CN117635630A - Automatic object segmentation method and system based on substation scenic spot cloud - Google Patents
Automatic object segmentation method and system based on substation scenic spot cloud Download PDFInfo
- Publication number
- CN117635630A CN117635630A CN202311321733.0A CN202311321733A CN117635630A CN 117635630 A CN117635630 A CN 117635630A CN 202311321733 A CN202311321733 A CN 202311321733A CN 117635630 A CN117635630 A CN 117635630A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- point
- level
- cloud data
- substation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 65
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000002372 labelling Methods 0.000 claims description 22
- 238000007781 pre-processing Methods 0.000 claims description 21
- 238000009826 distribution Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 13
- 230000008520 organization Effects 0.000 claims description 13
- 238000000638 solvent extraction Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 9
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000005192 partition Methods 0.000 claims 1
- 238000004364 calculation method Methods 0.000 abstract description 7
- 238000012545 processing Methods 0.000 abstract description 4
- 230000002776 aggregation Effects 0.000 abstract description 3
- 238000004220 aggregation Methods 0.000 abstract description 3
- 238000004088 simulation Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000012212 insulator Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an automatic object segmentation method and system based on substation scenic spot cloud. The multi-scale characteristics of the point cloud data can be captured, and the effective representation can be obtained from the global to the local. This is particularly important in the case where there are scale differences and multi-scale information in the processing point cloud data; the Pointet++ can extract features on each local area, acquire more global context information by using hierarchical aggregation operation, fully utilize the local and global information to perform more accurate segmentation tasks, and can operate more efficiently on each local area, thereby reducing calculation and memory requirements. Meanwhile, through the hierarchical structure, the hierarchy of the segmentation operation can be selected pertinently, and the calculation efficiency is further improved.
Description
Technical Field
The invention relates to the technical field of computer vision and image processing, in particular to an automatic object segmentation method and system based on substation scenic spot cloud.
Background
Three-dimensional vision technology has grown more mature over the years and is widely used in industrial production. In order to improve the management intelligent degree of the transformer substation, the three-dimensional technology can be utilized to comprehensively display data, images and videos of the transformer substation. When the three-dimensional modeling is carried out, because the scene of the transformer substation is huge and complex, in order to ensure the accuracy and the high efficiency of the model acquisition, the laser scanning can be adopted to acquire the point cloud data. The currently used point cloud segmentation method comprises a graph-based method: modeling point cloud data into a graph structure, wherein points of the point cloud serve as nodes of the graph, and dividing by calculating edges and edge characteristics among the nodes; region-based methods: dividing the point cloud data into areas with continuity, and marking the areas by using a clustering or segmentation algorithm; deep learning-based method: the method utilizes deep learning, such as convolutional neural networks, graph convolution neural networks and attention mechanisms, to segment point cloud data; feature-based methods: segmentation is performed by extracting features in the point cloud data, such as shape descriptors, normals, curvature, color, and the like. The features are then analyzed and segmented using only clustering or other algorithms. Method based on aggregate characteristics: and (3) segmenting the point cloud data by utilizing the aggregate characteristics such as the position, the distance, the normal direction and the like of the point cloud. For example, a reference plane-based segmentation method, a curvature-based segmentation method, and the like.
The above methods depend on the distribution condition of the object or the selection of initial seed points to a great extent, so that the robustness is not strong. The simultaneous point cloud file contains a large number of points, and a plurality of useless noise points exist in the simultaneous point cloud file, so that the point cloud needs to be clustered and segmented, and effective points are extracted so as to reconstruct a three-dimensional model.
Disclosure of Invention
The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides the automatic object segmentation method and the system based on the substation scenic spot cloud, which solve the problem that the traditional method at present depends on the distribution condition of objects or the selection of initial seed points to a great extent, so that the robustness is not strong; the simultaneous point cloud file contains a large number of points, and there are a plurality of useless noise points, so that the effective points cannot be extracted.
In order to solve the technical problems, the invention provides the following technical scheme:
in a first aspect, the invention provides an automatic object segmentation method based on a substation scenic spot cloud, which comprises the following steps:
acquiring cloud data of a scenic spot and corresponding labeling information;
preprocessing the live-action point cloud data by a self-adaptive layering method, wherein the preprocessing divides the point cloud data into different levels and arranges an organization mode of points;
extracting features of the point clouds of each level, and fusing the features of the point clouds of different levels to obtain global and comprehensive feature representations;
using a full connection layer to conduct category prediction on each point, partitioning and semantic marking on point clouds according to prediction results, and training and optimizing;
and dividing the new live-action point cloud by using the trained network to obtain semantic labels of each point so as to divide, extract and analyze different objects according to the labels.
As a preferable scheme of the automatic object segmentation method based on the substation scenic spot cloud, the invention comprises the following steps: the obtaining of the real scenic spot cloud data and the corresponding labeling information, wherein the obtaining of the real scenic spot cloud data comprises,
acquiring geometric shapes and position information of buildings, equipment, transmission lines and heat distribution;
the corresponding labeling information comprises category labeling, bounding box labeling and point cloud labeling.
As a preferable scheme of the automatic object segmentation method based on the substation scenic spot cloud, the invention comprises the following steps: preprocessing the live-action point cloud data by an adaptive layering method, wherein the preprocessing divides the point cloud data into different levels and arranges the organization modes of points, and comprises,
constructing a hierarchical structure according to the distribution and characteristics of point clouds by an octree method to obtain point clouds of different levels, wherein the method specifically comprises the steps of setting four levels;
the first level comprises cloud data of the whole transformer station, and global characteristics of the transformer station are provided;
the second level divides the point cloud data of the substation heat distribution into different heat distribution areas according to the point cloud data;
the third level divides the cloud data of the power transformation station into different equipment function areas according to the distribution and complexity of the cloud data;
the fourth level meshing according to the device pass-through type and function combinations inside each zone.
As a preferable scheme of the automatic object segmentation method based on the substation scenic spot cloud, the invention comprises the following steps: the arrangement of the points is used for obtaining point clouds with different densities and resolutions of different layers, and the arrangement comprises the following steps of,
setting a first neighborhood radius threshold value in the first level and the second level in a local neighborhood point cloud mode to determine corresponding local neighborhood;
setting a second neighborhood radius threshold value in the third level and the fourth level in a local neighborhood point cloud mode to determine corresponding local neighborhood;
the first neighborhood radius threshold is greater than the second neighborhood radius threshold.
As a preferable scheme of the automatic object segmentation method based on the substation scenic spot cloud, the invention comprises the following steps: further comprising, after determining the local neighborhood, setting a resolution of each level according to the relative position of the points or the resolution projected onto different grids;
the first level and the second level set a first resolution;
setting a second resolution by the third level and the fourth level;
the first resolution is less than the second resolution;
in the hierarchical structure, for each level of point cloud data, local coordinates with respect to that level are calculated for use in the segmentation stage.
As a preferable scheme of the automatic object segmentation method based on the substation scenic spot cloud, the invention comprises the following steps: feature extraction is performed for each level of point cloud, and point cloud features of different levels are fused to obtain global and comprehensive feature representations, including,
at each level after self-adaptive layering, performing feature extraction on local point cloud data through a Pointernet++ grid structure, wherein the feature extraction comprises multi-layer perception and feature extraction on the local point cloud;
using the full connection layer to conduct category prediction on each point, conducting segmentation and semantic marking on the point cloud according to the prediction result, conducting training and optimization, including,
aiming at a specific point cloud segmentation task, introducing a segmentation head network into the Pointernet++ to output semantic segmentation labels of each point, and adjusting the segmentation head according to task requirements;
in the training phase, point cloud data with known labels are used for supervised learning, parameters of the network are updated through back propagation and optimization algorithms, and the aim is to minimize the loss function between the predicted label and the real label.
As a preferable scheme of the automatic object segmentation method based on the substation scenic spot cloud, the invention comprises the following steps: dividing the new live-action point cloud by using the trained network to obtain semantic labels of each point so as to divide, extract and analyze different objects according to the labels, including,
in the test stage, the trained Pointnet++ network model is used for dividing the new point cloud data, forward propagation is carried out on each point, and corresponding labels are distributed to the points according to the final dividing result.
In a second aspect, the present invention provides a system for automatic object segmentation based on substation point cloud, comprising,
the acquisition module is used for acquiring the cloud data of the scenic spots and the corresponding labeling information;
the preprocessing module is used for preprocessing the live-action point cloud data through a self-adaptive layering method, and the preprocessing divides the point cloud data into different levels and arranges an organization mode of points;
the fusion module is used for extracting the characteristics of the point clouds of each level and fusing the characteristics of the point clouds of different levels to obtain global and comprehensive characteristic representations;
the training module is used for carrying out category prediction on each point by using the full-connection layer, carrying out segmentation and semantic marking on the point cloud according to the prediction result, and carrying out training and optimization;
the segmentation module is used for segmenting the new live-action point cloud by utilizing the trained network to obtain semantic labels of each point so as to segment, extract and analyze different objects according to the labels.
In a third aspect, the present invention provides a computing device comprising:
a memory and a processor;
the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions, and the computer executable instructions realize the steps of the automatic object segmentation method based on the substation scenic spot cloud when being executed by the processor.
In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions which, when executed by a processor, implement the steps of the substation-based method for automatically partitioning objects of a real-scene point cloud.
Compared with the prior art, the invention has the beneficial effects that: according to the invention, the distribution and the characteristics of the point cloud data are dynamically adjusted and sampled by the self-adaptive layering method, so that each level has proper points and distribution. The multi-scale characteristics of the point cloud data can be captured, and the effective representation can be obtained from the global to the local. This is particularly important in the case where there are scale differences and multi-scale information in the processing point cloud data; the Pointet++ can extract features on each local area, acquire more global context information by using hierarchical aggregation operation, fully utilize the local and global information to perform more accurate segmentation tasks, and can operate more efficiently on each local area, thereby reducing calculation and memory requirements. Meanwhile, through the hierarchical structure, the hierarchy of the segmentation operation can be selected pertinently, and the calculation efficiency is further improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Wherein:
fig. 1 is an overall flow diagram of an object automatic segmentation method based on a substation scenic spot cloud according to an embodiment of the invention.
Detailed Description
So that the manner in which the above recited objects, features and advantages of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
While the embodiments of the present invention have been illustrated and described in detail in the drawings, the cross-sectional view of the device structure is not to scale in the general sense for ease of illustration, and the drawings are merely exemplary and should not be construed as limiting the scope of the invention. In addition, the three-dimensional dimensions of length, width and depth should be included in actual fabrication.
Also in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer", etc. are based on the directions or positional relationships shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the system or element to be referred to must have a specific direction, be constructed and operated in the specific direction, and thus should not be construed as limiting the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected, and coupled" should be construed broadly in this disclosure unless otherwise specifically indicated and defined, such as: can be fixed connection, detachable connection or integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or may be indirectly connected through an intermediate medium, or may be a communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
Example 1
Referring to fig. 1, for one embodiment of the present invention, there is provided an object automatic segmentation method based on a substation scenic spot cloud, including:
s1: acquiring cloud data of a scenic spot and corresponding labeling information;
further, the method for obtaining the cloud data of the real scenic spot and the corresponding labeling information, wherein the step of obtaining the cloud data of the real scenic spot comprises the steps of,
acquiring geometric shapes and position information of buildings, equipment, transmission lines and heat distribution;
the corresponding labeling information comprises category labeling, bounding box labeling and point cloud labeling.
It should be noted that the acquired data may be obtained by a laser scanner, a thermal imager, etc., and the laser scanner may measure the distance and position of the object to generate the point cloud data. Three-dimensional point cloud data of each part of the transformer substation can be obtained through a laser scanner; the thermal imager may measure infrared radiation of the object to produce a thermal map. The thermal image data of each part of the transformer substation can be obtained through the thermal imager, so that the data set contains enough samples to represent each aspect of the scene of the transformer substation. Meanwhile, the collected live-action data is marked, so that more semantic information and scene understanding can be provided.
S2: preprocessing the real scenic spot cloud data by a self-adaptive layering method, dividing the point cloud data into different levels by preprocessing, and arranging an organization mode of points;
furthermore, the cloud data of the scenic spots is preprocessed by an adaptive layering method, the preprocessing divides the cloud data into different levels and arranges the organization modes of the spots, including,
constructing a hierarchical structure according to the distribution and characteristics of point clouds by an octree method to obtain point clouds of different levels, wherein the method specifically comprises the steps of setting four levels;
the first level comprises cloud data of the whole transformer station, and global characteristics of the transformer station are provided;
the second level divides the point cloud data of the substation heat distribution into different heat distribution areas according to the point cloud data;
the third level divides the cloud data of the power transformation station into different equipment function areas according to the distribution and complexity of the cloud data;
the fourth level meshing according to the device pass-through type and function combinations inside each zone.
It should be noted that, for the substation scenario, 4 levels are preferably established to construct in consideration of the calculation amount and the level complexity, and each level can perform subsequent local neighborhood setting according to its own characteristics. Meanwhile, each level is dynamically adjusted and sampled according to the density and characteristics of the point cloud by an octree (octree) method, so that each level can be ensured to have proper points and distribution, for example, sub-sampling or segmentation according to the density of the points.
It should be further noted that the purpose of the adaptive layering is to divide the substation cloud data into multiple layers according to the live-action type, so that the point clouds of different layers have different densities and resolutions, and the purpose of the adaptive layering is to capture the multi-scale characteristics of the point cloud data.
Furthermore, the point cloud of different layers is obtained by arranging the point organization mode with different densities and resolutions, which comprises the following steps,
setting a first neighborhood radius threshold value in the first level and the second level in a local neighborhood point cloud mode to determine corresponding local neighborhood;
setting a second neighborhood radius threshold value in the third level and the fourth level in a local neighborhood point cloud mode to determine corresponding local neighborhood;
the first neighborhood radius threshold is greater than the second neighborhood radius threshold.
It should be noted that the local neighborhood point cloud mode is selected to be applicable to a multi-device scene of a transformer substation, so that the local structure and the form of the point cloud can be reserved, the detection and the analysis of the details of a specific area are facilitated, and meanwhile, the local neighborhood point cloud is more suitable for detecting specific devices or executing semantic segmentation tasks.
Meanwhile, in order to reduce the calculated amount, only two radius thresholds are set between the layers, so that the suitability of the calculated amount is ensured according to different area divisions while the proper density is ensured, and the calculation efficiency is improved.
Still further, after determining the local neighborhood, setting a resolution of each level based on the relative locations of the points or the resolutions projected to the different grids;
the first level and the second level set a first resolution;
setting a second resolution by the third level and the fourth level;
the first resolution is less than the second resolution;
in the hierarchical structure, for each level of point cloud data, local coordinates with respect to that level are calculated for use in the segmentation stage.
It should be noted that setting different resolutions for different levels of identified segmentation requirements will help preserve local structure and morphology information in a more detailed way and increase the corresponding computational efficiency. The step of determining the hierarchy according to the relative positions or the resolutions projected to different grids helps to determine the segmentation of the point cloud and the division of the local area, and the organization mode of the local neighborhood point cloud further processes the point cloud data of each hierarchy so as to preserve local features and extract detailed information.
For the resolution, the first resolution may be set to 1 meter/cell and the second resolution may be set to 0.2 meter/cell.
S3: extracting features of the point clouds of each level, and fusing the features of the point clouds of different levels to obtain global and comprehensive feature representations;
further, feature extraction is performed for each level of point cloud, and point cloud features of different levels are fused to obtain global and comprehensive feature representations, including,
and at each level after self-adaptive layering, performing feature extraction on the local point cloud data through a Pointernet++ grid structure, wherein the feature extraction comprises multi-layer perception and feature extraction on the local point cloud.
It should be noted that, using a Pointnet++ network to perform the special extraction, by stacking a plurality of Pointnet++ modules, the point cloud data can be projected onto a specific space (nodes of octree), features can be extracted using a plurality of MLPs (multi-layer perceptrons), and the local features can be aggregated by using the maximum pooling operation, so that the multi-scale features of the point cloud data can be gradually extracted, which will provide a feature representation with rich semantic information for each point of the point cloud; the hierarchical aggregation operation can be utilized to obtain more global context information, and the network can fully utilize the local and global information to perform more accurate segmentation tasks.
S4: using the full connection layer to conduct category prediction on each point, conducting segmentation and semantic marking on the point cloud according to the prediction result, conducting training and optimization, including,
aiming at a specific point cloud segmentation task, introducing a segmentation head network into the Pointernet++ to output semantic segmentation labels of each point, and adjusting the segmentation head according to task requirements;
in the training phase, point cloud data with known labels are used for supervised learning, parameters of the network are updated through back propagation and optimization algorithms, and the aim is to minimize the loss function between the predicted label and the real label.
S5: dividing the new live-action point cloud by using the trained network to obtain semantic labels of each point so as to divide, extract and analyze different objects according to the labels, including,
in the test stage, the trained Pointnet++ network model is used for dividing the new point cloud data, forward propagation is carried out on each point, and corresponding labels are distributed to the points according to the final dividing result.
It should be noted that direct Pointnet++ processing of the entire high-resolution point cloud may result in excessive computational complexity. The my schema Pointet++ can operate more efficiently on each local area, reducing computation and memory requirements. Meanwhile, through the hierarchical structure, the hierarchy of the segmentation operation can be selected pertinently, and the calculation efficiency is further improved.
The above is a schematic scheme of the object automatic segmentation method based on the substation scenic spot cloud in this embodiment. It should be noted that, the technical solution of the system for automatically dividing the object based on the substation real-scene point cloud and the technical solution of the method for automatically dividing the object based on the substation real-scene point cloud belong to the same concept, and in this embodiment, details of the technical solution of the system for automatically dividing and pushing the object based on the substation real-scene point cloud are not described in detail, and all reference may be made to the description of the technical solution of the method for automatically dividing the object based on the substation real-scene point cloud.
In this embodiment, an object automatic segmentation system based on a substation scenic spot cloud includes:
the acquisition module is used for acquiring the cloud data of the scenic spots and the corresponding labeling information;
the preprocessing module is used for preprocessing the real scenic spot cloud data through a self-adaptive layering method, dividing the point cloud data into different levels and arranging an organization mode of the points;
the fusion module is used for extracting the characteristics of the point clouds of each level and fusing the characteristics of the point clouds of different levels to obtain global and comprehensive characteristic representations;
the training module is used for carrying out category prediction on each point by using the full-connection layer, carrying out segmentation and semantic marking on the point cloud according to the prediction result, and carrying out training and optimization;
the segmentation module is used for segmenting the new live-action point cloud by utilizing the trained network to obtain semantic labels of each point so as to segment, extract and analyze different objects according to the labels.
The embodiment also provides a computing device, which is suitable for the situation of automatic object segmentation based on the substation scenic spot cloud, and comprises the following steps:
a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the method for automatically dividing the object based on the substation scenic spot cloud according to the embodiment.
The present embodiment also provides a storage medium, on which a computer program is stored, which when executed by a processor implements the method for implementing automatic object segmentation based on a substation real-scene point cloud as set forth in the above embodiment.
The storage medium proposed in the present embodiment belongs to the same inventive concept as the method for implementing automatic object segmentation based on the substation scenic spot cloud proposed in the above embodiment, and technical details not described in detail in the present embodiment can be seen in the above embodiment, and the present embodiment has the same beneficial effects as the above embodiment.
From the above description of embodiments, it will be clear to a person skilled in the art that the present invention may be implemented by means of software and necessary general purpose hardware, but of course also by means of hardware, although in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as a floppy disk, a Read Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, etc., including several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to execute the method of the embodiments of the present invention.
Example 2
Referring to tables 1-3, for one embodiment of the invention, an object automatic segmentation method based on a substation scenic spot cloud is provided, and in order to verify the beneficial effects, an exemplary simulation application scene and a comparison scheme are provided for scientific demonstration.
Acquiring cloud data of a scenic spot and corresponding labeling information;
preprocessing the real scenic spot cloud data by a self-adaptive layering method, dividing the point cloud data into different levels by preprocessing, and arranging an organization mode of points;
extracting features of the point clouds of each level, and fusing the features of the point clouds of different levels to obtain global and comprehensive feature representations;
using a full connection layer to conduct category prediction on each point, partitioning and semantic marking on point clouds according to prediction results, and training and optimizing;
and dividing the new live-action point cloud by using the trained network to obtain semantic labels of each point so as to divide, extract and analyze different objects according to the labels.
According to the above scheme, specific patterns in a part of the substation hierarchy level are shown in table 1.
Table 1 specific modes in partial substation hierarchy level
In the test stage, the trained Pointnet++ network model is used for dividing the new point cloud data, and corresponding labels are distributed to each point. Table 2 is an example showing the results of partitioning point cloud data using the pointe++:
TABLE 2 segmentation results
Table 2 shows the result of dividing the example point cloud data, and according to the prediction tags, the functional area or the device type to which each point belongs can be determined.
The selection of the point cloud organization mode is related to S2, and comparison is performed through simulation data, see table 3.
Table 3 data example of local neighborhood point cloud mode in substation simulation scene
In the above table, the local density is the power demand density or the personnel density of a certain area in the substation. The vibration amplitude is used to evaluate the degree of vibration of the substation equipment. Current fluctuations are used to describe the degree of instability of the grid load. The temperature fluctuation is used for evaluating the temperature change condition of the substation equipment. The class label indicates the class of objects in the point cloud, such as transformers and insulators.
By selecting the local neighborhood point cloud, the parameter values of the local neighborhood point cloud are smaller for the same category (the transformer or the insulator). This shows that the local neighborhood point cloud approach has better capture capability for specific targets in the substation simulation scenario.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention.
Claims (10)
1. The automatic object segmentation method based on the substation scenic spot cloud is characterized by comprising the following steps of:
acquiring cloud data of a scenic spot and corresponding labeling information;
preprocessing the live-action point cloud data by a self-adaptive layering method, wherein the preprocessing divides the point cloud data into different levels and arranges an organization mode of points;
extracting features of the point clouds of each level, and fusing the features of the point clouds of different levels to obtain global and comprehensive feature representations;
using a full connection layer to conduct category prediction on each point, partitioning and semantic marking on point clouds according to prediction results, and training and optimizing;
and dividing the new live-action point cloud by using the trained network to obtain semantic labels of each point so as to divide, extract and analyze different objects according to the labels.
2. The method for automatically partitioning an object based on a real-scene point cloud of a transformer substation according to claim 1, wherein the acquiring real-scene point cloud data and corresponding labeling information includes,
acquiring geometric shapes and position information of buildings, equipment, transmission lines and heat distribution;
the corresponding labeling information comprises category labeling, bounding box labeling and point cloud labeling.
3. The method for automatically partitioning the object based on the substation scenic spot cloud according to claim 1 or 2, wherein the preprocessing is performed on the scenic spot cloud data by an adaptive layering method, the preprocessing divides the point cloud data into different levels and arranges the organization modes of the points,
constructing a hierarchical structure according to the distribution and characteristics of point clouds by an octree method to obtain point clouds of different levels, wherein the method specifically comprises the steps of setting four levels;
the first level comprises cloud data of the whole transformer station, and global characteristics of the transformer station are provided;
the second level divides the point cloud data of the substation heat distribution into different heat distribution areas according to the point cloud data;
the third level divides the cloud data of the power transformation station into different equipment function areas according to the distribution and complexity of the cloud data;
the fourth level meshing according to the device pass-through type and function combinations inside each zone.
4. The method for automatically partitioning objects based on a substation real-time point cloud as set forth in claim 3, wherein said arranging points is organized in such a way as to obtain said point clouds of different levels having different densities and resolutions, specifically including,
setting a first neighborhood radius threshold value in the first level and the second level in a local neighborhood point cloud mode to determine corresponding local neighborhood;
setting a second neighborhood radius threshold value in the third level and the fourth level in a local neighborhood point cloud mode to determine corresponding local neighborhood;
the first neighborhood radius threshold is greater than the second neighborhood radius threshold.
5. The method for automatically segmenting an object based on a substation real-scene point cloud according to claim 4, further comprising, after determining the local neighborhood, setting the resolution of each hierarchy according to the relative positions of the points or the resolutions projected to different grids;
the first level and the second level set a first resolution;
setting a second resolution by the third level and the fourth level;
the first resolution is less than the second resolution;
in the hierarchical structure, for each level of point cloud data, local coordinates with respect to that level are calculated for use in the segmentation stage.
6. The method for automatically partitioning an object based on a substation real-scene point cloud as set forth in claim 5, wherein feature extraction is performed on the point cloud of each level, and point cloud features of different levels are fused to obtain global and comprehensive feature representations, including,
at each level after self-adaptive layering, performing feature extraction on local point cloud data through a Pointernet++ grid structure, wherein the feature extraction comprises multi-layer perception and feature extraction on the local point cloud;
using the full connection layer to conduct category prediction on each point, conducting segmentation and semantic marking on the point cloud according to the prediction result, conducting training and optimization, including,
aiming at a specific point cloud segmentation task, introducing a segmentation head network into the Pointernet++ to output semantic segmentation labels of each point, and adjusting the segmentation head according to task requirements;
in the training phase, point cloud data with known labels are used for supervised learning, parameters of the network are updated through back propagation and optimization algorithms, and the aim is to minimize the loss function between the predicted label and the real label.
7. The method for automatically partitioning objects based on the substation point cloud according to claim 1 or 6, wherein the new point cloud is partitioned by using a trained network to obtain semantic tags of each point, so as to partition, extract and analyze different objects according to the tags, comprising,
in the test stage, the trained Pointnet++ network model is used for dividing the new point cloud data, forward propagation is carried out on each point, and corresponding labels are distributed to the points according to the final dividing result.
8. A system for automatically dividing objects based on a substation scenic spot cloud is characterized by comprising,
the acquisition module is used for acquiring the cloud data of the scenic spots and the corresponding labeling information;
the preprocessing module is used for preprocessing the live-action point cloud data through a self-adaptive layering method, and the preprocessing divides the point cloud data into different levels and arranges an organization mode of points;
the fusion module is used for extracting the characteristics of the point clouds of each level and fusing the characteristics of the point clouds of different levels to obtain global and comprehensive characteristic representations;
the training module is used for carrying out category prediction on each point by using the full-connection layer, carrying out segmentation and semantic marking on the point cloud according to the prediction result, and carrying out training and optimization;
the segmentation module is used for segmenting the new live-action point cloud by utilizing the trained network to obtain semantic labels of each point so as to segment, extract and analyze different objects according to the labels.
9. An electronic device, comprising:
a memory and a processor;
the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, where the computer executable instructions when executed by the processor implement the steps of the substation perspective point cloud-based object automatic segmentation method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing computer-executable instructions which, when executed by a processor, implement the steps of the substation perspective cloud-based object automatic segmentation method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311321733.0A CN117635630A (en) | 2023-10-12 | 2023-10-12 | Automatic object segmentation method and system based on substation scenic spot cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311321733.0A CN117635630A (en) | 2023-10-12 | 2023-10-12 | Automatic object segmentation method and system based on substation scenic spot cloud |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117635630A true CN117635630A (en) | 2024-03-01 |
Family
ID=90020602
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311321733.0A Pending CN117635630A (en) | 2023-10-12 | 2023-10-12 | Automatic object segmentation method and system based on substation scenic spot cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117635630A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117994527A (en) * | 2024-04-03 | 2024-05-07 | 中国空气动力研究与发展中心低速空气动力研究所 | Point cloud segmentation method and system based on region growth |
-
2023
- 2023-10-12 CN CN202311321733.0A patent/CN117635630A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117994527A (en) * | 2024-04-03 | 2024-05-07 | 中国空气动力研究与发展中心低速空气动力研究所 | Point cloud segmentation method and system based on region growth |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sankaranarayanan et al. | Learning from synthetic data: Addressing domain shift for semantic segmentation | |
| US11132551B2 (en) | Systems and methods for skyline prediction for cyber-physical photovoltaic array control | |
| CN111797716A (en) | Single target tracking method based on Siamese network | |
| CN110288586A (en) | A kind of multiple dimensioned transmission line of electricity defect inspection method based on visible images data | |
| CN113240688A (en) | Integrated flood disaster accurate monitoring and early warning method | |
| CN116843845A (en) | Spatial data integration method and system applied to digital twin city | |
| CN109034035A (en) | Pedestrian's recognition methods again based on conspicuousness detection and Fusion Features | |
| CN112085072B (en) | Cross-modal retrieval method of sketch retrieval three-dimensional model based on space-time characteristic information | |
| CN117635630A (en) | Automatic object segmentation method and system based on substation scenic spot cloud | |
| CN110472495B (en) | Deep learning face recognition method based on graphic reasoning global features | |
| Xiao et al. | Building segmentation and modeling from airborne LiDAR data | |
| CN118154996B (en) | Three-dimensional scene point cloud classification method for multi-scale depth feature aggregation | |
| Laupheimer et al. | The importance of radiometric feature quality for semantic mesh segmentation | |
| CN114782357A (en) | Self-adaptive segmentation system and method for transformer substation scene | |
| CN115019163A (en) | City factor identification method based on multi-source big data | |
| CN117691592A (en) | Photovoltaic output prediction method and device, electronic equipment and storage medium | |
| CN116958476A (en) | Building visual modeling method and system based on BIM data | |
| CN116310967A (en) | Chemical plant safety helmet wearing detection method based on improved YOLOv5 | |
| CN116704378A (en) | Homeland mapping data classification method based on self-growing convolution neural network | |
| Wang et al. | Efficient multi-plane extraction from massive 3D points for modeling large-scale urban scenes | |
| Tang et al. | Research on the Network Map Service Technology of Remote Sensing Image Intelligent Conversion Based on Gan Model | |
| CN116740721B (en) | Finger sentence searching method, device, electronic equipment and computer storage medium | |
| CN108764514A (en) | A kind of photovoltaic power generation power prediction method based on concurrent operation | |
| CN118172507B (en) | Digital twinning-based three-dimensional reconstruction method and system for fusion of transformer substation scenes | |
| CN117671181B (en) | Method and system for constructing smart city contracture map based on big data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |