A method, device, equipment and storage medium for identifying utility poles

Abstract

The present invention discloses a method, device, equipment and storage medium for identifying utility poles. The method includes: obtaining orthophoto image data of utility poles at a preset angle; constructing an initial utility pole detection model, and training the initial utility pole detection model by using the orthophoto image data as training samples to obtain a final trained utility pole detection model and its model weights; constructing a three-dimensional model according to the orthophoto image data of utility poles at a preset angle, and obtaining a weight file of the utility poles in the three-dimensional model according to the three-dimensional model and a preset utility pole height threshold; calculating a recognition weight after regression compensation according to the model weights and the weight file, and inputting the recognition weight into a preset utility pole recognition model so that the preset utility pole recognition model loaded with the recognition weight identifies the utility poles. The present invention solves the technical problems of low accuracy and efficiency of utility poles in orthophoto images in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of computer recognition technology, and particularly to a method, device, equipment and storage medium for identifying utility poles. Background Art

[0002] Utility poles play a crucial role in the infrastructure of modern cities and rural areas, supporting power lines, communication lines and other important public facilities. As China's power lines and communication networks continue to develop, the number of utility poles is also increasing. Currently, most of the crashes of plant protection drones are due to untimely obstacle avoidance, especially obstacles such as wires and utility poles, in addition to improper human operation or sudden machine failures.

[0003] Currently, traditional computer vision methods face the following challenges in identifying utility poles: 1. Regarding complex backgrounds, utility poles are usually located in complex backgrounds and may be blocked by trees, buildings or other debris, making it a complex task to accurately extract utility poles from images; 2. Regarding lighting changes, the appearance of utility poles may be affected by natural light, and images taken at different times and under different weather conditions may exhibit different brightness and contrast, which poses difficulties for identification; 3. Regarding the diversity of utility poles, the shapes and materials of utility poles are diverse, and sometimes they may blend in with the surrounding environment, making their detection and classification more complex; 4. Regarding detection speed, in some application scenarios, it is necessary to quickly and accurately detect utility poles, such as for urban planning or power line maintenance, so detection speed is also an important consideration.

[0004] Therefore, there is an urgent need for a method that can improve the accuracy and efficiency of utility pole identification under orthophotos. Summary of the Invention

[0005] The present invention provides a method, device, equipment and storage medium for identifying utility poles to solve the technical problem of low accuracy and efficiency of utility poles under orthophotos in the prior art.

[0006] To solve the above technical problem, an embodiment of the present invention provides a method for identifying utility poles, including:

[0007] Obtain orthophoto image data of a utility pole at a preset angle;

[0008] Construct an initial utility pole detection model, and train the initial utility pole detection model by using the orthophoto image data as training samples to obtain a final trained utility pole detection model and its model weights;

[0009] Construct a 3D model based on the orthophoto image data of the utility pole at a preset angle, and obtain the weight file of the utility pole in the 3D model according to the 3D model and the preset utility pole height threshold;

[0010] Calculate the recognition weight after regression compensation according to the model weight and the weight file, and input the recognition weight into the preset utility pole recognition model, so that the preset utility pole recognition model after loading the recognition weight can recognize the utility pole.

[0011] As a preferred solution, the obtaining of the orthophoto image data of the utility pole at a preset angle specifically includes:

[0012] Obtain the orthophoto image data of the utility pole at a number of preset angles; wherein, the number of the preset angles is at least three.

[0013] As a preferred solution, the constructing of the initial utility pole detection model and training the initial utility pole detection model with the orthophoto image data as the training samples to obtain the finally trained utility pole detection model and its model weight specifically includes:

[0014] Annotate the orthophoto image data of the utility pole at any one of the preset angles to obtain the orthophoto image data of the utility pole after annotation and the unannotated orthophoto image data at this preset angle;

[0015] Construct an initial utility pole detection model, and use the annotated orthophoto image data and the unannotated orthophoto image data as training samples, and then input the training samples into the initial utility pole detection model for training until the loss function of the initial utility pole detection model converges, so as to obtain the trained utility pole detection model and the model weight corresponding to the current state of this utility pole detection model.

[0016] As a preferred solution, the constructing of the 3D model based on the orthophoto image data of the utility pole at a preset angle and obtaining the weight file of the utility pole in the 3D model according to the 3D model and the preset utility pole height threshold specifically includes:

[0017] Integrate the orthophoto image data of the utility pole at all preset angles respectively to obtain the corresponding data set at each preset angle;

[0018] Perform 3D reconstruction in sequence according to the data set at each preset angle to obtain the initial 3D model corresponding to each preset angle;

[0019] According to the preset utility pole height threshold, screen the data of the initial 3D models respectively to obtain the 3D models corresponding to each preset angle and the weight files of the utility poles in each 3D model.

[0020] As a preferred solution, the three-dimensional reconstruction is sequentially performed on the data sets at each preset angle to obtain an initial three-dimensional model corresponding to each preset angle, which specifically includes:

[0021] The image data of the data sets at each preset angle is sequentially separated to collect the point cloud data corresponding to each data set;

[0022] The point cloud data corresponding to each data set is sequentially subjected to three-dimensional modeling, so that during each three-dimensional modeling process, the point cloud data corresponding to the data set is classified and separated to generate point cloud clustering centers corresponding to different categories. Then, the skeleton is drawn and classified and optimized according to the point cloud clustering centers, and the optimized skeleton is subjected to texture integration and texturing to obtain the initial three-dimensional model corresponding to the data set until each data set has a corresponding initial three-dimensional model.

[0023] As a preferred solution, the regression-compensated recognition weight is calculated according to the model weight and the weight file, and the recognition weight is input into a preset electric pole recognition model, so that the preset electric pole recognition model after loading the recognition weight recognizes the electric pole, which specifically includes:

[0024] According to a preset search algorithm, the model weight and the weight file are integrated to obtain the regression-compensated recognition weight;

[0025] The recognition weight is saved to obtain a recognition weight file, and the recognition weight file is sent to the preset electric pole recognition model, so that after the preset electric pole recognition model loads the recognition weight file, it recognizes the electric poles in the orthophoto image obtained in real time.

[0026] As a preferred solution, the preset search algorithm is the sparrow search algorithm; the integration of the model weight and the weight file to obtain the regression-compensated recognition weight specifically includes:

[0027] The model weight and the weight file are initialized to convert the model weight and the weight file into the corresponding mapping position information in the sparrow search algorithm;

[0028] Take the initialized model weights as the discoverer of the sparrow search algorithm, and the initialized weight file as the joiner of the sparrow search algorithm. Add a preset danger constraint condition to the discoverer and the joiner, so that the discoverer updates the mapped position information according to a preset fitness function, and the mapped position information of the joiner is updated according to the updated mapped position information of the discoverer. Until the preset stop condition is reached, output the optimal solution among the current mapped position information of the discoverer and the current mapped position information of the joiner as the recognition weight for regression compensation after weight integration.

[0029] Correspondingly, the present invention also provides an identification device for utility poles, including: an acquisition module, a training module, a three-dimensional module, and a regression module;

[0030] The acquisition module is used to acquire the orthophoto image data of the utility pole at a preset angle;

[0031] The training module is used to construct an initial utility pole detection model, and train the initial utility pole detection model by using the orthophoto image data as training samples to obtain the final trained utility pole detection model and its model weights;

[0032] The three-dimensional module is used to construct a three-dimensional model according to the orthophoto image data of the utility pole at a preset angle, and obtain the weight file of the utility pole in the three-dimensional model according to the three-dimensional model and a preset utility pole height threshold;

[0033] The regression module is used to calculate the recognition weight after regression compensation according to the model weights and the weight file, and input the recognition weight into a preset utility pole recognition model, so that the preset utility pole recognition model loaded with the recognition weight recognizes the utility pole.

[0034] Correspondingly, the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for identifying a utility pole as described in any one of the above.

[0035] Correspondingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method for identifying a utility pole as described in any one of the above.

[0036] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0037] The technical solution of the present invention constructs a pole detection model and its corresponding model weights by obtaining the orthophoto image data of poles at a preset angle, ensuring accurate identification of poles in the orthophoto image and reducing misidentification. Furthermore, a three-dimensional model is constructed through the orthophoto image data of poles at a preset angle, and thus a weight file of the poles in the three-dimensional model is obtained. By integrating the model weights and the weight file, the recognition weights after regression compensation are obtained, and the pole recognition model with the loaded recognition weights is used to identify poles. While maintaining high accuracy, the speed of pole identification is improved, making it applicable to large-scale datasets and real-time applications, avoiding the problem of model overfitting. Moreover, through the model weights of the pole detection model and the weight file of the three-dimensional model, more multi-dimensional model weight setting parameters for pole identification can be provided to improve the model recognition efficiency of poles. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 : is a flowchart of the steps of a method for identifying poles provided by an embodiment of the present invention;

[0039] Figure 2 : is a flowchart of pole identification provided by an embodiment of the present invention;

[0040] Figure 3 : is a structural diagram of an apparatus for identifying poles provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Embodiment 1

[0043] Please refer to Figure 1 , a method for identifying poles provided by an embodiment of the present invention, includes the following steps S101-S104:

[0044] Step S101: Obtain the orthophoto image data of poles at a preset angle.

[0045] As a preferred solution of this embodiment, the obtaining of the orthophoto image data of poles at a preset angle specifically includes:

[0046] Obtain the orthophoto image data of poles at a plurality of preset angles; wherein, the number of the preset angles is at least three.

[0047] In this embodiment, the orthophoto is a remote sensing image with orthographic projection properties. Due to the influence of internal state changes of the sensor (such as optical system distortion, non-linearity of the scanning system, etc.), external states (such as attitude changes), and surface conditions (such as the curvature of the earth, terrain undulation) during imaging, the original remote sensing images all have varying degrees of distortion and distortion. The geometric processing of remote sensing images not only extracts spatial information, such as drawing contour lines, but also resamples the image gray level according to the correct geometric relationship to form a new orthophoto. Therefore, compared with the image data obtained at ordinary angles, the orthophoto data of the utility poles has higher recognition accuracy and success rate for utility poles, avoiding the problem of large recognition errors in the training recognition model.

[0048] In this embodiment, the orthophoto data can be collected by a drone. Preferably, the number of preset angles can be three, namely 45°, 60°, and 90°. The gimbal of the drone will be set at three different angles, namely 45°, 60°, and 90°, so as to capture the orthophoto data of the utility poles from different perspectives.

[0049] Step S102: Construct an initial utility pole detection model, and use the orthophoto data as training samples to train the initial utility pole detection model to obtain the final trained utility pole detection model and its model weights.

[0050] As a preferred solution of this embodiment, the constructing of the initial utility pole detection model, and using the orthophoto data as training samples to train the initial utility pole detection model to obtain the final trained utility pole detection model and its model weights specifically includes:

[0051] Label the orthophoto data of the utility poles at any one of the preset angles to obtain the orthophoto data of the labeled utility poles and the unlabeled orthophoto data at this preset angle; construct an initial utility pole detection model, and use the labeled orthophoto data and the unlabeled orthophoto data as training samples, and then input the training samples into the initial utility pole detection model for training until the loss function of the initial utility pole detection model converges, to obtain the trained utility pole detection model and the model weights corresponding to the current state of this utility pole detection model.

[0052] In this embodiment, the orthophoto data of the utility poles at each preset angle can be used as sample data for training the initial utility pole detection model. However, since using too much data may cause overfitting during modeling and will waste a large amount of computing resources and the operator's time, the embodiment of the present invention only needs to train the model with the orthophoto data of the utility poles at any one of the preset angles, which can save a large amount of time and computing resources compared with the full-angle data training.

[0053] In this embodiment, please refer to Figure 2 , it is possible to train the orthophoto image data of the utility pole at 45°, 60°, or 90°. Preferably, the data with a pan-tilt angle of 90° is used for subsequent processing. At the same time, before image annotation, it is also necessary to perform data cleaning on the orthophoto image data at 90°, that is, based on the clearest and most useful images provided by the pan-tilt angle of 90° for training and analysis, and eliminate the unclear and obviously useless photos for model training.

[0054] In this embodiment, each collected orthophoto image data needs to be annotated. The purpose of annotation is to clearly locate the utility poles in the image. Exemplarily, a labeling tool (such as Label Img) can be used to draw a rectangular box to identify the position of the utility pole, and add corresponding labels to each utility pole, such as "raw" or "mature", where "raw" represents the unannotated orthophoto image data, and "mature" represents the annotated orthophoto image data to indicate the state of the utility pole. All these annotation results will be saved in a TXT file for subsequent training use.

[0055] In this embodiment, an initial utility pole detection model is built. Preferably, the YOLOv8 object detection model can be used as the initial utility pole detection model. The construction of the model may include defining the network architecture, selecting appropriate loss functions, and setting hyperparameters. After the model is built, it is also necessary to adjust the important parameters of the model, such as the number of iterations, optimizer type, learning rate, etc. Further, in the process of training the utility pole detection model, by inputting the images with annotation information into the model, the model will perform backpropagation iteration to optimize the weights and parameters. The training process may continue for multiple iterations to ensure that the model can accurately identify the utility poles. Among them, the training process may also include using the training set and the validation set to evaluate the model performance to avoid overfitting. The training set and the validation set are constructed from the orthophoto image data at the same angle. Finally, select the weights of the last set of model training saved during training as the final model weights, and save the trained model weights for subsequent step operations.

[0056] Step S103: Construct a 3D model based on the orthophoto image data of the utility pole at a preset angle, and obtain the weight file of the utility pole in the 3D model according to the 3D model and the preset utility pole height threshold.

[0057] As a preferred solution of this embodiment, the step of constructing a 3D model based on the orthophoto image data of the utility pole at a preset angle, and obtaining the weight file of the utility pole in the 3D model according to the 3D model and the preset utility pole height threshold specifically includes:

[0058] Integrate the orthophoto image data of the utility poles at all preset angles respectively to obtain the corresponding data sets at each preset angle; perform 3D reconstruction on the data sets at each preset angle in sequence to obtain the initial 3D models corresponding to each preset angle; according to the preset utility pole height threshold, screen the data of the initial 3D models respectively to obtain the 3D models corresponding to each preset angle and the weight files of the utility poles in each 3D model.

[0059] In this embodiment, integrate and clean the orthophoto image data of the utility poles at all preset angles to eliminate unclear and obviously useless photos for model training. Preferably, since the orthophoto image data at 90° has been data-cleaned in step S102, this step only needs to integrate and clean the orthophoto image data at 45° and 60°. Thus, data screening and data stitching are performed on the three angles of 45°, 60°, and 90° to obtain the data sets required for the 3D modeling part. Among them, each angle corresponds to a data set of orthophoto image data.

[0060] In this embodiment, screen the utility pole data in the generated 3D models. Specifically, the recognition height threshold of the utility poles can be set to 10 meters to exclude objects that do not meet the utility pole height standard, so as to reduce the computational complexity and noise in the subsequent recognition process, and then obtain the weight files of the utility poles in each 3D model.

[0061] As a preferred solution of this embodiment, the step of performing 3D reconstruction on the data sets at each preset angle in sequence to obtain the initial 3D models corresponding to each preset angle specifically includes:

[0062] Separate the image data of the data sets at each preset angle in sequence to collect the point cloud data corresponding to each data set; perform 3D modeling on the point cloud data corresponding to each data set in sequence, so that in the process of each 3D modeling, classify and separate the point cloud data corresponding to the data set to generate the point cloud clustering centers corresponding to different categories, and then draw the skeleton and optimize its classification according to the point cloud clustering centers, and integrate the texture and map the optimized skeleton to obtain the initial 3D model corresponding to the data set until each data set has a corresponding initial 3D model.

[0063] In this embodiment, exemplarily, software tools such as Pix4Dmapper or similar can be used to perform 3D reconstruction on the orthophoto image data of the utility poles collected at all preset angles respectively. The 3D reconstruction process includes steps such as image matching, point cloud generation, and texture mapping to create accurate 3D models, and then obtain the weight files of the utility pole data in each 3D model; among them, the number of weight files is at least one.

[0064] In this embodiment, point cloud generation and texture mapping can bring a certain improvement in accuracy. The steps of point cloud generation and texture mapping generally include: separating the point cloud of image data acquisition, separating point clouds of different categories, generating point cloud clustering, drawing the skeleton and its classification optimization, and integrating texture and mapping; among them, the point clouds of different categories include: separating the point cloud of utility poles, the point cloud of citrus trees, and the point cloud of roads and obstacles.

[0065] It can be understood that the 3D point cloud model constructed in the embodiment of the present invention is based on the 3D model in the current scene. Exemplarily, since the orthophoto data for constructing the 3D model has three different angles, the constructed 3D model is also based on the 3D models at three different angles of 45°, 60°, and 90°. For the orthophoto data at one angle, it represents the information of utility poles in the current scene collected by the current UAV. Therefore, at the same angle, the more orthophoto data collected by the UAV, the higher the accuracy of the constructed 3D point cloud model.

[0066] Step S104: Calculate the recognition weight after regression compensation according to the model weight and the weight file, and input the recognition weight into a preset utility pole recognition model, so that the preset utility pole recognition model loaded with the recognition weight can recognize utility poles.

[0067] As a preferred solution of this embodiment, the calculating the recognition weight after regression compensation according to the model weight and the weight file, and inputting the recognition weight into a preset utility pole recognition model, so that the preset utility pole recognition model loaded with the recognition weight can recognize utility poles specifically includes:

[0068] Integrate the weights of the model weight and the weight file according to a preset search algorithm to obtain the recognition weight after regression compensation; save the recognition weight to obtain a recognition weight file, and send the recognition weight file to the preset utility pole recognition model, so that after the preset utility pole recognition model loads the recognition weight file, it can recognize the utility poles in the orthophoto map obtained in real time.

[0069] In this embodiment, through the preset search algorithm, the weights of the model weight and the weight file can be integrated, and then the recognition weight after regression compensation can be obtained. That is, this weight can be directly used to determine the recognition weight parameters of the preset utility pole recognition model to improve the accuracy of the preset utility pole recognition model in recognizing utility poles.

[0070] In this embodiment, exemplarily, a corresponding recognition weight file is obtained through the saved recognition weights, so that the recognition weight file can be carried on the drone to perform real-time testing and recognition of utility poles under an orthophoto map.

[0071] In this embodiment, before implementing the recognition of utility poles, it can be detected whether the corresponding recognition weights for this scenario or that can be used have been constructed. If so, the recognition weights can be directly called to achieve real-time monitoring of utility poles.

[0072] As a preferred solution of this embodiment, the preset search algorithm is the sparrow search algorithm; the weight integration of the model weights and the weight file to obtain the recognition weights after regression compensation specifically includes:

[0073] Initialize the model weights and the weight file so that the model weights and the weight file are converted into the corresponding mapping position information in the sparrow search algorithm; use the initialized model weights as the discoverer in the sparrow search algorithm, and use the initialized weight file as the joiner in the sparrow search algorithm, and add a preset danger constraint condition to the discoverer and the joiner, so that the discoverer updates the mapping position information according to the preset fitness function, and the mapping position information of the joiner is updated according to the updated mapping position information of the discoverer. After reaching the preset stop condition, output the optimal solution among the current mapping position information of the discoverer and the current mapping position information of the joiner as the recognition weights after regression compensation by weight integration.

[0074] In this embodiment, the sparrow search algorithm is a swarm intelligence optimization algorithm, which can perform weight integration on the model weights and the weight file to obtain the final optimal weight after regression compensation as the recognition weights. Specifically, the sparrow search algorithm abstracts rules by learning from biological behaviors, introduces the discoverer as shown in formulas (1) and (2); introduces the joiner as shown in formulas (3) and (4) and updates the position of the discoverer or the discoverer aware of danger as shown in formulas (5) and (6) to achieve the best search result, that is, taking this optimal solution as the final optimal recognition weights.

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] Among them, t represents the current iteration number, iter max is a constant representing the maximum number of iterations, X ij represents the position information of the i-th discoverer or joiner in the j-th dimension, and α ∈ (0, 1] is a random number. R 2 (R 2 ∈ [0, 1]) and ST (ST ∈ [0.5, 1]) respectively represent preset warning values, Q is a random number subject to a normal distribution. L represents a 1×d matrix, where each element in the matrix is all 1.

[0082] X p is the optimal position occupied by the current joiner, X worst represents the current globally worst position, A represents a 1×d matrix, where each element is randomly assigned 1 or -1, and A+ = AT(AAT)-1. X best is the current globally optimal position; β, as a step size control parameter, is a random number subject to a normal distribution with a mean of 0 and a variance of 1. K ∈ [-1, 1] is a random number representing the moving direction of the discoverer or joiner and is also a step size control parameter; f i is the fitness value of the current individual of the discoverer or joiner; fg and fw are the current globally best and worst fitness values respectively. ε is the smallest constant to avoid a zero denominator.

[0083] Implementing the above embodiments has the following effects:

[0084] The technical solution of the present invention obtains the orthophoto image data of the telegraph pole at a preset angle, ensures accurate identification of the telegraph pole under the orthophoto image, reduces misidentification, constructs a telegraph pole detection model and its corresponding model weights, and then constructs a three-dimensional model through the orthophoto image data of the telegraph pole at a preset angle, so as to obtain the weight file of the telegraph pole in the three-dimensional model, so that through the integration of the model weights and the weight file, the recognition weight after regression compensation is obtained, and the telegraph pole is recognized by presetting the telegraph pole recognition model by loading the recognition weight. While maintaining high accuracy, the speed of telegraph pole recognition is improved, making it applicable to large-scale data sets and real-time applications, avoiding the problem of model overfitting, and through the model weights of the telegraph pole detection model and the weight file of the three-dimensional model, more multi-dimensional model weight setting parameters for telegraph pole recognition can be provided to improve the model recognition efficiency of the telegraph pole.

[0085] Embodiment 2

[0086] Please refer to the figure, which is an identification device for a utility pole provided by the present invention, including: an acquisition module 201, a training module 202, a three-dimensional module 203, and a regression module 204.

[0087] The acquisition module 201 is used to acquire the orthophoto image data of the utility pole at a preset angle.

[0088] The training module 202 is used to construct an initial utility pole detection model, and by using the orthophoto image data as training samples, train the initial utility pole detection model to obtain the final trained utility pole detection model and its model weights.

[0089] The three-dimensional module 203 is used to construct a three-dimensional model according to the orthophoto image data of the utility pole at a preset angle, and obtain the weight file of the utility pole in the three-dimensional model according to the three-dimensional model and a preset utility pole height threshold.

[0090] The regression module 204 is used to calculate the recognition weight after regression compensation according to the model weights and the weight file, and input the recognition weight into a preset utility pole recognition model, so that the preset utility pole recognition model after loading the recognition weight recognizes the utility pole.

[0091] As a preferred solution, the acquisition of the orthophoto image data of the utility pole at a preset angle specifically includes:

[0092] Acquire the orthophoto image data of the utility pole at a plurality of preset angles; wherein, the number of the preset angles is at least three.

[0093] As a preferred solution, the construction of the initial utility pole detection model, and by using the orthophoto image data as training samples, training the initial utility pole detection model to obtain the final trained utility pole detection model and its model weights specifically includes:

[0094] Annotate the orthophoto image data of the utility pole at any one of the preset angles, so as to obtain the orthophoto image data of the utility pole after annotation and the unannotated orthophoto image data at this preset angle;

[0095] Construct an initial utility pole detection model, and use the annotated orthophoto image data and the unannotated orthophoto image data as training samples, and thus input the training samples into the initial utility pole detection model for training until the loss function of the initial utility pole detection model converges, to obtain the trained utility pole detection model and the model weights corresponding to the current state of this utility pole detection model.

[0096] As a preferred solution, constructing a three-dimensional model based on the orthophoto image data of the utility pole at a preset angle, and obtaining a weight file of the utility pole in the three-dimensional model according to the three-dimensional model and a preset utility pole height threshold, specifically includes:

[0097] Integrate the orthophoto image data of the utility pole at all preset angles respectively to obtain a corresponding data set for each preset angle;

[0098] Perform three-dimensional reconstruction on the data set of each preset angle in sequence to obtain an initial three-dimensional model corresponding to each preset angle;

[0099] According to the preset utility pole height threshold, screen the data of the initial three-dimensional model respectively to obtain a three-dimensional model corresponding to each preset angle and a weight file of the utility pole in each three-dimensional model.

[0100] As a preferred solution, performing three-dimensional reconstruction on the data set of each preset angle in sequence to obtain an initial three-dimensional model corresponding to each preset angle, specifically includes:

[0101] Separate the image data of the data set of each preset angle in sequence to collect the point cloud data corresponding to each data set;

[0102] Perform three-dimensional modeling on the point cloud data corresponding to each data set in sequence. During each three-dimensional modeling process, classify and separate the point cloud data corresponding to the data set to generate point cloud clustering centers corresponding to different categories. Then, draw the skeleton and optimize its classification according to the point cloud clustering centers, and integrate the texture and map the optimized skeleton to obtain the initial three-dimensional model corresponding to the data set until each data set has a corresponding initial three-dimensional model.

[0103] As a preferred solution, calculating the recognition weight after regression compensation according to the model weight and the weight file, and inputting the recognition weight into a preset utility pole recognition model, so that the preset utility pole recognition model after loading the recognition weight recognizes the utility pole, specifically includes:

[0104] Integrate the model weight and the weight file according to a preset search algorithm to obtain the recognition weight after regression compensation;

[0105] Save the recognition weight to obtain a recognition weight file, and send the recognition weight file to the preset utility pole recognition model, so that after the preset utility pole recognition model loads the recognition weight file, it recognizes the utility pole in the orthophoto image obtained in real time.

[0106] As a preferred solution, the preset search algorithm is the sparrow search algorithm; the integration of the model weights and the weight file to obtain the recognition weights after regression compensation specifically includes:

[0107] Initialize the model weights and the weight file so that the model weights and the weight file are converted into the corresponding mapping position information in the sparrow search algorithm;

[0108] Take the initialized model weights as the discoverer of the sparrow search algorithm and the initialized weight file as the joiner of the sparrow search algorithm, and add a preset danger constraint condition to the discoverer and the joiner, so that the discoverer updates the mapping position information according to a preset fitness function, and the mapping position information of the joiner is updated according to the updated mapping position information of the discoverer. After reaching the preset stop condition, output the optimal solution among the current mapping position information of the discoverer and the current mapping position information of the joiner as the recognition weights after regression compensation by weight integration.

[0109] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described device can refer to the corresponding process in the foregoing method embodiment and will not be elaborated here.

[0110] Implementing the above embodiments has the following effects:

[0111] The technical solution of the present invention obtains the orthophoto image data of the utility pole at a preset angle to ensure accurate identification of the utility pole under the orthophoto image and reduce misidentification, and constructs a utility pole detection model and its corresponding model weights. Furthermore, a three-dimensional model is constructed through the orthophoto image data of the utility pole at a preset angle, so as to obtain the weight file of the utility pole in the three-dimensional model. By integrating the model weights and the weight file, the recognition weights after regression compensation are obtained, and the utility pole is identified by presetting the utility pole recognition model by loading the recognition weights. While maintaining high accuracy, the speed of utility pole recognition is improved, making it applicable to large-scale data sets and real-time applications, avoiding the problem of model overfitting, and providing more multi-dimensional model weight setting parameters for utility pole recognition through the model weights of the utility pole detection model and the weight file of the three-dimensional model to improve the model recognition efficiency of the utility pole.

[0112] Embodiment III

[0113] Correspondingly, the present invention also provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for identifying a utility pole according to any one of the above embodiments.

[0114] The terminal device of this embodiment includes: a processor, a memory, and a computer program and computer instructions stored in the memory and executable on the processor. When the processor executes the computer program, it implements each step in the first embodiment above, such as Figure 1 the steps S101 to S104 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above device embodiment, such as the three-dimensional module 203.

[0115] Exemplarily, the computer program can be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the terminal device. For example, the three-dimensional module 203 is used to construct a three-dimensional model based on the orthophoto image data of the utility pole at a preset angle, and obtain the weight file of the utility pole in the three-dimensional model according to the three-dimensional model and the preset utility pole height threshold.

[0116] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the schematic diagram is only an example of the terminal device, and does not constitute a limitation on the terminal device. It may include more or fewer components than shown, or combine some components, or different components. For example, the terminal device may further include input / output devices, network access devices, buses, etc.

[0117] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the terminal device, and connects various parts of the entire terminal device through various interfaces and lines.

[0118] The memory can be used to store the computer program and / or modules. By running or executing the computer program and / or modules stored in the memory, and invoking the data stored in the memory, the processor realizes various functions of the terminal device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function, etc.; the data storage area can store the data created according to the use of the mobile terminal, etc. In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, memory, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0119] Among them, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0120] Embodiment 4

[0121] Correspondingly, the present invention also provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program. Among them, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the pole identification method described in any one of the above embodiments.

[0122] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. In particular, it is pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying a utility pole, characterized in that: include: Obtain orthophoto data of electric poles at a preset angle; Constructing an initial utility pole detection model, and training the initial utility pole detection model by using the orthophoto data as training samples to obtain a final utility pole detection model and its model weight after training; Constructing a three-dimensional model based on orthophoto data of the utility pole at a preset angle, and obtaining a weight file of the utility pole in the three-dimensional model based on the three-dimensional model and a preset utility pole height threshold; The orthophoto data of the electric poles at all preset angles are integrated to obtain the corresponding data set at each preset angle; Perform three-dimensional reconstruction in sequence according to the data set of each preset angle to obtain an initial three-dimensional model corresponding to each preset angle; Separate the image data of each data set at a preset angle in turn, so as to acquire the point cloud data corresponding to each data set; Performing three-dimensional modeling on the point cloud data corresponding to each data set in turn, so that in each three-dimensional modeling process, the point cloud data corresponding to the data set is classified and separated, thereby generating point cloud cluster centers corresponding to different categories, and then performing skeleton drawing and classification optimization based on the point cloud cluster centers, and performing texture integration and mapping on the optimized skeleton to obtain an initial three-dimensional model corresponding to the data set, until each data set has a corresponding initial three-dimensional model; According to the preset pole height threshold, the initial three-dimensional model is respectively screened to obtain a three-dimensional model corresponding to each preset angle and a weight file corresponding to the pole in each three-dimensional model; Calculate the recognition weight after regression compensation according to the model weight and the weight file, and input the recognition weight into the preset electric pole recognition model, so that the preset electric pole recognition model loaded with the recognition weight can recognize the electric pole; According to a preset search algorithm, the model weight and the weight file are weight-integrated to obtain a recognition weight after regression compensation; The recognition weight is saved to obtain a recognition weight file, and the recognition weight file is sent to a preset utility pole recognition model, so that after the preset utility pole recognition model loads the recognition weight file, the utility pole in the orthophoto map acquired in real time is recognized.

2. A method for identifying a utility pole according to claim 1, characterized in that: The obtaining of orthophoto data of the electric pole at a preset angle specifically includes: Orthophoto data of electric poles at a plurality of preset angles are obtained; wherein the number of the preset angles is at least three.

3. A method for identifying a utility pole as claimed in claim 2, characterized in that: The initial utility pole detection model is constructed, and the initial utility pole detection model is trained by using the orthophoto data as training samples to obtain the final utility pole detection model after training and its model weight, specifically including: Annotate the orthophoto data of the electric pole at any preset angle, so as to obtain the annotated orthophoto data and unannotated orthophoto data of the electric pole at the preset angle; An initial utility pole detection model is constructed, and the labeled orthophoto data and the unlabeled orthophoto data are used as training samples, so that the training samples are input into the initial utility pole detection model for training until the loss function of the initial utility pole detection model converges, thereby obtaining a trained utility pole detection model and a model weight corresponding to the current state of the utility pole detection model.

4. A method for identifying a utility pole as claimed in claim 3, characterized in that: The preset search algorithm is a sparrow search algorithm; the weight integration of the model weight and the weight file to obtain the recognition weight after regression compensation specifically includes: Initializing the model weight and the weight file so that the model weight and the weight file are converted into corresponding mapping position information in the sparrow search algorithm; The initialized model weight is used as the finder of the sparrow search algorithm, and the initialized weight file is used as the joiner of the sparrow search algorithm. Preset danger constraints are added to the finder and the joiner, so that the finder updates the mapping position information according to a preset fitness function, and the mapping position information of the joiner is updated according to the updated mapping position information of the finder, until the preset stop condition is reached, and the optimal solution of the mapping position information of the current finder and the mapping position information of the current joiner is output as the recognition weight of regression compensation after weight integration.

5. A device for identifying a utility pole, characterized in that: include: Acquisition module, training module, three-dimensional module and regression module; The acquisition module is used to acquire orthophoto data of the electric pole at a preset angle; The training module is used to construct an initial utility pole detection model, and train the initial utility pole detection model by using the orthophoto data as a training sample to obtain a final utility pole detection model and its model weight after training; The three-dimensional module is used to construct a three-dimensional model according to the orthophoto data of the electric pole at a preset angle, and obtain a weight file of the electric pole in the three-dimensional model according to the three-dimensional model and a preset electric pole height threshold; The orthophoto data of the electric poles at all preset angles are integrated to obtain the corresponding data set at each preset angle; Perform three-dimensional reconstruction in sequence according to the data set of each preset angle to obtain an initial three-dimensional model corresponding to each preset angle; Separate the image data of each data set at a preset angle in turn, so as to acquire the point cloud data corresponding to each data set; Performing three-dimensional modeling on the point cloud data corresponding to each data set in turn, so that in each three-dimensional modeling process, the point cloud data corresponding to the data set is classified and separated, thereby generating point cloud cluster centers corresponding to different categories, and then performing skeleton drawing and classification optimization based on the point cloud cluster centers, and performing texture integration and mapping on the optimized skeleton to obtain an initial three-dimensional model corresponding to the data set, until each data set has a corresponding initial three-dimensional model; According to the preset pole height threshold, the initial three-dimensional model is respectively screened to obtain a three-dimensional model corresponding to each preset angle and a weight file corresponding to the pole in each three-dimensional model; The regression module is used to calculate the recognition weight after regression compensation according to the model weight and the weight file, and input the recognition weight into the preset electric pole recognition model, so that the preset electric pole recognition model loaded with the recognition weight can recognize the electric pole; According to a preset search algorithm, the model weight and the weight file are weight-integrated to obtain a recognition weight after regression compensation; The recognition weight is saved to obtain a recognition weight file, and the recognition weight file is sent to a preset utility pole recognition model, so that after the preset utility pole recognition model loads the recognition weight file, the utility pole in the orthophoto map acquired in real time is recognized.

6. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for identifying a utility pole according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the utility pole identification method according to any one of claims 1 to 4.

Images