VOCs fluid three-dimensional reconstruction method and device based on neural fluid field

Abstract

The invention discloses a VOCs fluid three-dimensional reconstruction method and device based on a neural fluid field, and belongs to the technical field of computers. The VOCs fluid three-dimensional reconstruction method based on the neural fluid field comprises the following steps: inputting an obtained sample fluid image corresponding to petrochemical gas in a sample time period in a target space and a sample label corresponding to the sample fluid image into a neural fluid field model to obtain a first fluid image corresponding to the petrochemical gas output by the neural fluid field model; constructing a target loss function based on at least one of the first fluid image, the sample fluid image and the sample label; and training the neural fluid field model based on the target loss function. According to the VOCs fluid three-dimensional reconstruction method based on the neural fluid field, the velocity field and the density field of the VOCs fluid can be recovered from multi-view video data, the method is suitable for processing the VOCs fluid lacking stable visual features, real-time monitoring, three-dimensional reconstruction and other applications can be better supported, and the detection precision is high.

Description

Technical Field

[0001] The present application belongs to the field of computer technology, and in particular, relates to a method and device for three-dimensional reconstruction of VOCs fluid based on a neural fluid field. Background Art

[0002] Petrochemical volatile organic compounds (VOCs) are a type of highly polluting gas with a wide range of sources. They have a significant impact on regional atmospheric ozone pollution and PM2.5 pollution. VOCs need to be monitored in real time in order to deal with the gas in a timely manner. In the related art, there is a method of reconstructing VOCs fluids using optical flow technology. Since VOCs fluids usually do not have a fixed shape and stable visual features, this method is difficult to accurately estimate the density and velocity of the fluid, making it difficult to reconstruct VOCs gas in three dimensions, and the detection accuracy is poor, making it difficult to apply to real-time monitoring of VOCs. Summary of the invention

[0003] The present application aims to solve at least one of the technical problems existing in the related art. To this end, the present application proposes a method and device for three-dimensional reconstruction of VOCs fluid based on neural fluid field. The neural fluid field model can recover the velocity field and density field of VOCs fluid from multi-view video data, which is suitable for processing VOCs fluid that lacks stable visual features, and can better support applications such as real-time monitoring and three-dimensional reconstruction, with high detection accuracy.

[0004] In a first aspect, the present application provides a method for three-dimensional reconstruction of VOCs fluid based on a neural fluid field, the method comprising:

[0005] Inputting the sample fluid image corresponding to the petrochemical gas in the target space within the sample period and the sample label corresponding to the sample fluid image into the neural fluid field model, obtaining a first fluid image corresponding to the petrochemical gas output by the neural fluid field model; the sample label includes at least one of the actual density field and the actual velocity field of the petrochemical gas in the target space;

[0006] constructing a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label;

[0007] The neural fluid field model is trained based on the target loss function; the neural fluid field model is used to reconstruct at least one of a target density field and a target velocity field of the petrochemical gas in the target space, and at least one of the target density field and the target velocity field is used to perform three-dimensional reconstruction of the fluid corresponding to the petrochemical gas.

[0008] According to the neural fluid field-based VOCs fluid three-dimensional reconstruction method provided in an embodiment of the present application, a neural fluid field model is constructed to restore the density field and velocity field of petrochemical gases in a sample fluid image to obtain a first fluid image, and a target loss function is constructed based on the first fluid image, the sample fluid image, and at least one of the sample labels corresponding to the sample fluid image. The neural fluid field model is then trained based on the target loss function to improve the precision and accuracy of the neural fluid field model, so that the neural fluid field model can restore the velocity field and density field of the VOCs fluid from multi-perspective video data. The neural fluid field model is suitable for processing VOCs fluids that lack stable visual features, can better support applications such as real-time monitoring and three-dimensional reconstruction, and has high detection accuracy.

[0009] In one embodiment of the present application, a method for three-dimensional reconstruction of VOCs fluid based on a neural fluid field is provided, wherein a target loss function is constructed based on at least one of the first fluid image, the sample fluid image, and the sample label, including:

[0010] Based on at least one of the first fluid image, the sample fluid image and the sample label, construct at least one of a density loss function, a projection loss function, a laminar regularization loss function and a rendering loss function;

[0011] The target loss function is obtained by fusing at least one of the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function.

[0012] In one embodiment of the present application, a method for three-dimensional reconstruction of VOCs fluid based on a neural fluid field is provided, wherein at least one of a density loss function, a projection loss function, a laminar regularization loss function, and a rendering loss function is constructed based on at least one of the first fluid image, the sample fluid image, and the sample label, including:

[0013] Constructing the density loss function based on the product of the diffusion rate of the density of the petrochemical gas along a target dimension among the multiple spatial dimensions and the speed of the petrochemical gas along the target dimension, and the density change rate of the petrochemical gas;

[0014] constructing the projection loss function based on the difference between the actual velocity field and the divergence corresponding to the actual velocity field;

[0015] Constructing the laminar regularization loss function based on the actual velocity field and the actual density field;

[0016] The rendering loss function is constructed based on the first fluid image and the sample fluid image.

[0017] In one embodiment of the present application, a method for three-dimensional reconstruction of VOCs fluid based on a neural fluid field is provided. The laminar regularization loss function is constructed based on the actual velocity field and the actual density field, including:

[0018] Constructing a hinge loss function based on the actual velocity field and the actual density field;

[0019] Based on the hinge loss function, the laminar regularization loss function is constructed.

[0020] In one embodiment of the present application, a method for three-dimensional reconstruction of VOCs fluid based on a neural fluid field is provided, wherein the rendering loss function is constructed based on the first fluid image and the sample fluid image, including:

[0021] Acquire the rendering brightness of the first fluid image at a target camera position and a target ray direction, and the observed brightness of the sample fluid image at the target camera position and the target ray direction;

[0022] The rendering loss function is constructed based on the difference between the rendering brightness and the observed brightness.

[0023] In a method for 3D reconstruction of VOCs fluid based on a neural fluid field according to an embodiment of the present application, the target loss function is obtained by fusing at least one of the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function, including:

[0024] Obtaining the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function, and the loss weight corresponding to the rendering loss function;

[0025] Based on the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function and the loss weight corresponding to the rendering loss function, the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function are weightedly summed to obtain the target loss function.

[0026] In one embodiment of the present application, a method for three-dimensional reconstruction of VOCs fluid based on a neural fluid field is provided, wherein the neural fluid field model is constructed based on the following steps:

[0027] The neural velocity field is constructed based on the neural graphics primitives, and the residual turbulence velocity field driven by the vortex particles is constructed based on the target motion equations;

[0028] The neural fluid field model is constructed based on the base neural velocity field and the residual turbulent velocity field driven by the vortex particles.

[0029] In a method for 3D reconstruction of VOCs fluid based on a neural fluid field according to an embodiment of the present application, after training the neural fluid field model based on the target loss function, the method further includes:

[0030] Acquire an initial fluid image corresponding to the petrochemical gas in the target space within a target time period;

[0031] The initial fluid image is input into a trained neural fluid field model to obtain a target fluid image corresponding to the petrochemical gas output by the trained neural fluid field model.

[0032] In a second aspect, the present application provides a VOCs fluid three-dimensional reconstruction device based on a neural fluid field, comprising:

[0033] A first processing module is used to input the sample fluid image corresponding to the petrochemical gas in the target space within the sample period and the sample label corresponding to the sample fluid image into the neural fluid field model to obtain a first fluid image corresponding to the petrochemical gas output by the neural fluid field model; the sample label includes at least one of the actual density field and the actual velocity field of the petrochemical gas in the target space;

[0034] A second processing module, configured to construct a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label;

[0035] The third processing module is used to train the neural fluid field model based on the target loss function; the neural fluid field model is used to reconstruct at least one of the target density field and the target velocity field of the petrochemical gas in the target space, and at least one of the target density field and the target velocity field is used to perform three-dimensional reconstruction of the fluid corresponding to the petrochemical gas.

[0036] According to the VOCs fluid three-dimensional reconstruction device based on neural fluid field provided in the embodiment of the present application, a neural fluid field model is constructed to restore the density field and velocity field of petrochemical gas in the sample fluid image to obtain a first fluid image, and a target loss function is constructed based on the first fluid image, the sample fluid image, and at least one of the sample labels corresponding to the sample fluid image. The neural fluid field model is then trained based on the target loss function to improve the precision and accuracy of the neural fluid field model, so that the neural fluid field model can restore the velocity field and density field of VOCs fluid from multi-perspective video data. It is suitable for processing VOCs fluids that lack stable visual features, can better support applications such as real-time monitoring and three-dimensional reconstruction, and has high detection accuracy.

[0037] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for three-dimensional reconstruction of VOCs fluid based on the neural fluid field as described in the first aspect above is implemented.

[0038] In a fourth aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the VOCs fluid three-dimensional reconstruction method based on the neural fluid field as described in the first aspect above.

[0039] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the VOCs fluid three-dimensional reconstruction method based on the neural fluid field as described in the first aspect above.

[0040] The above one or more technical solutions in the embodiments of the present application have at least one of the following technical effects:

[0041] A neural fluid field model is constructed to restore the density field and velocity field of petrochemical gases in a sample fluid image to obtain a first fluid image, and a target loss function is constructed based on the first fluid image, the sample fluid image, and at least one of the sample labels corresponding to the sample fluid image. The neural fluid field model is then trained based on the target loss function to improve the precision and accuracy of the neural fluid field model, so that the neural fluid field model can restore the velocity field and density field of VOCs fluid from multi-perspective video data. This is suitable for processing VOCs fluids that lack stable visual features, can better support applications such as real-time monitoring and three-dimensional reconstruction, and has high detection accuracy.

[0042] Furthermore, by constructing physical losses, the velocity field inferred by the neural fluid field model is divergence free and can drive the transmission of the density field. In addition, the density field is regularized based on differentiable volume rendering and physical fluid transmission constraints to improve the accuracy of the neural fluid field model in recovering 2D video data to a continuous 3D density field. By weightedly combining the multiple physical losses constructed, the parameters of the neural fluid field model can be jointly optimized based on all loss terms, so that the fluid density and velocity fields recovered from the video are consistent with both physical laws and the observed visual data, thereby improving the prediction precision and accuracy of the neural fluid field model.

[0043] Furthermore, by adopting a hybrid neural velocity representation, the VOCs fluid velocity field is decomposed into a basic neural velocity field and a residual turbulent velocity field driven by vortex particles. The large-scale flow characteristics of the VOCs fluid can be obtained using the basic neural velocity field, and the small-scale vortex characteristics of the VOCs fluid can be obtained based on the residual turbulent velocity field driven by vortex particles, so that the neural fluid field model can better obtain the turbulent characteristics of the fluid velocity.

[0044] Furthermore, by constructing and training a neural fluid field model and using it to recover the fluid density and velocity field of VOCs gas from video data, three-dimensional reconstruction of VOCs fluid smoke is achieved, which can be used to detect VOCs leakage and trace its source, etc., with high detection precision and accuracy.

[0045] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0047] Figure 1 It is a schematic diagram of the process of the VOCs fluid three-dimensional reconstruction method based on the neural fluid field provided in the embodiment of the present application;

[0048] Figure 2 It is a schematic diagram of the principle of the VOCs fluid three-dimensional reconstruction method based on the neural fluid field provided in the embodiment of the present application;

[0049] Figure 3 It is a schematic diagram of the results of the VOCs fluid three-dimensional reconstruction method based on the neural fluid field provided in the embodiment of the present application;

[0050] Figure 4 It is a schematic structural diagram of a VOCs fluid three-dimensional reconstruction device based on a neural fluid field provided in an embodiment of the present application;

[0051] Figure 5 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0053] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0054] In combination with the accompanying drawings, the VOCs fluid three-dimensional reconstruction method based on the neural fluid field, the VOCs fluid three-dimensional reconstruction device based on the neural fluid field, the electronic device and the readable storage medium provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0055] Among them, the VOCs fluid three-dimensional reconstruction method based on the neural fluid field can be applied to the terminal, and can be specifically executed by the hardware or software in the terminal.

[0056] The terminal includes, but is not limited to, a portable communication device such as a mobile phone or tablet computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad). It should also be understood that in some embodiments, the terminal may not be a portable communication device, but a desktop computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad).

[0057] In the following various embodiments, a terminal including a display and a touch-sensitive surface is described. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, a mouse and a joystick.

[0058] The embodiment of the present application provides a method for three-dimensional reconstruction of VOCs fluid based on neural fluid field. The execution subject of the method for three-dimensional reconstruction of VOCs fluid based on neural fluid field can be an electronic device or a functional module or functional entity in the electronic device that can implement the method for three-dimensional reconstruction of VOCs fluid based on neural fluid field. The electronic devices mentioned in the embodiment of the present application include but are not limited to mobile phones, tablet computers, computers, cameras and wearable devices. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field provided in the embodiment of the present application is explained below by taking the electronic device as the execution subject as an example.

[0059] like Figure 1 As shown, the VOCs fluid three-dimensional reconstruction method based on the neural fluid field includes: step 110, step 120 and step 130.

[0060] Step 110, inputting the sample fluid image corresponding to the petrochemical gas in the target space within the sample period and the sample label corresponding to the sample fluid image into the neural fluid field model, and obtaining a first fluid image corresponding to the petrochemical gas output by the neural fluid field model;

[0061] In this step, the petrochemical gas is petrochemical volatile organic compounds (VOCs).

[0062] The sample fluid image may include a fluid video corresponding to the VOCs collected under at least one collection viewing angle, and the fluid video includes multiple video frames.

[0063] The sample fluid image may include an image or video actually collected, and may also include an image or video corresponding to the petrochemical gas obtained by simulation.

[0064] For example, multiple cameras at fixed positions may be arranged in the target space to obtain multi-view video data of the same scene.

[0065] The sample period is a collection period corresponding to the sample fluid image. For example, video data of petrochemical gas from time A to time B may be collected.

[0066] The sample fluid image is used to characterize the flow state of petrochemical gas during the sample period.

[0067] The sample label includes at least one of an actual density field and an actual velocity field of the petrochemical gas in the target space.

[0068] The actual density field is used to characterize the density distribution of the VOCs fluid in the target space. The actual density field may include the rate of change of the density of the VOCs fluid over time and the diffusion rate of the density along multiple spatial dimensions.

[0069] The actual velocity field is used to characterize the flow direction and velocity of the VOCs fluid. The actual velocity field may include the velocity vectors of the fluid at each point in space.

[0070] The neural fluid field model is used to recover the density field and velocity field of the VOCs fluid from the sample fluid image, and then render the density field and velocity field of the VOCs fluid to the first fluid image.

[0071] The neural fluid field model can be trained end-to-end to ensure that the neural fluid field model can recover the density field and velocity field of VOCs fluid from multi-view video data.

[0072] In actual implementation, multi-view video data of the same scene can be captured by multiple cameras at fixed positions, and the evaluation can be performed using synthetic simulations using real data.

[0073] The real data set can be obtained based on the ScalarFlow data set. The real data of the first five scenes can be determined as the real data set. The real data set in each scene includes the video data collected under five acquisition perspectives. The five cameras are in fixed positions throughout the entire acquisition process and are evenly distributed on a 120° arc centered on the rising smoke.

[0074] The frame rate of each video data is 150 frames, the resolution is 1062×600, and the video data has been processed to remove background information. For each scene, four video data can be used for training and one video data can be used for testing.

[0075] It is also possible to evaluate 3D velocity and density using synthetic simulations. Synthetic data can be created through the ScalarFlow synthetic dataset generation code, including scenarios with different inflow sources and different viscosities. For example, five scenarios with different inflow sources and high viscosity and another five scenarios with lower viscosity can be generated for model training.

[0076] Step 120: construct a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label;

[0077] In this step, the objective loss function is used to characterize the difference between the output of the neural fluid field model and the actual fluid state.

[0078] The objective loss function can be used to constrain the density field and velocity field output by the neural fluid field model to be physically reasonable.

[0079] In the actual execution process, different sub-loss functions can be constructed based on different physical quantities. For example, a sub-loss function for constraining the density field output by the model can be constructed based on density, and a sub-loss function for constraining the velocity field output by the model can be constructed based on velocity. Multiple sub-loss functions can be fused to obtain the target loss function.

[0080] Step 130: Train the neural fluid field model based on the target loss function.

[0081] In this step, the neural fluid field model is used to reconstruct at least one of a target density field and a target velocity field of petrochemical gas in a target space.

[0082] The neural fluid field model can recover the density field and velocity field of petrochemical gas from the image data corresponding to the petrochemical gas.

[0083] After reconstructing the target density field and target velocity field of petrochemical gas in the target space, the VOCs fluid can be reconstructed in three dimensions, thereby realizing the 3D visualization of the detected VOCs fluid.

[0084] The neural fluid field model can also be used for applications such as fluid simulation, editing, and future prediction.

[0085] The neural fluid field model can predict the future state of the fluid based on historical data and can be applied to fields such as environmental pollution early warning and industrial process control.

[0086] In the process of training the neural fluid field model, the target loss function can update the corresponding model parameters in the neural fluid field model based on the loss value between the predicted value and the true value to reduce the loss between the predicted value and the true value, making the output result of the neural fluid field model more precise and accurate.

[0087] According to the neural fluid field-based VOCs fluid three-dimensional reconstruction method provided in an embodiment of the present application, a neural fluid field model is constructed to restore the density field and velocity field of petrochemical gases in a sample fluid image to obtain a first fluid image, and a target loss function is constructed based on the first fluid image, the sample fluid image, and at least one of the sample labels corresponding to the sample fluid image. The neural fluid field model is then trained based on the target loss function to improve the precision and accuracy of the neural fluid field model, so that the neural fluid field model can restore the velocity field and density field of the VOCs fluid from multi-perspective video data. The neural fluid field model is suitable for processing VOCs fluids that lack stable visual features, can better support applications such as real-time monitoring and three-dimensional reconstruction, and has high detection accuracy.

[0088] In some embodiments, step 120 may include:

[0089] Based on at least one of the first fluid image, the sample fluid image, and the sample label, construct at least one of a density loss function, a projection loss function, a laminar regularization loss function, and a rendering loss function;

[0090] At least one of the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function is fused to obtain the target loss function.

[0091] In this embodiment, the density loss function is constructed, and the density transmission equation can be used to ensure that the velocity field learned by the neural fluid field model can reasonably transmit the density field.

[0092] To construct the projection loss function, a pressure projection solver can be used to make the velocity field divergent, so as to meet the physical properties of incompressible fluids and ensure physical rationality.

[0093] Constructing a laminar regularized loss function can improve velocity reconstruction in laminar flow (a state where the fluid flows slowly and steadily), and thus improve the velocity field reconstructed by the neural fluid field model.

[0094] The density field can be regularized based on differentiable volume rendering and physical fluid transport constraints.

[0095] Based on at least one of the first fluid image, the sample fluid image and the sample label, at least one loss function can be constructed, and then by fusing the at least one loss function, a target loss function can be obtained.

[0096] In some embodiments, constructing at least one of a density loss function, a projection loss function, a laminar regularization loss function, and a rendering loss function based on at least one of the first fluid image, the sample fluid image, and the sample label may include:

[0097] A density loss function is constructed based on the product of the diffusion rate of the density of the petrochemical gas along the target dimension in multiple spatial dimensions and the velocity of the petrochemical gas along the target dimension, as well as the density change rate of the petrochemical gas;

[0098] Based on the difference between the actual velocity field and the divergence corresponding to the actual velocity field, a projection loss function is constructed;

[0099] Based on the actual velocity field and the actual density field, a laminar regularization loss function is constructed;

[0100] A rendering loss function is constructed based on the first fluid image and the sample fluid image.

[0101] In this embodiment, the target dimension is any one of the multiple spatial dimensions. For example, the multiple spatial dimensions may include x, y, and z.

[0102] The velocity field may include three components, representing velocities in the x, y, and z directions respectively.

[0103] For a target dimension in multiple spatial dimensions, a density loss function can be constructed based on the diffusion rate of the density of the petrochemical gas along the target dimension, the speed of the petrochemical gas along the target dimension, and the density change rate of the petrochemical gas.

[0104] In actual implementation, the density loss can be constructed Use the density transport equation to ensure that the learned velocity field can reasonably transfer the density field. The density transport equation in incompressible flow can be used. Introducing physical information supervision signals, the density loss function is as follows:

[0105]

[0106] in, is the density loss function, is the expectation operator, which is used to average the loss over the entire data set. x, y, and z are three spatial dimensions. σ is the density. u, v, and w are the velocities of petrochemical gas in the x, y, and z directions, respectively. is the rate of change of density over time (i.e., density change rate), and are the diffusion rates of density along the three spatial dimensions, respectively.

[0107] The density penalty states that the three-dimensional velocity field (u, v, w) at any given time t should transmit the density σ so that it evolves into the density field at the next time.

[0108] The divergence corresponding to the actual velocity field can be calculated by the pressure projection solver, and then the projection loss function is constructed according to the difference between the actual velocity field and the divergence corresponding to the actual velocity field.

[0109] In the actual implementation process, the projection loss is realized The pressure projection solver can be used to make the velocity field divergent, conform to the physical properties of incompressible fluids, and ensure physical rationality.

[0110] By using the pressure projection solver, the learned velocity field is constrained to be divergence-free, and the projection loss function is as follows:

[0111]

[0112] in, is the projection loss function, The expectation of the spatial coordinates (x, y, z) and time t, that is, the average loss over the entire flow field and time period, is u p is the velocity field calculated by the pressure projection solver, which is the projection of the original velocity field u on the divergence-free subspace, ||*|| 2 is the squared Euclidean norm, which is used to calculate the difference between the actual velocity field and the divergence corresponding to the actual velocity field.

[0113] In some embodiments, constructing a laminar regularization loss function based on the actual velocity field and the actual density field may include:

[0114] Based on the actual velocity field and the actual density field, a hinge loss function is constructed;

[0115] Based on the hinge loss function, a laminar regularization loss function is constructed.

[0116] In this embodiment, when the actual velocity field and the actual density field meet the conditions, the hinge loss function will have an impact on the laminar regularization loss function.

[0117] The laminar regularization loss function can encourage the neural fluid field model to output non-zero velocity fields in high-density areas, thereby improving velocity reconstruction in laminar areas.

[0118] In the actual implementation process, the laminar regularization loss function can be constructed based on the following formula:

[0119]

[0120] in, is the laminar regularization loss function, is the expectation of the spatial coordinates (x, y, z) and time t, that is, the loss is averaged over the entire flow field and time period, max(0,γσ-||u||) is the hinge loss function, and the hinge loss function contributes to the laminar regularization loss when γσ-||u||>0, that is, the loss function will only increase when the size of the velocity field u is less than a threshold determined by γ and density σ.

[0121] γ is a hyperparameter used to scale the threshold according to the velocity unit. The hyperparameter can be adjusted according to actual conditions to balance the accuracy and regularization strength of the velocity field.

[0122] σ is the density, which is used to characterize the mass distribution of the fluid. In laminar flow, the partial derivative of density may be zero, but in areas where there is a difference in mass distribution, the absolute value of density can be non-zero.

[0123] ||u|| is the modulus (or size) of the velocity field u, that is, the absolute value of the velocity, which is used to characterize the speed of fluid movement.

[0124] In some embodiments, constructing a rendering loss function based on the first fluid image and the sample fluid image may include:

[0125] Acquire the rendering brightness of the first fluid image at the target camera position and the target ray direction, and the observed brightness of the sample fluid image at the target camera position and the target ray direction;

[0126] Based on the difference between the rendered brightness and the observed brightness, a rendering loss function is constructed.

[0127] In this embodiment, the target camera position and target ray direction may be determined based on the user.

[0128] The rendering brightness is used to characterize the luminosity rendered at the target camera position and the target ray direction, and the observed brightness is the luminosity observed at the target camera position and the target ray direction.

[0129] The difference between the rendering brightness and the observed brightness may be calculated based on a structural similarity algorithm or a square Euclidean norm, or may be calculated based on other algorithms. The selection may be based on user needs and is not limited in this application.

[0130] In the actual implementation process, density, appearance and speed can be learned together, and based on the visual imaging signal and the physical fluid transport constraints, a rendering loss function as shown in the following formula is constructed:

[0131]

[0132] in, is the rendering loss function, which measures the difference between the rendered image (first fluid image) and the observed (real) image (sample fluid image), is the expectation operator, which is used to average the loss over the entire dataset, o is the camera position, d is the ray direction, t is the time, and L render (o, d) is the luminosity (brightness) rendered at camera position o and direction d, L observe (o, d): is the luminosity (brightness) observed at camera position o and direction d, ||*|| 2 is the squared Euclidean norm, used to calculate the difference between two photometric values.

[0133] In some embodiments, fusing at least one of a density loss function, a projection loss function, a laminar regularization loss function, and a rendering loss function to obtain a target loss function may include:

[0134] Get the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function, and the loss weight corresponding to the rendering loss function;

[0135] Based on the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function and the loss weight corresponding to the rendering loss function, the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function are weightedly summed to obtain the target loss function.

[0136] In this embodiment, the objective loss function can be designed The rendering loss function, density loss function, projection loss function and laminar regularization loss function are combined for joint optimization.

[0137] A variety of weight combination schemes can be set, and then the performance of the neural fluid field model under various weight combination schemes can be evaluated based on the training set and the validation set, and the weight combination scheme can be selected based on the validation results.

[0138] Alternatively, the weights corresponding to each loss function may be optimized based on other methods, which are not limited in this application.

[0139] In the actual execution process, the target loss function can be constructed based on the following formula:

[0140]

[0141] in, is the target loss function, is the rendering loss function, βrender is the loss weight corresponding to the rendering loss function, is the density loss function, β density is the loss weight corresponding to the density loss function, is the projection loss function, β proj is the loss weight corresponding to the projection loss function, is the laminar regularization loss function, β laminar is the weight corresponding to the laminar regularization loss function.

[0142] The density loss function, projection loss function, laminar regularization loss function and rendering loss function can jointly optimize the model parameters to ensure that the fluid density and velocity fields recovered from the video are both physically consistent and consistent with the observed visual data.

[0143] According to the method for three-dimensional reconstruction of VOCs fluid based on neural fluid field provided in the embodiment of the present application, by constructing physical losses, the velocity field inferred by the neural fluid field model is divergence free and can drive the transmission of the density field; in addition, the density field is regularized based on differentiable volume rendering and physical fluid transmission constraints to improve the accuracy of the neural fluid field model in recovering from 2D video data to a continuous 3D density field. By performing a weighted combination of multiple constructed physical losses, the parameters of the neural fluid field model can be jointly optimized based on all loss terms, so that the fluid density and velocity fields recovered from the video are consistent with both physical laws and observed visual data, thereby improving the prediction precision and accuracy of the neural fluid field model.

[0144] like Figure 2 As shown, in some embodiments, the neural fluid field model can be constructed based on the following steps:

[0145] The neural velocity field is constructed based on the neural graphics primitives, and the residual turbulence velocity field driven by the vortex particles is constructed based on the target motion equations;

[0146] A neural fluid field model is constructed based on the base neural velocity field and the residual turbulent velocity field driven by vortex particles.

[0147] In this embodiment, the Instant Neural Graphics Primitives is a graphics foundation that uses neural networks as basic building blocks to express and render scene surfaces through mathematical functions.

[0148] The neural velocity field can be used to obtain the large-scale flow motion of VOCs fluid.

[0149] The target equations of motion may include the Navier-Stokes equations (NS equations), which are equations of motion that characterize the conservation of momentum for a viscous incompressible fluid.

[0150] The residual turbulent velocity field driven by vortex particles can be used to obtain the small-scale vortex details of VOCs fluid.

[0151] That is, the underlying flow field can be decomposed into the basal neural velocity field and the residual turbulent velocity field driven by the vortex particles.

[0152] In the actual implementation process, a hybrid neural velocity representation can be designed, including the basic neural velocity field and the residual turbulent velocity field driven by vortex particles, to capture the turbulent characteristics of the VOCs fluid velocity. The specific steps include:

[0153] S1: Set the bottom flow field u=u base +u vort Decomposed into basis neural velocity field u base and the residual vorticity driving velocity field u vort The base neural velocity field captures the large-scale flow motion, and the vorticity driven velocity represents the small-scale vortex details.

[0154] S2: Create the basis neural velocity field (u base ), extended to the temporal domain using instanced neural graphics primitives (iNGP) to capture large-scale flows.

[0155] S3: Create the residual turbulent velocity field driven by the vortex particles (u vort ), a particle-based approach is used to represent small-scale vortex details. vort By exploiting the physical structure of turbulence to complement the base flow, the physical model behind vorticity-driven flow can be expressed in the curled form of the Navier-Stokes equations:

[0156]

[0157] in, is the vorticity, u is the velocity field, v is the dynamic viscosity, and f is the external volume force. In the case of inviscid fluid, there is only the vortex stretching term. is not zero, the equation becomes the vorticity transport equation, which means that the vorticity is stretched and advected by the velocity flow.

[0158] In the vortex particle method, a particle-based vorticity representation can be used, which is low-dimensional and easy to evolve in time, and can be embedded in the cyclic physical flow structure.

[0159] Vortex particles induced vorticity driven u vort (x,t) can be expressed as:

[0160]

[0161] Among them, u vort(x, t) is the residual turbulent velocity field driven by the vortex particles, N p is the unit vector from the particle position to x, is the adjusted vorticity, intensity I p ,Location and vorticity The triplet state represents a vortex particle p at a certain timestamp t.

[0162] K is a Gaussian distribution, used to smooth data. Among them, r is the width parameter of the kernel.

[0163] S4: We can assume that most of the energy is captured by the base flow and only use the base flow to transport the vortex particles. First, we can learn the base flow and fix it to learn the residual flow. Second, we can get the learning parameters from the particle transport based on the seeding strategy, seed enough vortex particles at high curl space-time positions, and pre-calculate all t and p trajectories through the learned base flow. and The learnable parameters can then be simplified to I p The redundant vortex particles are automatically suppressed by learning to have zero intensity. By simplifying step S4, the situation of poor local minimum value caused by directly learning vortex particles can be effectively avoided.

[0164] According to the VOCs fluid three-dimensional reconstruction method based on neural fluid field provided in an embodiment of the present application, by adopting a hybrid neural velocity representation, the VOCs fluid velocity field is decomposed into a basic neural velocity field and a residual turbulent velocity field driven by vortex particles. The large-scale flow characteristics of the VOCs fluid can be obtained using the basic neural velocity field, and the small-scale vortex characteristics of the VOCs fluid can be obtained based on the residual turbulent velocity field driven by vortex particles, so that the neural fluid field model can better obtain the turbulent characteristics of the fluid velocity.

[0165] like Figure 3 As shown, Figure 3 (d) illustrates a sample fluid image, and the sample fluid image corresponds to a sample label. Figure 3 (a) and Figure 3 (b) illustrates a fluid image obtained by processing a sample fluid image based on relevant technologies. The inventors have tested and found that the neural fluid field model provided in the embodiment of the present application can better restore the fluid density and velocity field of VOCs gas from the sample fluid image, and obtain the following: Figure 3 (c) shows the fluid image. The neural fluid field model provided in the embodiment of the present application has a high accuracy in recovering the density and velocity of VOCs fluid from the video.

[0166] In some embodiments, after step 130, the method may further include:

[0167] Acquire the initial fluid image corresponding to the petrochemical gas in the target space within the target time period;

[0168] The initial fluid image is input into the trained neural fluid field model to obtain the target fluid image corresponding to the petrochemical gas output by the trained neural fluid field model.

[0169] In this embodiment, the target period is a collection period in an actual application process.

[0170] The initial fluid image is the multi-view video data actually collected.

[0171] The target fluid image is a predicted fluid image output by the neural fluid field model. The target fluid image is used to characterize the density field and velocity field corresponding to the petrochemical gas recovered from the initial fluid image.

[0172] By inputting the actually collected multi-view video data into the trained neural fluid field model, the restored VOCs gas fluid density and velocity field output by the neural fluid field model can be obtained.

[0173] According to the neural fluid field-based VOCs fluid three-dimensional reconstruction method provided in the embodiment of the present application, by constructing and training a neural fluid field model, and using the neural fluid field model to restore the fluid density and velocity field of the VOCs gas from video data, three-dimensional reconstruction of VOCs fluid smoke is achieved, which can be used to detect VOCs leakage and trace its source, etc., and the detection precision and accuracy are high.

[0174] The following is a description of the VOCs fluid three-dimensional reconstruction device based on the neural fluid field provided in the present application. The VOCs fluid three-dimensional reconstruction device based on the neural fluid field described below and the VOCs fluid three-dimensional reconstruction method based on the neural fluid field described above can be referenced to each other.

[0175] The VOCs fluid 3D reconstruction method based on the neural fluid field provided in the embodiment of the present application can be performed by a VOCs fluid 3D reconstruction device based on the neural fluid field. In the embodiment of the present application, the VOCs fluid 3D reconstruction method based on the neural fluid field is performed by a VOCs fluid 3D reconstruction device based on the neural fluid field as an example to illustrate the VOCs fluid 3D reconstruction method based on the neural fluid field provided in the embodiment of the present application.

[0176] The embodiment of the present application also provides a VOCs fluid three-dimensional reconstruction device based on the neural fluid field.

[0177] like Figure 4 As shown, the VOCs fluid three-dimensional reconstruction device based on the neural fluid field includes: a first processing module 410, a second processing module 420 and a third processing module 430.

[0178] The first processing module 410 is used to input the sample fluid image corresponding to the petrochemical gas in the target space within the sample period and the sample label corresponding to the sample fluid image into the neural fluid field model to obtain the first fluid image corresponding to the petrochemical gas output by the neural fluid field model; the sample label includes at least one of the actual density field and the actual velocity field of the petrochemical gas in the target space;

[0179] A second processing module 420, configured to construct a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label;

[0180] The third processing module 430 is used to train the neural fluid field model based on the target loss function; the neural fluid field model is used to reconstruct at least one of the target density field and the target velocity field of the petrochemical gas in the target space.

[0181] According to the VOCs fluid three-dimensional reconstruction device based on neural fluid field provided in the embodiment of the present application, a neural fluid field model is constructed to restore the density field and velocity field of petrochemical gas in the sample fluid image to obtain a first fluid image, and a target loss function is constructed based on the first fluid image, the sample fluid image, and at least one of the sample labels corresponding to the sample fluid image. The neural fluid field model is then trained based on the target loss function to improve the precision and accuracy of the neural fluid field model, so that the neural fluid field model can restore the velocity field and density field of VOCs fluid from multi-perspective video data. It is suitable for processing VOCs fluids that lack stable visual features, can better support applications such as real-time monitoring and three-dimensional reconstruction, and has high detection accuracy.

[0182] In some embodiments, the second processing module 420 may also be used to:

[0183] Based on at least one of the first fluid image, the sample fluid image, and the sample label, construct at least one of a density loss function, a projection loss function, a laminar regularization loss function, and a rendering loss function;

[0184] At least one of the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function is fused to obtain the target loss function.

[0185] In some embodiments, the second processing module 420 may also be used to:

[0186] A density loss function is constructed based on the product of the diffusion rate of the density of the petrochemical gas along the target dimension in multiple spatial dimensions and the velocity of the petrochemical gas along the target dimension, as well as the density change rate of the petrochemical gas;

[0187] Based on the difference between the actual velocity field and the divergence corresponding to the actual velocity field, a projection loss function is constructed;

[0188] Based on the actual velocity field and the actual density field, a laminar regularization loss function is constructed;

[0189] A rendering loss function is constructed based on the first fluid image and the sample fluid image.

[0190] In some embodiments, the second processing module 420 may also be used to:

[0191] Based on the actual velocity field and the actual density field, a hinge loss function is constructed;

[0192] Based on the hinge loss function, a laminar regularization loss function is constructed.

[0193] In some embodiments, the second processing module 420 may also be used to:

[0194] Acquire the rendering brightness of the first fluid image at the target camera position and the target ray direction, and the observed brightness of the sample fluid image at the target camera position and the target ray direction;

[0195] Based on the difference between the rendered brightness and the observed brightness, a rendering loss function is constructed.

[0196] In some embodiments, the second processing module 420 may also be used to:

[0197] Get the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function, and the loss weight corresponding to the rendering loss function;

[0198] Based on the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function and the loss weight corresponding to the rendering loss function, the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function are weightedly summed to obtain the target loss function.

[0199] In some embodiments, the VOCs fluid 3D reconstruction device based on the neural fluid field may further include a fourth processing module for constructing a neural fluid field model based on the following steps:

[0200] The neural velocity field is constructed based on the neural graphics primitives, and the residual turbulence velocity field driven by the vortex particles is constructed based on the target motion equations;

[0201] A neural fluid field model is constructed based on the base neural velocity field and the residual turbulent velocity field driven by vortex particles.

[0202] In some embodiments, the VOCs fluid 3D reconstruction device based on neural fluid field may further include a fifth processing module for acquiring an initial fluid image corresponding to petrochemical gas in the target space within a target period after training the neural fluid field model based on the target loss function;

[0203] The initial fluid image is input into the trained neural fluid field model to obtain the target fluid image corresponding to the petrochemical gas output by the trained neural fluid field model.

[0204] The VOCs fluid three-dimensional reconstruction device based on the neural fluid field in the embodiment of the present application can be an electronic device, or a component in an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or it can be other devices other than a terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a car electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) device, a robot, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., and the embodiment of the present application is not specifically limited.

[0205] The VOCs fluid 3D reconstruction device based on the neural fluid field in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an IOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0206] The VOCs fluid three-dimensional reconstruction device based on the neural fluid field provided in the embodiment of the present application can achieve Figures 1 to 3 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0207] In some embodiments, Figure 5As shown, an embodiment of the present application further provides an electronic device 500, including a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the program is executed by the processor 501, each process of the above-mentioned embodiment of the method for three-dimensional reconstruction of VOCs fluid based on neural fluid field is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0208] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0209] On the other hand, the present application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the various processes of the above-mentioned embodiment of the VOCs fluid three-dimensional reconstruction method based on the neural fluid field, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0210] On the other hand, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the various processes of the above-mentioned embodiment of the VOCs fluid three-dimensional reconstruction method based on the neural fluid field, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0211] On the other hand, an embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned embodiment of the VOCs fluid three-dimensional reconstruction method based on the neural fluid field, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0212] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0213] The device embodiments described above are merely illustrative, 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.

[0214] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution can be essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0215] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for three-dimensional reconstruction of VOCs fluid based on neural fluid field, characterized in that: include: Inputting the sample fluid image corresponding to the petrochemical gas in the target space within the sample period and the sample label corresponding to the sample fluid image into the neural fluid field model, obtaining a first fluid image corresponding to the petrochemical gas output by the neural fluid field model; the sample label includes at least one of the actual density field and the actual velocity field of the petrochemical gas in the target space; constructing a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label; The neural fluid field model is trained based on the target loss function; the neural fluid field model is used to reconstruct at least one of a target density field and a target velocity field of the petrochemical gas in the target space, and at least one of the target density field and the target velocity field is used to perform three-dimensional reconstruction of the fluid corresponding to the petrochemical gas.

2. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to claim 1, characterized in that: The constructing a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label comprises: Based on at least one of the first fluid image, the sample fluid image and the sample label, construct at least one of a density loss function, a projection loss function, a laminar regularization loss function and a rendering loss function; The target loss function is obtained by fusing at least one of the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function.

3. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to claim 2, characterized in that: The constructing at least one of a density loss function, a projection loss function, a laminar regularization loss function and a rendering loss function based on at least one of the first fluid image, the sample fluid image and the sample label comprises: Constructing the density loss function based on the product of the diffusion rate of the density of the petrochemical gas along a target dimension among the multiple spatial dimensions and the speed of the petrochemical gas along the target dimension, and the density change rate of the petrochemical gas; constructing the projection loss function based on the difference between the actual velocity field and the divergence corresponding to the actual velocity field; Constructing the laminar regularization loss function based on the actual velocity field and the actual density field; The rendering loss function is constructed based on the first fluid image and the sample fluid image.

4. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to claim 3, characterized in that: The step of constructing the laminar regularization loss function based on the actual velocity field and the actual density field comprises: Constructing a hinge loss function based on the actual velocity field and the actual density field; Based on the hinge loss function, the laminar regularization loss function is constructed.

5. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to claim 3, characterized in that: The constructing the rendering loss function based on the first fluid image and the sample fluid image includes: Acquire the rendering brightness of the first fluid image at a target camera position and a target ray direction, and the observed brightness of the sample fluid image at the target camera position and the target ray direction; The rendering loss function is constructed based on the difference between the rendering brightness and the observed brightness.

6. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to claim 2, characterized in that: The fusing at least one of the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function to obtain the target loss function includes: Obtaining the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function, and the loss weight corresponding to the rendering loss function; Based on the loss weight corresponding to the density loss function, the loss weight corresponding to the projection loss function, the loss weight corresponding to the laminar regularization loss function and the loss weight corresponding to the rendering loss function, the density loss function, the projection loss function, the laminar regularization loss function and the rendering loss function are weightedly summed to obtain the target loss function.

7. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to any one of claims 1 to 6, characterized in that: The neural fluid field model is constructed based on the following steps: The neural velocity field is constructed based on the neural graphics primitives, and the residual turbulence velocity field driven by the vortex particles is constructed based on the target motion equations; The neural fluid field model is constructed based on the base neural velocity field and the residual turbulent velocity field driven by the vortex particles.

8. The method for three-dimensional reconstruction of VOCs fluid based on neural fluid field according to any one of claims 1 to 6, characterized in that: After training the neural fluid field model based on the target loss function, the method further includes: Acquire an initial fluid image corresponding to the petrochemical gas in the target space within a target time period; The initial fluid image is input into a trained neural fluid field model to obtain a target fluid image corresponding to the petrochemical gas output by the trained neural fluid field model.

9. A VOCs fluid 3D reconstruction device based on neural fluid field, characterized in that: include: A first processing module is used to input the sample fluid image corresponding to the petrochemical gas in the target space within the sample period and the sample label corresponding to the sample fluid image into the neural fluid field model to obtain a first fluid image corresponding to the petrochemical gas output by the neural fluid field model; the sample label includes at least one of the actual density field and the actual velocity field of the petrochemical gas in the target space; A second processing module, configured to construct a target loss function based on at least one of the first fluid image, the sample fluid image, and the sample label; The third processing module is used to train the neural fluid field model based on the target loss function; the neural fluid field model is used to reconstruct at least one of the target density field and the target velocity field of the petrochemical gas in the target space, and at least one of the target density field and the target velocity field is used to perform three-dimensional reconstruction of the fluid corresponding to the petrochemical gas.

10. An electronic 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 program, the VOCs fluid three-dimensional reconstruction method based on the neural fluid field as described in any one of claims 1 to 8 is implemented.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for three-dimensional reconstruction of VOCs fluid based on neural fluid field as described in any one of claims 1 to 8 is implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for three-dimensional reconstruction of VOCs fluid based on neural fluid field as described in any one of claims 1 to 8 is implemented.

Images