Method and system for detecting icing thickness of overhead power line based on structure sensing framework

Abstract

The invention discloses an overhead power line icing thickness detection method and system based on a structure perception framework, and the method comprises the steps: firstly obtaining point cloud data of a power transmission line, preprocessing the point cloud data, and inputting the preprocessed point cloud data into a structure perception semantic segmentation network; the network adopts a four-level encoder-decoder architecture, an encoder extracts multi-scale features through local space encoding, cross-scale space attention fusion and local aggregation operation, and a decoder realizes feature fusion in combination with a jumper connection mechanism and outputs a lead semantic segmentation result; extracting a conductor point set based on a segmentation result and carrying out Euclidean clustering strand splitting; performing biplane projection and polynomial fitting on each strand of wire to construct a three-dimensional center line; and finally calculating the distance from the point to the center line, selecting a peripheral point set to calculate the wire envelope radius, and combining the radius of the bare wire to obtain the icing thickness. Measurement deviation caused by point cloud shielding, wire bending and asymmetric icing is effectively solved, millimeter-level precision non-contact ice thickness detection in a complex environment is achieved, and reliable technical support is provided for power grid disaster prevention.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a method and system for detecting ice thickness of overhead power lines based on a structure perception framework. Background Art

[0002] Currently, ice thickness detection primarily relies on image processing or physical sensing methods. One common approach involves image-based geometric contour extraction. Using drones to capture transmission line images, techniques such as image enhancement, edge detection, and contour fitting are used to estimate the outer radius of the ice-covered conductor. This is then compared with the known radius of the bare conductor to infer the ice thickness. However, these methods are highly dependent on the external environment. Images are easily affected by factors such as occlusion, uneven lighting, and changes in viewing angle, resulting in reduced recognition accuracy and an inability to obtain three-dimensional structural information, limiting their practical application in complex environments. Another existing technology relies on physical model-based sensing methods. For example, tension sensors or inclinometers are deployed on conductors. Combined with meteorological data, these methods estimate conductor stress and deformation changes. This in turn establishes a physical response model of the conductor-ice layer to infer ice thickness. While these methods offer high accuracy, they require high sensor deployment and calibration requirements, resulting in high equipment costs and complex maintenance. They are difficult to deploy on a large scale and lack non-contact detection capabilities.

[0003] In recent years, with the advancement of unmanned aerial vehicle (UAV)-LiDAR (LiDAR) technology, some studies have proposed using three-dimensional point cloud modeling methods for contactless ice thickness estimation. This typically employs a two-stage strategy: first, semantic segmentation is performed on the point cloud to extract the conductor region; then, centerline fitting and cross-sectional analysis are performed to calculate the ice thickness based on radial expansion. However, in complex corridor scenarios, conductor point clouds are characterized by being slender, sparse, and subject to frequent occlusion. Existing segmentation methods struggle to accurately preserve conductor geometry, often resulting in breakage and blurred boundaries. This leads to large deviations in centerline fitting, unstable outer shell point extraction, and discontinuous and inaccurate thickness estimation results. Therefore, existing technologies still have significant deficiencies in point cloud structure preservation, boundary clarity, and estimation accuracy. A new ice thickness detection method with strong structural continuity, high boundary discrimination, and robustness is urgently needed. This paper addresses these issues by proposing an ice thickness estimation method that integrates structure-aware semantic segmentation with geometric modeling and analysis. This method effectively addresses the structural and estimation accuracy degradation caused by conductor sparsity and occlusion. Summary of the Invention

[0004] In response to the above technical problems, the present invention provides a method and system for detecting ice thickness on overhead power lines based on a structure perception framework.

[0005] The technical solution adopted by the present invention to solve the technical problem is: A method for detecting ice thickness of overhead power lines based on a structure perception framework, the method comprising the following steps: S100: UAV LiDAR collects point cloud data from overhead power lines, preprocesses the point cloud data, and uses a multi-layer perceptron to perform feature dimensionality upgrade on the preprocessed point cloud data. S200: Input the upgraded point cloud data into a structure-aware semantic segmentation network, and extract the semantic features of the wire point cloud through an encoder-decoder architecture. The encoder consists of a four-layer structure, each of which sequentially performs local spatial encoding, cross-scale spatial attention fusion, and local aggregation operations to generate multi-scale fused features. The decoder consists of a symmetrical four-layer structure, upsampling through nearest neighbor interpolation and fusion with the encoder features through jump connection. Finally, the shared fully connected classification head outputs the wire semantic segmentation results. S300: Extracting wire point sets based on semantic segmentation results, separating multiple wires through Euclidean clustering, and fitting the 3D centerline trajectory of each wire. S400: For each conductor strand, calculate the shortest Euclidean distance from each conductor point to the fitted centerline. Based on the statistical distribution of the distance from the conductor point to the centerline, extract the outer shell point set and calculate the conductor envelope radius. Combined with the bare conductor radius, estimate the ice thickness.

[0006] Preferably, each layer of the encoder in S200 includes a local spatial coding module, a cross-level spatial coding module and a local aggregation module; After the features of the input encoder are downsampled, the local spatial encoding module aggregates neighborhood features through grouping operations and extracts local geometric structures through maximum pooling and MLP; The cross-level spatial encoding module concatenates the output of the local spatial encoding module of the current layer with the features of the previous layer, constructs the query vector Q, key vector K, and value vector V through the MLP-ReLU module, and calculates the attention weights and weighted fusion through matrix multiplication; The local aggregation module fuses the output features of the local spatial encoding module and the cross-level spatial encoding module.

[0007] Preferably, the local spatial encoding module includes dual-branch processing. In the first branch, the input features are grouped to aggregate neighborhood points, and local features are extracted through maximum pooling and MLP. In the second branch, the input features are directly transformed by MLP, and the dual-branch outputs are spliced ​​and then subjected to MLP to generate local enhanced features.

[0008] Preferably, the cross-level spatial encoding module concatenates the output of the local spatial encoding module of the current layer with the features of the previous layer, specifically: ; in, is the feature of the previous layer, is the current layer feature, T(·) represents the projection function, is the feature after splicing; The cross-scale spatial correlation features are generated through the attention weight calculation module, specifically: ; in, , , Each consists of two layers of MLP, which is used to map input features to query, key and value representations in the attention space. Represents the attention weight matrix, which is used to characterize the relationship strength between each point and is used to weight the aggregation value features. represents the dimension of the key vector; Output weighted fusion features to the next encoding layer, specifically: ; in, It is an adjustable fusion weight used to balance the importance of local features and global attention enhancement features. Represents the output of the cross-level spatial encoding module.

[0009] Preferably, S300 includes: S310: extracting a wire category point set based on the semantic segmentation result output by S200, performing isolated point elimination and connectivity filtering on the wire category point set, and removing local noise and misclassified points; S320: Based on the adjacency relationship established by the Euclidean distance between points, a region-growing clustering method is used to divide the conductor category point set into several spatially connected subsets, each corresponding to a different strand of conductors. A distance threshold is set during the clustering process to ensure spatial continuity of conductors in the same strand, and a minimum cluster size is set to filter out fragmented clustering results. S330: For each conductor point set, based on its spatial distribution characteristics, a local polynomial fitting method is used to construct a three-dimensional centerline trajectory.

[0010] Preferably, S330 includes: S331: Project the traverse point to 、 On a plane; S332: Perform n-order polynomial fitting on the coordinate points in each projection plane to construct a centerline trajectory model: ; in, is the coordinate component in three-dimensional space; and They are Plane and The polynomial coefficients of the plane fitting curve represent the Second and The weight of the secondary item; and is the order of the polynomial used to control the complexity of the fitting curve; , is the index of the number of terms in the polynomial expansion; S333: Use the least squares method to fit the coefficients in the center line fitting model and Solve to minimize the vertical distance residual of the wire point cloud relative to the fitting curve in the corresponding projection plane; when Taking the plane as an example, let the projected wire point set be , then the fitting coefficient Obtained by solving the following optimization problem: ; Similarly, in Construct a fitting model in the plane and set the projection point to be , then the coefficient satisfy: ; The two minimization problems mentioned above are both linear least squares fitting problems. By constructing a normal equation system or a pseudo-inverse matrix to solve them, the final fitting coefficients are used to reconstruct the centerline trajectory equation of the wire in three-dimensional space. , specifically: ; in, is the curve parameter on the center line, taking a one-dimensional variable .

[0011] Preferably, S400 includes: S410: For each conductor strand , calculate each traverse point Shortest Euclidean distance to the fitted centerline: ; in, Indicates the The wire point to the Minimum radial distance between the center lines of the stranded conductors; For the The three-dimensional space coordinates of the traverse points; Indicates the The centerline trajectory function of the strand conductor, is the curve parameter on the center line; Represents the Euclidean norm in two-dimensional or three-dimensional space; S420: According to the statistical feature extraction rules, the outer points with a preset ratio ranked low among all radius values ​​are selected to form the outer shell point set. ,Right now: ; Among them, the percentile function Indicates that all After sorting the values ​​from small to large, the points ranked between 85% and 100% are selected as peripheral candidate points, excluding internal points in dense areas or noise interference; S430: Shell point set Based on this, we construct an estimate of the wire's envelope radius. , specifically: ; S440: Bare wire radius corresponding to the actual wire model , calculate the final ice thickness estimate , specifically: ; in, is the radius of the conductor envelope containing ice obtained by statistics, is the wire body radius, which can be obtained from the database or on-site measurement. is the estimated radial ice thickness in millimeters.

[0012] An overhead power line ice thickness detection system based on a structure-aware framework includes: The point cloud data acquisition and preprocessing module is used for the UAV laser radar to collect overhead power line point cloud data, preprocess the point cloud data, and perform feature dimensionality upgrade on the preprocessed point cloud data through a multi-layer perceptron; The structure-aware semantic segmentation module is used to input the upgraded point cloud data into the structure-aware semantic segmentation network and extract the semantic features of the wire point cloud through an encoder-decoder architecture. The encoder consists of a four-layer structure, each layer sequentially performs local spatial encoding, cross-scale spatial attention fusion, and local aggregation operations to generate multi-scale fused features. The decoder consists of a symmetrical four-layer structure, upsampling through nearest neighbor interpolation and fusion with the encoder features through jump connection, and finally outputs the wire semantic segmentation results through a shared fully connected classification head. The 3D centerline trajectory fitting module is used to extract the wire point set based on the semantic segmentation results, separate multiple wires through Euclidean clustering, and perform 3D centerline trajectory fitting on each wire. The ice thickness estimation module is used to calculate the shortest Euclidean distance from each conductor point to the fitted centerline for each conductor strand. Based on the statistical distribution of the distance from the conductor point to the centerline, the outer shell point set is extracted and the conductor envelope radius is calculated. The ice thickness is estimated in combination with the bare conductor radius.

[0013] 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 processor implements the steps of a method for detecting ice thickness of overhead power lines based on a structure perception framework.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for detecting ice thickness of overhead power lines based on a structure perception framework.

[0015] The above-mentioned method and system for detecting ice thickness on overhead power lines based on a structure-aware framework significantly improves the accuracy and robustness of detecting ice thickness on overhead power lines in complex environments through the collaborative innovation of a structure-aware semantic segmentation network and geometric modeling. The structure-aware segmentation network effectively enhances the semantic continuity of conductor point clouds in occlusion and fracture scenarios through a cross-scale spatial attention fusion mechanism, providing a high-integrity conductor point set for subsequent geometric modeling. The three-dimensional centerline segmented fitting method combines biplane projection with local polynomial regression to overcome the distortion problem of curved conductor trajectories, reducing the centerline reconstruction error to the millimeter level. The shell point statistical ice thickness estimation strategy significantly suppresses measurement deviations caused by point cloud sparsity and asymmetric icing by extracting some peripheral points and calculating the squared average radius. The entire method achieves non-contact, fully automatic detection, supports millimeter-level precision ice thickness analysis of transmission lines, and provides highly reliable technical support for power grid anti-icing disaster prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a method for detecting ice thickness on overhead power lines based on a structure perception framework in one embodiment of the present invention; Figure 2 Schematic diagram of the structure of the structure-aware semantic segmentation network in one embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below with reference to the accompanying drawings.

[0018] In one embodiment, Figure 1 and 2 As shown, a method for detecting ice thickness of overhead power lines based on a structure perception framework comprises the following steps: S100: UAV LiDAR collects point cloud data from overhead power lines, preprocesses the point cloud data, and uses a multi-layer perceptron to perform feature dimensionality upgrade on the preprocessed point cloud data. S200: Input the upgraded point cloud data into a structure-aware semantic segmentation network, and extract the semantic features of the wire point cloud through an encoder-decoder architecture. The encoder consists of a four-layer structure, each of which sequentially performs local spatial encoding, cross-scale spatial attention fusion, and local aggregation operations to generate multi-scale fused features. The decoder consists of a symmetrical four-layer structure, upsampling through nearest neighbor interpolation and fusion with the encoder features through jump connection. Finally, the shared fully connected classification head outputs the wire semantic segmentation results. S300: Extracting wire point sets based on semantic segmentation results, separating multiple wires through Euclidean clustering, and fitting the 3D centerline trajectory of each wire. S400: For each conductor strand, calculate the shortest Euclidean distance from each conductor point to the fitted centerline. Based on the statistical distribution of the distance from the conductor point to the centerline, extract the outer shell point set and calculate the conductor envelope radius. Combined with the bare conductor radius, estimate the ice thickness.

[0019] Specifically, the present invention proposes a structure-aware framework (SAF) for detecting ice thickness on overhead transmission lines. This framework integrates a two-stage processing flow of semantic segmentation and geometric modeling. Taking into account the sparsity, slenderness, and occlusion sensitivity of the conductor point cloud structure, it constructs an ice thickness estimation method with strong structure preservation and robustness.

[0020] To address the issues of conductor structure breakage and blurred boundaries in complex environments, a three-dimensional point cloud semantic segmentation network based on an encoding-decoding structure was designed. This network integrates a key improvement module: the Cross-level Spatial Encoding (CSE) module. This module introduces information interaction between multiple resolutions during feature extraction, enhancing the ability to express the spatial continuity of conductor structures at different scales and avoiding semantic interruptions caused by downsampling. This network accurately extracts a complete set of conductor points from the original point cloud, providing a reliable foundation for geometric modeling. Based on high-quality conductor point clouds, the present invention introduces a radial thickness estimation method based on centerline fitting and shell point statistical analysis. By combining spatial structure continuity modeling with slice-level thickness analysis, the present invention effectively suppresses estimation errors caused by local noise and point cloud sparsity, achieving millimeter-level ice thickness detection accuracy.

[0021] Furthermore, in S100, point cloud data is input ,in is the batch size, For points, is the number of input feature channels, through the multi-layer perceptron Upgrade to ; Point cloud data The input is sent to the segmentation network with an encoder-decoder structure. The multi-scale semantic features are extracted and fused in sequence through the four-layer encoder and four-layer decoder. Finally, the input is sent to a classification head to obtain the semantic prediction result of each point. The output is .

[0022] In one embodiment, each layer of the encoder in S200 includes a local spatial coding module, a cross-level spatial coding module, and a local aggregation module; After the features of the input encoder are downsampled, the local spatial encoding module aggregates neighborhood features through grouping operations and extracts local geometric structures through maximum pooling and MLP; The cross-level spatial encoding module concatenates the output of the local spatial encoding module of the current layer with the features of the previous layer, constructs the query vector Q, key vector K, and value vector V through the MLP-ReLU module, and calculates the attention weights and weighted fusion through matrix multiplication; The local aggregation module fuses the output features of the local spatial encoding module and the cross-level spatial encoding module.

[0023] Specifically, the point cloud features of the input encoder , firstly through the grid downsampling operation, adjusted to , then enter the local space encoding module to obtain feature output .

[0024] This feature result is processed along two paths: one is input to the cross-level spatial encoding module CSE, and the other is concatenated with the output of the CSE module and passed to the encoder of the next layer. The input of the CSE module consists of two sources: the local encoded features of the current stage, and the local features output by the previous encoder. The two are concatenated to generate a cross-layer structure representation.

[0025] The features output by the CSE module and the downsampling results are fused again through the local aggregation module to generate the fused features. .

[0026] In one embodiment, the local spatial encoding module includes a dual-branch processing. In the first branch, the input features are grouped to aggregate neighborhood points, and local features are extracted through maximum pooling and MLP. In the second branch, the input features are directly transformed by MLP, and the dual-branch outputs are concatenated and then subjected to MLP to generate local enhanced features.

[0027] In one embodiment, the cross-level spatial encoding module concatenates the output of the local spatial encoding module of the current layer with the features of the previous layer, specifically: ; in, is the feature of the previous layer, is the current layer feature, T(·) represents the projection function, is the feature after splicing; The cross-scale spatial correlation features are generated through the attention weight calculation module, specifically: ; in, , , Each consists of two layers of MLP, which is used to map input features to query, key and value representations in the attention space. Represents the attention weight matrix, which is used to characterize the relationship strength between each point and is used to weight the aggregation value features. represents the dimension of the key vector; Output weighted fusion features to the next encoding layer, specifically: ; in, It is an adjustable fusion weight used to balance the importance of local features and global attention enhancement features. Represents the output of the cross-level spatial coding module. After comparing multiple sets of experiments under typical power transmission scenarios, the results show that The best structural continuity and segmentation accuracy can be achieved when , which has been set as the default hyperparameter in model training.

[0028] In one embodiment, upsampling is performed by nearest neighbor interpolation and fusion is performed by jump connection with encoder features in S200, specifically: The input of the decoder stage is reduced in dimension through the MLP layer, and the features are restored to the point cloud shape using nearest neighbor interpolation, which is then fused with the encoder output features obtained by skipping the upper layer as the decoder output.

[0029] Specifically, for the decoder stage, the input features are , which is reduced to , and use nearest neighbor interpolation to restore the features to the point cloud shape , and then fused with the encoder output features obtained by skipping the upper layer as the decoder output.

[0030] In one embodiment, S300 includes: S310: extracting a wire category point set based on the semantic segmentation result output by S200, performing isolated point elimination and connectivity filtering on the wire category point set, and removing local noise and misclassified points; S320: Based on the adjacency relationship established by the Euclidean distance between points, a region-growing clustering method is used to divide the conductor category point set into several spatially connected subsets, each corresponding to a different strand of conductors. A distance threshold is set during the clustering process to ensure spatial continuity of conductors in the same strand, and a minimum cluster size is set to filter out fragmented clustering results. S330: For each conductor point set, based on its spatial distribution characteristics, a local polynomial fitting method is used to construct a three-dimensional centerline trajectory.

[0031] In one embodiment, S330 includes: S331: Project the traverse point to 、 On a plane; S332: Perform n-order polynomial fitting on the coordinate points in each projection plane to construct a centerline trajectory model: ; in, is the coordinate component in three-dimensional space; and They are Plane and The polynomial coefficients of the plane fitting curve represent the Second and The weight of the secondary item; and is the order of the polynomial used to control the complexity of the fitting curve; , is the index of the number of terms in the polynomial expansion; S333: Use the least squares method to fit the coefficients in the center line fitting model and Solve to minimize the vertical distance residual of the wire point cloud relative to the fitting curve in the corresponding projection plane; when Taking the plane as an example, let the projected wire point set be , then the fitting coefficient Obtained by solving the following optimization problem: ; Similarly, in Construct a fitting model in the plane and set the projection point to be , then the coefficient satisfy: ; The two minimization problems mentioned above are both linear least squares fitting problems. By constructing a normal equation system or a pseudo-inverse matrix to solve them, the final fitting coefficients are used to reconstruct the centerline trajectory equation of the wire in three-dimensional space. , specifically: ; in, is the curve parameter on the center line, taking a one-dimensional variable .

[0032] Specifically, the core purpose of this step is to generate a geometrically smooth and structurally continuous center reference line, which is used to model the actual conductor structure and thus achieve relative geometric measurement of ice cover. Compared with traditional cross-section interception methods, this method has stronger trajectory integrity and robustness, and is particularly suitable for centerline reconstruction in curved conductors and obscured areas.

[0033] In one embodiment, S400 includes: S410: For each conductor strand , calculate each traverse point Shortest Euclidean distance to the fitted centerline: ; in, Indicates the The wire point to the Minimum radial distance between the center lines of the stranded conductors; For the The three-dimensional space coordinates of the traverse points; Indicates the The centerline trajectory function of the strand conductor, is the curve parameter on the center line; Represents the Euclidean norm in two-dimensional or three-dimensional space; Furthermore, all The values ​​are used to construct the radial distribution of points to support the subsequent shell point extraction and thickness statistical analysis.

[0034] S420: According to the statistical feature extraction rules, the outer points with a preset ratio ranked low among all radius values ​​are selected to form the outer shell point set. ,Right now: ; Among them, the percentile function Indicates that all After sorting the values ​​from small to large, the points ranked between 85% and 100% are selected as peripheral candidate points, excluding internal points in dense areas or noise interference;

[0035] S430: Shell point set Based on this, we construct an estimate of the wire's envelope radius. , specifically: ; S440: Bare wire radius corresponding to the actual wire model , calculate the final ice thickness estimate , specifically: ; in, is the radius of the conductor envelope containing ice obtained by statistics, is the wire body radius, which can be obtained from the database or on-site measurement. is the estimated radial ice thickness in millimeters.

[0036] The above-mentioned method for detecting ice thickness on overhead power lines based on a structure-aware framework, through the combined idea of ​​"semantic structure enhancement + shell geometric modeling", constructs a high-precision, non-contact ice thickness detection framework for practical engineering applications. It not only improves the structural integrity of conductor extraction but also enhances the robustness of geometric modeling. It can be widely applied to automatic inspection tasks of transmission lines under different voltage levels and terrain conditions.

[0037] In one embodiment, a system for detecting ice thickness on overhead power lines based on a structure-aware framework is provided, comprising: The point cloud data acquisition and preprocessing module is used for the UAV laser radar to collect overhead power line point cloud data, preprocess the point cloud data, and perform feature dimensionality upgrade on the preprocessed point cloud data through a multi-layer perceptron; The structure-aware semantic segmentation module is used to input the upgraded point cloud data into the structure-aware semantic segmentation network and extract the semantic features of the wire point cloud through an encoder-decoder architecture. The encoder consists of a four-layer structure, each layer sequentially performs local spatial encoding, cross-scale spatial attention fusion, and local aggregation operations to generate multi-scale fused features. The decoder consists of a symmetrical four-layer structure, upsampling through nearest neighbor interpolation and fusion with the encoder features through jump connection, and finally outputs the wire semantic segmentation results through a shared fully connected classification head. The 3D centerline trajectory fitting module is used to extract the wire point set based on the semantic segmentation results, separate multiple wires through Euclidean clustering, and perform 3D centerline trajectory fitting on each wire. The ice thickness estimation module is used to calculate the shortest Euclidean distance from each conductor point to the fitted centerline for each conductor strand. Based on the statistical distribution of the distance from the conductor point to the centerline, the outer shell point set is extracted and the conductor envelope radius is calculated. The ice thickness is estimated in combination with the bare conductor radius.

[0038] Regarding the specific definition of an overhead power line ice thickness detection system based on a structure perception framework, please refer to the definition of an overhead power line ice thickness detection method based on a structure perception framework above, which will not be repeated here. The various modules in the above-mentioned overhead power line ice thickness detection system based on a structure perception framework can be implemented in whole or in part through software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0039] 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 processor implements the steps of a method for detecting ice thickness of overhead power lines based on a structure perception framework.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for detecting ice thickness of overhead power lines based on a structure perception framework.

[0041] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0042] The above is a detailed introduction to the method and system for detecting ice thickness on overhead power lines based on a structure-sensing framework provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for detecting ice thickness on overhead power lines based on a structure-aware framework, characterized in that: The method comprises the following steps: S100: UAV LiDAR collects point cloud data from overhead power lines, preprocesses the point cloud data, and uses a multi-layer perceptron to perform feature dimensionality upgrade on the preprocessed point cloud data. S200: Input the upgraded point cloud data into a structure-aware semantic segmentation network, and extract the semantic features of the wire point cloud through an encoder-decoder architecture. The encoder consists of a four-layer structure, each of which sequentially performs local spatial encoding, cross-scale spatial attention fusion, and local aggregation operations to generate multi-scale fused features. The decoder consists of a symmetrical four-layer structure, upsampling through nearest neighbor interpolation and fusion with the encoder features through jump connection. Finally, the shared fully connected classification head outputs the wire semantic segmentation results. S300: Extracts the wire point set based on the semantic segmentation results, separates multiple wires through Euclidean clustering, and constructs a 3D centerline for each wire using biplane projection and polynomial fitting. S400: For each conductor strand, calculate the shortest Euclidean distance from each conductor point to the fitted centerline. Based on the statistical distribution of the distance from the conductor point to the centerline, extract the outer shell point set and calculate the conductor envelope radius. Combined with the bare conductor radius, estimate the ice thickness.

2. The method according to claim 1, characterized in that Each layer of the encoder in S200 includes a local spatial coding module, a cross-level spatial coding module, and a local aggregation module; After the features of the input encoder are downsampled, the local spatial encoding module aggregates neighborhood features through grouping operations and extracts local geometric structures through maximum pooling and MLP; The cross-level spatial encoding module concatenates the output of the local spatial encoding module of the current layer with the features of the previous layer, constructs the query vector Q, key vector K, and value vector V through the MLP-ReLU module, and calculates the attention weights and weighted fusion through matrix multiplication; The local aggregation module fuses the output features of the local spatial encoding module and the cross-level spatial encoding module.

3. The method according to claim 2, characterized in that The local spatial encoding module includes a two-branch processing. In the first branch, the input features are grouped to aggregate neighborhood points, and local features are extracted through maximum pooling and MLP. In the second branch, the input features are directly transformed by MLP, and the outputs of the two branches are concatenated and then processed by MLP to generate local enhanced features.

4. The method according to claim 3, characterized in that The cross-level spatial encoding module concatenates the output of the local spatial encoding module of the current layer with the features of the previous layer, specifically: ； in, is the feature of the previous layer, is the current layer feature, T(·) represents the projection function, is the feature after splicing; The cross-scale spatial correlation features are generated through the attention weight calculation module, specifically: ； in, , , Each consists of two layers of MLP, which is used to map input features to query, key and value representations in the attention space. Represents the attention weight matrix, which is used to characterize the relationship strength between each point and is used to weight the aggregation value features. represents the dimension of the key vector; Output weighted fusion features to the next encoding layer, specifically: ； in, It is an adjustable fusion weight used to balance the importance of local features and global attention enhancement features. Represents the output of the cross-level spatial encoding module.

5. The method according to claim 4, characterized in that S300 includes: S310: extracting a wire category point set based on the semantic segmentation result output by S200, performing isolated point elimination and connectivity filtering on the wire category point set, and removing local noise and misclassified points; S320: Based on the adjacency relationship established by the Euclidean distance between points, a region-growing clustering method is used to divide the conductor category point set into several spatially connected subsets, each corresponding to a different strand of conductors. A distance threshold is set during the clustering process to ensure spatial continuity of conductors in the same strand, and a minimum cluster size is set to filter out fragmented clustering results. S330: For each conductor point set, based on its spatial distribution characteristics, a local polynomial fitting method is used to construct a three-dimensional centerline trajectory.

6. The method according to claim 5, characterized in that The S330 includes: S331: Project the traverse point to 、 On a plane; S332: Perform n-order polynomial fitting on the coordinate points in each projection plane to construct a centerline trajectory model: ； in, is the coordinate component in three-dimensional space; and They are Plane and The polynomial coefficients of the plane fitting curve represent the The second and The weight of the secondary item; and is the order of the polynomial used to control the complexity of the fitting curve; , is the index of the number of terms in the polynomial expansion; S333: Use the least squares method to fit the coefficients in the center line fitting model and Solve to minimize the vertical distance residual of the wire point cloud relative to the fitting curve in the corresponding projection plane; when Taking the plane as an example, let the projected wire point set be , then the fitting coefficient Obtained by solving the following optimization problem: ； Similarly, in Construct a fitting model in the plane and set the projection point to be , then the coefficient satisfy: ； The two minimization problems mentioned above are both linear least squares fitting problems. By constructing a normal equation system or a pseudo-inverse matrix to solve them, the final fitting coefficients are used to reconstruct the centerline trajectory equation of the wire in three-dimensional space. , specifically: ； in, is the curve parameter on the center line, taking a one-dimensional variable .

7. The method according to claim 6, characterized in that S400 includes: S410: For each conductor strand , calculate each traverse point Shortest Euclidean distance to the fitted centerline: ； in, Indicates the The wire point to the Minimum radial distance between the center lines of the stranded conductors; For the The three-dimensional space coordinates of the traverse points; Indicates the The centerline trajectory function of the strand conductor, is the curve parameter on the center line; Represents the Euclidean norm in two-dimensional or three-dimensional space; S420: According to the statistical feature extraction rules, the outer points with a preset ratio ranked low among all radius values ​​are selected to form the outer shell point set. ,Right now: ； Among them, the percentile function Indicates that all After sorting the values ​​from small to large, the points ranked between 85% and 100% are selected as peripheral candidate points, excluding internal points in dense areas or noise interference; S430: Shell point set Based on this, we construct an estimate of the wire's envelope radius. , specifically: ； S440: Bare wire radius corresponding to the actual wire model , calculate the final ice thickness estimate , specifically: ； in, is the radius of the conductor envelope containing ice obtained by statistics, is the wire body radius, which can be obtained from the database or on-site measurement. is the estimated radial ice thickness in millimeters.

8. An overhead power line ice thickness detection system based on a structure perception framework, characterized in that: include: The point cloud data acquisition and preprocessing module is used for the UAV laser radar to collect overhead power line point cloud data, preprocess the point cloud data, and perform feature dimensionality upgrade on the preprocessed point cloud data through a multi-layer perceptron; The structure-aware semantic segmentation module is used to input the upgraded point cloud data into the structure-aware semantic segmentation network and extract the semantic features of the wire point cloud through an encoder-decoder architecture. The encoder consists of a four-layer structure, each layer sequentially performs local spatial encoding, cross-scale spatial attention fusion, and local aggregation operations to generate multi-scale fused features. The decoder consists of a symmetrical four-layer structure, upsampling through nearest neighbor interpolation and fusion with the encoder features through jump connection, and finally outputs the wire semantic segmentation results through a shared fully connected classification head. The 3D centerline trajectory fitting module is used to extract the wire point set based on the semantic segmentation results, separate multiple wires through Euclidean clustering, and perform 3D centerline trajectory fitting on each wire. The ice thickness estimation module is used to calculate the shortest Euclidean distance from each conductor point to the fitted centerline for each conductor strand. Based on the statistical distribution of the distance from the conductor point to the centerline, the outer shell point set is extracted and the conductor envelope radius is calculated. The ice thickness is estimated in combination with the bare conductor radius.

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 according to any one of claims 1 to 7 are implemented.

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 according to any one of claims 1 to 7 are implemented.

Images