Power equipment three-dimensional point cloud reconstruction method and related device

Abstract

The invention belongs to a point cloud reconstruction method, and provides a power equipment three-dimensional point cloud reconstruction method and a related device for solving the technical problems that the accuracy of reconstructed and restored object form is low and key features are difficult to capture during detail restoration in two common curved surface reconstruction methods at present. And inputting into a three-dimensional point cloud reconstruction model of the power equipment to obtain a reconstruction result. Wherein the calculation method in the three-dimensional point cloud reconstruction model of the power equipment comprises local feature weighting, global feature fusion and occupancy prediction, the local feature weighting is based on the similarity of query point and neighborhood point features, the global feature fusion is based on the global feature weighting, and the occupancy prediction is realized through linear layer decoding. The detail reconstruction precision is effectively improved, and the accuracy of overall reconstruction is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to a point cloud reconstruction method, and particularly relates to a three-dimensional point cloud reconstruction method for power equipment and a related device. BACKGROUND

[0002] Digital three-dimensional models are increasingly applied to the power industry because they can intuitively and vividly express three-dimensional spatial information. When digital three-dimensional models are applied in substations, the variety of equipment and the complexity of the structure put higher requirements on the authenticity and work efficiency of three-dimensional reconstruction.

[0003] To achieve accurate and efficient surface reconstruction, many methods have emerged in the field of three-dimensional reconstruction. Common surface reconstruction methods mainly include two categories: explicit reconstruction and implicit reconstruction. (1) Explicit reconstruction is a method of directly constructing a three-dimensional object surface geometric model. The generated polygon mesh model is convenient for visualization and geometric processing. However, this method has very high requirements for the quality of input point cloud data. When the point cloud has noise, irregular distribution, or is relatively sparse, it is difficult to accurately reconstruct and restore the object shape, and holes and distortions may easily occur. (2) Implicit reconstruction represents the point cloud surface by fitting an implicit function and extracts the implicit function isosurface to reconstruct the surface model. Although this method can quickly outline the object contour, it has deficiencies in object detail restoration and is difficult to accurately capture key features of complex structures. SUMMARY

[0004] The application provides a three-dimensional point cloud reconstruction method for power equipment and a related device to solve the technical problems of low accuracy in reconstructing and restoring object shape and difficulty in capturing key features during detail restoration.

[0005] To achieve the above-mentioned purpose, the application adopts the following technical solutions: In a first aspect, the application provides a three-dimensional point cloud reconstruction method for power equipment, comprising: obtaining power equipment point cloud data; inputting the power equipment point cloud data into a three-dimensional point cloud reconstruction model for power equipment to obtain a reconstruction result; the calculation method in the three-dimensional point cloud reconstruction model for power equipment includes local feature weighting, global feature fusion, and occupancy prediction; wherein the local feature weighting is based on the similarity of query point and neighborhood point features, the global feature fusion is based on global feature weighting, and the occupancy prediction is realized by linear layer decoding.

[0006] Further, after obtaining the power equipment point cloud data, the method further comprises: performing noise removal processing and thinning processing on the power equipment point cloud data.

[0007] Further, the method of local feature weighting comprises: selecting a query point from the power equipment point cloud data; calculating relative position encodings of the query point and the neighbor points; point-by-point converting feature vectors of the query point and the neighbor points through a multi-layer perception, and then obtaining an attention vector according to the feature vector of the query point and the feature vectors of the neighbor points; combining the relative position encodings and the attention vector to calculate a local attention weight; performing weighted summation on the neighbor point features using the local attention weight to obtain updated local features, and completing local feature weighting.

[0008] Further, the method of point-by-point converting feature vectors of the query point and the neighbor points through a multi-layer perception, and then obtaining an attention vector according to the feature vector of the query point and the feature vectors of the neighbor points, comprises:

[0009] wherein, the attention vector is denoted as, the multi-layer perception of two linear layers and a ReLU nonlinear activation function is denoted as, the multi-layer perception is denoted as the multi-layer perception is denoted as the multi-layer perception is denoted as the multi-layer perception is denoted as the query point is denoted as, the neighbor points are denoted as.

[0010] Further, the method of global feature fusion comprises: generating a query vector and a key vector through mapping of features of all points in the power equipment point cloud data; performing weighted summation on the features of all points in the power equipment point cloud data using a global attention weight to obtain global features; fusing the updated local features and the global features in a concatenation manner to obtain a final feature representation, and completing global feature fusion.

[0011] Further, the method of occupancy prediction comprises: decoding the final feature representation into a binary logit regression vector through a linear layer; obtaining an occupancy probability of the query point through a softmax function based on the binary logit regression vector, and completing reconstruction.

[0012] Further, the loss function of the three-dimensional point cloud reconstruction model of the power equipment comprises:

[0013] wherein, the loss function is denoted as, represents the total number of points in the power equipment point cloud data, represents the learnable weight coefficient, represents the real occupancy label of the point i represents the occupancy probability of the power equipment three-dimensional point cloud reconstruction model prediction point i

[0014] In a second aspect, the present application provides a power equipment three-dimensional point cloud reconstruction system, comprising: a data module for obtaining power equipment point cloud data; a reconstruction module for inputting the power equipment point cloud data into a three-dimensional point cloud reconstruction model of the power equipment to obtain a reconstruction result; the calculation method in the three-dimensional point cloud reconstruction model of the power equipment comprises local feature weighting, global feature fusion and occupancy prediction; wherein the local feature weighting is based on the similarity of the query point and the neighborhood point features, the global feature fusion is based on the global feature weighting, and the occupancy prediction is realized by a linear layer decoding.

[0015] In a third aspect, the present application provides an electronic device, comprising a memory and one or more processors; the memory is coupled with the processor; wherein the memory stores computer program code, the computer program code comprises computer instructions, when the computer instructions are executed by the processor, the electronic device executes the steps of the power equipment three-dimensional point cloud reconstruction method.

[0016] In a fourth aspect, the present application provides a computer readable storage medium, characterized in that the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the power equipment three-dimensional point cloud reconstruction method.

[0017] Compared with the prior art, the present application has the following beneficial effects: ​​The application provides a three-dimensional point cloud reconstruction method of a power equipment. After obtaining point cloud data of the power equipment, the point cloud data is input into a three-dimensional point cloud reconstruction model of the power equipment to obtain a reconstruction result. The calculation method in the three-dimensional point cloud reconstruction model of the power equipment comprises local feature weighting, global feature fusion and occupation prediction. The local feature weighting is based on the similarity of query point features and neighborhood point features, the global feature fusion is based on global feature weighting, and the occupation prediction is realized through linear layer decoding. The method changes the fusion method of simple weighting or averaging in the surface reconstruction process, and can focus on key parts such as the curved surface and edge area of the power equipment. In addition, in order to fully capture global structure information, global feature weighting is designed, global features are obtained through a global attention mechanism, local features and global features are fused, the overall structure information of the power equipment is obtained, and the problem of deformity is avoided. The detail reconstruction precision is effectively improved, and the accuracy of the overall reconstruction is ensured.

[0018] The application also provides a three-dimensional point cloud reconstruction system of a power equipment, an electronic device and a computer storage medium, which have all the advantages of the three-dimensional point cloud reconstruction method of the power equipment. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0020] Figure 1 FIG. 1 is a flowchart of a three-dimensional point cloud reconstruction method of a power equipment according to an embodiment of the application; Figure 2 FIG. 2 is another flowchart of a three-dimensional point cloud reconstruction method of a power equipment according to an embodiment of the application; Figure 3 FIG. 3 is a schematic diagram of point cloud data of a substation after preprocessing in an embodiment of the application; Figure 4 FIG. 4 is a structural schematic diagram of a three-dimensional point cloud reconstruction model of a power equipment in an embodiment of the application; Figure 5 FIG. 5 is a flowchart of point cloud weighting in an embodiment of the application; Figure 6 FIG. 6 is a schematic diagram of global feature aggregation in an embodiment of the application; Figure 7 FIG. 7 is a result schematic diagram of a comparison between the method of the application and other three reconstruction models in an embodiment of the application; Figure 8 FIG. 8 is a qualitative analysis comparison diagram of the application and other three models on ShapeNet in an embodiment of the application. Figure 9 A schematic diagram of a three-dimensional point cloud reconstruction system for power equipment. DETAILED DESCRIPTION

[0021] To make the objects, technical solutions, and advantages of the embodiments of the present application clearer, the following will be combined with the accompanying drawings for the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application but not all the embodiments. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0023] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0024] In the description of the embodiments of the present application, it should be noted that if the terms “upper”, “lower”, “horizontal”, “inner”, and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship when the product of the present application is used, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms “first”, “second”, and the like are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0025] In addition, if the term “horizontal” appears, it does not mean that the component must be absolutely horizontal, but can be slightly inclined. For example, “horizontal” only means that its direction is relatively more horizontal than “vertical”, and does not mean that the structure must be completely horizontal, but can be slightly inclined.

[0026] In the description of the embodiments of the application, it should be explained that, unless otherwise explicitly specified and limited, if the terms "set", "install", "connect", "connect" appear, they should be understood in a broad sense, for example, they can be fixedly connected, or can be detachably connected, or integrally connected; can be mechanically connected, or can be electrically connected; can be directly connected, or can be indirectly connected through an intermediate medium, or can be connected inside two elements. For those skilled in the art, the specific meanings of the above terms in the application can be understood according to the specific circumstances.

[0027] There are a large number of free-form surfaces and thin-walled structures in power equipment such as transformer radiators and GIS equipment bellows. When collecting point clouds, local data is easily missing due to metal reflection, shielding between equipment, etc. Explicit reconstruction relies on high-quality point clouds, and incomplete data will directly cause holes or non-manifold structures in the grid model. The material difference of the surface coating of power equipment may also cause a sudden change in reflectivity of laser scanning, resulting in uneven point cloud density and jagged distortion of the curved surface after reconstruction. In addition, substation equipment is usually composed of conductors, insulators, mechanical mechanisms and other components, and explicit reconstruction is difficult to handle complex topological relationships. Furthermore, in the power inspection scene, the unmanned aerial vehicle or handheld device may be affected by electromagnetic interference when collecting point clouds, causing data jitter. The grid optimization algorithm of explicit reconstruction has high computational complexity, which is difficult to meet the demand of rapid modeling on site.

[0028] Therefore, the three-dimensional reconstruction of power equipment needs to break through the limitations of existing methods in complex surface processing, detail preservation, multi-source data fusion, etc., and develop customized reconstruction technology for industry scenarios.

[0029] Based on the above situation, the application provides a three-dimensional point cloud reconstruction method for power equipment and related devices, which will be further described in detail below in combination with embodiments and drawings.

[0030] As Figure 1 shown, it is a flowchart of the three-dimensional point cloud reconstruction method for power equipment of the application, which can include: S101, acquiring power equipment point cloud data.

[0031] Specifically, the three-dimensional point cloud reconstruction method for power equipment in the application can first acquire point cloud data of the power equipment through a laser scanner, a depth camera or other three-dimensional scanning devices. These point cloud data contain a large number of discrete points on the surface of the power equipment, each point having three-dimensional coordinate information, which is used to represent the geometric shape of the power equipment.

[0032] S102, input the point cloud data of the power equipment into the 3D point cloud reconstruction model of the power equipment to obtain the reconstruction result. The calculation methods in the 3D point cloud reconstruction model of the power equipment include: local feature weighting, global feature fusion, and occupancy prediction. Among them, local feature weighting is based on the similarity of features between the query point and its neighboring points, global feature fusion is based on global feature weighting, and occupancy prediction is achieved through linear layer decoding.

[0033] It should be noted that the query point refers to the center point of the local region where features need to be enhanced, such as the edge of an insulator skirt. Neighboring points are points within a radius *r* centered on the query point, such as other points on the same skirt. In local feature weighting, a self-attention mechanism is used to calculate the feature similarity between the query point and its neighboring points. Features of neighboring points with high similarity are assigned higher weights, strengthening key local features. In occupancy prediction, a multilayer perceptron maps the fused features to occupancy probabilities. Binarization is used to determine whether a voxel is occupied by a device, achieving the conversion from point cloud to 3D mesh.

[0034] This application introduces local and global self-attention mechanisms on the basis of implicit reconstruction models, which enhances the model's ability to express local features while reconstructing a complete surface. The model then interpolates neighboring points using inferred self-attention weights, enabling it to accurately focus on key complex areas and effectively improve the detail and realism of the reconstructed surface.

[0035] like Figure 2 The diagram shown is another flowchart illustrating the three-dimensional point cloud reconstruction method for power equipment according to this application, which may include: S201, Data Preparation.

[0036] LiDAR is used to collect data on substation power equipment for training and testing. In this embodiment, a 220kV substation in northern my country is used as the data collection target, and LiDAR is used to collect data on the substation power equipment for training and testing. LiDAR generates high-precision three-dimensional point cloud data by emitting a laser beam and measuring the time difference of the reflected signal.

[0037] Due to the high noise level in substation point cloud data, preprocessing is performed first. Specifically, to reduce the impact of noise, voxel filtering can be used to denoise the point cloud data and improve data quality. Voxel filtering is a spatial partitioning-based point cloud downsampling and denoising method. It divides the point cloud into a uniform three-dimensional voxel grid and replaces the original points with representative points within each voxel, achieving data compression and noise suppression. Afterward, the denoised point cloud data is thinned to reduce the number of data points while preserving the basic shape characteristics of the insulator, providing data support for subsequent model training. Point cloud thinning is a further process of reducing data volume based on denoising, aiming to balance computational efficiency and data accuracy. Its core principle is to reduce redundancy while maintaining key point cloud information through spatial sampling strategies or feature preservation mechanisms. Figure 3 The image shows a schematic diagram of the preprocessed substation cloud data.

[0038] S202, Construct a 3D point cloud reconstruction model of the power equipment.

[0039] This application fully leverages the advantages of point convolution in surface reconstruction, improving the attention-based weighted neural network structure by introducing a self-attention mechanism. This enhances the model's ability to represent local features while achieving complete surface reconstruction. It effectively solves the problem that existing surface reconstruction methods struggle to capture features of key areas such as surfaces and edge regions.

[0040] (1) Three-dimensional point cloud reconstruction model architecture of power equipment.

[0041] In this application, the three-dimensional point cloud reconstruction model of the power equipment is used for surface reconstruction of the power equipment, such as... Figure 4 The diagram shows the structure of a 3D point cloud reconstruction model of power equipment. The 3D point cloud reconstruction model of power equipment mainly consists of three parts: a local feature weighting module, a global feature fusion module, and an occupancy prediction module. Specifically, the local feature weighting module first bases its calculations on the similarity of features between the query point and its neighboring points, combined with relative position encoding. To calculate attention weights Then, by weighted summation of the features of neighboring points, the local features of the query point are obtained. The global feature fusion module first obtains global features using a global feature weighting mechanism, and then fuses the local and global features by concatenation to obtain the final feature representation, thereby enhancing the global perception capability of the features. The occupancy prediction module decodes the fused features into occupancy probabilities through a linear layer, which are used to achieve surface reconstruction of the target point cloud. The specific implementation steps are as follows: S1, local feature weighting.

[0042] like Figure 5The flowchart shown is a point cloud weighted flowchart. Different weighting coefficients are obtained through self-attention mechanisms, so that more attention can be paid to key parts such as curved surfaces and edge regions during reconstruction.

[0043] S1.1, select m query points from the original point cloud data using farthest point sampling .

[0044] S1.2, calculate trainable position encoding, as shown in formula (1), using encoding function Two linear layers and a ReLU nonlinear activation function multilayer perceptron) to calculate the relative position encoding of the query point and the neighborhood point . (1) In the formula, and are the three-dimensional coordinates of the query point i and the domain point j .

[0045] S1.3, get attention vector . As shown in formula (2), first convert the feature vectors of the query point and the neighborhood point point by point through a multilayer perceptron and a multilayer perceptron. Then, a multilayer perceptron with two linear layers and a ReLU nonlinear activation function is used to obtain the attention vector.

[0046] (2) Where, represents a multilayer perceptron with two linear layers and a ReLU nonlinear activation function.

[0047] S1.4, combine the relative position encoding and the attention vector to calculate the local attention weight . As shown in formula (3), the local attention weight is obtained by a normalization function softmax.

[0048] (3) Where, represents the normalization function softmax function.

[0049] S1.5, as shown in formula 4, use the local attention weight to perform weighted summation on the neighborhood point feature to obtain the updated local feature . Wherein, the subset is A set of points within a local neighborhood.

[0050] (4) S2, Global Feature Aggregation.

[0051] like Figure 6 The diagram illustrates global feature aggregation. Under the global attention mechanism, features from all points are globally aggregated to capture the relationships between long-distance points. The specific steps are as follows: S2.1, as shown in formula (5), calculate the global attention weights and generate query vectors and key vectors by mapping the features of all points.

[0052] (5) in, It is a query vector. It is a key vector.

[0053] S2.2, using global attention weights on the features of all points. We perform a weighted summation to obtain the global features. : (6) in, N It is the total number of points in the power equipment point cloud data, and the features of all points. Weighted summation.

[0054] S2.3, as shown in Formula 7, merges local and global features by concatenation to obtain the final feature representation. : (7) S3, Occupancy Prediction.

[0055] S3.1, Obtain the regression vector, as shown in formula (8), and represent the final fused features through a linear layer D. Decoded into a binary logit regression vector The binary logit regression vector Used to classify position q as occupied or unoccupied.

[0056] (8) S3.2, based on the regression vector The location is classified as occupied or unoccupied, and then converted into an occupancy probability using the softmax function.

[0057] S4, loss function.

[0058] S4.1, the embodiment improves the loss function, as shown in equation (9), increases the weight coefficient, so that it can pay more attention to the key parts of the power equipment: (9) wherein, is the true occupancy label (0 represents unoccupied, and 1 represents occupied) of the point i is the occupancy probability of the three-dimensional point cloud reconstruction model prediction point of the power equipment, i is a learnable weight coefficient, is the loss function.

[0059] S203, training and reasoning of the three-dimensional point cloud reconstruction model of the power equipment.

[0060] The three-dimensional point cloud reconstruction model of the power equipment constructed in the application is trained and tested, and different power equipment is reconstructed. In the implementation of the embodiment, an Intel Xeon Platinum 8352V CPU and a NVIDIA GeForce RTX3080 GPU are used on a 64-bit server, and Ubuntu 20.04.3 LTS is used as the operating system to ensure stability and compatibility. In the model training process, the Adam optimizer can be used, the learning rate is set to 0.001, and the batch size is flexibly adjusted according to the number of input points. When the number of input points is 3000, the batch size is 16, and when the number of input points is 10000, the batch size is 8. The model training and parameter optimization are completed through 300,000 iterations.

[0061] The local feature weighting module first partitions the point cloud data by farthest point sampling, and then calculates the weights of the neighbor points using the self-attention mechanism to generate enhanced local feature representation. This process not only accurately captures the detailed features of the local geometric structure, but also effectively suppresses the interference of noise points and non-key neighbors, significantly improving the authenticity of the model for complex curved surfaces. The global feature fusion module includes a feature extraction layer and a fusion layer. In the feature extraction layer, the features of all points are globally aggregated to capture the relationship between long-distance points. The feature fusion layer fuses the local features and global features extracted before in a splicing manner as the final feature representation. Finally, the fused features are input into the occupancy prediction module for surface reconstruction of the target point cloud.

[0062] In order to verify the technical effects of the application, the power equipment three-dimensional point cloud reconstruction method of the application and other three reconstruction models are used for reconstruction, as shown in Figure 7 , which is a result diagram for comparing the method of the application with other three reconstruction models. Figure 7 ​In the table, (A) before behavior reconstruction, (B) after behavior reconstruction, (1) the SPR (Surface Plasmon Resonance) model is used, (2) the ConvONet (Convolutional Occupancy Network) model is used, (3) the poco (Policy Composition) model is used, and (4) the method of the application is used. It can be seen from the analysis that the method of the application can better reconstruct the detailed structure.

[0063] As shown in the table, (1) the SPR (Surface Plasmon Resonance) model is used, (2) the ConvONet (Convolutional Occupancy Network) model is used, (3) the poco (Policy Composition) model is used, and (4) the method of the application is used. It can be seen from the analysis that the method of the application can better reconstruct the detailed structure. Figure 8 As shown in the table, (1) the SPR (Surface Plasmon Resonance) model is used, (2) the ConvONet (Convolutional Occupancy Network) model is used, (3) the poco (Policy Composition) model is used, and (4) the method of the application is used. It can be seen from the analysis that the method of the application can better reconstruct the detailed structure.

[0064] As shown in the table, (1) the SPR (Surface Plasmon Resonance) model is used, (2) the ConvONet (Convolutional Occupancy Network) model is used, (3) the poco (Policy Composition) model is used, and (4) the method of the application is used. It can be seen from the analysis that the method of the application can better reconstruct the detailed structure.

[0065] Table 1 Comparison results of the application with three other models in ShapeNet

[0066] As can be seen from Table 1, the application can not only effectively suppress the interference of noise of the substation point cloud data, but also improve the reality of surface reconstruction and reduce the problems of holes and distortion.

[0067] In the processing of traditional three-dimensional reconstruction methods, a large amount of calculation and storage operations need to be performed on the three-dimensional space. With the increase of model complexity, the amount of calculation increases exponentially, which not only leads to a significant increase in calculation time, but also puts high requirements on the performance of computer hardware. When facing objects with complex geometric structures and fine textures, traditional three-dimensional reconstruction techniques often have difficulty in accurately capturing and restoring their detailed features. For example, small holes, sharp edges, and complex curved surfaces on the surface of an object are prone to detail distortion or information loss during the reconstruction process. The present application changes the method of using simple weighting or averaging for fusion in the curved surface reconstruction process, uses a trainable position encoding module and a point cloud weighting module to obtain different weighting coefficients, so that the reconstruction process can focus more on the key parts such as the curved surface and edge area of the power equipment. In addition, in order to fully capture the global structural information, a global feature weighting is designed, the global feature is obtained through the global attention mechanism, and the local feature and the global feature are fused to obtain the overall structural information of the power equipment to avoid the problem of deformity. Finally, in order to adapt to the complex structure of the power equipment, a weight coefficient is added to the loss function, so that it can pay more attention to the key areas.

[0068] The specific advantages mainly manifest in the following three aspects: (1) Improve the accuracy of detail reconstruction. The present application introduces a trainable position encoding module and a point cloud weighting module, which accurately focus on the key parts such as the curved surface and edge of the power equipment by calculating the relative position encoding and feature similarity weight between points. When dealing with structures such as insulators and bushings, it can effectively avoid the problems of blurred and missing details.

[0069] (2) Ensure the accuracy of the overall structure. The present application aggregates all point features through the global attention mechanism, establishes long-distance point correlation, and deeply fuses with local features. It solves the problem that traditional reconstruction techniques lack effective global information capture mechanisms, which can easily cause the overall structure of the model to be deformed and the scale to be out of proportion.

[0070] (3) Accelerate model training and convergence. The present application introduces a dynamic weight coefficient in the loss function, which gives higher weight to the prediction error of key areas such as curved surfaces and edges, guiding the model to prioritize optimizing the reconstruction effect of core parts. According to the actual measurement, under the same training data set, the convergence speed of the three-dimensional point cloud reconstruction model of the power equipment of the present application can be effectively improved by 1.2 times.

[0071] As shown in Figure 9 , it is a schematic diagram of the three-dimensional point cloud reconstruction system of the power equipment of the present application, which can include: a data module for obtaining power equipment point cloud data; The reconstruction module is configured to input the power equipment point cloud data into a three-dimensional point cloud reconstruction model of the power equipment to obtain a reconstruction result. The calculation method in the three-dimensional point cloud reconstruction model of the power equipment includes local feature weighting, global feature fusion, and occupancy prediction. The local feature weighting is based on the similarity of the query point and the neighborhood point features. The global feature fusion is based on global feature weighting. The occupancy prediction is realized by linear layer decoding.

[0072] It should be noted that in several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of each module is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components can be or can not be physically separated, and the components displayed as modules can be a physical unit or multiple physical units, that is, can be located in one place, or can be distributed to multiple different places. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment scheme.

[0073] In addition, each module in each embodiment of the present application can be integrated in one processing unit, or each module can be physically present alone, or two or more modules can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0074] The embodiment of the present application also provides an electronic device, which can include one or more processors, memories and communication interfaces.

[0075] The memory, the communication interface and the processor are coupled. For example, the memory, the communication interface and the processor can be coupled together through a bus.

[0076] The communication interface is configured to perform data transmission with other devices. The memory stores computer program code. The computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the steps of the power equipment three-dimensional point cloud reconstruction method.

[0077] The processor can be a processor or a controller, for example, can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, transistor logic device, hardware component or any combination thereof. It can implement or execute the various exemplary logical blocks, modules and circuits described in combination with the present disclosure. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The processor can be used to support the electronic device to perform the method steps provided in the above embodiments.

[0078] The bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0079] The computer readable storage medium provided by the embodiments of the present application stores a computer program, and the computer program is executed by the processor to implement the steps of the power equipment three-dimensional point cloud reconstruction method.

[0080] The computer readable storage medium involved in the present application includes random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD ROM, or any other form of storage medium known in the technical field.

[0081] The above is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for three-dimensional point cloud reconstruction of power equipment, characterized in that, The method comprises the following steps: acquiring power equipment point cloud data; inputting the power equipment point cloud data into a three-dimensional point cloud reconstruction model of the power equipment to obtain a reconstruction result; the calculation method in the three-dimensional point cloud reconstruction model of the power equipment comprises local feature weighting, global feature fusion and occupancy prediction; wherein the local feature weighting is based on the similarity of the features of a query point and neighbor points, the global feature fusion is based on global feature weighting, and the occupancy prediction is realized by linear layer decoding.

2. The method of claim 1, wherein, After the power equipment point cloud data is acquired, the method further comprises the following steps: performing noise removal processing and thinning processing on the power equipment point cloud data. 3.The method of claim 1, wherein, The method of local feature weighting comprises the following steps: selecting a query point from the power equipment point cloud data; calculating the relative position encoding of the query point and neighbor points; point-by-point converting the feature vectors of the query point and the neighbor points by using a multi-layer perception machine, and then obtaining an attention vector according to the feature vector of the query point and the feature vector of the neighbor points; combining the relative position encoding and the attention vector to calculate a local attention weight; performing weighted summation on the neighbor point features by using the local attention weight to obtain updated local features, and completing the local feature weighting.

4. The method of claim 3, wherein, The method of point-by-point converting the feature vectors of the query point and the neighbor points by using a multi-layer perception machine, and then obtaining an attention vector according to the feature vector of the query point and the feature vector of the neighbor points comprises the following steps: wherein, denotes an attention vector, denotes a multi-layer perceptron of two linear layers and one ReLU non-linear activation function, denotes a multi-layer perceptron, denotes a multi-layer perceptron, denotes a query point, denotes a neighborhood point.

5. The method of claim 1, wherein, The method of global feature fusion comprises the following steps: generating a query vector and a key vector by mapping the features of all points in the power equipment point cloud data; performing weighted summation on the features of all points in the power equipment point cloud data by using a global attention weight to obtain global features; combining the updated local features and the global features in a splicing manner to obtain a final feature representation, and completing the global feature fusion.

6. The method of claim 1, wherein, The method of occupancy prediction comprises the following steps: decoding the final feature representation into a binary logit regression vector by using a linear layer; obtaining the occupancy probability of the query point by using a softmax function based on the binary logit regression vector, and completing the reconstruction.

7. The method of claim 1, wherein, The loss function of the three-dimensional point cloud reconstruction model of the power equipment comprises the following steps: wherein, represents a loss function, represents the total number of points in the power equipment point cloud data, represents a learnable weight coefficient, represents a point i of a real occupancy label, represents an occupancy probability of a three-dimensional point cloud reconstruction model predicted point i of the power equipment.

8. A power equipment three-dimensional point cloud reconstruction system, characterized in that, The method comprises the following steps: a data module for acquiring power equipment point cloud data; a reconstruction module for inputting the power equipment point cloud data into a three-dimensional point cloud reconstruction model of the power equipment to obtain a reconstruction result; the calculation method in the three-dimensional point cloud reconstruction model of the power equipment comprises local feature weighting, global feature fusion and occupancy prediction; wherein the local feature weighting is based on the similarity of the features of a query point and neighbor points, the global feature fusion is based on global feature weighting, and the occupancy prediction is realized by linear layer decoding.

9. An electronic device, comprising: The method comprises the following steps: a memory and one or more processors; the memory is coupled with the processor; wherein the memory stores computer program code, the computer program code comprises computer instructions, when the computer instructions are executed by the processor, the electronic device executes the steps of the power equipment three-dimensional point cloud reconstruction method in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer program stored in the computer readable storage medium is executed by the processor to realize the steps of the power equipment three-dimensional point cloud reconstruction method in any one of claims 1-8.

Images