A flame three-dimensional multi-physical field reconstruction method based on light field focus stack

Abstract

The application provides a flame three-dimensional multi-physical field reconstruction method based on an optical field focus stack, comprising the following steps: imaging a blackbody radiation source with different temperatures by using an optical field camera, and establishing and outputting a fitting relationship between response values of each color channel of the optical field refocus image and corresponding spectral radiation intensity according to the Planck law; acquiring a flame optical field focus stack image sequence of the flame optical field original image by using a super-resolution reconstruction method based on a sub-aperture image; and outputting temperature distribution and soot concentration distribution of a corresponding flame slice image by using a three-dimensional convolutional neural network model fused with radiation transmission physical information constraint. The method directly constructs an implicit correlation model between a flame focus slice and its three-dimensional physical field by using the three-dimensional convolutional neural network model fused with the radiation transmission physical information constraint, and can improve the generalization ability and interpretability of the model, so that the joint reconstruction work of the physical field is more effectively realized.

Description

Technical Field

[0001] This invention relates to the field of flame physics field measurement technology, and more particularly to a method for three-dimensional multiphysics field reconstruction of flames based on light field focus stacking. Background Technology

[0002] As a core energy source for industrial combustion equipment and aerospace propulsion systems, the accurate measurement of flame temperature and soot concentration during the combustion process of hydrocarbon fuels is crucial for improving energy conversion efficiency and achieving clean combustion. Due to the high spatial complexity and temporal variability of the flame flow field within the combustion chamber of actual power equipment, achieving high-dimensional, high-spatiotemporal resolution, and accurate in-situ measurement of the flame physical field under limited field of view and confined space conditions is a bottleneck problem that urgently needs to be solved. Flame measurement technology based on radiation light field imaging utilizes a light field camera to simultaneously acquire the intensity, spectrum, position, and angle of flame radiation light in a single exposure. This theoretically and technically solves the problem that single-path detection systems are unsuitable for three-dimensional reconstruction of unsteady, asymmetric flames, and also avoids the complexity of multi-path measurement systems and the problems of signal synchronization.

[0003] However, the performance of current flame light field imaging measurement methods is still limited by their physical field reconstruction algorithms. Current reconstruction algorithms are broadly divided into layered light field reconstruction methods and volumetric light field reconstruction methods. The former iteratively deconvolves the layered flame image to calculate the original radiation intensity distribution of each layer, while the latter uses a radiative transfer model and its inverse problem solving algorithm to estimate the photothermal parameter group of the flame mesh. Both methods involve solving high-dimensional nonlinear inverse problems, which have inherent drawbacks such as ill-posedness, initial value dependence, and susceptibility to noise interference. Although various deconvolution joint reconstruction methods and intelligent optimization algorithms for the radiative inverse problem have been developed, the accuracy, spatial resolution, and robustness of the reconstruction still need improvement.

[0004] Focus stacking imaging technology overcomes the depth-of-field limitations of traditional optical imaging. By using a large number of focus slices with equally spaced focus depths along the depth direction, it acquires the focus, defocus, depth, and spatial correlation information of the target under test. The light field focus stacking imaging method combines the focus stacking imaging mechanism with the characteristics of light field imaging. It acquires four-dimensional light field information of the target under test through a single exposure of a light field camera, and then uses a digital refocusing algorithm to construct a focus stack containing different focus depths. This method can acquire spatial information and focus cues of the target along the depth direction in real time. It is typically used for the 3D reconstruction of opaque targets with local contrast, structure, and patterns. However, for semi-transparent radiation-participating media such as high-temperature luminous flames, complex physical processes such as radiation transmission, absorption, and scattering are involved in light propagation, and related reconstruction methods still require further research.

[0005] Current focus stack reconstruction methods suffer from the following problems: First, existing methods rely on complex calibration or multi-path systems, resulting in cumbersome processes and difficulty in applying them to dynamic flames. Second, the reconstruction algorithms lack sufficient depth resolution and often involve solving high-dimensional inverse problems, leading to low computational efficiency, sensitivity to initial values ​​and noise, and poor robustness. Finally, existing data-driven models lack explicit physical mechanism constraints, making it difficult to guarantee the interpretability and physical consistency of the reconstruction results. Summary of the Invention

[0006] In view of this, the purpose of this invention is to propose a three-dimensional multiphysics reconstruction method for flames based on light field focus stack, so as to solve the technical problems of unsatisfactory reconstruction accuracy, efficiency and robustness of existing flame physics reconstruction algorithms based on light field imaging.

[0007] The technical means employed in this invention are as follows: A method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking includes the following steps: Using a light field camera to image blackbody radiation sources at different temperatures, the response values ​​of each color channel in the light field refocused image are established and output according to Planck's law. x λ With corresponding spectral radiance I λ Fitting relationship between I λ = f ( x λ ); A light field camera is used to image the flame, and the original image of the flame's light field is acquired and output. A super-resolution reconstruction method based on sub-aperture images is used to obtain a sequence of flame light field focus stack images from the original flame light field image. N H × N W × N L , where NH is the horizontal resolution of the flame refocusing image. N W The resolution of the flame refocusing image in the vertical direction. N L The number of slices in the flame light field focus stack; A three-dimensional convolutional neural network model that utilizes fused radiative transport physical information constraints, based on fitting relationships... I λ = f ( x λ This function takes a sequence of flame light field focus stacked images as input and outputs the temperature distribution of the corresponding flame slice image. NH × N W × N L and carbon soot concentration distribution N H × N W × N L .

[0008] Furthermore, obtaining the fitting relationship specifically includes: Place the blackbody radiation source in front of the light field camera lens, making the emitting surface of the blackbody radiation source perpendicular to the camera's optical axis, and adjust it to the same working distance as flame imaging. Within a preset temperature range, original images of the light field of the blackbody emitting surface at different temperatures are acquired at fixed intervals. Each original light field image is refocused to obtain a refocused image focused at the original depth of focus. The average grayscale value of each color channel within a preset calibration region in each refocused image is extracted and used as the response value at that temperature. x λ ; The theoretical values ​​of spectral radiance at corresponding wavelengths at each temperature were calculated based on Planck's law. I λ ; By fitting the response value x λ Compared with theoretical value I λ Establish the fitting relationship for each color channel. I λ = f ( x λ ).

[0009] Furthermore, when using a light field camera to image a flame, the positions of the flame and the light field camera are adjusted so that the flame imaging area is located in the central area of ​​the image sensor, and the pixel ratio of the flame imaging area in the horizontal and vertical directions of the image is 60% to 80%.

[0010] Furthermore, when using a light field camera to image a flame, if the effective radiation area of ​​the flame accounts for less than a preset value in the image, edge detection is performed on the original flame light field image, and the effective pixel area containing only flame radiation information is extracted and output as the original flame light field image.

[0011] Furthermore, the flame light field focus stack image sequence N H × N W × NL The relationship is: H / N H ≈ W / N W ≈ L / N L in, H The height of the target area of ​​the flame. W The width of the target area of ​​the flame. L The radial depth of the target area of ​​the flame.

[0012] Furthermore, the super-resolution reconstruction method based on sub-aperture images includes: Based on the principle of digital refocusing, the subpixel displacement between sub-aperture images at different focusing depths is determined; Based on sub-pixel displacement, the sub-aperture image is registered and back-mapped. The fused sub-aperture images are used to generate a sequence of flame light field focus stack images with improved spatial resolution.

[0013] Furthermore, the overall loss function of the three-dimensional convolutional neural network model that integrates radiative transfer physical information constraints includes a data loss term. Physical consistency loss term The overall loss function is as follows:

[0014] Data loss items as follows:

[0015] Physical consistency loss term as follows:

[0016] in, These are temperature sample observations. These are the observed values ​​of carbon soot concentration. This is a predicted temperature value. This is the predicted value for soot concentration. These are the predicted values ​​of spectral radiance calculated from the model predictions; The spectral radiance corresponds to the channel response value of the actual flame slice image. ω These are the weighting coefficients.

[0017] Furthermore, the three-dimensional convolutional neural network model adopts the U-net architecture, including an encoder, a bottleneck layer, and a decoder; The encoder is used to extract local slice features under adjacency spatial dependencies by stacking multiple convolutional layers; The bottleneck layer includes a spatial association module composed of pixel attention blocks, spatial attention blocks, and focus attention blocks. The bottleneck layer is used to extract stack global features under distance-based spatial dependencies. The decoder uses the extracted high-dimensional features to progressively upsample and restore the details and spatial dimensions of the reconstructed target, and the multi-scale features extracted by each convolutional layer in the encoder are skipped to the corresponding convolutional layers in the decoder.

[0018] The present invention also provides a storage medium comprising a stored program, wherein, when the program is executed, any of the above-described methods for three-dimensional multiphysics reconstruction of flames based on a light field focus stack are performed.

[0019] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any of the above-described flame three-dimensional multiphysics reconstruction methods based on light field focus stacks through the computer program.

[0020] Compared with the prior art, the present invention has the following advantages: A single-field camera is used as the flame radiation information acquisition device. The full-field radiation information of the flame is obtained through a single exposure. There is no need to calibrate the internal parameters of the camera and the point spread function, which simplifies the measurement and reconstruction process and can be used for dynamic reconstruction of asymmetric flames.

[0021] Introducing the flame radiation focus stack into the physical field reconstruction process can fully utilize the spatial correlation information in the stack data with depth information as a clue, effectively improving the accuracy, efficiency, and radial depth resolution of flame 3D reconstruction.

[0022] Compared with existing flame physics field reconstruction methods, which suffer from high nonlinearity, ill-posedness, long computation time, and low reconstruction resolution, this method directly constructs an implicit correlation model between the flame focal slice and its three-dimensional physics field by integrating a three-dimensional convolutional neural network model constrained by forward radiative transmission physical information. This improves the model's generalization ability and interpretability, thereby more effectively achieving joint reconstruction of the physics field. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the method of the present invention; Figure 2 These are refocused images of the light field of the blackbody radiation source plane at different temperatures according to the present invention. Figure 3 This is a flowchart of the flame super-resolution focus stack reconstruction based on sub-aperture images according to the present invention; Figure 4 This is a sequence of super-resolution focused stacked images of flames according to the present invention; Figure 5 This is a diagram of the 3D CNN reconstruction model that incorporates radiative transfer information constraints according to the present invention. Figure 6 This is a network architecture diagram of the 3D CNN model for reconstructing the three-dimensional physical field of flame in this invention; Figure 7 This is a distribution diagram of the three-dimensional temperature field reconstructed by the present invention in the radial depth direction. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] like Figure 1 As shown, this invention provides a method for three-dimensional multiphysics reconstruction of flames based on light field focus stacks, comprising the following steps: Step 1: Image blackbody radiation sources at different temperatures using a light field camera, and determine the response values ​​of the R, G, and B channels of the refocused light field image according to Planck's law. xλ Its spectral radiance I λ Fitting relationship I λ = f ( x λ ).

[0028] A blackbody radiation source is fixed in front of the light field camera lens, ensuring its emitting surface is perpendicular to the camera's optical axis. The working distance is adjusted to match the subsequent imaging of the flame target, and the imaging area is ensured to be centered on the image sensor. Based on the predicted flame temperature, the temperature variation range and fixed interval of the blackbody radiation source are set, and raw light field images of the blackbody's emitting surface at different temperatures are acquired sequentially. The spectral radiant intensities of the R, G, and B channels at a given temperature can be calculated using Planck's law. I λ Accordingly, the image grayscale value of the corresponding channel extracted from the image is the response value of the image sensor to that spectral radiance. x λ Since this method uses a sequence of flame light field focus stack images as reconstruction input data, the grayscale information of the refocused image is used for calibration. The specific process is as follows: First, the original light field images at each temperature are preprocessed with bilateral filtering to suppress noise and retain the edge information of the sub-images. Then, based on the preprocessed image, a refocused image focused at the original focus depth is calculated and generated. Furthermore, the average pixel value within the calibration area at the center of the refocused image is statistically analyzed. x λ And its relationship with spectral radiance is obtained through polynomial fitting. I λ Functional relationship I λ = f ( x λ ).

[0029] In this embodiment, the temperature variation range of the blackbody radiation source plane is set to 1123 ~ 1373 K, with an interval of 50 K. The refocusing images at each temperature are calculated as follows: Figure 2 As shown, the selected center calibration area is shown below. Figure 2 The area marked by the red box in the middle. The wavelength corresponding to the R channel is obtained by performing a quadratic polynomial fitting between the average gray value of pixels within the calibration area and the corresponding spectral radiance. λ R Quantitative relationship between spectral radiance at 0.61 μm and grayscale values ​​of refocused images: (1) The calibration relationship between channels G and B is implemented in the same way.

[0030] Step 2: Use a light field camera to image the flame and acquire the raw image of the flame's light field. To ensure sufficient light field data is acquired and to prevent data confusion caused by lens edge distortion, the flame imaging area should be located at the center of the image sensor, and the horizontal and vertical proportions should be controlled between 60% and 80%.

[0031] The laminar flow diffusion flame generator is placed in front of the light field camera lens. Its position and the working distance of the light field camera are adjusted to align the flame center with the camera's optical axis, while simultaneously ensuring that the flame imaging area is centered on the image sensor and its field of view is within the aforementioned range. After the flame has stabilized and fully developed, the light field camera is used to capture the original light field image of the flame.

[0032] If the effective radiation area of ​​the flame occupies a small proportion in the image (such as presenting a long and thin shape), in order to improve the efficiency of subsequent reconstruction and reduce the amount of invalid background pixels to be processed, the Sobel or Candy operator can be used to perform edge detection on the original image and extract the effective pixel area containing the flame radiation information.

[0033] Step 3: Based on the requirements of three-dimensional spatial resolution, a super-resolution reconstruction method based on sub-aperture images is adopted to obtain a sequence of flame light field focal stack images. N H × N W × N L ,in N H and N W For flame refocusing image resolution. N L This represents the number of slices in the flame light field focus stack. To ensure the accuracy and efficiency of the reconstruction, the following three conditions should be met: H / N H ≈ W / N W ≈ L / N L ( H , W and L These are the height, width, and radial depth of the flame target area, respectively.

[0034] The four-dimensional sampling of flame radiation acquired by a single exposure of a light field camera can be parameterized as follows: ,in L The intensity of light radiation. and These are the coordinates of the intersection points of the light rays with the planes of the main lens and the image sensor, respectively. FThis is the distance between the two planes (i.e., the depth of focus). When the depth of focus is adjusted to... At that time, for the same ray of light, there are According to radiosity theory, the depth of focus can be obtained by weighted integral over the angle of light rays. Refocused image at: (2) In the formula, Indicates focus depth The directional radiation force received at the plane. This formula describes the directional radiation force at any focusing depth. Flame images at any location can be obtained through a four-dimensional light field. Calculated. Define the operator. , will focus on depth F Light field sampling at the location Transform into depth of focus Image at: (3) By adjusting the focus depth coefficient α It can calculate the corresponding different focusing depths. αF The refocused image is then used to construct the flame light field focus stack transformation. δ : (4) As shown in equation (2), the digital refocusing process can be equivalent to moving and superimposing sub-aperture images at different angles. Therefore, the resolution of the refocused image is limited by the resolution of the sub-aperture images, i.e., the spatial sampling resolution determined by the number of microlens units, resulting in a low and non-adjustable resolution. However, since the light field information itself has redundancy and correlation, the super-resolution reconstruction method based on sub-aperture images can be used to improve the resolution of the refocused image.

[0035] Based on the projection relationship between the same object point and corresponding image points on adjacent sub-aperture images during optical field refocusing, the relative displacement between adjacent sub-aperture images can be determined as follows: (5) In the formula, , These represent the relative displacements of the sub-aperture image in the vertical and horizontal directions, respectively. The center distance between adjacent sub-apertures, where 1 D The main lens aperture diameter, n The number of sub-apertures is 1 (i.e., the number of pixels covered by the sub-image). d Let be the pixel size. From equation (5), it can be seen that, except for the original focal position ( α= 1) Apart from the sub-aperture images at other focusing depths, there are sub-pixel displacements between them. Therefore, spatial super-resolution reconstruction of the flame light field focus stack can be achieved through sub-pixel registration and back-mapping of the sub-aperture images. The specific process is as follows: Figure 3 As shown.

[0036] In this embodiment, the axial height of the cylindrical flame model H 0.2 m, width W and radial depth L Both are 0.04 m, and the resolution of the flame super-resolution refocusing image is 200 (H) × 40 (W). According to the aforementioned proportional relationship: H / N H ≈ W / N W ≈ L / N L The number of slices in the focal stack of the flame light field can be determined. N L The value is 40. Following the procedure described in this step, flame focus stack reconstruction is performed to obtain a 200 × 40 × 40 super-resolution flame focus stack image sequence. Figure 4 As shown, the voxel resolution corresponding to this sequence in three-dimensional space is 1 mm.

[0037] Step 4: Using a 3D convolutional neural network model constrained by fused radiative transport physical information, the sequence of flame light field focal stack images is used as the input layer, and the output layer is the temperature distribution of the corresponding flame slice image. N H × N W × N L and carbon soot concentration distribution N H × N W × N L .

[0038] To improve the generalization ability and interpretability of the reconstruction model, this invention introduces physical information related to forward radiative transfer as constraints, based on data-driven principles. For example... Figure 5 As shown, physical information is incorporated into the reconstruction network model by imposing physical constraints on the loss function. The loss function of this model consists of two parts: one part is the sample observation value (temperature). Carbon soot concentration ) and predicted value (temperature) Carbon soot concentration The data-driven loss term is generated by the deviation between the two values, and the mean squared error is used as the loss function: (6) The second part is the physical consistency loss term resulting from the computational constraints of medium radiative transport. For any slice image in the flame focus stack, the thickness is... s The radiative transfer within the slice medium can be equivalent to the emission of the slice medium itself, with a spectral emission intensity of... In the model's prediction process, it is assumed that the predicted values ​​of temperature and soot concentration can be obtained through the nonlinear fitting ability of the network. and If equation (7) holds true, then the model's predicted value is relatively accurate.

[0039] (7) In the formula, These are the predicted values ​​of spectral radiance calculated from the model predictions; Let be the spectral radiance corresponding to the channel response value of the actual flame slice image. Based on this, the loss function for this part is defined as: (8) To balance the relationship between the two types of constraints, weighting coefficients are set. ω (Hyperparameters) represent the contribution of physical information constraints, then the expression for the total loss function of the network model is: (9) The architecture of the three-dimensional convolutional neural network model (3D CNN) is as follows: Figure 6 As shown, considering the local and global spatial correlation characteristics of the flame light field focus stack data, U-net is adopted as the basic architecture, and a spatial correlation module is introduced in the bottleneck layer to ensure the correlation of global spatial information. The network consists of three parts: encoder, bottleneck layer and decoder. The encoder extracts slice local features under the spatial dependency of "adjacency" through multiple convolutional layers. The bottleneck layer is a spatial correlation module composed of pixel attention blocks, spatial attention blocks and focus attention blocks, which mainly extracts stack global features under the spatial dependency of "distance". The decoder uses the extracted high-dimensional features to gradually upsample and restore the details and spatial dimensions of the reconstructed target. In addition, the multi-scale features extracted by each convolutional layer in the encoder are skipped to the corresponding convolutional layers in the decoder, which helps the decoder to reconstruct feature details more accurately.

[0040] In the bottleneck layer, pixel attention blocks generate pixel-level weights through Conv1×1×1 convolution and Sigmoid activation, and enhance local details through element-wise multiplication; spatial attention blocks extract spatial context using Conv k×k×k convolution, calculate regional importance through weight matrix, and also highlight key spatial structures through element-wise multiplication; while focal attention blocks are designed specifically for multi-focal stacked images, and fuse the local and global correlations of different focal slices through matrix multiplication and element-wise addition operations to achieve full-focus clear image synthesis.

[0041] In this embodiment, the super-resolution focal stack image sequence of the flame light field obtained in step three (200 × 40 × 40) is input into a trained 3D CNN model that incorporates radiative transport physical information constraints, and the corresponding three-dimensional temperature distribution (200 × 40 × 40) and soot concentration distribution (200 × 40 × 40) are directly reconstructed. Figure 7 The figure shows the reconstructed three-dimensional temperature field distribution along the radial depth direction.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking, characterized in that, Includes the following steps: Using a light field camera to image blackbody radiation sources at different temperatures, the response values ​​of each color channel in the light field refocused image are established and output according to Planck's law. x λ With corresponding spectral radiance I λ Fitting relationship between I λ = f ( x λ ); A light field camera is used to image the flame, and the original image of the flame's light field is acquired and output. A super-resolution reconstruction method based on sub-aperture images is used to obtain a sequence of flame light field focus stack images from the original flame light field image. N H × N W × N L ,in N H The horizontal resolution of the flame refocusing image. N W The resolution of the flame refocusing image in the vertical direction. N L The number of slices in the flame light field focus stack; A three-dimensional convolutional neural network model that utilizes fused radiative transport physical information constraints, based on fitting relationships... I λ = f ( x λ This function takes a sequence of flame light field focus stacked images as input and outputs the temperature distribution of the corresponding flame slice image. N H × N W × N L and carbon soot concentration distribution N H × N W × N L .

2. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, Obtaining the fitted relationship specifically includes: Place the blackbody radiation source in front of the light field camera lens, making the emitting surface of the blackbody radiation source perpendicular to the camera's optical axis, and adjust it to the same working distance as flame imaging. Within a preset temperature range, original images of the light field of the blackbody emitting surface at different temperatures are acquired at fixed intervals. Each original light field image is refocused to obtain a refocused image focused at the original depth of focus. The average grayscale value of each color channel within a preset calibration region in each refocused image is extracted and used as the response value at that temperature. x λ ; The theoretical values ​​of spectral radiance at corresponding wavelengths at each temperature were calculated based on Planck's law. I λ ; By fitting the response value x λ Compared with theoretical value I λ Establish the fitting relationship for each color channel. I λ = f ( x λ ).

3. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, When using a light field camera to image a flame, adjust the position of the flame and the light field camera so that the flame imaging area is located in the center of the image sensor, and the pixel ratio of the flame imaging area in the horizontal and vertical directions of the image is 60% to 80%.

4. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, When imaging a flame using a light field camera, if the effective radiation area of ​​the flame accounts for less than a preset value in the image, edge detection is performed on the original flame light field image, and the effective pixel area containing only flame radiation information is extracted and output as the original flame light field image.

5. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, Flame light field focus stack image sequence N H × N W × N L The relationship is: H / N H ≈ W / N W ≈ L / N L in, H The height of the target area of ​​the flame. W The width of the target area of ​​the flame. L The radial depth of the target area of ​​the flame.

6. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, The super-resolution reconstruction method based on sub-aperture images includes: Based on the principle of digital refocusing, the subpixel displacement between sub-aperture images at different focusing depths is determined; Based on sub-pixel displacement, the sub-aperture image is registered and back-mapped. The fused sub-aperture images are used to generate a sequence of flame light field focus stack images with improved spatial resolution.

7. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, The overall loss function of the three-dimensional convolutional neural network model constrained by fused radiative transfer physical information includes a data loss term. Physical consistency loss term The overall loss function is as follows: Data loss items as follows: Physical consistency loss term as follows: in, These are temperature sample observations. These are the observed values ​​of carbon soot concentration. This is a predicted temperature value. This is the predicted value for soot concentration. These are the predicted values ​​of spectral radiance calculated from the model predictions; The spectral radiance corresponds to the channel response value of the actual flame slice image. ω These are the weighting coefficients.

8. The method for three-dimensional multiphysics reconstruction of flames based on light field focus stacking according to claim 1, characterized in that, The three-dimensional convolutional neural network model adopts the U-net architecture, including an encoder, a bottleneck layer, and a decoder; The encoder is used to extract local slice features under adjacency spatial dependencies by stacking multiple convolutional layers; The bottleneck layer includes a spatial association module composed of pixel attention blocks, spatial attention blocks, and focus attention blocks. The bottleneck layer is used to extract stack global features under distance-based spatial dependencies. The decoder uses the extracted high-dimensional features to progressively upsample and restore the details and spatial dimensions of the reconstructed target, and the multi-scale features extracted by each convolutional layer in the encoder are skipped to the corresponding convolutional layers in the decoder.

9. A storage medium, characterized in that, The storage medium includes a stored program, wherein when the program is executed, it performs the flame three-dimensional multiphysics reconstruction method based on light field focus stack as described in any one of claims 1 to 8.

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, The processor executes the flame three-dimensional multiphysics reconstruction method based on light field focus stack as described in any one of claims 1 to 8 through the computer program.

Images