Water-entry cavitation bubble gas-liquid interface reconstruction method based on binocular vision technology and artificial intelligence strategy

Abstract

The invention discloses a water entry cavitation bubble gas-liquid interface reconstruction method based on a binocular vision technology and an artificial intelligence strategy, and relates to the crossing field of computer vision and fluid mechanics. In order to solve the technical defects that in the prior art, the consistency of image processing precision and physical rules cannot be considered at the same time, and stable, accurate and physically reasonable three-dimensional reconstruction of a high-speed complex water-entry cavitation interface is difficult to realize, the technical scheme provided by the invention comprises the following steps: acquiring a water-entry cavitation image sequence; performing preprocessing to obtain a clear cavitation boundary image; based on the clear cavitation bubble boundary image, extracting a cavitation bubble gas-liquid interface contour in each frame of image; obtaining three-dimensional point cloud data of the corresponding frame; generating a high-quality point cloud for reconstruction; generating a cavitation bubble gas-liquid interface three-dimensional curved surface; and performing physical consistency correction on the three-dimensional curved surface to obtain a final reconstruction result. The method can be applied to gas-liquid interface three-dimensional reconstruction and multiphase flow dynamic monitoring work in the high-speed water entry process.

Description

Technical Field

[0001] It involves the intersection of computer vision and fluid mechanics, as well as the field of artificial intelligence technology, and specifically involves the reconstruction of the gas-liquid interface of water-entering cavitation bubbles. Background Art

[0002] Water entry cavitation refers to a gas-liquid two-phase mixing phenomenon that forms at the tail or around an object due to its interaction with the fluid when the object enters the water at a certain speed. It is widely present in marine engineering and defense fields such as aircraft landing on the water surface, ship wake propulsion, and underwater vehicle movement. In the research of fluid dynamics, cavitation dynamics, and hydrodynamic optimization design, the three-dimensional interface morphology, dynamic evolution characteristics, and fluid interaction behavior of water entry cavitation are the core focus. Especially in the environment of high-speed water entry and multiphase disturbance, accurately obtaining the gas-liquid interface morphology of the cavitation is of great engineering significance for evaluating fluid loads, optimizing cavitation control strategies, and improving the compressive strength of structures.

[0003] Existing technical methods for measuring the interface of water-entering cavitation bubbles mainly include laser-induced fluorescence (PLIF), particle image velocimetry (PIV), X-ray imaging, and the recently emerging computer vision reconstruction method.

[0004] Laser-induced fluorescence (PLIF) technology allows for two-dimensional identification of gas-liquid interfaces by labeling fluid interfaces with fluorescent dyes. However, this method relies on a uniform distribution of the tracer, and the fluorescence signal is easily attenuated in high-speed motion or strong scattering environments, resulting in insufficient imaging clarity and difficulty meeting the requirements for high-resolution three-dimensional reconstruction.

[0005] Although particle image velocimetry (PIV) technology can reconstruct the velocity vector information of the flow field, its ability to reconstruct the three-dimensional interface itself is limited. Especially in the cavitation region, due to problems such as sparse particles and insufficient scattering intensity, the interface is not fully captured.

[0006] X-ray imaging technology has penetrating capabilities and can be used to observe the internal structure of cavitation, but the equipment is expensive and the experimental conditions are harsh, making it difficult to widely deploy on conventional hydrodynamic experimental platforms.

[0007] With the development of computer vision technology, binocular vision systems have been increasingly used in fluid visualization measurement due to their simple structure, low cost, and non-contact measurement. This technology uses two cameras to capture images from different angles, calculates the image parallax, and then recovers the target's three-dimensional form. Promising results have been achieved in 3D reconstruction of static objects or objects with regular structures. Some research has also attempted to apply this technology to complex flow interface scenarios, such as cavitation entering water.

[0008] However, directly applying traditional binocular vision technology to the reconstruction of the gas-liquid interface of water-entering bubbles still faces the following challenges:

[0009] Insufficient image clarity and difficulty in edge extraction: The bubble interface is affected by factors such as water refraction, scattering, and surface reflection, which reduces image quality and makes it impossible for traditional edge detection algorithms (such as Canny and Sobel) to accurately extract the interface contour.

[0010] Poor stability of stereo matching algorithms: The morphology of the cavitation interface changes dramatically during high-speed evolution. Traditional image feature point matching algorithms (such as SIFT, SURF, ORB, etc.) are easily affected by occlusion, noise, and local deformation, and cannot stably obtain high-quality disparity information.

[0011] The reconstruction results have weak physical consistency: Traditional binocular reconstruction relies only on image information and lacks the constraints of fluid mechanics knowledge. The obtained three-dimensional results may not be consistent with physical reality, and there will be problems such as distortion and drift.

[0012] Therefore, the existing technology cannot simultaneously take into account the accuracy of image processing and the consistency of physical laws, and it is difficult to achieve stable, accurate, and physically reasonable three-dimensional reconstruction of the high-speed and complex water-entry cavitation interface. Summary of the Invention

[0013] To address the technical shortcomings of the prior art, which are that the prior art cannot simultaneously take into account the accuracy of image processing and the consistency of physical laws, and is difficult to achieve stable, accurate, and physically reasonable three-dimensional reconstruction of high-speed and complex water-entry cavitation interfaces, the present invention provides the following technical solutions:

[0014] A method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy, comprising:

[0015] The steps of acquiring a sequence of water-entering cavitation images synchronously captured by a binocular high-speed camera;

[0016] The step of preprocessing the image sequence to obtain a clear cavitation boundary image;

[0017] Based on the clear cavitation boundary image, extract the cavitation gas-liquid interface contour in each frame image;

[0018] The step of obtaining three-dimensional point cloud data of the corresponding frame according to the bubble gas-liquid interface contour and binocular parallax information in each frame of the image;

[0019] Performing denoising and sparse optimization processing on three-dimensional point cloud data to generate high-quality point cloud for reconstruction;

[0020] The step of generating a three-dimensional surface of the cavitation gas-liquid interface using a surface reconstruction algorithm based on the high-quality point cloud used for reconstruction;

[0021] The step of performing physical consistency correction on the three-dimensional surface to obtain the final reconstruction result.

[0022] Furthermore, a preferred embodiment is provided, wherein the preprocessing includes multi-scale filtering and image enhancement.

[0023] Furthermore, a preferred embodiment is provided, which calls a deep learning segmentation model to extract the contour of the cavitation gas-liquid interface in each frame of the image.

[0024] Furthermore, a preferred embodiment is provided, which uses a neural network model to perform stereo matching to obtain three-dimensional point cloud data of corresponding frames.

[0025] Furthermore, a preferred embodiment is provided, in which fluid mechanics equations are introduced into the neural network as physical constraints to perform physical consistency correction on the three-dimensional surface.

[0026] Based on the same inventive concept, the present invention also provides a device for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy, comprising:

[0027] A module for acquiring a sequence of water-entering cavitation images collected synchronously by a binocular high-speed camera;

[0028] A module for preprocessing image sequences to obtain clear cavitation boundary images;

[0029] A module that extracts the contour of the cavitation gas-liquid interface in each frame of the image based on a clear cavitation boundary image;

[0030] A module for obtaining the 3D point cloud data of the corresponding frame based on the bubble gas-liquid interface contour and binocular parallax information in each frame of image;

[0031] A module that performs denoising and sparse optimization on 3D point cloud data to generate high-quality point clouds for reconstruction;

[0032] A module that generates a three-dimensional surface of the cavitation gas-liquid interface using a surface reconstruction algorithm based on high-quality point clouds for reconstruction;

[0033] A module that performs physical consistency correction on 3D surfaces to obtain the final reconstruction result.

[0034] Based on the same inventive concept, the present invention also provides a binocular vision shooting experimental system for implementing the method described above, including:

[0035] Experimental water tank, used to hold liquid;

[0036] Experimental pellets designed to create cavitation bubbles upon entry into water;

[0037] Experimental light source, used to provide high-brightness lighting;

[0038] The experimental calibration plate is placed inside the water tank and is used for camera calibration and 3D reconstruction coordinate system establishment;

[0039] Two high-speed cameras are arranged at two different angles outside the water tank, forming a binocular vision shooting structure, which is used to synchronously capture images of the entry process.

[0040] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program. When the computer program is read by a computer, the computer executes the method described above.

[0041] Based on the same inventive concept, the present invention also provides a computer, comprising a processor and a storage medium. When the processor reads the computer program stored in the storage medium, the computer executes the method described above.

[0042] Based on the same inventive concept, the present invention also provides a computer program product, which is a computer program. When the computer program is executed, the method described above is implemented.

[0043] Compared with the prior art, the technical solution provided by the present invention is beneficial in that:

[0044] This solution, by building an experimental system consisting of a high-speed camera, a calibration plate, and a high-brightness LED light source, combined with a hydrophobic-coated sphere device, effectively enhances the imaging contrast of the cavitation interface during water entry and reduces the blurring of the interface under high-speed water entry. Compared with traditional experimental systems using natural light or unstructured light sources, this solution significantly improves image clarity and the recognition of cavitation boundaries, providing a higher-quality raw data foundation for subsequent image processing and 3D reconstruction.

[0045] This solution uses a binocular vision system for image acquisition, combined with multi-scale filtering and image enhancement preprocessing, to effectively suppress underwater noise and strong reflection interference, improving the representation of cavitation boundary layer structures in the image. Compared to traditional processing methods using single-scale edge detection operators, this preprocessing strategy improves adaptability to complex multi-scale structures and enhances the contour extraction of cavitation interfaces.

[0046] This solution introduces the Segment Anything Model (SAM) for intelligent cavitation contour recognition, addressing the low accuracy of traditional edge extraction algorithms under optical distortion. The SAM model uses deep learning to segment complex gas-liquid interface morphologies, enabling more stable two-dimensional boundary extraction. Compared to Canny or grayscale thresholding methods, this method offers superior boundary accuracy and occlusion robustness, making it particularly suitable for scenarios where the interface undergoes drastic changes during water entry.

[0047] This solution incorporates the RAFT-Stereo neural network model, based on optical flow constraints, into the stereo matching process. This enables refined disparity estimation in high-gradient and irregular interface regions, improving the accuracy of point cloud reconstruction for complex cavitation surfaces. Compared to traditional SIFT or ORB feature point matching, RAFT-Stereo's time-series perception effectively reduces mismatches and depth jumps, improving point cloud density and accuracy.

[0048] After acquiring the initial point cloud, this solution removes outliers through RANSAC and reconstructs and optimizes the data using a combination of Gaussian filtering and voxel grid methods, effectively improving the smoothness and stability of the 3D reconstructed model. Compared to conventional Delaunay or Ball Pivoting reconstruction algorithms, this method is more suitable for point clouds with complex boundary variations and non-uniform density, enhancing the ability to preserve interface details.

[0049] This solution introduces physical information neural networks (PINNs), using the Navier-Stokes equations from fluid dynamics as physical constraints for neural network training, thereby enhancing the physical plausibility of the reconstruction results. Unlike 3D reconstruction models trained solely on image data, PINNs effectively avoid "fictitious" interfaces, resulting in reconstructions that more closely resemble actual fluid behavior, meeting the demands of high-precision engineering simulations and emulations.

[0050] This solution incorporates a Kalman filter algorithm into the dynamic evolution processing phase, enabling state prediction and assimilation updates of time series data, enhancing the ability to track cavitation evolution trends in real time. Compared to traditional frame-by-frame reconstruction, this method dynamically fuses consecutive frames, reducing mutation errors and improving the 3D model's response speed and accuracy to interface changes during high-speed water entry.

[0051] It can be applied to the three-dimensional reconstruction of the gas-liquid interface and the dynamic monitoring of multiphase flow during high-speed water entry. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of the method for reconstructing the gas-liquid interface of water-entering cavitation;

[0053] Figure 2 Schematic diagram of the binocular vision shooting experimental system;

[0054] Figure 3 The effect diagram of obtaining the contour information of the water-entering cavitation interface.

[0055] Among them, 1 represents the experimental water tank; 2 represents the experimental light source; 3 represents the experimental ball; 4 represents the calibration plate; 5 represents the high-speed camera; and 6 represents the computer. DETAILED DESCRIPTION

[0056] In order to make the advantages and benefits of the technical solution provided by the present invention more clearly reflected, the technical solution provided by the present invention is now further described in detail with reference to the accompanying drawings, specifically:

[0057] Implementation 1: This implementation provides a method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy, including:

[0058] The step of acquiring a sequence of water-entering cavitation images synchronously captured by a binocular high-speed camera 5;

[0059] The step of preprocessing the image sequence to obtain a clear cavitation boundary image;

[0060] Based on the clear cavitation boundary image, extract the cavitation gas-liquid interface contour in each frame image;

[0061] The step of obtaining three-dimensional point cloud data of the corresponding frame according to the bubble gas-liquid interface contour and binocular parallax information in each frame of the image;

[0062] Performing denoising and sparse optimization processing on three-dimensional point cloud data to generate high-quality point cloud for reconstruction;

[0063] The step of generating a three-dimensional surface of the cavitation gas-liquid interface using a surface reconstruction algorithm based on the high-quality point cloud used for reconstruction;

[0064] The step of performing physical consistency correction on the three-dimensional surface to obtain the final reconstruction result.

[0065] in:

[0066] The preprocessing includes multi-scale filtering and image enhancement.

[0067] The deep learning segmentation model is called to extract the contour of the cavitation gas-liquid interface in each frame image.

[0068] The neural network model is used for stereo matching to obtain the three-dimensional point cloud data of the corresponding frame.

[0069] Fluid mechanics equations are introduced into the neural network as physical constraints to perform physical consistency correction on the three-dimensional surface.

[0070] Implementation Method 2: This implementation method further explains the technical solution provided in Implementation Method 1. Specifically:

[0071] like Figure 1 As shown in FIG, a method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy specifically includes the following steps:

[0072] Step 1: Experimental system construction and data collection preparation

[0073] Construct an experimental platform and configure imaging and optical conditions to obtain image data of the water-entry cavitation formation process.

[0074] Specifically, an experimental platform was constructed, including an experimental water tank 1, a high-brightness LED light source, a hydrophobic-coated ball entering the water, a calibration plate 4 placed within the water tank, a binocular high-speed camera 5 system, and a data acquisition computer 6. The high-brightness LED light source provided stable and uniform illumination, enhancing the contrast of cavitation contour imaging. The hydrophobic material was sprayed on the ball surface to reduce resistance in the water and induce stable cavitation formation. The calibration plate 4 was used to calibrate the binocular camera system and establish a three-dimensional coordinate transformation model. The high-speed camera 5 simultaneously recorded the ball's entry into the water and the entire cavitation evolution process, generating a high-resolution time-series image sequence that served as input for subsequent image processing.

[0075] Step 2: Image preprocessing and interface contour feature extraction

[0076] The image sequence is preprocessed and the cavitation boundary is extracted to obtain high-quality two-dimensional interface data.

[0077] Specifically, image enhancement and multi-scale filtering are performed on the acquired binocular image sequences to suppress background noise, enhance boundary information, and improve contrast in dark or bright areas. Subsequently, the Segment Anything Model (SAM) is used to perform deep learning-based intelligent segmentation on each image frame, automatically identifying and extracting the cavitation gas-liquid interface contour. The output interface contour information serves as input for the next step of stereo matching and 3D point cloud reconstruction.

[0078] Step 3: Acquisition of 3D point cloud based on stereo matching

[0079] Disparity estimation is performed through a neural network model to reconstruct the preliminary three-dimensional point cloud of the cavitation interface.

[0080] Specifically, based on the boundary information in the image pair, an optical flow constraint model is applied, assuming that the cavitation interface maintains constant brightness across consecutive frames. The RAFT-Stereo model is then used to estimate the disparity field. As a recurrent neural network architecture, RAFT-Stereo has the ability to understand temporal context, enabling stable and accurate stereo matching even under conditions of drastic changes in the cavitation interface. Ultimately, a dense 3D point cloud of the cavitation interface is output for each frame, serving as input for the next step of surface reconstruction.

[0081] Step 4: 3D point cloud denoising and reconstruction optimization

[0082] De-noising and spatial structure optimization are performed on the preliminary point cloud to improve reconstruction accuracy.

[0083] Specifically, the RANSAC algorithm is used to remove outliers from the point cloud data. Mean and Gaussian filtering are then used to further suppress local noise and smooth the point cloud structure. The denoised point cloud is then fed into the Voxel Grid module, which normalizes the three-dimensional space to reduce data volume while preserving geometric detail. The optimized point cloud data serves as the basis for surface reconstruction and is then fed into the surface fitting module.

[0084] Step 5: Construction of cavitation interface three-dimensional surface

[0085] The physical consistency algorithm is used to perform three-dimensional surface fitting on the point cloud data to obtain the high-precision cavitation interface morphology.

[0086] Specifically, based on the optimized point cloud, the Poisson surface reconstruction method is used to solve the point cloud implicit function based on the Poisson equation to construct a continuous and smooth three-dimensional gas-liquid interface. The Poisson reconstruction method is suitable for processing dense and uneven point clouds. It can effectively maintain the geometric continuity of the cavitation interface in complex shapes such as boundaries and sharp corners, and output a high-quality surface model that provides input for dynamic evolution analysis.

[0087] Step 6: High-fidelity optimization of the 3D interface and correction of physical constraints

[0088] Fluid physics constraints are introduced to correct the reconstruction results and enhance physical consistency.

[0089] Specifically, a physical-informed neural network (PINN) is used to optimize the Poisson reconstruction results. The Navier-Stokes equations are used as physical constraints in the network training loss, and the objective function is constructed together with the image reconstruction error. PINNs incorporate prior knowledge of fluid mechanics into the 3D modeling process, effectively correcting for deformations that violate physical laws in traditional methods, thereby improving the stability, credibility, and scientific nature of the reconstructed model.

[0090] Step 7: Dynamic Evolution Modeling and Time Series Assimilation

[0091] Model and predict the temporal changes of the three-dimensional structure of the cavitation bubble to enhance the system's ability to describe interface evolution.

[0092] Specifically, a recursive Kalman filter algorithm is introduced to fuse and estimate the state of time series image data. By weightedly fusing the reconstruction result at the current moment with the estimation result at the previous moment, the influence of external noise and algorithm errors is reduced, the accuracy of predicting the dynamic evolution trend of the gas-liquid interface is improved, and high-fidelity modeling of the complex cavitation formation, evolution, and dissipation process is achieved.

[0093] In terms of the binocular vision shooting experimental system, the aim is to obtain dynamic image data of the cavitation gas-liquid interface generated by the ball entering the water at high speed, providing raw data support for subsequent 3D reconstruction and image recognition. The structure of the system is as follows Figure 2 As shown, it includes the following main components:

[0094] The experimental water tank 1 is a closed, transparent structure that holds a certain amount of experimental water and serves as a carrier space for the cavitation generated by the water-entering balls. The transparent material of the water tank helps reduce optical distortion when the camera captures images.

[0095] Experimental ball 3 is a controlled-release falling object, coated with a hydrophobic coating to reduce liquid resistance during entry and induce stable cavitation. The ball freely falls from above the water surface, simulating a high-speed entry into the water.

[0096] Experimental light source 2 is a high-brightness LED array light source, which is installed on the side or top of the water tank to provide high-uniformity and high-color temperature background lighting, enhance the imaging contrast of the gas-liquid interface, and ensure that the cavitation boundary is clearly discernible in the image.

[0097] The calibration plate 4 is set inside the water tank and has known feature points and patterns. It is used to perform geometric correction and coordinate transformation of the binocular camera system to ensure that the image acquisition data can be accurately restored to three-dimensional points in the physical space.

[0098] The system consists of two high-speed cameras 5, mounted on either side of the water tank, forming a binocular vision system that simultaneously captures image sequences from different angles during the cavitation process. The cameras are equipped with high-speed shutters and high frame rate imaging capabilities to adapt to the high-speed evolution of cavitation.

[0099] Computer 6 is used for image acquisition control, data storage, and subsequent processing. It is connected to the binocular camera via a high-speed interface (such as USB 3.0 or Gigabit Ethernet) to achieve synchronous triggering and real-time image stream transmission. After acquisition, it performs image preprocessing, contour extraction, and data backup.

[0100] The experimental ball 3 is placed in a release mechanism above the water tank and can fall freely into the water tank, generating cavitation during its entry into the water.

[0101] The experimental light source 2 is set on the side or top of the water tank. On the basis of ensuring the same-side illumination of the high-speed camera, the supplementary light source optimizes the imaging quality of the cavitation contour and improves the contrast of the interface feature points.

[0102] The binocular high-speed camera 5 is fixed on both sides or the front of the water tank, facing the calibration plate 4 set in the middle of the water tank and the water entry area at a certain angle, forming a stable parallax baseline and realizing three-dimensional visual reconstruction capability.

[0103] The calibration plate 4 is fixed at the bottom of the water tank or near the shooting area. Its characteristic pattern is clearly visible in the camera field of view, supporting system calibration and establishing a mapping relationship between image coordinates and physical coordinates.

[0104] The cameras perform frame-level time matching through synchronization trigger lines or software synchronization mechanisms to ensure the consistency of captured images and avoid reconstruction errors caused by time differences.

[0105] All collected images are transmitted to the computer 6 via a data line for storage and serve as input data for subsequent image processing and neural network reconstruction.

[0106] Implementation Method 3: Combination Figure 3 , the water bubble interface profile information acquisition effect diagram, illustrates this embodiment, this embodiment further describes the above-mentioned technical solution in detail through specific examples, specifically:

[0107] 1. Experimental system construction

[0108] The experimental system includes: an experimental water tank 1, an experimental light source 2, an experimental sphere 3, an experimental calibration plate 4 (inside the water tank), a high-speed camera 5, and a computer. A high-brightness LED light source is used to enhance interface contrast. A hydrophobic coating is sprayed on the sphere surface to reduce frictional resistance during water entry, facilitating the formation of a stable, streamlined cavitation bubble. High-speed camera 5 captures high-resolution images of the cavitation bubble's formation and evolution during water entry.

[0109] 2. Image preprocessing and feature extraction

[0110] Experimental images were preprocessed using binocular camera calibration technology. Multi-scale filtering was further employed to suppress image noise and highlight features at different scales. Image enhancement was used to improve the optical distinction between different flow structures by correcting overly dark or overly bright areas in the image. The Segment Anything Model (SAM) was introduced to intelligently segment flow structures and obtain temporal information on the interface contours of the water-entry cavitation bubble.

[0111] 3. Acquisition of 3D gas-liquid interface point cloud data

[0112] A stereo matching method based on optical flow constraints is adopted. It is assumed that the brightness of a certain point in the image does not change with time. The motion in the two frames of images is described by the optical flow equation. Furthermore, combined with the RAFT-Stereo model based on the recurrent neural network (RNN) architecture, the disparity estimation is gradually refined, the learning and reconstruction effects of complex interface features are enhanced, and the preliminary acquisition and processing of gas-liquid interface point cloud data are achieved.

[0113] 4. High-precision construction of three-dimensional gas-liquid interface

[0114] A sparse point cloud optimization algorithm is used, combined with mean and Gaussian filtering for noise suppression. RANSAC (Random Sampling Consensus Algorithm) is used to remove outliers and extract the expected target points for refined processing. Furthermore, a voxel grid is used to partition the spatial region, and a Poisson surface reconstruction method is used to recover the continuous water-entry cavitation interface from the discrete point cloud data by solving the Poisson equation, completing the high-precision three-dimensional construction of the gas-liquid interface.

[0115] 5. High-fidelity reconstruction of three-dimensional gas-liquid interface

[0116] The Physical Information Neural Network (PINNS) method was introduced, incorporating the fluid dynamics equations (Navier-Stokes equations) as physical constraints. The residual term of the physical equations was incorporated into the neural network's loss function, which includes both data loss and physical constraints to ensure that the reconstruction results conform to the laws of fluid physics. Time series data assimilation was performed, employing a recursive Kalman filter algorithm to improve the ability to predict interface dynamic evolution, supporting research on issues such as water entry cavitation.

[0117] The invention has the following advantages

[0118] 1. High precision: Combining binocular vision technology with artificial intelligence strategies to improve the spatial resolution of complex gas-liquid interfaces.

[0119] 2. Strong robustness: To address optical distortion and noise issues, the matching algorithm is optimized to resolve boundary layer structural distortion.

[0120] 3. Good physical consistency: Fluid mechanics equation constraints are introduced to ensure that the reconstruction results conform to actual physical laws.

[0121] 4. Strong applicability in complex environments: Suitable for reconstruction of flow fields in complex environments such as high-speed water entry processes and multiphase flows.

[0122] The above further describes the technical solution provided by the present invention in detail through several specific embodiments in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific embodiments described above are not intended to limit the present invention. Any reasonable modification and improvement of the present invention, combination of embodiments and equivalent replacement based on the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy, characterized in that: include: The steps of acquiring a sequence of water-entering cavitation images synchronously captured by a binocular high-speed camera; The step of preprocessing the image sequence to obtain a clear cavitation boundary image; Based on the clear cavitation boundary image, extract the cavitation gas-liquid interface contour in each frame image; The step of obtaining three-dimensional point cloud data of the corresponding frame according to the bubble gas-liquid interface contour and binocular parallax information in each frame of the image; Performing denoising and sparse optimization processing on three-dimensional point cloud data to generate high-quality point cloud for reconstruction; The step of generating a three-dimensional surface of the cavitation gas-liquid interface using a surface reconstruction algorithm based on the high-quality point cloud used for reconstruction; The step of performing physical consistency correction on the three-dimensional surface to obtain the final reconstruction result.

2. The method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy according to claim 1 is characterized in that: The preprocessing includes multi-scale filtering and image enhancement.

3. The method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy according to claim 1 is characterized in that: The deep learning segmentation model is called to extract the contour of the cavitation gas-liquid interface in each frame image.

4. The method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy according to claim 1 is characterized in that: The neural network model is used for stereo matching to obtain the three-dimensional point cloud data of the corresponding frame.

5. The method for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy according to claim 1 is characterized in that: Fluid mechanics equations are introduced into the neural network as physical constraints to perform physical consistency correction on the three-dimensional surface.

6. A device for reconstructing the gas-liquid interface of a water-entering cavitation bubble based on binocular vision technology and artificial intelligence strategy, characterized in that: include: A module for acquiring a sequence of water-entering cavitation images collected synchronously by a binocular high-speed camera; A module for preprocessing image sequences to obtain clear cavitation boundary images; A module that extracts the contour of the cavitation gas-liquid interface in each frame of the image based on a clear cavitation boundary image; A module for obtaining the 3D point cloud data of the corresponding frame based on the bubble gas-liquid interface contour and binocular parallax information in each frame of image; A module that performs denoising and sparse optimization on 3D point cloud data to generate high-quality point clouds for reconstruction; A module that generates a three-dimensional surface of the cavitation gas-liquid interface using a surface reconstruction algorithm based on high-quality point clouds for reconstruction; A module that performs physical consistency correction on 3D surfaces to obtain the final reconstruction result.

7. A binocular vision shooting experimental system, characterized in that: The method for implementing claim 1 comprises: Experimental water tank, used to hold liquid; Experimental pellets designed to create cavitation bubbles upon entry into water; Experimental light source, used to provide high-brightness lighting; The calibration plate is placed inside the water tank and is used for camera calibration and establishing the 3D reconstruction coordinate system; Two high-speed cameras are arranged at two different angles outside the water tank, forming a binocular vision shooting structure, which is used to synchronously capture images of the entry process.

8. A computer storage medium for storing a computer program, characterized in that When the computer program is read by a computer, the computer executes the method according to claim 1 .

9. A computer comprising a processor and a storage medium, characterized in that When the processor reads the computer program stored in the storage medium, the computer executes the method according to claim 1 .

10. A computer program product, being a computer program, characterized in that When the computer program is executed, the method according to claim 1 is implemented.

Images