Intelligent classification method and system based on point cloud data multi-dimensional feature extraction

Abstract

The invention discloses an intelligent classification method and system based on point cloud data multi-dimensional feature extraction, and relates to the technical field of deep learning, and the method comprises the steps: collecting point cloud data, and carrying out the preprocessing of the point cloud data; extracting features based on the preprocessed point cloud data, and performing adaptive fusion by using the relative change rate of the features to generate a fusion feature vector; constructing a hybrid activation multi-layer sensing network model by using the fused feature vectors; performing multi-scale processing on the original point cloud data, extracting features, generating a multi-scale fusion high-dimensional feature vector through splicing, and transmitting the multi-scale fusion high-dimensional feature vector to a hybrid activation multi-layer sensing network model; introducing an attention mechanism for the hybrid activation multilayer sensing network model, and making a classification decision; according to the method, the hybrid activation multi-layer sensing network model can be quickly converged in the training process, and the classification precision is improved; and when high-dimensional fusion feature vectors are processed, better classification performance is shown, and the overall classification effect and robustness are improved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to an intelligent classification method and system based on multi-dimensional feature extraction of point cloud data. Background Art

[0002] With the development of three-dimensional sensing technology, point cloud data has been widely used in autonomous driving, robot navigation, 3D reconstruction and other fields; point cloud data, as a discrete expression of three-dimensional space, can accurately describe the geometric shape and spatial distribution of objects, and therefore plays an important role in intelligent classification and target recognition; traditional point cloud data processing methods usually rely on extraction and classification models based on geometric features. Although these methods can provide a certain degree of accuracy, due to the sparsity, irregularity and high dimensionality of point cloud data, traditional methods have problems such as high computational complexity and insufficient feature extraction when processing large-scale point cloud data; for this reason, point cloud classification methods combined with deep learning have gradually received attention, among which models such as multilayer perceptrons and convolutional neural networks have been introduced to enhance the automatic feature extraction and classification capabilities of point cloud data.

[0003] However, existing technologies still have many shortcomings when dealing with multi-scale point cloud feature fusion and classification. First, existing methods lack an adaptive fusion mechanism for feature changes at different scales, resulting in insufficient feature expression capabilities. Second, most traditional classification models use a single activation function, which cannot effectively combine the nonlinear relationship between different types of features, limiting the classification performance of the model. Third, existing models do not fully introduce attention mechanisms when processing high-dimensional features, making it difficult to focus on important feature dimensions or spatial regions, resulting in a decrease in classification accuracy. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent classification method based on multi-dimensional feature extraction of point cloud data to solve the problem of low classification accuracy of point cloud data.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides an intelligent classification method based on multi-dimensional feature extraction of point cloud data, which includes collecting point cloud data and preprocessing the point cloud data; extracting features based on the preprocessed point cloud data, and adaptively fusing the features using the relative change rate of the features to generate a fused feature vector; using the fused feature vector to construct a hybrid activated multi-layer perception network model; performing multi-scale processing on the original point cloud data to extract features, generating a multi-scale fused high-dimensional feature vector by splicing and transmitting it to the hybrid activated multi-layer perception network model; introducing an attention mechanism into the hybrid activated multi-layer perception network model to make classification decisions.

[0008] As a preferred solution of the intelligent classification method based on multi-dimensional feature extraction of point cloud data described in the present invention, wherein: the point cloud data is collected and preprocessed, and the specific steps are as follows:

[0009] Collect 3D point cloud data;

[0010] Statistical filtering is used for denoising and downsampling is performed using a voxel grid filter.

[0011] As a preferred solution of the intelligent classification method based on multi-dimensional feature extraction of point cloud data described in the present invention, wherein: the extracted features are adaptively fused using the relative change rate of the features to generate a fused feature vector, and the specific steps are as follows:

[0012] Use principal component analysis PCA to extract geometric features from 3D point cloud data;

[0013] Based on the reflection intensity in the 3D point cloud data, physical features are obtained;

[0014] Based on the adaptation of the relative change rate of features, the extracted geometric features and physical features are fused to obtain the fused feature vector F f , the expression is:

[0015]

[0016] Where Z is the normalization factor, Δ geo is the relative rate of change of the geometric feature, Δ phy is the relative rate of change of the physical characteristic, f geo (i) is the i-th geometric feature, f phy (j) is the jth physical feature, i is the index of the geometric feature, j is the index of the physical feature, m is the total number of physical features, and n is the total number of geometric features.

[0017] As a preferred solution of the intelligent classification method based on multi-dimensional feature extraction of point cloud data described in the present invention, the specific steps of constructing a hybrid activation multi-layer perception network model are as follows:

[0018] The fused feature vector is input into the multi-layer hidden layer, and a linear transformation is performed to introduce nonlinear capabilities. The activation result f(x) is obtained by processing it through a mixed activation function. The expression is:

[0019] f(x)=α×ReLU(x)+(1-α)×Tanh(x);

[0020] Among them, α is the adaptive parameter, ReLU(x) is the activation function, Tanh(x) is the hyperbolic tangent activation function, and x is the input data;

[0021] Based on the activation results, the activation results are mapped to the category space of the classification task through linear transformation;

[0022] Set a classification threshold, and select the category whose category score in the category space reaches the set threshold as the prediction result of the mixed activation multi-layer perception network model;

[0023] The cross entropy loss function is used to measure the difference between the prediction results of the mixed activation multi-layer perceptron model and the true category label;

[0024] The parameters of the mixed activation multilayer perception network model are optimized using the Adam optimization algorithm.

[0025] As a preferred solution of the intelligent classification method based on multi-dimensional feature extraction of point cloud data described in the present invention, wherein: the original point cloud data is subjected to multi-scale processing, features are extracted, and multi-scale fusion high-dimensional feature vectors are generated by splicing and transmitted to the hybrid activation multi-layer perception network model. The specific steps are as follows:

[0026] The original point cloud data is processed at multiple scales, and geometric and physical features are extracted from the point cloud data at each scale to generate feature vectors at each scale.

[0027] Concatenate feature vectors of different scales to generate multi-scale fused high-dimensional feature vectors;

[0028] The multi-scale fused high-dimensional feature vector is input into the hybrid activation multi-layer perception network model for multi-dimensional feature processing, the multi-scale fused high-dimensional feature vector is compressed, and the classification result is output.

[0029] As a preferred solution of the intelligent classification method based on multi-dimensional feature extraction of point cloud data described in the present invention, the specific steps of introducing the attention mechanism and making classification decisions are as follows:

[0030] Introducing channel attention mechanism and spatial attention mechanism into the mixed activation multi-layer perception network model;

[0031] The channel attention mechanism is used to analyze different channels of the multi-scale fused high-dimensional feature vector, and the spatial attention mechanism is used to highlight the most important spatial region of the multi-scale fused high-dimensional feature vector to obtain a feature map.

[0032] The feature map is input into the hybrid activation multi-layer perceptron network model to compress the multi-scale fused high-dimensional feature vector into the same dimension as the number of categories in the category space to generate category probabilities;

[0033] A probability threshold is set, and the category whose category probability reaches the set threshold is selected as the final prediction result of the mixed activation multi-layer perception network model.

[0034] As a preferred solution of the intelligent classification method based on multi-dimensional feature extraction of point cloud data described in the present invention, the specific steps of introducing the channel attention mechanism and the spatial attention mechanism are as follows:

[0035] Input a multi-scale fused high-dimensional feature vector, perform global average pooling on each channel of the multi-scale fused high-dimensional feature vector, and obtain an average pooling value of each channel;

[0036] Perform global maximum pooling on each channel of the multi-scale fusion high-dimensional feature vector to obtain the maximum pooling value of each channel;

[0037] Concatenate the average pooling value and the maximum pooling value to obtain a fused pooling feature vector;

[0038] Input the fused pooled feature vector into the fully connected layer to generate channel attention weights;

[0039] The multi-scale fusion high-dimensional feature vector is weighted by using the channel attention weight to generate a weighted multi-scale fusion high-dimensional feature map;

[0040] Input weighted multi-scale fusion high-dimensional feature map and perform spatial dimensioning;

[0041] Perform global average pooling on the spatial dimension of the weighted multi-scale fusion high-dimensional feature map to generate a global average pooling spatial map;

[0042] Perform global maximum pooling on the spatial dimension of the weighted multi-scale fusion high-dimensional feature map to generate a global maximum pooling spatial map;

[0043] The global average pooling spatial map and the global maximum pooling spatial map are spliced ​​to generate a fused feature map;

[0044] Perform convolution operation on the fused feature map to obtain the spatial attention weight map;

[0045] The spatial attention weight map is used to adjust the weighted multi-scale fusion high-dimensional feature map to obtain the final weighted multi-scale fusion high-dimensional feature map.

[0046] In a second aspect, the present invention provides an intelligent classification system based on multi-dimensional feature extraction of point cloud data, comprising an acquisition and processing module, an adaptive fusion module, a construction module, a multi-scale processing module and a decision module; the acquisition and processing module is used to acquire point cloud data and pre-process the point cloud data; the adaptive fusion module is used to extract features based on the pre-processed point cloud data, and adaptively fuse the features using the relative change rate to generate a fused feature vector; the construction module is used to construct a hybrid activated multi-layer perceptual network model using the fused feature vector; the multi-scale processing module is used to perform multi-scale processing on the original point cloud data, extract features, and generate a multi-scale fused high-dimensional feature vector by splicing and transmit it to the hybrid activated multi-layer perceptual network model; the decision module is used to introduce an attention mechanism into the hybrid activated multi-layer perceptual network model to make classification decisions.

[0047] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the intelligent classification method based on multi-dimensional feature extraction of point cloud data as described in the first aspect of the present invention.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent classification method based on multi-dimensional feature extraction of point cloud data as described in the first aspect of the present invention.

[0049] The beneficial effects of the present invention are as follows: the present invention enhances the nonlinear expression ability of different features by combining the advantages of the two activation functions, ReLU and Tanh; ReLU can provide better gradient propagation when processing sparse data, while Tanh performs well when processing smooth nonlinear relationships; and the adaptive parameter α enables the network to flexibly adjust the combination ratio of the two activation functions according to different types of features, thereby improving the generalization ability of the hybrid activation multilayer perception network model; in addition, the combination of the cross entropy loss function and the Adam optimization algorithm enables the hybrid activation multilayer perception network model to converge quickly and improve the classification accuracy during the training process; the hybrid activation mechanism effectively solves the limitations of a single activation function, exhibits better classification performance when processing high-dimensional fused feature vectors, and improves the overall classification effect and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0051] Figure 1 This is a flow chart of the intelligent classification method based on multi-dimensional feature extraction of point cloud data in Example 1.

[0052] Figure 2 This is a module diagram of the intelligent classification system based on multi-dimensional feature extraction of point cloud data in Example 1. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0056] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides an intelligent classification method based on multi-dimensional feature extraction of point cloud data, comprising the following steps:

[0057] S1. Collect point cloud data and pre-process the point cloud data.

[0058] Furthermore, laser radar is used to collect 3D point cloud data; each point contains 3D coordinates (x, y, z) and additional attributes (such as reflection intensity); the collection should cover all angles of the target object to ensure the comprehensiveness and density of the data;

[0059] Statistical filtering is used for denoising, the distance from each point to its neighboring points is calculated, and those points whose distance from the neighboring points is greater than the set threshold are eliminated to ensure the smoothness and effectiveness of the data;

[0060] The number of points in point cloud data is usually very large, and direct processing will increase the computational overhead; downsampling is performed through voxel grid filters; voxel grid filters divide the point cloud into a fixed-size cubic grid (such as a voxel size of 0.05 meters) and use the centroid point within the voxel to represent all points in the voxel, thereby reducing the number of points while retaining the overall structure of the point cloud;

[0061] Eliminate points with too low reflection intensity in point cloud data. These points usually do not contain valuable information and may interfere with subsequent feature extraction and classification.

[0062] It should be noted that the point cloud data is fully collected and has undergone noise removal, downsampling and invalid data elimination, which significantly improves the validity and quality of the data; statistical filtering effectively removes noise points, and voxel grid filters reduce computational overhead while retaining the overall structure of the point cloud. Eliminating points with low reflection intensity ensures the validity of the data, ultimately making subsequent feature extraction and classification more efficient and accurate, and improving overall processing performance and robustness.

[0063] S2. Based on the preprocessed point cloud data, features are extracted and adaptively fused using the relative change rate of the features to generate a fused feature vector.

[0064] Furthermore, principal component analysis (PCA) is used to extract geometric features from 3D point cloud data. For each point, the covariance matrix is ​​calculated based on its neighborhood point set, and the normal vector and curvature of the point are calculated through the distribution of eigenvalues. The curvature reflects the local geometric shape characteristics of the point, such as flatness, convexity or concave.

[0065] Based on the reflection intensity in the 3D point cloud data, physical features are obtained; it reflects the physical properties of the object surface, such as material, color, smoothness, etc. The ability of the object surface to reflect the laser beam will affect the reflection intensity, which makes the reflection intensity an important physical feature of the object surface characteristics;

[0066] Through adaptation based on the relative rate of change of features, the extracted geometric features and physical features are fused to obtain the fused feature vector F f , the expression is:

[0067]

[0068] Where Z is the normalization factor, Z = Δ geo +Δ phy , ensuring the numerical stability of the fused feature vector, Δ geo is the relative rate of change of the geometric feature, Δ phy is the relative rate of change of the physical characteristic, f geo (i) is the i-th geometric feature (normal vector, curvature, etc.), f phy (j) is the jth physical feature (such as reflection intensity, etc.), i is the index of the geometric feature, j is the index of the physical feature, m is the total number of physical features, that is, the total number of physical features extracted from the point cloud data, and n is the total number of geometric features, that is, the total number of geometric features extracted from the point cloud data.

[0069] It should be noted that, assuming that the relative change rate of the geometric feature is 0.6, the relative change rate of the physical feature is 0.4, and there are three geometric features, f geo (1) = 1.2, f geo (2) = 1.5, f geo (3) = 1.3, the physical characteristics are 2, respectively f phy (1) = 0.8, f phy (2) = 0.9, the normalization factor is 1;

[0070] First, calculate the weight coefficient of the geometric feature:

[0071] Calculate the weighting coefficients for the physical features:

[0072] Then calculate the sum of geometric features:

[0073]

[0074] Calculate the sum of physical characteristics:

[0075] Substitute the formula to calculate the fused feature vector:

[0076] F f =1×(0.6×4+0.4×1.7)=3.08;

[0077] It should be noted that the geometric features of the point cloud data are extracted through principal component analysis, and the physical features (such as surface material and smoothness) are extracted based on the reflection intensity, and then the relative change rate of the geometric features and the physical features are used for adaptive fusion to generate a fused feature vector; this fusion comprehensively considers the local geometric shape and physical properties, ensures the accuracy and comprehensiveness of the feature expression, and improves the accuracy and robustness of subsequent classification.

[0078] S3. Use the fused feature vector to construct a hybrid activation multi-layer perception network model.

[0079] Furthermore, the input layer is defined as the fused feature vector, and the output layer is the prediction result;

[0080] Define multiple hidden layers; here we define three hidden layers;

[0081] The fused feature vector is input into the multi-layer hidden layer, and a linear transformation is performed to introduce nonlinear capabilities. The activation result f(x) is obtained by processing it through a mixed activation function. The expression is:

[0082] f(x)=α×ReLU(x)+(1-α)×Tanh(x);

[0083] Among them, f(x) is a weighted combination of ReLU and Tanh functions, α is an adaptive parameter with a value range of [0, 1], which is used to control the weight ratio of ReLU and Tanh functions in the combination, ReLU(x) is an activation function used to process input x; the function of ReLU is to set the negative part to 0 and keep the positive part unchanged, Tanh(x) is a hyperbolic tangent activation function, x is the input data, it is a smooth nonlinear function, Tanh function maps input x to the interval [0, 1], 1-α is the weight of Tanh function, indicating the contribution of Tanh in the combination, as α changes, 1-α will also change, thereby adjusting the influence of Tanh in the combination;

[0084] The second hidden layer receives the activation result of the first layer, generates a new feature vector, processes it again through the hybrid activation function, and the activation result is input to the third hidden layer;

[0085] The third hidden layer receives the activation result of the second layer, generates a new feature vector, and processes it through the mixed activation function. The activation result is used as the output result.

[0086] Based on the activation results, the activation results are mapped to the category space of the classification task through linear transformation;

[0087] Set the classification threshold, and select the category whose category score reaches the set threshold in the category space as the prediction result of the mixed activation multi-layer perception network model; the output layer generates a vector in which each element represents the score of a certain category, and through the Softmax function, select the category with the highest category score (the category that reaches the set threshold) as the final prediction result; threshold setting refers to setting a fixed probability value (such as 0.5) in the classification task. When the model predicts that the score of a certain category exceeds the threshold, it is used as the final prediction result. If no category reaches the threshold, the result is output by selecting the category with the maximum score or other processing mechanisms.

[0088] The cross entropy loss function is used to measure the difference between the prediction results of the mixed activation multi-layer perceptron model and the true category label; the cross entropy loss function compares the model's predicted probability with the true category label to calculate the model's current error;

[0089] The parameters of the mixed activation multi-layer perceptron network model are optimized by the Adam optimization algorithm. In order to minimize the cross entropy loss, we use the back propagation algorithm to calculate the parameter gradients of each layer. Then, the parameters of the network are updated according to these gradients using the Adam optimization algorithm, which is an adaptive gradient descent algorithm that can automatically adjust the learning rate to make the model converge faster.

[0090] It should be noted that by fusing the feature vector input into the multilayer perceptron with three hidden layers, the combination of ReLU and Tanh activation functions is controlled by adaptive parameters to enhance the nonlinear expression ability of the network. The multiple hidden layers gradually extract deep features, measure the prediction error by the cross entropy loss function, and adaptively adjust the learning rate in combination with the Adam optimization algorithm to accelerate the convergence of the model. This model can not only accurately capture complex features, but also improve the accuracy and robustness of classification.

[0091] S4. Perform multi-scale processing on the original point cloud data, extract features, generate multi-scale fusion high-dimensional feature vectors through splicing, and transmit them to the hybrid activation multi-layer perception network model.

[0092] In order to capture the global and local features of point cloud data, the original point cloud data is processed at multiple scales, and geometric and physical features are extracted from the point cloud data at each scale to generate feature vectors at each scale. Point clouds of different scales are generated by downsampling different voxel resolutions (such as 0.02 meters, 0.05 meters, and 0.1 meters). Smaller voxel sizes retain more detailed information, while larger voxel sizes reflect global information. This multi-scale processing helps capture features of different granularities. Feature vectors of each scale are generated by extracting geometric features such as normal vectors and curvature, as well as physical features such as reflection intensity.

[0093] The feature vectors of different scales are spliced ​​to generate a multi-scale fused high-dimensional feature vector; this spliced ​​multi-scale fused high-dimensional feature vector contains multi-scale information from micro to macro, and can more comprehensively describe the shape and material of the object;

[0094] The multi-scale fused high-dimensional feature vector is input into the hybrid activation multi-layer perception network model for multi-dimensional feature processing, the multi-scale fused high-dimensional feature vector is compressed, and the classification result is output by using the previously trained classification ability.

[0095] It should be noted that by downsampling the point cloud data at different resolutions through multi-scale processing, it is possible to capture multi-scale features from details to the global, and generate multi-scale feature vectors containing geometric features and physical features; these feature vectors are spliced ​​into multi-scale fused high-dimensional feature vectors to comprehensively describe the shape and material of the object; the spliced ​​multi-scale fused high-dimensional feature vectors are input into the mixed-activated multi-layer perception network, which effectively compresses the data through multi-dimensional feature processing and outputs the classification results, thereby improving the classification accuracy and the perception ability of multi-scale information.

[0096] S5. Introduce the attention mechanism into the mixed activation multi-layer perception network model to make classification decisions.

[0097] Furthermore, the channel attention mechanism and spatial attention mechanism are introduced into the mixed activation multi-layer perception network model;

[0098] The channel attention mechanism is used to analyze different channels of the multi-scale fused high-dimensional feature vector, and higher weights are automatically assigned to important channels. The weight of each channel is generated through global pooling operations (average pooling and maximum pooling). These weights are used to control the importance of each feature channel, and a weighted multi-scale fused high-dimensional feature map is obtained. The spatial attention mechanism is used to highlight the most important spatial area of ​​the multi-scale fused high-dimensional feature vector. The spatial attention generates spatial weights through convolution operations, so that the classification model can focus more on the key areas of the input data, thereby improving the classification accuracy and obtaining the final multi-scale fused high-dimensional feature map.

[0099] The multi-scale fusion high-dimensional feature map is input into the hybrid activation multi-layer perceptron network model, and the multi-scale fusion high-dimensional feature vector is compressed into the same dimension as the number of categories in the category space to generate category probabilities; the weighted multi-scale fusion high-dimensional feature map is input into the classifier, where a fully connected layer is used to compress the high-dimensional features into the same dimension as the number of categories; then the output of the fully connected layer is passed into the Softmax function to generate the probability distribution of the category, and the Softmax function converts the output value of the network into a probability value, which represents the predicted probability of each category;

[0100] Set a probability threshold and select the category whose category probability reaches the set threshold as the final prediction result of the mixed activation multi-layer perception network model; set a probability threshold, if the probability value of a category is greater than or equal to the threshold, then the category will be selected as the final prediction result; this threshold can ensure that the category output by the model has sufficient confidence and avoid selecting categories with too low probability; set it to a fixed value, usually between 0 and 1, for example, set the threshold to 0.5, indicating that only categories with category probability greater than or equal to 0.5 will be selected as the final prediction result; if no category has a probability of 0.5, other processing mechanisms may be triggered (such as selecting the category with the highest probability, or through further model correction processing).

[0101] S5.1. Introduce channel attention mechanism and spatial attention mechanism, specifically,

[0102] Input a multi-scale fused high-dimensional feature vector, perform global average pooling on each channel of the multi-scale fused high-dimensional feature vector, and obtain an average pooling value of each channel;

[0103] Perform global maximum pooling on each channel of the multi-scale fused high-dimensional feature vector to obtain the maximum pooling value of each channel and capture the most significant features of the channel;

[0104] The average pooling value and the maximum pooling value are concatenated to obtain a fused pooling feature vector; this vector contains the global statistical information (average and maximum value) of each channel, representing the importance of each channel;

[0105] Input the fused pooled feature vector into the fully connected layer to generate the attention weight for each channel;

[0106] The multi-scale fusion high-dimensional feature vector is weighted by using the channel attention weight to generate a weighted multi-scale fusion high-dimensional feature map;

[0107] Input weighted multi-scale fusion high-dimensional feature map and perform spatial dimensioning;

[0108] Perform global average pooling on the spatial dimension of the weighted multi-scale fusion high-dimensional feature map to generate a global average pooling spatial map;

[0109] Perform global maximum pooling on the spatial dimension of the weighted multi-scale fusion high-dimensional feature map to generate a global maximum pooling spatial map;

[0110] The global average pooling spatial map and the global maximum pooling spatial map are spliced ​​to generate a fused feature map;

[0111] Perform convolution operation on the fused feature map to obtain the spatial attention weight map;

[0112] The spatial attention weight map is used to adjust the weighted multi-scale fusion high-dimensional feature map (Hadamard product) to obtain the final weighted multi-scale fusion high-dimensional feature map, so that the hybrid activation multi-layer perception network model pays more attention to important spatial areas.

[0113] It should be noted that by introducing channel and spatial attention mechanisms, the mixed activation multi-layer perception network can automatically assign higher weights to important channels and key spatial areas; the channel attention mechanism captures the importance of each channel through global pooling, and the spatial attention mechanism focuses on key areas through convolution operations; this mechanism enhances the network's selective attention to multi-scale features, enabling the classification model to more accurately identify and classify complex scenes, improve classification accuracy and robustness, and output more confident prediction results.

[0114] The present embodiment also provides an intelligent classification system based on multi-dimensional feature extraction of point cloud data, including: an acquisition and processing module, an adaptive fusion module, a construction module, a multi-scale processing module and a decision module; the acquisition and processing module is used to acquire point cloud data and pre-process the point cloud data; the adaptive fusion module is used to extract features based on the pre-processed point cloud data, and adaptively fuse the features using the relative change rate of the features to generate a fused feature vector; the construction module is used to use the fused feature vector to construct a hybrid activated multi-layer perception network model; the multi-scale processing module is used to perform multi-scale processing on the original point cloud data, extract features, and generate a multi-scale fused high-dimensional feature vector by splicing and transmit it to the hybrid activated multi-layer perception network model; the decision module is used to introduce an attention mechanism into the hybrid activated multi-layer perception network model to make classification decisions.

[0115] This embodiment also provides a computer device, which is suitable for the case of an intelligent classification method based on multi-dimensional feature extraction of point cloud data, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent classification method based on multi-dimensional feature extraction of point cloud data as proposed in the above embodiment.

[0116] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0117] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the intelligent classification method based on multi-dimensional feature extraction of point cloud data proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0118] In summary, the present invention enhances the nonlinear expression ability of different features by combining the advantages of ReLU and Tanh activation functions; ReLU can provide better gradient propagation when processing sparse data, while Tanh performs well when processing smooth nonlinear relationships; and the adaptive parameter α, the network can flexibly adjust the combination ratio of the two activation functions according to different types of features, thereby improving the generalization ability of the hybrid activation multilayer perceptron model; in addition, the combination of the cross entropy loss function and the Adam optimization algorithm enables the hybrid activation multilayer perceptron model to converge quickly and improve the classification accuracy during the training process; the hybrid activation mechanism effectively solves the limitations of a single activation function, shows better classification performance when processing high-dimensional fused feature vectors, and improves the overall classification effect and robustness.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent classification method based on multi-dimensional feature extraction of point cloud data, characterized in that: include, Collect point cloud data and pre-process the point cloud data; Based on the preprocessed point cloud data, features are extracted and adaptively fused using the relative change rate of the features to generate a fused feature vector; Using the fused feature vectors, a hybrid activation multi-layer perception network model is constructed; Perform multi-scale processing on the original point cloud data, extract features, generate multi-scale fusion high-dimensional feature vectors through splicing, and transmit them to the hybrid activation multi-layer perception network model; Introducing an attention mechanism into the mixed-activation multilayer perceptron model to make classification decisions.

2. The intelligent classification method based on multi-dimensional feature extraction of point cloud data according to claim 1, characterized in that: The point cloud data is collected and preprocessed, and the specific steps are: Collect 3D point cloud data; Statistical filtering is used for denoising and downsampling is performed using a voxel grid filter.

3. The intelligent classification method based on multi-dimensional feature extraction of point cloud data according to claim 2, characterized in that: The extracting features and adaptively fusing the features using the relative change rate to generate a fused feature vector are specifically performed as follows: Use principal component analysis PCA to extract geometric features from 3D point cloud data; Based on the reflection intensity in the 3D point cloud data, physical features are obtained; Based on the adaptation of the relative change rate of features, the extracted geometric features and physical features are fused to obtain the fused feature vector F f , the expression is: Where Z is the normalization factor, Δ geo is the relative rate of change of the geometric feature, Δ phy is the relative rate of change of the physical characteristic, f geo (i) is the i-th geometric feature, f phy (j) is the jth physical feature, i is the index of the geometric feature, j is the index of the physical feature, m is the total number of physical features, and n is the total number of geometric features.

4. The intelligent classification method based on multi-dimensional feature extraction of point cloud data according to claim 3, characterized in that: The specific steps of constructing a hybrid activation multi-layer perception network model are as follows: Define the input layer as the fused feature vector, and the output layer as the prediction result; The fused feature vector is input into the multi-layer hidden layer, and a linear transformation is performed to introduce nonlinear capabilities. The activation result f(x) is obtained by processing it through a mixed activation function. The expression is: f(x)=α×ReLU(x)+(1-α)×Tanh(x); Among them, α is the adaptive parameter, ReLU(x) is the activation function, Tanh(x) is the hyperbolic tangent activation function, and x is the input data; Based on the activation results, the activation results are mapped to the category space of the classification task through linear transformation; Set a classification threshold, and select the category whose category score in the category space reaches the set threshold as the prediction result of the mixed activation multi-layer perception network model; The cross entropy loss function is used to measure the difference between the prediction results of the mixed activation multi-layer perceptron model and the true category label; The parameters of the mixed activation multilayer perception network model are optimized using the Adam optimization algorithm.

5. The intelligent classification method based on multi-dimensional feature extraction of point cloud data according to claim 4, characterized in that: The original point cloud data is processed at multiple scales, features are extracted, and multi-scale fusion high-dimensional feature vectors are generated by splicing and transmitted to the hybrid activation multi-layer perception network model. The specific steps are: The original point cloud data is processed at multiple scales, and geometric and physical features are extracted from the point cloud data at each scale to generate feature vectors at each scale. Concatenate feature vectors of different scales to generate multi-scale fused high-dimensional feature vectors; The multi-scale fused high-dimensional feature vector is input into the hybrid activation multi-layer perception network model for multi-dimensional feature processing, the multi-scale fused high-dimensional feature vector is compressed, and the classification result is output.

6. The intelligent classification method based on multi-dimensional feature extraction of point cloud data according to claim 5, characterized in that: The introduction of the attention mechanism to make classification decisions has the following specific steps: Introducing channel attention mechanism and spatial attention mechanism into the mixed activation multi-layer perception network model; The channel attention mechanism is used to analyze different channels of the multi-scale fused high-dimensional feature vector, and the spatial attention mechanism is used to highlight the most important spatial region of the multi-scale fused high-dimensional feature vector to obtain a feature map. The feature map is input into the hybrid activation multi-layer perceptron network model to compress the multi-scale fused high-dimensional feature vector into the same dimension as the number of categories in the category space to generate category probabilities; A probability threshold is set, and the category whose category probability reaches the set threshold is selected as the final prediction result of the mixed activation multi-layer perception network model.

7. The intelligent classification method based on multi-dimensional feature extraction of point cloud data according to claim 6, characterized in that: The specific steps of introducing the channel attention mechanism and the spatial attention mechanism are as follows: Input a multi-scale fused high-dimensional feature vector, perform global average pooling on each channel of the multi-scale fused high-dimensional feature vector, and obtain an average pooling value of each channel; Perform global maximum pooling on each channel of the multi-scale fusion high-dimensional feature vector to obtain the maximum pooling value of each channel; Concatenate the average pooling value and the maximum pooling value to obtain a fused pooling feature vector; Input the fused pooled feature vector into the fully connected layer to generate channel attention weights; The multi-scale fusion high-dimensional feature vector is weighted by using the channel attention weight to generate a weighted multi-scale fusion high-dimensional feature map; Input weighted multi-scale fusion high-dimensional feature map and perform spatial dimensioning; Perform global average pooling on the spatial dimension of the weighted multi-scale fusion high-dimensional feature map to generate a global average pooling spatial map; Perform global maximum pooling on the spatial dimension of the weighted multi-scale fusion high-dimensional feature map to generate a global maximum pooling spatial map; The global average pooling spatial map and the global maximum pooling spatial map are spliced ​​to generate a fused feature map; Perform convolution operation on the fused feature map to obtain the spatial attention weight map; The spatial attention weight map is used to adjust the weighted multi-scale fusion high-dimensional feature map to obtain the final weighted multi-scale fusion high-dimensional feature map.

8. An intelligent classification system based on multi-dimensional feature extraction of point cloud data, based on the intelligent classification method based on multi-dimensional feature extraction of point cloud data according to any one of claims 1 to 7, characterized in that: Including, acquisition and processing module, adaptive fusion module, construction module, multi-scale processing module and decision module; The acquisition and processing module is used to acquire point cloud data and pre-process the point cloud data; The adaptive fusion module is used to extract features based on the preprocessed point cloud data, and adaptively fuse the features using the relative change rate to generate a fused feature vector; The construction module is used to construct a mixed activation multi-layer perception network model using the fused feature vectors; The multi-scale processing module is used to perform multi-scale processing on the original point cloud data, extract features, generate multi-scale fusion high-dimensional feature vectors by splicing, and transmit them to the hybrid activation multi-layer perception network model; The decision module is used to introduce an attention mechanism into the mixed activation multi-layer perception network model to make classification decisions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent classification method based on multi-dimensional feature extraction of point cloud data as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent classification method based on multi-dimensional feature extraction of point cloud data as described in any one of claims 1 to 7 are implemented.

Images