A dense power transmission channel identification method and device based on a two-stage deep network

Abstract

The present application relates to power transmission line risk management and control technical field, especially a kind of dense power transmission channel identification method and equipment based on two-stage deep network.The present application is greatly simplified the scale of point cloud data by grid division and simplification processing to the point cloud data of the power transmission line scene to be identified;After the feature point cloud data of power transmission line is extracted by power transmission line feature information extraction network, it is sent into dense power transmission channel identification network for identification, so as to obtain accurate dense power transmission channel data.Compared with the traditional manual identification method, the present application can improve the accuracy and identification efficiency of identification, so as to facilitate maintenance personnel to accurately risk management and control dense channel area.

Description

Technical Field

[0001] This invention relates to the field of transmission line risk management technology, and in particular to a method and device for identifying dense transmission channels based on a two-stage deep network. Background Technology

[0002] According to the National Energy Administration's definition of dense power transmission corridors, these corridors refer to areas with a width not exceeding 600 meters that contain at least two ultra-high-voltage direct current (UHVDC) lines or at least five 500 kV or higher transmission lines. Currently, my country has several such dense power transmission corridor areas. These areas have complex environments, and natural disasters such as wildfires or earthquakes can cause large-scale power outages in a short period. Therefore, areas with dense power transmission corridors are key areas for power grid companies to inspect and manage for risks.

[0003] Currently, the identification of dense power transmission corridors mainly relies on manual methods. This involves first identifying high-voltage transmission lines within the same area, then using distance measurement tools in software to measure areas with dense lines, and finally determining whether an area constitutes a dense corridor based on the measurement results. This manual identification method suffers from low efficiency and is prone to missing dense corridor areas. Therefore, efficient, accurate, and automated technical means for identifying dense power transmission corridors are currently lacking.

[0004] Therefore, there is a need for a method to identify dense power transmission channels with higher efficiency and accuracy. Summary of the Invention

[0005] The purpose of this invention is to overcome the problems of low identification efficiency and unstable accuracy of special corridors in the prior art, and to provide a method and device for identifying dense power transmission channels based on a two-stage deep network.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution: A method for identifying dense power transmission channels based on a two-stage deep network includes the following steps: S1: Obtain point cloud data of the power transmission line scene to be identified; S2: Divide the point cloud data into grids, and simplify the point cloud data in each grid to output simplified point cloud data; S3: Input the simplified point cloud data into the pre-constructed power transmission line feature information extraction network to extract the feature point cloud data of the power transmission line; the feature point cloud data includes the three-dimensional coordinates and direction features of the power transmission line; S4: Input the feature point cloud data into the pre-constructed dense power transmission channel identification network; S5: The dense power transmission channel identification network determines whether there are dense channels within the scene of the power transmission line to be identified; If it exists, output the three-dimensional coordinates of all transmission lines within the dense channel; If not, output the result that there are no dense power transmission channels in the scene of the power transmission line to be identified; S2 includes the following steps: S21: Divide the point cloud data into grids according to a preset grid size; S22: Calculate the minimum Z-axis coordinate of each grid, and add the minimum Z-axis coordinate to the preset simplification height as the simplification threshold of the current grid; S23: Filter out point cloud data in each grid whose Z-axis coordinate is less than the corresponding simplification threshold; S24: After all the point cloud data of the grids has been simplified, the output is the simplified point cloud data.

[0007] As a preferred embodiment of the present invention, the transmission line feature information extraction network in S3 includes three sequentially connected feature extraction modules, a feature cross-fusion module that cross-fused the output features of the three feature extraction modules, and a dual-branch output module that extracts feature point cloud data of the transmission line based on the fused feature information. The power transmission line feature information extraction network includes the following operational steps: S31: Input the simplified point cloud data into the first feature extraction module and output the extracted features once; input the extracted features once into the second feature extraction module and output the extracted features twice; input the extracted features twice into the third feature extraction module and output the extracted features three times. S32: The feature cross-fusion module performs cross-fusion on the features extracted in the first, second, and third extractions, and outputs the fused feature information; S33: The dual-branch output module outputs the probability that each point in the simplified point cloud data belongs to the transmission line, as well as the direction feature, based on the fused feature information; S34: Based on a preset probability threshold, filter out point clouds that do not belong to the transmission line, and output the three-dimensional coordinates and orientation features of the remaining point cloud data as feature point cloud data.

[0008] As a preferred embodiment of the present invention, the feature extraction module includes a K-feature nearest neighbor grouping layer and a feature extraction operator connected in sequence; and includes the following operating steps: K-feature nearest neighbor grouping layer: The input is a point cloud data of N points. The feature distance between each point and the other points is calculated. The K points with the farthest feature distance from the current point are selected as grouping points. After all points are calculated, the point cloud group data of N×(K+1) points is output, where K is a set value. Feature extraction operators:

[0009] In the formula, I This represents the input data; G() represents the K-feature nearest neighbor grouping layer; Conv() represents the convolution operation; BN() represents the batch normalization operation; ( ) represents the ReLU activation function; Max() represents the max pooling operation; f represents the feature extraction result.

[0010] As a preferred embodiment of the present invention, the feature cross-fusion module in S32 includes the following operating steps:

[0011]

[0012]

[0013] In the formula, f1, f2, and f3 represent the first-order, second-order, and third-order extracted features output by the three feature extraction modules, respectively; CAT() represents the feature concatenation operation; MHA() represents the point cloud multi-head attention mechanism; f 12 This represents the feature resulting from the cross-fusion of the output features of the first and second operations; f 23 This represents the feature resulting from the cross-fusion of the output features of the second and third operations; f 123 This represents the feature resulting from the cross-fusion of the output features from the three operations.

[0014] As a preferred embodiment of the present invention, the dual-branch output module in S33 includes the following operating steps: Branch Road 1:

[0015] Branch Road 2:

[0016] In the formula, FC() represents a fully connected operation; BN() represents a batch normalization operation; ( ) represents the ReLU activation function; DO() represents Dropout processing; ( ) represents the softmax activation function; ( ) represents the sigmoid activation function; P represents the probability that the point belongs to the transmission line; D represents the direction of the transmission line, including the direction features (dx, dy) in both the X and Y directions.

[0017] As a preferred embodiment of the present invention, the dense power transmission channel identification network in S4 includes a directional grouping layer, a graph convolutional feature extraction module, a feature extraction operator, a distance attention weighting module, and a fully connected output module connected in sequence; the dense power transmission channel identification network includes the following operating steps: S41: The directional grouping layer is used to group the feature point cloud data of the transmission line; S42: The feature point cloud data after grouping is sequentially processed by the graph convolution feature extraction module, the feature extraction operator, and the distance attention weighting module to extract features. S43: The fully connected output module determines the probability that each point is in a dense power transmission channel based on the characteristics output in S42.

[0018] As a preferred embodiment of the present invention, S41 includes the following steps: Input the feature point cloud data of the power transmission line, which includes the three-dimensional coordinates and orientation features of M power transmission line points; Choose any two points i and j, and calculate the spatial straight-line distance between the i-th point and the j-th point based on the three-dimensional coordinates of the two points, i∈[1,M]; j∈[1,M]; The expression for calculating the direction angle between the i-th point and the j-th point of the transmission line is:

[0019] In the formula, θ ij d represents the direction angle between the i-th point and the j-th point; xi and d yi The directional characteristics of point i; Δx ij and Δy ij represents the coordinate deviation between point i and point j on the X and Y axes; arctan() represents the arctangent operation; | represents the absolute value calculation; Using the i-th point as the aggregation point, the farthest point sampling algorithm is used to select L grouping points that are farthest apart from each other and whose direction angle is less than 10° as grouping data, resulting in M ​​groups × (L+1) points of transmission line grouping data, where L is a set value.

[0020] As a preferred embodiment of the present invention, S42 includes the following steps: Graph convolution feature extraction module:

[0021] In the formula, f4 represents the features output by the graph convolution feature extraction module; ( ) represents the ReLU activation function; BN() represents batch normalization operation; Conv() represents convolution operation; CAT() represents feature concatenation operation; G2 represents the three-dimensional coordinates of the transmission line group data; dist represents the spatial distance from the group point to the aggregation point in the group data; D represents the direction feature of each point in the group data. The operation flow of the feature extraction operator is as follows: ; In the formula, f5 represents the feature output by the feature extraction operator; Max() represents the max pooling operation; The calculation process of the distance attention weighting module is as follows:

[0022] In the formula, f6 represents the feature output by the distance attention weighting module; Represents the dot product operation of matrices; ( ) represents the softmax activation function.

[0023] As a preferred embodiment of the present invention, the operation flow of the fully connected output module in S43 is as follows:

[0024] In the formula, Q is the probability that the current calculation point is within a dense power transmission channel; ( ) represents the sigmoid activation function; FC() represents the fully connected operation; DO() represents the Dropout processing; When Q is greater than the preset threshold, the current calculation point is located within the dense power transmission channel; otherwise, it is not located within the dense power transmission channel.

[0025] A dense power transmission channel identification device based on a two-stage deep network includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform any of the above-described dense power transmission channel identification methods based on a two-stage deep network.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention significantly simplifies the scale of point cloud data by dividing and refining it into grids. Then, feature point cloud data of the transmission lines is extracted using a transmission line feature information extraction network and fed into a dense transmission channel identification network for recognition, thus obtaining accurate dense transmission channel data. Compared with traditional manual identification methods, this invention improves the accuracy and efficiency of identification, enabling maintenance personnel to accurately manage risks in dense transmission channel areas. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a dense power transmission channel identification method based on a two-stage deep network as described in Embodiment 1 of the present invention. Figure 2 This is a flowchart illustrating a dense power transmission channel identification method based on a two-stage deep network as described in Embodiment 2 of the present invention. Figure 3 This is a schematic diagram of the architecture of the line feature extraction network in a dense transmission channel identification method based on a two-stage deep network as described in Embodiment 2 of the present invention. Figure 4 This is a schematic diagram of the architecture of the dense power transmission channel identification network in the dense power transmission channel identification method based on a two-stage deep network described in Embodiment 2 of the present invention; Figure 5 This is a schematic diagram of the final recognition effect in the dense power transmission channel identification method based on a two-stage deep network described in Embodiment 2 of the present invention. Figure 6 This is a schematic diagram of the structure of a dense power transmission channel identification device based on a two-stage deep network, as described in Embodiment 3 of the present invention. Detailed Implementation

[0028] The present invention will be further described in detail below with reference to experimental examples and specific embodiments. However, this should not be construed as limiting the scope of the above-mentioned subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0029] Example 1 like Figure 1 As shown, a method for identifying dense power transmission channels based on a two-stage deep network includes the following steps: S1: Obtain point cloud data of the power transmission line scene to be identified.

[0030] S2: Divide the point cloud data into grids, and simplify the point cloud data in each grid to output simplified point cloud data.

[0031] S3: Input the simplified point cloud data into the pre-constructed power transmission line feature information extraction network to extract the feature point cloud data of the power transmission line; the feature point cloud data includes the three-dimensional coordinates and direction features of the power transmission line.

[0032] S4: Input the feature point cloud data into the pre-constructed dense power transmission channel identification network.

[0033] S5: The dense power transmission channel identification network determines whether there are dense channels in the scene of the power transmission line to be identified.

[0034] If it exists, output the three-dimensional coordinates of all transmission lines within the dense channel; If it does not exist, output the result that there are no dense power transmission channels in the scene of the power transmission line to be identified.

[0035] Example 2 This embodiment is a specific implementation of the dense transmission channel identification method based on a two-stage deep network described in Embodiment 1, such as... Figure 2 As shown, it includes the following steps: S1: Point cloud data acquisition: Acquire point cloud data of the power transmission line scene to be identified.

[0036] Furthermore, in this embodiment, airborne lidar is used to collect point cloud data of the power transmission line scene. The specific steps are as follows: S11: Determine the area for collecting point cloud data of transmission lines, and then formulate the flight path of the UAV; S12: The drone is equipped with a lidar to collect point cloud data of the power transmission line scene along the flight route.

[0037] S2: Point cloud data simplification: The point cloud data is divided into grids, and the point cloud data in each grid is simplified to output simplified point cloud data.

[0038] S21: Divide the point cloud data into grids according to a preset grid size (e.g., 10m × 10m); S22: Calculate the minimum Z-axis coordinate Zmin for each grid, and add the minimum Z-axis coordinate Zmin to the preset simplification height as the simplification threshold for the current grid; Since 500kV and above high-voltage transmission lines are at least 10 meters above the ground, point cloud data with a height less than Zmin+10 meters within the grid area are removed, and the remaining point cloud data are the point cloud data of transmission lines and towers.

[0039] S23: Filter out point cloud data in each grid whose Z-axis coordinate is less than the corresponding simplification threshold; remove point cloud data of objects such as ground, vegetation, and buildings.

[0040] S24: After all the point cloud data of the grids has been simplified, the output is the simplified point cloud data.

[0041] S3: Line feature extraction: Input the simplified point cloud data into the pre-constructed transmission line feature information extraction network to extract the feature point cloud data of the transmission line; the feature point cloud data includes the three-dimensional coordinates and direction features of the transmission line.

[0042] like Figure 3 As shown, the power transmission line feature information extraction network includes three sequentially connected feature extraction modules, a feature cross-fusion module (FCFM) that cross-fuses the output features of the three feature extraction modules, and a dual-branch output module (DBOM) that extracts feature point cloud data of the power transmission line based on the fused feature information. The power transmission line feature information extraction network includes the following operating steps: S31: Input the simplified point cloud data into the first feature extraction module and output the extracted features once; input the extracted features once into the second feature extraction module and output the extracted features twice; input the extracted features twice into the third feature extraction module and output the extracted features three times. The feature extraction module includes a K-feature nearest neighbor grouping layer (KFNNL) and a feature extraction operator (FEO) connected in sequence; it includes the following operation steps: K-feature nearest neighbor grouping layer: The input is a point cloud data of N points. The feature distance between each point and the other points is calculated. The K points with the farthest feature distance from the current point are selected as grouping points. After all points are calculated, the point cloud group data of N×(K+1) points is output, where K is a set value. Feature extraction operators: Its working expression is:

[0043] In the formula, I represents the input data; G() represents the K-feature nearest neighbor grouping layer (i.e., extracting K points that are closest to the center point feature); Conv() represents the convolution operation (in this embodiment, the convolution kernel size involved in the convolution operation is 1×1); BN() represents the batch normalization operation; ( ) represents the ReLU activation function; Max() represents the max pooling operation; f represents the feature extraction result.

[0044] S32: Cross-fuse the features extracted in the first, second, and third extractions, and output the fused feature information; Feature cross-fusion module:

[0045]

[0046]

[0047] In the formula, f1, f2, and f3 represent the output features after one feature extraction module operation, two feature extraction module operations, and three feature extraction module operations, respectively; CAT() represents the feature concatenation operation; MHA() represents the point cloud multi-head attention mechanism; f 12 This represents the feature resulting from the cross-fusion of the output features of the first and second operations; f 23 This represents the feature resulting from the cross-fusion of the output features of the second and third operations; f 123 This represents the feature resulting from the cross-fusion of the output features from the three operations.

[0048] S33: The dual-branch output module outputs the probability that each point in the simplified point cloud data belongs to the transmission line, as well as the direction feature, based on the fused feature information; Dual-branch output module: Branch Road 1:

[0049] Branch Road 2:

[0050] In the formula, FC() represents a fully connected operation; BN() represents a batch normalization operation; ( ) represents the ReLU activation function; DO() represents Dropout processing, which in this embodiment masks all nodes in the fully connected layer with a probability of 20%. ( ) represents the softmax activation function; ( ) represents the sigmoid activation function; P represents the probability that the point belongs to the transmission line; D represents the direction of the transmission line, including the direction features in the X and Y directions (dx, dy).

[0051] S34: Based on a preset probability threshold, filter out point clouds that do not belong to the transmission line, and output the three-dimensional coordinates and orientation features of the remaining point cloud data as feature point cloud data.

[0052] This step achieves secondary data simplification by removing point clouds of non-transmission lines while retaining the original 3D coordinates and directional features of the transmission line point clouds.

[0053] Furthermore, the power transmission line feature information extraction network is pre-trained using the labeled training dataset until the model converges.

[0054] S4: Dense Channel Identification: Input the feature point cloud data into a pre-constructed dense power transmission channel identification network.

[0055] like Figure 4 As shown, the dense power transmission channel identification network includes a Directed Grouping Layer (DGL), a Graph Convolutional Feature Extraction Module (GCFEM), a Feature Extraction Operator (FEO), a Distance Attention Weighted Module (DAWM), and a Fully Connected Output Module (FCOM) connected in sequence; the dense power transmission channel identification network includes the following operating steps: S41: The directional grouping layer is used to group the feature point cloud data of the transmission line; Input the feature point cloud data of the power transmission line, which includes the three-dimensional coordinates and orientation features of M power transmission line points; Choose any two points i and j, and calculate the spatial straight-line distance between the i-th point and the j-th point based on the three-dimensional coordinates of the two points, i∈[1,M]; j∈[1,M]; The expression for calculating the direction angle between the i-th point and the j-th point of the transmission line is:

[0056] In the formula, θ ij d represents the direction angle between the i-th point and the j-th point; xi and d yi The directional characteristics of point i; Δx ij and Δy ij represents the coordinate deviation between point i and point j on the X and Y axes; arctan() represents the arctangent operation; | represents the absolute value calculation; Using the i-th point as the aggregation point, the farthest point sampling algorithm is used to select L grouping points that are farthest apart from each other and whose direction angle is less than 10° as grouping data, resulting in M ​​groups × (L+1) points of transmission line grouping data, where L is a set value.

[0057] S42: The feature point cloud data after grouping is sequentially processed by the graph convolution feature extraction module, the feature extraction operator, and the distance attention weighting module to extract features. Graph convolution feature extraction module:

[0058] In the formula, f4 represents the features output by the graph convolution feature extraction module; ( ) represents the ReLU activation function; BN() represents batch normalization operation; Conv() represents convolution operation; CAT() represents feature concatenation operation; G2 represents the three-dimensional coordinates of the transmission line group data; dist represents the spatial distance from the group point to the aggregation point in the group data; D represents the direction feature of each point in the group data. The operation flow of the feature extraction operator is as follows: ; In the formula, f5 represents the feature output by the feature extraction operator; Max() represents the max pooling operation; The calculation process of the distance attention weighting module is as follows:

[0059] In the formula, f6 represents the feature output by the distance attention weighting module; Represents the dot product operation of matrices; ( ) represents the softmax activation function.

[0060] S43: The fully connected output module determines the probability that each point is in a dense power transmission channel based on the characteristics output in S42.

[0061] The operation flow of the fully connected output module is as follows:

[0062] In the formula, Q is the probability that the current calculation point is within a dense power transmission channel; ( ) represents the sigmoid activation function; FC() represents the fully connected operation; DO() represents the Dropout processing; When Q is greater than the preset threshold (in this embodiment, the preset threshold = 0.6), the current calculation point is located within the dense power transmission channel; otherwise, it is not located within the dense power transmission channel.

[0063] Furthermore, the dense power transmission channel identification network is pre-trained using a labeled training dataset until the model converges.

[0064] S5: Recognition result output: The dense power transmission channel recognition network determines whether there are dense channels in the scene of the power transmission line to be identified.

[0065] If present, output the three-dimensional coordinates of all transmission lines within the dense channel; this data can be transcribed into a txt file for output to other devices, or... Figure 5 As shown, the output is in the form of an image.

[0066] If it does not exist, output the result that there are no dense power transmission channels in the scene of the power transmission line to be identified.

[0067] Example 3 like Figure 6As shown, a dense power transmission channel identification device based on a two-stage deep network includes at least one processor, a memory communicatively connected to the at least one processor, and at least one input / output interface communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enables the at least one processor to perform the dense power transmission channel identification method based on a two-stage deep network described in the foregoing embodiments. The input / output interface may include a display, keyboard, mouse, and USB interface for inputting and outputting data.

[0068] Furthermore, the dense transmission channel identification device based on two-stage deep networks can be a desktop computer, mobile phone, tablet computer, wearable device, or other device capable of deep information identification based on two-stage deep networks.

[0069] Furthermore, the processor may include one or more processing cores. The processor connects various parts within the dense power transmission channel identification device based on a two-stage deep network using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory, and by calling data stored in memory. Optionally, the processor may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor and may be implemented separately using a communication chip.

[0070] The memory may include random access memory (RAM) or read-only memory (ROM). The memory can be used to store instructions, programs, code, code sets, or instruction sets, such as instructions or code sets used to implement the dense transmission channel identification method based on a two-stage deep network provided in this application embodiment. The memory may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the various method embodiments described above, etc. The data storage area may also store data created during the use of the dense transmission channel identification device based on a two-stage deep network (such as a mapping table of modulation sequences and depth, image data, spectrogram data, etc.).

[0071] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0072] When the integrated units of the present invention are implemented as software functional units and sold or used as independent products, they can also be stored in a computer-readable storage medium. The computer-readable storage medium stores program code, which can be called by a processor to execute the methods described in the above method embodiments. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes electronic memories such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium includes a non-transitory computer-readable storage medium. The computer-readable storage medium has storage space for program code that executes any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code can be compressed, for example, in an appropriate form.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying dense power transmission channels based on a two-stage deep network, characterized in that, Includes the following steps: S1: Obtain point cloud data of the power transmission line scene to be identified; S2: Divide the point cloud data into grids, and simplify the point cloud data in each grid to output simplified point cloud data; S3: Input the simplified point cloud data into the pre-constructed power transmission line feature information extraction network to extract the feature point cloud data of the power transmission line; The feature point cloud data includes the three-dimensional coordinates and orientation features of the power transmission line; S4: Input the feature point cloud data into the pre-constructed dense power transmission channel identification network; S5: The dense power transmission channel identification network determines whether there are dense channels within the scene of the power transmission line to be identified; If it exists, output the three-dimensional coordinates of all transmission lines within the dense channel; If not, output the result that there are no dense power transmission channels in the scene of the power transmission line to be identified; S2 includes the following steps: S21: Divide the point cloud data into grids according to a preset grid size; S22: Calculate the minimum Z-axis coordinate of each grid, and add the minimum Z-axis coordinate to the preset simplification height as the simplification threshold of the current grid; S23: Filter out point cloud data in each grid whose Z-axis coordinate is less than the corresponding simplification threshold; S24: After all the point cloud data of the grids has been simplified, the output is the simplified point cloud data.

2. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 1, characterized in that, The S3 transmission line feature information extraction network includes three sequentially connected feature extraction modules, a feature cross-fusion module that cross-fused the output features of the three feature extraction modules, and a dual-branch output module that extracts feature point cloud data of the transmission line based on the fused feature information. The power transmission line feature information extraction network includes the following operational steps: S31: Input the simplified point cloud data into the first feature extraction module and output the extracted features once; input the extracted features once into the second feature extraction module and output the extracted features twice; input the extracted features twice into the third feature extraction module and output the extracted features three times. S32: The feature cross-fusion module performs cross-fusion on the features extracted in the first, second, and third extractions, and outputs the fused feature information; S33: The dual-branch output module outputs the probability that each point in the simplified point cloud data belongs to the transmission line, as well as the direction feature, based on the fused feature information; S34: Based on a preset probability threshold, filter out point clouds that do not belong to the transmission line, and output the three-dimensional coordinates and orientation features of the remaining point cloud data as feature point cloud data.

3. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 2, characterized in that, The feature extraction module includes a K-feature nearest neighbor grouping layer and a feature extraction operator connected in sequence; The following steps are included: K-feature nearest neighbor grouping layer: The input is a point cloud data of N points. The feature distance between each point and the other points is calculated. The K points with the farthest feature distance from the current point are selected as grouping points. After all points are calculated, the point cloud group data of N×(K+1) points is output, where K is a set value. Feature extraction operators: In the formula, I This represents the input data; G() represents the K-feature nearest neighbor grouping layer; Conv() represents the convolution operation; BN() represents the batch normalization operation; ( ) represents the ReLU activation function; Max() represents the max pooling operation; f represents the feature extraction result.

4. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 3, characterized in that, The feature cross-fusion module in S32 includes the following operating steps: In the formula, f1, f2 and f3 represent the first-order extracted features, second-order extracted features and third-order extracted features output by the three feature extraction modules, respectively; CAT() represents the feature concatenation operation; MHA() represents the multi-head attention mechanism for point clouds; f 12 This represents the feature resulting from the cross-fusion of the output features of the first and second operations; f 23 This represents the feature resulting from the cross-fusion of the output features of the second and third operations; f 123 This represents the feature resulting from the cross-fusion of the output features from the three operations.

5. A method for identifying dense power transmission channels based on a two-stage deep network according to claim 4, characterized in that, The dual-branch output module in S33 includes the following operating steps: Branch Road 1: Branch Road 2: In the formula, FC() represents a fully connected operation; BN() represents a batch normalization operation; ( ) represents the ReLU activation function; DO() represents Dropout processing; ( ) represents the softmax activation function; ( ) represents the sigmoid activation function; P represents the probability that the point belongs to the transmission line; D represents the direction of the transmission line, including the direction features in the X and Y directions (dx, dy).

6. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 1, characterized in that, The dense power transmission channel identification network in S4 includes a directional grouping layer, a graph convolutional feature extraction module, a feature extraction operator, a distance attention weighting module, and a fully connected output module connected in sequence; the dense power transmission channel identification network includes the following operating steps: S41: The directional grouping layer is used to group the feature point cloud data of the transmission line; S42: The feature point cloud data after grouping is sequentially processed by the graph convolution feature extraction module, the feature extraction operator, and the distance attention weighting module to extract features. S43: The fully connected output module determines the probability that each point is in a dense power transmission channel based on the characteristics output in S42.

7. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 6, characterized in that, S41 includes the following steps: Input the feature point cloud data of the power transmission line, which includes the three-dimensional coordinates and orientation features of M power transmission line points; Choose any two points i and j, and calculate the spatial straight-line distance between the i-th point and the j-th point based on the three-dimensional coordinates of the two points, i∈[1,M]; j∈[1,M]; The expression for calculating the direction angle between the i-th point and the j-th point of the transmission line is: In the formula, θ ij d represents the direction angle between the i-th point and the j-th point; xi and d yi The directional characteristics of point i; Δx ij and Δy ij represents the coordinate deviation between point i and point j on the X and Y axes; arctan() represents the arctangent operation; | represents the absolute value calculation; Using the i-th point as the aggregation point, the farthest point sampling algorithm is used to select L grouping points that are farthest apart from each other and whose direction angle is less than 10° as grouping data, resulting in M ​​groups × (L+1) points of transmission line grouping data, where L is a set value.

8. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 7, characterized in that, S42 Includes the following steps: Graph convolution feature extraction module: In the formula, f4 represents the features output by the graph convolution feature extraction module; ( ) represents the ReLU activation function; BN() represents batch normalization operation; Conv() represents convolution operation; CAT() represents feature concatenation operation; G2 represents the three-dimensional coordinates of the transmission line group data; dist represents the spatial distance from the group point to the aggregation point in the group data; D represents the direction feature of each point in the group data. The operation flow of the feature extraction operator is as follows: ； In the formula, f5 represents the feature output by the feature extraction operator; Max() represents the max pooling operation; The calculation process of the distance attention weighting module is as follows: In the formula, f6 represents the feature output by the distance attention weighting module; Represents the dot product operation of matrices; ( ) represents the softmax activation function.

9. The method for identifying dense power transmission channels based on a two-stage deep network according to claim 8, characterized in that, The operation flow of the fully connected output module in S43 is as follows: In the formula, Q is the probability that the current calculation point is within a dense power transmission channel; ( ) represents the sigmoid activation function; FC() represents the fully connected operation; DO() represents the Dropout processing; When Q is greater than the preset threshold, the current calculation point is located within the dense power transmission channel; otherwise, it is not located within the dense power transmission channel.

10. A dense power transmission channel identification device based on a two-stage deep network, characterized in that, The method includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform a dense transmission channel identification method based on a two-stage deep network according to any one of claims 1 to 9.

Images