A digital simulation method for the lunar surface and internal structure of a three-dimensional engine

Abstract

The present invention belongs to the field of digital simulation, and specifically discloses a digital simulation method of the lunar surface and internal structure of a three-dimensional engine, the method comprising: data collection, lunar data processing, building a three-dimensional model, simulation and rendering, and interaction and simulation. The present invention constructs a denoising autoencoder neural network, regards the missing values ​​of the lunar data as the noise of the lunar data, interpolates the lunar data through an unsupervised learning method, and has good generalization ability in various forms of lunar data processing; uses orthogonal correction to correct the lunar surface image, eliminates image distortion caused by terrain undulations and camera position changes, extracts surface normal information from the corrected image, and uses the surface normal information to optimize the digital model, thereby obtaining a high-resolution lunar digital model.

Description

Technical Field

[0001] The present invention relates to the field of digital simulation, and in particular to a method for digitally simulating the lunar surface and internal structure of a three-dimensional engine. Background Art

[0002] The digital model of lunar structure helps people better understand the characteristics and evolution of the moon and its interaction with the earth and other celestial bodies in the solar system by simulating various physical, chemical and geological processes on the moon. Currently, the lunar exploration mission is still ongoing. The lack of lunar data has led to an incomplete understanding of the structure and current state of the moon, which has restricted the development of scientific research. The lunar data is huge and complex. Traditional interpolation methods may have performance problems when processing such high-dimensional, severely nonlinear and unconventional data sets. When transmitting lunar image data to the earth, it is easy to have low resolution of lunar image data due to signal attenuation, limited storage capacity, cost considerations or limitations of lunar exploration mission objectives, which cannot provide enough detailed information, affecting the visual effect, accuracy and reliability of modeling. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a digital simulation method of the lunar surface and internal structure of a three-dimensional engine. In view of the fact that the current lunar exploration mission is still in progress, the lack of lunar data has led to incomplete understanding of the structure and current state of the moon, which has restricted the development of scientific research. The lunar data is huge and complex. When the traditional interpolation method processes such a data set with high dimension, serious nonlinearity and unconventional form, there may be performance problems. The present invention constructs a denoising autoencoder neural network, regards the missing values ​​of the lunar data as the noise of the lunar data, interpolates the lunar data through an unsupervised learning method, has good generalization ability in various forms of lunar data processing, and provides complete data for subsequent three-dimensional modeling; in view of the technical problem that when the lunar image data is transmitted to the earth, it is easy to cause the low resolution of the lunar image data due to signal attenuation, limited storage capacity, cost considerations or restrictions on the objectives of the lunar exploration mission, and cannot provide sufficient detail information, affecting the visual effect, accuracy and reliability of the modeling, the present invention uses orthogonal correction to correct the lunar surface image, eliminates the image distortion caused by terrain undulation and camera position change, extracts surface normal information from the corrected image, and uses the surface normal information to optimize the digital model, thereby obtaining a high-resolution lunar digital model.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a method for digitally simulating the lunar surface and internal structure of a three-dimensional engine, which specifically includes the following steps:

[0005] Step S1: Data collection, collecting high-precision lunar data from lunar exploration missions, the lunar data including lunar surface images, lunar surface data and lunar internal structure data, and removing invalid and obviously erroneous lunar data;

[0006] Step S2: lunar data processing, using an unsupervised denoising autoencoder neural network to perform multiple interpolations on the missing parts of the lunar data to obtain complete lunar data;

[0007] Step S3: Establish a three-dimensional model, use three-dimensional modeling software to create digital models of the lunar surface structure and lunar internal structure, and use surface normal information to improve the resolution of the lunar surface structure to obtain a high-resolution lunar digital model;

[0008] Step S4: simulation and rendering, designing the architecture of the three-dimensional engine, rendering the texture and lighting of the high-resolution digital lunar model, and obtaining the digital moon;

[0009] Step S5: Interaction and simulation: design a user interaction interface to allow users to scale, rotate and slice the digital moon.

[0010] Furthermore, step S2: lunar data processing specifically includes the following steps:

[0011] Step S21: Establish and initialize a denoising autoencoder neural network as a data supplement model, introduce a nonlinear neural network into the data supplement model, input the lunar data into the data supplement model for training, and perform forward transmission in the hidden layer of the nonlinear neural network. The formula used is as follows:

[0012] ;

[0013] In the formula, is the output of the hidden layer, is the weight matrix between the hidden layer nodes, and the superscript in brackets is the number of hidden layers. is the total number of hidden layers, is the input lunar data, is the bias vector of the hidden layer, is the activation function, the activation function is the exponential linear unit, is the activation function of the last layer;

[0014] Step S22: artificially add noise to the lunar data, mark the missing values ​​in the lunar data as 0, and obtain lunar labeled data, which is set to have M missing values;

[0015] Step S23: Use the dropout supplementation technique to regularize the lunar data. During training, randomly delete the nodes in the hidden layer. Record the nodes before the regularization process as the original nodes. Calculate the probability that the original nodes are retained during training. Use the probability to scale the hidden layer. Use the autoencoder to compress the lunar data into low-dimensional data. The formula used is as follows:

[0016] ;

[0017] In the formula, It is the low-dimensional data after the moon data is compressed. is the total number of hidden layers after scaling, is the Bernoulli vector in the regularization process, It is the moon marking data;

[0018] Step S24: Use a decoder to map the low-dimensional data into reconstructed data. The reconstructed data has the same probability distribution and dimension as the lunar data. The reconstructed data is passed backward in the hidden layer and reaches the output layer to obtain supplementary data. The formula used is as follows:

[0019] ;

[0020] In the formula, It is to reconstruct the data;

[0021] Step S25: Use the root mean square error and cross entropy loss function to define the loss function of the data supplementation model, preset the training threshold, and when the loss function is equal to the training threshold, the training ends, otherwise, the training continues. The formula used is as follows:

[0022] ;

[0023] In the formula, is the loss function of the data supplementation model, is the missing indicator matrix, is the training cycle, Yes Iterations;

[0024] Step S26: After the training is completed, the 0 in the lunar data is replaced by the supplementary data, and this is repeated M times to obtain the complete lunar data.

[0025] Furthermore, step S3: establishing a three-dimensional model specifically includes the following steps:

[0026] Step S31: Use 3D modeling software to create a rough digital model of the moon and construct a reflection model. Assuming that the reflection of light on the moon surface is diffuse reflection, the process of generating the moon surface image is as follows:

[0027] ;

[0028] In the formula, is an image of the lunar surface, is the surface albedo, is the light source intensity, is the surface reflectivity model, are the coordinates on the lunar surface;

[0029] Step S32: correcting the lunar surface image using orthogonal correction to eliminate image distortion caused by terrain undulations and camera position changes, thereby obtaining a corrected image;

[0030] Step S33: Using the least mean square method, the image data intensity matrix of the corrected image is calculated. The formula used is as follows:

[0031] ;

[0032] In the formula, is the image dataset intensity matrix, is the unit light matrix, is the matrix of the unit surface normal;

[0033] Step S34: Calculate the high-resolution surface normal of the spatial high-resolution and corrected image through the pseudo-inverse of the unit light source matrix. The formula used is as follows:

[0034] ;

[0035] In the formula, yes The pseudo-inverse matrix of is the high-resolution surface normal;

[0036] Step S35: Check the longitude and latitude of the high-resolution surface normal, align the high-resolution surface normal with the lunar digital rough model and fuse them to obtain an optimized digital model. The difference between the depth of the lunar digital rough model and the depth of the optimized digital model is used to define the 3D point depth error at point p. The formula used is as follows:

[0037] ;

[0038] In the formula, is the depth error of the 3D point, is the depth of the rough digital model of the Moon, is the depth of the optimized digital model, is the control parameter;

[0039] Step S36: preset a distance threshold, define the points within the distance threshold at point p as the neighboring points of point p, and calculate the surface normal error. The formula used is as follows:

[0040] ;

[0041] In the formula, is the surface normal error, and are points adjacent to the current depth point. represents a line segment composed of all adjacent points, is the surface normal;

[0042] Step S37: Express Ed and En as the sum of LMS, and generate the objective function as follows:

[0043] ;

[0044] In the formula, is the objective function, is the relative importance of the error;

[0045] Step S38: When the objective function takes the minimum value, the resolution of the constructed lunar digital model is the highest, and a high-resolution lunar digital model is obtained.

[0046] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0047] (1) As the lunar exploration mission is still ongoing, the lack of lunar data has led to an incomplete understanding of the structure and current state of the moon, which has restricted the development of scientific research. The lunar data is huge and complex. Traditional interpolation methods may have performance problems when processing such high-dimensional, severely nonlinear and unconventional data sets. The present invention constructs a denoising autoencoder neural network, which regards the missing values ​​of the lunar data as the noise of the lunar data, and interpolates the lunar data through an unsupervised learning method. It has good generalization ability in processing various forms of lunar data and provides complete data for subsequent three-dimensional modeling.

[0048] (2) In order to address the technical problem that when transmitting lunar image data to the Earth, the lunar image data has low resolution and cannot provide sufficient detail information due to signal attenuation, limited storage capacity, cost considerations or limitations of lunar exploration mission objectives, thus affecting the visual effect, accuracy and reliability of modeling, the present invention uses orthogonal correction to correct the lunar surface image, eliminates image distortion caused by terrain undulations and camera position changes, extracts surface normal information from the corrected image, and uses the surface normal information to optimize the digital model, thereby obtaining a high-resolution lunar digital model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of the steps of a method for digitally simulating the lunar surface and internal structure of a three-dimensional engine provided by the present invention.

[0050] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0052] Example 1: See Figure 1 This embodiment provides a method for digitally simulating the lunar surface and internal structure of a three-dimensional engine, which specifically includes the following steps:

[0053] Step S1: Data collection, collecting high-precision lunar data from lunar exploration missions, the lunar data including lunar surface images, lunar surface data and lunar internal structure data, and removing invalid and obviously erroneous lunar data;

[0054] Step S2: lunar data processing, using an unsupervised denoising autoencoder neural network to perform multiple interpolations on the missing parts of the lunar data to obtain complete lunar data;

[0055] Step S3: Establish a three-dimensional model, use three-dimensional modeling software to create digital models of the lunar surface structure and lunar internal structure, and use surface normal information to improve the resolution of the lunar surface structure to obtain a high-resolution lunar digital model;

[0056] Step S4: simulation and rendering, designing the architecture of the three-dimensional engine, rendering the texture and lighting of the high-resolution digital lunar model, and obtaining the digital moon;

[0057] Step S5: Interaction and simulation: design a user interaction interface to allow users to scale, rotate and slice the digital moon.

[0058] Example 2, see Figure 1 This embodiment is based on the above embodiment, step S2: lunar data processing, specifically including the following steps:

[0059] Step S21: Establish and initialize a denoising autoencoder neural network as a data supplement model, introduce a nonlinear neural network into the data supplement model, input the lunar data into the data supplement model for training, and perform forward transmission in the hidden layer of the nonlinear neural network. The formula used is as follows:

[0060] ;

[0061] In the formula, is the output of the hidden layer, is the weight matrix between the hidden layer nodes, and the superscript in brackets is the number of hidden layers. is the total number of hidden layers, is the input lunar data, is the bias vector of the hidden layer, is the activation function, the activation function is the exponential linear unit, is the activation function of the last layer;

[0062] Step S22: artificially add noise to the lunar data, mark the missing values ​​in the lunar data as 0, and obtain lunar labeled data, which is set to have M missing values;

[0063] Step S23: Use the dropout supplementation technique to regularize the lunar data. During training, randomly delete the nodes in the hidden layer. Record the nodes before the regularization process as the original nodes. Calculate the probability that the original nodes are retained during training. Use the probability to scale the hidden layer. Use the autoencoder to compress the lunar data into low-dimensional data. The formula used is as follows:

[0064] ;

[0065] In the formula, It is the low-dimensional data after the moon data is compressed. is the total number of hidden layers after scaling, is the Bernoulli vector in the regularization process, It is the moon marking data;

[0066] Step S24: Use a decoder to map the low-dimensional data into reconstructed data. The reconstructed data has the same probability distribution and dimension as the lunar data. The reconstructed data is passed backward in the hidden layer and reaches the output layer to obtain supplementary data. The formula used is as follows:

[0067] ;

[0068] In the formula, It is to reconstruct the data;

[0069] Step S25: Use the root mean square error and cross entropy loss function to define the loss function of the data supplementation model, preset the training threshold, and when the loss function is equal to the training threshold, the training ends, otherwise, the training continues. The formula used is as follows:

[0070] ;

[0071] In the formula, is the loss function of the data supplementation model, is the missing indicator matrix, is the training cycle, Yes Iterations;

[0072] Step S26: After the training is completed, the 0 in the lunar data is replaced by the supplementary data, and this is repeated M times to obtain the complete lunar data.

[0073] Through the above operations, the lunar exploration mission is still ongoing. The missing lunar data leads to incomplete understanding of the structure and current state of the moon, which limits the development of scientific research. The lunar data is huge and complex. Traditional interpolation methods may have performance problems when processing such high-dimensional, severely nonlinear and unconventional data sets. The present invention constructs a denoising autoencoder neural network, regards the missing values ​​of the lunar data as the noise of the lunar data, interpolates the lunar data through an unsupervised learning method, has good generalization ability in various forms of lunar data processing, and provides complete data for subsequent three-dimensional modeling.

[0074] Example 3, see Figure 1 This embodiment is based on the above embodiment, step S3: establishing a three-dimensional model, specifically comprising the following steps:

[0075] Step S31: Use 3D modeling software to create a rough digital model of the moon and construct a reflection model. Assuming that the reflection of light on the moon surface is diffuse reflection, the process of generating the moon surface image is as follows:

[0076] ;

[0077] In the formula, is an image of the lunar surface, is the surface albedo, is the light source intensity, is the surface reflectivity model, are the coordinates on the lunar surface;

[0078] Step S32: correcting the lunar surface image using orthogonal correction to eliminate image distortion caused by terrain undulations and camera position changes, thereby obtaining a corrected image;

[0079] Step S33: Using the least mean square method, the image data intensity matrix of the corrected image is calculated. The formula used is as follows:

[0080] ;

[0081] In the formula, is the image dataset intensity matrix, is the unit light matrix, is the matrix of the unit surface normal;

[0082] Step S34: Calculate the high-resolution surface normal of the spatial high-resolution and corrected image through the pseudo-inverse of the unit light source matrix. The formula used is as follows:

[0083] ;

[0084] In the formula, yes The pseudo-inverse matrix of is the high-resolution surface normal;

[0085] Step S35: Check the longitude and latitude of the high-resolution surface normal, align the high-resolution surface normal with the lunar digital rough model and fuse them to obtain an optimized digital model. The difference between the depth of the lunar digital rough model and the depth of the optimized digital model is used to define the 3D point depth error at point p. The formula used is as follows:

[0086] ;

[0087] In the formula, is the depth error of the 3D point, is the depth of the rough digital model of the Moon, is the depth of the optimized digital model, is the control parameter;

[0088] Step S36: preset a distance threshold, define the points within the distance threshold at point p as the neighboring points of point p, and calculate the surface normal error. The formula used is as follows:

[0089] ;

[0090] In the formula, is the surface normal error, and are points adjacent to the current depth point. represents a line segment composed of all adjacent points, is the surface normal;

[0091] Step S37: Express Ed and En as the sum of LMS, and generate the objective function as follows:

[0092] ;

[0093] In the formula, is the objective function, is the relative importance of the error;

[0094] Step S38: When the objective function takes the minimum value, the resolution of the constructed lunar digital model is the highest, and a high-resolution lunar digital model is obtained.

[0095] Through the above operations, the technical problem that when transmitting lunar image data to the earth, the resolution of the lunar image data is low, which is easy to be caused by signal attenuation, limited storage capacity, cost considerations or limitations of the lunar exploration mission objectives, and cannot provide sufficient detail information, affecting the visual effect, accuracy and reliability of modeling, the present invention uses orthogonal correction to correct the lunar surface image, eliminates image distortion caused by terrain undulations and changes in camera position, extracts surface normal information from the corrected image, and uses the surface normal information to optimize the digital model, thereby obtaining a high-resolution lunar digital model.

[0096] Embodiment 4: This embodiment is based on the above embodiment. The lunar data described in Embodiment 1 is the original data obtained from the Planetary Data System (PDS). The lunar surface image is affected by the camera orbit, light source, and terrain. When correcting the lunar surface image in Embodiment 3, the feature points of the lunar surface image are first extracted, and the camera position error is corrected using the bundle adjustment process. After selecting the lunar surface image with the selected feature points, a corrected image of the same size is generated, and the corrected image is projected according to the feature points of each image.

[0097] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0098] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0099] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for digitally simulating the lunar surface and internal structure using a three-dimensional engine, characterized in that: The specific steps include: Step S1: Data collection, collecting high-precision lunar data from lunar exploration missions, the lunar data including lunar surface images, lunar surface data and lunar internal structure data, and removing invalid and obviously erroneous lunar data; Step S2: lunar data processing, using an unsupervised denoising autoencoder neural network to perform multiple interpolations on the missing parts of the lunar data to obtain complete lunar data; Step S3: Establish a three-dimensional model, use three-dimensional modeling software to create digital models of the lunar surface structure and lunar internal structure, and use surface normal information to improve the resolution of the lunar surface structure to obtain a high-resolution lunar digital model, which specifically includes the following steps: Step S31: Use 3D modeling software to create a rough digital model of the moon and construct a reflection model. Assuming that the reflection of light on the moon surface is diffuse reflection, the process of generating the moon surface image is as follows: ； In the formula, is an image of the lunar surface, is the surface albedo, is the light source intensity, is the surface reflectivity model, are the coordinates on the lunar surface; Step S32: correcting the lunar surface image using orthogonal correction to eliminate image distortion caused by terrain undulations and camera position changes, thereby obtaining a corrected image; Step S33: Using the least mean square method, the image data intensity matrix of the corrected image is calculated. The formula used is as follows: ； In the formula, is the image dataset intensity matrix, is the unit light matrix, is the matrix of the unit surface normal, is the surface albedo; Step S34: Calculate the high-resolution surface normal of the spatial high-resolution and corrected image through the pseudo-inverse of the unit light source matrix. The formula used is as follows: ； In the formula, yes The pseudo-inverse matrix of is the high-resolution surface normal; Step S35: Check the longitude and latitude of the high-resolution surface normal, align the high-resolution surface normal with the lunar digital rough model and fuse them to obtain an optimized digital model. The difference between the depth of the lunar digital rough model and the depth of the optimized digital model is used to define the 3D point depth error at point p. The formula used is as follows: ； In the formula, is the depth error of the 3D point, is the depth of the rough digital model of the Moon, is the depth of the optimized digital model, is the control parameter; Step S36: preset a distance threshold, define the points within the distance threshold at point p as the neighboring points of point p, and calculate the surface normal error. The formula used is as follows: ； In the formula, is the surface normal error, and are points adjacent to the current depth point. represents a line segment composed of all adjacent points, is the surface normal; Step S37: Express Ed and En as the sum of LMS, and generate the objective function as follows: ； In the formula, is the objective function, is the relative importance of the error; Step S38: When the objective function takes the minimum value, the resolution of the constructed lunar digital model is the highest, and a high-resolution lunar digital model is obtained; Step S4: simulation and rendering, designing the architecture of the three-dimensional engine, rendering the texture and lighting of the high-resolution digital lunar model, and obtaining the digital moon; Step S5: Interaction and simulation: design a user interaction interface to allow users to scale, rotate and slice the digital moon.

2. The method for digitally simulating the lunar surface and internal structure of a three-dimensional engine according to claim 1, characterized in that: Step S2, lunar data processing, specifically includes the following steps: Step S21: Establish and initialize a denoising autoencoder neural network as a data supplement model, introduce a nonlinear neural network into the data supplement model, input the lunar data into the data supplement model for training, and perform forward transmission in the hidden layer of the nonlinear neural network. The formula used is as follows: ； In the formula, is the output of the hidden layer, is the weight matrix between the hidden layer nodes, and the superscript in brackets is the number of hidden layers. is the total number of hidden layers, is the input lunar data, is the bias vector of the hidden layer, is the activation function, the activation function is the exponential linear unit, is the activation function of the last layer; Step S22: artificially add noise to the lunar data, mark the missing values ​​in the lunar data as 0, and obtain lunar labeled data, which is set to have M missing values; Step S23: Use the dropout supplementation technique to regularize the lunar data. During training, randomly delete the nodes in the hidden layer. Record the nodes before the regularization process as the original nodes. Calculate the probability that the original nodes are retained during training. Use the probability to scale the hidden layer. Use the autoencoder to compress the lunar data into low-dimensional data. The formula used is as follows: ； In the formula, It is the low-dimensional data after the moon data is compressed. is the total number of hidden layers after scaling, is the Bernoulli vector in the regularization process, It is the moon marking data; Step S24: Use a decoder to map the low-dimensional data into reconstructed data. The reconstructed data has the same probability distribution and dimension as the lunar data. The reconstructed data is passed backward in the hidden layer and reaches the output layer to obtain supplementary data. The formula used is as follows: ； In the formula, It is to reconstruct the data; Step S25: Use the root mean square error and cross entropy loss function to define the loss function of the data supplementation model, preset the training threshold, and when the loss function is equal to the training threshold, the training ends, otherwise, the training continues. The formula used is as follows: ； In the formula, is the loss function of the data supplementation model, is the missing indicator matrix, is the training cycle, Yes Iterations; Step S26: After the training is completed, the 0 in the lunar data is replaced by the supplementary data, and this is repeated M times to obtain the complete lunar data.

Images