A 3D imaging system and measurement method for electric power drone survey

Abstract

The present invention discloses a 3D imaging system and measurement method for electric power drone survey, which relates to the field of drone electric power survey. A drone equipped with visible light and thermal imaging cameras is used to move in the air to survey and collect visible light image data and thermal imaging image data of the power network, identify network nodes and connecting lines through a deep learning algorithm, and construct a two-dimensional topological map; then, the temperature data in the thermal imaging image data is combined to assign weights to the nodes, and an improved genetic algorithm is used to plan the inspection path, so as to achieve priority inspection of high-risk nodes while reducing the total inspection distance; during the inspection process according to the inspection path, the three-dimensional coordinates of the nodes and the visible light and thermal imaging images are obtained through the drone sensor, and the three-dimensional topological map of the power network is constructed and the faulty nodes are marked. The present invention has the advantages of quickly generating a three-dimensional topological map, preferentially detecting faulty nodes, and significantly reducing the total inspection distance.

Description

Technical Field

[0001] The present invention relates to the field of UAV power survey, and more specifically, to a 3D imaging system and a measurement method for power UAV survey. Background Art

[0002] With the rapid development of drone technology, the power industry has gradually begun to use drones for exploration and inspection of power networks. Drones are equipped with high-resolution sensors, such as visible light cameras, thermal imaging cameras, and lidar, which can obtain real-time data of power facilities in a relatively short period of time, especially in high-voltage towers, transmission lines, and remote areas that are difficult to reach, replacing traditional manual inspection methods and significantly improving the efficiency and safety of inspections. The flexibility of drones has led to their widespread use in the operation and maintenance of power systems, especially in the regular inspection of power networks, fault diagnosis, and equipment health monitoring, showing great advantages.

[0003] Although drone technology is widely used in power inspections, there are still significant challenges in the exploration and modeling of power networks. Power networks often involve complex terrain environments and variable line layouts. For example, in some power networks in mountainous areas, forests, or urban high-rise buildings, the obstruction of power facilities and terrain fluctuations will affect the comprehensiveness and accuracy of data collection, resulting in a long 3D modeling process, and the modeling speed cannot keep up with the changes in power facilities and networks.

[0004] In addition, the existing exploration and modeling lacks a global perspective and cannot fully monitor the power network. Before the modeling is completed, it is often impossible to quickly understand the overall structure of the power network, making it difficult to quickly discover potential fault points or risk nodes in a timely manner.

[0005] To this end, a new technical solution is needed that can quickly obtain the overall picture of the power network and the three-dimensional topological model during the exploration process, and can detect potential faults of power facilities as soon as possible in the early stages. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a 3D imaging system and a measurement method for electric power UAV survey to solve the problems mentioned in the background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A 3D imaging system for electric power drone survey, characterized by comprising:

[0009] A drone configured with a visible light camera and a thermal imaging camera for power network survey; the drone is used to move in the air to survey the power network of an area;

[0010] An image recognition module is used to receive the two-dimensional visible light image data captured by the visible light camera, and identify the nodes and connecting lines in the power network through an image recognition algorithm, thereby forming a two-dimensional topological structure diagram;

[0011] A temperature monitoring module is used to receive the thermal imaging data captured by the thermal imaging camera, obtain the temperature data of each node in the power network, and assign a weight to each node according to the temperature data;

[0012] The path planning module is used to plan the inspection path of the UAV based on the two-dimensional topological structure diagram of the power network and the node weights using an optimization algorithm. The constraints of the optimization algorithm include a short total inspection distance and priority access to high-weight nodes.

[0013] A three-dimensional positioning module, used to obtain the three-dimensional coordinates of each node through sensors on the drone during the drone's inspection process according to the inspection path;

[0014] The fault identification module is used to determine whether a node is faulty based on the node images taken by the visible light camera and the thermal imaging camera during the drone inspection process; manual judgment or artificial intelligence judgment can be used, and the artificial intelligence judgment can use an image recognition algorithm or a multimodal generative large model.

[0015] The 3D modeling module constructs a 3D topology map through the 3D coordinates of each node and marks the fault conditions of each node.

[0016] In a specific embodiment, the nodes include transformer substations, distribution substations, power plants, control centers, and grid intersections; the connecting lines include transmission lines and cables.

[0017] In a specific embodiment, the image recognition module uses an image recognition algorithm based on deep learning. In another embodiment, manual labeling is even possible.

[0018] In a specific embodiment, the weight assigned to each node is Satisfies the following formula:

[0019] ;

[0020] in:

[0021] is the temperature of node i;

[0022] Rating safe temperatures for power nodes;

[0023] is the preset temperature alarm threshold, where > .

[0024] In a specific embodiment, the path planning module uses an improved genetic algorithm to generate an inspection path through the following steps:

[0025] 1) Use the node sequence of the path as a chromosome to construct the initial population;

[0026] 2) Set the fitness function to:

[0027] ;

[0028] in:

[0029] F is the fitness value;

[0030] S is the total length of the path;

[0031] For Node Sequential ranking in the path;

[0032] is the weight adjustment coefficient;

[0033] n is the number of nodes in the power network;

[0034] 3) performing genetic operations, including sequential crossover and mutation, and selecting a new generation of population based on the fitness function;

[0035] 4) Iterate step 3) until the fitness is stable or reaches the preset maximum number of generations, and output the optimal inspection path.

[0036] The system of the present invention further comprises:

[0037] Wireless communication module, used for data transmission between the UAV and the ground control station;

[0038] The data processing center is used to receive, store and process data from drones to generate a three-dimensional model of the power network.

[0039] The present invention also provides a measurement method for 3D imaging of electric power drone survey, comprising the following steps:

[0040] Step 1: A drone moves in the air to survey the power network of a region, wherein the drone is equipped with a visible light camera and a thermal imaging camera for power network survey;

[0041] Step 2: receiving the two-dimensional visible light image data transmitted by the drone, and identifying the nodes and connecting lines in the power network through an image recognition algorithm, thereby forming a two-dimensional topological structure diagram;

[0042] Step 3: receiving the thermal imaging data transmitted by the drone, obtaining the temperature data of each node in the power network, and assigning a weight to each node according to the temperature data;

[0043] Step 4: According to the two-dimensional topological structure diagram of the power network and the node weights, the inspection path of the drone is planned using the optimization algorithm, where the constraints of the optimization algorithm include short total inspection distance and priority access to high-weight nodes;

[0044] Step 5: The drone inspects the power network according to the inspection path, and obtains the three-dimensional coordinates of each node through sensors on the drone;

[0045] Step 6: Determine whether the node is faulty based on the node images taken by the drone;

[0046] Step 7: Construct a 3D topology map using the 3D coordinates of each node and mark the fault condition of each node.

[0047] More specifically, in step 2, the image recognition module adopts an image recognition algorithm based on deep learning.

[0048] More specifically, the weight assigned to each node Satisfies the following formula:

[0049] ;

[0050] in:

[0051] is the temperature of node i;

[0052] Rating safe temperatures for power nodes;

[0053] is the preset temperature alarm threshold, where > ;

[0054] More specifically, the path planning module uses an improved genetic algorithm to generate an inspection path through the following steps:

[0055] 1) Use the node sequence of the path as a chromosome to construct the initial population;

[0056] 2) Set the fitness function to:

[0057] ;

[0058] in:

[0059] F is the fitness value;

[0060] S is the total length of the path;

[0061] For Node Sequential ranking in the path;

[0062] is the weight adjustment coefficient;

[0063] n is the number of nodes in the power network;

[0064] 3) performing genetic operations, including sequential crossover and mutation, and selecting a new generation of population based on the fitness function;

[0065] 4) Iterate step 3) until the fitness is stable or reaches the preset maximum number of generations, and output the optimal inspection path.

[0066] The advantages of the present invention over the prior art are as follows:

[0067] The present invention integrates the visible light camera and thermal imaging camera carried by the drone, as well as advanced image recognition and three-dimensional modeling algorithms, so that the system can quickly construct a three-dimensional topological map of the power network. During the flight, the two-dimensional images collected by the drone are processed by a deep learning algorithm to quickly identify the various nodes and connecting lines in the power network to form a two-dimensional topological structure; then, the three-dimensional coordinates of each node are obtained through the sensors on the drone, and integrated and calculated in the data processing center, and finally a high-precision three-dimensional model is constructed. This process not only greatly shortens the time of on-site inspection and data processing, but also improves the automation level of image and data processing, making the entire three-dimensional topological map generation process efficient and real-time.

[0068] The present invention assigns weights to each node through analysis of thermal imaging data, and can prioritize the location and detection of nodes that may have a risk of failure. Specifically, the temperature monitoring module obtains the temperature data of each node in the power network in real time, and weights the nodes according to the preset safety temperature and alarm threshold; in the path planning process, the improved genetic algorithm makes full use of this weight information to ensure that high-risk (high-weight) nodes are visited first in the inspection path. In this way, the system can focus on checking key nodes that may have abnormalities in the shortest time, effectively improving the efficiency of early fault identification and maintenance, and reducing the risk of accidents.

[0069] The present invention optimizes the inspection path through an improved genetic algorithm, which not only meets the requirement of giving priority to high-weight nodes, but also greatly reduces the total inspection distance of the UAV. The algorithm adopts strategies such as node sequence chromosome encoding, sequential crossover and probability mutation in the design to fine-tune the entire path, thereby optimizing the path distance while ensuring priority inspection of high-risk nodes. The optimized path can effectively reduce the flight time and energy consumption of the UAV and improve the overall efficiency of the inspection operation.

[0070] In summary, the present invention provides a 3D imaging system and measurement method for electric power drone survey, which can quickly generate a three-dimensional topological map of the power network, preferentially detect high-risk nodes that may have faults, and significantly reduce the total inspection distance through optimization algorithms, providing an efficient and accurate technical solution for the intelligent monitoring and maintenance of the power network. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is the overall logic block diagram of the system of the present invention;

[0072] Figure 2 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0073] The specific implementation of the present invention will be described below in conjunction with the accompanying drawings.

[0074] like Figure 1 The overall logic block diagram of the system of the present invention is shown, including:

[0075] A drone equipped with a visible light camera and a thermal imaging camera for power network survey; the drone is used to move in the air to survey the power network of an area;

[0076] An image recognition module is used to receive the two-dimensional visible light image data transmitted by the drone and identify the nodes and connecting lines in the power network through an image recognition algorithm, thereby forming a two-dimensional topological structure diagram;

[0077] The temperature monitoring module is used to receive the thermal imaging data transmitted by the drone, obtain the temperature data of each node in the power network, and assign weights to each node based on the temperature data;

[0078] The path planning module is used to plan the inspection path of the UAV based on the two-dimensional topological structure diagram of the power network and the node weights using an optimization algorithm. The constraints of the optimization algorithm include a short total inspection distance and priority access to high-weight nodes.

[0079] The three-dimensional positioning module is used to obtain the three-dimensional coordinates of each node through the sensors on the drone during the inspection process according to the inspection path;

[0080] The fault identification module determines whether a node is faulty based on the node images taken by the drone;

[0081] The 3D modeling module constructs a 3D topology map through the 3D coordinates of each node and marks the fault conditions of each node.

[0082] Therefore, the present invention actually discloses a measurement method for 3D imaging of power drone survey, such as Figure 2 The general flow chart includes the following steps:

[0083] Step 1: A drone moves in the air to survey the power network of a region, wherein the drone is equipped with a visible light camera and a thermal imaging camera for power network survey;

[0084] Step 2: Receive the two-dimensional visible light image data transmitted by the drone, and identify the nodes and connecting lines in the power network through the image recognition algorithm, thereby forming a two-dimensional topological structure diagram;

[0085] Step 3: Receive the thermal imaging data transmitted by the drone, obtain the temperature data of each node in the power network, and assign weights to each node based on the temperature data;

[0086] Step 4: According to the two-dimensional topological structure diagram of the power network and the node weights, the inspection path of the drone is planned using the optimization algorithm, where the constraints of the optimization algorithm include short total inspection distance and priority access to high-weight nodes;

[0087] Step 5: The drone inspects the power network according to the inspection path and obtains the three-dimensional coordinates of each node through the sensors on the drone;

[0088] Step 6: Determine whether the node is faulty based on the node images taken by the drone;

[0089] Step 7: Construct a 3D topology map using the 3D coordinates of each node and mark the fault condition of each node.

[0090] Among them, in the fault identification module and step 6, two methods can be used to judge the fault. The simplest one is manual judgment. Professionals can rely on their own experience and expertise to identify abnormal conditions of equipment, such as cracks, corrosion, overheating, etc. by viewing the node pictures taken by the drone. This method is intuitive and has high accuracy. It can be used when the number of nodes is small.

[0091] In addition, automatic image recognition (including recognition of visible light images and thermal imaging images) can be used.

[0092] More specifically, deep learning algorithms are used to automatically analyze node images (including visible light images and thermal imaging images) taken by drones to identify abnormal conditions of equipment. Specifically, the system first pre-processes the image, such as denoising and contrast enhancement, and then uses deep learning models such as convolutional neural networks to extract image features for classification and judgment.

[0093] A simpler method is to directly use the current multimodal generative large model. By inputting pictures and questions, the large model can determine the abnormality of the equipment. For example, GPT4o, DeepSeek, etc. Compared with manual judgment, the image recognition method has the advantages of high efficiency, accuracy, and scalability. It can process a large amount of image data, reduce human intervention, and improve inspection efficiency and accuracy. In addition, the image recognition method can also be combined with other technologies, such as thermal imaging, lidar, etc., to further improve the comprehensiveness and accuracy of fault detection.

[0094] Among them, nodes may include substations, distribution stations, power plants, control centers and grid intersections; connecting lines include transmission lines and cables.

[0095] In one embodiment, each node is assigned a weight Satisfies the following formula:

[0096] ;

[0097] in:

[0098] is the temperature of node i;

[0099] Rating safe temperatures for power nodes;

[0100] is the preset temperature alarm threshold, where > ;

[0101] From this formula we can see that when When the weight , indicating that no priority testing is required; when When the weight 0, and The larger the value, the weight The larger the value of is, the greater the priority is. ≥ hour, ≥1.

[0102] The so-called high-weight nodes can be defined as 0 node.

[0103] The path planning module can use an improved genetic algorithm to generate an inspection path through the following steps:

[0104] 1) Use the node sequence of the path as a chromosome to construct an initial population; the initial population can be a random arrangement generated randomly, or a greedy path based on weights, or a combination of the above two. In the greedy path, high-weight nodes are connected first;

[0105] That is to say: each inspection path is represented by a fixed-length permutation code, each position in this permutation corresponds to a node in the power grid, and the entire sequence reflects the order in which the drone actually visits the nodes during the inspection. In other words, if there are N nodes in the power grid, then each chromosome can be regarded as an array of length N, each array element identifies a specific node, and the order of the nodes in the array is exactly the order of the inspection path.

[0106] The initial population can be a randomly generated arrangement, or it can include a path constructed by a greedy algorithm using weight information. When constructing this path, high-weight nodes are connected first, ensuring that these key nodes can be visited in a better order. Such an initial population has both random diversity and heuristic strategies, which provides a better starting point for subsequent genetic operations and helps the algorithm quickly converge to the optimal or near-optimal path that meets the requirements of short total inspection distance and priority access to high-weight nodes.

[0107] 2) The fitness function can be set as:

[0108] ;

[0109] in:

[0110] F is the fitness value;

[0111] S is the total length of the path;

[0112] For Node The sequential ranking in the path, the sequential ranking starts from 1;

[0113] is the weight adjustment coefficient, which is used to balance the total distance and priority;

[0114] n is the number of nodes in the power network;

[0115] Analyzing this formula, we can see that S The smaller the fitness value FThe larger it is, the greater the probability that the path will be retained in the genetic algorithm. This is in line with expectations, because we want the total length of the drone inspection to be as short as possible, so that it is more efficient;

[0116] Similarly, the weight term The smaller the fitness value F If you want to make the weight item smaller, you need to make the weight The larger value in corresponds to a smaller order ranking. For example, multiply the order value 1 by the largest weight, multiply the order value 2 by the smaller weight, and so on. The weight items after all the superposition will be as small as possible, so as to ensure that the nodes that may have faults can be inspected preferentially and the nodes can be discovered as early as possible.

[0117] As the simplest example, assuming that only one node has a possible fault, that is, the weights of other nodes are all 0, then in the best path, the node with the fault must be at order 1, so that the fitness function can be as large as possible.

[0118] The subsequent steps are all conventional steps in genetic algorithms:

[0119] 3) Perform genetic operations, including sequential crossover and mutation, and select the new generation population based on the fitness function;

[0120] More specifically, at this stage, each individual in the current population (i.e., the chromosome representing the inspection path) is first scored according to the pre-designed fitness function. Then, the following specific operations are used to generate a new generation of population:

[0121] Crossover operation, using sequential crossover:

[0122] In specific implementation, a continuous gene substring (for example, containing multiple continuous nodes and preferably containing high-weight nodes) is randomly selected from the two parent chromosomes, and this substring is directly copied to the corresponding position of the offspring. Then, according to the order of appearance of the remaining nodes in the other parent chromosome, the vacant positions are filled in sequence to ensure that the offspring chromosome maintains a legal node arrangement. This operation not only retains the structural advantages of the key area (continuous area of ​​high-weight nodes), but also mixes the information of other parts of the parent.

[0123] Mutation operation:

[0124] The offspring chromosomes are locally randomly adjusted with a preset lower mutation probability (e.g. 0.05 or 5%). The specific method can be to use node exchange or reversal subsequences, focusing on appropriately perturbing the access order around high-weight nodes. This can not only jump out of the local optimum, but also maintain a good path structure overall.

[0125] Select an action:

[0126] Based on the score calculated by the fitness function, methods such as roulette selection, tournament selection or elite retention strategy are used to select individuals with higher fitness from the candidate solutions generated by crossover and mutation to enter the next generation of population. The elite retention strategy ensures that the best individuals in each generation will not be lost due to random operations, while maintaining the diversity of the overall population and promoting the algorithm to converge to the global optimal solution.

[0127] Through the above crossover, mutation and selection operations, a new generation of population is gradually formed, and the excellent access order of key nodes can be effectively retained and transmitted.

[0128] 4) Termination conditions:

[0129] Iterate step 3) until the fitness is stable or reaches the maximum number of generations, and output the optimal inspection path. More specifically:

[0130] After each generation of genetic operations is completed, it is necessary to determine whether the algorithm termination conditions are met. Commonly used specific methods include:

[0131] Set the change range of the best individual or average fitness in the population for several consecutive generations (for example, 10 consecutive generations) to be lower than a predefined threshold (for example, the change rate is less than 0.1%). When it is detected that the overall fitness of the population tends to be stable, it is considered that the algorithm has basically converged and the iteration can be terminated.

[0132] At the same time, a maximum number of generations is set (for example, 500 or 1000 generations) to prevent unlimited iterations when the problem is more complex or the search space is larger. After reaching this number of generations, regardless of whether the fitness of the population is still changing, the iteration is forced to end, and the individual with the best fitness in the current population is output as the final inspection path.

[0133] By combining the two termination strategies of fitness stability and maximum algebra, it can be ensured that the algorithm can be terminated in time when the ideal convergence state is reached, and can also avoid excessive calculation time caused by extreme situations. In this way, the generation and selection of a new generation of populations will continue to drive the algorithm towards the optimal solution that not only ensures the shortest total inspection path but also gives priority to visiting high-risk nodes, until the termination conditions are met and the optimal inspection path is output.

[0134] In addition, the image recognition module of the present invention adopts an image recognition algorithm based on deep learning. Image recognition algorithm is a relatively common field, and its basic processes are not much different. Here we only roughly describe the algorithm description for this scenario, but this is only under a specific embodiment and does not mean that only these methods can be used. Even to some extent, nodes and connecting lines can be manually marked, so that there is no need to train the image recognition algorithm. For image recognition algorithm:

[0135] First, we can build a high-quality dataset, including a large number of power network images from the perspective of drones. These images need to be carefully annotated: nodes are usually annotated as detection boxes or category labels in semantic segmentation, and connecting lines need to be annotated as lines or segmentation masks. After data annotation, data enhancement techniques (such as rotation, scaling, color transformation, etc.) can be used to expand the diversity of the dataset to cope with various complex scenarios that may be encountered in practical applications.

[0136] In terms of model architecture, a multi-task learning framework can be used to combine target detection with semantic segmentation. For example, for node detection, existing target detection networks (such as Faster R-CNN, YOLO series or RetinaNet) can be used to locate and identify different types of nodes. Considering that the size, shape and background environment of power network nodes may vary greatly, feature pyramid networks (FPN) can be used to process multi-scale information, and self-attention mechanisms or context information fusion modules can be used to improve the accuracy of detection. For the extraction of connecting lines, a parallel semantic segmentation branch can be designed or a strategy combining edge detection with image segmentation can be adopted, such as using networks such as U-Net, DeepLab or HRNet to perform pixel-level segmentation on power lines to distinguish between background and lines; image post-processing steps can also be introduced, such as extracting straight line features based on Hough transform, further refining the detection results, and combining the probability map output by the network with traditional image processing methods to improve the continuity and integrity of the line.

[0137] During the training process, the design of the loss function needs to take into account the characteristics of different tasks. For the target detection branch, classification loss (such as cross entropy loss) and bounding box regression loss (such as Smooth L1 or IoU Loss) can be used; for the semantic segmentation branch, pixel-level cross entropy loss is usually used, and Dice Loss or Focal Loss can be added to deal with the problem of category imbalance. In order to make the two branches promote each other on the shared feature extraction layer, a suitable loss weighting strategy can be set during joint training, so that the model can focus on the precise positioning of the nodes and accurately capture the structural information of the lines during the optimization process. It should be noted that since power lines are usually thin and easily disturbed by background noise, special emphasis should be placed on the labeling of thin line samples in the training data, and multi-scale supervision strategies may be used to enhance the capture of details.

[0138] After the model training is completed, a post-processing module needs to be designed to improve the overall performance of the system. For node detection results, non-maximum suppression (NMS) can be used to remove duplicate detection frames; for line segmentation results, morphological operations (such as expansion and corrosion) can be used to smooth the prediction results, and then the connection algorithm (such as the shortest path algorithm based on graph theory) can be used to connect the scattered line segments into a continuous network. At the same time, prior knowledge (such as the characteristics that power lines are usually continuous and as straight as possible or slowly curved) can be used to reasonably constrain the detection results. Combined with the detection results of nodes and lines, a graph model can also be constructed, with the detected nodes as the vertices of the graph and the lines as the edges. Through graph matching, topology verification and other methods, it is ensured that the final output power grid structure is consistent with the actual physical network. Real-time performance and resource usage need to be considered when deploying the system. Technologies such as model compression, quantization or knowledge distillation can be used to reduce the number of model parameters, improve the inference speed, and achieve fast response in edge devices or cloud environments. The entire system should also have a certain degree of robustness and be able to handle image quality degradation caused by factors such as lighting, weather, and terrain. By continuously collecting field data, iteratively updating the model, and combining feedback from domain experts, we ultimately built a deep learning system that can stably and accurately extract power network structure information from drone images.

[0139] In a further embodiment, the system not only has basic data acquisition and image processing capabilities, but also adds a wireless communication module and a data processing center to achieve efficient data interaction and three-dimensional model construction between the UAV and the ground control station. The wireless communication module adopts advanced wireless transmission protocols (such as LTE, 5G or dedicated wireless links) to ensure that the high-resolution images, sensor data and location information collected by the UAV during flight can be transmitted to the ground control station in real time and stably. This module not only supports two-way transmission of data and realizes instant feedback of task instructions and status information, but also has the functions of anti-interference and adaptive bandwidth allocation to ensure that the stability of the communication link can be maintained in complex environments. At the same time, as the core node of the system, the data processing center is responsible for receiving a large amount of data transmitted from the wireless communication module, and classifying, storing and processing it. The data processing center is embedded with a dedicated image and signal processing server. By integrating a variety of image processing algorithms, three-dimensional reconstruction algorithms and deep learning models, the image data of the power network is processed in real time, key grid nodes and connecting line information are extracted, and a high-precision three-dimensional model of the power network is generated based on these data. This three-dimensional model can not only intuitively display the overall structure of the power network, but also can be used for subsequent fault diagnosis, maintenance planning and real-time monitoring. To meet the needs of massive data processing, the data processing center supports distributed storage and parallel computing, and also provides functions such as data backup, logging and security encryption to ensure that the data transmission, storage and processing of the entire system are both efficient and safe. Through the collaborative work of the wireless communication module and the data processing center, the system realizes the instant transmission of drone-collected data and the rapid generation of high-quality three-dimensional models, providing solid technical support for the intelligent monitoring and management of power networks.

[0140] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A 3D imaging system for electric power drone survey, characterized in that: include: A drone configured with a visible light camera and a thermal imaging camera for power network survey; the drone is used to move in the air to survey the power network of an area; An image recognition module is used to receive the two-dimensional visible light image data captured by the visible light camera, and identify the nodes and connecting lines in the power network through an image recognition algorithm, thereby forming a two-dimensional topological structure diagram; A temperature monitoring module is used to receive the thermal imaging data captured by the thermal imaging camera, obtain the temperature data of each node in the power network, and assign a weight to each node according to the temperature data; The path planning module is used to plan the inspection path of the UAV based on the two-dimensional topological structure diagram of the power network and the node weights using an optimization algorithm. The constraints of the optimization algorithm include a short total inspection distance and priority access to high-weight nodes. A three-dimensional positioning module, used to obtain the three-dimensional coordinates of each node through sensors on the drone during the drone's inspection process according to the inspection path; A fault identification module, used to determine whether a node is faulty according to the node images taken by the visible light camera and the thermal imaging camera during the inspection of the drone; The 3D modeling module constructs a 3D topology map through the 3D coordinates of each node and marks the fault conditions of each node; The path planning module uses an improved genetic algorithm to generate an inspection path through the following steps: 1) Use the node sequence of the path as a chromosome to construct the initial population; 2) Set the fitness function to: ； in: F is the fitness value; S is the total length of the path; For Node Sequential ranking in the path; is the weight adjustment coefficient; is the number of nodes in the power network; The weight assigned to each node; 3) performing genetic operations, including sequential crossover and mutation, and selecting a new generation of population based on the fitness function; 4) Iterate step 3) until the fitness is stable or reaches the preset maximum number of generations, and output the optimal inspection path.

2. The electric power drone survey 3D imaging system according to claim 1, characterized in that: The nodes include transformer substations, distribution substations, power plants, control centers and grid intersections; the connecting lines include transmission lines and cables.

3. The electric power drone survey 3D imaging system according to claim 1, characterized in that: The image recognition module adopts an image recognition algorithm based on deep learning.

4. The electric power UAV survey 3D imaging system according to claim 1, characterized in that: The weight assigned to each node Satisfies the following formula: ； in: is the temperature of node i; Rating safe temperatures for power nodes; is the preset temperature alarm threshold, where > .

5. The electric power drone survey 3D imaging system according to claim 1, characterized in that: The system further comprises: Wireless communication module, used for data transmission between the UAV and the ground control station; The data processing center is used to receive, store and process data from drones to generate a three-dimensional model of the power network.

6. A measurement method for 3D imaging of power UAV survey, characterized in that: The following steps are involved: Step 1: A drone moves in the air to survey the power network of a region, wherein the drone is equipped with a visible light camera and a thermal imaging camera for power network survey; Step 2: receiving the two-dimensional visible light image data transmitted by the drone, and identifying the nodes and connecting lines in the power network through an image recognition algorithm, thereby forming a two-dimensional topological structure diagram; Step 3: receiving the thermal imaging data transmitted by the drone, obtaining the temperature data of each node in the power network, and assigning a weight to each node according to the temperature data; Step 4: According to the two-dimensional topological structure diagram of the power network and the node weights, the inspection path of the drone is planned using the optimization algorithm, where the constraints of the optimization algorithm include short total inspection distance and priority access to high-weight nodes; Step 5: The drone inspects the power network according to the inspection path, and obtains the three-dimensional coordinates of each node through sensors on the drone; Step 6: Determine whether the node is faulty based on the node images taken by the visible light camera and thermal imaging camera on the drone; Step 7: Construct a 3D topology map using the 3D coordinates of each node and mark the fault conditions of each node; The path planning method uses an improved genetic algorithm to generate an inspection path through the following steps: 1) Use the node sequence of the path as a chromosome to construct the initial population; 2) Set the fitness function to: ； in: F is the fitness value; S is the total length of the path; For Node Sequential ranking in the path; is the weight adjustment coefficient; is the number of nodes in the power network; The weight assigned to each node; 3) performing genetic operations, including sequential crossover and mutation, and selecting a new generation of population based on the fitness function; 4) Iterate step 3) until the fitness is stable or reaches the preset maximum number of generations, and output the optimal inspection path.

7. The method for measuring 3D imaging of electric power drone survey according to claim 6 is characterized in that: In step 2, the image recognition method adopts an image recognition algorithm based on deep learning.

8. The method for measuring 3D imaging of electric power drone survey according to claim 6 is characterized in that: The weight assigned to each node Satisfies the following formula: ； in: is the temperature of node i; Rating safe temperatures for power nodes; The preset temperature alarm threshold, where > .

Images