Surface texture reconstruction method and device for non-contact measurement test, equipment and medium

Abstract

The invention provides a surface texture reconstruction method and device for a non-contact measurement test, equipment and a medium, and the method comprises the steps: obtaining a data set of the shape of a test model, dividing the data set into a training set and a test set, and enabling the data set to comprise a data pair composed of photo data and labels; constructing a texture reconstruction restoration end-to-end model comprising a preprocessing module, a texture map prediction module of an encoder-decoder structure and a restoration module for texture mapping; training the texture reconstruction and restoration end-to-end model in a distributed manner by adopting the training data set to obtain a trained texture reconstruction and restoration end-to-end model; inputting the input data in the test set and the test model into the trained texture reconstruction and reduction end-to-end model to obtain a three-dimensional model with surface physical information distribution; the problems that in non-contact measurement test result analysis, a traditional analysis means based on a two-dimensional picture is large in judgment difficulty, many in human factors, distorted in physical parameter distribution and low in efficiency are solved.

Description

Technical Field

[0001] The invention relates to the technical field of aerodynamic test, and in particular to a surface texture reconstruction method, device, equipment and medium for non-contact measurement test. Background Art

[0002] In non-contact measurement test tasks such as pressure measurement and heat measurement of models, the test results need to be verified and analyzed. The existing method is to use cameras at different angles to take pictures, and the pictures are processed by the program to verify with the CFD (Computational Fluid Dynamics) data, and the distribution and change trend of physical quantities on the model surface are analyzed through the pictures. This method has certain problems in practical applications.

[0003] Affected by lighting and shooting angle, photos from different angles may reflect the physical quantities of the same position of the model differently, which increases the human factors in data judgment and interpretation and the communication cost with customers.

[0004] Two-dimensional photos cannot reflect the real surface flow information and can only be used for qualitative display. They cannot quantitatively analyze the development trend and change gradient of aerodynamic parameters, and it is difficult to obtain a physical explanation for the distribution of aerodynamic parameters. It is not conducive to exploring the physical mechanism behind the flow and cannot be applied to aerodynamic characteristics analysis and design.

[0005] Traditional reconstruction methods such as DLT (Digital Image Correlation for Large-scale Testing) form a partial three-dimensional surface by identifying the position of pressure measuring holes, and fuse multiple perspectives to form the final three-dimensional surface flow field. The processing process is cumbersome and inefficient, and is only applicable to simple blunt-headed bodies without rudders and protrusions. It is essentially interpolation and has low calculation accuracy.

[0006] In recent years, the rapid development of artificial intelligence vision has provided more advanced technical tools for vertical research. ViT (Vision Transformer) is based on the self-attention mechanism of Transformer, which can effectively capture the global dependencies in the image. It also has parallel capabilities and high flexibility, and its performance is much better than that of traditional CNN. The Encoder-Decoder structure model based on ViT and attention block gradually replaces the traditional convolution-deconvolution model, providing a new solution for the three-dimensional reconstruction of surface flow fields.

[0007] Therefore, it is urgent to propose a surface texture reconstruction method for non-contact measurement tests to solve the problems existing in traditional analysis methods in the analysis of non-contact measurement test results, such as difficulty in judgment, many human factors, distortion of physical parameter distribution, and low efficiency. Summary of the invention

[0008] In order to overcome the problems existing in the related art, the present disclosure provides a surface texture reconstruction method, device, equipment and medium for non-contact measurement test, so as to solve the technical problems existing in the traditional analysis means in the analysis of non-contact measurement test results in the related art, such as great difficulty in judgment, many human factors, distortion of physical parameter distribution and low efficiency.

[0009] One or more embodiments of the present specification provide a surface texture reconstruction method for a non-contact measurement test, comprising the following steps:

[0010] Obtaining a data set of the test model's appearance, divided into a training set and a test set, wherein the data set includes data pairs consisting of photo data and labels;

[0011] Construct an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping;

[0012] Distributed training of the texture reconstruction and restoration end-to-end model is performed using the training data set to obtain a trained texture reconstruction and restoration end-to-end model;

[0013] The input data in the test set and the test model are input into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution.

[0014] Preferably, the preprocessing module performs shape recognition, cropping and transformation on the photo data of the test model shape and uses it as the input of the texture map prediction module;

[0015] The texture map prediction module uses ViT as an encoder and an attention-based Transformer and a deep deconvolutional network as a decoder to obtain a predicted texture map;

[0016] The restoration module restores the data set of the test model shape and the texture map into a textured three-dimensional mesh model.

[0017] Preferably, the adopting the training data set to distribute train the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model specifically comprises the following steps:

[0018] Normalizing and standardizing the training set;

[0019] Use pre-trained weights for large scenarios to initialize weights;

[0020] Construct a loss function to train the texture map prediction module;

[0021] The trained weights are loaded into the texture map prediction module, and together with the preprocessing module and the restoration module, form an end-to-end model for texture reconstruction and restoration.

[0022] Perform DDP-based distributed training on supercomputing nodes.

[0023] Preferably, the data set for obtaining the test model shape is divided into a training set and a test set, and the data set contains data pairs consisting of input data and labels, and specifically includes the following steps:

[0024] Obtaining a half-mold of each test model shape in the test model shape data set, performing mesh conformal parameterization on the half-mold, and mapping the half-mold onto a plane square to obtain a UV unfolded image;

[0025] Rendering the color of the half-mold mesh onto the UV unfolded map to obtain a texture map;

[0026] The photo data of the test model shape and the texture map are combined into a data pair including the photo data and a label.

[0027] One or more embodiments of this specification provide a surface texture reconstruction device for a non-contact measurement test, including a data acquisition module, a model building module, a model training module, and a reconstruction module;

[0028] The data acquisition module is used to acquire a data set of the test model's appearance, which is divided into a training set and a test set, and the data set contains data pairs consisting of photo data and labels;

[0029] The model construction module is used to construct an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping;

[0030] The model training module is used to use the training data set to distribute the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model;

[0031] The reconstruction module is used to input the input data in the test set and the test model into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution.

[0032] Preferably, the model building module includes a preprocessing unit, a prediction unit and a restoration unit;

[0033] The preprocessing unit is used for the preprocessing module to perform shape recognition, cropping and transformation on the photo data of the test model shape as input to the texture map prediction module;

[0034] The prediction unit is used for the texture map prediction module to use ViT as an encoder and the Transformer based on the attention mechanism and the deep deconvolution network as a decoder to obtain a predicted texture map;

[0035] The restoration unit is used for the restoration module to restore the data set of the test model shape and the texture map into a textured three-dimensional mesh model.

[0036] Preferably, the model training module includes a processing unit, an initialization unit, a training unit, a model building unit and a distributed training unit;

[0037] The processing unit is used to perform normalization and standardization on the training set;

[0038] The initialization unit is used to perform weight initialization using pre-trained weights in a large scene;

[0039] The training unit is used to construct a loss function to train the texture map prediction module;

[0040] The model building unit is used to load the trained weights into the texture map prediction module, and form an end-to-end texture reconstruction and restoration model with the preprocessing module and the restoration module;

[0041] The distributed training unit is used to perform DDP-based distributed training on supercomputing nodes.

[0042] Preferably, the data acquisition module includes a mapping unit, a rendering unit and a composition unit;

[0043] The mapping unit is used to obtain a half-mold of each test model shape in the test model shape data set, perform grid conformal parameterization on the half-mold, and map it onto a plane square to obtain a UV unfolded map;

[0044] The rendering unit is used to render the color of the half-mold mesh onto the UV unfolded map to obtain a texture map;

[0045] The composition unit is used to combine the photo data of the test model's appearance and the texture map into a data pair including the photo data and a label.

[0046] One or more embodiments of the present specification provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the surface texture reconstruction method of the non-contact measurement test as described above when executing the computer program.

[0047] One or more embodiments of the present specification provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the surface texture reconstruction method of the non-contact measurement test described above.

[0048] The present invention provides a surface texture reconstruction method, device, equipment and medium for non-contact measurement test, which has the advantages of obtaining a data set of the test model's appearance, which is divided into a training set and a test set, wherein the data set contains data pairs consisting of photo data and labels, and can train and verify the texture reconstruction and restoration end-to-end model in a targeted manner to ensure that the model learns appropriate features and laws; constructing a texture reconstruction and restoration end-to-end model including a preprocessing module, a texture map prediction module of an encoder-decoder structure and a restoration module for texture mapping, thereby realizing integrated processing of the entire process from input data to final output of a three-dimensional model with surface physical information distribution, reducing the connection problems and error accumulation of intermediate links, and helping to improve the overall processing efficiency. rate and accuracy; using the training data set to distribute the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model, the distributed computing resources can be used to accelerate the training process, and the model can better fit the data features, and then after the test set data is input into the trained model, a three-dimensional model with surface physical information distribution can be reconstructed more accurately; the input data in the test set and the test model are input into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution, so that the reconstructed texture is not only accurate in visual presentation, but also can reflect the distribution of physical properties such as material and roughness on the surface, reducing the difficulty of data discrimination and human factors in data interpretation. At the same time, the three-dimensional shape can more realistically reflect the flow information on the aircraft surface, and can be used to quantitatively analyze the development trend and change gradient of aerodynamic parameters, which is conducive to exploring the physical mechanism behind the flow, or can be applied to aerodynamic characteristic analysis and design. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0050] Figure 1 A schematic flow chart of a surface texture reconstruction method for a non-contact measurement test provided for one or more embodiments of this specification;

[0051] Figure 2 A flow chart of a surface texture reconstruction method for a non-contact measurement test provided in one or more embodiments of this specification;

[0052] Figure 3 This is a flowchart of Example 1 of this specification;

[0053] Figure 4 A schematic diagram of the structure of a surface texture reconstruction device for a non-contact measurement test provided for one or more embodiments of this specification;

[0054] Figure 5 A schematic diagram of the structure of a computer device provided for one or more embodiments of this specification. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the protection scope of the present invention.

[0056] The present invention is described in detail below in conjunction with specific implementation methods and the accompanying drawings.

[0057] Method Embodiment

[0058] According to an embodiment of the present invention, a surface texture reconstruction method for a non-contact measurement test is provided, such as Figure 1 FIG. 1 is a flow chart of a surface texture reconstruction method for a non-contact measurement test provided in this embodiment. The surface texture reconstruction method for a non-contact measurement test according to an embodiment of the present invention comprises the following steps:

[0059] S110, obtaining a data set of the test model shape, divided into a training set and a test set, wherein the data set includes data pairs consisting of photo data and labels, and generating paired data {input, label} that can be used for training by mesh parameterization and color rendering of non-contact measurement photos of the same type of test shape and the corresponding three-dimensional shape, wherein the input is a two-dimensional photo and the label is a texture map containing surface color information. The data is divided into a training set and a test set for model training and testing.

[0060] S120, constructing an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping.

[0061] S130, using the training data set to distribute train the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model.

[0062] S140, input the input data in the test set and the test model into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution. Specifically, load the trained weights into the texture map prediction module, and form a texture reconstruction and restoration end-to-end model with the preprocessing module. Input one or more two-dimensional photos of the test model with surface information texture outside the training data set at different angles into the texture reconstruction model to obtain a predicted texture map. Input the texture map and the test model into the constructed restoration model, and the restoration model predicts the texture map according to the texture map Tex(u i ,v i )=(R i ,G i ,B i ) and digital-analog position diagram Pos(u i ,v i )=(x i ,y i ,z i ), and obtain a 3D model with surface physical information distribution. In particular, when the texture in a photo is blocked, multiple photos from different angles can be used for texture prediction to restore the complete surface texture.

[0063] The method provided in this embodiment obtains a data set of the test model's appearance, which is divided into a training set and a test set. The data set contains data pairs consisting of photo data and labels, and can train and verify the end-to-end texture reconstruction and restoration model in a targeted manner to ensure that the model learns appropriate features and rules; constructs an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping, thereby realizing integrated processing of the entire process from input data to final output of a three-dimensional model with surface physical information distribution, reducing the connection problems and error accumulation in the intermediate links, and helping to improve the overall processing efficiency and accuracy; using the training data The distributed training of the texture reconstruction and restoration end-to-end model is used to obtain a trained texture reconstruction and restoration end-to-end model, which can accelerate the training process by using distributed computing resources, and can make the model better fit the data features, and then after the test set data is input into the trained model, a three-dimensional model with surface physical information distribution is reconstructed more accurately; the input data in the test set and the test model are input into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution, so that the reconstructed texture is not only accurate in visual presentation, but also can reflect the distribution of physical properties such as material and roughness on the surface, reducing the difficulty of data discrimination and human factors in data interpretation. At the same time, the three-dimensional shape can more realistically reflect the flow information on the aircraft surface, and can be used to quantitatively analyze the development trend and change gradient of aerodynamic parameters, which is conducive to exploring the physical mechanism behind the flow, or can be applied to aerodynamic characteristic analysis and design.

[0064] In one embodiment, a preprocessing module is constructed, and the function of the preprocessing module is to preprocess the image. The preprocessing module performs shape recognition, cropping and transformation on the photo data of the test model shape as the input of the texture map prediction module. The shape recognition of the preprocessing module is realized by a pre-trained CNN. Cropping and transformation convert the location of the recognized shape into a picture of a specific size as the input of the texture map prediction module.

[0065] A texture map prediction module is constructed. The texture map prediction module uses ViT as the encoder and the Transformer based on the attention mechanism and the deep deconvolution network as the decoder. The w*h*3 image is encoded by the encoder to generate a small-size multi-channel image feature vector n*n*c1. The feature vector passes through the Transformer to obtain texture tokens with a dimension of 3*m*m*c2, and finally restored to the w*h*3 texture map through deconvolution.

[0066] Construct a restoration module, which restores the test model shape data set and the texture map into a textured three-dimensional mesh model. First, the mesh of the digital model half-model is parameterized and rendered to obtain a position map representing geometric information. Specifically, Pos(u i ,v i )=(x i ,y i ,z i ). Then, the position map index relationship is used to restore it to a three-dimensional digital model grid. Then, the pixel points of the texture map and the grid points are matched to obtain a three-dimensional grid model with texture.

[0067] The method provided in this embodiment can effectively preprocess the data set to provide suitable input, use the encoder and decoder of specific structure to accurately predict the texture map, and finally restore the relevant data to a textured three-dimensional mesh model through the restoration module, thereby realizing efficient and accurate construction from the test model shape data set to a textured three-dimensional model, which can be used for texture reconstruction under non-contact measurement tests and broaden the application scenarios.

[0068] In one embodiment, S130, using the training data set to distribute train the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model, specifically includes the following steps:

[0069] Normalize the photo values ​​in the training set to the interval [0,1], and then perform standardization. Specifically, transform the pixel value x / 255 of the photo to the interval [0,1], and then calculate the mean value of the pixel and standard deviation σ, through standardization.

[0070] The weights of the texture reconstruction and restoration end-to-end model are initialized using the pre-trained weights in a large scene, and the trained weights are loaded into the texture map prediction module, and together with the pre-processing module, the texture reconstruction and restoration end-to-end model is formed. Among them, the ViT model uses the pre-trained weights in a large scene, and the loss function selects MSE or MAE. MAE can better resist noise. The optimizer selects the built-in Adam method of pytorch, and the learning rate adopts an exponential decay strategy.

[0071] The photos in the training set are input into the texture prediction module, and the loss function is constructed by combining the obtained output data with the labels in the training set to train the texture map prediction module.

[0072] The trained weights are loaded into the texture map prediction module, and together with the preprocessing module and the restoration module, form an end-to-end texture reconstruction and restoration model.

[0073] To speed up the training, distributed training based on DDP (Distributed Data Parallel) is performed on supercomputing nodes.

[0074] The method provided in this embodiment can efficiently optimize the end-to-end model of texture reconstruction and restoration by normalizing and standardizing the training set, initializing and building the model using pre-trained weights, training according to the loss function, and combining distributed training of supercomputing nodes, thereby improving its accuracy and generalization ability in texture reconstruction tasks.

[0075] In one embodiment, S110, obtaining a data set of the test model shape, which is divided into a training set and a test set, wherein the data set contains data pairs consisting of input data and labels, specifically includes the following steps:

[0076] Obtain the half-mold of each test model shape in the test model shape data set, perform mesh conformal parameterization on the half-mold, and map it to a plane square to obtain a UV expansion map. o All of them are composed of their adjacent points P i Decision, P o =∑ {i：[o，i]∈Edges} λ oi P i .

[0077] The color of the half-mold mesh is rendered onto the UV unfolded map to obtain a texture map. i , v i )=(R i , G i , B i ),u i , V i is the pixel coordinate in the texture map, R i , G i , B i For Texture Figure 3 Channel value.

[0078] The photo data of the test model shape and the texture map are combined into a data pair consisting of photo data and labels. Each shape generates a texture map and forms a pair of data {photo, texture map} with the shape photo. All data are divided into two parts: training set and test set in a ratio of 8:2.

[0079] The method provided in this embodiment obtains a UV unfolded image and a texture image by performing operations such as mesh conformal parameterization on the test model's outer shape half-mold, and forms a data pair with the photo data, thereby providing effective and matching basic data for subsequent texture reconstruction model training.

[0080] The technical solution is further described below through specific embodiments:

[0081] Example 1

[0082] like Figure 2 As shown, it is a workflow diagram of Example 1 provided in this embodiment. This example constructs a model in an end-to-end manner and performs texture reconstruction and digital model mapping. Specifically, it includes the following steps:

[0083] 1. Construction of missile shape dataset: Through mesh parameterization and rendering of historical test data, a missile shape dataset for training is constructed, including non-contact measurement photos of missiles of specific sizes and corresponding texture maps.

[0084] 2. Construction of an end-to-end model for texture reconstruction and restoration: Construction of a preprocessing module, a texture map prediction module based on the encoding and decoding structure of ViT and Transformer, and an end-to-end model for texture reconstruction and restoration of the restoration model.

[0085] 3. Training of texture reconstruction and restoration model: Use the missile shape dataset constructed in step 1 to perform distributed training on the texture map prediction module in the texture reconstruction and restoration end-to-end model to obtain a texture map prediction module with texture map reconstruction capabilities.

[0086] 4. Restoration of three-dimensional surface flow field information: Connect the trained texture map prediction module to the end-to-end model, input the missile photos and test numerical models after the test, quickly generate the surface texture map and map it to the three-dimensional missile shape, and obtain the three-dimensional missile shape with surface physical parameter distribution.

[0087] like Figure 2 Shown is a flow chart of a texture reconstruction method for a non-contact measurement test provided in this embodiment, which is specifically described as follows.

[0088] Step 1: Construction of missile shape dataset:

[0089] (1.1) Collect missile photos and 3D shapes from historical test data, perform mesh conformal parameterization on the half-mold, and embed it into the square uv plane. Each point P within the square boundary o The positions of its neighboring points P i Decision, P o =∑ {i：[o,i]∈Edges} λ oi P i .

[0090] (1.2) Render the missile's half-model color information to the square uv plane in the previous step to obtain a texture map. Specifically, Tex(u i ,v i )=(R i,G i ,B i ),u i ,v i is the pixel coordinate in the texture map, R i ,G i ,B i For Texture Figure 3 Channel value. The texture map contains the three-dimensional color information of the missile, so it can be mapped to the three-dimensional shape.

[0091] (1.3) A texture map is generated for each missile shape in the data set, and a pair of data {photo, texture map} is formed with the shape photo. The size of the photo must be consistent with the input dimension of the texture map prediction module, such as 512×512. All data are divided into a training set and a test set in a ratio of 8:2 for model training and testing.

[0092] Step 2: Construction of the end-to-end model for texture reconstruction and restoration:

[0093] (2.1) Construct a preprocessing module. First, a pre-trained CNN model or an open source Object detection model is needed to identify the location of the missile in the original photo. The model then performs cropping and transformation operations to convert the location of the identified shape into a picture of a certain size (such as 512×512) as the input of the texture map prediction module. This operation can be performed in the openCV environment. Secondly, grid parameterization methods and rendering are added to the preprocessing module to automatically process the input test model into a position map.

[0094] (2.2) Construct a texture map prediction module. The model is an encoder-decoder structure, where the encoder is a visual Transformer-ViT. Specifically, the mainstream DINO ViT can be used. The decoder is a Transformer+deconvolution network. A picture of a certain size n*n*3 is encoded by the encoder to generate a small-size multi-channel feature vector n*n*C, and then restored to a n*n*3 texture map by the decoder. In particular, for convenience, the image size is set to be equal in length and width.

[0095] (2.3) Construct a restoration module. The index relationship is obtained through the order of pixels in the digital model position map, and the position map is restored to a triangular mesh shape, usually in .obj format. Then a mapping algorithm is written to paste the predicted texture map onto the digital model mesh to obtain a restoration module. The above three models together constitute an end-to-end 3D reconstruction model.

[0096] Step 3: Training of the end-to-end model for texture reconstruction and restoration:

[0097] (3.1) Normalization and standardization of missile training and validation set photos. It is easier to converge the model with standardized data. Specifically, transform the pixel value x / 255 of the photo to the interval [0,1], and then calculate the mean value of the pixel and standard deviation σ, through standardization.

[0098] (3.2) Model weight initialization and training. For the encoder, the pre-trained DINO ViT weights are used as the initial weights. After the weights are initialized, the training data photos are input into the texture map prediction module in the texture reconstruction and restoration end-to-end model. The output and label of the model are used to construct the loss function for training. The loss function is the mean absolute error MAE. where y i is the output of the model, is the label value corresponding to the shape. The optimizer selects the built-in Adam method of pytorch, and the learning rate adopts the exponential decay strategy.

[0099] (3.3) Training testing and acceleration. During the training process, every fixed number of rounds, the images in the test set are input into the texture map prediction module in the texture reconstruction and restoration end-to-end model, and the error between the model output and the label is compared to test the training effect and generalization ability of the model. At the same time, in order to speed up the training speed, distributed training based on DDP can be performed on supercomputing nodes.

[0100] Step 4: Restoration of three-dimensional surface flow field information:

[0101] (4.1) Load the trained weights into the texture map prediction module, and form an end-to-end texture reconstruction and restoration model with the preprocessing module and the restoration module. Input the missile photo with surface texture and the missile digital model into the end-to-end texture reconstruction and restoration model, and the three-dimensional shape of the missile with the surface flow field information distribution can be quickly obtained.

[0102] (4.2) In particular, when it is necessary to analyze the distribution of surface flow field information on the front and back sides of the model, two photos taken at different angles can be used to input the texture reconstruction model to obtain two texture maps at different angles. The texture maps are then pasted to the corresponding model positions to obtain the three-dimensional surface texture distribution of the entire missile.

[0103] Device Embodiment

[0104] According to an embodiment of the present invention, a surface texture reconstruction device for a non-contact measurement test is provided, such as Figure 4 As shown, it is a structural schematic diagram of the surface texture reconstruction device of the non-contact measurement test provided in this embodiment. The surface texture reconstruction device of the non-contact measurement test according to the embodiment of the present invention includes a data acquisition module 41, a model building module 42, a model training module 43 and a reconstruction module 44.

[0105] The data acquisition module 41 is used to acquire a data set of the test model's appearance, which is divided into a training set and a test set. The data set contains data pairs consisting of photo data and labels.

[0106] The model construction module 42 is used to construct an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping.

[0107] The model training module 43 is used to use the training data set to distribute train the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model.

[0108] The reconstruction module 44 is used to input the input data in the test set and the test model into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution.

[0109] In the device provided in this embodiment, the data acquisition module 41 obtains a data set of the test model's appearance, which is divided into a training set and a test set. The data set contains data pairs consisting of photo data and labels, which can train and verify the texture reconstruction and restoration end-to-end model in a targeted manner to ensure that the model learns appropriate features and laws; the model construction module 42 constructs a texture reconstruction and restoration end-to-end model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping, thereby realizing the integrated processing of the entire process from input data to the final output of a three-dimensional model with surface physical information distribution, reducing the connection problems and error accumulation of the intermediate links, and helping to improve the overall processing efficiency and accuracy; the model training module 4 3. The training data set is used to distribute the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model, which can accelerate the training process by using distributed computing resources, and can make the model better fit the data features, and then after the test set data is input into the trained model, a three-dimensional model with surface physical information distribution can be reconstructed more accurately; the reconstruction module 44 inputs the input data in the test set and the test model into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution, so that the reconstructed texture is not only accurate in visual presentation, but also can reflect the distribution of physical characteristics such as material and roughness on the surface, reducing the difficulty of data discrimination and human factors in data interpretation. At the same time, the three-dimensional shape can more realistically reflect the flow information on the aircraft surface, can be used to quantitatively analyze the development trend and change gradient of aerodynamic parameters, is conducive to exploring the physical mechanism behind the flow, or can be applied to aerodynamic characteristic analysis and design.

[0110] In one embodiment, the model building module 42 includes a preprocessing unit 4201 , a prediction unit 4202 , and a restoration unit 4203 .

[0111] The preprocessing unit 4201 is used for the preprocessing module to perform shape recognition, cropping and transformation on the photo data of the test model shape and use it as the input of the texture map prediction module.

[0112] The prediction unit 4202 is used for the texture map prediction module to use ViT as an encoder and the Transformer based on the attention mechanism and the deep deconvolution network as a decoder to obtain a predicted texture map.

[0113] The restoration unit 4203 is used for the restoration module to restore the data set of the test model shape and the texture map into a textured three-dimensional mesh model.

[0114] The device provided in this embodiment can effectively pre-process the data set to provide suitable input, use the encoder and decoder of specific structure to accurately predict the texture map, and finally restore the relevant data to a textured three-dimensional mesh model through the restoration module, thereby realizing efficient and accurate construction from the test model shape data set to the textured three-dimensional model, which can be used for texture reconstruction under non-contact measurement tests and broaden the application scenarios.

[0115] In one embodiment, the model training module 43 includes a processing unit 4301, an initialization unit 4302, a training unit 4303, a model building unit 4304 and a distributed training unit 4305.

[0116] The processing unit 4301 is used to normalize and standardize the training set.

[0117] The initialization unit 4302 is used to perform weight initialization using pre-trained weights in a large scene.

[0118] The training unit 4303 is used to construct a loss function to train the texture map prediction module.

[0119] The model building unit 4304 is used to load the trained weights into the texture map prediction module, and form an end-to-end texture reconstruction and restoration model with the preprocessing module and the restoration module.

[0120] The distributed training unit 4305 is used to perform DDP-based distributed training on supercomputing nodes.

[0121] The device provided in this embodiment can efficiently optimize the texture reconstruction and restoration end-to-end model by normalizing and standardizing the training set, initializing and building the model using pre-trained weights, training according to the loss function, and combining the distributed training of supercomputing nodes, thereby improving its accuracy and generalization ability in texture reconstruction tasks.

[0122] In one embodiment, the data acquisition module 41 includes a mapping unit 4101 , a rendering unit 4102 and a composition unit 4103 .

[0123] The mapping unit 4101 is used to obtain the half-mold of each test model shape in the data set of the test model shape, perform grid conformal parameterization on the half-mold, and map it onto a plane square to obtain a UV unfolded map.

[0124] The rendering unit 4102 is used to render the color of the half-mold mesh onto the UV unfolded map to obtain a texture map.

[0125] The composition unit 4103 is used to combine the photo data of the test model's appearance and the texture map into a data pair including the photo data and a label.

[0126] The device provided in this embodiment obtains a UV unfolded image and a texture image by performing operations such as mesh conformal parameterization on the test model's outer shape half-mold, and forms a data pair with the photo data, thereby providing effective and matching basic data for subsequent texture reconstruction model training.

[0127] The embodiment of the present invention is an apparatus embodiment corresponding to the above-mentioned method embodiment. The specific operations of the processing steps of each module can be understood by referring to the description of the method embodiment, which will not be repeated here.

[0128] like Figure 5 As shown, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the surface texture reconstruction method in the above-mentioned embodiment is implemented, or when the computer program is executed by a processor, the surface texture reconstruction method in the non-contact measurement test in the above-mentioned embodiment is implemented.

[0129] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0130] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0131] Finally, 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and the contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

Claims

1. A surface texture reconstruction method for a non-contact measurement test, characterized in that: The following steps are involved: Obtaining a data set of the test model's appearance, divided into a training set and a test set, wherein the data set includes data pairs consisting of photo data and labels; Construct an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping; Distributed training of the texture reconstruction and restoration end-to-end model is performed using the training data set to obtain a trained texture reconstruction and restoration end-to-end model; The input data in the test set and the test model are input into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution.

2. The surface texture reconstruction method of the non-contact measurement test according to claim 1, characterized in that: The pre-processing module performs shape recognition, cropping and transformation on the photo data of the test model shape and uses it as input of the texture map prediction module; The texture map prediction module uses ViT as an encoder and an attention-based Transformer and a deep deconvolutional network as a decoder to obtain a predicted texture map; The restoration module restores the data set of the test model shape and the texture map into a textured three-dimensional mesh model.

3. The surface texture reconstruction method of the non-contact measurement test according to claim 1, characterized in that: The adopting of the training data set to distribute the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model specifically comprises the following steps: Normalizing and standardizing the training set; Use pre-trained weights for large scenarios to initialize weights; Construct a loss function to train the texture map prediction module; The trained weights are loaded into the texture map prediction module, and together with the preprocessing module and the restoration module, form an end-to-end model for texture reconstruction and restoration. Perform DDP-based distributed training on supercomputing nodes.

4. The surface texture reconstruction method of the non-contact measurement test according to claim 1, characterized in that: The data set for obtaining the test model shape is divided into a training set and a test set, and the data set contains data pairs consisting of input data and labels, and specifically includes the following steps: Obtaining a half-mold of each test model shape in the test model shape data set, performing mesh conformal parameterization on the half-mold, and mapping the half-mold onto a plane square to obtain a UV unfolded image; Rendering the color of the half-mold mesh onto the UV unfolded map to obtain a texture map; The photo data of the test model shape and the texture map are combined into a data pair including the photo data and a label.

5. A surface texture reconstruction device for non-contact measurement test, characterized in that: It includes data acquisition module, model building module, model training module and reconstruction module; The data acquisition module is used to acquire a data set of the test model's appearance, which is divided into a training set and a test set, and the data set contains data pairs consisting of photo data and labels; The model construction module is used to construct an end-to-end texture reconstruction and restoration model including a preprocessing module, a texture map prediction module of an encoder-decoder structure, and a restoration module for texture mapping; The model training module is used to use the training data set to distribute the texture reconstruction and restoration end-to-end model to obtain a trained texture reconstruction and restoration end-to-end model; The reconstruction module is used to input the input data in the test set and the test model into the trained texture reconstruction and restoration end-to-end model to obtain a three-dimensional model with surface physical information distribution.

6. The surface texture reconstruction device for non-contact measurement test according to claim 5, characterized in that: The model building module includes a preprocessing unit, a prediction unit and a restoration unit; The preprocessing unit is used for the preprocessing module to perform shape recognition, cropping and transformation on the photo data of the test model shape as input to the texture map prediction module; The prediction unit is used for the texture map prediction module to use ViT as an encoder and the Transformer based on the attention mechanism and the deep deconvolution network as a decoder to obtain a predicted texture map; The restoration unit is used for the restoration module to restore the data set of the test model shape and the texture map into a textured three-dimensional mesh model.

7. The surface texture reconstruction device for non-contact measurement test according to claim 5, characterized in that: The model training module includes a processing unit, an initialization unit, a training unit, a model building unit and a distributed training unit; The processing unit is used to perform normalization and standardization on the training set; The initialization unit is used to perform weight initialization using pre-trained weights in a large scene; The training unit is used to construct a loss function to train the texture map prediction module; The model building unit is used to load the trained weights into the texture map prediction module, and form an end-to-end texture reconstruction and restoration model with the preprocessing module and the restoration module; The distributed training unit is used to perform DDP-based distributed training on supercomputing nodes.

8. The surface texture reconstruction device for non-contact measurement test according to claim 5, characterized in that: The data acquisition module includes a mapping unit, a rendering unit and a composition unit; The mapping unit is used to obtain a half-mold of each test model shape in the test model shape data set, perform grid conformal parameterization on the half-mold, and map it onto a plane square to obtain a UV unfolded map; The rendering unit is used to render the color of the half-mold mesh onto the UV unfolded map to obtain a texture map; The composition unit is used to combine the photo data of the test model's appearance and the texture map into a data pair including the photo data and a label.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the surface texture reconstruction method of the non-contact measurement test according to any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the surface texture reconstruction method of the non-contact measurement test according to any one of claims 1 to 4 are implemented.

Images