3D point cloud classification segmentation method based on adaptive Laplacian convolutional network

Abstract

The invention discloses a 3D point cloud classification segmentation method based on an adaptive Laplacian convolutional network, and the method comprises the steps: carrying out the network training through employing a public data set after preprocessing and sampling grouping, carrying out the improvement of a conventional Laplacian convolutional network, introducing a vector attention method into a feature extraction part, achieving the adaptive adjustment of a weight, and achieving the classification segmentation of a 3D point cloud. And establishing a skip connection mode between modules of the feature extraction part to realize multi-scale fusion of features. And designing an enhancement module to enhance the extracted local feature information. And finally, classifying and segmenting the 3D point cloud to be detected by the trained network. According to the method, the problem that local information is difficult to extract due to irregular characteristics of 3D point clouds is solved, and a new solution thought is provided for solving the problem that performance of a part of networks is suddenly reduced when facing sparse point clouds and noisy point clouds. Moreover, the method pursues lightweight design, is good in real-time performance, and can accurately complete the classification and segmentation task of the 3D point cloud.

Description

Technical Field

[0001] The present invention relates to the technical field of 3D point cloud feature extraction and classification segmentation, and in particular to a 3D point cloud classification segmentation method based on an adaptive Laplace convolutional network. Background Art

[0002] Extracting features directly from 3D point clouds for classification and segmentation is an urgent need in applications such as robotics, autonomous driving, and augmented reality. Unlike 2D images, point clouds are disordered and unstructured, making it challenging to design neural networks to process them.

[0003] There are two approaches to this problem. One is to design a geometric structure extractor and perform convolution operations to extract local geometric features of the point cloud. However, since standard convolution is poorly applicable to 3D point clouds and insensitive to the connections between point cloud data, this approach suffers from low extraction performance. The other approach is to use a multi-head attention mechanism to extract global features. However, this approach incurs significant computational overhead, hindering the ability to achieve rapid iteration in real-time tasks. Furthermore, both approaches suffer from a significant performance degradation when processing sparse and noisy point clouds, resulting in poor robustness. Summary of the Invention

[0004] This invention aims to overcome the shortcomings and deficiencies of existing technologies by proposing a 3D point cloud classification and segmentation method based on an adaptive Laplacian convolutional network. This method effectively extracts local and global features from point cloud data, while maintaining a lightweight network structure and minimizing unnecessary overhead. Furthermore, this method exhibits robustness, avoiding the problem of a sudden drop in accuracy when processing sparse point clouds or those affected by noise.

[0005] To achieve the above objectives, the present invention provides a technical solution: a 3D point cloud classification and segmentation method based on an adaptive Laplace convolutional network, comprising the following steps:

[0006] 1) Use a public 3D point cloud dataset that contains ground truth point cloud category labels and several point clouds, and preprocess the dataset to obtain uniform and evenly distributed point cloud data;

[0007] 2) Using the farthest point sampling algorithm and the sphere query algorithm, the point cloud data obtained after preprocessing is sampled and grouped, and the point cloud data is divided into several sphere groups. Each sphere group has a center point and each sphere group contains all the point clouds within a sphere formed by a specified radius;

[0008] 3) Performing feature extraction on the multiple spherical groups obtained through sampling and grouping to extract corresponding feature vectors, including center point features and neighborhood features, where the neighborhood features refer to features of other point clouds in the spherical group except the center point;

[0009] 4) Design a Laplacian feature convolution mechanism that includes vector attention. This mechanism adaptively assigns attention weights to feature vectors and strengthens the center point features and neighborhood features of the point cloud in the spherical group based on the size of the attention weights, achieving convolution between the center point features and neighborhood features.

[0010] 5) Design an enhancement module that includes similarity matching calculation to further enhance the extracted feature vectors to improve the accuracy of network training;

[0011] 6) Using the feature vector to train an adaptive Laplace convolutional network, using multi-classification cross entropy as the loss function, and outputting the corresponding point cloud category; wherein the adaptive Laplace convolutional network is an improved Laplace convolutional network, which is improved by adding vector attention to the original Laplace convolutional network, adaptively obtaining Laplace weights, adding a multi-scale fusion method of skip connections, inputting point clouds, and using the spherical domain groups obtained by group sampling to perform feature extraction respectively, obtaining local and global features of the point cloud data, so that the overall features of the corresponding space are more complete;

[0012] 7) Apply the trained adaptive Laplace convolutional network and input the 3D point cloud data to be tested into the network to obtain the point cloud category and the segmentation effect of each component.

[0013] Furthermore, in step 2), the point cloud data obtained after preprocessing is sampled and grouped. The specific process is as follows:

[0014] Input point cloud data U={u1,u2,...,u n}, where u n Represents the nth point cloud; the center point set U cm ={u c1 ,u c2 ,...,u cm} is selected using the iterative farthest point sampling algorithm, where u cm represents the mth center point; U cm = FPS{U}, for the sampled center point set U cm , use the ball query algorithm to determine each center point set U according to the distance cm Neighborhood N cm ={u cm |u m1 ,u m2 ,...,u mj}, where u mj Indicates u cm The adjacent points of the jth point as the center, N cm Indicates u cmThe spherical region group with the center point N cm =queryBall{U cm}, N cm By U cm Obtained through the ball query algorithm.

[0015] Furthermore, in step 3), the adaptive Laplace backbone part of the adaptive Laplace convolutional network is used to extract features of the spherical domain group, and the extracted feature vector includes center point features and neighborhood features; the adaptive Laplace backbone part is composed of 6 convolution modules with different parameters stacked together, and these convolution modules are fused through multi-scale features to ensure that more comprehensive local information is extracted. The convolution module not only includes linear layers, convolution operations and activation functions, but also combines a vector attention mechanism to improve the effect and accuracy of feature extraction.

[0016] Furthermore, in step 4), the specific implementation of the vector attention and Laplacian feature convolution mechanism is as follows:

[0017] Vector attention: randomly generate a set of parameters op that obey the standard normal distribution, normalize op and convert it into learnable parameters, weight W a By the op and N after the softmax function cm The dot product is obtained, and N cm With all center point sets U cm ={u c1 ,u c2 ,...,u cm} are spliced ​​together through the fully connected layer and finally the output F is obtained on , as shown in the following formula:

[0018] F on =Conv1d(concat(U cm ,sum(W a N cm )))

[0019] In the formula, sum is a summation operation, concat is a concatenate operation, which is used to concatenate along the vector dimension. The concatenated data is then passed through Conv1d, which is a one-dimensional convolution operation. Vector attention uses the attention method to calculate the weight.

[0020] Laplace feature convolution mechanism: The input of the Laplace convolution module is denoted as F in , the weighted sum of the Laplace features learned by training the neural network is expressed as F a , F a and F in The relationship between them is expressed by the following formula:

[0021] F a =∑softmax(θ·(N cm -u cm ))·(N cm -u cm )

[0022] F in =N cm -u cm

[0023] Where, θ=(θ1,θ2,..,θ m ) is the parameter of neural network learning, θ m represents the mth parameter learned; the Laplace feature convolution mechanism introduces an adaptive strategy in deep learning, making it not only equivalent to using a neural network layer to infer weights, but also introducing higher flexibility and accuracy in point cloud classification and segmentation tasks. It adopts a discretization method that is completely different from the existing technology, abandoning the dependence on uniform discretization or specific geometric information, thereby achieving a deeper understanding of complex geometric shapes, enabling the network to dynamically strengthen the relationship between the center point features and the neighborhood features in the spherical domain group according to the characteristics of the specific data.

[0024] Furthermore, in step 5), the enhancement module is specifically implemented as follows:

[0025] Similarity matching is used to calculate the relationship between neighborhood features and the true point cloud category labels, and then linear interpolation is performed with the vector features obtained from the adaptive Laplace backbone to enhance the features. The similarity matching calculation formula is as follows:

[0026] L pre =exp(-γ·(1-∑F s ·L t T ))·L t T

[0027] Where, F s is the feature for similarity matching calculation, L t is the ground truth point cloud category label, L t T For L t The matrix transpose of , γ is the measured matching coefficient, which is used to control the sensitivity of the matching calculation process. The linear interpolation formula is as follows:

[0028] L cls =λ q ·L pre +λ n ·F b

[0029] Where, L pre is the result of similarity matching calculation, F b It is the sum of the features extracted by the adaptive Laplace backbone, and the two are linearly interpolated to obtain the final prediction value L cls ,λ q and λ n are measured parameters, which are used to control the degree of interpolation between the two, thereby achieving the regulatory effect of feature enhancement; by introducing the enhancement module of similarity matching calculation, the real label information and the extracted features can be effectively integrated, so that the network can distinguish the importance of features during the training process.

[0030] Furthermore, in step 6), the adaptive Laplacian convolutional network processes point cloud segmentation and classification, and the loss function adopted is multi-classification cross entropy to minimize the distance between the predicted value and the true point cloud category label; the output of the network is a probability value indicating the probability of belonging to N categories; for multi-classification problems, the true point cloud category label is a one-hot encoded vector indicating which category the sample belongs to; the output of the entire network is a vector indicating the probability of each category, which is processed by the softmax function; assuming there are N points, each point has C categories, the probability distribution of the network output is P, and the true point cloud category label is Y; the loss function for point cloud classification is expressed as:

[0031]

[0032] Where N is the number of points, C is the number of categories, and Y ij is the true label of point i belonging to category j, and the probability of the point predicted by the network belonging to category j is P ij , category j is used as the output of the network, and the trained adaptive Laplace convolutional network can be used to complete the classification and segmentation task of point cloud.

[0033] Furthermore, in step 7), the specific steps of applying the trained adaptive Laplacian convolutional network to complete the 3D point cloud classification and segmentation are as follows:

[0034] 71) Scanning the object to be classified and segmented by the sensor to obtain the complete 3D point cloud data of the object to be measured, and feeding the obtained 3D point cloud data into the trained adaptive Laplace convolutional network after preprocessing;

[0035] 72) After step 71), the acquired point cloud data is sampled and grouped to obtain different spherical domain groups. Corresponding features are extracted from the different spherical domain groups. A Laplace feature convolution mechanism is used to perform deep local feature extraction. Finally, a feature enhancement module is used to enhance the extracted global and local features.

[0036] 73) After the feature extraction in step 72) above, the category of the object to be tested is output for the classification task; and the segmentation effect of each component of the object to be tested is output for the segmentation task.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] 1. The present invention applies Laplace convolution to 3D point cloud data, which has a strong ability to capture local details of the point cloud and can effectively extract features.

[0039] 2. This paper proposes a vector attention method that uses a neural network to adaptively learn Laplace weights, so that the Laplace convolution no longer relies on a uniform discretization method, but can be adaptively adjusted according to different tasks.

[0040] 3. The neural network training of the present invention adopts a lightweight design concept, which reduces redundant calculations while ensuring high classification and segmentation accuracy.

[0041] 4. The present invention can process relatively sparse and noisy point clouds and has strong robustness.

[0042] 5. The neural network structure of the present invention introduces a point cloud feature enhancement branch, which can provide more feature information and make classification and segmentation more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a logical flow diagram of the present invention.

[0044] Figure 2 Schematic diagram of an example of a neural network structure designed for the present invention.

[0045] Figure 3 Schematic diagram of the vector attention structure designed for this invention.

[0046] Figure 4 This is an example effect diagram of the neural network classification and segmentation designed for the present invention. DETAILED DESCRIPTION

[0047] The present invention will be described in further detail below with reference to the embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0048] This embodiment discloses a 3D point cloud classification and segmentation method based on an adaptive Laplace convolutional network, which uses a deep learning algorithm and is trained using a public 3D point cloud dataset. The logical process is as follows: Figure 1 As shown, the neural network structure is as Figure 2 As shown, the vector attention structure is as follows Figure 3 As shown, the classification and segmentation effect is as follows Figure 4 As shown, it includes the following steps:

[0049] 1) Use a public 3D point cloud dataset that contains ground truth point cloud category labels and several point clouds, and preprocess the dataset to obtain uniform and evenly distributed point cloud data;

[0050] The public 3D point cloud dataset contains ground truth point cloud category labels and several point clouds. Classification training is performed on the ModelNet40 dataset and the ScanObjNN dataset. The ModelNet40 dataset contains 40 categories of CAD mesh models, of which 9,843 are trained and 2,468 are tested; the ScanObjNN

[41] dataset contains 15,000 objects, which are divided into 15 categories and 2,902 unique object instances in the real world. The preprocessing step is to read the point cloud dataset, distinguish between training data and test data, and then downsample the point clouds in the dataset to ensure that the number of point clouds fed into the network is uniform and evenly distributed.

[0051] 2) Using the farthest point sampling algorithm and the sphere query algorithm, the pre-processed point cloud data is sampled and grouped into several sphere groups. Each sphere group has a center point. A sphere group refers to all point clouds contained in a sphere with a certain radius.

[0052] Input point cloud data U={u1,u2,...,u n}, where u n Represents the nth point cloud; the center point set U cm ={u c1 ,u c2 ,...,u cm} is selected using the iterative farthest point sampling algorithm, where u cm represents the mth center point; U cm = FPS{U}, for the sampled center point set U cm , use the ball query algorithm to determine each center point set U according to the distance cm Neighborhood N cm ={u cm |u m1 ,u m2 ,...,u mj}, where u mj Indicates u cm The adjacent points of the jth point as the center, N cm Indicates u cm The spherical region group with the center point N cm =queryBall{U cm}, N cm By U cm Obtained through the ball query algorithm;

[0053] The basic principle of farthest point sampling is to select a starting point from the dataset and then iteratively select the point farthest from the currently selected point set. This method ensures that the sampling points are relatively evenly distributed in space, effectively preserving the geometric structure of the dataset. The algorithm steps are as follows:

[0054] S1: Assume that the input set P contains n points, select a point p0 from the set P as the starting point, and obtain the sampling point set S = {p0};

[0055] S2: Calculate the distances from all points to p0, form an array L, select the point corresponding to the maximum value as p1, and update the sampling point set S = {p0, p1};

[0056] S3: Calculate the distance from all points to p1, for each point p i If the distance between it and p1 is less than L[i], then update L[i]=d(p i ,p1), L always stores the shortest distance between each point and the sampling point set S;

[0057] S4: Select the point corresponding to the maximum value in L as p2, and update the sampling point set S = {p0, p1, p2};

[0058] S5: Repeat steps S2-S4 until the sampling is completed.

[0059] 3) Perform feature extraction on the multiple spherical groups obtained by sampling and grouping, and extract the corresponding feature vectors, including center point features and neighborhood features, where neighborhood features refer to the features of other point clouds in the spherical group except the center point;

[0060] like Figure 2 As shown in the figure, the feature extraction part is the adaptive Laplacian backbone of the adaptive Laplacian convolutional network, which is composed of six stacked convolutional modules with different parameters. Multi-scale feature fusion is performed between the convolutional modules, allowing the adaptive Laplacian backbone to extract more comprehensive local information. The convolutional module not only includes linear layers, convolution operations, and activation functions, but also incorporates a vector attention mechanism.

[0061] 4) Design a Laplacian feature convolution mechanism that includes vector attention. This mechanism adaptively assigns attention weights to feature vectors and strengthens the center point features and neighborhood features of the point cloud in the spherical group based on the size of the attention weights, achieving convolution between the center point features and neighborhood features.

[0062] like Figure 3As shown in the figure, it is a specific flow chart of vector attention. Vector attention enables the network to adaptively learn the features of different local areas. The weights will continue to learn and change according to training instead of being fixed to the default value. This helps to capture the complex local structure information in the 3D point cloud, thereby improving the performance of the network.

[0063] The specific steps of vector attention are: randomly generate a set of parameters op that obey the standard normal distribution, normalize op and convert them into learnable parameters. a By the op and N after the softmax function cm The dot product is obtained. And N cm With all center point sets U cm ={u c1 ,u c2 ,...,u cm}, and spliced ​​together through the fully connected layer. Finally, the output F on , as shown in the formula:

[0064] F on =Conv1d(concat(U cm ,sum(W a N cm )))

[0065] Among them, sum is a summation operation, concat is a concatenate operation, which is used to splice along the vector dimension. The spliced ​​data is then passed through Conv1d, which is a conventional one-dimensional convolution operation. Vector attention uses the attention method to calculate the weight;

[0066] The specific implementation of the Laplace feature convolution mechanism: the input of the Laplace convolution module is denoted as F in , the weighted sum of the Laplace features learned by training the neural network is expressed as F a , F a and F in The relationship between can be expressed by the following formula:

[0067] F a =∑softmax(θ·(N cm -u cm ))·(N cm -u cm )

[0068] F in =N cm -u cm

[0069] where N cm is the neighborhood, u cm is the center point, θ=(θ1,θ2,..,θm ) is the parameter of neural network learning, θ m Represents the mth parameter learned. The Laplace feature convolution mechanism is equivalent to using a neural network layer to infer weights and is applicable to corresponding point cloud classification and segmentation tasks. Compared with the traditional Laplace convolution method, its discretization method is more flexible and no longer relies on uniform discretization or specific geometric information. The Laplace feature convolution mechanism introduces adaptive strategies in deep learning, making it not only equivalent to using traditional neural network layers to infer weights, but also introducing higher flexibility and accuracy in point cloud classification and segmentation tasks. The innovation of this method lies in that it adopts a discretization method that is completely different from existing technologies, abandoning the dependence on uniform discretization or specific geometric information, thereby achieving a deeper understanding of complex geometric shapes, enabling the model to dynamically strengthen the relationship between the center point features and neighborhood features in the spherical domain group according to the characteristics of the specific data, and showing stronger robustness and generalization capabilities compared to traditional methods.

[0070] 5) Design an enhancement module that includes similarity matching calculation to further enhance the extracted feature vectors to improve the accuracy of network training;

[0071] Similarity matching is used to calculate the relationship between neighborhood features and the true point cloud category labels, and then linear interpolation is performed with the vector features obtained from the adaptive Laplace backbone to enhance the features. The similarity matching calculation formula is as follows:

[0072] L pre =exp(-γ·(1-∑F s ·L t T ))·L t T

[0073] Where, F s is the feature for similarity matching calculation, L t is the ground truth point cloud category label, L t T For L t The matrix transpose of , γ is the measured matching coefficient, which is used to control the sensitivity of the matching calculation process. The linear interpolation formula is as follows:

[0074] L cls =λ q ·L pre +λ n ·F b

[0075] Where, L pre is the result of similarity matching calculation, F bIt is the sum of the features extracted by the adaptive Laplace backbone, and the two are linearly interpolated to obtain the final prediction value L cls ,λ q and λ n are measured parameters that control the degree of interpolation between the two, thereby achieving a regulatory effect on feature enhancement. By introducing an enhanced module for similarity matching calculations, it is possible to effectively integrate real label information with extracted features, allowing the network to differentiate the importance of features during training. This strategy not only improves the accuracy of feature representation but also enhances the network's sensitivity to complex data information.

[0076] 6) Use the feature vector to train the adaptive Laplace convolutional network, use multi-classification cross entropy as the loss function, and output the corresponding point cloud category. The adaptive Laplace convolutional network is an improved Laplace convolutional network. Its improvement is to add vector attention to the original Laplace convolutional network, adaptively obtain Laplace weights, add a multi-scale fusion method with skip connections, input point clouds, and use the spherical domain groups obtained by group sampling to extract features separately, obtain local and global features of the point cloud data, and make the overall features of the corresponding space more complete;

[0077] The adaptive Laplacian convolutional network mainly handles point cloud segmentation and classification. The loss function used is multi-class cross entropy to minimize the distance between the predicted value and the true point cloud category label. The output of the network is a probability value indicating the probability of belonging to N categories. For multi-classification problems, the true point cloud category label is a one-hot encoded vector indicating which category the sample belongs to. The output of the entire network is a vector indicating the probability of each category, which is processed by the softmax function. Assuming there are N points, each point has C categories, the probability distribution of the network output is P, and the true point cloud category label is Y. The loss function for point cloud classification is expressed as:

[0078]

[0079] Where N is the number of points, C is the number of categories, and Y ij is the true label of point i belonging to category j, and the probability of the point predicted by the network belonging to category j is P ij , category j is used as the output of the network, and the trained adaptive Laplace convolutional network can be used to complete the classification and segmentation task of point cloud.

[0080] 7) Apply the trained adaptive Laplace convolutional network and input the 3D point cloud data to the network to obtain the point cloud category and component segmentation results. The specific steps are as follows:

[0081] 71) Scanning the object to be classified and segmented by the sensor to obtain the complete 3D point cloud data of the object to be measured, and feeding the obtained 3D point cloud data into the trained adaptive Laplace convolutional network after preprocessing;

[0082] 72) After step 71), the acquired point cloud data is sampled and grouped to obtain different spherical domain groups. Corresponding features are extracted from the different spherical domain groups. A Laplace feature convolution mechanism is used to perform deep local feature extraction. Finally, a feature enhancement module is used to enhance the extracted global and local features.

[0083] 73) After the feature extraction in step 72) above, the category of the object to be tested is output for the classification task; and the segmentation effect of each component of the object to be tested is output for the segmentation task.

[0084] The following details the experimental results of this experiment:

[0085] According to the final results of the network, the accuracy and speed of the improved network are evaluated from the overall accuracy (OA), average category accuracy (mAcc), average category intersection over union (mIoU), parameter count (Params) and floating-point operation size (FLOPs) indicators.

[0086] The model size comparison between the adaptive Laplacian convolutional network and other networks is shown in Table 1.

[0087] Table 1

[0088] method Params.(M) FLOPs(G) PointNet 3.5 0.9 PointConv 18.6 - KpConv 15.2 - PointMLP 13.2 31.3 Adaptive Laplacian Convolutional Network 4.5 10.5

[0089] The accuracy comparison results of the adaptive Laplace convolutional network and other algorithms are shown in Tables 2 and 3 below.

[0090] Table 2

[0091] method mAcc (%) OA (%) PointNet 63.4 68.2 SpiderCNN 69.8 73.7 PointNet++ 75.4 77.9 PointCNN 75.1 78.5 Dgcnn 73.6 78.1 DRNet 78.0 80.3 GBNet 77.8 80.5 PRANet 79.1 82.1 Adaptive Laplacian Convolutional Network 82.1 84.2

[0092] Table 3

[0093]

[0094]

[0095] Where input represents the dimensions contained in the input to the network, xyz represents the three-dimensional coordinates (x, y, z), and nr represents the normal vector coordinates (n x ,n y ,n z ) information. The number of input points indicates the number of points for preprocessing sampling.

[0096] The results in the table above show that the adaptive Laplace convolutional network Dgcnn has obvious advantages in accuracy and speed compared with traditional point cloud feature extraction classification and segmentation networks PointNet and PointNet++. In addition, the adaptive Laplace convolutional network is more lightweight in terms of model size. In general, the classification and segmentation accuracy of the present invention is higher and the real-time performance is also better.

[0097] The comparison results of the adaptive Laplace convolutional network segmentation instance and the true value are as follows Figure 4 As shown in the figure, it can be seen that the components are clearly layered, the predicted categories are correct, and they are basically close to the facts.

[0098] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A 3D point cloud classification and segmentation method based on an adaptive Laplace convolutional network, characterized in that: The following steps are involved: 1) Use a public 3D point cloud dataset that contains true point cloud category labels and several point clouds, and preprocess the dataset to obtain point cloud data with a uniform quantity and uniform distribution; 2) Using the farthest point sampling algorithm and the ball query algorithm, the point cloud data obtained after preprocessing is sampled and grouped, and the point cloud data is divided into several spherical domain groups, each of which has a center point and contains all point clouds within a sphere formed by a specified radius; 3) Performing feature extraction on the multiple spherical domain groups obtained by sampling and grouping, and extracting corresponding feature vectors, including center point features and neighborhood features, wherein the neighborhood features refer to features of other point clouds in the spherical domain group except the center point; 4) Design a Laplace feature convolution mechanism that includes vector attention, use the Laplace feature convolution mechanism to adaptively allocate attention weights to the feature vectors, and strengthen the center point features and neighborhood features of the point cloud in the spherical group according to the size of the attention weights, thereby realizing the convolution between the center point features and the neighborhood features; 5) Design an enhancement module including similarity matching calculation to further enhance the extracted feature vectors to improve the accuracy of network training; 6) Using feature vectors to train an adaptive Laplace convolutional network, using multi-classification cross entropy as a loss function, and outputting corresponding point cloud categories; wherein the adaptive Laplace convolutional network is an improved Laplace convolutional network, and its improvement is to add vector attention on the basis of the original Laplace convolutional network, adaptively obtain Laplace weights, add a multi-scale fusion method of skip connections, input point clouds, and use the spherical domain groups obtained by group sampling to extract features respectively, obtain local and global features of point cloud data, so that the overall features of the corresponding space are more perfect; 7) Apply the trained adaptive Laplace convolutional network and input the 3D point cloud data to be tested into the network to obtain the point cloud category and the segmentation effect of each component.

2. The 3D point cloud classification and segmentation method based on the adaptive Laplace convolutional network according to claim 1, characterized in that: In step 2), the point cloud data obtained after preprocessing is sampled and grouped. The specific process is as follows: Input point cloud data U = {u1,u2,...,u n }, where u n Represents the nth point cloud; the center point set U cm = {u c1 ,u c2 ,...,u cm } is selected using the iterative farthest point sampling algorithm, where u cm represents the mth center point; U cm = FPS{U}, for the sampled center point set U cm , use the ball query algorithm to determine each center point set U according to the distance cm Neighborhood N cm = {u cm |u m1 ,u m2 ,...,u mj }, where u mj Indicates u cm The neighboring points of the jth point as the center, N cm Indicates u cm The spherical region group with N as the center point, cm =queryBall{U cm },N cm By U cm Obtained through the ball query algorithm.

3. The 3D point cloud classification and segmentation method based on the adaptive Laplace convolutional network according to claim 2, characterized in that: In step 3), the adaptive Laplace backbone part of the adaptive Laplace convolutional network is used to extract features from the spherical domain group, and the extracted feature vector includes center point features and neighborhood features; The adaptive Laplacian backbone is composed of 6 stacked convolution modules with different parameters. These convolution modules are fused with multi-scale features to ensure the extraction of more comprehensive local information. The convolution module not only includes linear layers, convolution operations and activation functions, but also combines vector attention mechanisms to improve the effect and accuracy of feature extraction.

4. The 3D point cloud classification and segmentation method based on the adaptive Laplace convolutional network according to claim 3 is characterized in that: In step 4), the specific implementation of the vector attention and Laplacian feature convolution mechanism is as follows: Vector attention: randomly generate a set of parameters op that follow a standard normal distribution, normalize op and convert them into learnable parameters, weights W a By the op and N after the softmax function cm The dot product is obtained, and N cm With all the center points U cm = {u c1 ,u c2 ,...,u cm } are spliced ​​together through the fully connected layer, and finally the output F is obtained on , as shown in the following formula: F on =Conv1d(concat(U cm ,sum(W a N cm ))) In the formula, sum is a summation operation, concat is a concatenate operation, and concatenation is performed along the vector dimension. The concatenated data is then passed through Conv1d, which is a one-dimensional convolution operation. Vector attention calculates weights through the attention method. Laplace feature convolution mechanism: The input through the Laplace convolution module is denoted as F in , the weighted sum of the Laplace features learned by training the neural network is expressed as F a , F a and F in The relationship between is expressed by the following formula: F a =Σsoftmax(θ·(N cm -u cm ))·(N cm -u cm ) F in =N cm -u cm In the formula, θ=(θ1,θ2,..,θ m ) is the parameter learned by the neural network, θ m represents the mth parameter learned; the Laplace feature convolution mechanism introduces an adaptive strategy in deep learning, which is not only equivalent to using a neural network layer to infer weights, but also introduces higher flexibility and accuracy in point cloud classification and segmentation tasks. It adopts a discretization method that is completely different from the existing technology, abandons the dependence on uniform discretization or specific geometric information, thereby achieving a deeper understanding of complex geometric shapes, and enables the network to dynamically strengthen the relationship between the center point features and the neighborhood features in the spherical domain group according to the characteristics of the specific data.

5. The 3D point cloud classification and segmentation method based on the adaptive Laplace convolutional network according to claim 4, characterized in that: In step 5), the specific implementation of the enhancement module is as follows: The relationship between the neighborhood features and the true point cloud category labels is calculated using similarity matching, and then linear interpolation is performed with the vector features obtained from the adaptive Laplace backbone to enhance the features. The similarity matching calculation formula is as follows: L pre =exp(-γ·(1-∑F s ·L t T ))·L t T In the formula, F s is the feature for similarity matching calculation, L t is the true point cloud category label, L t T For L t The matrix transpose of , γ is the measured matching coefficient, which is used to control the sensitivity of the matching calculation process. The linear interpolation formula is as follows: L cls =λ q ·L pre +λ n ·F b Where, L pre is the result of similarity matching calculation, F b is the sum of the features extracted from the adaptive Laplace backbone, and the two are linearly interpolated to obtain the final prediction value L cls ,λ q and λ n are measured parameters, which are used to control the degree of interpolation between the two, thereby achieving the regulatory effect of feature enhancement; by introducing the enhancement module of similarity matching calculation, the real label information and the extracted features can be effectively integrated, so that the network can distinguish the importance of features during the training process.

6. The 3D point cloud classification and segmentation method based on the adaptive Laplace convolutional network according to claim 5, characterized in that: In step 6), the adaptive Laplace convolutional network processes point cloud segmentation and classification, and the loss function adopted is multi-classification cross entropy to minimize the distance between the predicted value and the true point cloud category label; the output of the network is a probability value indicating the probability of belonging to N categories; for multi-classification problems, the true point cloud category label is a one-hot encoded vector indicating which category the sample belongs to; the output of the entire network is a vector indicating the probability of each category, which is processed by the softmax function; assuming that there are N points, each point has C categories, the probability distribution of the network output is P, and the true point cloud category label is Y; the loss function for point cloud classification is expressed as: In the formula, N is the number of points, C is the number of categories, and Y is ij is the true label of point i belonging to category j, and the probability predicted by the network that the point belongs to category j is P ij , category j is taken as the output of the network, and the trained adaptive Laplace convolutional network can be used to complete the classification and segmentation task of point cloud.

7. The 3D point cloud classification and segmentation method based on the adaptive Laplace convolutional network according to claim 6, characterized in that: In step 7), the specific steps of applying the trained adaptive Laplace convolutional network to complete the 3D point cloud classification and segmentation are as follows: 71) Scanning the object to be classified and segmented by the sensor to obtain the complete 3D point cloud data of the object to be measured, and sending the obtained 3D point cloud data to the trained adaptive Laplace convolutional network after preprocessing; 72) After the processing in step 71), the acquired point cloud data is sampled and grouped to obtain different spherical domain groups, corresponding features are extracted according to the different spherical domain groups, and a Laplace feature convolution mechanism is used to perform deep local feature extraction, and finally a feature enhancement module is used to enhance the extracted global and local features; 73) After the feature extraction in step 72) above, the category of the object to be tested is output for the classification task; and the segmentation effect of each component of the object to be tested is output for the segmentation task.

Images