Efficient denoising power transmission line point cloud processing method and system

Abstract

The invention discloses an efficient denoising power transmission line point cloud processing method and system, and relates to the technical field of three-dimensional point cloud data processing and intelligent analysis, and the method comprises the steps: carrying out the segmentation processing of original power transmission line point cloud data, dividing the point cloud into power equipment parts, carrying out the local density analysis of the segmented point cloud, and generating a density distribution diagram; based on the density distribution diagram and the local geometric features of the point clouds, multi-stage filtering is carried out, denoising model parameters are optimized, a statistical filtering method is combined to remove noise points, curvature features and normal vector consistency features of the point clouds are extracted, classification feature vectors are generated, and based on the classification feature vectors, fine classification is carried out on the point clouds. Through data preprocessing, multi-stage filtering, self-supervised learning, feature extraction and refined classification, the technical defects of insufficient multi-modal feature fusion efficiency and poor equipment feature adaptability in power transmission line point cloud denoising are effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-dimensional point cloud data processing and intelligent analysis, and particularly relates to a power transmission line point cloud processing method and system with efficient denoising. BACKGROUND

[0002] In recent years, with the increasing demand for intelligent inspection in the power industry, efficient processing technology of power transmission line point cloud data has gradually become a research hotspot. Point cloud data, as an important form of three-dimensional data, can truly reflect the geometric structure and spatial distribution information of the power transmission line. However, due to the influence of factors such as vegetation and weather in the actual environment, the obtained point cloud data often contains a large amount of noise and redundant information, which seriously affects the subsequent data analysis and processing. Therefore, how to efficiently denoise the power transmission line point cloud data has become the key to improving the efficiency and accuracy of power inspection.

[0003] At present, the power transmission line point cloud denoising technology mainly relies on two categories of methods, namely statistical filtering and geometric analysis. Traditional statistical filtering methods, such as radius filtering and voxel downsampling, are simple and easy to implement, but have significant limitations. Such methods usually rely on fixed threshold parameters and are difficult to distinguish between regular structures of power equipment and dynamic vegetation noise. For example, shock absorbers are easily confused with tree branch point cloud features, resulting in poor denoising effect. Geometric analysis methods understand the geometric structure of point cloud through curvature, normal vector and other features, which improves the denoising accuracy to some extent, but is greatly affected by environmental interference, which easily causes edge blur at the conductor connection and affects the integrity of the equipment features. In recent years, deep learning technology has shown great potential in point cloud denoising. However, due to the lack of feature data sets in the power field, existing deep learning solutions have a domain migration bottleneck in power transmission line point cloud denoising, with low recognition rate and high computational cost, which is difficult to meet the needs of real-time inspection by unmanned aerial vehicles. Although hybrid methods combining statistical filtering and machine learning have appeared in recent years, they have prominent feature segmentation problems and lack dynamic adjustment mechanisms, resulting in insufficient multi-modal feature fusion efficiency, missing key area protection, and low hardware energy efficiency ratio, which have not been effectively solved. SUMMARY

[0004] In view of the above existing problems, the present application provides a power transmission line point cloud processing method and system with efficient denoising, to solve the problems of poor denoising effect, edge blur at the conductor connection caused by large environmental interference, difficulty in meeting the needs of real-time inspection by unmanned aerial vehicles, and lack of dynamic adjustment mechanism leading to insufficient multi-modal feature fusion efficiency in the prior art.

[0005] To solve the above technical problems, a power transmission line point cloud processing method with efficient denoising is proposed, which comprises,

[0006] The original power transmission line point cloud data is segmented, the point cloud is divided into power equipment components, local density analysis is performed on the segmented point cloud, and a density distribution map is generated; based on the density distribution map and the local geometric features of the point cloud, multi-level filtering is performed, and the noise removal model parameters are optimized, and the statistical filtering method is combined to remove noise points; the curvature features and the normal vector consistency features of the point cloud are extracted, and a classification feature vector is generated, and the point cloud is classified based on the classification feature vector.

[0007] As a preferred scheme of the efficient denoising power transmission line point cloud processing method, the segmentation processing includes: based on the local curvature difference and the normal vector angle of the point cloud, the region is expanded, the point cloud is divided into conductors, insulators and tower components based on the geometric features of the point cloud, and the local density analysis is performed on the segmented component point cloud to generate a density distribution map.

[0008] The local density analysis includes constructing a neighborhood centered on each point in each segmented component point cloud, calculating the density distribution of the points in the neighborhood, and generating a density distribution map.

[0009] As a preferred scheme of the efficient denoising power transmission line point cloud processing method, the multi-level filtering includes: based on the density distribution map and the local curvature features of the point cloud, first-level filtering and second-level filtering are performed.

[0010] The first-level filtering includes coarse screening with a large voxel size to reduce the data amount; the second-level filtering includes reducing the grid size when the local curvature is higher than a threshold value and the density is higher than a threshold value, and increasing the grid size when the local curvature is less than or equal to a threshold value and the density is less than or equal to a threshold value.

[0011] As a preferred scheme of the efficient denoising power transmission line point cloud processing method, the optimization of the noise removal model parameters includes reconstructing the point cloud and calculating the reconstruction error, analyzing the neighborhood distance distribution of the point cloud by combining the statistical filtering method, and removing abnormal points.

[0012] The optimization of the noise removal model parameters further includes constructing a point cloud reconstruction task, inputting a noisy point cloud and predicting a clean version, and optimizing the model parameters by calculating the reconstruction error between the predicted point cloud and the real point cloud.

[0013] The statistical filtering method includes calculating the average distance of each point to K nearest neighbors, determining the noise points and removing them when the average distance exceeds a threshold value, and performing secondary denoising on the point cloud by combining the reconstruction error and the statistical filtering result.

[0014] As a preferred scheme of the efficient denoising power transmission line point cloud processing method, the extraction of the curvature features of the point cloud includes calculating the curvature of each point through eigenvalue decomposition of the covariance matrix.

[0015] wherein the covariance matrix is constructed by the coordinates of the neighborhood of the point, and the minimum eigenvalue represents the local curvature; the normal vector consistency feature comprises calculating the variance of the normal vector included angle of each point in the neighborhood, and the region with the variance lower than a preset threshold is determined as the normal vector consistency;

[0016] The curvature formula is calculated as:

[0017]

[0018] wherein δ i is the curvature of the point p i in the point cloud, λ min is the minimum eigenvalue of the covariance matrix, λ1, λ2 and λ3 are eigenvalues of the covariance matrix of the point p i in the point cloud, and i is a variable index;

[0019] The covariance matrix is represented as:

[0020]

[0021] wherein C is the covariance matrix of the point p i in the point cloud, K is the number of neighboring points of the point p i in the point cloud, p j is a point in the neighborhood of the point p i in the point cloud, μ i is the mean vector of the neighboring points of the point p i in the point cloud, T is a transpose operation, and i and j are variable indexes.

[0022] As a preferred scheme of the power transmission line point cloud processing method with high efficiency de-noising, the generating the classification feature vector comprises: based on the optimized de-noising model, extracting the curvature feature and the normal vector consistency feature of the point cloud, and fusing the curvature, the normal vector consistency and the density feature into the classification feature vector through a dynamic weight distribution model;

[0023] The dynamic weight distribution model comprises: based on grid search, iteratively adjusting the weight parameter on the training set, and minimizing the classification loss function, which is represented by the formula:

[0024]

[0025] wherein w i is the feature weight of the point p i in the point cloud, α and β are weight parameters, δ i is the curvature of the point p i in the point cloud, ρ i is the local density of the point p i in the point cloud, and K is the number of neighboring points of the point p ithe number of neighboring points of point p j the local density of points in the neighborhood of point p i j the curvature of points in the neighborhood of point p i i and j are variable indices.

[0026] As a preferred scheme of the power transmission line point cloud processing method with high efficiency de-noising provided by the application, the fine classification of the point cloud comprises fine classification of the point cloud by using a classification feature vector through a deep learning model, and power equipment feature points are retained and residual noise points are removed.

[0027] The deep learning model comprises inputting the classification feature vector into a multi-scale graph convolution network, outputting a point cloud category probability, training the multi-scale graph convolution network through a cross-entropy loss function, and introducing a Dropout layer and L2 regularization, and random rotation, translation and noise disturbance are applied to the input point cloud to simulate complex scene data distribution.

[0028] The multi-scale graph convolution network comprises a first layer of convolution for extracting local geometric features, a second layer of convolution for fusing global topological relationships, and a third layer for outputting classification results through a fully connected layer; and the training of the multi-scale graph convolution network comprises joint optimization of classification loss and reconstruction loss, and end-to-end training through weighted summation.

[0029] As a preferred scheme of the power transmission line point cloud processing system with high efficiency de-noising provided by the application, the system comprises a point cloud segmentation and density analysis module, a multi-level filtering and model optimization module, a feature extraction and dynamic fusion module, and a point cloud classification module.

[0030] The point cloud segmentation and density analysis module comprises a point cloud segmentation unit and a density analysis unit, which are used for dividing original power transmission line point cloud data into power equipment components according to geometric features, calculating the neighborhood density distribution of each point based on a Gaussian kernel function in the segmented component point cloud, and generating a density distribution map.

[0031] The multi-level filtering and model optimization module comprises a multi-level filtering unit, a self-supervised learning unit and a statistical filtering unit, which are used for removing redundant noise points through two-level filtering, optimizing de-noising model parameters through point cloud reconstruction error, removing abnormal points based on neighborhood distance distribution, and combining with the self-supervised learning result for secondary de-noising.

[0032] The feature extraction and dynamic fusion module comprises a covariance matrix feature decomposition unit, a normal vector consistency analysis unit and a dynamic weight allocation unit, which are used for calculating curvature features through covariance matrix decomposition, analyzing normal vector consistency through variance of neighborhood normal vector angles, identifying complex structure regions, dynamically allocating weights according to curvature, density and normal vector consistency, optimizing weight parameters, and minimizing classification loss.​

[0033] The point cloud classification module is used to input the classification feature vector into the multi-scale graph convolutional network, output the point cloud category probability through local and global feature fusion, and perform model training.

[0034] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a method for efficiently denoising a transmission line point cloud are implemented.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for efficiently denoising a transmission line point cloud.

[0036] The beneficial effects of the present invention are as follows: Through geometric feature-based segmentation processing and local density analysis, the present invention accurately divides point clouds into conductors, insulators, and tower power equipment components, and generates density distribution maps, effectively solving the problem of inaccurate component identification in complex scenarios by traditional methods, and providing a structured data foundation for subsequent processing; adopting a multi-level dynamic filtering strategy based on density and curvature features, by adaptively adjusting the grid size, it efficiently removes noise while retaining key equipment details, and achieves a more refined denoising effect in areas with overlapping vegetation; through self-supervised learning to optimize model parameters and combining statistical filtering, a dual denoising mechanism is constructed, which not only reduces the dependence on labeled data but also significantly improves the model's generalization ability in changing environments; innovatively integrates multimodal features such as curvature, normal vector consistency, and density, generates classification feature vectors through dynamic weight allocation, and uses a multi-scale graph convolutional network for refined classification. This enables the present invention to effectively distinguish and remove various types of noise while retaining the key features of power equipment, significantly improving the point cloud denoising precision and classification accuracy in complex scenarios, and providing a highly reliable data processing solution for intelligent inspection of transmission lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 An overall flow chart of an efficient denoising method for processing transmission line point clouds provided by one embodiment of the present invention.

[0039] Figure 2 A system solution flow chart of a transmission line point cloud processing system with efficient denoising provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the above objectives, features and advantages of the present application more clear and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work should fall within the protection scope of the present application.

[0041] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced without the specific details. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to obscure the present application. Accordingly, the present application is not limited to the embodiments described herein.

[0042] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. The "in one embodiment" appearing in different places in the specification does not mean the same embodiment, nor does it mean that the embodiment is mutually exclusive with other embodiments or is selected.

[0043] The present application is described in detail with reference to the accompanying drawings. In the detailed description of the embodiments of the present application, the cross-sectional view of the device structure is partially enlarged without the general proportion for the convenience of description, and the schematic diagram is only an example, which should not limit the scope of protection of the present application. In addition, the three-dimensional spatial dimensions of length, width and depth should be included in actual manufacture.

[0044] Meanwhile, in the description of the present application, it should be noted that the terms "upper, lower, inner and outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first, second or third" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance.

[0045] Unless otherwise specifically defined and limited, the terms "mounting, connecting, connection" in the present application should be understood broadly, for example: it can be fixed connection, detachable connection or integral connection; it can also be mechanical connection, electrical connection or direct connection, it can also be indirectly connected through intermediate medium, or it can be the communication between two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0046] Example 1, with reference to Figure 1For the first embodiment of the present application, the embodiment provides a power transmission line point cloud processing method with high denoising efficiency, comprising:

[0047] S1: performing segmentation processing on original power transmission line point cloud data, dividing the point cloud into power equipment components, and performing local density analysis on the segmented point cloud to generate a density distribution map.

[0048] Further, the point cloud data segmentation adopts a region growing method based on geometric features to segment the original point cloud data into conductor, insulator, and tower components; the region growing method selects points with curvature less than or equal to a curvature threshold and consistent normal vector direction as seed points, and performs region expansion when the geometric similarity condition is met in the neighborhood.

[0049] In the embodiments of the present application, the segmentation processing includes a segmentation method based on point cloud local curvature difference and normal vector angle for region expansion.

[0050] In an alternative embodiment, the segmentation processing includes preprocessing the original point cloud, reducing the data volume through voxel grid downsampling, setting an Euclidean distance threshold, clustering the point cloud, grouping points with a distance less than the threshold into the same component, screening the clustering results in combination with prior knowledge of power equipment, dividing the conductor, insulator, and tower, and performing local density analysis on the segmented component point cloud.

[0051] In another alternative embodiment, the segmentation processing includes extracting intensity and color histogram features using the intensity information or RGB information of the laser radar point cloud, dividing the point cloud into different feature regions through K-means clustering, manually labeling typical sample regions according to power equipment features, training a support vector machine classifier based on the labeling results, and completing component segmentation.

[0052] It should be noted that the geometric similarity condition is that the absolute value of the curvature difference of two points in the neighborhood is less than the curvature threshold, and the normal vector angle is greater than the normal vector angle threshold.

[0053] Further, the local density analysis includes calculating the local density of each point through Gaussian kernel density estimation to generate a point cloud density map, and the formula is as follows:

[0054]

[0055] wherein, p i is the local density of point p i in the point cloud, K is the number of neighboring points of point p i in the point cloud, p j is a point in the neighborhood of point p i in the point cloud, and sigma is a Gaussian kernel bandwidth parameter. iFor the original point in the point cloud, i and j are variable indexes.

[0056] S2: Based on the density distribution map and the local geometric features of the point cloud, multi-level filtering is performed, and the noise point is removed by combining the statistical filtering method and optimizing the denoising model parameters.

[0057] Further, the multi-level filtering includes voxel filtering, which significantly reduces the data amount by dividing the point cloud space into a uniform voxel grid and retaining only one representative point in each voxel. The first level filtering uses large voxel size for coarse screening, and the second level filtering includes reducing the grid size when the local curvature is higher than the threshold and the density is higher than the threshold, and increasing the grid size when the local curvature is less than or equal to the threshold and the density is less than or equal to the threshold.

[0058] The formula of the dynamic voxel size is:

[0059] s = s max -α*δ-β*ρ

[0060] Where s max is the maximum voxel size, s is the obtained voxel size, α and β are weight parameters, δ is the curvature, and ρ is the density.

[0061] Further, the denoising model is optimized by self-supervised learning. The self-supervised learning optimizes the model parameters by constructing an unsupervised task, reducing the dependence on labeled data, and for each point, the model predicts a clean version and optimizes the model through reconstruction error; a statistical method is used to remove noise points, which calculates the average distance of each point to the nearest neighbor point, and when the average distance exceeds the threshold, it is determined as a noise point.

[0062] The formula of the reconstruction error optimization model is:

[0063]

[0064] Where L self is the total reconstruction error, N is the number of points in the neighborhood of point p i in the point cloud, f i (p) is the reconstruction error of point p i in the point cloud, is the clean version point predicted by the model, i is the variable index, and p i is the original point in the point cloud.

[0065] The statistical method calculation formula is:

[0066]

[0067] Where d i is the average distance of point p i to K nearest neighbors, p iis the original point in the point cloud, K is the point p in the point cloud i The number of neighboring points, p j is the point p in the point cloud i The points in the neighborhood of , i and j are variable indices.

[0068] S3: Extract the curvature features and normal vector consistency features of the point cloud and generate a classification feature vector. Based on the classification feature vector, perform fine classification on the point cloud.

[0069] In an embodiment of the present application, extracting curvature features of the point cloud includes calculating the curvature of each point by eigenvalue decomposition of the covariance matrix.

[0070] In an optional embodiment, the extraction of curvature features of the point cloud includes performing elevation normalization on the point cloud, calculating the maximum height difference within a local neighborhood as a roughness feature, fitting the local plane through the least squares method, using the normal vector angle as a flatness feature, and inputting the roughness and flatness features into a classification model.

[0071] In another optional embodiment, the extraction of curvature features of the point cloud includes dividing the point cloud into a voxel grid of fixed size, counting the number of points in each voxel, the average height, and the height variance as features, calculating the feature differences of adjacent voxels through a sliding window, and inputting the voxel feature matrix into a 2D convolutional network for classification.

[0072] The covariance matrix is ​​constructed by the neighborhood coordinates of the point, and the minimum eigenvalue represents the local curvature; the normal vector consistency feature includes calculating the variance of the normal vector angle of each point in the neighborhood, and the area where the variance is lower than a preset threshold is determined to have consistent normal vectors;

[0073] The formula for calculating curvature is expressed as:

[0074]

[0075] Among them, δ i is the point p in the point cloud i The curvature, λ min is the minimum eigenvalue of the covariance matrix, λ1, λ2 and λ3 are the midpoint p in the point cloud i The eigenvalues ​​of the covariance matrix of the point cloud are represented by the eigenvalues ​​of the covariance matrix, where i is the variable index. The curvature reflects the smoothness of the local surface of the point cloud. The high curvature area corresponds to the edge and complex geometric structure (such as the edge of the insulator), and the low curvature area corresponds to the plane and flat surface (such as the tower plane). The principal component analysis (PCA) is used to accelerate the calculation of the eigenvalues ​​of the covariance matrix, and the KD tree is combined to optimize the neighborhood search to improve the computational efficiency of large-scale point clouds.

[0076] The normal vector is the eigenvector corresponding to the minimum eigenvalue of the covariance matrix, which is implemented through principal component analysis. The normal vector represents the orientation of the local surface of the point cloud and is used to identify the characteristic direction of planar areas or complex geometric structures. The smoothness of the local surface is judged by calculating the angle between the normal vectors of the point cloud in the neighborhood. Areas with high consistency of normal vectors are planar areas (such as tower surfaces), and areas with low consistency are complex geometric structures (such as vegetation noise). The combination of curvature and normal vectors can more comprehensively characterize the geometric characteristics of the point cloud.

[0077] The covariance matrix is ​​expressed as:

[0078]

[0079] Where C is the point p in the point cloud i The covariance matrix of point p in the point cloud i The number of neighboring points, p j is the point p in the point cloud i points in the neighborhood of μ i is the point p in the point cloud i The mean vector of the neighboring points, T is the transpose operation, and i and j are variable indices.

[0080] Furthermore, generating the classification feature vector includes extracting curvature features and normal vector consistency features of the point cloud based on the optimized denoising model, and fusing the curvature, normal vector consistency and density features into the classification feature vector through a dynamic weight allocation model;

[0081] The dynamic weight allocation model includes iteratively adjusting the weight parameters on the training set based on grid search to minimize the classification loss function. The formula is expressed as:

[0082]

[0083] Among them, w i is the point p in the point cloud i The feature weights, α and β are weight parameters, δ i is the point p in the point cloud i The curvature, ρ i is the point p in the point cloud i The local density of point p in the point cloud i The number of neighbors, ρ j is the point p in the point cloud i The local density of points in the neighborhood of j is the point p in the point cloud i The curvature of the point in the neighborhood of , i and j are variable indices.

[0084] According to the curvature and the density, the weight is dynamically allocated, the high curvature and high density area (such as vegetation) has low weight, and the low curvature and low density area (such as equipment) has high weight, so that the feature fusion effect is optimized, and the parameters are optimized through grid search, and it is ensured that the weight allocation model is adapted to different scenes.

[0085] The fine classification of the point cloud includes fine classification of the point cloud by using the classification feature vector through the deep learning model, retaining the power equipment feature points and removing residual noise points.

[0086] The deep learning model includes inputting the classification feature vector into a multi-scale graph convolution network, outputting a point cloud category probability, training the multi-scale graph convolution network through a cross-entropy loss function, and introducing a Dropout layer and L2 regularization to prevent model overfitting, and applying random rotation, translation and noise disturbance to the input point cloud to simulate complex scene data distribution and improve the robustness of the model in complex scenes.

[0087] The multi-scale graph convolution network includes a first layer of convolution to extract local geometric features, a second layer of convolution to fuse global topological relationships, and a third layer to output classification results through a fully connected layer; training the multi-scale graph convolution network includes jointly optimizing classification loss and reconstruction loss, and performing end-to-end training through weighted summation.

[0088] The formula of the cross-entropy loss function is:

[0089]

[0090] Wherein, L class is the cross-entropy loss, N is the number of points in the neighborhood of point p i in the point cloud, y i is the true label, is the model prediction probability, and i is the variable index.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

[0092] Embodiment 2, refer to Figure 2 , is a second embodiment of the present application, which provides a high-efficiency denoising power transmission line point cloud processing system, including a point cloud segmentation and density analysis module, a multi-stage filtering and model optimization module, a feature extraction and dynamic fusion module, and a point cloud classification module.

[0093] The point cloud segmentation and density analysis module comprises a point cloud segmentation unit and a density analysis unit, and is used for dividing original power transmission line point cloud data into power equipment components according to geometric features, and generating a density distribution map by calculating the neighborhood density distribution of each point in the segmented component point cloud based on a Gaussian kernel function.

[0094] The multi-stage filtering and model optimization module comprises a multi-stage filtering unit, a self-supervised learning unit and a statistical filtering unit, and is used for removing redundant noise points through two-stage filtering, optimizing denoising model parameters through point cloud reconstruction error, removing abnormal points based on neighborhood distance distribution, and performing secondary denoising in combination with self-supervised learning results.

[0095] The feature extraction and dynamic fusion module comprises a covariance matrix feature decomposition unit, a normal vector consistency analysis unit and a dynamic weight distribution unit, and is used for calculating curvature features through covariance matrix decomposition, analyzing normal vector consistency through variance of neighborhood normal vector angles, identifying complex structure regions, dynamically distributing weights according to curvature, density and normal vector consistency, optimizing weight parameters, and minimizing classification loss.

[0096] The point cloud classification module is used for inputting the classification feature vector into a multi-scale graph convolution network, outputting point cloud category probability through local and global feature fusion, and performing model training.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

[0098] Embodiment 3, the third embodiment of the present application, is different from the first two embodiments in that:

[0099] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0100] The logic and / or steps represented in the flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. Just as an example, "computer-readable medium" can be any medium that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium.

[0101] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer. In some embodiments, the computer-readable medium can be a transmission medium that carries the program for use by, or in connection with, an instruction execution system, apparatus, or device.

[0102] It should be understood that aspects of the application can be implemented in hardware, software, firmware, or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques, which are well known in the art of hardware implementation, can be used: a hybrid of the above techniques, a mixture of two or more of the above techniques, or a combination of the above techniques with other techniques not listed above.

Claims

1. An efficient denoising method for processing power line point clouds, characterized by: include, Segment the original transmission line point cloud data, divide the point cloud into power equipment components, and perform local density analysis on the segmented point cloud to generate a density distribution map; Based on the density distribution map and local geometric features of the point cloud, multi-level filtering is performed, and the denoising model parameters are optimized, and the noise points are removed by combining the statistical filtering method; The curvature features and normal vector consistency features of the point cloud are extracted, and a classification feature vector is generated. Based on the classification feature vector, the point cloud is refinedly classified.

2. The method for processing power transmission line point clouds with high efficiency denoising according to claim 1, characterized in that: The segmentation process includes performing region expansion based on local curvature differences and normal vector angles of the point cloud, dividing the point cloud into conductors, insulators, and tower components based on geometric features of the point cloud, and performing local density analysis on the segmented component point clouds to generate a density distribution map; Local density analysis involves constructing a neighborhood with each point as the center in each component point cloud after segmentation, calculating the density distribution of points in the neighborhood, and generating a density distribution map.

3. The method for processing power transmission line point clouds with high efficiency denoising according to claim 2, characterized in that: The multi-stage filtering includes performing first-stage filtering and second-stage filtering based on the density distribution map and the local curvature characteristics of the point cloud; The first level of filtering involves coarse screening with a large voxel size to reduce the amount of data. The second level of filtering involves reducing the grid size when the local curvature is higher than the threshold and the density is higher than the threshold, and increasing the grid size when the local curvature is less than or equal to the threshold and the density is less than or equal to the threshold.

4. The method for processing power transmission line point clouds with high efficiency denoising according to claim 3, characterized in that: Optimizing the denoising model parameters includes reconstructing the point cloud and calculating the reconstruction error, analyzing the neighborhood distance distribution of the point cloud in combination with a statistical filtering method, and removing abnormal points; Optimizing the denoising model parameters also includes constructing a point cloud reconstruction task, inputting a noisy point cloud and predicting a clean version, calculating the reconstruction error between the predicted point cloud and the real point cloud, and back-propagating the optimization model parameters; The statistical filtering method includes calculating the average distance from each point to its K nearest neighbor points. When the average distance exceeds the threshold, it is determined to be a noise point and removed. The point cloud is subjected to secondary denoising by combining the reconstruction error with the statistical filtering results.

5. The method for processing power transmission line point clouds with high efficiency denoising according to claim 4, characterized in that: Extracting the curvature features of the point cloud includes calculating the curvature of each point by eigenvalue decomposition of the covariance matrix; The covariance matrix is ​​constructed by the neighborhood coordinates of the point, and the minimum eigenvalue represents the local curvature; the normal vector consistency feature includes calculating the variance of the normal vector angle of each point in the neighborhood, and the area where the variance is lower than a preset threshold is determined to have consistent normal vectors; The formula for calculating curvature is expressed as: Among them, δ i is the point p in the point cloud i The curvature, λ min is the minimum eigenvalue of the covariance matrix, λ1, λ2 and λ3 are the midpoint p in the point cloud i The eigenvalues ​​of the covariance matrix, where i is the variable index; The covariance matrix is ​​expressed as: Where C is the point p in the point cloud i The covariance matrix of point p in the point cloud i The number of neighboring points, p j is the point p in the point cloud i points in the neighborhood of μ i is the point p in the point cloud i The mean vector of the neighboring points, T is the transpose operation, and i and j are variable indices.

6. The method for processing power transmission line point clouds with high efficiency denoising according to claim 5, characterized in that: Generating the classification feature vector includes extracting curvature features and normal vector consistency features of the point cloud based on the optimized denoising model, and fusing the curvature, normal vector consistency and density features into the classification feature vector through a dynamic weight allocation model; The dynamic weight allocation model includes iteratively adjusting the weight parameters on the training set based on grid search to minimize the classification loss function. The formula is expressed as: Among them, w i is the point p in the point cloud i The feature weights, α and β are weight parameters, δ i is the point p in the point cloud i The curvature, ρ i is the point p in the point cloud i The local density of point p in the point cloud i The number of neighbors, ρ j is the point p in the point cloud i The local density of points in the neighborhood of j is the point p in the point cloud i The curvature of the point in the neighborhood of , i and j are variable indices.

7. The method for processing power transmission line point clouds with high efficiency denoising according to claim 6, characterized in that: The fine classification of the point cloud includes using the classification feature vector to finely classify the point cloud through a deep learning model, retaining the characteristic points of the power equipment and removing the residual noise points; The deep learning model involves inputting classification feature vectors into a multi-scale graph convolutional network, outputting point cloud category probabilities, training the multi-scale graph convolutional network using a cross-entropy loss function, introducing a dropout layer and L2 regularization, and applying random rotation, translation, and noise perturbations to the input point cloud to simulate complex scene data distribution. The multi-scale graph convolutional network includes a first-layer convolution to extract local geometric features, a second-layer convolution to fuse global topological relationships, and a third-layer to output classification results through a fully connected layer; training the multi-scale graph convolutional network includes jointly optimizing the classification loss and reconstruction loss, and performing end-to-end training through weighted summation.

8. A system using the efficient denoising method for processing power transmission line point clouds according to any one of claims 1 to 7, characterized in that: Including point cloud segmentation and density analysis module, multi-level filtering and model optimization module, feature extraction and dynamic fusion module and point cloud classification module; The point cloud segmentation and density analysis module includes a point cloud segmentation unit and a density analysis unit, which is used to divide the original transmission line point cloud data into power equipment components according to geometric features, and calculate the neighborhood density distribution of each point in the segmented component point cloud based on the Gaussian kernel function to generate a density distribution map; The multi-stage filtering and model optimization module includes a multi-stage filtering unit, a self-supervised learning unit, and a statistical filtering unit, which is used to remove redundant noise points through two-stage filtering, optimize the denoising model parameters through point cloud reconstruction error, eliminate abnormal points based on neighborhood distance distribution, and perform secondary denoising in combination with the self-supervised learning results; The feature extraction and dynamic fusion module includes a covariance matrix feature decomposition unit, a normal vector consistency analysis unit, and a dynamic weight allocation unit, which is used to calculate curvature features through covariance matrix decomposition, and analyze normal vector consistency through the variance of neighborhood normal vector angles to identify complex structure areas, dynamically allocate weights based on curvature, density, and normal vector consistency, and optimize weight parameters to minimize classification loss; The point cloud classification module is used to input the classification feature vector into the multi-scale graph convolutional network, output the point cloud category probability through local and global feature fusion, and perform model training.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for processing a transmission line point cloud with efficient denoising are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for processing a transmission line point cloud with high efficiency denoising according to any one of claims 1 to 7 are implemented.

Images