[0222] Tree height and dimensions: The height (in meters) and crown diameter (in meters) of each tree are given values.

[0223] Building height: The building height (unit: meters) is 30 meters.

[0224] Meteorological conditions: wind speed 20 m / s, temperature 25°C, humidity 60%.

[0225] The comprehensive risk assessment results are shown in Table 1:

[0226] Table 1: Comprehensive Risk Assessment Results

[0227] Based on the risk assessment, the following analysis can be performed:

[0228] Eucalyptus: Eucalyptus has the highest overall risk score (0.92) because it is close to power transmission lines (30 meters), has a large height and canopy, and is at greater risk of falling over due to wind speed.

[0229] Pine trees: Pine trees have a high risk score (0.90). Despite the close proximity (25 meters) and the large size of the trees themselves, the wind speed weighting factor is significant, resulting in a high risk of them falling over.

[0230] Buildings: Buildings pose a lower risk (0.60). Although they are far away, their height and structure do not pose a threat to power lines as easily as trees.

[0231] Moso bamboo: Moso bamboo has a lower risk (0.58) because it is smaller in height and canopy and is farther away, so it is less affected by wind speed.

[0232] Ground and water: Since ground and water do not directly affect the structure of transmission lines, the risk scores are relatively low (ground 0.30, water 0.25).

[0233] Therefore, we can conclude that:

[0234] High-risk areas: Trees such as eucalyptus, pine, and camphor trees, especially those close to power transmission lines and in areas with high wind speeds, should be prioritized for monitoring and clearing to prevent trees from falling and affecting power transmission lines.

[0235] Medium-risk areas: Trees such as fir, paper mulberry, and oak, although the risk is low, still need to be regularly inspected and monitored to prevent the risk from increasing due to weather changes, especially when weather conditions are poor, and emergency measures should be taken.

[0236] Low-risk areas: Buildings, bamboo, ground, and water have little impact on power transmission lines, but periodic inspections should be conducted to ensure they are not subject to other potential threats.

[0237] In this embodiment of the application, the method may also include step S600: performing accuracy analysis on the multi-factor weighted scoring model, specifically including:

[0238] All types of obstacles are evaluated using four metrics: accuracy, recall, F1 score, and precision.

[0239] The accuracy of the model is evaluated based on the numerical range of each indicator on different types of obstacles.

[0240] Specifically, precision represents the proportion of obstacles correctly identified by the model out of all obstacles. Recall represents the proportion of all actual obstacles successfully identified by the model. The F1 score is the harmonic mean of precision and recall. Precision represents the proportion of points identified as obstacles by the model that are actually obstacles.

[0241] For example, nine common obstacles, including trees, buildings, water bodies, and ground surfaces, were identified and their risks assessed in detail. Each obstacle underwent multiple model tests and data validations. Specific results are shown below. Figure 3 As shown in Table 2:

[0242] Table 2: Model Evaluation Index Results

[0243]

[0244] Combination Figure 3 The accuracy of the model is analyzed in Table 2:

[0245] ① Regarding tree identification accuracy:

[0246] For tall trees (such as eucalyptus and pine), the model generally achieves high recognition accuracy, with an accuracy rate exceeding 95%. This is because the trees have distinctive features, and especially with the support of 3D point cloud data and thermal imaging images, the model can accurately distinguish the differences between the trees and their surrounding environment.

[0247] For smaller or shorter trees (such as bamboo and tung trees), the accuracy rate of identification is slightly lower, but still remains above 90%, indicating that the model has a good ability to identify trees of different sizes.

[0248] ②Accuracy of identifying buildings and water bodies:

[0249] The building identification accuracy was 93% and the recall rate was 90%, reflecting that the model can effectively identify buildings around transmission lines and conduct accurate risk assessments. The size and shape characteristics of the buildings make them easily distinguishable from other obstacles.

[0250] The accuracy rate for water body identification was 92%, and the recall rate was 88%. Water bodies have relatively simple morphological characteristics and usually have obvious reflective properties, so the model can efficiently identify water body areas.

[0251] ③ Ground and other obstacles:

[0252] The accuracy rate for ground-based obstacle identification is relatively low (90%) because the ground typically lacks obvious obstacle features, making the model susceptible to noise interference when classifying obstacles. However, since the ground usually does not pose a direct threat to power transmission lines, the risk assessment focuses on obstacles such as trees and buildings.

[0253] For other small obstacles (such as bushes), the model's accuracy remains above 85%.

[0254] Overall, the model performs excellently in recognizing all types of obstacles. It demonstrates high accuracy in identifying obstacles such as trees, buildings, and water bodies, with accuracy exceeding 90%. Particularly noteworthy is its accuracy for identifying large trees like eucalyptus and pine, which surpasses 97%. The model's precision and recall are both outstanding, proving its effectiveness in handling complex real-world scenarios. While factors such as wind speed, tree height and size, and point cloud data density have some impact on model accuracy, the optimized BDF Bayesian data fusion method effectively integrates multiple data sources, improving robustness and accuracy. Furthermore, the model comprehensively considers the spatial distribution, size, and meteorological conditions of different obstacle types, providing accurate support for risk assessment of transmission lines. It can also be flexibly applied in various types of transmission line environments, adapting to environmental monitoring needs of varying scales and complexities, and providing effective technical support for the safe maintenance of transmission lines.

[0255] Simultaneously, it can also provide early warnings of potential dangerous obstacles (such as fallen trees, tall buildings near power lines, etc.). The entire obstacle identification and risk assessment process can be automated, greatly improving work efficiency. It is no longer necessary to inspect every area of ​​the transmission line one by one; instead, targeted inspections and handling can be carried out based on the system's automatic identification and assessment results. This not only saves a significant amount of labor costs but also improves the efficiency and accuracy of monitoring.

[0256] Example 3 illustrates a schematic scheme for a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion. It should be noted that the technical solution of this system for three-dimensional fusion of multi-model transmission line channels based on BDF Bayesian data fusion is based on the same concept as the aforementioned method for three-dimensional fusion of multi-model transmission line channels based on BDF Bayesian data fusion. Details not described in detail in this embodiment can be found in the description of the aforementioned method for three-dimensional fusion of multi-model transmission line channels based on BDF Bayesian data fusion.

[0257] This embodiment also provides another multi-model transmission line channel three-dimensional fusion system based on BDF Bayesian data fusion, including:

[0258] The acquisition module is used to acquire 3D data, image data, and meteorological data of the environment surrounding multi-source transmission lines;

[0259] The fusion module is used to fuse registered multi-source data using the BDF Bayesian fusion strategy.

[0260] The fusion model construction module is used to generate a three-dimensional fusion model of the transmission line channel based on the fused data;

[0261] The recognition module is used to construct a three-dimensional mesh model based on the three-dimensional fusion model and to identify objects through deep learning algorithms to obtain various obstacles in the transmission line channel;

[0262] The scoring module is used to combine the geometric features of obstacles with meteorological data to construct a multi-factor weighted scoring model, obtain the risk score of the obstacle, and conduct hazard analysis.

[0263] This embodiment also provides a computer device applicable to a three-dimensional fusion of multi-model transmission line channels based on BDF Bayesian data fusion, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the three-dimensional fusion method of multi-model transmission line channels based on BDF Bayesian data fusion as proposed in the above embodiment.

[0264] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as proposed in the above embodiment.

[0265] The storage medium proposed in this embodiment and the method for implementing a three-dimensional fusion of multi-model transmission line channels based on BDF Bayesian data fusion proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0266] From the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0267] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion, characterized in that, include: Acquire 3D data, image data, and meteorological data of the environment surrounding multi-source transmission lines; The registered multi-source data are fused using the BDF Bayesian fusion strategy. A three-dimensional fusion model of the transmission line corridor is generated based on the fused data; Based on the 3D fusion model, a 3D mesh model is constructed, and object recognition is performed through deep learning algorithms to obtain various obstacles in the power transmission line channel. By combining the geometric features of obstacles with meteorological data, a multi-factor weighted scoring model is constructed to obtain the risk score of obstacles and conduct hazard analysis.

2. The three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in claim 1, characterized in that, The registered multi-source data will be fused using a BDF Bayesian fusion strategy, including: Define a state vector that describes the state of the environment; Based on the defined state vector, the prior probabilities of each type of data are constructed. Based on the spatial location, category, size and ambient temperature of each obstacle, a probability distribution is defined for each obstacle type to obtain the prior probability model of the corresponding obstacle. Acquire observation data from various sensors of the corresponding obstacles, including the spatial coordinates and reflection intensity information of each point in the three-dimensional data, the object appearance information in the image data, thermal imaging data, and meteorological data; Each sensor's observation data for the corresponding obstacle is defined as its own likelihood function. The likelihood functions of each obstacle are weighted to obtain a joint likelihood function. The joint likelihood function is then combined with the prior probability model of the obstacle, and the posterior probability is calculated using Bayes' theorem. The state vector of the corresponding obstacle is updated according to the posterior probability, thereby realizing the fusion of multi-source data of the corresponding obstacle.

3. The three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in claim 2, characterized in that, A 3D fusion model of the transmission line corridor is generated based on the fused data, including: By integrating data from multiple sources, integrated point cloud data is obtained. By combining the updated state vectors of the corresponding obstacles with the integrated point cloud data, a three-dimensional model of all obstacles is obtained, generating a three-dimensional fusion model of the power line channel.

4. The three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in claim 3, characterized in that, Based on the 3D fusion model, a 3D mesh model is constructed, including: Based on the point cloud data in the 3D fusion model, a normal vector is calculated for each point in the point cloud for surface reconstruction; The Poisson reconstruction algorithm is used to transform the point cloud into a continuous surface of obstacles; Map the attributes in the fusion model to the obstacle reconstruction surface; The reconstructed continuous surface model is transformed into a three-dimensional mesh model by applying Delaunay triangulation, generating a polygonal mesh. The attributes in the fusion model are assigned to the mesh vertices to obtain a 3D mesh model.

5. The three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in claim 4, characterized in that, Object recognition was performed using deep learning algorithms to identify various obstacles in the power transmission line corridor, including: Extract the vertices of the 3D mesh model into point cloud form and retain the point cloud information; Analyze the relationships between points using local feature analysis of point cloud computing to extract geometric features; extract color and edge features from image data; extract temperature features from thermal imaging data; Based on the extracted features, multimodal feature vectors are constructed to form a point cloud dataset; The PointNet++ algorithm is used to classify and segment feature point clouds to identify obstacle types; By using RGB images and thermal imaging data and fusing multimodal information, the obstacle type and probability are output.

6. The three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in claim 5, characterized in that, By combining obstacle geometric features with meteorological data, a multi-factor weighted scoring model is constructed to obtain the obstacle risk score, including: Obtain the type of obstacle; Calculate the shortest distance between the obstacle and the power transmission line; Calculate the height and dimensions of the obstacle; Acquire meteorological factors to assess the impact of meteorological conditions on obstacle stability; By using a weighted scoring method, the overall risk of obstacles in each area is scored, resulting in an obstacle risk score.

7. The three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in claim 6, characterized in that, Also includes: Accuracy analysis of the multi-factor weighted scoring model was conducted, specifically including: All types of obstacles are evaluated using four metrics: accuracy, recall, F1 score, and precision. The accuracy of the model is evaluated based on the numerical range of each indicator on different types of obstacles.

8. A three-dimensional fusion system for multi-model transmission line channels based on BDF Bayesian data fusion, using the method described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire 3D data, image data, and meteorological data of the environment surrounding multi-source transmission lines; The fusion module is used to fuse registered multi-source data using the BDF Bayesian fusion strategy. The fusion model construction module is used to generate a three-dimensional fusion model of the transmission line channel based on the fused data; The recognition module is used to construct a three-dimensional mesh model based on the three-dimensional fusion model and to identify objects through deep learning algorithms to obtain various obstacles in the transmission line channel; The scoring module is used to combine the geometric features of obstacles with meteorological data to construct a multi-factor weighted scoring model, obtain the risk score of the obstacle, and conduct hazard analysis.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the multi-model transmission line channel three-dimensional fusion method based on BDF Bayesian data fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a processor, implement the steps of the three-dimensional fusion method for multi-model transmission line channels based on BDF Bayesian data fusion as described in any one of claims 1 to 7.

Images