A point cloud classification neural network optimization method and device based on dimension transformation, an electronic device, and a medium
By embedding a dimensionality transformation module into a point cloud classification neural network, and utilizing high-dimensional projection and skip connections to optimize point cloud features, the problem of lost spatial location information is solved, thereby improving the accuracy and efficiency of point cloud classification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU FRONTIER ELECTRIC TECH
- Filing Date
- 2023-06-01
- Publication Date
- 2026-05-01
AI Technical Summary
Existing point cloud classification methods based on deep neural networks often lose spatial location information during the computation process, affecting classification accuracy.
By embedding a dimensionality transformation module into a point cloud classification neural network, the point cloud features are nonlinearly transformed to a high-dimensional space through high-dimensional projection. Combined with skip connections and feature recombination, linear separability is improved.
It improves the accuracy of point cloud classification, reduces computational complexity, and solves the problem of lost spatial location information, making it suitable for various deep learning point cloud classification neural networks.
Smart Images
Figure CN116612329B_ABST
Abstract
Description
A method, apparatus, electronic device, and medium for optimizing point cloud classification neural networks based on dimensionality transformation. Technical Field
[0001] This invention relates to the field of 3D point cloud classification technology, specifically to a point cloud classification neural network optimization method, device, electronic device, and medium based on dimensional transformation. Background Technology
[0002] 3D point cloud processing has numerous applications in automation, autonomous driving, and AR, with 3D point cloud classification being a crucial part of this process. With the rise of deep learning, point cloud classification methods based on deep neural networks have become mainstream due to their significantly higher accuracy compared to traditional methods. Representative methods include PointNet++ (point cloud classification and segmentation network), PointCNN (dynamic graph convolutional neural network), and SpiderCNN (anisotropic separable set abstraction network). Since these methods are largely derived from corresponding computer vision and natural language processing methods, their main structures include convolution and clustering commonly used in computer vision, as well as self-attention mechanisms from natural language processing. These general structures do not consider the spatial characteristics of 3D point clouds themselves. Therefore, existing deep neural network-based point cloud classification methods often suffer from a loss of spatial location information during computation, affecting the final classification accuracy. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device, electronic device and medium for optimizing point cloud classification neural networks based on dimensional transformation. It fully considers the spatial structural characteristics of point clouds, improves the linear separability of point clouds through high-dimensional projection, solves the problem of spatial location information loss during the operation, and further improves the classification accuracy of the network.
[0004] To achieve the above objectives, the present invention is implemented using the following technical solution:
[0005] In a first aspect, the present invention provides a point cloud classification neural network optimization method based on dimensionality transformation. The point cloud classification neural network includes a feature extraction module and a classification module. The feature extraction module consists of multiple sets of neural network layers that appear in the same order. Each set of repeating network layers constitutes a sub-module.
[0006] The optimization method includes:
[0007] Obtain the point cloud classification neural network to be optimized;
[0008] In the feature extraction module of the point cloud classification neural network, a pre-built dimension transformation module is embedded between each sub-module; wherein, the dimension transformation module is used to non-linearly transform the point cloud features to be classified into a high-dimensional space through high-dimensional projection, so as to improve the linear separability of the point cloud features to be classified.
[0009] The network parameters of the point cloud classification neural network with the embedded dimension change module are trained using the dataset to obtain an optimized point cloud classification neural network.
[0010] In conjunction with the first aspect, preferably, the step of nonlinearly transforming the point cloud features to be classified into a high-dimensional space through high-dimensional projection includes:
[0011] Obtain the point cloud features to be classified;
[0012] The acquired point cloud features are linearly reconstructed to obtain linearly reconstructed features;
[0013] The linear recombination features are grouped to obtain multiple sets of features;
[0014] Each set of features is subjected to nonlinear transformation to obtain a high-dimensional feature set;
[0015] The high-dimensional feature set and the acquired point cloud features are concatenated in the feature space to obtain point cloud features with high-dimensional features.
[0016] In conjunction with the first aspect, preferably, the step of linearly reconstructing the acquired point cloud features includes:
[0017] Based on the acquired point cloud features X, the feature dimensions of the linearly reconstructed features Y are determined using a preset scaling factor S:
[0018]
[0019] In the formula, c Y c is the feature dimension of the linear recombination feature Y; X Let X be the feature dimension of the point cloud feature, and X be the feature dimension of the point cloud feature. x i This represents the i-th element in the point cloud feature X, where i ∈ [1, c]. X ]; Indicates rounding down;
[0020] Based on the feature dimension c of linear recombination feature Y Y The acquired point cloud features X are linearly reconstructed using a fully connected layer to obtain the linearly reconstructed features Y, calculated as follows:
[0021]
[0022]
[0023] In the formula, w ij Indicates the parameters of the fully connected layer; y j Let j represent the j-th element in the linear recombination feature Y, where j∈[1,c]. Y ].
[0024] In conjunction with the first aspect, preferably, the step of grouping the linear recombination features includes:
[0025] The linear recombination feature Y is grouped according to a preset grouping number N, and the calculation formula is as follows:
[0026]
[0027] G = {g1, g2, ... g} k …,g N}
[0028] In the formula, The first element in the linear recombination feature Y represents the... 1 element, d k =(k-1)d, g k Let G represent the k-th feature group after grouping; G represents the set of N features.
[0029] In conjunction with the first aspect, the preferred steps for performing nonlinear transformation processing on each group of features include:
[0030] Each group of features g k The products obtained by multiplying each pair of elements in the set are used as the new feature set g′ after nonlinear processing. k The element is calculated using the following formula:
[0031] g′ k ={y p y q |p∈[1,N],q∈[1,N]}
[0032] In the formula, y p y q Represents the k-th feature group, feature g. k Multiply each p-th element by the q-th element.
[0033] Based on each new feature group g ′ k To obtain the high-dimensional feature set G ′ As shown below:
[0034] G ′ ={g ′ 1g ′ 2,…,g ′ k …,g ′ N}
[0035] In conjunction with the first aspect, preferably, the step of concatenating the high-dimensional feature set with the acquired point cloud features in the feature space includes:
[0036] The high-dimensional feature set G is expressed by the following formula. ′ Perform the expand operation:
[0037] H={z|z∈g ′ k ,k∈[1,N]}
[0038] In the formula, H represents the expanded feature;
[0039] The feature H and the acquired point cloud feature X are concatenated in the feature space through skip connections to obtain the point cloud feature W with high-dimensional features:
[0040] W = H∪X.
[0041] In conjunction with the first aspect, preferably, the step of training a point cloud classification neural network with embedded dimensionality variation modules using a dataset to optimize network parameters includes:
[0042] Obtain the ModelNet40 dataset and divide it into training samples and test samples;
[0043] The training samples are input into the point cloud classification neural network embedded with the dimension change module for training to obtain the network's prediction results;
[0044] The cross-entropy loss function is used to calculate the distance between the network's prediction results and the training samples;
[0045] The network is repeatedly trained based on the calculated distance until a preset number of training iterations or the loss function converges, at which point training stops, resulting in the final optimized point cloud classification neural network.
[0046] Secondly, the present invention provides a point cloud classification neural network optimization device based on dimensionality transformation. The point cloud classification neural network includes a feature extraction module and a classification module. The feature extraction module consists of multiple sets of neural network layers appearing in the same order, and each set of repeatedly appearing network layers constitutes a sub-module. The device includes:
[0047] The acquisition module is used to acquire the point cloud classification neural network to be optimized;
[0048] An embedding module is used to embed pre-built dimension transformation modules between each sub-module in the feature extraction module of the point cloud classification neural network; wherein, the dimension transformation module is used to non-linearly transform the point cloud features to be classified into a high-dimensional space through high-dimensional projection, so as to improve the linear separability of the point cloud features to be classified.
[0049] The training optimization module is used to train the point cloud classification neural network with embedded dimension change module using the dataset to optimize the network parameters, so as to obtain the optimized point cloud classification neural network.
[0050] Thirdly, the present invention provides an electronic device, including a processor and a storage medium;
[0051] The storage medium is used to store instructions;
[0052] The processor is configured to operate according to the instructions to perform the steps of the point cloud classification neural network optimization method based on dimensional transformation as described in any of the first aspects.
[0053] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the point cloud classification neural network optimization method based on dimensional transformation as described in any of the first aspects.
[0054] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:
[0055] This invention improves the classification accuracy of a point cloud classification neural network by embedding a dimensionality transformation module. The dimensionality transformation module non-linearly transforms the point cloud features to be classified into a higher-dimensional space through high-dimensional projection, improving the linear separability of the features and thus reducing classification difficulty and increasing accuracy. Furthermore, this invention does not rely on any specific point cloud classification neural network; the dimensionality transformation module can be easily embedded and function normally in various deep learning-based point cloud classification neural networks.
[0056] The dimension transformation module designed in this invention also introduces skip connections and uses recombination and grouping methods to control the dimension of the transformed features, reducing the computational complexity of feature transformation and limiting the growth of the overall computational overhead of the neural network. It effectively solves the problem of spatial location information loss during the operation and further improves the classification accuracy of the network. Attached Figure Description
[0057] Figure 1 is a schematic diagram of the point cloud classification neural network optimization method based on dimension transformation provided in an embodiment of the present invention;
[0058] Figure 2 is a schematic diagram of the process by which the dimension change module improves the linear separability of the features of the classified point cloud provided in the embodiment of the present invention;
[0059] Figure 3 is a schematic diagram of the network structure before and after the embedded dimension transformation module provided in the embodiment of the present invention;
[0060] Figure 4 is a block diagram of the structural principle of the point cloud classification neural network optimization device based on dimension transformation provided in an embodiment of the present invention. Detailed Implementation
[0061] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.
[0062] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0063] Example 1:
[0064] As shown in Figure 1, this embodiment of the invention introduces a point cloud classification neural network optimization method based on dimensionality transformation. Referring to Figure 3, the point cloud classification neural network includes a feature extraction module and a classification module. The feature extraction module consists of multiple sets of neural network layers appearing in the same order, and the set of each set of repeating network layers constitutes a sub-module. The point cloud classification neural network optimization method based on dimensionality transformation provided in this embodiment specifically includes the following steps:
[0065] Step 1: Obtain the point cloud classification neural network to be optimized;
[0066] Step 2: A pre-built dimension transformation module is embedded between each sub-module in the feature extraction module of the point cloud classification neural network; wherein, the dimension transformation module is used to non-linearly transform the point cloud features to be classified into a high-dimensional space through high-dimensional projection, so as to improve the linear separability of the point cloud features to be classified.
[0067] Figure 3 shows a schematic diagram of the structure of the dimension change module embedded in the point cloud classification neural network.
[0068] Step 3: Use the dataset to train the point cloud classification neural network with the embedded dimension change module to optimize the network parameters, so as to obtain the optimized point cloud classification neural network.
[0069] As an embodiment of the present invention, referring to Figure 2, step 2, which involves nonlinearly transforming the classified point cloud features to a high-dimensional space using high-dimensional projection, includes:
[0070] Step a: Obtain the point cloud features to be classified;
[0071] Specifically, the point cloud features obtained are the point cloud features output by the submodule connected to the module that changes in this dimension.
[0072] Step b: Perform linear reconstruction on the acquired point cloud features to obtain linearly reconstructed features;
[0073] As an embodiment of the present invention, step b specifically includes:
[0074] Based on the acquired point cloud features X, the feature dimensions of the linearly reconstructed features Y are determined using a preset scaling factor S:
[0075]
[0076] In the formula, c Y c is the feature dimension of the linear recombination feature Y; X Let X be the feature dimension of the point cloud feature, and X be the feature dimension of the point cloud feature. x i This represents the i-th element in the point cloud feature X, where i ∈ [1, c]. X ]; This indicates rounding down; further explanation is that the preset scaling factor S is a coefficient greater than 1, determined based on requirements and device performance.
[0077] Based on the feature dimension c of linear recombination feature Y Y The acquired point cloud features X are linearly reconstructed using a fully connected layer to obtain the linearly reconstructed features Y, calculated as follows:
[0078]
[0079]
[0080] In the formula, w ij Indicates the parameters of the fully connected layer; y j Let j represent the j-th element in the linear recombination feature Y, where j∈[1,c]. Y ].
[0081] Step c: Group the linear recombination features to obtain multiple sets of features;
[0082] Furthermore, the linear recombination feature Y is grouped according to a preset grouping number N, and the calculation formula is as follows:
[0083]
[0084] G = {g1, g2, ... g} k …,g N}
[0085] In the formula, The first element in the linear recombination feature Y represents the... 1 element, d k =(k-1)d, g k Let G represent the k-th feature group after grouping; G represents the set of N feature groups. Furthermore, the preset number of groups N should be divisible by c. Y .
[0086] Step d: Perform nonlinear transformation on each group of features to obtain a high-dimensional feature set;
[0087] As an embodiment of the present invention, the nonlinear transformation process includes:
[0088] Each group of features g k The products obtained by multiplying each pair of elements in the set are used as the new feature set g after nonlinear processing. ′ k The element is calculated using the following formula:
[0089] g ′ k ={y p y q |p∈[1,N],q∈[1,N]}
[0090] In the formula, y p y q Represents the k-th feature group, feature g. k Multiply each p-th element by the q-th element.
[0091] Based on each new feature group g ′ k To obtain the high-dimensional feature set G ′ As shown below:
[0092] G ′ ={g ′ 1g ′ 2,…,g ′ k …,g ′ N}
[0093] Step e: Concatenate the high-dimensional feature set with the acquired point cloud features in the feature space to obtain point cloud features with high-dimensional features.
[0094] As one embodiment of the present invention, the splicing process includes:
[0095] The high-dimensional feature set G is expressed by the following formula. ′ Perform the expand operation:
[0096] H={z|z∈g ′ k ,k∈[1,N]}
[0097] In the formula, H represents the expanded feature;
[0098] The feature H and the acquired point cloud feature X are concatenated in the feature space through skip connections to obtain the point cloud feature W with high-dimensional features:
[0099] W = H∪X.
[0100] It should be further explained that after the dimensionality transformation module nonlinearly transforms the point cloud features to be classified into a high-dimensional space, the feature dimension of feature H is calculated. Adjust the input parameters of the submodule connected to the module that changes this dimension, and modify the dimension of its input features to c. H .
[0101] As an embodiment of the present invention, step 3 of the present invention provides the following step: optimizing the network using the AdamW optimizer. The step of training the point cloud classification neural network with embedded dimension change modules using a dataset to optimize the network parameters includes:
[0102] Obtain the ModelNet40 dataset and divide it into training samples and test samples;
[0103] The training samples are input into the point cloud classification neural network embedded with the dimension change module for training to obtain the network's prediction results;
[0104] The cross-entropy loss function is used to calculate the distance between the network's prediction results and the training samples;
[0105] The network is repeatedly trained based on the calculated distance until a preset number of training iterations or the loss function converges, at which point training stops, resulting in the final optimized point cloud classification neural network.
[0106] Furthermore, in this embodiment of the invention, different classification neural networks were used to classify point cloud features on the ModelNet40 dataset to obtain the classification accuracy of each method. The comparison results are shown in Table 1.
[0107] Table 1
[0108] Overall accuracy of the method (%): PointNet++ 91.9, PointCNN 92.2, SpiderCNN 92.4, Method of this invention (PN++) 92.9 surface
[0109] As can be seen from Table 1, the point cloud classification neural network obtained by optimizing PointNet++ using the method of this invention significantly outperforms the original PointNet++ network in terms of overall accuracy of point cloud feature classification, and surpasses several other existing methods. Therefore, compared with the prior art, this invention effectively utilizes the spatial features of point clouds and improves the accuracy of point cloud classification.
[0110] Example 2:
[0111] As shown in Figure 4, this embodiment of the invention provides a point cloud classification neural network optimization device based on dimensionality transformation, which can be used to implement the method described in Embodiment 1. The device includes:
[0112] The acquisition module is used to acquire the point cloud classification neural network to be optimized;
[0113] An embedding module is used to embed pre-built dimension transformation modules between each sub-module in the feature extraction module of the point cloud classification neural network; wherein, the dimension transformation module is used to non-linearly transform the point cloud features to be classified into a high-dimensional space through high-dimensional projection, so as to improve the linear separability of the point cloud features to be classified.
[0114] The training optimization module is used to train the point cloud classification neural network with embedded dimension change module using the dataset to optimize the network parameters, so as to obtain the optimized point cloud classification neural network.
[0115] The point cloud classification neural network optimization device based on dimension transformation provided in this embodiment of the invention is based on the same technical concept as the point cloud classification neural network optimization method based on dimension transformation provided in Embodiment 1, and can produce the beneficial effects described in Embodiment 1. For the contents not described in detail in this embodiment, please refer to Embodiment 1.
[0116] Example 3:
[0117] This invention provides an electronic device, including a processor and a storage medium;
[0118] The storage medium is used to store instructions;
[0119] The processor is configured to operate according to instructions to perform steps according to any of the methods in Embodiment 1.
[0120] Example 4:
[0121] This invention provides a computer-readable storage medium storing a computer program thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of any of the methods in Embodiment 1.
[0122] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0123] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.
[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.
[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.
[0126] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A point cloud classification neural network optimization method based on dimensionality transformation, characterized in that, The point cloud classification neural network includes a feature extraction module and a classification module. The feature extraction module consists of multiple sets of neural network layers appearing in the same order, with each set of repeating network layers forming a sub-module. The optimization method includes: obtaining the point cloud classification neural network to be optimized; embedding pre-constructed dimension transformation modules between each sub-module in the feature extraction module of the point cloud classification neural network; wherein, the dimension transformation module is used to non-linearly transform the point cloud features to be classified into a high-dimensional space through high-dimensional projection to improve the linear separability of the point cloud features to be classified; the high-dimensional projection method... The steps of nonlinearly transforming the point cloud features to be classified into a high-dimensional space include: acquiring the point cloud features to be classified; linearly recombining the acquired point cloud features to obtain linearly recombined features; grouping the linearly recombined features to obtain multiple sets of features; performing nonlinear transformation processing on each set of features to obtain a high-dimensional feature set; concatenating the high-dimensional feature set with the acquired point cloud features in the feature space to obtain point cloud features with high-dimensional features; and training the point cloud classification neural network with embedded dimension transformation modules using the dataset to optimize the network parameters to obtain an optimized point cloud classification neural network.
2. The point cloud classification neural network optimization method based on dimensionality transformation according to claim 1, characterized in that, The steps for linearly reconstructing the acquired point cloud features include: based on the acquired point cloud features... Combined with the preset scaling factor Determine the characteristics of linear recombination Feature dimensions: In the formula, linear recombination characteristics Feature dimensions; Point cloud features Feature dimensions, point cloud features ， Representing point cloud features The first in One element, Indicates rounding down; based on linear recombination features Feature Dimensions The acquired point cloud features are processed using fully connected layers. Perform linear recombination to obtain linear recombination characteristics. The calculation formula is as follows: ， In the formula, Indicates the parameters of the fully connected layer; Represents linear recombination characteristics The first in One element, 。 3. The point cloud classification neural network optimization method based on dimensionality transformation according to claim 2, characterized in that, The step of grouping the linear recombination features includes: according to a preset number of groups. For the linear recombination feature The formula for grouping is as follows: ， In the formula, Represents linear recombination characteristics The first in One element, ， ， ； Indicates the first group after grouping Group features; express A set of group features.
4. The point cloud classification neural network optimization method based on dimensionality transformation according to claim 3, characterized in that, The steps of performing nonlinear transformation on each group of features include: transforming each group of features... The products obtained by multiplying each pair of elements in the set are used as new feature sets after nonlinear processing. The element is calculated using the following formula: In the formula, Indicates the first Group features The first in Each element and the first Multiply each element by its corresponding new feature set. To obtain a high-dimensional feature set As shown below: 。 5. The point cloud classification neural network optimization method based on dimensionality transformation according to claim 4, characterized in that, The step of concatenating the high-dimensional feature set with the acquired point cloud features in the feature space includes: combining the high-dimensional feature set using the following formula... Perform the expand operation: In the formula, Represents the expanded features; [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] With the acquired point cloud features By splicing the data in the feature space using skip connections, point cloud features with high-dimensional characteristics are obtained. ： 。 6. The point cloud classification neural network optimization method based on dimensionality transformation according to claim 1 or 5, characterized in that, The steps for training a point cloud classification neural network with embedded dimension change modules using a dataset to optimize network parameters include: acquiring the ModelNet40 dataset and dividing it into training samples and test samples; inputting the training samples into the point cloud classification neural network with embedded dimension change modules for training to obtain the network's prediction results; calculating the distance between the network's prediction results and the training samples using the cross-entropy loss function; and repeatedly training the network based on the calculated distance until a preset number of training iterations is reached or the loss function converges, at which point training stops to obtain the final optimized point cloud classification neural network.
7. A point cloud classification neural network optimization device based on dimensionality transformation, characterized in that, The point cloud classification neural network includes a feature extraction module and a classification module. The feature extraction module consists of multiple sets of neural network layers appearing in the same order, with each set of repeating network layers constituting a sub-module. The device includes: an acquisition module for acquiring the point cloud classification neural network to be optimized; and an embedding module for embedding pre-constructed dimension transformation modules between each sub-module in the feature extraction module of the point cloud classification neural network. The dimension transformation module is used to non-linearly transform the point cloud features to be classified into a high-dimensional space through high-dimensional projection, thereby improving the linear separability of the point cloud features. The method involves nonlinearly transforming the point cloud features to be classified into a high-dimensional space, including: acquiring the point cloud features to be classified; linearly recombining the acquired point cloud features to obtain linearly recombined features; grouping the linearly recombined features to obtain multiple sets of features; performing nonlinear transformation processing on each set of features to obtain a high-dimensional feature set; concatenating the high-dimensional feature set with the acquired point cloud features in the feature space to obtain point cloud features with high-dimensional features; and training an optimization module to train the point cloud classification neural network with embedded dimension transformation module using the dataset to optimize the network parameters, thereby obtaining an optimized point cloud classification neural network.
8. An electronic device, characterized in that, It includes a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the point cloud classification neural network optimization method based on dimensional transformation according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the point cloud classification neural network optimization method based on dimensional transformation as described in any one of claims 1 to 6.
Citation Information
Patent Citations
Workpiece classification method and device, computer readable medium and electronic equipment
CN115410042A