A Point Cloud Classification / Segmentation Method Based on Preserving Local Geometric Contours and Global Topological Structures

Abstract

The present invention discloses a point cloud classification / segmentation method based on preserving local geometric contours and global topological structures. First, through a neighborhood relationship construction module, a neighborhood relationship graph is constructed based on spatial distances. Among them, spatial distance calculation and the k-nearest neighbor algorithm accurately capture complex geometric and semantic information in point cloud data, enhancing the accuracy and robustness of feature extraction. At the same time, the self-diffusion mechanism of the neighborhood relationship graph further enhances the correlation between local and global features of the point cloud, enabling the model to better capture the geometric contours (shapes) and global topological structure features of the point cloud. In this way, the geometric distribution characteristics in the point cloud data are fully considered, enhancing the sensitivity of the model to local geometric details of the point cloud, effectively enhancing the efficiency of local feature aggregation, reducing the influence of redundant information, avoiding feature redundancy caused by methods that only rely on the nearest neighbor method to extract features, improving the accuracy of point cloud classification and segmentation tasks, and at the same time strengthening the sharing of global information to achieve a significant improvement in the accuracy and robustness of point cloud classification and segmentation tasks.

Description

Technical Field

[0001] The present invention belongs to the technical field of dialogue recommendation, and more specifically, relates to a point cloud classification / segmentation method based on maintaining local geometric contours and global topological structures. Background Art

[0002] Deep neural networks have been widely applied to computer vision and pattern recognition tasks. In this context, point cloud analysis, as a popular research direction in the field of three-dimensional understanding, has received extensive attention from the academic and industrial communities in recent years. With the progress of technology, the acquisition methods of point cloud data have become increasingly intelligent and convenient, and there are various methods, such as optical radar (LiDAR) laser detection, three-dimensional model calculation to generate point clouds, and three-dimensional reconstruction through two-dimensional images. During the processing of point clouds, point cloud data of the same category usually has similar information such as reflection intensity and color, and often exhibits similar macroscopic features and local geometric features. Therefore, it is possible to form feature vectors using these features to classify and extract laser point clouds. Point cloud classification, as a basic point cloud analysis task, is widely applied to multiple fields such as security detection, target object recognition, medical image analysis, and three-dimensional reconstruction.

[0003] Unlike images represented by regular two-dimensional grids, point clouds are an unordered and irregular set of points. Additionally, natural properties such as sparsity and noise in point clouds further limit the performance of traditional algorithms. Therefore, directly applying standard deep learning techniques to point cloud data poses a great challenge. Nevertheless, thanks to the rapid development of deep neural networks, significant progress has been made in many point cloud analysis tasks, including 3D shape classification, part segmentation, and semantic segmentation. Limited by the unstructured form of point clouds, existing methods mainly rely on neighborhood point feature aggregation algorithms, which achieve message passing between non-consecutive points in deep learning. Current feature aggregation algorithms are mainly divided into local feature aggregation and non-local feature aggregation. For example, in a patent application for invention published on October 20, 2023, with publication number CN116912561A, a method for classifying and segmenting point cloud data based on a multi-view adaptive graph convolutional network is disclosed. By constructing a graph network in the offline stage, the collected point cloud data is input into an adaptive rotation matrix generator to obtain adaptively rotated point cloud data and generate three multi-view projection maps; the point cloud data and the three multi-view projection maps are respectively constructed to obtain global information graph data and local information graph data, and then input into a global feature extraction network and a local feature extraction network respectively. After extracting the global feature and the local feature, the two are fused and input into the output head of a functional neural network, and the loss function is calculated based on the obtained result to realize the training of the graph network; in the online stage, the point cloud data collected in real time is input into the trained graph network to obtain the point cloud classification or segmentation result. In the present invention, the point cloud data is adaptively rotated to obtain an optimized viewing angle, and according to the geometric features and spatial distribution of the point cloud data, the position of the viewing angle is flexibly adjusted, and the connection between points with different depths on a specific projection plane is introduced, so as to better capture the local and global features of the point cloud data and make the composition more accurate and robust. The global feature extraction network of this patent application for invention is a six-layer cascaded neural network with a max pooling layer, and the local feature extraction network is a two-layer cascaded neural network, which simply uses a convolutional neural network to extract the global feature and the local feature. Some other methods perform feature extraction through different distance measurement methods or different neighborhood relationship graph construction methods,

[0004] However, the existing feature aggregation methods have the following problems: 1. The local neighborhood construction method based on spatial distance usually relies on the geometric distance between points to select neighbors. This method is vulnerable to the change of the density of the point cloud distribution in the point cloud. In sparse regions, the neighbor points may be insufficient, resulting in incomplete local feature information. 2. The method based on feature distance can transcend the spatial limitation when capturing the global structure and capture points that are far away but have similar features. However, this method is prone to the problem of feature redundancy because its dependence on features causes similar points to be frequently included in the neighborhood, especially in the case where the feature distribution is highly similar or repetitive. This redundancy will cause the over-concentration of information, resulting in a decrease in the information difference within the neighborhood, thereby weakening the sensitivity of the model to local details. 3. The existing methods use a simple splicing method to fuse local and global information. Although it can retain the information of both scales to a certain extent, this way often leads to insufficient utilization of information. For point cloud scenarios with complex spatial structures or rich semantic levels, local features cannot fully play their roles in the global context only through splicing. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a point cloud classification / segmentation method based on maintaining local geometric contours and global topological structures, so as to fully consider geometric contour information, that is, the geometric features of the parts where the points are located, and topological structure information for feature aggregation, and improve the accuracy and robustness of point cloud classification / segmentation.

[0006] To achieve the above invention purpose, the point cloud classification / segmentation method based on maintaining local geometric contours and global topological structures of the present invention is characterized in that it includes the following steps:

[0007] (1). Use a multi-stage cascaded feature extractor to extract features from the point cloud data

[0008] The first-stage feature extractor takes the three-dimensional coordinates of the points in the point cloud data as input features, and the subsequent feature extractors take the features output by the previous-stage feature extractor as input features. Each stage of the feature extractor includes a neighborhood relationship construction module, a neighborhood relationship diffusion module, a feature extraction network, and a farthest point sampling module. In each stage of the feature extractor, the input features are respectively input into the neighborhood relationship construction module;

[0009] 1.1). Obtain the neighborhood relationship graph

[0010] In the neighborhood relationship construction module, first calculate the spatial distance between each point and other points according to the input features, and then determine the neighborhood point set of each point from the point cloud data according to the spatial distance, that is, construct the neighborhood relationship graph A by selecting the first k minimum spatial distances, and convert the asymmetric neighborhood relationship graph A into a symmetric matrix and perform normalization processing. The normalization of the neighborhood relationship graph is as follows:

[0011]

[0012] Among them, A represents the neighborhood relationship graph, and D represents the degree matrix of the neighborhood relationship graph;

[0013] 1.2), Using the self-diffusion mechanism to perform feature propagation and enhancement

[0014] In the neighborhood relationship diffusion module, the obtained neighborhood relationship graph is used to perform feature propagation and enhancement by the self-diffusion mechanism, so as to quickly expand to the surface contour points of the point cloud. Finally, the top k maximum values and indexes on the specified dimension in the diffused neighborhood relationship graph are determined as the input features of the feature extraction network. Among them, the calculation formula of the self-diffusion mechanism is:

[0015]

[0016] Among them, The subscript n represents the diffusion times, the superscripts n + 1 and n - i represent powers, and α is a user-defined parameter used to adjust the local neighborhood field of view;

[0017] 1.3), Extract features

[0018] The feature extraction network is a multi-layer cascaded neural network. Each layer of the neural network consists of two two-dimensional convolutional layers and one max-pooling layer. Its update function is a neighborhood feature aggregation function with residual connections, specifically:

[0019]

[0020] Among them, h i (n) is the input feature of the i-th point in the n-th layer of the neural network, h i (n+1) is the output feature of the i-th point in the n-th layer of the neural network, p i is the point cloud coordinate of the i-th point, N(i) represents the set of neighborhood points of the i-th point, j represents the j-th point in the set of neighborhood points N(i), p j is the point cloud coordinate of the j-th point in the set of neighborhood points N(i), Θ and Φ are the parameters of the two two-dimensional convolutional layers respectively, max() is the max-pooling function, and the comma in the parentheses represents concatenation;

[0021] The input features of the feature extraction network are input into the first layer of the neural network, and the output features of the last layer of the feature extraction network are the extracted features;

[0022] 1.4), Furthest point sampling

[0023] In the farthest point sampling module, the farthest point sampling method is adopted. Based on the extracted features, the farthest points are iteratively selected to extract a representative point set from the point cloud data. The features corresponding to each point in the representative point set constitute the features of the point cloud data.

[0024] (2) Point cloud classification / splitting

[0025] Using the classification / splitting output head, classification / splitting is performed according to the features of the point cloud data to obtain the point cloud classification / splitting result.

[0026] The object of the present invention is achieved as follows.

[0027] The point cloud classification / splitting method based on local geometric contour and global topological structure preservation of the present invention first constructs a neighborhood relationship graph through the neighborhood relationship construction module according to the spatial distance. Among them, the spatial distance calculation and the k-nearest neighbor algorithm accurately capture the complex geometric and semantic information in the point cloud data, enhancing the accuracy and robustness of feature extraction. At the same time, the self-diffusion mechanism of the neighborhood relationship graph further enhances the correlation between the local and global features of the point cloud, enabling the model to better capture the geometric contour (shape) and global topological structure features of the point cloud. In this way, the geometric distribution characteristics in the point cloud data are fully considered, enhancing the sensitivity of the model to the local geometric details of the point cloud, effectively enhancing the efficiency of local feature aggregation, reducing the influence of redundant information, avoiding feature redundancy caused by the method of only relying on the nearest neighbor method to extract features, improving the accuracy of the point cloud classification and splitting tasks, and at the same time strengthening the sharing of global information, achieving a significant improvement in the accuracy and robustness of the point cloud classification and splitting tasks.

[0028] Through the organic combination of modular design and innovative technologies, the present invention shows significant advantages in task adaptability, expansion ability, and combining multi-view composition. When the point cloud is sparse or has uneven density, the k-nearest neighbor algorithm can be replaced by the ball query algorithm and the ball query radius can be adjusted or the number of self-diffusions of the neighborhood relationship graph can be adjusted to achieve adaptive optimization of neighborhood selection. At the same time, the self-diffusion algorithm of the neighborhood relationship graph is essentially a form of feature propagation and can be seamlessly combined with other graph neural network technologies (such as graph convolution, graph attention, etc.) to further enhance the model performance. Brief Description of the Drawings

[0029] Figure 1 is a flowchart of a specific implementation manner of the point cloud classification / splitting method based on local geometric contour and global topological structure preservation of the present invention;

[0030] Figure 2 is the overall network structure diagram of the point cloud classification / splitting method based on local geometric contour and global topological structure preservation of the present invention;

[0031] Figure 3It is the self-diffusion effect diagram of the neighborhood relationship graph for any point in the point cloud data, where (a) is before self-diffusion and (b) is after self-diffusion;

[0032] Figure 4 It is the point cloud segmentation effect diagram of 'airplane', 'chair', 'table' and 'table lamp' in the ShapNetPart dataset. Detailed implementation manners

[0033] The following describes the detailed implementation manners of the present invention with reference to the accompanying drawings, so that those skilled in the art can better understand the present invention. It should be particularly noted that in the following description, when the detailed descriptions of known functions and designs may dilute the main content of the present invention, these descriptions will be ignored here.

[0034] Embodiment 1

[0035] Figure 1 It is a flowchart of a specific implementation manner of the point cloud classification / segmentation method based on local geometric contour and global topological structure preservation of the present invention.

[0036] In this embodiment, as Figure 1 shown, the point cloud classification / segmentation method based on local geometric contour and global topological structure preservation of the present invention includes the following steps:

[0037] Step S1: Use a multi-stage cascaded feature extractor to extract features from the point cloud data

[0038] In this embodiment, as Figure 2 shown, a four-stage cascaded feature extractor 1-4 is used to extract fused features from the point cloud data. The first-stage feature extractor takes the three-dimensional coordinates of the points in the point cloud data as input features, and the subsequent feature extractors take the features output by the previous-stage feature extractor as input features. Each stage of the feature extractor includes a neighborhood relationship construction module, a neighborhood relationship diffusion module, a feature extraction network, and a farthest point sampling module. In each stage of the feature extractor, the input features are respectively input into the neighborhood relationship construction module.

[0039] In this embodiment, point cloud data with three-dimensional coordinates P = [x, y, z] is obtained through lidar (LiDAR) scanning or a three-dimensional model generation tool, and the data size is N×3, where N is the number of points in the point cloud.

[0040] Step S1.1: Obtain the neighborhood relationship graph

[0041] In the neighborhood relationship construction module, first calculate the spatial distance between each point and other points based on the input features, and then determine the neighborhood point set of each point from the point cloud data according to the spatial distance, that is, construct the neighborhood relationship graph A by selecting the first k minimum spatial distances, convert the asymmetric neighborhood relationship graph A into a symmetric matrix and perform normalization processing. The normalization of the neighborhood relationship graph is as follows:

[0042]

[0043] where A represents the neighborhood relationship graph and D represents the degree matrix of the neighborhood relationship graph.

[0044] In this embodiment, the neighborhood relationship construction module consists of a spatial distance calculation algorithm, a k-nearest neighbor algorithm, a neighborhood relationship graph symmetrization algorithm, and a neighborhood relationship graph normalization algorithm. Among them: the spatial distance calculation algorithm is used to calculate the spatial distance between each point and other points, the k-nearest neighbor algorithm is used to determine the neighborhood point set of each point from the point cloud data according to the spatial distance, and this algorithm constructs a neighborhood relationship graph by selecting the first k minimum values. The neighborhood relationship graph symmetrization algorithm is used to convert the asymmetric neighborhood relationship graph into a symmetric matrix to meet the positive definiteness of the matrix diffusion kernel, and the neighborhood relationship graph normalization algorithm is used to normalize the graph structure data, thereby alleviating the possible numerical instability problems during the node diffusion process of the neighborhood relationship graph.

[0045] In this embodiment, the spatial distance calculation algorithm is used to calculate the Euclidean distance between two points, and the calculation formula is:

[0046]

[0047] where p1(x1, y1, z1) and p2(x2, y2, z2) are any two points in the point cloud.

[0048] Step S1.2: Use the self-diffusion mechanism to propagate and strengthen features

[0049] In the neighborhood relationship diffusion module, use the self-diffusion mechanism to propagate and strengthen the features of the obtained neighborhood relationship graph, so as to quickly expand to the surface contour points of the point cloud. Finally, determine the first k maximum values and indexes in the specified dimension of the diffused neighborhood relationship graph as the input features of the feature extraction network. Among them, the self-diffusion mechanism calculation formula is:

[0050]

[0051] where the subscript n represents the diffusion times, the superscripts n + 1 and n - i represent powers, α is a user-defined parameter, the smaller α is, the larger the field of view is. In this embodiment, it is set to 0.01 according to experience.

[0052] This self-diffusion mechanism can enhance the correlation between local and global features of the point cloud, enabling the model to better capture the geometric contour (shape) and global topological structure features of the point cloud.

[0053] In this embodiment, n = 5, and the self-diffusion effect diagram of the neighborhood relationship graph is as Figure 3 shown. The neighborhood before diffusion is the nearest neighborhood based on the central point. Due to ignoring the geometric structure of the point cloud, some points belonging to different components are included in the neighborhood of the central point, resulting in noise in feature extraction. After using the self-diffusion method, the neighborhood of the central point gradually expands from local to global, while avoiding the introduction of noise points. Such a mechanism enables the global neighborhood to be locally constrained, and local information is also refined and enhanced within the global scope, thus achieving a consistent expression of global and local in the entire point cloud structure, more efficiently capturing the multi-scale relationships in the point cloud, and significantly improving the information utilization rate and expression ability.

[0054] Step S1.3: Extract features

[0055] The feature extraction network described above is a multi-layer cascaded neural network. Each layer of the neural network consists of two two-dimensional convolutional layers and one max-pooling layer. Its update function is a neighborhood feature aggregation function with residual connections, specifically:

[0056]

[0057] where, h i (n) is the input feature of the i-th point in the n-th layer of the neural network, h i (n+1) is the output feature of the i-th point in the n-th layer of the neural network, p i is the point cloud coordinate of the i-th point, N(i) represents the set of neighborhood points of the i-th point, j represents the j-th point in the set of neighborhood points N(i), p j is the point cloud coordinate of the j-th point in the set of neighborhood points N(i), Θ and Φ are the parameters of the two two-dimensional convolutional layers respectively, max() is the max-pooling function, and the comma in the parentheses represents concatenation.

[0058] The input feature of the feature extraction network is input into the first layer of the neural network, and the output feature of the last layer of the feature extraction network is the extracted feature.

[0059] In this embodiment, the feature extraction network is a five-layer cascaded neural network. Among them: the first layer is a graph convolutional layer with an input dimension of 3 and an output dimension of 32, the second layer is a graph convolutional layer with an input dimension of 32 and an output dimension of 64, the third layer is a graph convolutional layer with an input dimension of 64 and an output dimension of 128, the fourth layer is a graph convolutional layer with an input dimension of 128 and an output dimension of 256, and the fifth layer is a graph convolutional layer with an input dimension of 256 and an output dimension of 512.

[0060] Step S1.4: Farthest Point Sampling

[0061] In the farthest point sampling module, the farthest point sampling method is adopted. Based on the extracted features, the farthest points are iteratively selected to extract a representative point set from the point cloud data while maintaining the integrity of the geometric structure information. The features corresponding to each point in the representative point set constitute the features of the point cloud data.

[0062] Step S2: Point Cloud Classification / Segmentation

[0063] A classification / segmentation output head is used to perform classification / segmentation according to the features of the point cloud data, and the point cloud classification / segmentation result is obtained.

[0064] In this embodiment, the classification / segmentation output head is a functional neural network output head including several layers of multi-layer perceptrons, and the output dimension is the number of categories in the classification task or the number of components to be segmented in the segmentation task.

[0065] The features of the point cloud data obtained in step S1 are input into the functional neural network output head, and the loss function is calculated according to the obtained result and then the network weights are updated. The functional neural network output head is a multi-layer perceptron with a series structure of two linear layers and one max pooling layer, and its output result is a vector with the dimension of the number of categories in the classification task, and the index corresponding to the maximum value is the final result predicted by the network.

[0066] The step of updating the network weights is as follows: Calculate the cross-entropy error between the obtained output result and the dataset label and backpropagate to update the network weights. For the update of the network weights, the publicly available dataset ModelNet40 is selected, in which there are 9843 point clouds in the training set, 2468 point clouds in the test set, and no validation set. 1024 points are randomly sampled from each point cloud. The optimizer used in the training process is the Adam optimizer with a learning rate of 0.001, and the number of training epochs is 300. The finally trained graph network is obtained, which includes a neighborhood relationship graph construction module, a neighborhood relationship graph diffusion module, a feature extraction network, a farthest point sampling module, and a functional neural network output head.

[0067] After specific actual experiments, using PyTorch as the underlying deep learning framework and the hardware environment of NVIDIA RTX4090 GPU (24GB), running the above method, on the officially divided test set, the average instance accuracy reaches 94.1%, which is 1.2% higher than the existing technology as the baseline, such as Wang et al. in "Dynamic graph cnn for learning on point clouds" ([J]. Acm Transactions On Graphics (tog), 2019, 38(5): 1-12).

[0068] Example 2

[0069] In this example, the ShapeNetPart dataset is selected for training and testing, which includes 12,137 training set samples, 1,870 validation set samples, and 2,874 test set samples. The number of points in each point cloud is approximately 2,000, and the specific number of point clouds may vary. During the training process, the Adam optimizer with a learning rate of 0.001 is adopted, and the number of training epochs is 200.

[0070] In the point cloud segmentation task, this example first uses the neighborhood relationship graph construction module to construct the corresponding point cloud neighborhood relationship graph. To ensure the numerical stability of the graph diffusion process, the neighborhood relationship graph symmetrization method and the neighborhood relationship graph normalization algorithm are further applied to symmetrize and normalize the neighborhood relationship graph. Subsequently, through the neighborhood relationship graph diffusion module and the feature extraction network, the feature information of the point cloud surface contour is further captured to ensure the effective focusing and enhancement of the feature information. On this basis, the farthest point sampling method is used to reduce the point cloud scale, thereby reducing the computational complexity without losing important information.

[0071] According to the requirements of a specific segmentation task, the features of the point cloud data are input into the functional neural network output head. This output head adopts a multi-layer perceptron structure and generates classification or segmentation results according to the input fused features, where the output dimension is the number of categories or parts of the point cloud. In the point cloud segmentation task, the prediction result of each point is the part category it belongs to.

[0072] In the experiment, PyTorch is used as the underlying deep learning framework, and the hardware environment of NVIDIA RTX 4090 GPU (24GB) is used for training and testing. After verification on the training set and the test set, the average instance accuracy on the test set reaches 86.4%, which is 1.3% higher than the existing technology; the average class accuracy is 83.3%, which is 1.0% higher than the existing technology.

[0073] Figure 4 are the point cloud segmentation effect diagrams of 'airplane', 'chair', 'table' and 'table lamp' in the ShapNetPart dataset. Among them, different colors represent different parts. From Figure 4 it can be seen that in categories such as 'airplane', 'chair' and 'table lamp', each part is accurately segmented, and the point cloud segmentation performance is better than the existing methods.

[0074] Although the above-described illustrative specific embodiments of the present invention have been described to facilitate understanding of the present invention by those skilled in the art, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

Claims

1. A point cloud classification / segmentation method based on preserving local geometric contours and global topological structures, characterized in that It includes the following steps: (1) Feature extraction is performed on the point cloud data using a multi-level cascaded feature extractor. The first-level feature extractor takes the three-dimensional coordinates of the points in the point cloud data as input features. The subsequent feature extractors take the features output by the previous-level feature extractor as input features. Each level of the feature extractor includes a neighborhood relationship construction module, a neighborhood relationship diffusion module, a feature extraction network, and a farthest point sampling module. In each level of the feature extractor, the input features are respectively input into the neighborhood relationship construction module. 1.1) Obtain the neighborhood relationship graph In the neighborhood relationship construction module, first calculate the spatial distance between each point and other points based on the input features. Then, determine the neighborhood point set of each point from the point cloud data according to the spatial distance, that is, construct the neighborhood relationship graph A by selecting the first k minimum spatial distances. Convert the asymmetric neighborhood relationship graph A into a symmetric matrix and perform normalization processing. The normalization of the neighborhood relationship graph is as follows: where A represents the neighborhood relationship graph, and D represents the degree matrix of the neighborhood relationship graph; 1.2) Use the self-diffusion mechanism for feature propagation and enhancement In the neighborhood relationship diffusion module, the obtained neighborhood relationship graph is used for feature propagation and enhancement using the self-diffusion mechanism, so as to quickly expand to the surface contour points of the point cloud. Finally, determine the first k maximum values and indices in the specified dimension of the diffused neighborhood relationship graph as the input features of the feature extraction network. The self-diffusion mechanism calculation formula is: Among them, the subscript n represents the number of diffusion times, the superscripts n + 1 and n - i represent powers, and α is a user-defined parameter used to adjust the local neighborhood field of view; 1.3) Extract features The feature extraction network is a multi-layer cascaded neural network. Each layer of the neural network consists of two two-dimensional convolutional layers and one max-pooling layer. Its update function is a neighborhood feature aggregation function with residual connections, specifically: Among them, h i (n) is the input feature of the i-th point in the n-th layer of the neural network, and h i (n+1) is the output feature of the i-th point in the n-th layer of the neural network. p i is the point cloud coordinate of the i-th point. N(i) represents the set of neighboring points of the i-th point. j represents the j-th point in the set of neighboring points N(i). p j is the point cloud coordinate of the j-th point in the set of neighboring points N(i). Θ and Φ are the parameters of two two-dimensional convolutional layers respectively. max() is the max pooling function. The comma in the parentheses indicates concatenation; The input features of the feature extraction network are input into the first layer of the neural network, and the output features of the last layer of the feature extraction network are the extracted features; 1.4) Farthest point sampling In the farthest point sampling module, the farthest point sampling method is adopted. Based on the extracted features, representative point sets are extracted from the point cloud data by iteratively selecting the farthest points. The features corresponding to each point in the representative point set constitute the features of the point cloud data. (2) Point cloud classification / segmentation A classification / segmentation output head is used to perform classification / segmentation according to the features of the point cloud data to obtain the point cloud classification / segmentation result.

2. The point cloud classification / segmentation method based on maintaining local geometric profiles and global topological structures according to claim 1, characterized in that The custom parameter in step 1.2) is set to 0.

01.

3. The point cloud classification / segmentation method based on maintaining local geometric profiles and global topological structures according to claim 1, characterized in that, The feature extraction network described in step 1.3) is a five-layer cascaded neural network. Among them: the first layer is a graph convolutional layer with an input dimension of 3 and an output dimension of 32, the second layer is a graph convolutional layer with an input dimension of 32 and an output dimension of 64, the third layer is a graph convolutional layer with an input dimension of 64 and an output dimension of 128, the fourth layer is a graph convolutional layer with an input dimension of 128 and an output dimension of 256, and the fifth layer is a graph convolutional layer with an input dimension of 256 and an output dimension of 512.

4. The point cloud classification / segmentation method based on local geometric contours and global structure preservation according to claim 1, characterized in that The classification / segmentation output head described in step (2) is a functional neural network output head including several layers of multi-layer perceptrons, and the output dimension is the number of categories in the classification task or the number of components to be segmented in the segmentation task.

Images