Acoustic boundary element method combined with nerve implicit three-dimensional reconstruction

Abstract

The invention belongs to the technical field of computer vision and acoustic boundary elements, and particularly relates to an acoustic boundary element method combined with neural implicit three-dimensional reconstruction, which comprises the following steps of: acquiring a group of acoustic images based on a single-beam forward-looking sonar; acquiring azimuth angle, elevation angle and range information corresponding to the image data, and calculating sampling points; a neural implicit three-dimensional reconstruction model is constructed, the neural implicit three-dimensional reconstruction model comprises an SDF field and an intensity field which are used for describing geometric information of the target, and the SDF field implicitly represents the three-dimensional shape of the target; constructing an acoustic body rendering equation by taking the image data and the corresponding azimuth information as training data; training a neural implicit three-dimensional reconstruction model; dense sampling is carried out on the trained neural implicit three-dimensional reconstruction model, and discrete representation of the boundary is obtained; and structure surface sound pressure distribution is calculated based on a fast multi-pole boundary element method. According to the method, the processing efficiency and flexibility of the complex geometric object can be improved while relatively high calculation precision is kept.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and acoustic boundary element technology, and specifically relates to an acoustic boundary element method combined with neural implicit three-dimensional reconstruction. Background Art

[0002] The acoustic boundary element method is a classic numerical calculation method that is widely used to solve problems such as acoustic propagation, scattering and radiation. The boundary element method greatly reduces the calculation scale by limiting the solution area to the geometric boundary, and is particularly suitable for acoustic problems in infinite domains. However, when dealing with complex geometric shapes, the geometric modeling and meshing process of the traditional boundary element method is usually very cumbersome, which easily leads to reduced computational efficiency and increased errors. Especially under complex three-dimensional geometric structures, the quality of mesh generation often directly affects the accuracy and stability of acoustic calculations. In recent years, with the development of deep learning technology, neural implicit reconstruction has received widespread attention in the field of three-dimensional geometric modeling. Neural implicit reconstruction can express complex three-dimensional geometric structures in a continuous form by learning the implicit function description of the geometric surface, avoiding the limitations of traditional explicit meshing methods. However, these methods are mainly used in the fields of computer vision and three-dimensional reconstruction, and have not yet been combined with the acoustic boundary element method, and have not fully utilized the potential of neural implicit modeling in solving acoustic problems. Therefore, how to combine neural implicit 3D reconstruction with the acoustic boundary element method to achieve the goal of improving the efficiency and accuracy of acoustic calculations while accurately representing complex geometric structures is a technical problem that needs to be solved urgently. Summary of the invention

[0003] In view of the technical problem that the geometric modeling and meshing process of the above-mentioned traditional boundary element method is usually very cumbersome when dealing with complex geometric shapes, which easily leads to reduced computational efficiency and increased errors, the present invention provides an acoustic boundary element method combined with neural implicit 3D reconstruction, which aims to map the two-dimensional information in the sonar image into an implicit field in three-dimensional space, optimize the network parameters by back propagation, and gradually approach the real surface of the object. By densely sampling the trained neural implicit 3D reconstruction model, a discrete representation of the boundary is obtained, and it is combined with the acoustic boundary element method to solve acoustic problems.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] An acoustic boundary element method combined with neural implicit three-dimensional reconstruction includes the following steps:

[0006] S1, collect a set of acoustic images based on single-beam forward-looking sonar;

[0007] S2, obtain the azimuth, elevation and range information corresponding to the image data, and calculate the sampling points;

[0008] S3, constructing a neural implicit 3D reconstruction model, including an SDF field and an intensity field for describing the target geometric information, and the SDF field implicitly represents the target 3D shape;

[0009] S4, using the image data and the corresponding orientation information as training data to construct an acoustic volume rendering equation;

[0010] S5. Based on back propagation, the SDF field and intensity field are jointly optimized through multiple constraints to train the neural implicit 3D reconstruction model.

[0011] S6, densely sampling the trained neural implicit 3D reconstruction model to obtain a discrete representation of the boundary;

[0012] S7. Calculate the sound pressure distribution on the structure surface based on the fast multipole boundary element method.

[0013] The method for calculating the sampling points in S2 is:

[0014] S21, obtaining the azimuth angle θ corresponding to the image data i ∈[θmax min ], elevation angle φ i ∈[φmax min ] and range information i ∈[rmax min ];

[0015] S22. Construct pixel sample set Randomly select from the image pixels at the same time Pixels whose intensities are higher than the threshold; let the set of these pixels be

[0016] S23. Construct a sampling point set A on the arc p : For each pixel Use layered sampling to sample on the arc corresponding to the pixel; set the elevation angle range [φmax min ] is divided into angles, the difference between two adjacent angles is Add random noise ~Uniform(0,1)Δφ to these angles to get a set of sampling points on the arc

[0017] S24. Construct a sampling point set on the sonar ray For each sampling point P on the arc p , the construction starts from the sonar acoustic center and ends at P p Sonar rays Sample uniformly on each ray points to obtain the sampling point set on the sonar ray

[0018] S25, the final sampling point set is composed of constitute.

[0019] The method for constructing the neural implicit 3D reconstruction model in S3 is:

[0020] S31. Construct SDF field: The spatial position x∈R of any sampling point in three-dimensional space 3 Mapped to its corresponding SDF value sdf∈R 1 , intermediate features f∈R 256 , γ is the position code, is a multi-layer fully connected neural network. is the learnable parameter of the network;

[0021] S32. Construct target density σ: where Φ s (x) = (1 + e ―sx ) ―1 is the sigmoid function, is the gradient information of the SDF field, and s is the learnable standard deviation;

[0022] S33. Construct intensity field: The spatial position x of any sampling point in three-dimensional space and the observation direction d∈R from the camera to the point 3 Mapped to its corresponding intensity value c∈R 3 , F ω is a multi-layer fully connected neural network, and ω is a learnable parameter of the network.

[0023] The method for constructing the acoustic volume rendering equation in S4 is:

[0024] S41, point set Convert to Cartesian coordinates and multiply by the transformation matrix from sonar to world coordinate system

[0025]

[0026] To transform the point set from the sonar local coordinate system to the world coordinate system;

[0027] S42, define the direction of each ray as a unit vector D(P p ):

[0028]

[0029] S43, obtaining intensity and SDF value: intensity and SDF value of the target corresponding to each sampling point predicted and output by the neural implicit 3D reconstruction model;

[0030] S44. Build target rendering items

[0031]

[0032] in, is the arc at (r,θ), is the disturbed point P on the arc p The distance, M(P p ) is the intensity field at point P p The predicted strength at T[P p ] is the discrete transmittance, defined as follows:

[0033]

[0034] where α[P i ] is defined as:

[0035]

[0036] Indicates discrete opacity.

[0037] The method for training the neural implicit 3D reconstruction model in S5 is:

[0038] S51. Construct SDF regularized loss function L eikonal :

[0039]

[0040] in, represents the gradient of the SDF field, and x is any point in three-dimensional space;

[0041] S52. Construct the square error photometric loss function L intensity :

[0042]

[0043] in, is the rendering intensity, I is the real intensity;

[0044] S53. Construct the total loss function L:

[0045] L=L intensity +λ·L eikonal

[0046] Here, λ is a hyperparameter and is set to 0.1.

[0047] The method for obtaining the discrete representation of the boundary in S6 is: after the optimization is completed, the SDF field is densely sampled to obtain the SDF value of each sampling point, and the isosurface is extracted by the Marching Cubes algorithm, and the three-dimensional surface model of the target is generated using the extracted isosurface to obtain the discrete representation of the boundary.

[0048] The method for calculating the sound pressure distribution on the structure surface in S7 is:

[0049] S71. Set boundary conditions and use the Helmholtz equation to establish the CBIE equation in the underwater sound scattering environment;

[0050] S72. Couple the CBIE equation and the HBIE equation to solve the problem of non-unique solution;

[0051] S73, Use regularization techniques to avoid singular integral problems;

[0052] S74. The discrete system equations are established by the point matching method, and the broadband fast multipole algorithm is used to accelerate the solution of the radiation and scattering fields on the surface of the structure.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention uses the compact representation of implicit function networks to reduce the time overhead of geometric modeling and mesh generation, and combines the fast multipole method with an efficient solution algorithm to reduce the time complexity of sound field calculation. In addition, the present invention combines the neural implicit reconstruction method with the acoustic boundary element method for the first time, innovatively solving the problem of efficient coupling between geometric modeling and acoustic solution, and providing a new solution to acoustic problems under complex geometric shapes. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0056] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0057] Figure 1is a flow chart of the steps of the present invention;

[0058] Figure 2 It is a schematic diagram of the network framework of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. These descriptions are only to further illustrate the features and advantages of the present invention, rather than to limit the claims of the present invention. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0060] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0061] See also Figure 1 The embodiment of the present invention provides an acoustic boundary element method combined with neural implicit three-dimensional reconstruction, comprising the following steps:

[0062] S1. Collect a set of acoustic images based on single-beam forward-looking sonar.

[0063] S2, obtain the azimuth, elevation and range information corresponding to the image data, and calculate the sampling points. The specific steps include:

[0064] S21, obtaining the azimuth angle θ corresponding to the image data i ∈[θmax min ], elevation angle φ i ∈[φmax min ] and range information r i ∈[rmax min ].

[0065] S22. Construct pixel sample set Randomly select from the image pixels at the same time Pixels whose intensities are higher than the threshold. Let the set of these pixels be

[0066] S23. Construct a sampling point set A on the arc p : For each pixel Use layered sampling to sample on the arc corresponding to the pixel. min ] is divided into angles, the difference between two adjacent angles is Add random noise ~Uniform(0,1)Δφ to these angles to get a set of sampling points on the arc

[0067] S24. Construct a sampling point set on the sonar ray For each sampling point P on the arc p , the construction starts from the sonar acoustic center and ends at P p Sonar rays Sample uniformly on each ray points to obtain the sampling point set on the sonar ray

[0068] S25, the final sampling point set is composed of constitute.

[0069] S3. Construct a neural implicit 3D reconstruction model, including an SDF field and an intensity field for describing the target geometric information. The SDF field implicitly represents the target 3D shape, such as Figure 2 The specific steps include:

[0070] S31. Construct SDF field: The spatial position x∈R of any sampling point in three-dimensional space 3 Mapped to its corresponding SDF value sdf∈R 1 , intermediate features f∈R 256 , γ is the position code, is a multi-layer fully connected neural network. are the learnable parameters of the network.

[0071] S32. Construct target density σ: where Φ s (x) = (1 + e ―sx ) ―1 is the sigmoid function, is the gradient information of the SDF field, and s is the learnable standard deviation.

[0072] S33. Construct intensity field: The spatial position x of any sampling point in three-dimensional space and the observation direction d∈R from the camera to the point 3 Mapped to its corresponding intensity value c∈R 3 , F ω is a multi-layer fully connected neural network, and ω is a learnable parameter of the network.

[0073] S4. Use the image data and the corresponding orientation information as training data to construct an acoustic volume rendering equation.

[0074] The specific steps include:

[0075] S41, point set Convert to Cartesian coordinates and multiply by the transformation matrix from sonar to world coordinate system

[0076]

[0077] to transform a point set from the sonar local coordinate system to the world coordinate system.

[0078] S42, define the direction of each ray as a unit vector D(P p ):

[0079]

[0080] S43, obtaining intensity and SDF value: the intensity and SDF value of the target corresponding to each sampling point predicted and output by the neural implicit 3D reconstruction model.

[0081] S44. Build target rendering items

[0082]

[0083] in, is the arc at (r,θ), is the disturbed point P on the arc p The distance, M(P p ) is the intensity field at point P p The predicted strength at T[P p ] is the discrete transmittance, defined as follows:

[0084]

[0085] where α[P i ] is defined as:

[0086]

[0087] Indicates discrete opacity.

[0088] S5. Based on back propagation, the SDF field and the intensity field are jointly optimized through multiple constraints to train the neural implicit 3D reconstruction model. The specific steps include:

[0089] S51. Construct SDF regularized loss function L eikonal :

[0090]

[0091] in, Represents the gradient of the SDF field, and x is any point in three-dimensional space.

[0092] S52. Construct the square error photometric loss function L intensity :

[0093]

[0094] in, is the rendering intensity and I is the real intensity.

[0095] S53. Construct the total loss function L:

[0096] L=L intensity +λ·L eikonal

[0097] Here, λ is a hyperparameter and is set to 0.1.

[0098] S6. Densely sample the trained neural implicit 3D reconstruction model to obtain a discrete representation of the boundary. The specific form is: after the optimization is completed, densely sample the SDF field to obtain the SDF value of each sampling point, and extract the isosurface through the Marching Cubes algorithm, and use the extracted isosurface to generate a 3D surface model of the target to obtain a discrete representation of the boundary.

[0099] S7. Calculate the sound pressure distribution on the structure surface based on the fast multipole boundary element method. The specific steps include:

[0100] S71. Set boundary conditions and use the Helmholtz equation to establish the CBIE equation in the underwater sound scattering environment.

[0101] S72. Couple the CBIE equation and the HBIE equation to solve the problem of non-unique solution.

[0102] S73. Use regularization techniques to avoid singular integral problems.

[0103] S74. The discrete system equations are established by the point matching method, and the broadband fast multipole algorithm is used to accelerate the solution of the radiation and scattering fields on the surface of the structure.

[0104] Only the preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention, and various changes should be included in the protection scope of the present invention.

Claims

1. An acoustic boundary element method combined with neural implicit three-dimensional reconstruction, characterized in that: The following steps are involved: S1, collect a set of acoustic images based on single-beam forward-looking sonar; S2, obtain the azimuth, elevation and range information corresponding to the image data, and calculate the sampling points; S3, constructing a neural implicit 3D reconstruction model, including an SDF field and an intensity field for describing the target geometric information, and the SDF field implicitly represents the target 3D shape; S4, using the image data and the corresponding orientation information as training data to construct an acoustic volume rendering equation; S5. Based on back propagation, the SDF field and intensity field are jointly optimized through multiple constraints to train the neural implicit 3D reconstruction model. S6, densely sampling the trained neural implicit 3D reconstruction model to obtain a discrete representation of the boundary; S7. Calculate the sound pressure distribution on the structure surface based on the fast multipole boundary element method.

2. The acoustic boundary element method combined with neural implicit three-dimensional reconstruction according to claim 1, characterized in that: The method for calculating the sampling points in S2 is: S21, obtaining the azimuth angle θ corresponding to the image data i ∈[θmax min ], elevation angle φ i ∈[φmax min ] and range information i ∈[rmax min ]; S22. Construct pixel sample set Randomly select from the image pixels at the same time Pixels whose intensities are higher than the threshold; let the set of these pixels be S23. Construct a sampling point set A on the arc p : For each pixel Use layered sampling to sample on the arc corresponding to the pixel; set the elevation angle range [φmax min ] is divided into angles, the difference between two adjacent angles is Add random noise ~Uniform(0,1)Δφ to these angles to get a set of sampling points on the arc S24. Construct a sampling point set on the sonar ray For each sampling point P on the arc p , the construction starts from the sonar acoustic center and ends at P p Sonar rays Sample uniformly on each ray points to obtain the sampling point set on the sonar ray S25, the final sampling point set is composed of constitute.

3. The acoustic boundary element method combined with neural implicit three-dimensional reconstruction according to claim 1, characterized in that: The method for constructing the neural implicit 3D reconstruction model in S3 is: S31. Construct SDF field: The spatial position x∈R of any sampling point in three-dimensional space 3 Mapped to its corresponding SDF value sdf∈R 1 , intermediate features f∈R 256 , γ is the position code, is a multi-layer fully connected neural network. is the learnable parameter of the network; S32. Construct target density σ: where Φ s (x) = (1 + e ―sx ) ―1 is the sigmoid function, is the gradient information of the SDF field, and s is the learnable standard deviation; S33. Construct intensity field: The spatial position x of any sampling point in three-dimensional space and the observation direction d∈R from the camera to the point 3 Mapped to its corresponding intensity value c∈R 3 , F ω is a multi-layer fully connected neural network, and ω is a learnable parameter of the network.

4. The acoustic boundary element method combined with neural implicit three-dimensional reconstruction according to claim 1, characterized in that: The method for constructing the acoustic volume rendering equation in S4 is: S41, point set Convert to Cartesian coordinates and multiply by the transformation matrix from sonar to world coordinate system To transform the point set from the sonar local coordinate system to the world coordinate system; S42, define the direction of each ray as a unit vector D(P p ): S43, obtaining intensity and SDF value: intensity and SDF value of the target corresponding to each sampling point predicted and output by the neural implicit 3D reconstruction model; S44. Build target rendering items in, is the arc at (r,θ), is the disturbed point P on the arc p The distance, M(P p ) is the intensity field at point P p The predicted strength at T[P p ] is the discrete transmittance, defined as follows: where α[P i ] is defined as: Indicates discrete opacity.

5. The acoustic boundary element method combined with neural implicit three-dimensional reconstruction according to claim 1, characterized in that: The method for training the neural implicit 3D reconstruction model in S5 is: S51. Construct SDF regularized loss function L eikonal : in, represents the gradient of the SDF field, and x is any point in three-dimensional space; S52. Construct the square error photometric loss function L intensity : in, is the rendering intensity, I is the real intensity; S53. Construct the total loss function L: L=L intensity +λ·L eikonal Here, λ is a hyperparameter and is set to 0.

1.

6. The acoustic boundary element method combined with neural implicit three-dimensional reconstruction according to claim 1, characterized in that: The method for obtaining the discrete representation of the boundary in S6 is: after the optimization is completed, the SDF field is densely sampled to obtain the SDF value of each sampling point, and the isosurface is extracted by the Marching Cubes algorithm, and the three-dimensional surface model of the target is generated using the extracted isosurface to obtain the discrete representation of the boundary.

7. The acoustic boundary element method combined with neural implicit three-dimensional reconstruction according to claim 1, characterized in that: The method for calculating the sound pressure distribution on the structure surface in S7 is: S71. Set boundary conditions and use the Helmholtz equation to establish the CBIE equation in the underwater sound scattering environment; S72. Couple the CBIE equation and the HBIE equation to solve the problem of non-unique solution; S73, Use regularization techniques to avoid singular integral problems; S74. The discrete system equations are established by the point matching method, and the broadband fast multipole algorithm is used to accelerate the solution of the radiation and scattering fields on the surface of the structure.

Images