Power line target extraction method based on random forest and one-class classification strategy

Abstract

The invention provides a power line target extraction method based on a random forest and a classification strategy. The method comprises the following steps: acquiring airborne LiDAR point cloud data and a multispectral orthoimage collected by an unmanned aerial vehicle multi-sensor platform; acquiring a feature set according to the airborne LiDAR point cloud data, and training a random forest binary classifier by using the airborne LiDAR point cloud data, the feature set and the multispectral orthoimage according to a random forest algorithm and a class-I classification strategy to obtain a PBL-Random Forest model; and acquiring airborne point cloud data to be classified, and performing power line classification on the airborne point cloud data to be classified by adopting the PBL-Random Forest model to generate a vector result of a power line. The problem that a model in the prior art is difficult to accurately predict the actual existence of the power line is solved.

Description

Technical Field

[0001] The present application relates to the technical field of power line target extraction, and in particular, to a power line target extraction method based on random forest and one-class classification strategy, a power line target extraction device based on random forest and one-class classification strategy, a computer-readable storage medium and an electronic device. Background Art

[0002] With the advancement of urbanization, the coverage of power lines in cities has gradually expanded, and the threat posed by tree growth to the safety of power lines has become increasingly significant. As trees grow, their branches will gradually approach power lines, causing faults such as short circuits and grounding, seriously affecting power supply and possibly threatening human life. The difficulty of maintaining power facilities has also increased accordingly. Therefore, accurate identification of power lines in the process of urbanization is of great significance for the rational planning of line corridors, regular maintenance of power facilities, timely detection and treatment of tree obstacles, and improving the safety and reliability of power supply, and promoting sustainable urban development.

[0003] Due to the development needs of UAV technology and earth observation, airborne laser radar (LiDAR) technology combines the technical advantages of high-precision ranging and hyperspectral imaging of LiDAR, which can achieve high-precision and high-resolution monitoring of ground objects, and has become a particularly important technology for obtaining accurate three-dimensional (3D) space technology. The point cloud data obtained by airborne laser radar (LiDAR) technology has high-precision, high-resolution and high-latitude geometric information, and can intuitively represent the shape, surface and texture of objects in space. Point cloud data can not only obtain high-precision three-dimensional structural information of power lines through high-density three-dimensional point sets, but also in complex environments, such as dense forests and urban high-rise buildings, where power lines are often blocked by vegetation, buildings, etc., point cloud data can directly obtain three-dimensional space information through laser scanning, which is not easily interfered by environmental factors and can more accurately identify blocked power lines.

[0004] Most existing point cloud datasets focus on the existence information of the target object, that is, the point cloud data itself, while often ignoring the importance of background or non-existence information. This imbalance in data composition makes it difficult for the model to accurately predict the actual existence of power lines because the model lacks background feature learning when the object does not exist. Traditional statistical methods, such as the maximum entropy method, usually require both existence and non-existence data for model training. However, in the actual acquisition of point cloud data, the acquisition of such paired data is often unrealistic, which limits the effective application of these traditional methods in point cloud classification tasks. Even if some methods can process point clouds containing only existence data, they can often only output a relative suitability index, but cannot provide a probability estimate of the actual existence of the object. This lack of probabilistic interpretation of the prediction results makes it difficult to compare different models and also affects the interpretability of the prediction results in practical applications. Summary of the invention

[0005] The main purpose of the present application is to provide a power line target extraction method based on random forest and one-class classification strategy, a power line target extraction device based on random forest and one-class classification strategy, a computer-readable storage medium and an electronic device, so as to at least solve the problem that the models in the prior art are difficult to accurately predict the actual existence of power lines.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a power line target extraction method based on random forest and one-class classification strategy is provided, including: obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by a multi-sensor platform of an unmanned aerial vehicle; obtaining a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality and plane fitting residual; according to the random forest algorithm and one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophotos are used to train a random forest binary classifier to obtain a PBL-Random Forest model; obtaining airborne point cloud data to be classified, and using the PBL-Random Forest model to classify the airborne point cloud data to be classified into power lines to generate vector results of power lines.

[0007] Optionally, acquiring height-based features according to the airborne LiDAR point cloud data includes: using a progressive triangulated irregular network densification filtering algorithm in LAStools software to identify ground points in the airborne LiDAR point cloud data to obtain first echo data and ground echo data; interpolating the first echo data and the ground echo data according to an ordinary Kriging method to obtain a digital surface model and a digital elevation model; determining a normalized digital surface model according to the digital surface model and the digital elevation model, defining a spherical neighborhood for each point on the normalized digital surface model, determining the height variance of all points in the spherical neighborhood, and assigning the height variance to the center point of the spherical neighborhood to obtain the height-based features.

[0008] Optionally, feature-based features are obtained according to the airborne LiDAR point cloud data, including: defining a spherical neighborhood centered on each point cloud in the airborne LiDAR point cloud data; determining a local covariance matrix of each point cloud according to the spherical neighborhood; performing eigendecomposition on the local covariance matrix to obtain a plurality of eigenvectors and a plurality of eigenvalues; and respectively determining relevant features of the eigenvectors and relevant features of the eigenvalues ​​to obtain the feature-based features, wherein the relevant features of the eigenvectors include linearity, flatness, sphericity and curvature change, and the relevant features of the eigenvalues ​​include verticality and plane fitting residuals.

[0009] Optionally, wherein the echo-based features include a total laser pulse echo number and an echo ratio, and obtaining the echo-based features according to the airborne LiDAR point cloud data includes: determining a numerical value of the laser pulse echo received by each point cloud in the airborne LiDAR point cloud data to obtain the total laser pulse echo number; determining the echo number of each laser pulse echo, and determining the echo ratio according to the total laser pulse echo number and the echo number.

[0010] Optionally, obtaining multi-scale features based on the airborne LiDAR point cloud data includes: obtaining a preset number of preset scales, and determining a spherical neighborhood with each preset scale as a radius based on the airborne LiDAR point cloud data; determining local geometric features of the spherical neighborhood corresponding to each preset scale to obtain the multi-scale features.

[0011] Optionally, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier, including: classifying the airborne LiDAR point cloud data to obtain labeled samples and unlabeled samples, wherein the labeled samples represent samples with the power lines labeled, and the unlabeled samples represent samples without the power lines labeled; training the forest binary classifier using the labeled samples and the unlabeled samples and the feature set, and then using the multispectral orthophoto to verify the output result of the random forest binary classifier during the training of the random forest binary classifier.

[0012] Optionally, after generating the vector result of the power line, the method further includes: determining the position and shape of the power line according to the vector result of the power line, so as to implement inspection work on the power line.

[0013] According to another aspect of the present application, a power line target extraction device based on random forest and one-class classification strategy is provided, including: a first acquisition unit, used to acquire airborne LiDAR point cloud data and multispectral orthophotos collected by a multi-sensor platform of an unmanned aerial vehicle; a second acquisition unit, used to acquire a feature set based on the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality and plane fitting residual; a training unit, used to train a random forest binary classifier using the airborne LiDAR point cloud data, the feature set and the multispectral orthophotos according to a random forest algorithm and a one-class classification strategy to obtain a PBL-Random Forest model; a classification unit, used to acquire airborne point cloud data to be classified, and use the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, and generate vector results of power lines.

[0014] According to another aspect of the present application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the power line target extraction methods based on random forest and one-class classification strategy.

[0015] According to another aspect of the present application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing any one of the power line target extraction methods based on random forest and one-class classification strategies.

[0016] Apply the technical solution of the present application to obtain airborne LiDAR point cloud data and multispectral orthophotos collected by a multi-sensor platform of an unmanned aerial vehicle; obtain a feature set based on the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residuals; according to the random forest algorithm and a one-class classification strategy, train the random forest binary classifier using airborne LiDAR point cloud data, feature sets, and multispectral orthophotos to obtain a PBL-Random Forest model; obtain airborne point cloud data to be classified, and use the PBL-Random Forest model to classify the airborne point cloud data to be classified to generate vector results of the power lines. The PBL-Random Forest model is obtained by using the random forest method and a one-class classification strategy in combination with airborne LiDAR point cloud data and feature sets, and the PBL-RandomForest model is used to predict and identify power lines, thereby improving the accuracy of identifying power lines and solving the actual problem that the existing model is difficult to accurately predict power lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 A hardware structure block diagram of a mobile terminal for performing a power line target extraction method based on a random forest and a one-class classification strategy provided in an embodiment of the present application is shown;

[0019] Figure 2 A schematic diagram of a flow chart of a method for extracting power line targets based on a random forest and a one-class classification strategy provided according to an embodiment of the present application is shown;

[0020] Figure 3 A schematic diagram of a flow chart of a power line target extraction method based on a random forest and a one-class classification strategy provided according to an embodiment of the present application is shown;

[0021] Figure 4 A calculation flow chart of a feature set provided according to an embodiment of the present application is shown;

[0022] Figure 5 A flowchart of generating a PBL-Random Forest model according to an embodiment of the present application is shown;

[0023] Figure 6A flow chart of power line classification using a PBL-Random Forest model provided in an embodiment of the present application is shown;

[0024] Figure 7 A structural block diagram of a power line target extraction device based on random forest and one-class classification strategy provided according to an embodiment of the present application is shown.

[0025] The above drawings include the following reference numerals:

[0026] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0027] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] As introduced in the background technology, it is difficult for the models of the prior art to accurately predict the actual existence of power lines. In order to solve the problem that the models of the prior art are difficult to accurately predict the actual existence of power lines, the embodiments of the present application provide a power line target extraction method based on random forest and one-class classification strategy, a power line target extraction device based on random forest and one-class classification strategy, a computer-readable storage medium and an electronic device.

[0031] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0032] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a method for extracting power line targets based on random forest and one-class classification strategy according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0033] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the power line target extraction method based on random forest and one-class classification strategy in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0034] In this embodiment, a power line target extraction method based on random forest and one-class classification strategy is provided, which runs on a mobile terminal, a computer terminal or a similar computing device. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0035] Figure 2 is a flow chart of a method for extracting power line targets based on random forest and one-class classification strategy according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0036] Step S201, obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by the drone multi-sensor platform;

[0037] Specifically, the airborne sensor is a device installed on the UAV platform for remote sensing detection during flight, including common multispectral sensors and laser radar point cloud detectors. The airborne sensor can collect multi-source remote sensing data, such as multispectral images and airborne laser radar point clouds. In this embodiment, the multispectral sensor is used to collect multispectral high-resolution orthophotos for ground observation, and the airborne laser radar detector is used to collect airborne laser radar point clouds of different scales of 1.5 meters, 2.5 meters and 3.5 meters for ground observation.

[0038] Step S202, acquiring a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual;

[0039] Among them, feature-based features refer to defining another feature or attribute based on a certain feature or attribute. This method is often used in the field of data analysis and machine learning to identify and extract key feature information. For example, suppose we have a set of data that contains students' academic performance and various personal characteristics, such as age, gender, family background, etc. We can determine which features have an important impact on academic performance by analyzing the relationship between these personal characteristics and academic performance. Then, based on these important features, we can build a model to predict students' academic performance. Through feature-based feature analysis, we can better understand the relationship between data, discover the patterns and laws hidden behind the data, and thus improve the accuracy and efficiency of data analysis and prediction.

[0040] Step S203, according to the random forest algorithm and the one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier to obtain a PBL-RandomForest model;

[0041] Specifically, the training model is obtained by training the airborne LiDAR point cloud dataset and the four extracted features through the random forest algorithm and the one-class classification strategy. Then, according to the quantitative relationship between the training model and the target model, it is calibrated through the validation dataset, and finally the target model PBL-Random Forest model is obtained.

[0042] Step S204, obtaining the airborne point cloud data to be classified, using the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, and generating a vector result of the power line.

[0043] Specifically, vectorization refers to the process of converting raster-format image data into vector data. After classifying the airborne point cloud data into power lines using the PBL-Random Forest model, vector lines are generated by connecting adjacent boundary points in the point cloud, and finally the vector results of the power lines are obtained.

[0044] Through this embodiment, the airborne LiDAR point cloud data and multispectral orthophotos collected by the multi-sensor platform of the unmanned aerial vehicle are obtained; a feature set is obtained based on the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residuals; according to the random forest algorithm and a one-class classification strategy, the airborne LiDAR point cloud data, feature set, and multispectral orthophotos are used to train the random forest binary classifier to obtain a PBL-RandomForest model; the airborne point cloud data to be classified is obtained, and the PBL-Random Forest model is used to classify the airborne point cloud data to be classified to generate vector results of the power lines. The PBL-Random Forest model is obtained by using the random forest method and a one-class classification strategy combined with the airborne LiDAR point cloud data and feature set, and the PBL-Random Forest model is used to predict and identify the power lines, thereby improving the accuracy of identifying the power lines and solving the actual problem that the existing model is difficult to accurately predict the power lines.

[0045] In a specific implementation process, the height-based features are obtained according to the above-mentioned airborne LiDAR point cloud data, including: using the progressive triangulated irregular network densification filtering algorithm in the LAStools software to identify the ground points in the above-mentioned airborne LiDAR point cloud data to obtain the first echo data and the ground echo data; interpolating the above-mentioned first echo data and the above-mentioned ground echo data according to the ordinary Kriging method to obtain a digital surface model and a digital elevation model; determining a normalized digital surface model according to the above-mentioned digital surface model and the above-mentioned digital elevation model, defining a spherical neighborhood for each point on the above-mentioned normalized digital surface model, determining the height variance of all points in the above-mentioned spherical neighborhood, and distributing the above-mentioned height variance to the center point of the above-mentioned spherical neighborhood to obtain the above-mentioned height-based features.

[0046] The method first removes discrete outliers in the airborne LiDAR point cloud data, and uses the progressive triangulated irregular network (TIN) densification filtering algorithm in the LAStools software to identify ground points; uses the ordinary kriging method to interpolate the first echo data and the ground echo data to obtain a 1-meter resolution digital surface model (DSM) and a digital elevation model (DEM); calculates the difference between the digital surface model (DSM) and the digital elevation model (DEM), which is the normalized digital surface model (nDSM); outputs the normalized digital surface model (nDSM) as the normalized height (Hn); defines a spherical neighborhood for each point on the normalized digital surface model (nDSM), calculates the variance of the height values ​​of all points in the neighborhood, and assigns the variance value to the center point to obtain the height variance (Hv).

[0047] Specifically, feature-based features are obtained according to the above-mentioned airborne LiDAR point cloud data, including: defining a spherical neighborhood centered on each point cloud in the above-mentioned airborne LiDAR point cloud data; determining a local covariance matrix of each of the above-mentioned point clouds according to the above-mentioned spherical neighborhood; performing eigendecomposition on the above-mentioned local covariance matrix to obtain multiple eigenvectors and multiple eigenvalues; respectively determining relevant features of the above-mentioned eigenvectors and relevant features of the eigenvalues ​​to obtain the above-mentioned feature-based features, wherein the relevant features of the above-mentioned eigenvectors include linearity, flatness, sphericity and curvature change, and the relevant features of the above-mentioned eigenvalues ​​include verticality and plane fitting residual.

[0048] This method first removes discrete outliers in the airborne LiDAR point cloud data. Since the point cloud data is n 3D points, it can be expressed as For each point, define a spherical neighborhood centered on the point and calculate the local covariance matrix of the point. cov Perform eigendecomposition to obtain eigenvectors (V1, V2, V3) and eigenvalues ​​(λ1≥λ2≥λ3≥0), calculate eigenvalue-related features, and linearity L λ, Flatness P λ , Sphericity S λ And the curvature change k, calculate the eigenvector related features, calculate the angle between the local normal vector and the vertical direction, and get the verticality N z ; Calculate the residual of the local plane fitting and get the plane fitting residual R z .

[0049] More specifically, the echo-based features include a total number of laser pulse echoes and an echo ratio. Acquiring the echo-based features according to the airborne LiDAR point cloud data includes: determining a numerical value of the laser pulse echo received by each point cloud in the airborne LiDAR point cloud data to obtain the total number of laser pulse echoes; determining an echo number of each of the laser pulse echoes, and determining the echo ratio according to the total number of laser pulse echoes and the echo number.

[0050] This method first removes discrete abnormal points in the airborne LiDAR point cloud data. For each point in the point cloud, the total number of laser pulse echoes received by the point is calculated to obtain the total number of echoes N. E ; For each echo, calculate the echo number, that is, the number of times the echo returns, and divide the echo number by the total number of echoes N E , get the echo ratio R E .

[0051] Furthermore, multi-scale features are obtained based on the above-mentioned airborne LiDAR point cloud data, including: obtaining a preset number of preset scales, and determining spherical neighborhoods with each of the above-mentioned preset scales as a radius based on the above-mentioned airborne LiDAR point cloud data; determining local geometric features of the above-mentioned spherical neighborhoods corresponding to each preset scale to obtain the above-mentioned multi-scale features.

[0052] Among them, the preset number of preset scales can be set to 1.5m, 2.5m and 3.5m;

[0053] This method repeatedly calculates the local geometric features of the neighborhood for each scale, including: height variance H V , Linearity L λ , Flatness P λ , Sphericity S λ , curvature change k, verticality N z and plane fitting residual R z .

[0054] Furthermore, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier, including: classifying the airborne LiDAR point cloud data to obtain labeled samples and unlabeled samples, wherein the labeled samples represent samples with the power lines labeled, and the unlabeled samples represent samples without the power lines labeled; the labeled samples and unlabeled samples and the feature set are used to train the forest binary classifier, and during the training of the random forest binary classifier, the multispectral orthophoto is used to verify the output result of the random forest binary classifier.

[0055] This method is based on the PBL (presence and background learning algorithm) method. By dividing the collected airborne LiDAR point cloud dataset into labeled or not, the labeled data is regarded as positive samples and the unlabeled data as unlabeled samples, and the probability of a single sample x being a positive sample (denoted as P(y=1|x)), that is, the target model P(y=1|x) is modeled.

[0056] First, positive samples and unlabeled samples are randomly selected at a ratio of 1:5, features are extracted and normalized to the range of [0,1], and then 75% of the samples are used as training data sets and 25% as validation data sets. The features in the training process have the same weight. Let s = 1 to indicate that x is collected as a positive sample, and s = 0 to indicate that x is collected as an unlabeled sample. Then, the random forest model is used for training to obtain the random forest binary classifier model P (s = 1 | x, η = 1), where η = 1 means that there is a background strategy.

[0057] Specifically, after generating the vector result of the power line, the method further comprises: determining the position and shape of the power line according to the vector result of the power line, so as to implement the inspection work on the power line.

[0058] This embodiment adopts the PBL (presence and background learning) method. This method thoroughly solves the problem of non-existence information that is often overlooked in traditional point cloud data sets, and achieves the comprehensiveness and depth of model training by cleverly integrating existence data and background data. In addition, the PBL method significantly improves the prediction accuracy of the existence status of objects by accurately utilizing these two types of complementary information sources. It is worth mentioning that this method also breaks through the limitation that traditional technology can only output relative suitability scores. It can accurately estimate the actual existence probability of an object under specific environmental covariates. This probabilistic prediction output not only greatly enhances the interpretability of the prediction results, but also provides a more objective and scientific measurement standard for performance comparison between different models.

[0059] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the power line target extraction method based on random forest and one-class classification strategy of the present application will be described in detail below in combination with specific embodiments.

[0060] This embodiment relates to a specific power line target extraction method based on random forest and one-class classification strategy, such as Figure 3 As shown, the following steps are included:

[0061] Step S1: Collect multispectral high-resolution orthophotos and airborne LiDAR point cloud data through the UAV multi-sensor platform. Airborne sensors are devices installed on the UAV platform for remote sensing detection during flight, including common multispectral sensors and LiDAR point cloud detectors. Airborne sensors can be used to collect multi-source remote sensing data, such as multispectral images and airborne LiDAR point clouds. In this embodiment, multispectral high-resolution orthophotos are obtained by ground observation using multispectral sensors, and airborne LiDAR point clouds of different scales of 1.5 meters, 2.5 meters and 3.5 meters are obtained by ground observation using airborne LiDAR detectors.

[0062] Step S2: For the airborne LiDAR point cloud data, the height features are calculated using the laser radar height data; the local spatial distribution is evaluated to obtain feature-based features; the echo features are obtained according to the number of laser pulse returns; and the local geometric features are repeatedly calculated to obtain multi-scale features.

[0063] like Figure 4 As shown, height-based features are generated through airborne LiDAR point cloud data to obtain normalized height (Hn) and height variance (Hv); feature-based features are generated to obtain eigenvalue-related features and eigenvector-related features; echo-based features are generated to obtain the total echo number NE and echo ratio RE; and multi-scale features are generated.

[0064] The steps of extracting height-based features from airborne LiDAR point cloud data are as follows:

[0065] 1) Eliminate discrete outliers in the airborne lidar point cloud;

[0066] 2) Use the progressive triangulated irregular network (TIN) densification filtering algorithm in LAStools software to identify ground points;

[0067] 3) Use the ordinary kriging method to interpolate the first echo data and ground echo data to obtain a 1-meter resolution digital surface model (DSM) and digital elevation model (DEM);

[0068] 4) Calculate the difference between the digital surface model (DSM) and the digital elevation model (DEM), which is the normalized digital surface model (nDSM);

[0069] 5) Output the normalized digital surface model (nDSM) as the normalized height (Hn);

[0070] 6) Define a spherical neighborhood for each point on the normalized digital surface model (nDSM), calculate the variance of the height values ​​of all points in the neighborhood, assign the variance value to the center point, and obtain the height variance (Hv).

[0071] Among them, extracting feature-based features from airborne LiDAR point cloud data specifically includes the following steps:

[0072] 1) Eliminate discrete outliers in the airborne lidar point cloud;

[0073] 2) Since the point cloud data is n 3D points, it can be expressed as For each point, define a spherical neighborhood centered on the point and calculate the local covariance matrix M of the point cov The local covariance matrix M cov The calculation formula is as follows:

[0074]

[0075] In the formula, is the mean of the points, and its calculation formula is as follows:

[0076]

[0077] Among them, O i is the i-th point in the point cloud dataset, T is the transposition operation, n is the total number of points in the point cloud dataset, and i is the index of the point.

[0078] 3) For the covariance matrix M cov Perform eigendecomposition to obtain eigenvectors (V1, V2, V3) and eigenvalues ​​(λ1≥λ2≥λ3≥0). The calculation formulas for eigenvectors (V1, V2, V3) and eigenvalues ​​(λ1≥λ2≥λ3≥0) are as follows:

[0079]

[0080] 4) Calculate the eigenvalue related features and linearity L λ , Flatness P λ , Sphericity S λ And the curvature change k, which is calculated as follows:

[0081]

[0082] Where λ1, λ2 and λ3 represent the three eigenvalues ​​of the covariance matrix respectively.

[0083] 5) Calculate the relevant features of the feature vector, calculate the angle between the local normal vector and the vertical direction, and obtain the verticality N z ; Calculate the residual of the local plane fitting and get the plane fitting residual R z .

[0084] Among them, extracting echo-based features from airborne LiDAR point cloud data specifically includes the following steps:

[0085] 1) Eliminate discrete outliers in the airborne lidar point cloud;

[0086] 2) For each point in the point cloud, calculate the total number of laser pulse echoes received by the point and obtain the total number of echoes N E ;

[0087] 3) For each echo, calculate the echo number, that is, the number of times the echo returns, and divide the echo number by the total number of echoes N. E , get the echo ratio R E .

[0088] Among them, extracting multi-scale features from airborne LiDAR point cloud data specifically includes the following steps:

[0089] 1) Determine the scales of interest, 1.5m, 2.5m and 3.5m;

[0090] 2) For each scale of interest, repeatedly calculate the local geometric features of the neighborhood, including: height variance H V , Linearity L λ , Flatness P λ , Sphericity S λ , curvature change k, verticality N z and plane fitting residual R z .

[0091] Step S3: The training model is obtained by training the airborne LiDAR point cloud dataset and the four extracted features through the random forest algorithm and the one-class classification strategy. Then, according to the quantitative relationship between the training model and the target model, the validation dataset is used for calibration, and finally the target model PBL-Random Forest model is obtained.

[0092] like Figure 5As shown, according to the PBL (presence and background learning algorithm) method, the present invention divides the collected airborne LiDAR point cloud data set into labeled ones, regards the labeled data as positive samples, and regards the unlabeled data as unlabeled samples, and models the probability that a single sample x is a positive sample (denoted as P(y=1|x)), that is, the target model P(y=1|x).

[0093] First, positive samples and unlabeled samples are randomly selected at a ratio of 1:5, features are extracted and normalized to the range of [0,1], and then 75% of the samples are used as training data sets and 25% as validation data sets. The features in the training process have the same weight. Let s = 1 to indicate that x is collected as a positive sample, and s = 0 to indicate that x is collected as an unlabeled sample. Then, the random forest model is used for training to obtain the random forest binary classifier model P (s = 1 | x, η = 1), where η = 1 means that there is a background strategy.

[0094] The specific operations for training the random forest binary classifier are:

[0095] 1) Use the Bootstraping method to randomly extract multiple sample subsets from the training set with replacement, and each subset is used to train a decision tree. The Bootstraping method is a method of estimating the distribution of a statistic or performing a hypothesis test by repeated sampling (extracting samples from the original data set with replacement).

[0096] 2) When constructing each decision tree, a portion of features are randomly selected as candidate features, and the best features are selected for node splitting;

[0097] 3) Each tree grows to its maximum depth without pruning;

[0098] 4) Use majority voting method to aggregate the prediction results of multiple decision trees;

[0099] 5) Tune the trained model, including adjusting hyperparameters (such as the number of decision trees, maximum number of features, maximum depth, etc.), optimizing the model structure, etc., to improve the performance and generalization ability of the model;

[0100] 6) Finally, the training model P(s=1|x,η=1) is obtained.

[0101] According to the conditional probability method and random sampling strategy, the quantitative relationship between the training model P(s=1|x,η=1) and the target model P(y=1|x) can be expressed as:

[0102]

[0103] Where c is a constant used for calibration and is estimated from the validation dataset O and is calculated as follows:

[0104]

[0105] Where n is the cardinality of the dataset O.

[0106] Finally, the target model P(y=1|x) is obtained, which is the PBL-Random Forest model.

[0107] Step S4: Use the PBL-Random Forest model to classify the airborne point cloud data into power lines, then vectorize the classification results to finally generate vector results of the power lines.

[0108] like Figure 6 As shown, in this embodiment, vectorization refers to the process of converting image data in raster format into vector data. After the airborne point cloud data is classified into power lines using the PBL-Random Forest model, vector lines are generated by connecting adjacent boundary points in the point cloud, and finally the vector result of the power lines is obtained.

[0109] In order to solve the problem that most existing point cloud datasets only focus on the point cloud data itself and ignore the importance of background or non-existence information in the target detection task, this embodiment adopts the PBL (presence and background learning algorithm) method, and uses both presence and background data for model training, which effectively solves the problem of missing non-existence information in point cloud data. PBL is a general single-class classification scheme that models the probability that a single sample x is a positive sample (denoted as P(y=1|x)) based on the input positive samples and unlabeled samples. Compared with traditional methods, the PBL method can estimate the actual existence probability of an object under given environmental covariates.

[0110] In order to improve the accuracy of the random forest algorithm in the point cloud classification task, this embodiment combines the random forest with the PBL method to form a PBL-Random Forest model. The random forest improves the robustness and prediction accuracy of the model by constructing multiple decision trees and outputting the average or mode of their prediction results. The use of random forest in PBL can further improve the prediction accuracy of the PBL model. Since PBL relies on accurate posterior probability estimates, the integrated characteristics of random forests make it better at processing complex and noisy data sets, thereby providing more reliable probability estimates.

[0111] The present invention adopts the PBL (presence and background learning) method. This method thoroughly solves the problem of non-existence information that is often neglected in traditional point cloud data sets, and achieves the comprehensiveness and depth of model training by cleverly integrating existence data and background data. In addition, the PBL method significantly improves the prediction accuracy of the existence state of objects by accurately utilizing these two types of complementary information sources. It is worth mentioning that this method also breaks through the limitation that traditional technology can only output relative suitability scores, and it can accurately estimate the actual existence probability of objects under specific environmental covariates. This probabilistic prediction output not only greatly enhances the interpretability of the prediction results, but also provides a more objective and scientific measurement standard for performance comparison between different models. In addition, the PBL-Random Forest model obtained by using random forest and one-class classification strategy does not use a ready-made classification model, but combines random forest with the PBL method, so that the PBL-Random Forest model can maintain a high classification accuracy when there is no information missing in the point cloud data.

[0112] The embodiment of the present application also provides a power line target extraction device based on random forest and a class classification strategy. It should be noted that the power line target extraction device based on random forest and a class classification strategy in the embodiment of the present application can be used to execute the power line target extraction method based on random forest and a class classification strategy provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0113] The following introduces a power line target extraction device based on random forest and one-class classification strategy provided in an embodiment of the present application.

[0114] Figure 7 is a schematic diagram of a power line target extraction device based on random forest and one-class classification strategy according to an embodiment of the present application. Figure 7 As shown, the device comprises:

[0115] The first acquisition unit 71 is used to acquire the airborne LiDAR point cloud data and multispectral orthophotos collected by the UAV multi-sensor platform;

[0116] A second acquisition unit 72 is used to acquire a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual;

[0117] A training unit 73 is used to train a random forest binary classifier using the airborne LiDAR point cloud data, the feature set, and the multispectral orthophoto according to a random forest algorithm and a one-class classification strategy to obtain a PBL-RandomForest model;

[0118] The classification unit 74 is used to obtain the airborne point cloud data to be classified, and use the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, so as to generate a vector result of the power line.

[0119] In this embodiment, the first acquisition unit is used to acquire the airborne LiDAR point cloud data and multispectral orthophotos collected by the unmanned aerial vehicle multi-sensor platform; the second acquisition unit is used to acquire a feature set based on the above-mentioned airborne LiDAR point cloud data, wherein the above-mentioned feature set includes height-based features, feature-based features, echo-based features and multi-scale features, and the above-mentioned multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality and plane fitting residual; the training unit is used to train the random forest binary classifier according to the random forest algorithm and a one-class classification strategy using the above-mentioned airborne LiDAR point cloud data, the above-mentioned feature set and the above-mentioned multispectral orthophotos to obtain a PBL-Random Forest model; the classification unit is used to acquire the airborne point cloud data to be classified, and use the above-mentioned PBL-Random Forest model to perform power line classification on the above-mentioned airborne point cloud data to be classified, and generate a vector result of the power line. The PBL-Random Forest model is obtained by combining the random forest method and one-class classification strategy with the airborne LiDAR point cloud data and feature set. The PBL-Random Forest model is used to predict and identify power lines, which improves the accuracy of identifying power lines and solves the actual problem that the existing technology models are difficult to accurately predict power lines.

[0120] As an optional solution, the second acquisition unit includes an identification module, an interpolation processing module, and a first definition module; the identification module is used to use the progressive triangulated irregular network densification filtering algorithm in the LAStools software to identify ground points in the above-mentioned airborne LiDAR point cloud data to obtain first echo data and ground echo data; the interpolation processing module is used to interpolate the above-mentioned first echo data and the above-mentioned ground echo data according to the ordinary Kriging method to obtain a digital surface model and a digital elevation model; the first definition module is used to determine a normalized digital surface model based on the above-mentioned digital surface model and the above-mentioned digital elevation model, define a spherical neighborhood for each point on the above-mentioned normalized digital surface model, determine the height variance of all points in the above-mentioned spherical neighborhood, and distribute the above-mentioned height variance to the center point of the above-mentioned spherical neighborhood to obtain the above-mentioned height-based feature.

[0121] An optional solution, the second acquisition unit includes a second definition module, a first determination module, a decomposition module and a second determination module; the second definition module is used to define a spherical neighborhood centered on each point cloud in the above-mentioned airborne LiDAR point cloud data; the first determination module is used to determine the local covariance matrix of each of the above-mentioned point clouds based on the above-mentioned spherical neighborhood; the decomposition module is used to perform eigendecomposition on the above-mentioned local covariance matrix to obtain multiple eigenvectors and multiple eigenvalues; the second determination module is used to respectively determine the relevant features of the above-mentioned eigenvectors and the relevant features of the eigenvalues ​​to obtain the above-mentioned feature-based features, wherein the relevant features of the above-mentioned eigenvectors include linearity, flatness, sphericity and curvature change, and the relevant features of the above-mentioned eigenvalues ​​include verticality and plane fitting residuals.

[0122] An optional solution, the second acquisition unit includes a third determination module and a fourth determination module, the third determination module is used to determine the numerical value of the laser pulse echo received by each point cloud in the above-mentioned airborne LiDAR point cloud data, and obtain the above-mentioned total laser pulse echo number; the fourth determination module is used to determine the echo number of each of the above-mentioned laser pulse echoes, and determine the above-mentioned echo ratio based on the above-mentioned total laser pulse echo number and the above-mentioned echo number.

[0123] An optional solution, the second acquisition unit includes a fifth determination module and a sixth determination module, the fifth determination module is used to obtain a preset number of preset scales, and determine the spherical neighborhoods with each of the above preset scales as the radius according to the above airborne LiDAR point cloud data; the sixth determination module is used to determine the local geometric features of the above spherical neighborhoods corresponding to each preset scale to obtain the above multi-scale features.

[0124] An optional solution, the training unit includes a classification processing module and a training module, the classification processing module is used to classify the above-mentioned airborne LiDAR point cloud data to obtain labeled samples and unlabeled samples, wherein the above-mentioned labeled samples represent samples with the above-mentioned power lines labeled, and the above-mentioned unlabeled samples represent samples without the above-mentioned power lines labeled; the training module is used to use the above-mentioned labeled samples and unlabeled samples and the above-mentioned feature set to train the above-mentioned forest binary classifier, and then in the process of training the above-mentioned random forest binary classifier, use the above-mentioned multispectral orthophoto to verify the output result of the above-mentioned random forest binary classifier.

[0125] In an optional solution, the device further includes a determination unit for determining the position and shape of the power line according to the vector result of the power line after generating the vector result of the power line, so as to implement inspection of the power line.

[0126] The power line target extraction device based on random forest and one-class classification strategy includes a processor and a memory. The first acquisition unit, the second acquisition unit, the training unit and the classification unit are stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0127] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the problem that the existing model is difficult to accurately predict the actual existence of the power line is solved by adjusting the kernel parameters.

[0128] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0129] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the power line target extraction method based on random forest and one-class classification strategy.

[0130] Specifically, the power line target extraction method based on random forest and one-class classification strategy includes:

[0131] Step S201, obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by the drone multi-sensor platform;

[0132] Step S202, acquiring a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual;

[0133] Step S203, according to the random forest algorithm and the one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier to obtain a PBL-RandomForest model;

[0134] Step S204, obtaining the airborne point cloud data to be classified, using the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, and generating a vector result of the power line.

[0135] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes the power line target extraction method based on random forest and one-class classification strategy when running.

[0136] Specifically, the power line target extraction method based on random forest and one-class classification strategy includes:

[0137] Step S201, obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by the drone multi-sensor platform;

[0138] Step S202, acquiring a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual;

[0139] Step S203, according to the random forest algorithm and the one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier to obtain a PBL-RandomForest model;

[0140] Step S204, obtaining the airborne point cloud data to be classified, using the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, and generating a vector result of the power line.

[0141] An embodiment of the present invention provides an electronic device, the device comprising a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, at least the following steps are implemented:

[0142] Step S201, obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by the drone multi-sensor platform;

[0143] Step S202, acquiring a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual;

[0144] Step S203, according to the random forest algorithm and the one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier to obtain a PBL-RandomForest model;

[0145] Step S204, obtaining the airborne point cloud data to be classified, using the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, and generating a vector result of the power line.

[0146] The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0147] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0148] Step S201, obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by the drone multi-sensor platform;

[0149] Step S202, acquiring a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual;

[0150] Step S203, according to the random forest algorithm and the one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier to obtain a PBL-RandomForest model;

[0151] Step S204, obtaining the airborne point cloud data to be classified, using the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, and generating a vector result of the power line.

[0152] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0153] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0154] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0155] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0156] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0157] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0158] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0159] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0160] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0161] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0162] 1) A power line target extraction method based on random forest and one-class classification strategy of the present application includes: obtaining airborne LiDAR point cloud data and multispectral orthophotos collected by a multi-sensor platform of an unmanned aerial vehicle; obtaining a feature set based on the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality and plane fitting residuals; according to the random forest algorithm and one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophotos are used to train a random forest binary classifier to obtain a PBL-Random Forest model; obtaining airborne point cloud data to be classified, using the PBL-Random Forest model to classify the airborne point cloud data to be classified, and generating vector results of the power lines. The PBL-Random Forest model is obtained by combining the random forest method and one-class classification strategy with the airborne LiDAR point cloud data and feature set. The PBL-Random Forest model is used to predict and identify power lines, which improves the accuracy of identifying power lines and solves the actual problem that the existing technology models are difficult to accurately predict power lines.

[0163] 2) A power line target extraction device based on random forest and one-class classification strategy of the present application includes: a first acquisition unit, used to acquire airborne LiDAR point cloud data and multispectral orthophotos collected by a multi-sensor platform of an unmanned aerial vehicle; a second acquisition unit, used to acquire a feature set based on the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality and plane fitting residuals; a training unit, used to train a random forest binary classifier based on a random forest algorithm and a one-class classification strategy using airborne LiDAR point cloud data, feature sets and multispectral orthophotos to obtain a PBL-Random Forest model; a classification unit, used to acquire airborne point cloud data to be classified, and use the PBL-Random Forest model to classify the airborne point cloud data to be classified to perform power line classification, and generate vector results of the power lines. The PBL-Random Forest model is obtained by combining the random forest method and one-class classification strategy with the airborne LiDAR point cloud data and feature set. The PBL-Random Forest model is used to predict and identify power lines, which improves the accuracy of identifying power lines and solves the actual problem that the existing technology models are difficult to accurately predict power lines.

[0164] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for extracting power line targets based on random forest and one-class classification strategy, characterized in that: include: Obtain airborne LiDAR point cloud data and multispectral orthophotos collected by the drone multi-sensor platform; Acquire a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual; According to the random forest algorithm and the one-class classification strategy, the airborne LiDAR point cloud data, the feature set and the multispectral orthophoto are used to train a random forest binary classifier to obtain a PBL-RandomForest model; The airborne point cloud data to be classified is obtained, and the PBL-Random Forest model is used to perform power line classification on the airborne point cloud data to be classified, so as to generate a vector result of the power line.

2. The method according to claim 1, characterized in that Acquiring height-based features based on the airborne LiDAR point cloud data includes: A progressive triangulated irregular network densification filtering algorithm in LAStools software is used to identify ground points in the airborne LiDAR point cloud data to obtain first echo data and ground echo data; According to the ordinary Kriging method, interpolation processing is performed on the first echo data and the ground echo data respectively to obtain a digital surface model and a digital elevation model; A normalized digital surface model is determined according to the digital surface model and the digital elevation model, a spherical neighborhood is defined for each point on the normalized digital surface model, the height variance of all points in the spherical neighborhood is determined, the height variance is distributed to the center point of the spherical neighborhood, and the height-based feature is obtained.

3. The method according to claim 1, characterized in that Acquiring feature-based features according to the airborne LiDAR point cloud data, including: Defining a spherical neighborhood centered on each point cloud in the airborne LiDAR point cloud data; Determining a local covariance matrix of each of the point clouds according to the spherical neighborhood; Performing eigendecomposition on the local covariance matrix to obtain a plurality of eigenvectors and a plurality of eigenvalues; The relevant features of the eigenvector and the relevant features of the eigenvalue are determined respectively to obtain the feature-based features, wherein the relevant features of the eigenvector include linearity, flatness, sphericity and curvature change, and the relevant features of the eigenvalue include perpendicularity and plane fitting residual.

4. The method according to claim 1, characterized in that: in, The echo-based features include the total number of laser pulse echoes and the echo ratio. The echo-based features are obtained according to the airborne LiDAR point cloud data, including: Determine the value of the laser pulse echo received by each point cloud in the airborne LiDAR point cloud data to obtain the total number of laser pulse echoes; The echo number of each laser pulse echo is determined, and the echo ratio is determined according to the total number of laser pulse echoes and the echo number.

5. The method according to claim 1, characterized in that Acquiring multi-scale features according to the airborne LiDAR point cloud data, including: Obtain a preset number of preset scales, and determine spherical neighborhoods with each of the preset scales as a radius according to the airborne LiDAR point cloud data; The local geometric features of the spherical neighborhood corresponding to each preset scale are determined to obtain the multi-scale features.

6. The method according to claim 1, characterized in that The airborne LiDAR point cloud data, the feature set, and the multispectral orthophoto are used to train a random forest binary classifier, including: Classify the airborne LiDAR point cloud data to obtain labeled samples and unlabeled samples, wherein the labeled samples represent samples with the power lines labeled, and the unlabeled samples represent samples without the power lines labeled; The labeled samples and unlabeled samples and the feature set are used to train the forest binary classifier, and in the process of training the random forest binary classifier, the multispectral orthophoto is used to verify the output result of the random forest binary classifier.

7. The method according to claim 1, characterized in that After generating the vector results of the power lines, the method further comprises: The position and shape of the power line are determined according to the vector result of the power line to implement inspection work on the power line.

8. A power line target extraction device based on random forest and one-class classification strategy, characterized in that: include: The first acquisition unit is used to acquire the airborne LiDAR point cloud data and multispectral orthophotos collected by the UAV multi-sensor platform; A second acquisition unit is used to acquire a feature set according to the airborne LiDAR point cloud data, wherein the feature set includes height-based features, feature-based features, echo-based features, and multi-scale features, and the multi-scale features include height variance, linearity, flatness, sphericity, curvature change, verticality, and plane fitting residual; A training unit is used to train a random forest binary classifier using the airborne LiDAR point cloud data, the feature set, and the multispectral orthophoto according to a random forest algorithm and a one-class classification strategy to obtain a PBL-RandomForest model; The classification unit is used to obtain the airborne point cloud data to be classified, and use the PBL-Random Forest model to perform power line classification on the airborne point cloud data to be classified, so as to generate a vector result of the power line.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the power line target extraction method based on random forest and one-class classification strategy as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing the power line target extraction method based on random forest and one-class classification strategy as described in any one of claims 1 to 7.

Images