Method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation

Abstract

This invention discloses a method for capturing the three-dimensional motion trajectory of methane bubbles during a deep-sea methane leakage simulation, comprising the following steps: S10: establishing a three-dimensional coordinate system during bubble motion; S20: conducting a methane leakage simulation experiment in a reactor to obtain the three-dimensional coordinates of the particle position at any moment during bubble motion; S30: calculating the rising rate of the bubble motion; S40: calculating the size of the bubble motion. By adopting the above settings, compared with existing methods for studying bubbles in laboratory simulations of deep-sea methane leakage experiments, this invention considers the influence of morphological changes caused by methane hydrate coating and bubble size changes caused by gas mass transfer within the methane bubble. This method is used to perform high-precision calculations of the rising trajectory, rate, and size of methane bubbles in a deep-sea methane leakage simulation device, in order to explore the influence mechanism of the three-dimensional motion trajectory and high-precision methane bubble size on methane mass transfer.

Description

Technical Field

[0001] This invention belongs to the field of marine science and marine engineering technology, and relates to a method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation. Background Technology

[0002] Natural gas hydrates, commonly known as combustible ice, are widely found in seafloor sediments along continental margins. Cold seeps are widely considered a common phenomenon in marine geology. The formation of submarine cold seeps is closely related to the decomposition and evolution of natural gas hydrates. Scientific research estimates that submarine cold seep systems leak between 400,000 and 12.2 million tons of methane into the upper ocean each year. During this leakage process, most of the methane is transported from the seafloor to the upper ocean in the form of gas bubbles, while some of the methane dissolves in the seawater. The methane leaking into the seawater can further dissolve into water bodies and even enter the atmosphere, posing a certain potential environmental impact.

[0003] It can be seen that the extreme environment of deep-sea methane leakage areas is often characterized by high pressure, low temperature and high corrosivity. At the same time, methane leakage in this area is characterized by intermittent leakage, large differences in leakage flux, large randomness of leakage time and large variation in methane concentration in the leakage area. As a result, in-situ methane leakage research is often accompanied by extremely high costs and great research difficulty.

[0004] Because deep-sea methane leakage has a certain impact on the seawater and atmospheric environment, the trajectory of deep-sea methane leakage is of significant research value. Existing technologies commonly used to study the three-dimensional trajectory of bubbles typically employ ship or oil extraction scenarios, and their analysis of bubble movement trajectories is relatively coarse, making them unsuitable for studying the trajectory of deep-sea methane leakage. Therefore, there is still room for improvement in how to perform high-precision calculations of the rising trajectory of methane bubbles in a methane leakage simulation device, under the influence of morphological changes caused by methane hydrate coating and bubble size changes caused by gas mass transfer within the methane bubbles, in order to explore the mechanism by which the three-dimensional trajectory of methane bubbles affects methane mass transfer. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the present invention aims to provide a method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation. This method can accurately calculate the rising trajectory, speed, and size of methane bubbles in a methane leakage simulation device under the influence of morphological changes caused by methane hydrate coating and changes in bubble size caused by gas mass transfer within the methane bubbles, in order to explore the mechanism by which the three-dimensional motion trajectory of methane bubbles affects methane mass transfer.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A method for capturing the three-dimensional motion trajectory of methane bubbles during a deep-sea methane leakage simulation includes the following steps:

[0008] S10: Establish a three-dimensional coordinate system during the bubble's motion;

[0009] S11: Set up two perpendicular sides of the reactor as perspective surfaces, set up camera units at positions corresponding to the two perspective surfaces outside the reactor, and make the camera end of the camera unit face the center of the reactor through the corresponding perspective surfaces. Connect the two camera units to the computer system and construct a three-dimensional coordinate system with X, Y and Z axes through the two perpendicularly set camera units.

[0010] S20: Obtain the three-dimensional coordinates of the bubble;

[0011] S21: A methane leakage simulation experiment was conducted in a reactor. A grid scale for measurement was set in the center of the reactor. The upward-moving bubble was recorded by two camera units. The position of the bubble in the constructed three-dimensional coordinate system was recorded by the two camera units in two planar coordinate systems respectively. The coordinates of the bubble particle position in the two planar coordinate systems recorded by the two camera units are (X, 0, Z) and (0, Y, Z) respectively. The two coordinates are integrated to obtain the three-dimensional coordinates of the particle position at any time during the bubble's movement.

[0012] S30: Calculate the rising rate of the bubble during its motion;

[0013] S31: Based on the three-dimensional coordinate system constructed in S10 and S20 and the method for obtaining the three-dimensional coordinate position of the bubble, when calculating the rising rate of the bubble, the images of the bubble at the corresponding frame number in the video recordings of the two camera units at time t1 and time t2 are obtained respectively. The three-dimensional coordinate positions of the bubble at time t1 and time t2 are obtained through the images and defined as (X1, Y1, Z1) and (X2, Y2, Z2) respectively. The difference between the three-dimensional coordinate positions of the bubble at time t1 and time t2 is divided by the corresponding time difference to obtain the rising rate of the bubble.

[0014] S40: Calculate the size of the bubble during its movement: Obtain the image of the bubble at the corresponding frame number in the video recording of the camera unit at the corresponding time point, obtain the outline of the bubble from the image, and obtain the size of the bubble outline from the grid ruler.

[0015] Furthermore, S10 also includes: S12: A grid scale is set in the center of the reactor; the camera unit captures images of the grid scale and acquires multiple images of the grid scale containing lens field of view distortion effects; the images containing field of view distortion effects are corrected and multiple corrected images of the grid scale are acquired; the multiple images containing lens field of view distortion effects and the corrected images are used to transfer learn the principle of the lens field of view correction function to construct a distortion parameter learning model; the grid scale images captured by the camera unit are input into the distortion parameter learning model to obtain and output the distortion parameters under the corresponding lens; the images are corrected using the distortion parameters and a relatively standard grid scale and a three-dimensional coordinate system are obtained.

[0016] Furthermore, prior to S20 and / or S30, the following is also included:

[0017] S13: Obtain video recording files of bubbles captured by two camera units, obtain the image corresponding to each frame in the video recording files, and perform bubble detail expansion and erosion processing on the bubble image in each frame to obtain the clear outline of the bubble and the position of the bubble particles.

[0018] S14: The bubble images processed by expansion and erosion in S13 are labeled with individual bubble recognition and bubble contour recognition to obtain a large number of labeled bubble images. Four-fifths of the bubble images are used to train an image detection model and a semantic segmentation model to obtain a bubble detection model. The remaining one-fifth of the bubble images are used to verify the trained bubble detection model. When the training and verification meet the corresponding accuracy, the final bubble detection model is mounted on and connected to the post-processing hardware device of the two camera units.

[0019] Furthermore, in S14, during the training of the bubble image for the image detection model and the semantic segmentation model, the transfer learning YOLOV8 model and the DeeplabV3+ model serve as the main structures of the image detection model and the semantic segmentation model, respectively.

[0020] Furthermore, prior to S31, the following method is also included: based on S20, the bubble video is recorded by two camera units, the video recording file is transferred into the bubble detection model, the bubble detection model is used for real-time detection, the bubble is identified and marked, and the marked bubble obtains the three-dimensional coordinate position of the bubble at the corresponding time node through the method of obtaining the bubble's three-dimensional coordinate position in S20.

[0021] Furthermore, in S40, the following method is employed:

[0022] S41: Obtain the distortion parameters of each lens image of the camera unit through the distortion parameter learning model of S12, set grid scales at multiple positions in the three-dimensional area of ​​the lens range of the camera unit in the reactor, and take a single image through the camera unit;

[0023] S42: Compare the length of the grid scale of the captured single image with the length of the grid scale after correction by the corresponding distortion parameters to obtain the distortion length. Fit the distortion parameters obtained in S41, the distortion length of the grid scale at different positions in the reactor, and the original scale length of the grid scale to a polynomial nonlinear relationship in the spatial coordinate system. Each time the relative distance between the camera unit and the perspective plane and / or the height of the camera unit is adjusted, steps S41 and S42 need to be restarted to perform a new polynomial nonlinear relationship fitting.

[0024] S43: After the deep-sea methane leakage simulation device is run, the position of the rising bubble in the three-dimensional spatial coordinate system is obtained. The height value of the bubble position is input into the polynomial nonlinear relationship in S42 and the fitting result is obtained to obtain the distortion parameter of the position of the bubble. The distortion parameter is used to obtain the ratio between the distorted bubble size and the actual size. Then, the actual size of the bubble outline is obtained through the ratio.

[0025] S44: Based on the bubble detection model trained in S14, the semantic segmentation model of the bubble detection model is used to fully present the outline of the bubble. The semantically segmented bubble outline is compared with the grid scale corrected by the corresponding distortion parameters to obtain the size of the bubble outline in the corresponding lens image. The distortion parameters of the bubble position calculated in S43 and the size of the bubble outline in the lens image are put into a polynomial nonlinear relationship fitting function to calculate the actual size of the bubble.

[0026] S45: Conduct a methane leakage simulation experiment in a reactor. The bubbles are close to spherical in shape. Calculate the volume, surface area, and buoyancy parameters of the bubbles based on the actual dimensions of the bubble profile obtained in S43 or S44.

[0027] The present invention has the following beneficial effects:

[0028] 1. This invention proposes a method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation. Compared with existing methods for studying bubbles in laboratory simulations of deep-sea methane leakage, this invention takes into account the effects of morphological changes caused by methane hydrate coating and bubble size changes caused by gas mass transfer within the methane bubble. This method is used to perform high-precision calculations of the rising trajectory, velocity, and bubble size of methane bubbles in the deep-sea methane leakage simulation device, in order to explore the mechanism by which the three-dimensional motion trajectory and high-precision methane bubble size affect methane mass transfer.

[0029] 2. The method proposed in this invention is based on a deep learning model. By utilizing existing deep-sea methane leakage simulation experimental data and a set of bubble images captured by cameras, it transfers and trains a correlation function model for calculating lens field-of-view distortion parameters to automatically identify captured bubble images, and a semantic segmentation model for classifying bubble contour markers. The addition of the deep learning model greatly improves the convenience, accuracy, and efficiency of calculating various parameters during bubble movement.

[0030] 3. The method proposed in this invention is based on a dual-camera unit device with the same horizontal shooting angle placed vertically. It introduces lens field-of-view distortion parameters through a deep learning model (i.e., distortion parameter learning model), and uses a grid ruler to correct and construct a three-dimensional spatial grid coordinate system that takes into account the lens field-of-view distortion conditions. This is used to accurately locate the actual spatial position of the bubble at any time, providing data support for calculating the bubble movement rate within any time period.

[0031] 4. In the method proposed in this invention, scales are set at multiple positions in front of the lens field of view inside the reactor. The lens field of view distortion parameters and the relative length parameters (i.e., distortion length) of the scales after shooting and comparison with the corrected grid scales are used to perform polynomial nonlinear fitting to construct the lens distortion loss function at any point inside the reactor. This further corrects the bubble size-related parameters. Combined with the bubble contour obtained by the semantic segmentation model under the lens, the actual contour size of the bubble at any position in the reactor space is calculated, providing a high-precision guarantee for further calculation of bubble-related size parameters. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the deep-sea methane leakage simulation device of the present invention.

[0033] Figure 2 This is a flowchart of the method for capturing the trajectory of methane bubbles during the deep-sea methane leakage simulation process of the present invention.

[0034] Figure 3 This is the bubble outline image after the deep learning model has been used to dilate and erode the bubble.

[0035] Figure 4 This is a network structure diagram of the deep learning image detection model used by the present invention for identifying bubble markers in images.

[0036] Figure 5 This is a network structure diagram of the deep learning semantic segmentation model used in this invention for calculating bubble contours in images.

[0037] Figure 6 This is a diagram showing the result of the semantic segmentation model of the present invention identifying and segmenting the bubble contour.

[0038] In the diagram: 1. Reactor; 2. Perspective view; 3. Camera unit; 4. Illumination unit. Detailed Implementation

[0039] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Terms such as “upper,” “inner,” “middle,” “left,” “right,” and “one” used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0040] This invention discloses a method for capturing the trajectory of methane bubbles during a deep-sea methane leakage simulation process, primarily employing a deep-sea methane leakage simulation device, such as... Figure 1 As shown, the device includes a reaction vessel 1, two camera units 3, and an illumination unit 4. The main structure of the reaction vessel 1 is cylindrical. Both sides of the reaction vessel 1 along the vertical direction are provided with a viewing surface 2, which is made of glass or transparent resin material, so that the bubbles inside the reaction vessel 1 can be observed from the outside through the viewing surface 2. The two camera units 3 are set outside the reaction vessel 1, respectively corresponding to and opposite to the two viewing surfaces 2. The two camera units are on the same horizontal plane, and the front and side shooting directions of the two camera units 3 are at a 90-degree angle. The illumination unit 4 is set outside the reaction vessel 1 and illuminates the reaction vessel 1 by emitting light to improve the shooting clarity of the camera units.

[0041] Based on the above description of the structure of the deep-sea methane leakage simulation device, the following further describes the methane bubble trajectory capture method of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0042] S10: Establish a three-dimensional coordinate system during the bubble's motion;

[0043] S11: Two perspective surfaces 2 are set on two sides perpendicular to the reactor 1. Camera units 3 are set on the outside of the reactor 1 at positions corresponding to the two perspective surfaces 2, and the camera ends of the camera units 3 are made to face the center of the reactor 1 through the corresponding perspective surfaces 2. The two camera units 3 are connected to the computer system. The lens view of one camera unit 3 establishes an X and Z axis plane coordinate system at the center position inside the reactor 1, and the lens view of the other camera unit 3 establishes a Y and Z axis plane coordinate system at the center position inside the reactor 1. Since the two coordinate systems have the same Z axis and are perpendicular to each other, a three-dimensional coordinate system with X, Y, and Z axes is constructed through the lens views of the two camera units 3. The origin of the constructed three-dimensional coordinate system coincides with the exact center position of the reactor 1.

[0044] S12: Because the image captured by camera unit 3 inside reactor 1 will have uneven unit size on the side of the coordinate system near the inner wall of reactor 1 due to lens field of view distortion, the distortion parameters of the corresponding image captured by camera unit 3 are obtained by setting S12 to correct the image. The specific method is as follows:

[0045] A grid scale is set in the center of the reactor 1. The camera unit 3 captures images of the grid scale and acquires multiple images of the grid scale that include lens field of view distortion effects. The images containing field of view distortion effects are corrected, and multiple corrected images of the grid scale are acquired. The lens field of view correction function principle is transferred to the multiple images containing lens field of view distortion effects and the corrected images to construct a distortion parameter learning model. The grid scale images captured by the camera unit 3 are input into the distortion parameter learning model to obtain and output the distortion parameters under the corresponding lens. The images are corrected using the distortion parameters, and a relatively standard grid scale and a three-dimensional coordinate system are obtained, wherein the origin of the three-dimensional coordinate system is kept consistent with the initial origin position.

[0046] In this embodiment, in order to facilitate the identification and positioning of the bubbles captured or recorded by the camera unit 3, so as to facilitate the subsequent positioning of the moving bubble's position and the real-time calculation of the moving bubble's rising rate, the following method is adopted:

[0047] S13: A methane leakage simulation experiment was conducted in reactor 1. The upward-moving bubbles were recorded by two camera units 3. The video recording files were preprocessed using a trained deep learning model to obtain the corresponding images for each frame in the video recording files. The bubble images in each frame were subjected to bubble detail expansion and erosion processing to obtain a clear outline of the bubbles and the positions of the bubble particles (e.g., Figure 3 (As shown).

[0048] S14: The bubble images that have undergone expansion and erosion processing in S13 are labeled with bubble individual recognition and bubble contour recognition to obtain a large number of labeled bubble images. In this embodiment, 3000 bubble images are labeled. Four-fifths of the bubble images are used to train an image detection model and a semantic segmentation model to obtain a bubble detection model. The remaining one-fifth of the bubble images are used to verify the trained bubble detection model. When the training and verification meet the corresponding accuracy, the final bubble detection model is mounted on and connected to the post-processing hardware of the two camera units 3.

[0049] In the training of image detection and semantic segmentation models using bubble images, transfer learning models YOLOv8 and DeeplabV3+ are used as the main structures for the image detection and semantic segmentation models, respectively (e.g., ...). Figure 4 and Figure 5 (As shown).

[0050] S20: Obtain the three-dimensional coordinate position of the bubble using the following method:

[0051] S21: A methane leakage simulation experiment is conducted in reactor 1. The upward-moving bubbles are recorded by two camera units 3. In this embodiment, the video shooting frame rate is 29.97 frames per second. Through S13 and S14, the position of the corresponding bubble in each frame of the video recording file can be obtained. Due to the presence of two camera units 3 set perpendicularly to each other, the position of the bubble in the constructed three-dimensional coordinate system is recorded by the two camera units 3 in two planar coordinate systems respectively. The coordinates of the bubble particle position in the two planar coordinate systems recorded by the two camera units 3 are (X, 0, Z) and (0, Y, Z) respectively. Integrating the two coordinates, the three-dimensional coordinates (X, Y, Z) of the particle position at any moment during the bubble's movement are obtained; thus, it is convenient to explore the three-dimensional motion trajectory of the bubble.

[0052] S30: Calculate the rising rate of the bubble during its motion using the following method:

[0053] S31: A methane leakage simulation experiment is conducted in reactor 1. Based on S20, two camera units 3 record video of bubbles. The video recording file is transferred to the bubble detection model in S14. After the bubble image in the video recording file is processed by expansion and corrosion, the bubble image in the video recording file is detected in real time using the image detection model of the bubble detection model. The bubble is identified and marked. The marked bubble is obtained by obtaining the three-dimensional coordinate position of the bubble at the corresponding time node through the method of obtaining the three-dimensional coordinate position of the bubble in S20.

[0054] Further, when calculating the rising rate of the bubble, the images of the bubble at the corresponding frame number in the video recordings of the two camera units 3 at time t1 and time t2 are obtained respectively. The three-dimensional coordinate positions of the bubble at time t1 and time t2 are obtained by the method described in S20, which are defined as (X1, Y1, Z1) and (X2, Y2, Z2) respectively. The difference between the three-dimensional coordinate positions of the bubble at time t1 and time t2 is divided by the corresponding time difference to obtain the rising rate of the bubble.

[0055] S40: Calculate the size of the bubble during its movement, i.e., acquire the image of the bubble at the corresponding frame number in the video recording of camera unit 3 at the corresponding time point, obtain the outline of the bubble from the image, and obtain the size of the bubble outline using a grid ruler; this process uses the following method:

[0056] S41: Obtain the distortion parameters of each lens image of camera unit 3 through the distortion parameter learning model of S12, set grid scales at multiple positions in the three-dimensional area of ​​the lens range of camera unit 3 in reactor 1, and take a single image through camera unit 3.

[0057] S42: Compare the length of the grid scale of the captured single image with the length of the grid scale after correction by the corresponding distortion parameters to obtain the distortion length. Then, perform polynomial nonlinear fitting of the spatial coordinate system using the distortion parameters obtained in S41, the distortion length of the grid scale at different positions within the reactor 1, and the original scale length of the grid scale. Specifically, after each adjustment of the relative distance between the camera unit 3 and the perspective plane 2 and / or the height of the camera unit 3, steps S41 and S42 need to be restarted to perform a new polynomial nonlinear fitting.

[0058] S43: After the deep-sea methane leakage simulation device is run, the three-dimensional coordinates of the rising bubble in the three-dimensional spatial coordinate system are obtained. The height value (i.e., Z coordinate value) of the bubble particle position is input into the polynomial nonlinear relationship in S42 and the fitting result is obtained to obtain the distortion parameter of the particle position. The ratio between the distorted bubble size and the actual size is obtained through the distortion parameter.

[0059] S44: Based on the bubble detection model trained in S14, the semantic segmentation model of the bubble detection model is used to fully present the outline of the bubble (e.g., Figure 6 As shown, the semantically segmented bubble outline is compared with the grid scale corrected by the corresponding distortion parameters to obtain the size of the bubble outline in the corresponding lens image. The distortion parameters of the bubble position calculated in S43 and the size of the bubble outline in the lens image are put into a polynomial nonlinear relationship fitting function to calculate the actual size of the bubble.

[0060] S45: Conduct a methane leakage simulation experiment in reactor 1. The bubbles are close to spherical in shape. Calculate the volume, surface area, and buoyancy parameters of the bubbles based on the actual dimensions of the bubble profile obtained in S43 or S44.

[0061] In summary, the present invention provides a method for capturing the trajectory of methane bubbles during a deep-sea methane leakage simulation. Considering the morphological changes caused by methane hydrate coating and the size changes caused by gas mass transfer within the methane bubbles, the method simulates deep-sea methane leakage in reactor 1, simultaneously simulating the rising trajectory, velocity, and size of methane bubbles for high-precision calculation. This aims to explore the mechanism by which the three-dimensional trajectory and precise size of methane bubbles affect methane mass transfer. Compared to existing technologies, the method of the present invention, through steps S10 to S40, solves problems such as the planarization of bubble rising process studies in simulated deep-sea methane leakage, the difficulty in accurately identifying complex bubble morphological changes, the difficulty in locating moving particles in bubbles, the cumbersome and inaccurate calculation process of motion parameters after acquiring bubble motion image data, and the difficulty in correcting lens distortion during bubble motion imaging.

[0062] In addition, the method of the present invention has the following main features:

[0063] 1) Deep learning models trained on big data, such as bubble detection models, image detection models, and semantic segmentation models, are used to identify bubble motion trajectories in real time and draw bubble outlines to provide high-precision data support for further calculation of bubble size;

[0064] 2) Based on two camera units 3 placed vertically at the same horizontal plane shooting angle, the lens field of view distortion parameters are introduced through the depth distortion parameter learning model. Using the grid ruler, a three-dimensional spatial grid coordinate system considering the lens field of view distortion is corrected and constructed to accurately locate the actual position of the bubble at any time.

[0065] 3) Set grid scales at multiple positions in front of the lens field of view inside reactor 1. Use the lens field of view distortion parameters and the relative length parameters of the grid scale after shooting and after correction to perform polynomial nonlinear fitting, construct the lens distortion loss function at any point inside the reactor, and further correct the parameters related to bubble size.

[0066] The embodiments of the present invention are not limited thereto. Based on the above description of the present invention, and using common technical knowledge and conventional means in the field, the present invention can be modified, replaced or combined in various other forms without departing from the basic technical idea of ​​the present invention, and all such modifications, replacements or combinations fall within the scope of protection of the present invention.

Claims

1. A method for capturing the three-dimensional motion trajectory of methane bubbles during a deep-sea methane leakage simulation process, characterized in that, Includes the following steps: S10: Establish a three-dimensional coordinate system during the bubble's motion; S11: Set up two perpendicular sides of the reactor as perspective surfaces, set up camera units at positions corresponding to the two perspective surfaces outside the reactor, and make the camera end of the camera unit face the center of the reactor through the corresponding perspective surfaces. Connect the two camera units to the computer system and construct a three-dimensional coordinate system with X, Y and Z axes through the two perpendicularly set camera units. S12: A grid scale is set in the center of the reactor. The camera unit captures images of the grid scale and acquires multiple images of the grid scale that include lens field of view distortion effects. The images containing lens field of view distortion effects are corrected, and multiple corrected images of the grid scale are acquired. The multiple images containing lens field of view distortion effects and the corrected images are used to transfer learn the principle of lens field of view correction function to construct a distortion parameter learning model. The grid scale images captured by the camera unit are input into the distortion parameter learning model to obtain and output the distortion parameters under the corresponding lens. The images are corrected by the distortion parameters, and a relatively standard grid scale and a three-dimensional coordinate system are obtained. S13: Obtain video recording files of bubbles captured by two camera units, obtain the image corresponding to each frame in the video recording files, and perform bubble detail expansion and erosion processing on the bubble image in each frame to obtain the clear outline of the bubble and the position of the bubble particles. S14: The bubble images that have undergone expansion and erosion processing in S13 are labeled with individual bubble recognition and bubble contour recognition to obtain a large number of labeled bubble images. 4 / 5 of the bubble images are used to train an image detection model and a semantic segmentation model to obtain a bubble detection model. The remaining 1 / 5 of the bubble images are used to verify the trained bubble detection model. When the training and verification meet the corresponding accuracy, the final bubble detection model is mounted on and connected to the post-processing hardware device of the two camera units. S20: Obtain the three-dimensional coordinates of the bubble; S21: A methane leakage simulation experiment was conducted in a reactor. A grid scale for measurement was set in the center of the reactor. The upward-moving bubble was recorded by two camera units. The position of the bubble in the constructed three-dimensional coordinate system was recorded by the two camera units in two planar coordinate systems respectively. The coordinates of the bubble particle position in the planar coordinate systems recorded by the two camera units are (X, 0, Z) and (0, Y, Z) respectively. The two coordinates are integrated to obtain the three-dimensional coordinates of the particle position at any time during the bubble's movement. S30: Calculate the rising rate of the bubble during its motion; S31: Based on the three-dimensional coordinate system constructed in S10 and S20 and the method for obtaining the three-dimensional coordinate position of the bubble, when calculating the rising rate of the bubble, the images of the bubble at the corresponding frame number in the video recordings of the two camera units at time t1 and time t2 are obtained respectively. The three-dimensional coordinate positions of the bubble at time t1 and time t2 are obtained through the images and defined as (X1, Y1, Z1) and (X2, Y2, Z2) respectively. The difference between the three-dimensional coordinate positions of the bubble at time t1 and time t2 is divided by the corresponding time difference to obtain the rising rate of the bubble. S40: Calculate the size of the bubble during its movement: Obtain the image of the bubble at the corresponding frame number in the video recording of the camera unit at the corresponding time point, obtain the outline of the bubble from the image, and obtain the size of the bubble outline from the grid ruler.

2. The method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation as described in claim 1, characterized in that, Before S31, the method further includes: recording bubble video through two camera units based on S20, transferring the video recording file into the bubble detection model, performing real-time detection through the bubble detection model, identifying and marking bubbles, and obtaining the three-dimensional coordinate position of the bubble at the corresponding time node through the method of obtaining the three-dimensional coordinate position of the bubble in S20.

3. The method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation as described in claim 1, characterized in that, In S40, the following method is used: S41: Obtain the distortion parameters of each lens image of the camera unit through the distortion parameter learning model of S12, set grid scales at multiple positions in the three-dimensional area of ​​the lens range of the camera unit in the reactor, and take a single image through the camera unit; S42: Compare the length of the grid scale of the captured single image with the length of the grid scale after correction by the corresponding distortion parameters to obtain the distortion length. Fit the distortion parameters obtained in S41, the distortion length of the grid scale at different positions in the reactor and the original scale length of the grid scale to a polynomial nonlinear relationship in the spatial coordinate system. S43: After the deep-sea methane leakage simulation device is run, the position of the rising bubble in the three-dimensional spatial coordinate system is obtained. The height value of the bubble position is input into the polynomial nonlinear relationship in S42 and the fitting result is obtained to obtain the distortion parameter of the position of the bubble. The ratio between the distorted bubble size and the actual size is obtained through the distortion parameter, and then the actual size of the bubble profile is obtained through the ratio.

4. The method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation as described in claim 3, characterized in that, The S40 also includes: S44: Based on the bubble detection model trained in S14, the semantic segmentation model of the bubble detection model is used to fully present the outline of the bubble. The semantically segmented bubble outline is compared with the grid scale corrected by the corresponding distortion parameters to obtain the size of the bubble outline in the corresponding lens image. The distortion parameters of the bubble position calculated in S43 and the size of the bubble outline in the lens image are put into a polynomial nonlinear relationship fitting function to calculate the actual size of the bubble.

5. The method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation as described in claim 4, characterized in that, The S40 also includes: S45: Conduct a methane leakage simulation experiment in a reactor. The bubbles are close to spherical in shape. Calculate the volume, surface area, and buoyancy parameters of the bubbles based on the actual dimensions of the bubble profile obtained in S43 or S44.

6. The method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation as described in any one of claims 3 to 5, characterized in that, In S40, after each adjustment of the relative distance between the camera unit and the perspective surface and / or the height of the camera unit, steps S41 and S42 need to be restarted to perform a new polynomial nonlinear relationship fitting.

7. The method for capturing the three-dimensional motion trajectory of methane bubbles during deep-sea methane leakage simulation as described in claim 1, characterized in that, In S14, during the training of the bubble image image detection model and the semantic segmentation model, the transfer learning YOLOV8 model and the DeeplabV3+ model serve as the main structures of the image detection model and the semantic segmentation model, respectively.

Images