A method for registering aerospace components and electronic equipment

Abstract

This application provides a registration method and electronic device for aerospace components. The method includes the following steps: preprocessing the original point cloud of the aerospace component to obtain a preprocessed point cloud; inputting the preprocessed point cloud into an encoder and a decoder to obtain a coarse point cloud; inputting the preprocessed point cloud and the coarse point cloud into a generator to obtain a complete point cloud; and registering the aerospace component based on the complete point cloud. This application solves the problem of missing point clouds by using a self-supervised learning framework and an attention mechanism, ensuring that the input point cloud is more complete, thereby significantly improving registration accuracy. Furthermore, this application combines multiple features and achieves multi-feature fusion through weighted summation, further improving the accuracy and reliability of registration. Simultaneously, through reasonable module design and optimization, the computational complexity is reduced, making this application more efficient when processing large-scale point clouds.

Description

Technical Field

[0001] This invention relates to the field of aerospace technology, and in particular to a method for registering aerospace components and an electronic device. Background Technology

[0002] With the continuous development of aerospace technology, the requirements for precision and reliability in the manufacturing and assembly of large aerospace components are becoming increasingly stringent. Traditional component registration methods typically rely on complete geometric information and complex measurement equipment. However, in practice, acquiring complete point cloud data often faces numerous challenges, such as sensor viewing angle limitations, occlusion, and data noise. These factors make high-precision component registration in the presence of missing information a pressing problem. Existing point cloud registration methods mainly rely on geometric features and topological structures for registration. The Iterative Closest Point (ICP) algorithm achieves registration by iteratively minimizing the distance between two point clouds, but its registration effect is often unsatisfactory when dealing with partially missing point cloud data. The Normal Distributions Transform (NDT) algorithm represents the geometric structure of point clouds by constructing a probability distribution. While it can handle partially missing data to some extent, its computational complexity is high, and its effectiveness is limited when facing complex scenarios. Summary of the Invention

[0003] This application provides a method and electronic device for aerospace component registration. By using self-supervised learning and attention mechanisms, it innovatively solves the problem of missing point cloud information. By combining multiple features such as edge curvature similarity and contour matching degree, it achieves high-precision and low-computational-complexity aerospace component registration.

[0004] This application provides a method for registering aerospace components, comprising the following steps: preprocessing the original point cloud of the aerospace component to obtain a preprocessed point cloud; inputting the preprocessed point cloud into an encoder and a decoder to obtain a coarse point cloud; inputting the preprocessed point cloud and the coarse point cloud into a generator to obtain a complete point cloud; and registering the aerospace component based on the complete point cloud; wherein the original point cloud is a residual point cloud, and the generator includes at least one attention feature enhancement layer.

[0005] In one embodiment, the preprocessing of the original point cloud of the aerospace component to obtain a preprocessed point cloud includes: sampling the farthest point of the original point cloud to obtain a sampled point cloud; normalizing the sampled point cloud to obtain a normalized point cloud; and dividing the normalized point cloud into regions to obtain a preprocessed point cloud.

[0006] In one embodiment, the step of sampling the farthest point from the original point cloud to obtain a sampled point cloud includes: arbitrarily selecting one or more point clouds from the original point cloud as an initial sampled point cloud set, and the remaining original point clouds as an initial unsampled point cloud set; for each point cloud in the current unsampled point cloud set, calculating the minimum distance from the unsampled point cloud to all point clouds in the current sampled point cloud set, and adding the unsampled point cloud corresponding to the maximum value of the minimum distance to the current sampled point cloud set; when the number of point clouds in the current sampled point cloud set reaches a preset threshold, using the point clouds in the current sampled point cloud set as the sampled point cloud.

[0007] In one embodiment, the step of dividing the normalized point cloud into regions to obtain a preprocessed point cloud includes: randomly initializing M center points based on the normalized point cloud; extracting neighborhood regions for each of the M center points based on the K-nearest neighbor algorithm, and dividing the normalized point cloud into M parts; further dividing each of the M parts into a first part, a second part, and a third part; and using the first part, the second part, and the third part as the preprocessed point cloud; wherein the first part is an encoder-observable region, the second part is a masked region, and the third part is an encoder-unobservable region.

[0008] In one embodiment, the step of inputting the preprocessed point cloud into an encoder and a decoder to obtain a coarse point cloud includes: inputting the preprocessed point cloud into an encoder to extract important feature information of the point cloud; and inputting the important feature information of the point cloud into a decoder to obtain the coarse point cloud.

[0009] In one embodiment, the encoder includes a first shared multilayer perceptron layer, a first max pooling layer, an extended layer, a second shared multilayer perceptron layer, and a second max pooling layer. The step of inputting the preprocessed point cloud into the encoder to extract important feature information of the point cloud includes: inputting the preprocessed point cloud into the first shared multilayer perceptron layer, performing linear transformation and nonlinear activation on each point cloud to extract shallow features; inputting the shallow features into the first max pooling layer to extract the maximum value of each point cloud in the feature dimension to obtain salient features; inputting the salient features into the extended layer to obtain extended features; wherein the extended features have the same feature dimension as the shallow features; inputting the shallow features and extended features into the second shared multilayer perceptron layer to obtain comprehensive features; and inputting the comprehensive features into the second max pooling layer to extract important feature information of the point cloud.

[0010] In one embodiment, the encoder includes a third shared multilayer perceptron layer and a reshaping layer; the step of inputting the important feature information of the point cloud into the decoder to obtain the coarse point cloud includes: inputting the important feature information of the point cloud into the third shared multilayer perceptron layer to extract decoded features; inputting the decoded features into the reshaping layer to convert the decoded features into features of a preset shape and format to obtain the coarse point cloud; wherein, the coarse point cloud includes at least the main geometric features of the original point cloud.

[0011] In one embodiment, the generator includes: a first multilayer perceptron layer, a second multilayer perceptron layer, three attention feature enhancement layers, and a third multilayer perceptron layer; the step of inputting the preprocessed point cloud and the coarse point cloud into the generator to obtain a complete point cloud includes: inputting the preprocessed point cloud and the coarse point cloud together into the first multilayer perceptron layer to extract first multi-level features; inputting the important feature information of the point cloud into the second multilayer perceptron layer to extract second multi-level features; superimposing the first multi-level features and the second multi-level features, and sequentially inputting them into the three attention feature enhancement layers to obtain attention features; superimposing the multi-level features and the attention features, and inputting them into the third multilayer perceptron layer to obtain the complete point cloud.

[0012] In one embodiment, registering the aerospace component based on the complete point cloud includes: obtaining the edge curvature similarity and contour matching degree of the aerospace component based on the complete point cloud; obtaining the multi-feature registration result of the aerospace component by weighted fusion based on the curvature similarity and contour matching degree; and registering the aerospace component when the multi-feature registration result meets preset conditions.

[0013] This application also provides an electronic device, which includes: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the above-described aerospace component registration method.

[0014] The solution provided in the above embodiments of this application innovatively solves the problem of missing point clouds by using a self-supervised learning framework and attention mechanism, ensuring that the input point cloud is more complete, thereby significantly improving the registration accuracy. Furthermore, this application combines multiple features (such as edge curvature similarity and contour matching degree) and achieves multi-feature fusion through weighted summation, further improving the accuracy and reliability of registration. Simultaneously, through reasonable module design and optimization, computational complexity is reduced, making this application more efficient in processing large-scale point clouds. Compared with traditional methods, this application can not only adaptively extract important feature regions, but also comprehensively consider the importance of each feature through a multi-feature fusion strategy, thereby achieving high-precision, low-computational-complexity aerospace component registration. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below.

[0016] Figure 1 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application;

[0017] Figure 2 This is a schematic flowchart of a space component registration method provided in an embodiment of this application;

[0018] Figure 3 This is a flowchart illustrating another aerospace component registration method provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0020] Similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] As mentioned above, achieving high-precision registration of aerospace components has become a pressing issue. In recent years, machine learning-based point cloud completion methods have gained attention. Deep learning models can learn and reconstruct complete point cloud information from partial point cloud data, thereby improving registration accuracy. In particular, self-supervised learning paradigms have demonstrated significant potential in point cloud completion. This method does not require complete point cloud information and is suitable for scenarios where obtaining complete point cloud information for large-sized aerospace components is difficult. By combining an encoder-decoder architecture, it is possible to efficiently learn and extract various features from incomplete point cloud information, thus achieving point cloud completion. However, existing self-supervised learning methods still have limitations when handling the registration of large aerospace components with high precision requirements, such as insufficient reconstruction accuracy and limited ability to capture complex geometries. Furthermore, attention mechanisms, as an important research direction in deep learning in recent years, have performed exceptionally well, especially in tasks requiring the capture of long-distance dependencies. Introducing attention mechanisms into point cloud processing can effectively reveal the fine geometric features of 3D shapes and enhance the understanding of spatial relationships between points, thus supporting high-precision point cloud registration. However, research on applying attention mechanisms to point cloud registration is still in its early stages, and its specific application effects and optimization strategies require further exploration. Combining self-supervised learning frameworks with attention mechanisms to reconstruct point clouds and thus improve registration accuracy is currently a key research focus.

[0022] To address the aforementioned problems, this application provides at least one aerospace component registration method, comprising the following steps: preprocessing the original point cloud of the aerospace component to obtain a preprocessed point cloud; inputting the preprocessed point cloud into an encoder and a decoder to obtain a coarse point cloud; inputting the preprocessed point cloud and the coarse point cloud into a generator to obtain a complete point cloud; and registering the aerospace component based on the complete point cloud; wherein the original point cloud is a partial point cloud, and the generator includes at least one attention feature enhancement module. This application integrates a self-supervised learning framework and an attention enhancement module, innovatively solving the problem of missing complete point cloud data in the field of large aerospace components, thereby improving the reliability and accuracy of large aerospace component registration.

[0023] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 100 is used to execute the aerospace component registration method provided in an embodiment of this application. Figure 1 As shown, the electronic device 100 includes: one or more processors 102 and one or more memories 104 storing processor-executable instructions. The processors 102 are configured to execute the spacecraft component registration method provided in the following embodiments of this application.

[0024] It is important to note that Figure 1 The electronic device described is only one possible device structure, and other possible device structures also exist. This application does not impose any restrictions.

[0025] The processor 102 may be a gateway, a smart terminal, or a device that includes a central processing unit (CPU), a graphics processing unit (GPU), or other forms of processing units with data processing capabilities and / or instruction execution capabilities. It can process data from other components in the electronic device 100 and control other components in the electronic device 100 to perform desired functions.

[0026] The memory 104 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 102 may execute the program instructions to implement the aerospace component registration method described below. Various application programs and various data, such as various data used and / or generated by the application programs, may also be stored in the computer-readable storage medium.

[0027] In one embodiment, Figure 1 The illustrated electronic device 100 may further include an input device 106, an output device 108, and a data acquisition device 110, these components being interconnected via a bus system 112 and / or other forms of connection mechanisms (not shown). It should be noted that... Figure 1 The components and structure of the electronic device 100 shown are merely exemplary and not limiting; the electronic device 100 may also have other components and structures as needed.

[0028] The input device 106 may be a device used by a user to input commands, and may include one or more of a keyboard, mouse, microphone, and touchscreen. The output device 108 may output various information (e.g., images or sounds) to the outside (e.g., a user), and may include one or more of a display, speaker, etc. The data acquisition device 110 may acquire raw point clouds of aerospace components. For example, the data acquisition device 110 may be an electromagnetic ultrasonic thickness gauge.

[0029] In one embodiment, the components in the example electronic device 100 used to implement the aerospace component registration method of the present application embodiment can be integrated or distributed. For example, the processor 102, memory 104, input device 106 and output device 108 can be integrated into one unit, while the data acquisition device 110 can be separated.

[0030] In one embodiment, the example electronic device 100 used to implement the aerospace component registration method of the present application embodiment can be implemented as an artificial intelligence (AI) server, a supercomputer, etc.

[0031] Figure 2 This is a flowchart illustrating a spacecraft component registration method according to an embodiment of this application. The method is executed by a spacecraft component registration system, which includes an encoder, a decoder, a generator, and a multi-feature registration module.

[0032] like Figure 2 As shown, the aerospace component registration method includes the following steps 201-204.

[0033] Step 201: Preprocess the original point cloud of the aerospace component to obtain a preprocessed point cloud.

[0034] Regarding aerospace components, these can be parts of various spacecraft (such as rockets), for example, they could be the petals of a spacecraft.

[0035] The original point cloud of aerospace components can be obtained by measuring and collecting data from the aerospace components using an electromagnetic ultrasonic thickness gauge. In this embodiment, the point cloud is also referred to as point cloud data or point cloud information. The original point cloud comes from different aerospace components, which will be assembled together later. The original point cloud includes at least the three-dimensional coordinates (x, y, z) of points on the surface of the measured aerospace component, the main geometric features of the measured aerospace component, and the detailed features of the measured aerospace component. The main features may be, for example, the outline and surface shape of the aerospace component, and the detailed features may be, for example, the curvature of the non-edge parts of the aerospace component. Since it is difficult to obtain a complete point cloud of aerospace components, especially large aerospace components, through measurement, the original point cloud is a partial point cloud. Steps 201-203 in this embodiment are the process of transforming the partial point cloud into a complete point cloud through a self-supervised learning framework and attention mechanism.

[0036] Preprocessed point clouds are the point clouds obtained after a series of processing steps on the original point clouds, and they serve as the input to subsequent models (also known as "networks"). The model combines a self-supervised learning framework and an attention mechanism, specifically including an encoder, decoder, and generator.

[0037] Step 201 specifically includes: sampling the farthest point of the original point cloud to obtain a sampled point cloud; normalizing the sampled point cloud to obtain a normalized point cloud; and dividing the normalized point cloud into regions to obtain a preprocessed point cloud.

[0038] The process of sampling the original point cloud to obtain a sampled point cloud specifically includes: arbitrarily selecting one or more point clouds from the original point cloud as an initial sampled point cloud set, and the remaining original point clouds as an initial unsampled point cloud set; for each point cloud in the current unsampled point cloud set, calculating the minimum distance from the unsampled point cloud to all point clouds in the current sampled point cloud set, and adding the unsampled point cloud corresponding to the maximum value of the minimum distance to the current sampled point cloud set; when the number of point clouds in the current sampled point cloud set reaches a preset threshold, the point clouds in the current sampled point cloud set are used as the sampled point cloud. It can be understood that the initial sampled point cloud set is the first current sampled point cloud set, and the initial unsampled point cloud set is the first current unsampled point cloud set.

[0039] The normalized point cloud is divided into regions to obtain a preprocessed point cloud. Specifically, this includes: randomly initializing M center points based on the normalized point cloud; extracting neighborhood regions for each of the M center points based on the K-nearest neighbor algorithm, and dividing the normalized point cloud into M parts; further dividing each of the M parts into a first part, a second part, and a third part; and using the first part, the second part, and the third part as the preprocessed point cloud; wherein the first part is the region observable by the encoder, the second part is the masked region, and the third part is the region unobservable by the encoder.

[0040] In one embodiment, step 201 is implemented as follows:

[0041] First, the original point cloud is sampled for the farthest point. The minimum distance between the unsampled point cloud and the sampled point cloud is continuously retained. The point corresponding to the maximum value among these minimum distances is selected as a new sampled point cloud and added to the sampled point cloud set. This process continues until a sufficient number of points have been sampled, ensuring uniform sampling as much as possible. The specific mathematical expression is as follows:

[0042]

[0043] Where Q is the set of unsampled point clouds, i.e., the unsampled point cloud set, and P is the set of sampled point clouds, i.e., the sampled point cloud set. i p is the unsampled point cloud, and p is the sampled point cloud. next This is the point cloud sampled next. argmax represents the independent variable that maximizes the objective function.

[0044] Then, the sampled point cloud is normalized, with the range set to [-1, 1], resulting in a normalized point cloud. This helps the model remain stable during training. The specific mathematical expression is as follows:

[0045]

[0046] x, y, and z are the coordinates of the sampled point cloud, respectively; x center y center , z center x represents the average coordinates of all sampled point clouds; max y max , z max These represent the maximum coordinates of the sampled point cloud; x min y min , z min These are the minimum coordinates of the sampled point cloud.

[0047] Next, based on the normalized point cloud, M center points are randomly initialized. The K-Nearest Neighbors (KNN) algorithm is used to extract neighborhood regions for each center point, thus segmenting the point cloud into M parts. Each part represents different local feature information of the point cloud, and each local region can capture different geometric shapes and texture features, helping the network understand the structural information of various locations on the object's surface. Each of the M parts is further divided into three parts: Part 1, Part 2, and Part 3. Part 1 P obs Representing the observable region of the network, the predicted coarse point cloud P is forced to be output by the decoder. c Align with this region. To further standardize point cloud completion, P is introduced. mis , representing the masked area that needs to be predicted, encouraging the network to fill in areas that are intentionally masked and naturally missing. The third group P latent Including segments not observed by the encoder allows for additional regularization. Effective point cloud region segmentation ensures the network can learn diverse feature information from the object's surface, thereby improving performance on the completion task. Note that the area ratio between the first, second, and third parts can be 20:40:4.

[0048] Finally, the first, second, and third parts are used as the preprocessed point cloud.

[0049] Step 202: Input the preprocessed point cloud into the encoder and decoder to obtain a coarse point cloud.

[0050] The encoder may include a first shared multilayer perceptron layer (ShareMLP), a first point-wise maxpool layer, an expansion layer, a second shared multilayer perceptron layer, and a second maxpool layer.

[0051] Regarding the decoder, it may include a third shared multilayer perceptron layer and a reshape layer.

[0052] Rough point clouds retain the main geometric features of the original point cloud, such as the outline of components and surface shape, but with reduced detail.

[0053] Step 202 specifically includes: inputting the preprocessed point cloud into the encoder to extract important feature information of the point cloud; inputting the important feature information of the point cloud into the decoder to obtain the coarse point cloud.

[0054] The preprocessed point cloud is input into an encoder to extract important feature information. Specifically, this includes: inputting the preprocessed point cloud into a first shared multilayer perceptron layer, performing linear transformation and nonlinear activation on each point cloud to extract shallow features; inputting the shallow features into a first max pooling layer to extract the maximum value of each point cloud in the feature dimension to obtain salient features; inputting the salient features into an extension layer to obtain extended features; wherein the extended features have the same feature dimension as the shallow features; inputting the shallow features and extended features into a second shared multilayer perceptron layer to obtain comprehensive features; and inputting the comprehensive features into a second max pooling layer to extract important feature information of the point cloud.

[0055] The process of inputting the important feature information of the point cloud into the decoder to obtain the coarse point cloud specifically includes: inputting the important feature information of the point cloud into the third shared multilayer perceptron layer to extract the decoded features; and inputting the decoded features into the reshaping layer to transform the decoded features into features of a preset shape and format to obtain the coarse point cloud.

[0056] Step 203: Input the preprocessed point cloud and coarse point cloud into the generator to obtain the complete point cloud.

[0057] The generator can include a first multilayer perceptron (MLP) layer, a second multilayer perceptron layer, three attention feature augmentation layers, and a third multilayer perceptron layer.

[0058] A complete point cloud is obtained by processing the original point cloud (i.e., the incomplete point cloud) using a model. It is a point cloud predicted by the model. Compared to the original point cloud, the complete point cloud can represent the features of aerospace components (such as contour features and surface features) in more detail, so as to complete the subsequent component registration operation more accurately.

[0059] Step 203 specifically includes: inputting the preprocessed point cloud and the coarse point cloud together into the first multilayer perceptron layer to extract the first multilevel features; inputting the important feature information of the point cloud into the second multilayer perceptron layer to extract the second multilevel features; superimposing the first multilevel features and the second multilevel features, and inputting them sequentially into three attention feature enhancement layers to obtain attention features; superimposing the multilevel features and the attention features, and inputting them into the third multilayer perceptron layer to obtain the complete point cloud.

[0060] Step 204: Register the aerospace components based on the complete point cloud.

[0061] Regarding registration, it can involve assembling aerospace components, such as assembling multiple melon lobes. As mentioned earlier, the original point cloud comes from different aerospace components; therefore, the complete point cloud obtained from the original point cloud also comes from different aerospace components. These aerospace components will be assembled in step 204. By using the complete point clouds of these aerospace components, the edge curvature similarity and contour matching degree between the aerospace components to be assembled can be obtained, and then the final registration can be completed based on these two features.

[0062] Step 204 specifically includes: obtaining the edge curvature similarity and contour matching degree of the aerospace component based on the complete point cloud; obtaining the multi-feature registration result of the aerospace component by weighted fusion based on the curvature similarity and contour matching degree; and registering the aerospace component when the multi-feature registration result meets the preset conditions.

[0063] Figure 3 This is a flowchart illustrating another aerospace component registration method according to an embodiment of this application. The method is executed by an aerospace component registration system, which includes an encoder, a decoder, a generator, and a multi-feature registration module. The method specifically includes the following steps:

[0064] Step 301: Measure and collect data on multiple aerospace components to be assembled using an electromagnetic ultrasonic thickness measuring device to obtain the original point cloud.

[0065] Multiple aerospace components to be assembled can be multiple segments of a rocket propellant tank, which need to be assembled, or registered.

[0066] Step 302: Preprocess the original point cloud to obtain a preprocessed point cloud.

[0067] The preprocessing of the original point cloud is the same as in step 201, and will not be repeated here.

[0068] Step 303: Input the preprocessed point cloud into the encoder to extract important feature information of the point cloud.

[0069] Step 303 may specifically include the following steps:

[0070] Step 3031: The preprocessed point cloud enters the first shared multilayer perceptron layer (ShareMLP1), and linear transformation and nonlinear activation are performed on each point to extract shallow features.

[0071] Step 3032: Input the shallow features into the first max pooling layer (Point-wise Maxpool1), perform max pooling on the features of each point, extract the maximum value of each point in the feature dimension, reduce the feature dimension, and at the same time retain the key feature information.

[0072] Step 3033: The pooled features are restored to the same feature dimension as the original point cloud through the Expand layer to ensure the spatial consistency of the features, so that each point can maintain the feature representation of its original position.

[0073] Step 3034: The shallow features and the expanded features are processed again through the shared multilayer perceptron layer (i.e., the input first shared multilayer perceptron layer (ShareMLP2)) to further extract features.

[0074] Step 3035: Perform point-wise max pooling again (that is, the output of step 3034 will be input into the second max pooling layer (Point-wise Maxpool2)) to extract important feature information of the point cloud.

[0075] Step 304: Input the important feature information of the point cloud into the decoder to obtain the coarse point cloud.

[0076] Step 304 may specifically include the following steps:

[0077] Step 3041: The feature vector output by the encoder enters the third shared multilayer perceptron layer (ShareMLP3) of the decoder for further processing and feature extraction;

[0078] Step 3042: Finally, the decoded feature vectors are transformed into the required shape and format through the reshape layer to obtain the coarse point cloud.

[0079] Step 305: Input the preprocessed point cloud and coarse point cloud into the generator to obtain the complete point cloud.

[0080] Step 305 may specifically include the following steps:

[0081] Step 3051: The coarse point cloud output from the decoder and the preprocessed point cloud are input together into the first multilayer perceptron layer (MLP1) to extract multi-level features. Simultaneously, the important feature information of the point cloud extracted in step 303 is further extracted using a separate second multilayer perceptron layer (MLP2). The feature matrices are combined by point-by-point stacking to obtain a fused feature matrix, ensuring that each point feature retains both local details and the global geometric structure of its shape. Local details refer to the geometric and shape features within a small area of ​​each point in the point cloud.

[0082] Step 3052: The fused feature matrix obtained after superposition is further fed into three attention feature enhancement layers (Attention Feature Augmentation1, Attention Feature Augmentation2, Attention Feature Augmentation3) to gradually upsample point features, capture details and global relationships in point cloud data. Each module dynamically weights the features through a self-attention mechanism to enhance important information and suppress irrelevant interference, and finally obtains attention features.

[0083] The specific implementation of step 3052 may include the following steps:

[0084] Step 30521: Initialize the query, key, and value matrix (Q, K, V);

[0085] Step 30522: Calculate the score for each attention head. For each head, the dot product of the query matrix Q and the key matrix K is calculated as the attention score. The specific mathematical expression is:

[0086]

[0087] Where, d k / h is a scaling factor to prevent the dot product value from becoming too large.

[0088] A softmax function is applied to the attention scores to obtain attention weights, which determine the importance of each value in the weighted summation. Then, the values ​​are weighted and summed using these attention weights to obtain the output of each head. The outputs of all attention heads are concatenated to form a new feature representation, and a linear transformation is used to generate the final multi-head attention output.

[0089] Step 30523: Based on the multi-head attention output, residual connections are used to add input features, directly passing the original input information to the next layer to avoid the gradient vanishing problem and help optimize the model. The residual results are calculated as follows:

[0090] F residual = Norm(Q+MultiHead(Q,K,V))

[0091] Where Q = XW Qu K = XW Ku V = XW Vu X is the input feature matrix, W is the weight matrix, and Qu, Ku, and Vu are used to calculate the weight matrices for the query, key, and value.

[0092] Step 30524: The residual results are input into the feedforward neural network. The feedforward neural network consists of two linear transformations and a ReLU activation function, using skip connections to enable the model to learn representations more effectively. The specific form of the feedforward neural network is as follows:

[0093] FFN(x) = max(0, xW1+b1)W2+b2

[0094] Where W1, W2, b1, and b2 are the parameters of the feedforward neural network, and x is the input of the feedforward neural network.

[0095] The specific implementation method for step 30524 is as follows:

[0096] Step 305241: Save the input point cloud feature vector for use in subsequent skip connections to ensure that the original point cloud information is not lost during network processing.

[0097] Step 305242: Perform layer normalization to adjust the mean and variance of the input to make it more suitable for subsequent network processing, thereby improving the stability and training speed of the model.

[0098] Step 305243: The normalized tensor is mapped to a higher-dimensional space through a linear mapping to capture more feature information.

[0099] Step 305244: Apply the GELU (Gaussian Error LinearUnit) activation function to the tensor after mapping to the high-dimensional space to enhance the ability to represent nonlinear features.

[0100] Step 305245: Map the activated high-dimensional feature vector back to the original feature dimension through a linear mapping.

[0101] Step 305246: Add the result obtained in the previous step to the previously saved input point cloud feature vector using a residual connection. Residual connections can alleviate the vanishing gradient problem and help preserve information from the original input. This summation result is then returned as the output of the feedforward neural network.

[0102] Step 30525: After processing by the feedforward network, residual connections are used again to add the output of the feedforward network to the normalized input in order to maintain the flow of information; then, layer normalization is applied to the summed result to help stabilize the training process and reduce the risk of gradient vanishing or exploding.

[0103] F output =LayerNorm(F FFN +F residual )

[0104] Step 30526: Through the backpropagation process, update the weights of the query and key according to the attention weight matrix to finally obtain the attention features.

[0105] Step 3053: Apply the third multilayer perceptron layer (MLP3) to the integrated and superimposed features of the multi-level features and the output features of the attention enhancement layer (i.e., attention features) to generate the final predicted complete point cloud, thereby realizing the completion and refinement of point cloud information.

[0106] Step 306: Register the aerospace components to be assembled based on the complete point cloud.

[0107] Step 306 may specifically include the following steps:

[0108] Step 3061: Calculate the curvature of the complete point cloud, obtain the curvature value in the neighborhood of each point, obtain the local shape information of the point cloud, and then construct a curvature map of the point cloud. Output the point cloud edge curvature similarity results between different aerospace components to be assembled. The specific process is as follows:

[0109] Step 30611: Calculating the normal vector of each point in the complete point cloud is a prerequisite for curvature calculation. The normal vector of each point is estimated using PCA (Principal Component Analysis) based on the covariance matrix of its neighboring points.

[0110] Step 30612: Select a neighborhood for each point and use the K-nearest neighbor method (with K neighborhood points) to calculate the average degree of change in the surface normal vector of the neighborhood, which is suitable for describing the curvature of the surface.

[0111] Step 30613: Construct a curvature map based on the curvature values ​​of the point cloud to display the curvature characteristics of each point in the point cloud and reflect the local shape changes of the point cloud.

[0112] Step 30614: The Euclidean distance metric is used to quantify the similarity of curvature distributions. The sum of squares of the curvature value differences between each pair of corresponding point clouds (corresponding points are on two aerospace components to be assembled) is calculated to obtain the total similarity score. The final edge curvature similarity calculation formula is as follows:

[0113]

[0114] C1 and C2 represent the curvature values ​​of different point clouds, which are located on two aerospace components to be assembled.

[0115] Step 3062: Extract contours using the geometric features of the complete point cloud data, and evaluate the contour matching degree between the different aerospace components to be assembled. The specific process is as follows:

[0116] Step 30621: Project the point cloud onto a two-dimensional plane. Select the XY plane as the optimal projection plane. The projected two-dimensional representation can help simplify the processing and highlight the contour information of the point cloud.

[0117] Step 30622: On the projected point cloud, perform edge detection technology to extract the contour and represent it as a set of points with important geometric information.

[0118] Step 30623: Based on the calculation principle of Hu moments, a similarity value is returned by comparing the shape features of the contours. The smaller the value, the more similar the contours are. The contour matching degree between the two point clouds is quantitatively evaluated, thereby achieving accurate point cloud registration.

[0119] Step 3063: Adopt a multi-feature-based matching strategy, comprehensively consider edge curvature similarity and contour matching features, and obtain multi-feature registration results by weighted fusion of different features, thereby improving matching accuracy.

[0120] The multi-feature registration result can be the registration accuracy of two aerospace components to be assembled. The specific calculation formula can be:

[0121] (Edge curvature similarity × Weight of edge curvature similarity + Contour matching degree × Weight of contour matching degree) × 100%

[0122] For example, if the focus is on the matching degree of the aerospace component contours, the weights of edge curvature similarity and contour matching degree can be set to 0.3:0.7, and the multi-feature registration result can be obtained.

[0123] Step 3064: When the multi-feature registration results meet the preset conditions, register the aerospace components.

[0124] When the registration accuracy of two aerospace components to be assembled is greater than the preset accuracy, it indicates that the two components meet the requirements for assembly and can be precisely assembled together. At this point, the registration operation can be performed on the two aerospace components.

[0125] The above embodiments integrate a self-supervised learning framework and an attention enhancement module, innovatively solving the problem of missing point cloud data for large aerospace components. By combining multiple features, including edge curvature and contour matching, multi-feature benchmark registration is achieved through weighted fusion, which significantly improves the accuracy of registration. At the same time, the reasonable module design and optimization reduce the computational complexity, making this application more efficient in processing large-scale point cloud data.

[0126] The devices and methods disclosed in the several embodiments provided in this application can also be implemented in other ways. The device and method embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0127] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0128] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

Claims

1. A method for registering aerospace components, characterized in that, Includes the following steps: The original point cloud of the aerospace component is preprocessed to obtain a preprocessed point cloud; The preprocessed point cloud is input into the encoder and decoder to obtain a coarse point cloud; The preprocessed point cloud and the coarse point cloud are input into the generator to obtain the complete point cloud; Based on the complete point cloud, the aerospace components are registered; The original point cloud is a residual point cloud; The step of inputting the preprocessed point cloud into the encoder and decoder to obtain a coarse point cloud includes: inputting the preprocessed point cloud into the encoder to extract important feature information of the point cloud; and inputting the important feature information of the point cloud into the decoder to obtain the coarse point cloud. The encoder includes a first shared multilayer perceptron layer, a first max pooling layer, an extended layer, a second shared multilayer perceptron layer, and a second max pooling layer. The step of inputting the preprocessed point cloud into the encoder to extract important feature information includes: inputting the preprocessed point cloud into the first shared multilayer perceptron layer, performing linear transformation and nonlinear activation on each point cloud to extract shallow features; inputting the shallow features into the first max pooling layer to extract the maximum value of each point cloud in the feature dimension to obtain salient features; inputting the salient features into the extended layer to obtain extended features; wherein the extended features have the same feature dimension as the shallow features; inputting the shallow features and extended features into the second shared multilayer perceptron layer to obtain comprehensive features; and inputting the comprehensive features into the second max pooling layer to extract important feature information of the point cloud. The decoder includes a third shared multilayer perceptron layer and a reshaping layer; the step of inputting the important feature information of the point cloud into the decoder to obtain the coarse point cloud includes: inputting the important feature information of the point cloud into the third shared multilayer perceptron layer to extract decoded features; inputting the decoded features into the reshaping layer to convert the decoded features into features of a preset shape and format to obtain the coarse point cloud; wherein, the coarse point cloud includes at least the main geometric features of the original point cloud; The generator includes: a first multilayer perceptron layer, a second multilayer perceptron layer, three attention feature enhancement layers, and a third multilayer perceptron layer. The step of inputting the preprocessed point cloud and the coarse point cloud into the generator to obtain a complete point cloud includes: inputting the preprocessed point cloud and the coarse point cloud together into the first multilayer perceptron layer to extract first multi-level features; inputting the important feature information of the point cloud into the second multilayer perceptron layer to extract second multi-level features; superimposing the first and second multi-level features and sequentially inputting them into the three attention feature enhancement layers to obtain attention features; and superimposing the multi-level features and attention features and inputting them into the third multilayer perceptron layer to obtain the complete point cloud.

2. The aerospace component registration method according to claim 1, characterized in that, The preprocessing of the original point cloud of the aerospace component to obtain a preprocessed point cloud includes: The original point cloud is sampled at its farthest point to obtain a sampled point cloud; The sampled point cloud is normalized to obtain a normalized point cloud; The normalized point cloud is divided into regions to obtain a preprocessed point cloud.

3. The aerospace component registration method according to claim 2, characterized in that, The step of sampling the farthest point from the original point cloud to obtain the sampled point cloud includes: One or more point clouds are randomly selected from the original point cloud as the initial sampled point cloud set, and the remaining original point clouds are used as the initial unsampled point cloud set. For each point cloud in the current unsampled point cloud set, calculate the minimum distance from the unsampled point cloud to all point clouds in the current sampled point cloud set, and add the unsampled point cloud corresponding to the maximum value of the minimum distance to the current sampled point cloud set; When the number of point clouds in the current sampling point cloud set reaches a preset threshold, the point cloud in the current sampling point cloud set is used as the sampling point cloud.

4. The aerospace component registration method according to claim 2, characterized in that, The step of dividing the normalized point cloud into regions to obtain a preprocessed point cloud includes: Based on the normalized point cloud, M center points are randomly initialized; Based on the K-nearest neighbor algorithm, neighborhood regions are extracted for each of the M center points, and the normalized point cloud is divided into M parts; For each of the M parts, divide it into a first part, a second part, and a third part; The first part, the second part, and the third part are used as the preprocessed point cloud; The first part is the region observable by the encoder, the second part is the masked region, and the third part is the region unobservable by the encoder.

5. The aerospace component registration method according to claim 1, characterized in that, The registration of the aerospace components based on the complete point cloud includes: Based on the complete point cloud, the edge curvature similarity and contour matching degree of the aerospace component are obtained; Based on the curvature similarity and contour matching degree, the multi-feature registration result of the aerospace component is obtained through weighted fusion; When the multi-feature registration result meets the preset conditions, the aerospace components are registered.

6. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; The processor is configured to execute the aerospace component registration method according to any one of claims 1-5.

Images