Submarine damage assessment method and system based on multi-source bubble energy analysis

Abstract

The invention discloses a submarine damage assessment method and system based on multi-source bubble energy analysis, and belongs to the technical field of underwater explosion experiments. The method comprises the following steps: acquiring image data, sound field data and pressure data of explosion bubbles in different dimensions in a submarine damage experiment process; generating explosion bubble three-dimensional point cloud data through a sub-pixel-level edge fitting algorithm and a three-dimensional point cloud algorithm; collapse characteristic data and sound field characteristic data are extracted through the sound field data; extracting bubble pulsation local features through the pressure data, and further calculating a global bubble collapse energy release value of the explosion bubbles; the method comprises the following steps: firstly, carrying out multi-source data fusion on an image, a sound field and pressure three-dimensional data through a Bayesian inversion algorithm to construct a more comprehensive and accurate three-dimensional energy density field, realizing explosion bubble actual pulsation behavior analysis of multiple data sources, and then constructing a bubble structure damage mapping model through a machine learning regression algorithm to realize explosion bubble actual pulsation behavior analysis. Therefore, accurate assessment of submarine damage is realized.

Description

Technical Field

[0001] This invention belongs to the field of underwater explosion experimental technology, and particularly relates to a method and system for submarine damage assessment based on multi-source bubble energy analysis. Background Technology

[0002] In the environment of deep-water explosions, the bubble pulsation behavior generated by the explosion has an extremely severe destructive impact on underwater target structures such as submarines. The bubbles formed by the explosion undergo a series of complex processes in the water, including rapid expansion, contraction, and collapse. This bubble pulsation behavior induces phenomena such as high-intensity nonlinear fluid-structure interaction, transient energy accumulation, and directional jet impact, thus causing significant and severe damage to underwater target structures, especially critical components such as the hull and rudders of submarines. The bubble pulsation behavior generated by deep-water explosions is a crucial component of the underwater explosion damage effect, severely impacting the structural integrity, navigation performance, and normal operation of internal systems of submarines.

[0003] Currently, submarine damage analysis based on bubble pulsation behavior typically employs simulation experiments and numerical studies. For example, numerical studies in actual deep-water explosion environments often use the confined volume method (FVM) and VOF interface capture technology to numerically simulate the entire process of underwater explosion bubbles, reconstructing bubble pulsation behavior. However, these methods are mostly based on single data sources, resulting in an inability to comprehensively analyze the actual bubble pulsation behavior. This leads to low overall accuracy, incomplete data, and poor repeatability in the analysis of bubble pulsation behavior. Furthermore, current modeling analyses largely rely on axisymmetric models or two-dimensional boundary element methods, which severely approximate non-spherical symmetric bubble behavior under complex boundaries, making it difficult to accurately predict bubble coupling responses and local energy accumulation phenomena. This results in significant inaccuracies in the analysis of bubble pulsation behavior, hindering accurate assessment of submarine damage. Therefore, there is an urgent need for a submarine damage assessment method and system based on multi-source bubble energy analysis to address the shortcomings of existing technologies. Summary of the Invention

[0004] This invention aims to provide a submarine damage assessment method and system based on multi-source bubble energy analysis to solve the above-mentioned technical problems. By analyzing multi-source data in three dimensions—shell image, sound field, and pressure—as well as using Yeats inversion algorithm and machine learning regression algorithm, the accuracy of submarine damage assessment is improved.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a submarine damage assessment method based on multi-source bubble energy analysis, comprising: Acquire image data, acoustic field data, and pressure data of the explosion bubbles during the submarine damage experiment; Based on a preset subpixel-level edge fitting algorithm, image features are extracted from the image data to determine the coordinate data of the exploding bubble; based on a preset 3D point cloud algorithm, the coordinate data of the exploding bubble is registered to generate 3D point cloud data of the exploding bubble. Based on the sound field data, the collapse feature data and sound field feature data of the exploding bubble are extracted, and based on the collapse feature data and sound field feature data, the sound field modeling data of the exploding bubble is constructed. Based on the pressure data, the local features of bubble pulsation of the exploded bubble are extracted, and the global bubble collapse energy release value of the exploded bubble is determined based on the local features of bubble pulsation. Based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploded bubble, the sound field modeling data of the exploded bubble, and the global bubble collapse energy release value, the three-dimensional energy field of the exploded bubble is reconstructed to determine the three-dimensional energy density field of the exploded bubble. Based on a preset machine learning regression algorithm and combined with pre-acquired historical data of submarine damage experiments, a bubble structure damage mapping model is constructed. The three-dimensional energy density field of the exploded bubble is input into the bubble structure damage mapping model to determine the submarine damage assessment result.

[0006] Understandably, this invention acquires image data, sound field data, and pressure data of the exploding bubbles from different dimensions during submarine damage experiments; then, it generates 3D point cloud data of the exploding bubbles using subpixel-level edge fitting algorithms and 3D point cloud algorithms, providing the spatial morphology of the exploding bubbles from the image dimension; it extracts collapse feature data and sound field feature data from the sound field data, and then uses the exploding bubble sound field modeling data to characterize the collapse mechanism of the exploding bubbles; finally, it extracts local features of bubble pulsation from the pressure data, and then calculates the global bubble collapse energy release value of the exploding bubbles, thereby realizing the energy analysis of the exploding bubbles from the pressure dimension. Subsequently, a Bayesian inversion algorithm was used to fuse multi-source data from three dimensions: image, sound field, and pressure, thereby constructing a more comprehensive and accurate three-dimensional energy density field. This enabled comprehensive analysis of the actual pulsation behavior of exploded bubbles from multiple data sources, avoiding the problems of low analysis accuracy, incomplete data, and poor repeatability caused by a single data source. Then, a bubble structure damage mapping model was constructed using a machine learning regression algorithm, establishing a generalizable mapping mechanism between the behavior parameters of exploded bubbles and the degree of structural damage. This enabled accurate assessment of submarine damage and improved the accuracy of submarine damage assessment based on the analysis of the pulsation behavior of exploded bubbles.

[0007] As a preferred embodiment, the step of extracting image features from the image data based on a preset sub-pixel level edge fitting algorithm to determine the coordinate data of the exploded bubble; and registering the coordinate data of the exploded bubble based on a preset 3D point cloud algorithm to generate 3D point cloud data of the exploded bubble, including: The image data includes several binocular images; Based on a preset subpixel-level edge fitting algorithm, pixel-level coarse edge extraction and subpixel-level edge localization are performed on each of the stereo images to extract image features from each of the stereo images and obtain the bubble coordinate data of each of the stereo images. A joint objective function for bubble volume constraint and point cloud distance volume is constructed, and the binocular image is registered by combining a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to generate 3D point cloud data of the exploded bubble.

[0008] This preferred scheme employs a subpixel-level edge fitting algorithm for pixel-level coarse edge extraction and subpixel-level edge localization. Subpixel-level edge localization, building upon pixel-level coarse extraction, significantly reduces quantization errors in the coordinate data of the exploding bubble. Subsequently, a bubble volume constraint is introduced to ensure that the registration of the binocular images conforms to the physical continuity of the volume changes of the exploding bubble during actual pulsation. Through a dual joint objective function of point cloud distance and volume, the generated 3D point cloud data of the exploding bubble is ensured to be not only geometrically aligned but also maintain reasonable volume, avoiding incorrect matching. This achieves spatial morphological representation of the exploding bubble from an image dimension, thereby improving the accuracy of subsequent 3D energy density field and submarine damage assessment results.

[0009] As a preferred embodiment, the step of performing pixel-level coarse edge extraction and sub-pixel-level edge localization on each of the stereo images based on a preset sub-pixel-level edge fitting algorithm, in order to extract image features from each of the stereo images and obtain the coordinate data of the exploding bubble in each of the stereo images, includes: Each of the stereo images is input into a preset graphics processing model to determine the image region of each stereo image; Based on a preset edge detection operator, edge detection is performed on each of the image regions to determine the edge mask map of each of the binocular images; Contour extraction is performed on the edge mask image to determine the contour point set for each binocular image; Gaussian filtering is applied to each of the stereo images, and the gradient data of each pixel in each stereo image after Gaussian filtering is calculated. Based on the gradient data, the neighboring pixels of each pixel in the contour point set of each stereo image are determined, and a preset fitting function is used to fit each pixel in the contour point set of each stereo image and its neighboring pixels to determine the sub-pixel position offset of each pixel in the contour point set of each stereo image. Based on the sub-pixel position offset, displacement compensation is performed on each pixel in the contour point set of each stereo image to determine the bubble contour point set of each stereo image. Based on the bubble outline point set of each stereo image, determine the pixel coordinates of the exploding bubble, the center coordinates of the exploding bubble, and the radius of the exploding bubble for each stereo image; The pixel coordinates, center coordinates, and radius of the exploding bubble in each stereo image are used as the exploding bubble coordinate data for each stereo image.

[0010] This preferred solution achieves sub-pixel-level edge localization based on pixel-level coarse extraction through image region determination, edge detection, contour extraction, Gaussian filtering, and sub-pixel localization. The preset image processing model effectively eliminates background interference, thus focusing on the area where the exploded bubble is located. Edge detection and contour extraction lock the boundary of the exploded bubble, while Gaussian filtering smooths image noise, avoiding interference from noise on subsequent high-precision gradient calculations. By calculating the gradient of the pixel and its neighbors and using a fitting function to calculate the sub-pixel position offset, displacement compensation is achieved, thereby realizing sub-pixel localization. This greatly reduces the quantization error of the exploded bubble coordinate data and improves the accuracy of subsequent three-dimensional energy density field and submarine damage assessment results.

[0011] As a preferred embodiment, the step of constructing a joint objective function for bubble volume constraint and point cloud distance volume, and combining it with a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to register the binocular image and generate 3D point cloud data of the exploded bubble includes: Obtain a bubble volume threshold and construct a bubble volume constraint based on the bubble volume threshold; Construct a point cloud distance sum of squares term and a point cloud volume difference squares term, and construct a joint objective function for point cloud distance and volume based on the point cloud distance sum of squares term and the point cloud volume difference squares term; The iteration objective is to minimize the joint objective function of point cloud distance and volume. Based on the preset 3D point cloud algorithm and the bubble volume constraint, the coordinate data of the exploded bubble in each stereo image is iterated to register the stereo images and generate 3D point cloud data of the exploded bubble.

[0012] This preferred scheme ensures that the registration of binocular images conforms to the physical continuity of volume changes of the exploding bubble during actual pulsation by introducing bubble volume constraints. By constructing a joint objective function of point cloud distance and volume through the sum of squared point cloud distances and the sum of squared point cloud volume differences, it ensures that the generated 3D point cloud data of the exploding bubble is not only aligned in geometric position but also maintains reasonable volume, avoiding incorrect matching. This achieves spatial morphological representation of the exploding bubble from the image dimension, thereby improving the accuracy of subsequent 3D energy density field and submarine damage assessment results.

[0013] As a preferred embodiment, the step of extracting collapse feature data and sound field feature data of the exploding bubble based on the sound field data, and constructing sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data, includes: The sound field data includes: sound pressure signal and sound pressure signal arrival time; Based on the difference in arrival time of the sound pressure signal, and combined with a preset sound source localization algorithm, the sound source localization data of the exploding bubble is determined. Based on the sound pressure signal, combined with the preset water density and water sound velocity, the sound intensity data of the exploding bubble is determined; Based on the sound intensity data and the sound source location data of the exploding bubble, the sound power data of the exploding bubble is determined. The sound pressure signal is subjected to a short-time Fourier transform to determine the sound field spectrum characteristics of the exploding bubble; Autocorrelation analysis was performed on the sound pressure signal to determine the pulsation period of the exploding bubble; Based on the sound source location data, sound intensity data, sound power data, pulsation period, and sound field spectrum characteristics of the exploding bubble, the sound field characteristic data of the exploding bubble are determined. Based on the sound pressure signal and the sound field spectrum characteristics, the collapse time of the exploding bubble is identified; Based on the sound source localization data and the image data, the collapse direction of the exploding bubble is identified; Based on the collapse time and collapse direction of the exploded bubble, the collapse characteristic data of the exploded bubble are determined; The collapse characteristic data and sound field characteristic data of the exploded bubble are used as the sound field modeling data of the exploded bubble.

[0014] This preferred scheme not only utilizes sound pressure signals to calculate sound intensity and power to assess sound energy levels, but also extracts the sound field spectrum characteristics and pulsation period of the bubble through short-time Fourier transform and autocorrelation analysis. Subsequently, it identifies the collapse time and direction of the exploding bubble. The collapse mechanism of the exploding bubble is characterized by the sound field modeling data of the exploding bubble, which comprehensively reflects the sound source characteristics and energy distribution of the exploding bubble, thereby improving the accuracy of subsequent three-dimensional energy density field and submarine damage assessment results.

[0015] As a preferred embodiment, the step of extracting the local features of bubble pulsation of the exploded bubble based on the pressure data, and determining the global bubble collapse energy release value of the exploded bubble based on the local features of bubble pulsation, includes: The pressure data includes: the pressure signal at each pressure signal measurement point; Wavelet decomposition and Fourier spectrum analysis were performed on the pressure signal to determine the local characteristics of bubble pulsation at each pressure signal measurement point of the exploding bubble; Based on the local characteristics of the bubble pulsation, the time-domain integral and frequency-domain integral of the exploding bubble at each pressure signal measurement point are determined; Based on the time-domain integral and frequency-domain integral, the estimated collapse energy of the exploding bubble at each pressure signal measurement point is determined; The estimated collapse energy of the exploding bubble at each pressure signal measurement point is input into a preset sound propagation model to determine the bubble collapse energy release value at each pressure signal measurement point. The global bubble collapse energy release value of the exploded bubble is determined by weighted averaging of the bubble collapse energy release value at each pressure signal measurement point.

[0016] This preferred scheme, through wavelet decomposition and Fourier spectrum analysis of the pressure signal, can precisely separate the pressure characteristics of different frequency bands and time periods, thereby extracting the local bubble pulsation features reflecting drastic changes in the local flow field. Subsequently, time-domain and frequency-domain integration based on the local bubble pulsation features can more accurately estimate the collapse energy estimate of each pressure signal measurement point. Furthermore, the attenuation and distortion of energy during propagation are compensated by an acoustic propagation model. Finally, weighted averaging improves the comprehensiveness of the global bubble collapse energy release value and avoids errors caused by single-point measurements. This enables energy analysis of the exploding bubble from a pressure dimension, improving the accuracy of subsequent three-dimensional energy density field and submarine damage assessment results.

[0017] As a preferred embodiment, the step of reconstructing the three-dimensional energy field of the exploded bubble based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploded bubble, the sound field modeling data of the exploded bubble, and the global bubble collapse energy release value, to determine the three-dimensional energy inversion data of the exploded bubble, includes: Based on a preset Bayesian inversion algorithm, the three-dimensional space of the exploding bubble is discretized into several voxels; Based on the three-dimensional point cloud data of the exploded bubble, an image constraint for the exploded bubble is constructed; Based on the sound field modeling data of the exploding bubble, an acoustic constraint for the exploding bubble is constructed. Based on the global bubble collapse energy release value, an explosion bubble pressure constraint is constructed; Based on the preset gradient descent method and the preset Markov chain Monte Carlo method, combined with the exploding bubble image constraint, exploding bubble acoustic constraint and exploding bubble pressure constraint, the energy density of each voxel is iteratively solved until the preset maximum number of iterations is met, and the optimal energy density of each voxel is determined. The three-dimensional energy density field of the exploding bubble is determined based on the optimal energy density of each voxel.

[0018] This preferred scheme discretizes the three-dimensional space of the exploding bubble into several voxels using a Bayesian inversion algorithm. Then, data from three different sources are used to construct exploding bubble image constraints, exploding bubble acoustic constraints, and exploding bubble pressure constraints. The exploding bubble image constraints ensure that the reconstructed three-dimensional energy density field morphology is consistent with the three-dimensional point cloud; the exploding bubble acoustic constraints ensure that the exploding bubble sound field modeling data matches the actual exploding bubble pulse behavior; and the exploding bubble pressure constraints ensure that the total energy release matches the global bubble collapse energy release value. Subsequently, gradient descent and Markov chain Monte Carlo methods are used to solve the problem, which not only finds the most probable energy distribution but also assesses the uncertainty of the energy distribution. This achieves accurate visualization and quantification of the energy density of the exploding bubble in three-dimensional space, enabling comprehensive analysis of the actual pulsating behavior of the exploding bubble from multiple data sources. This avoids the problems of low analysis accuracy, incomplete data, and poor repeatability caused by a single data source, thus improving the accuracy of submarine damage assessment based on exploding bubble pulsating behavior analysis.

[0019] As a preferred embodiment, the step of constructing a bubble structure damage mapping model based on a preset machine learning regression algorithm and pre-acquired historical data of submarine damage experiments, and inputting the three-dimensional energy density field of the exploded bubble into the bubble structure damage mapping model to determine the submarine damage assessment result includes: Based on the three-dimensional energy density field of the exploded bubble, a feature vector for submarine damage assessment is constructed; The initial machine learning model was obtained, and the submarine was divided into several functional areas. Based on the functional areas, the pre-acquired historical data of submarine damage experiments was divided to construct model training data. Based on a preset machine learning regression algorithm, the initial machine learning model is trained using model training data to construct a bubble structure damage mapping model. The submarine damage assessment feature vector is input into the bubble structure damage mapping model to determine the maximum stress result, maximum deformation result, and damage level of each functional area of ​​the submarine. The submarine damage assessment results are determined based on the maximum stress, maximum deformation, and damage level of each functional area.

[0020] This preferred solution divides the submarine into several functional areas, enabling modular and detailed analysis of submarine damage. By constructing a bubble structure damage mapping model through machine learning regression algorithms, a generalizable mapping mechanism between the behavior parameters of the exploding bubble and the degree of structural damage is established, thereby achieving accurate assessment of submarine damage and improving the accuracy of submarine damage assessment based on the analysis of exploding bubble pulsation behavior.

[0021] Accordingly, this invention provides a submarine damage assessment system based on multi-source bubble energy analysis, including: a multi-source data acquisition module for exploding bubbles, an image data analysis module, a sound field data analysis module, a pressure signal analysis module, a multi-source bubble energy analysis module, and a submarine damage assessment module; The multi-source data acquisition module for explosive bubbles is used to acquire image data, sound field data, and pressure data of explosive bubbles during submarine damage experiments. The image data analysis module is used to extract image features from the image data based on a preset subpixel-level edge fitting algorithm to determine the coordinate data of the exploding bubble; and to register the coordinate data of the exploding bubble based on a preset three-dimensional point cloud algorithm to generate three-dimensional point cloud data of the exploding bubble. The sound field data analysis module is used to extract the collapse feature data and sound field feature data of the exploding bubble based on the sound field data, and to construct the sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data. The pressure signal analysis module is used to extract the local features of bubble pulsation of the exploded bubble based on the pressure data, and to determine the global bubble collapse energy release value of the exploded bubble based on the local features of bubble pulsation. The multi-source bubble energy analysis module is used to reconstruct the three-dimensional energy field of the exploding bubble based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploding bubble, the sound field modeling data of the exploding bubble, and the global bubble collapse energy release value, and to determine the three-dimensional energy density field of the exploding bubble. The submarine damage assessment module is used to construct a bubble structure damage mapping model based on a preset machine learning regression algorithm and pre-acquired historical data of submarine damage experiments. The three-dimensional energy density field of the exploded bubble is input into the bubble structure damage mapping model to determine the submarine damage assessment result.

[0022] As a preferred embodiment, the image data analysis module includes: an image data analysis unit; In the image data analysis unit, the image data includes several binocular images; The image data analysis unit is used to perform pixel-level coarse edge extraction and sub-pixel-level edge localization on each of the stereo images based on a preset sub-pixel-level edge fitting algorithm, so as to extract image features from each of the stereo images and obtain the explosion bubble coordinate data of each of the stereo images. A joint objective function for bubble volume constraint and point cloud distance volume is constructed, and the binocular image is registered by combining a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to generate 3D point cloud data of the exploded bubble.

[0023] Understandably, this system acquires image data, sound field data, and pressure data of the exploding bubbles from different dimensions during submarine damage experiments. Then, it generates 3D point cloud data of the exploding bubbles using sub-pixel-level edge fitting and 3D point cloud algorithms, providing the spatial morphology of the exploding bubbles from an image perspective. Collapse feature data and sound field feature data are extracted from the sound field data, and the collapse mechanism of the exploding bubbles is characterized using the sound field modeling data. Finally, local features of bubble pulsation are extracted from the pressure data, and the global bubble collapse energy release value is calculated, thus achieving energy analysis of the exploding bubbles from a pressure perspective. Subsequently, a Bayesian inversion algorithm was used to fuse multi-source data from three dimensions: image, sound field, and pressure, thereby constructing a more comprehensive and accurate three-dimensional energy density field. This enabled comprehensive analysis of the actual pulsation behavior of exploded bubbles from multiple data sources, avoiding the problems of low analysis accuracy, incomplete data, and poor repeatability caused by a single data source. Then, a bubble structure damage mapping model was constructed using a machine learning regression algorithm, establishing a generalizable mapping mechanism between the behavior parameters of exploded bubbles and the degree of structural damage. This enabled accurate assessment of submarine damage and improved the accuracy of submarine damage assessment based on the analysis of the pulsation behavior of exploded bubbles. Attached Figure Description

[0024] Figure 1 A flowchart illustrating the steps of a submarine damage assessment method based on multi-source bubble energy analysis, provided in this embodiment of the invention; Figure 2 This is a schematic diagram of a submarine damage assessment system based on multi-source bubble energy analysis, provided as an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0026] Example 1 Please refer to Figure 1 , Figure 1 The flowchart of a submarine damage assessment method based on multi-source bubble energy analysis provided in this embodiment of the invention includes steps S101 to S106.

[0027] Step S101: Acquire image data, sound field data, and pressure data of the explosion bubbles during the submarine damage experiment.

[0028] In one optional embodiment, during the submarine damage experiment, four 10,000fps underwater high-speed cameras are used to capture images of the exploding bubble. The images captured by the underwater high-speed cameras are pre-processed using an NVIDIA Jetson TX2 to obtain image data of the exploding bubble. A 6-node sonar array is set up with a working frequency band of 10-200kHz and a TDOA (Time Difference of Arrival) positioning accuracy of ±0.1°. The sound field data of the exploding bubble is obtained through the 6-node sonar array. A 12-channel pressure sensor network with a resolution of 0.001MPa and a dynamic range of 160dB is set up to obtain pressure data of the exploding bubble.

[0029] Step S102: Based on a preset subpixel-level edge fitting algorithm, extract image features from the image data to determine the coordinate data of the exploding bubble; based on a preset three-dimensional point cloud algorithm, register the coordinate data of the exploding bubble to generate three-dimensional point cloud data of the exploding bubble.

[0030] In this embodiment, the step of extracting image features from the image data based on a preset sub-pixel level edge fitting algorithm to determine the coordinate data of the exploded bubble; and registering the coordinate data of the exploded bubble based on a preset 3D point cloud algorithm to generate 3D point cloud data of the exploded bubble, includes: The image data includes several binocular images; Based on a preset subpixel-level edge fitting algorithm, pixel-level coarse edge extraction and subpixel-level edge localization are performed on each of the stereo images to extract image features from each of the stereo images and obtain the bubble coordinate data of each of the stereo images. A joint objective function for bubble volume constraint and point cloud distance volume is constructed, and the binocular image is registered by combining a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to generate 3D point cloud data of the exploded bubble.

[0031] This embodiment employs a subpixel-level edge fitting algorithm for pixel-level coarse edge extraction and subpixel-level edge localization. Subpixel-level edge localization, building upon pixel-level coarse extraction, significantly reduces quantization errors in the coordinate data of the exploding bubble. Subsequently, a bubble volume constraint is introduced to ensure that the registration of the binocular images conforms to the physical continuity of the volume changes of the exploding bubble during actual pulsation. Through a dual joint objective function of point cloud distance and volume, the generated 3D point cloud data of the exploding bubble is ensured to be not only geometrically aligned but also maintain reasonable volume, avoiding incorrect matching. This achieves spatial morphological representation of the exploding bubble from an image dimension, thereby improving the accuracy of subsequent 3D energy density field and submarine damage assessment results.

[0032] In this embodiment, the step of performing pixel-level coarse edge extraction and sub-pixel-level edge localization on each of the stereo images based on a preset sub-pixel-level edge fitting algorithm, in order to extract image features from each of the stereo images and obtain the coordinate data of the exploding bubble in each of the stereo images, includes: Each of the stereo images is input into a preset graphics processing model to determine the image region of each stereo image; Based on a preset edge detection operator, edge detection is performed on each of the image regions to determine the edge mask map of each of the binocular images; Contour extraction is performed on the edge mask image to determine the contour point set for each binocular image; Gaussian filtering is applied to each of the stereo images, and the gradient data of each pixel in each stereo image after Gaussian filtering is calculated. Based on the gradient data, the neighboring pixels of each pixel in the contour point set of each stereo image are determined, and a preset fitting function is used to fit each pixel in the contour point set of each stereo image and its neighboring pixels to determine the sub-pixel position offset of each pixel in the contour point set of each stereo image. Based on the sub-pixel position offset, displacement compensation is performed on each pixel in the contour point set of each stereo image to determine the bubble contour point set of each stereo image. Based on the bubble outline point set of each stereo image, determine the pixel coordinates of the exploding bubble, the center coordinates of the exploding bubble, and the radius of the exploding bubble for each stereo image; The pixel coordinates, center coordinates, and radius of the exploding bubble in each stereo image are used as the exploding bubble coordinate data for each stereo image.

[0033] In one optional embodiment, the preset graphics processing model is set to the YOLOv7 depth model. Two sets of binocular camera systems are formed using four underwater high-speed cameras to capture several binocular images. Each binocular image is input into the YOLOv7 depth model to obtain the image region output by the YOLOv7. Then, the edge detection operator is set to the Canny edge detection operator. The image region is processed using the Canny edge detection operator to obtain a binary edge mask. Finally, the outermost closed contour in the edge mask is extracted to obtain the contour point set. For the contour point set Each pixel in Its satisfaction The coordinates are integers; then a Gaussian filter is applied to the stereo image, and the gradient data of each pixel in the Gaussian-filtered stereo image is calculated, including the gradient magnitude. and gradient direction Then for the contour point set Each pixel in In its gradient direction In the direction perpendicular to (i.e., the direction of the normal), take The distance between the front and back pixels (i.e., the distance between the front and back pixels) , (where the normal vector is a unit vector), to obtain two neighboring pixels; then, the gray values ​​of these three pixels are processed by a quadratic function. , and By fitting, we can obtain ,in, , and All are fitting coefficients. Represents pixels Then solve first derivative The zero solution yields the subpixel position offset. Then the subpixel position offset With pixels The coordinates of the edges are added together to achieve displacement compensation, thus obtaining the sub-pixel coordinates of the edge points. Therefore, the bubble outline point set is ; The corresponding two-dimensional coordinates are the pixel coordinates of the exploded bubble; then, the least squares method is used to fit it into a circle, the radius of the circle is the radius of the exploded bubble, and the center of the circle is the coordinate of the center of the exploded bubble.

[0034] It should be noted that the Canny edge detection operator is a classic edge detection algorithm widely used in computer vision, image segmentation, and object recognition. Gaussian filtering is a linear smoothing filtering technique based on the mathematical properties of the Gaussian function, widely used in image processing, signal processing, and computer vision; its core objective is to suppress high-frequency noise (such as Gaussian noise) in images or signals through local weighted averaging while preserving overall trends and edge information. The Least Squares Method is a mathematical optimization method that fits data or estimates model parameters by minimizing the sum of squared errors.

[0035] This embodiment achieves sub-pixel-level edge localization based on pixel-level coarse extraction through image region determination, edge detection, contour extraction, Gaussian filtering, and sub-pixel localization. The preset image processing model can effectively eliminate background interference, thereby focusing on the area where the exploded bubble is located. Edge detection and contour extraction lock the boundary of the exploded bubble, and Gaussian filtering smooths image noise, avoiding noise interference with subsequent high-precision gradient calculation. By calculating the gradient of the pixel and its neighbors, and using the fitting function to calculate the sub-pixel position offset, displacement compensation is achieved, thereby realizing sub-pixel localization. This greatly reduces the quantization error of the exploded bubble coordinate data and improves the accuracy of subsequent three-dimensional energy density field and submarine damage assessment results.

[0036] In this embodiment, the step of constructing a joint objective function for bubble volume constraint and point cloud distance volume, and combining it with a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to register the binocular image and generate 3D point cloud data of the exploded bubble includes: Obtain a bubble volume threshold and construct a bubble volume constraint based on the bubble volume threshold; Construct a point cloud distance sum of squares term and a point cloud volume difference squares term, and construct a joint objective function for point cloud distance and volume based on the point cloud distance sum of squares term and the point cloud volume difference squares term; The iteration objective is to minimize the joint objective function of point cloud distance and volume. Based on the preset 3D point cloud algorithm and the bubble volume constraint, the coordinate data of the exploded bubble in each stereo image is iterated to register the stereo images and generate 3D point cloud data of the exploded bubble.

[0037] In an optional embodiment, the bubble volume threshold Set to a very small value, the bubble volume constraint is set as follows: , The volume of the bubble in the current frame. The bubble volume is from the previous frame; the sum of squared point cloud distances is... The squared term of the point cloud volume difference is Therefore, the joint objective function for point cloud distance and volume is: The point cloud distance sum of squares term is used to sum the squared distances of all matching point pairs. Minimizing this term is to align the two point clouds geometrically. The point cloud volume difference squared term is a physical constraint term, which is calculated and estimated under the current transformation matrix (R, t). and Minimizing the square of the difference is to ensure that the bubble volume does not change abruptly, which is in accordance with the laws of fluid dynamics. It is the regularization coefficient (a hyperparameter greater than 0), used to balance the importance of geometric fitting accuracy and physical plausibility; the 3D point cloud algorithm is set as the ICP algorithm, first taking the 3D point cloud data (exploded bubble coordinate data) of a stereo image, denoted as the source point cloud. It is an N x 3 matrix; the other image is the 3D point cloud data of the exploded bubble coordinates (exploded bubble coordinate data), denoted as the target point cloud. It is an M-row, 3-column matrix; under the premise of satisfying the bubble volume constraint, for the source point cloud Each point in In the target point cloud Find the corresponding point with the nearest Euclidean distance in the middle. Then, the Singular Value Decomposition (SVD) method was used to initially calculate an optimal transformation that minimizes only the sum of squared distance terms in the point cloud. ;Will Application to source cloud Temporary point cloud obtained The bubble volume in the current frame is calculated using either the Convex Hull or Alpha Shape algorithm. After that and As initial values, nonlinear optimizers such as the Levenberg-Marquardt algorithm are used to iteratively solve the problem with the goal of minimizing the joint objective function of point cloud distance and volume. The problem is solved until the transformation matrix converges (the change is less than the threshold) or the maximum number of iterations is reached, thus completing the registration of the stereo images and generating 3D point cloud data of the exploding bubble.

[0038] It should be noted that the ICP algorithm (Iterative Closest Point) is a registration algorithm used to calculate the optimal geometric transformation. Its core idea is to iteratively search for the closest point correspondence between two point cloud sets (or surfaces) and calculate the rigid body transformation (rotation and translation) that minimizes the distance error between these corresponding points, ultimately aligning the two point clouds to the same coordinate system.

[0039] This embodiment introduces bubble volume constraints to ensure that the registration of binocular images conforms to the physical continuity of volume changes of the exploding bubble during actual pulsation. By constructing a joint objective function of point cloud distance and volume using the sum of squared point cloud distances and the sum of squared point cloud volume differences, it ensures that the generated 3D point cloud data of the exploding bubble is not only geometrically aligned but also maintains reasonable volume, avoiding incorrect matching. This achieves spatial morphological representation of the exploding bubble from the image dimension, thereby improving the accuracy of subsequent 3D energy density field and submarine damage assessment results.

[0040] Step S103: Based on the sound field data, extract the collapse feature data and sound field feature data of the exploding bubble, and construct the sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data.

[0041] In this embodiment, the step of extracting collapse feature data and sound field feature data of the exploding bubble based on the sound field data, and constructing sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data, includes: The sound field data includes: sound pressure signal and sound pressure signal arrival time; Based on the difference in arrival time of the sound pressure signal, and combined with a preset sound source localization algorithm, the sound source localization data of the exploding bubble is determined. Based on the sound pressure signal, combined with the preset water density and water sound velocity, the sound intensity data of the exploding bubble is determined; Based on the sound intensity data and the sound source location data of the exploding bubble, the sound power data of the exploding bubble is determined. The sound pressure signal is subjected to a short-time Fourier transform to determine the sound field spectrum characteristics of the exploding bubble; Autocorrelation analysis was performed on the sound pressure signal to determine the pulsation period of the exploding bubble; Based on the sound source location data, sound intensity data, sound power data, pulsation period, and sound field spectrum characteristics of the exploding bubble, the sound field characteristic data of the exploding bubble are determined. Based on the sound pressure signal and the sound field spectrum characteristics, the collapse time of the exploding bubble is identified; Based on the sound source localization data and the image data, the collapse direction of the exploding bubble is identified; Based on the collapse time and collapse direction of the exploded bubble, the collapse characteristic data of the exploded bubble are determined; The collapse characteristic data and sound field characteristic data of the exploded bubble are used as the sound field modeling data of the exploded bubble.

[0042] In one optional embodiment, the sound field data includes: sound pressure signal and sound pressure signal arrival time, specifically, the sound pressure signal received by each node in the sonar array and the sound pressure signal arrival time; the difference in sound pressure signal arrival time at different nodes in the sonar array is calculated, and the TDOA algorithm is used to obtain the sound source localization data of the exploding bubble, which is the three-dimensional spatial location of the bubble pulsation; since the square of the sound pressure is proportional to the sound intensity, and the sound intensity is the sound energy flux density, therefore, the formula is used to... , For sound intensity, For sound pressure, For the density of water, Given the speed of sound in water, calculate the sound intensity data of the exploding bubble; then use the formula... , The distance from each node in the sonar array to the explosion bubble is obtained based on the sound source localization data. The radiated sound power is the sound power data of the exploding bubble. A short-time Fourier transform (STFT) is then performed on the sound pressure signal to obtain the sound field spectrum characteristics of the exploding bubble. The energy distribution in the sound field spectrum characteristics reflects the severity of the bubble's collapse and the frequency domain characteristics of energy release. Autocorrelation analysis is then performed on the sound pressure signal to determine the pulsation period of the exploding bubble. The pulsation period can also be determined by detecting the spectral peaks of the sound field spectrum characteristics, with the interval between the main peaks representing the pulsation period of the exploding bubble. Since the sound pressure signal and the sound field spectrum characteristics exhibit an extremely short, high-amplitude pulse at the moment of the bubble's collapse, the collapse moment can be identified by detecting the sound pressure signal or the sound field spectrum characteristics and using envelope or wavelet transform modulus maxima. The collapse location of the exploding bubble is then obtained through sound source localization data, and the collapse direction is determined by observing the bubble's morphology (e.g., jet direction) through image data. For example, if the sound source localization data shows that the collapse location of the exploding bubble is significantly deviated from the bubble's geometric center, and the image data shows that one side of the bubble is indented, the collapse direction can be determined.

[0043] It should be noted that the Short-Time Fourier Transform (STFT) is a joint time-frequency analysis method. It uses a sliding window function to segment a non-stationary signal into short time intervals, performs a Fourier transform within each interval, and generates a three-dimensional spectrum of time, frequency, and amplitude. Autocorrelation is a measure of the correlation between the same signal or data sequence at different time or spatial points, used to reveal the inherent periodicity, repetitive patterns, or dependencies in the data. Wavelet transform modulus maxima extract information about abrupt changes in the signal by finding local maxima of the wavelet transform coefficients and tracing the changes of these maxima with scale. The lines connecting these maxima can be seen as a representation of the signal's envelope at different resolutions.

[0044] This embodiment not only utilizes sound pressure signals to calculate sound intensity and power to assess sound energy levels, but also extracts the sound field spectrum characteristics and pulsation period of the bubble through short-time Fourier transform and autocorrelation analysis. Subsequently, it identifies the collapse time and direction of the exploding bubble. The collapse mechanism of the exploding bubble is characterized by the sound field modeling data of the exploding bubble, which comprehensively reflects the sound source characteristics and energy distribution of the exploding bubble, thereby improving the accuracy of subsequent three-dimensional energy density field and submarine damage assessment results.

[0045] Step S104: Extract the local bubble pulsation features of the exploded bubble based on the pressure data, and determine the global bubble collapse energy release value of the exploded bubble based on the local bubble pulsation features.

[0046] In this embodiment, the step of extracting the local features of bubble pulsation of the exploding bubble based on the pressure data, and determining the global bubble collapse energy release value of the exploding bubble based on the local features of bubble pulsation, includes: The pressure data includes: the pressure signal at each pressure signal measurement point; Wavelet decomposition and Fourier spectrum analysis were performed on the pressure signal to determine the local characteristics of bubble pulsation at each pressure signal measurement point of the exploding bubble; Based on the local characteristics of the bubble pulsation, the time-domain integral and frequency-domain integral of the exploding bubble at each pressure signal measurement point are determined; Based on the time-domain integral and frequency-domain integral, the estimated collapse energy of the exploding bubble at each pressure signal measurement point is determined; The estimated collapse energy of the exploding bubble at each pressure signal measurement point is input into a preset sound propagation model to determine the bubble collapse energy release value at each pressure signal measurement point. The global bubble collapse energy release value of the exploded bubble is determined by weighted averaging of the bubble collapse energy release value at each pressure signal measurement point.

[0047] In an optional embodiment, wavelet decomposition uses Daubechies (dbN) or Symlets (symN) wavelets, whose tight support and approximate symmetry are more conducive to capturing transient impact features. Multi-resolution analysis of the pressure signal is performed through wavelet decomposition, decomposing the pressure signal into different frequency bands to separate the low-frequency pressure fluctuations generated by the exploding bubble pulsation and the high-frequency impact components generated by the collapse event, while filtering out some environmental noise. Then, Fourier spectrum analysis is performed on the separated frequency band signals related to the exploding bubble, transforming them to the frequency domain, observing their spectral characteristics, and calculating the pulse energy of the exploding bubble to obtain the collapse pulse signal. The collapse pulse signal is used as a local feature of the bubble pulsation (pulse). The energy is proportional to the square of the spectral amplitude. Then, the time-domain integral (integrating with respect to the square of the pressure) or frequency-domain integral (integrating with respect to the square of the spectral amplitude) of the collapse pulse signal is calculated as the time-domain integral and frequency-domain integral of the pressure signal measurement point. The time-domain integral or frequency-domain integral is selected as the estimated collapse energy of the pressure signal measurement point. Then, the sound propagation model (the spherical wave attenuation model is selected in this embodiment) is used to back-calculate the bubble collapse energy release value of the exploded bubble at each pressure signal measurement point. Then, the weighted average of all bubble collapse energy release values ​​is performed to obtain the global bubble collapse energy release value.

[0048] It should be noted that wavelet decomposition is a signal analysis method based on wavelet transform. It achieves localized analysis in the time and frequency domain by decomposing the signal into wavelet coefficients of different scales (frequency) and locations. Fourier Spectrum Analysis, based on the core technology of Fourier transform, decomposes a time-domain signal into sine / cosine components of different frequencies, revealing the signal's frequency components, intensity (amplitude spectrum), and phase relationship (phase spectrum). The Spherical Wave Attenuation Model describes the physical law that the energy of a wave with a concentric spherical wavefront (such as sound waves, electromagnetic waves, or seismic waves generated by a point source) attenuates with increasing propagation distance when propagating in a free field.

[0049] This embodiment uses wavelet decomposition and Fourier spectrum analysis to finely separate pressure characteristics at different frequency bands and time periods, thereby extracting local bubble pulsation features that reflect drastic changes in the local flow field. Subsequent time-domain and frequency-domain integration based on these bubble pulsation features allows for a more accurate estimation of the collapse energy at each pressure signal measurement point. Furthermore, an acoustic propagation model compensates for energy attenuation and distortion during propagation. Weighted averaging improves the comprehensiveness of the global bubble collapse energy release value, avoiding errors caused by single-point measurements. This achieves energy analysis of the exploding bubble from a pressure perspective, improving the accuracy of subsequent three-dimensional energy density field and submarine damage assessment results.

[0050] Step S105: Based on the preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploded bubble, the sound field modeling data of the exploded bubble, and the global bubble collapse energy release value, the three-dimensional energy field of the exploded bubble is reconstructed to determine the three-dimensional energy density field of the exploded bubble.

[0051] In this embodiment, the step of reconstructing the three-dimensional energy field of the exploded bubble based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploded bubble, the sound field modeling data of the exploded bubble, and the global bubble collapse energy release value, to determine the three-dimensional energy inversion data of the exploded bubble, includes: Based on a preset Bayesian inversion algorithm, the three-dimensional space of the exploding bubble is discretized into several voxels; Based on the three-dimensional point cloud data of the exploded bubble, an image constraint for the exploded bubble is constructed; Based on the sound field modeling data of the exploding bubble, an acoustic constraint for the exploding bubble is constructed. Based on the global bubble collapse energy release value, an explosion bubble pressure constraint is constructed; Based on the preset gradient descent method and the preset Markov chain Monte Carlo method, combined with the exploding bubble image constraint, exploding bubble acoustic constraint and exploding bubble pressure constraint, the energy density of each voxel is iteratively solved until the preset maximum number of iterations is met, and the optimal energy density of each voxel is determined. The three-dimensional energy density field of the exploding bubble is determined based on the optimal energy density of each voxel.

[0052] In one optional embodiment, the three-dimensional space of the exploding bubble is discretized into several voxels using a Bayesian inversion algorithm. Then the energy density was set to Based on the 3D point cloud data of the exploding bubble, image constraints for the exploding bubble are constructed. The image constraints for the exploding bubble are as follows: ; This is the 3D point cloud data of the exploding bubble. Let be the image projection matrix, which represents the projection relationship from the three-dimensional energy field to the edges of the two-dimensional image. Each element in Represents the first Energy density of individual elements on the first The contribution of each pixel can be determined by factors such as the geometric line of sight from the voxel to the camera and the point spread function; this embodiment does not impose further limitations on these factors. Then, based on the exploding bubble sound field modeling data, exploding bubble acoustic constraints are constructed. The exploding bubble acoustic constraints are... ; Data for modeling the sound field of exploding bubbles; The acoustic propagation matrix represents the propagation relationship from the three-dimensional energy field to each node of the sonar array. Each element in Represents the first Individual elements, acting as sound sources, release energy that affects the first element in the sonar array. The contribution of each node is determined by the sound propagation attenuation law and time delay; based on the global bubble collapse energy release value, an explosion bubble pressure constraint is constructed, and the explosion bubble pressure constraint is... ; This represents the global energy release value from bubble collapse. This is the pressure transmission matrix, which represents the transmission relationship from the three-dimensional energy field to the location of the pressure signal measurement point. Each element in Represents the first Drastic energy changes (such as collapse) of individual elements on the first The influence of pressure readings at each pressure signal measurement point is determined by the propagation attenuation and time relationship of the pressure wave. Then, the image constraints, acoustic constraints, and pressure constraints of the exploding bubble are constructed as a least-squares optimization problem (maximizing the posterior probability). A preliminary solution is quickly obtained using gradient descent (or conjugate gradient method). Using the solution obtained by gradient descent as the initial point, the mean and variance of the energy density of each voxel are calculated using the Markov chain Monte Carlo method (MCMC), thus obtaining the final reconstruction result and uncertainty assessment. After satisfying the preset maximum number of iterations, the optimal energy density of each voxel is determined, generating a three-dimensional energy density field with a resolution of 10 cm³ (in VTK visualization file format).

[0053] It should be noted that Bayesian inversion is a parameter estimation method based on Bayesian statistical theory. It combines prior information (initial knowledge of unknown parameters) with observed data, using Bayes' theorem to calculate the posterior probability distribution of model parameters. Gradient descent is an optimization algorithm based on the first derivative (gradient), iteratively adjusting model parameters to minimize the objective function (such as a loss function). Markov Chain Monte Carlo (MCMC) is a statistical sampling technique based on Markov chains and Monte Carlo integrals. It constructs a Markov chain whose stationary distribution matches the target probability distribution, thereby drawing samples from a complex distribution to approximate the statistical value.

[0054] This embodiment uses a Bayesian inversion algorithm to discretize the three-dimensional space of the exploding bubble into several voxels. Then, data from three different sources are used to construct exploding bubble image constraints, exploding bubble acoustic constraints, and exploding bubble pressure constraints. The exploding bubble image constraints ensure that the reconstructed three-dimensional energy density field morphology is consistent with the three-dimensional point cloud; the exploding bubble acoustic constraints ensure that the exploding bubble sound field modeling data matches the actual exploding bubble pulse behavior; and the exploding bubble pressure constraints ensure that the total energy release matches the global bubble collapse energy release value. Then, the gradient descent method and Markov chain Monte Carlo method are used to solve the problem. This not only finds the most likely energy distribution but also assesses the uncertainty of the energy distribution. Thus, accurate visualization and quantification of the energy density of the exploding bubble in three-dimensional space are achieved. This enables comprehensive analysis of the actual pulsating behavior of the exploding bubble from multiple data sources, avoiding the problems of low analysis accuracy, incomplete data, and poor repeatability caused by a single data source. This improves the accuracy of submarine damage assessment based on the analysis of exploding bubble pulsating behavior.

[0055] Step S106: Based on the preset machine learning regression algorithm and combined with the pre-acquired historical data of submarine damage experiments, construct a bubble structure damage mapping model, input the three-dimensional energy density field of the exploded bubble into the bubble structure damage mapping model, and determine the submarine damage assessment result.

[0056] In this embodiment, the step of constructing a bubble structure damage mapping model based on a preset machine learning regression algorithm and pre-acquired historical data of submarine damage experiments, and inputting the three-dimensional energy density field of the exploded bubble into the bubble structure damage mapping model to determine the submarine damage assessment result includes: Based on the three-dimensional energy density field of the exploded bubble, a feature vector for submarine damage assessment is constructed; The initial machine learning model was obtained, and the submarine was divided into several functional areas. Based on the functional areas, the pre-acquired historical data of submarine damage experiments was divided to construct model training data. Based on a preset machine learning regression algorithm, the initial machine learning model is trained using model training data to construct a bubble structure damage mapping model. The submarine damage assessment feature vector is input into the bubble structure damage mapping model to determine the maximum stress result, maximum deformation result, and damage level of each functional area of ​​the submarine. The submarine damage assessment results are determined based on the maximum stress, maximum deformation, and damage level of each functional area.

[0057] In one optional embodiment, the initial machine learning model is set as an XGBoost model. The pre-acquired historical data of submarine damage experiments includes: finite element simulation results of the submarine and historical deep-water explosion experiment data. The finite element simulation results can be obtained through underwater explosion fluid-structure interaction simulation using finite element simulation software or self-written code, including but not limited to: stress time history, plastic strain, deformation displacement, and other response data of the submarine's various functional areas (such as pressure hull, rudder, and connecting compartments) under different explosion bubble parameters. The historical deep-water explosion experiment data includes but is not limited to: structural response data measured by various sensors (such as strain gauges and accelerometers), and observation records of the macroscopic damage morphology of the structure after the experiment (such as crack length and indentation depth). Then, the pre-acquired historical data of submarine damage experiments is divided according to functional areas to construct model training data. The XGBoost model is trained using the gradient boosting algorithm with mean squared error (MSE) or cross-entropy as the loss function, and the feature space adaptive technique is used. The model is fine-tuned using the Adaptation method to form a bubble structure damage mapping model. Then, data such as the collapse location (x, y, z), maximum radius, pulsation period, local energy density peak, and collapse direction vector of the exploded bubble are extracted from the three-dimensional energy density field to construct a submarine damage assessment feature vector. The submarine damage assessment feature vector is then input into the bubble structure damage mapping model to obtain the maximum stress result, maximum deformation result, and damage level of each functional area of ​​the submarine.

[0058] It's important to note that Feature Space Adaptation (MSA) is a technique that dynamically adjusts or optimizes feature representations to enable a model to adapt to different data distributions or task requirements. Its core lies in mapping data from both the source and target domains to the same feature space and using metrics such as Maximum Mean Difference (MMD) to reduce the distributional differences between the two domains within that space, thereby improving the model's generalization ability in the target domain. This process typically involves feature extraction, distribution alignment, and parameter optimization, aiming to address the performance degradation caused by inconsistent cross-domain data distributions.

[0059] This embodiment divides the submarine into several functional areas, enabling modular and detailed analysis of submarine damage. By constructing a bubble structure damage mapping model through machine learning regression algorithms, a generalizable mapping mechanism between the behavior parameters of the exploding bubble and the degree of structural damage is established, thereby achieving accurate assessment of submarine damage and improving the accuracy of submarine damage assessment based on the analysis of exploding bubble pulsation behavior.

[0060] This embodiment acquires image data, sound field data, and pressure data of the exploded bubble from different dimensions during a submarine damage experiment. Then, it generates 3D point cloud data of the exploded bubble using a subpixel-level edge fitting algorithm and a 3D point cloud algorithm, providing the spatial morphology of the exploded bubble from an image perspective. Collapse feature data and sound field feature data are extracted from the sound field data, and the collapse mechanism of the exploded bubble is characterized using the sound field modeling data. Subsequently, local features of bubble pulsation are extracted from the pressure data, and the global bubble collapse energy release value is calculated, thus achieving energy analysis of the exploded bubble from a pressure perspective. By fusing multi-source data from three dimensions—image, sound field, and pressure—using a Bayesian inversion algorithm, a more comprehensive and accurate three-dimensional energy density field is constructed. This enables comprehensive analysis of the actual pulsation behavior of exploded bubbles from multiple data sources, avoiding the problems of low analysis accuracy, incomplete data, and poor repeatability caused by single data sources. Subsequently, a bubble structure damage mapping model is constructed using a machine learning regression algorithm, establishing a generalizable mapping mechanism between exploded bubble behavior parameters and the degree of structural damage. This enables accurate assessment of submarine damage and improves the accuracy of submarine damage assessment based on the analysis of exploded bubble pulsation behavior.

[0061] Example 2 Please refer to Figure 2 , Figure 2 A schematic diagram of a submarine damage assessment system based on multi-source bubble energy analysis provided in an embodiment of the present invention includes: an explosion bubble multi-source data acquisition module 201, an image data analysis module 202, an acoustic field data analysis module 203, a pressure signal analysis module 204, a multi-source bubble energy analysis module 205, and a submarine damage assessment module 206. The multi-source data acquisition module 201 for explosive bubbles is used to acquire image data, sound field data and pressure data of explosive bubbles during the submarine damage experiment. The image data analysis module 202 is used to extract image features from the image data based on a preset subpixel-level edge fitting algorithm to determine the coordinate data of the exploding bubble; and to register the coordinate data of the exploding bubble based on a preset three-dimensional point cloud algorithm to generate three-dimensional point cloud data of the exploding bubble. The sound field data analysis module 203 is used to extract the collapse feature data and sound field feature data of the exploding bubble based on the sound field data, and to construct the sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data. The pressure signal analysis module 204 is used to extract the local features of bubble pulsation of the exploded bubble based on the pressure data, and to determine the global bubble collapse energy release value of the exploded bubble based on the local features of bubble pulsation. The multi-source bubble energy analysis module 205 is used to reconstruct the three-dimensional energy field of the exploding bubble based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploding bubble, the sound field modeling data of the exploding bubble, and the global bubble collapse energy release value, and to determine the three-dimensional energy density field of the exploding bubble. The submarine damage assessment module 206 is used to construct a bubble structure damage mapping model based on a preset machine learning regression algorithm and pre-acquired historical data of submarine damage experiments, and input the three-dimensional energy density field of the exploded bubble into the bubble structure damage mapping model to determine the submarine damage assessment result.

[0062] In this embodiment, the image data analysis module 202 includes: an image data analysis unit; In the image data analysis unit, the image data includes several binocular images; The image data analysis unit is used to perform pixel-level coarse edge extraction and sub-pixel-level edge localization on each of the stereo images based on a preset sub-pixel-level edge fitting algorithm, so as to extract image features from each of the stereo images and obtain the explosion bubble coordinate data of each of the stereo images. A joint objective function for bubble volume constraint and point cloud distance volume is constructed, and the binocular image is registered by combining a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to generate 3D point cloud data of the exploded bubble.

[0063] In this embodiment, the image data analysis unit includes: an explosion bubble coordinate data acquisition subunit; The explosion bubble coordinate data acquisition subunit is used to input each of the binocular images into a preset graphics processing model to determine the image region of each of the binocular images; Based on a preset edge detection operator, edge detection is performed on each of the image regions to determine the edge mask map of each of the binocular images; Contour extraction is performed on the edge mask image to determine the contour point set for each binocular image; Gaussian filtering is applied to each of the stereo images, and the gradient data of each pixel in each stereo image after Gaussian filtering is calculated. Based on the gradient data, the neighboring pixels of each pixel in the contour point set of each stereo image are determined, and a preset fitting function is used to fit each pixel in the contour point set of each stereo image and its neighboring pixels to determine the sub-pixel position offset of each pixel in the contour point set of each stereo image. Based on the sub-pixel position offset, displacement compensation is performed on each pixel in the contour point set of each stereo image to determine the bubble contour point set of each stereo image. Based on the bubble outline point set of each stereo image, determine the pixel coordinates of the exploding bubble, the center coordinates of the exploding bubble, and the radius of the exploding bubble for each stereo image; The pixel coordinates, center coordinates, and radius of the exploding bubble in each stereo image are used as the exploding bubble coordinate data for each stereo image.

[0064] In this embodiment, the image data analysis unit includes: a subunit for acquiring three-dimensional point cloud data of the exploded bubble; The three-dimensional point cloud data acquisition subunit for the exploding bubble is used to acquire the bubble volume threshold and construct bubble volume constraints based on the bubble volume threshold. Construct a point cloud distance sum of squares term and a point cloud volume difference squares term, and construct a joint objective function for point cloud distance and volume based on the point cloud distance sum of squares term and the point cloud volume difference squares term; The iteration objective is to minimize the joint objective function of point cloud distance and volume. Based on the preset 3D point cloud algorithm and the bubble volume constraint, the coordinate data of the exploded bubble in each stereo image is iterated to register the stereo images and generate 3D point cloud data of the exploded bubble.

[0065] In this embodiment, the sound field data analysis module 203 includes: a sound field data analysis unit; In the sound field data analysis unit, the sound field data includes: sound pressure signal and sound pressure signal arrival time; The sound field data analysis unit is used to determine the sound source location data of the exploding bubble based on the difference in arrival time of the sound pressure signal and in combination with a preset sound source location algorithm. Based on the sound pressure signal, combined with the preset water density and water sound velocity, the sound intensity data of the exploding bubble is determined; Based on the sound intensity data and the sound source location data of the exploding bubble, the sound power data of the exploding bubble is determined. The sound pressure signal is subjected to a short-time Fourier transform to determine the sound field spectrum characteristics of the exploding bubble; Autocorrelation analysis was performed on the sound pressure signal to determine the pulsation period of the exploding bubble; Based on the sound source location data, sound intensity data, sound power data, pulsation period, and sound field spectrum characteristics of the exploding bubble, the sound field characteristic data of the exploding bubble are determined. Based on the sound pressure signal and the sound field spectrum characteristics, the collapse time of the exploding bubble is identified; Based on the sound source localization data and the image data, the collapse direction of the exploding bubble is identified; Based on the collapse time and collapse direction of the exploded bubble, the collapse characteristic data of the exploded bubble are determined; The collapse characteristic data and sound field characteristic data of the exploded bubble are used as the sound field modeling data of the exploded bubble.

[0066] In this embodiment, the pressure signal analysis module 204 includes: a pressure signal analysis unit; In the pressure signal analysis unit, the pressure data includes: the pressure signal at each pressure signal measurement point; The pressure signal analysis unit is used to perform wavelet decomposition and Fourier spectrum analysis on the pressure signal to determine the local characteristics of bubble pulsation of the exploding bubble at each pressure signal measurement point. Based on the local characteristics of the bubble pulsation, the time-domain integral and frequency-domain integral of the exploding bubble at each pressure signal measurement point are determined; Based on the time-domain integral and frequency-domain integral, the estimated collapse energy of the exploding bubble at each pressure signal measurement point is determined; The estimated collapse energy of the exploding bubble at each pressure signal measurement point is input into a preset sound propagation model to determine the bubble collapse energy release value at each pressure signal measurement point. The global bubble collapse energy release value of the exploded bubble is determined by weighted averaging of the bubble collapse energy release value at each pressure signal measurement point.

[0067] In this embodiment, the multi-source bubble energy analysis module 205 includes: a multi-source bubble energy analysis unit; The multi-source bubble energy analysis unit is used to discretize the three-dimensional space of the exploding bubble into several voxels based on a preset Bayesian inversion algorithm. Based on the three-dimensional point cloud data of the exploded bubble, an image constraint for the exploded bubble is constructed; Based on the sound field modeling data of the exploding bubble, an acoustic constraint for the exploding bubble is constructed. Based on the global bubble collapse energy release value, an explosion bubble pressure constraint is constructed; Based on the preset gradient descent method and the preset Markov chain Monte Carlo method, combined with the exploding bubble image constraint, exploding bubble acoustic constraint and exploding bubble pressure constraint, the energy density of each voxel is iteratively solved until the preset maximum number of iterations is met, and the optimal energy density of each voxel is determined. The three-dimensional energy density field of the exploding bubble is determined based on the optimal energy density of each voxel.

[0068] In this embodiment, the submarine damage assessment module 206 includes: a submarine damage assessment unit; The submarine damage assessment unit is used to construct a submarine damage assessment feature vector based on the three-dimensional energy density field of the explosion bubble; The initial machine learning model was obtained, and the submarine was divided into several functional areas. Based on the functional areas, the pre-acquired historical data of submarine damage experiments was divided to construct model training data. Based on a preset machine learning regression algorithm, the initial machine learning model is trained using model training data to construct a bubble structure damage mapping model. The submarine damage assessment feature vector is input into the bubble structure damage mapping model to determine the maximum stress result, maximum deformation result, and damage level of each functional area of ​​the submarine. The submarine damage assessment results are determined based on the maximum stress, maximum deformation, and damage level of each functional area.

[0069] This embodiment acquires image data, sound field data, and pressure data of the exploded bubble from different dimensions during a submarine damage experiment. Then, it generates 3D point cloud data of the exploded bubble using a subpixel-level edge fitting algorithm and a 3D point cloud algorithm, providing the spatial morphology of the exploded bubble from an image perspective. Collapse feature data and sound field feature data are extracted from the sound field data, and the collapse mechanism of the exploded bubble is characterized using the sound field modeling data. Subsequently, local features of bubble pulsation are extracted from the pressure data, and the global bubble collapse energy release value is calculated, thus achieving energy analysis of the exploded bubble from a pressure perspective. By fusing multi-source data from three dimensions—image, sound field, and pressure—using a Bayesian inversion algorithm, a more comprehensive and accurate three-dimensional energy density field is constructed. This enables comprehensive analysis of the actual pulsation behavior of exploded bubbles from multiple data sources, avoiding the problems of low analysis accuracy, incomplete data, and poor repeatability caused by single data sources. Subsequently, a bubble structure damage mapping model is constructed using a machine learning regression algorithm, establishing a generalizable mapping mechanism between exploded bubble behavior parameters and the degree of structural damage. This enables accurate assessment of submarine damage and improves the accuracy of submarine damage assessment based on the analysis of exploded bubble pulsation behavior.

[0070] In summary, this invention acquires image data, sound field data, and pressure data of the exploding bubble from different dimensions during a submarine damage experiment. Then, it generates 3D point cloud data of the exploding bubble using a subpixel-level edge fitting algorithm and a 3D point cloud algorithm, providing the spatial morphology of the exploding bubble from an image perspective. Collapse feature data and sound field feature data are extracted from the sound field data, and the collapse mechanism of the exploding bubble is characterized using the sound field modeling data. Finally, local features of bubble pulsation are extracted from the pressure data, and the global bubble collapse energy release value is calculated, thus achieving energy analysis of the exploding bubble from a pressure perspective. Subsequently, a Bayesian inversion algorithm was used to fuse multi-source data from three dimensions: image, sound field, and pressure, thereby constructing a more comprehensive and accurate three-dimensional energy density field. This enabled comprehensive analysis of the actual pulsation behavior of exploded bubbles from multiple data sources, avoiding the problems of low analysis accuracy, incomplete data, and poor repeatability caused by a single data source. Then, a bubble structure damage mapping model was constructed using a machine learning regression algorithm, establishing a generalizable mapping mechanism between the behavior parameters of exploded bubbles and the degree of structural damage. This enabled accurate assessment of submarine damage and improved the accuracy of submarine damage assessment based on the analysis of the pulsation behavior of exploded bubbles.

[0071] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A submarine damage assessment method based on multi-source bubble energy analysis, characterized in that, include: Acquire image data, acoustic field data, and pressure data of the explosion bubbles during the submarine damage experiment; Based on a preset subpixel-level edge fitting algorithm, image features are extracted from the image data to determine the coordinate data of the exploding bubble; The coordinate data of the exploded bubble are registered based on a preset 3D point cloud algorithm to generate 3D point cloud data of the exploded bubble. Based on the sound field data, the collapse feature data and sound field feature data of the exploding bubble are extracted, and based on the collapse feature data and sound field feature data, the sound field modeling data of the exploding bubble is constructed. Based on the pressure data, the local features of bubble pulsation of the exploded bubble are extracted, and the global bubble collapse energy release value of the exploded bubble is determined based on the local features of bubble pulsation. Based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploded bubble, the sound field modeling data of the exploded bubble, and the global bubble collapse energy release value, the three-dimensional energy field of the exploded bubble is reconstructed to determine the three-dimensional energy density field of the exploded bubble. Based on a preset machine learning regression algorithm and combined with pre-acquired historical data of submarine damage experiments, a bubble structure damage mapping model is constructed. The three-dimensional energy density field of the exploded bubble is input into the bubble structure damage mapping model to determine the submarine damage assessment result.

2. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 1, characterized in that, The image feature extraction based on the preset subpixel-level edge fitting algorithm is performed on the image data to determine the coordinate data of the exploding bubble; The coordinate data of the exploded bubble are registered based on a preset 3D point cloud algorithm to generate 3D point cloud data of the exploded bubble, including: The image data includes several binocular images; Based on a preset subpixel-level edge fitting algorithm, pixel-level coarse edge extraction and subpixel-level edge localization are performed on each of the stereo images to extract image features and obtain the bubble coordinate data of each stereo image. A joint objective function for bubble volume constraint and point cloud distance volume is constructed, and the binocular image is registered by combining a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to generate 3D point cloud data of the exploded bubble.

3. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 2, characterized in that, The algorithm based on a preset sub-pixel level edge fitting performs pixel-level coarse edge extraction and sub-pixel level edge localization on each of the stereo images to extract image features from each stereo image, obtaining the coordinate data of the exploding bubble in each stereo image, including: Each of the stereo images is input into a preset graphics processing model to determine the image region of each stereo image; Based on a preset edge detection operator, edge detection is performed on each of the image regions to determine the edge mask map of each of the binocular images; Contour extraction is performed on the edge mask image to determine the contour point set for each binocular image; Gaussian filtering is applied to each of the stereo images, and the gradient data of each pixel in each stereo image after Gaussian filtering is calculated. Based on the gradient data, the neighboring pixels of each pixel in the contour point set of each stereo image are determined, and a preset fitting function is used to fit each pixel in the contour point set of each stereo image and its neighboring pixels to determine the sub-pixel position offset of each pixel in the contour point set of each stereo image. Based on the sub-pixel position offset, displacement compensation is performed on each pixel in the contour point set of each stereo image to determine the bubble contour point set of each stereo image. Based on the bubble outline point set of each stereo image, determine the pixel coordinates of the exploding bubble, the center coordinates of the exploding bubble, and the radius of the exploding bubble for each stereo image; The pixel coordinates, center coordinates, and radius of the exploding bubble in each stereo image are used as the exploding bubble coordinate data for each stereo image.

4. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 3, characterized in that, The process of constructing a joint objective function for bubble volume constraint and point cloud distance volume, and combining it with a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to register the binocular image and generate 3D point cloud data of the exploded bubble includes: Obtain a bubble volume threshold and construct a bubble volume constraint based on the bubble volume threshold; Construct a point cloud distance sum of squares term and a point cloud volume difference squares term, and construct a joint objective function for point cloud distance and volume based on the point cloud distance sum of squares term and the point cloud volume difference squares term; The iteration objective is to minimize the joint objective function of point cloud distance and volume. Based on the preset 3D point cloud algorithm and the bubble volume constraint, the coordinate data of the exploded bubble in each stereo image is iterated to register the stereo images and generate 3D point cloud data of the exploded bubble.

5. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 1, characterized in that, The step of extracting collapse feature data and sound field feature data of the exploding bubble based on the sound field data, and constructing sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data, includes: The sound field data includes: sound pressure signal and sound pressure signal arrival time; Based on the difference in arrival time of the sound pressure signal, and combined with a preset sound source localization algorithm, the sound source localization data of the exploding bubble is determined. Based on the sound pressure signal, combined with the preset water density and water sound velocity, the sound intensity data of the exploding bubble is determined; Based on the sound intensity data and the sound source location data of the exploding bubble, the sound power data of the exploding bubble is determined. The sound pressure signal is subjected to a short-time Fourier transform to determine the sound field spectrum characteristics of the exploding bubble; Autocorrelation analysis was performed on the sound pressure signal to determine the pulsation period of the exploding bubble; Based on the sound source location data, sound intensity data, sound power data, pulsation period, and sound field spectrum characteristics of the exploding bubble, the sound field characteristic data of the exploding bubble are determined. Based on the sound pressure signal and the sound field spectrum characteristics, the collapse time of the exploding bubble is identified; Based on the sound source localization data and the image data, the collapse direction of the exploding bubble is identified; Based on the collapse time and collapse direction of the exploded bubble, the collapse characteristic data of the exploded bubble are determined; The collapse characteristic data and sound field characteristic data of the exploded bubble are used as the sound field modeling data of the exploded bubble.

6. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 1, characterized in that, The step of extracting local features of bubble pulsation from the exploded bubble based on the pressure data, and determining the global bubble collapse energy release value of the exploded bubble based on the local features of bubble pulsation, includes: The pressure data includes: the pressure signal at each pressure signal measurement point; Wavelet decomposition and Fourier spectrum analysis were performed on the pressure signal to determine the local characteristics of bubble pulsation at each pressure signal measurement point of the exploding bubble; Based on the local characteristics of the bubble pulsation, the time-domain integral and frequency-domain integral of the exploding bubble at each pressure signal measurement point are determined; Based on the time-domain integral and frequency-domain integral, the estimated collapse energy of the exploding bubble at each pressure signal measurement point is determined; The estimated collapse energy of the exploding bubble at each pressure signal measurement point is input into a preset sound propagation model to determine the bubble collapse energy release value at each pressure signal measurement point. The global bubble collapse energy release value of the exploded bubble is determined by weighted averaging of the bubble collapse energy release value at each pressure signal measurement point.

7. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 6, characterized in that, The method, based on a preset Bayesian inversion algorithm, combines the three-dimensional point cloud data of the exploded bubble, the sound field modeling data of the exploded bubble, and the global bubble collapse energy release value to reconstruct the three-dimensional energy field of the exploded bubble, determining the three-dimensional energy inversion data of the exploded bubble, including: Based on a preset Bayesian inversion algorithm, the three-dimensional space of the exploding bubble is discretized into several voxels; Based on the three-dimensional point cloud data of the exploded bubble, an image constraint for the exploded bubble is constructed; Based on the sound field modeling data of the exploding bubble, an acoustic constraint for the exploding bubble is constructed. Based on the global bubble collapse energy release value, an explosion bubble pressure constraint is constructed; Based on the preset gradient descent method and the preset Markov chain Monte Carlo method, combined with the exploding bubble image constraint, exploding bubble acoustic constraint and exploding bubble pressure constraint, the energy density of each voxel is iteratively solved until the preset maximum number of iterations is met, and the optimal energy density of each voxel is determined. The three-dimensional energy density field of the exploding bubble is determined based on the optimal energy density of each voxel.

8. The submarine damage assessment method based on multi-source bubble energy analysis as described in claim 1, characterized in that, The process involves constructing a bubble structure damage mapping model based on a preset machine learning regression algorithm and pre-acquired historical data of submarine damage experiments. The three-dimensional energy density field of the exploded bubble is then input into the bubble structure damage mapping model to determine the submarine damage assessment result, including: Based on the three-dimensional energy density field of the exploded bubble, a feature vector for submarine damage assessment is constructed; The initial machine learning model was obtained, and the submarine was divided into several functional areas. Based on the functional areas, the pre-acquired historical data of submarine damage experiments was divided to construct model training data. Based on a preset machine learning regression algorithm, the initial machine learning model is trained using model training data to construct a bubble structure damage mapping model. The submarine damage assessment feature vector is input into the bubble structure damage mapping model to determine the maximum stress result, maximum deformation result, and damage level of each functional area of ​​the submarine. The submarine damage assessment results are determined based on the maximum stress, maximum deformation, and damage level of each functional area.

9. A submarine damage assessment system based on multi-source bubble energy analysis, characterized in that, include: The system includes a multi-source data acquisition module for exploding bubbles, an image data analysis module, a sound field data analysis module, a pressure signal analysis module, a multi-source bubble energy analysis module, and a submarine damage assessment module. The multi-source data acquisition module for explosive bubbles is used to acquire image data, sound field data, and pressure data of explosive bubbles during submarine damage experiments. The image data analysis module is used to extract image features from the image data based on a preset subpixel-level edge fitting algorithm to determine the coordinate data of the exploding bubble. The coordinate data of the exploded bubble are registered based on a preset 3D point cloud algorithm to generate 3D point cloud data of the exploded bubble. The sound field data analysis module is used to extract the collapse feature data and sound field feature data of the exploding bubble based on the sound field data, and to construct the sound field modeling data of the exploding bubble based on the collapse feature data and sound field feature data. The pressure signal analysis module is used to extract the local features of bubble pulsation of the exploding bubble based on the pressure data, and to determine the global bubble collapse energy release value of the exploding bubble based on the local features of bubble pulsation. The multi-source bubble energy analysis module is used to reconstruct the three-dimensional energy field of the exploding bubble based on a preset Bayesian inversion algorithm, combined with the three-dimensional point cloud data of the exploding bubble, the sound field modeling data of the exploding bubble, and the global bubble collapse energy release value, and to determine the three-dimensional energy density field of the exploding bubble. The submarine damage assessment module is used to construct a bubble structure damage mapping model based on a preset machine learning regression algorithm and pre-acquired historical data of submarine damage experiments. The three-dimensional energy density field of the exploded bubble is input into the bubble structure damage mapping model to determine the submarine damage assessment result.

10. A submarine damage assessment system based on multi-source bubble energy analysis as described in claim 9, characterized in that, The image data analysis module includes: an image data analysis unit; In the image data analysis unit, the image data includes several binocular images; The image data analysis unit is used to perform pixel-level coarse edge extraction and sub-pixel-level edge localization on each of the stereo images based on a preset sub-pixel-level edge fitting algorithm, so as to extract image features from each of the stereo images and obtain the explosion bubble coordinate data of each of the stereo images. A joint objective function for bubble volume constraint and point cloud distance volume is constructed, and the binocular image is registered by combining a preset 3D point cloud algorithm and the coordinate data of the exploded bubble to generate 3D point cloud data of the exploded bubble.

Images