Transmission conductor point cloud segmentation method and device

Abstract

The invention discloses a power transmission conductor point cloud segmentation method and device, relates to the field of power transmission line data processing, and solves the technical problem that the point cloud segmentation precision of a power transmission conductor is low due to the fact that conductor characteristics are not considered in the prior art. Obtaining a plurality of corresponding first segmentation data; wherein the first segmentation model is obtained by training based on an initial segmentation model, and a linear constraint layer is arranged in each module in the initial segmentation model; a plurality of connectivity scores are obtained through corresponding calculation according to the plurality of pieces of first segmentation data, and the plurality of pieces of first segmentation data are screened according to the plurality of connectivity scores to obtain second segmentation data; and obtaining a segmentation result of the point cloud data of the plurality of transmission conductors according to the second segmentation data. According to the invention, the point cloud segmentation precision of the transmission conductor can be improved.

Description

Technical Field

[0001] The present application relates to the field of power transmission line data processing, and in particular to a method and device for segmenting a power transmission line point cloud. Background Art

[0002] When conducting line risk assessment on transmission lines, it is necessary to understand the structure and surrounding environment of the transmission lines. When understanding the structure of transmission lines, the collection and segmentation of point clouds of transmission lines are essential pre-processes. Among them, point cloud segmentation of transmission lines refers to the classification of various infrastructures in transmission lines into corresponding categories, such as conductors, towers, crossarms and insulators. As a relatively large facility in transmission lines, effective point cloud segmentation of transmission lines can help assess the line risk of transmission lines at distances. In the existing point cloud segmentation technology for transmission lines, there are the following defects: on the one hand, the point cloud data of transmission lines obtained by drone inspection and three-dimensional laser scanning technology may have problems of uneven density, low resolution and serious noise interference; on the other hand, the inherent characteristics of the slenderness of the transmission lines themselves and the interweaving with other infrastructures hinder the effective identification and segmentation of point cloud data, increasing the complexity of point cloud segmentation.

[0003] The point cloud segmentation methods in the prior art include point cloud segmentation methods based on edge information and point cloud segmentation methods based on neural networks. The point cloud segmentation method based on edge information identifies edge information by identifying brightness mutation points and then identifies the transmission line to complete point cloud segmentation. The robustness and segmentation accuracy of point cloud data with uneven density or severe noise are very low. The point cloud segmentation method based on neural networks autonomously learns point cloud features through neural networks, and the segmentation accuracy is higher than that of the point cloud segmentation method based on edge information. However, the learning of neural networks is random, and it is difficult to accurately learn the inherent characteristics of transmission lines, resulting in actual point cloud segmentation accuracy that is lower than expected. Both methods have failed to solve the above technical defects and improve the point cloud segmentation accuracy of transmission lines. Therefore, the technical problem of how to effectively utilize the characteristics of transmission lines to improve the point cloud segmentation accuracy of transmission lines still needs to be solved. Summary of the invention

[0004] The present application provides a method and device for segmenting a point cloud of a power transmission line, so as to solve the technical problem that the prior art fails to consider the characteristics of the line, resulting in low accuracy of the point cloud segmentation of the power transmission line.

[0005] According to a first aspect of an embodiment of the present application, a method for segmenting a transmission line point cloud is provided, comprising:

[0006] Inputting a plurality of transmission line point cloud data to be segmented into a first segmentation model to obtain a plurality of corresponding first segmentation data; wherein the first segmentation model is obtained by training based on an initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model;

[0007] According to the plurality of first segmented data, a plurality of connectivity scores are correspondingly calculated, and the plurality of first segmented data are screened according to the plurality of connectivity scores to obtain second segmented data;

[0008] A segmentation result of the plurality of transmission line point cloud data is obtained according to the second segmentation data.

[0009] The present application first inputs a plurality of transmission line point cloud data to be segmented into a first segmentation model obtained by training an initial segmentation model based on which a linear constraint layer is set for each module to obtain a plurality of first segmentation data. The linear constraint layer can force the model to learn the linear characteristics of the wire, thereby improving the accuracy of the model output. The plurality of first segmentation data are then screened by calculating the connectivity score to obtain the second segmentation data, thereby obtaining the segmentation results of the plurality of transmission line point clouds. By calculating the connectivity and performing post-screening processing, other infrastructure other than the transmission line can be mathematically excluded, thereby improving the accuracy of the second segmentation data and thus improving the point cloud segmentation accuracy of the transmission line.

[0010] In certain embodiments of the present application, before inputting the plurality of transmission line point cloud data to be segmented into the first segmentation model, the method further includes:

[0011] Performing Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds;

[0012] Performing feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features; wherein the categories of the line point cloud features include point cloud local curvature and point cloud normal vector;

[0013] The plurality of first transmission line point clouds are combined with the corresponding plurality of line point cloud features to obtain a plurality of transmission line point cloud data.

[0014] The present application first performs Gaussian denoising on multiple initial transmission line point clouds to obtain multiple first transmission line point clouds, which can remove point cloud noise and reduce the interference of point cloud noise on point cloud segmentation. Then, feature construction is performed on the multiple first transmission line point clouds, and the multiple first transmission line point clouds are combined with the constructed multiple line point cloud features, which can increase the feature expression ability of the transmission line point cloud, thereby improving the accuracy of subsequent segmentation using the model.

[0015] In certain embodiments of the present application, the performing feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features specifically includes:

[0016] The local curvature of the point cloud is specifically:

[0017]

[0018] Among them, δ i is the local curvature of the point cloud of the i-th first transmission line point cloud, λ i0 ,λi1,λ i2 is the local eigenvalue of the i-th first transmission line point cloud with a preset first radius as the neighborhood;

[0019] The point cloud normal vector is specifically:

[0020]

[0021] Among them, M i is the point cloud covariance matrix of the i-th first transmission line point cloud with the preset second radius as the neighborhood, ±[A i , B i , C i ] T is the point cloud normal vector of the i-th first transmission line point cloud.

[0022] In certain embodiments of the present application, the first segmentation model is obtained by training based on the initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model, specifically including:

[0023] Input the historical transmission line point cloud data into an initial segmentation model with preset initial parameters, perform iterative training based on linear semantic fusion loss, optimize the initial parameters during iteration, complete the training when the initial parameters of the initial segmentation model reach the preset values, and obtain and output the first segmentation model;

[0024] The linear constraint layer is specifically:

[0025] L(x) = Wx + b;

[0026] Among them, L(x) is the output of the linear constraint layer, W is the weight matrix, x is the input vector of the linear constraint layer, and b is the bias term;

[0027] The linear semantic fusion loss is specifically:

[0028] L=αL seg +βL linear ;

[0029] Among them, L, L seg , L linearThey are linear semantic fusion loss, semantic segmentation loss and linear constraint penalty respectively. α and β are semantic segmentation weight and linear constraint weight respectively.

[0030] By setting a linear constraint layer in each module of the initial segmentation model, the present application can force the model to learn the linear features of the conductor, thereby improving the accuracy of the model output. At the same time, the initial segmentation model is iteratively trained based on the linear semantic fusion loss. While ensuring the semantic segmentation accuracy of the model, the model's recognition and segmentation accuracy of the linear features of the conductor can be improved, thereby improving the accuracy of the model output and improving the point cloud segmentation accuracy of the transmission line.

[0031] In certain embodiments of the present application, the step of calculating a plurality of connectivity scores according to the plurality of first segmented data, and filtering the plurality of first segmented data according to the plurality of connectivity scores to obtain the second segmented data specifically includes:

[0032] Acquire corresponding multiple connected areas according to the multiple transmission line point cloud data;

[0033] Corresponding to the plurality of first segmented data, respectively calculating connectivity scores of the plurality of connected regions;

[0034] The connectivity score is specifically:

[0035]

[0036] Among them, C i is the i-th connected region, F(C i ) is the connectivity score of the i-th connected area, Area(C i ) is the area of ​​the i-th connected region, Perimeter(C i ) is the perimeter of the i-th connected region, Boundary(C i ) represents the set of arcs that form the boundary of the i-th connected region, is the maximum curvature of the boundary of the ith connected region, is the boundary curvature of arc s, T(s), T′(s) are the unit tangent vector and the derivative of the unit tangent vector of arc s, w1, w2 are weight coefficients;

[0037] According to the connectivity scores of the plurality of connected regions, the plurality of first segmented data are screened based on a preset threshold to obtain second segmented data.

[0038] The present application first obtains corresponding multiple connected areas based on multiple transmission line point cloud data, then respectively calculates the connectivity scores of the multiple connected areas, and screens the multiple first segmented data according to the connectivity scores of each connected area and a preset threshold to obtain second segmented data. Other infrastructure other than transmission lines can be excluded by mathematical means, thereby improving the accuracy of the second segmented data, and further improving the point cloud segmentation accuracy of the transmission line.

[0039] According to a second aspect of the implementation of the present application, there is provided a transmission line point cloud segmentation device, comprising a model segmentation module, a score screening module and a result acquisition module;

[0040] The model segmentation module is used to input a plurality of transmission line point cloud data to be segmented into a first segmentation model to obtain a plurality of corresponding first segmentation data; wherein the first segmentation model is obtained by training based on an initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model;

[0041] The score screening module is used to calculate a plurality of connectivity scores according to the plurality of first segmented data, and screen the plurality of first segmented data according to the plurality of connectivity scores to obtain second segmented data;

[0042] The result acquisition module is used to obtain the segmentation results of the plurality of transmission line point cloud data according to the second segmentation data.

[0043] In some embodiments of the present application, a point cloud preprocessing module is also included; the point cloud preprocessing module includes a Gaussian denoising unit, a feature construction unit and a feature combination unit;

[0044] The Gaussian denoising unit is used to perform Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds;

[0045] The feature construction unit is used to perform feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features; wherein the categories of the line point cloud features include point cloud local curvature and point cloud normal vector;

[0046] The feature combination unit is used to combine the multiple first transmission line point clouds with the corresponding multiple line point cloud features to obtain multiple transmission line point cloud data.

[0047] In certain embodiments of the present application, when the feature construction unit is used to perform feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features, the step specifically includes:

[0048] The local curvature of the point cloud is specifically:

[0049]

[0050] Among them, δ i is the local curvature of the point cloud of the i-th first transmission line point cloud, λ i0 ,λ i1 ,λ i2 is the local eigenvalue of the i-th first transmission line point cloud with a preset first radius as the neighborhood;

[0051] The point cloud normal vector is specifically:

[0052]

[0053] Among them, M i is the point cloud covariance matrix of the i-th first transmission line point cloud with the preset second radius as the neighborhood, ±[A i , B i , C i ] T is the point cloud normal vector of the i-th first transmission line point cloud.

[0054] In certain embodiments of the present application, the first segmentation model is obtained by training based on the initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model, specifically including:

[0055] Input the historical transmission line point cloud data into an initial segmentation model with preset initial parameters, perform iterative training based on linear semantic fusion loss, optimize the initial parameters during iteration, complete the training when the initial parameters of the initial segmentation model reach the preset values, and obtain and output the first segmentation model;

[0056] The linear constraint layer is specifically:

[0057] L(x) = Wx + b;

[0058] Among them, L(x) is the output of the linear constraint layer, W is the weight matrix, x is the input vector of the linear constraint layer, and b is the bias term;

[0059] The linear semantic fusion loss is specifically:

[0060] L=αL seg +βL linear ;

[0061] Among them, L, L seg , L linear They are linear semantic fusion loss, semantic segmentation loss and linear constraint penalty respectively. α and β are semantic segmentation weight and linear constraint weight respectively.

[0062] In certain embodiments of the present application, the score screening module includes a connected region acquisition unit, a connected score calculation unit, and a point cloud data screening unit;

[0063] The connected area acquisition unit is used to acquire the corresponding multiple connected areas according to the multiple transmission line point cloud data;

[0064] The connectivity score calculation unit is used to calculate the connectivity scores of the plurality of connected regions respectively according to the plurality of first segmented data;

[0065] The connectivity score is specifically:

[0066]

[0067] Among them, C i is the i-th connected region, P(C i ) is the connectivity score of the i-th connected area, Area(C i ) is the area of ​​the i-th connected region, Perimeter(C i ) is the perimeter of the i-th connected region, Boundary(C i ) represents the set of arcs that form the boundary of the i-th connected region, is the maximum curvature of the boundary of the ith connected region, is the boundary curvature of arc s, T(s), T′(s) are the unit tangent vector and the derivative of the unit tangent vector of arc s, w1, w2 are weight coefficients;

[0068] The point cloud data screening unit is used to screen the plurality of first segmented data based on a preset threshold according to the connectivity scores of the plurality of connected areas to obtain second segmented data.

[0069] The present application first inputs a plurality of transmission line point cloud data to be segmented into a first segmentation model obtained by training an initial segmentation model based on which a linear constraint layer is set for each module to obtain a plurality of first segmentation data. The linear constraint layer can force the model to learn the linear characteristics of the wire, thereby improving the accuracy of the model output. The plurality of first segmentation data are then screened by calculating the connectivity score to obtain the second segmentation data, thereby obtaining the segmentation results of the plurality of transmission line point clouds. By calculating the connectivity and performing post-screening processing, other infrastructure other than the transmission line can be mathematically excluded, thereby improving the accuracy of the second segmentation data and thus improving the point cloud segmentation accuracy of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 : A schematic diagram of a flow chart of a method for segmenting a transmission line point cloud according to certain embodiments of the present application;

[0071] Figure 2 : A module structure diagram of a transmission line point cloud segmentation device shown in certain embodiments of the present application. DETAILED DESCRIPTION

[0072] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below in conjunction with the accompanying drawings are exemplary and are only used to explain some embodiments of the present application, and should not be construed as limitations on the embodiments of the present application. Based on the embodiments shown in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0073] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, unless otherwise clearly and specifically defined, "multiple" and "several" mean two or more.

[0074] In order to assess the line risk of transmission lines, it is necessary to effectively segment the point cloud data of transmission lines and classify different infrastructures into corresponding categories. However, there is a problem of noise interference in the point cloud segmentation of transmission lines. At the same time, the linear characteristics of slender transmission lines and their interweaving with other infrastructures also hinder the effectiveness and efficiency of point cloud segmentation, increasing the complexity of point cloud segmentation.

[0075] In the existing technology, the point cloud segmentation method based on edge information is difficult to adapt to the environment with severe noise, and the point cloud segmentation method based on neural network is difficult to learn the inherent characteristics of the transmission line, and the actual point cloud segmentation accuracy is not as good as expected. Both methods are difficult to improve the point cloud segmentation accuracy of the transmission line. Therefore, how to effectively utilize the characteristics of the transmission line to improve the point cloud segmentation accuracy of the transmission line still needs to be solved.

[0076] Based on the above technical background, please refer to Figure 1 The embodiment of the present application provides a method for segmenting a transmission line point cloud, including steps S101 to S103, each of which is specifically as follows:

[0077] Step S101: input multiple transmission line point cloud data to be segmented into a first segmentation model to obtain corresponding multiple first segmentation data; wherein the first segmentation model is trained based on an initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model.

[0078] In certain embodiments of the present application, before inputting the plurality of transmission line point cloud data to be segmented into the first segmentation model, the method further includes:

[0079] Performing Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds;

[0080] Performing feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features; wherein the categories of the line point cloud features include point cloud local curvature and point cloud normal vector;

[0081] The plurality of first transmission line point clouds are combined with the corresponding plurality of line point cloud features to obtain a plurality of transmission line point cloud data.

[0082] In certain embodiments of the present application, performing Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds specifically includes:

[0083] Selecting one initial transmission line point cloud in turn as the current point cloud, and obtaining a first transmission line point cloud corresponding to the current point cloud according to a preset denoising method, until the processing of the multiple initial transmission line point clouds is completed;

[0084] Wherein, the denoising method comprises:

[0085] The center point of the current point cloud is used as the denoising center point, the denoising center point is used as the sphere center, and the sphere with a preset length as the radius is used as the neighborhood of the current point cloud. The current point cloud is subjected to Gaussian filtering and denoising to obtain the first transmission line point cloud corresponding to the current point cloud;

[0086] The first transmission line point cloud is specifically:

[0087]

[0088] Among them, P f is the filtered point cloud, P i is the coordinate value of the i-th neighborhood point in the neighborhood, N is the total number of neighborhood points in the neighborhood, is the Gaussian kernel, σ is the standard deviation, and d is the distance from the i-th neighborhood point to the denoising center point.

[0089] In some optional embodiments of the present application, the preferred default value of the standard deviation is 1.

[0090] In certain embodiments of the present application, the performing feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features specifically includes:

[0091] The local curvature of the point cloud is specifically:

[0092]

[0093] Among them, δ i is the local curvature of the point cloud of the i-th first transmission line point cloud, λ i0 ,λ i1 ,λ i2 is the local eigenvalue of the i-th first transmission line point cloud with a preset first radius as the neighborhood;

[0094] The point cloud normal vector is specifically:

[0095]

[0096] Among them, M i is the point cloud covariance matrix of the i-th first transmission line point cloud with the preset second radius as the neighborhood, ±[A i , B i , C i ] T is the point cloud normal vector of the i-th first transmission line point cloud.

[0097] The present application first performs Gaussian denoising on multiple initial transmission line point clouds to obtain multiple first transmission line point clouds, which can remove point cloud noise and reduce the interference of point cloud noise on point cloud segmentation. Then, feature construction is performed on the multiple first transmission line point clouds, and the multiple first transmission line point clouds are combined with the constructed multiple line point cloud features, which can increase the feature expression ability of the transmission line point cloud, thereby improving the accuracy of subsequent segmentation using the model.

[0098] In certain embodiments of the present application, the first segmentation model is obtained by training based on the initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model, specifically including:

[0099] Input the historical transmission line point cloud data into an initial segmentation model with preset initial parameters, perform iterative training based on linear semantic fusion loss, optimize the initial parameters during iteration, complete the training when the initial parameters of the initial segmentation model reach the preset values, and obtain and output the first segmentation model;

[0100] The linear constraint layer is specifically:

[0101] L(x) = Wx + b;

[0102] Among them, L(x) is the output of the linear constraint layer, W is the weight matrix, x is the input vector of the linear constraint layer, and b is the bias term;

[0103] The linear semantic fusion loss is specifically:

[0104] L=αLseg +βL linear ;

[0105] Among them, L, L seg , L linear They are linear semantic fusion loss, semantic segmentation loss and linear constraint penalty respectively. α and β are semantic segmentation weight and linear constraint weight respectively.

[0106] In some optional implementations of the present application, the preferred value scheme of the semantic segmentation weight is 0.7, and the preferred value scheme of the linear constraint weight is 0.3.

[0107] By setting a linear constraint layer in each module of the initial segmentation model, the present application can force the model to learn the linear features of the conductor, thereby improving the accuracy of the model output. At the same time, the initial segmentation model is iteratively trained based on the linear semantic fusion loss. While ensuring the semantic segmentation accuracy of the model, the model's recognition and segmentation accuracy of the linear features of the conductor can be improved, thereby improving the accuracy of the model output and improving the point cloud segmentation accuracy of the transmission line.

[0108] Step S102: Calculate and obtain a plurality of connectivity scores according to the plurality of first segmented data, and filter the plurality of first segmented data according to the plurality of connectivity scores to obtain second segmented data.

[0109] In certain embodiments of the present application, the step of calculating a plurality of connectivity scores according to the plurality of first segmented data, and filtering the plurality of first segmented data according to the plurality of connectivity scores to obtain the second segmented data specifically includes:

[0110] Acquire corresponding multiple connected areas according to the multiple transmission line point cloud data;

[0111] Corresponding to the plurality of first segmented data, respectively calculating connectivity scores of the plurality of connected regions;

[0112] The connectivity score is specifically:

[0113]

[0114] Among them, C i is the i-th connected region, F(C i ) is the connectivity score of the i-th connected area, Area(C i ) is the area of ​​the i-th connected region, Perimeter(C i ) is the perimeter of the i-th connected region, Boundary(C i ) represents the set of arcs that form the boundary of the i-th connected region, is the maximum curvature of the boundary of the ith connected region, is the boundary curvature of arc s, T(s), T′(s) are the unit tangent vector and the derivative of the unit tangent vector of arc s, w1, w2 are weight coefficients;

[0115] According to the connectivity scores of the plurality of connected regions, the plurality of first segmented data are screened based on a preset threshold to obtain second segmented data.

[0116] In certain optional implementations of the present application, the preferred values ​​of the weight coefficients w1 and w2 are 0.35 and 0.65.

[0117] In certain embodiments of the present application, the filtering of the plurality of first segmented data based on a preset threshold according to the connectivity scores of the plurality of connected regions to obtain the second segmented data specifically includes:

[0118] Selecting one first segmented data in turn as current processing data, and obtaining a screening result corresponding to the current processing data according to a preset screening method, until the screening of the plurality of first segmented data is completed, and obtaining and outputting the second segmented data;

[0119] Wherein, the preset screening method includes:

[0120] If the connectivity score of the connected area corresponding to the currently processed data is higher than the preset threshold, the screening result of the currently processed data is "retained";

[0121] If the connectivity score of the connected area corresponding to the currently processed data is not higher than the preset threshold, the screening result of the currently processed data is "removed".

[0122] Obviously, the second segmented data is the data whose screening result is "retained" in the first segmented data.

[0123] In certain optional implementations of the present application, the preferred value scheme for the preset threshold is 0.6.

[0124] Considering calculating the connectivity score of the connected area corresponding to the first segmentation data is because the linear characteristics of the transmission line will cause the boundary curvature of the transmission line to be lower than that of other infrastructure. By calculating the connectivity score and based on the relationship between the connectivity score and the preset threshold, the first segmentation data can be screened and optimized, thereby achieving the purpose of improving the segmentation accuracy of the transmission line point cloud.

[0125] The present application first obtains corresponding multiple connected areas based on multiple transmission line point cloud data, then respectively calculates the connectivity scores of the multiple connected areas, and screens the multiple first segmented data according to the connectivity scores of each connected area and a preset threshold to obtain second segmented data. Other infrastructure other than transmission lines can be excluded by mathematical means, thereby improving the accuracy of the second segmented data, and further improving the point cloud segmentation accuracy of the transmission line.

[0126] Step S103: obtaining a segmentation result of the plurality of transmission line point cloud data according to the second segmentation data.

[0127] In some optional embodiments of the present application, a preferred solution of the segmentation result of the plurality of transmission line point cloud data is the second segmentation data.

[0128] The present application first inputs a plurality of transmission line point cloud data to be segmented into a first segmentation model obtained by training an initial segmentation model based on which a linear constraint layer is set for each module to obtain a plurality of first segmentation data. The linear constraint layer can force the model to learn the linear characteristics of the wire, thereby improving the accuracy of the model output. The plurality of first segmentation data are then screened by calculating the connectivity score to obtain the second segmentation data, thereby obtaining the segmentation results of the plurality of transmission line point clouds. By calculating the connectivity and performing post-screening processing, other infrastructure other than the transmission line can be mathematically excluded, thereby improving the accuracy of the second segmentation data and thus improving the point cloud segmentation accuracy of the transmission line.

[0129] Corresponding to the previous method, see Figure 2 , the embodiment of the present application provides a transmission line point cloud segmentation device, including a model segmentation module 210, a score screening module 220 and a result acquisition module 230;

[0130] The model segmentation module 210 is used to input a plurality of transmission line point cloud data to be segmented into a first segmentation model to obtain a plurality of corresponding first segmentation data; wherein the first segmentation model is obtained by training based on an initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model;

[0131] The score screening module 220 is used to calculate a plurality of connectivity scores according to the plurality of first segmented data, and screen the plurality of first segmented data according to the plurality of connectivity scores to obtain second segmented data;

[0132] The result acquisition module 230 is used to obtain the segmentation result of the plurality of transmission line point cloud data according to the second segmentation data.

[0133] In some embodiments of the present application, a point cloud preprocessing module 240 is further included; the point cloud preprocessing module 240 includes a Gaussian denoising unit 241, a feature construction unit 242 and a feature combination unit 243;

[0134] The Gaussian denoising unit 241 is used to perform Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds;

[0135] The feature construction unit 242 is used to perform feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features; wherein the categories of the line point cloud features include point cloud local curvature and point cloud normal vector;

[0136] The feature combination unit 243 is used to combine the multiple first transmission line point clouds with the corresponding multiple line point cloud features to obtain multiple transmission line point cloud data.

[0137] In certain embodiments of the present application, when the feature construction unit 242 is used to perform feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features, the steps specifically include:

[0138] The local curvature of the point cloud is specifically:

[0139]

[0140] Among them, δ i is the local curvature of the point cloud of the i-th first transmission line point cloud, λ i0 ,λ i1 ,λ i2 is the local eigenvalue of the i-th first transmission line point cloud with a preset first radius as the neighborhood;

[0141] The point cloud normal vector is specifically:

[0142]

[0143] Among them, M i is the point cloud covariance matrix of the i-th first transmission line point cloud with the preset second radius as the neighborhood, ±[A i , B i , C i ] T is the point cloud normal vector of the i-th first transmission line point cloud.

[0144] In certain embodiments of the present application, the first segmentation model is obtained by training based on the initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model, specifically including:

[0145] Input the historical transmission line point cloud data into an initial segmentation model with preset initial parameters, perform iterative training based on linear semantic fusion loss, optimize the initial parameters during iteration, complete the training when the initial parameters of the initial segmentation model reach the preset values, and obtain and output the first segmentation model;

[0146] The linear constraint layer is specifically:

[0147] L(x) = Wx + b;

[0148] Among them, L(x) is the output of the linear constraint layer, W is the weight matrix, x is the input vector of the linear constraint layer, and b is the bias term;

[0149] The linear semantic fusion loss is specifically:

[0150] L=αL seg +βL linear ;

[0151] Among them, L, L seg , L linear They are linear semantic fusion loss, semantic segmentation loss and linear constraint penalty respectively. α and β are semantic segmentation weight and linear constraint weight respectively.

[0152] In some embodiments of the present application, the score screening module 220 includes a connected region acquisition unit 221, a connected score calculation unit 222, and a point cloud data screening unit 223;

[0153] The connected area acquisition unit 221 is used to acquire corresponding multiple connected areas according to the multiple transmission line point cloud data;

[0154] The connectivity score calculation unit 222 is used to calculate the connectivity scores of the plurality of connected regions respectively according to the plurality of first segmented data;

[0155] The connectivity score is specifically:

[0156]

[0157] Among them, C i is the i-th connected region, F(C i ) is the connectivity score of the i-th connected area, Area(C i ) is the area of ​​the i-th connected region, Perimeter(C i ) is the perimeter of the i-th connected region, Boundary(C i ) represents the set of arcs that form the boundary of the i-th connected region, is the maximum curvature of the boundary of the ith connected region, is the boundary curvature of arc s, T(s), T′(s) are the unit tangent vector and the derivative of the unit tangent vector of arc s, w1, w2 are weight coefficients;

[0158] The point cloud data screening unit 223 is used to screen the plurality of first segmented data based on a preset threshold according to the connectivity scores of the plurality of connected regions to obtain second segmented data.

[0159] The present application first inputs a plurality of transmission line point cloud data to be segmented into a first segmentation model obtained by training an initial segmentation model based on which a linear constraint layer is set for each module to obtain a plurality of first segmentation data. The linear constraint layer can force the model to learn the linear characteristics of the wire, thereby improving the accuracy of the model output. The plurality of first segmentation data are then screened by calculating the connectivity score to obtain the second segmentation data, thereby obtaining the segmentation results of the plurality of transmission line point clouds. By calculating the connectivity and performing post-screening processing, other infrastructure other than the transmission line can be mathematically excluded, thereby improving the accuracy of the second segmentation data and thus improving the point cloud segmentation accuracy of the transmission line.

[0160] It should be understood that the device provided in the embodiment of the present application corresponds to the aforementioned method, and a transmission line point cloud segmentation device provided in the embodiment of the present application can implement a transmission line point cloud segmentation method provided in any embodiment of the present application.

[0161] Adaptively, the embodiments of the present application also provide a computer device and a computer-readable storage medium.

[0162] The computer device comprises: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor;

[0163] Wherein, when the processor executes the computer program, a transmission line point cloud segmentation method of the present application is implemented.

[0164] The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute a transmission line point cloud segmentation method of the present application.

[0165] The above are some implementation methods of the present application, and the purpose, technical scheme and beneficial effects of the present application are further described in detail. It should be clear that the above-mentioned partial implementation methods of the present application cannot be understood as limiting the present application. It is particularly pointed out that for those skilled in the art, any changes, modifications, equivalent substitutions and variations made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for segmenting a transmission line point cloud, characterized in that: include: Inputting a plurality of transmission line point cloud data to be segmented into a first segmentation model to obtain a plurality of corresponding first segmentation data; wherein the first segmentation model is obtained by training based on an initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model; According to the plurality of first segmented data, a plurality of connectivity scores are correspondingly calculated, and the plurality of first segmented data are screened according to the plurality of connectivity scores to obtain second segmented data; A segmentation result of the plurality of transmission line point cloud data is obtained according to the second segmentation data.

2. A method for segmenting a transmission line point cloud according to claim 1, characterized in that: Before inputting the plurality of transmission line point cloud data to be segmented into the first segmentation model, the method further includes: Performing Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds; Performing feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features; wherein the categories of the line point cloud features include point cloud local curvature and point cloud normal vector; The plurality of first transmission line point clouds are combined with the corresponding plurality of line point cloud features to obtain a plurality of transmission line point cloud data.

3. A method for segmenting a transmission line point cloud according to claim 2, characterized in that: The step of constructing features on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features specifically includes: The local curvature of the point cloud is specifically: Among them, δ i is the local curvature of the point cloud of the i-th first transmission line point cloud, λ i0 ,λ i1 ,λ i2 is the local eigenvalue of the i-th first transmission line point cloud with a preset first radius as the neighborhood; The point cloud normal vector is specifically: Among them, M i is the point cloud covariance matrix of the i-th first transmission line point cloud with the preset second radius as the neighborhood, ±[A i , B i , ,C i ] T is the point cloud normal vector of the i-th first transmission line point cloud.

4. A method for segmenting a transmission line point cloud according to claim 1, characterized in that: The first segmentation model is obtained by training based on the initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model, specifically including: Input the historical transmission line point cloud data into an initial segmentation model with preset initial parameters, perform iterative training based on linear semantic fusion loss, optimize the initial parameters during iteration, complete the training when the initial parameters of the initial segmentation model reach the preset values, and obtain and output the first segmentation model; The linear constraint layer is specifically: L(x) = Wx + b; Among them, L(x) is the output of the linear constraint layer, W is the weight matrix, x is the input vector of the linear constraint layer, and b is the bias term; The linear semantic fusion loss is specifically: L=αL seg +βL linear ; Among them, L, L seg , L linear They are linear semantic fusion loss, semantic segmentation loss and linear constraint penalty respectively. α and β are semantic segmentation weight and linear constraint weight respectively.

5. A method for segmenting a transmission line point cloud according to claim 1, characterized in that: The step of calculating a plurality of connectivity scores according to the plurality of first segmented data, and filtering the plurality of first segmented data according to the plurality of connectivity scores to obtain second segmented data specifically includes: Acquire corresponding multiple connected areas according to the multiple transmission line point cloud data; Corresponding to the plurality of first segmented data, respectively calculating connectivity scores of the plurality of connected regions; The connectivity score is specifically: Among them, C i is the i-th connected region, F(C i ) is the connectivity score of the i-th connected area, Area(C i ) is the area of ​​the i-th connected region, Perimeter(C i ) is the perimeter of the i-th connected region, Boundary(C i ) represents the set of arcs that form the boundary of the i-th connected region, is the maximum curvature of the boundary of the ith connected region, is the boundary curvature of arc s, T(s), T′(s) are the unit tangent vector and the derivative of the unit tangent vector of arc s, w1, w2 are weight coefficients; According to the connectivity scores of the plurality of connected regions, the plurality of first segmented data are screened based on a preset threshold to obtain second segmented data.

6. A transmission line point cloud segmentation device, characterized in that: It includes model segmentation module, score screening module and result acquisition module; The model segmentation module is used to input a plurality of transmission line point cloud data to be segmented into a first segmentation model to obtain a plurality of corresponding first segmentation data; wherein the first segmentation model is obtained by training based on an initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model; The score screening module is used to calculate a plurality of connectivity scores according to the plurality of first segmented data, and screen the plurality of first segmented data according to the plurality of connectivity scores to obtain second segmented data; The result acquisition module is used to obtain the segmentation results of the plurality of transmission line point cloud data according to the second segmentation data.

7. A transmission line point cloud segmentation device according to claim 6, characterized in that: It also includes a point cloud preprocessing module; the point cloud preprocessing module includes a Gaussian denoising unit, a feature construction unit and a feature combination unit; The Gaussian denoising unit is used to perform Gaussian denoising on the multiple initial transmission line point clouds to be segmented to obtain multiple first transmission line point clouds; The feature construction unit is used to perform feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features; wherein the categories of the line point cloud features include point cloud local curvature and point cloud normal vector; The feature combination unit is used to combine the multiple first transmission line point clouds with the corresponding multiple line point cloud features to obtain multiple transmission line point cloud data.

8. The transmission line point cloud segmentation device according to claim 7, characterized in that: When the feature construction unit is used to perform feature construction on the plurality of first transmission line point clouds to obtain a plurality of line point cloud features, the feature construction unit specifically includes: The local curvature of the point cloud is specifically: Among them, δ i is the local curvature of the point cloud of the i-th first transmission line point cloud, λ i0 ,λ i1 ,λ i2 is the local eigenvalue of the i-th first transmission line point cloud with a preset first radius as the neighborhood; The point cloud normal vector is specifically: Among them, M i is the point cloud covariance matrix of the i-th first transmission line point cloud with the preset second radius as the neighborhood, ±[A i , B i , C i ] T is the point cloud normal vector of the i-th first transmission line point cloud.

9. The transmission line point cloud segmentation device according to claim 6, characterized in that: The first segmentation model is obtained by training based on the initial segmentation model, and a linear constraint layer is set in each module of the initial segmentation model, specifically including: Input the historical transmission line point cloud data into an initial segmentation model with preset initial parameters, perform iterative training based on linear semantic fusion loss, optimize the initial parameters during iteration, complete the training when the initial parameters of the initial segmentation model reach the preset values, and obtain and output the first segmentation model; The linear constraint layer is specifically: L(x) = Wx + b; Among them, L(x) is the output of the linear constraint layer, W is the weight matrix, x is the input vector of the linear constraint layer, and b is the bias term; The linear semantic fusion loss is specifically: L=αL seg +βL linear ; Among them, L, L seg , L linear They are linear semantic fusion loss, semantic segmentation loss and linear constraint penalty respectively. α and β are semantic segmentation weight and linear constraint weight respectively.

10. The transmission line point cloud segmentation device according to claim 6, characterized in that: The score screening module includes a connected area acquisition unit, a connected score calculation unit and a point cloud data screening unit; The connected area acquisition unit is used to acquire the corresponding multiple connected areas according to the multiple transmission line point cloud data; The connectivity score calculation unit is used to calculate the connectivity scores of the plurality of connected regions respectively according to the plurality of first segmented data; The connectivity score is specifically: Among them, C i is the i-th connected region, F(C i ) is the connectivity score of the i-th connected area, Area(C i ) is the area of ​​the i-th connected region, Perimeter(C i ) is the perimeter of the i-th connected region, Boundary(C i ) represents the set of arcs that form the boundary of the i-th connected region, is the maximum curvature of the boundary of the ith connected region, is the boundary curvature of arc s, T(s), T′(s) are the unit tangent vector and the derivative of the unit tangent vector of arc s, w1, w2 are weight coefficients; The point cloud data screening unit is used to screen the plurality of first segmented data based on a preset threshold according to the connectivity scores of the plurality of connected areas to obtain second segmented data.

Images