A method for dispersed phase identification and completion in multiphase flow based on SAM neural network

Through the multi-phase flow dispersed phase recognition and completion method based on SAM neural network, the limitations of the existing technology when processing complex multi-phase flow dispersed phase images are solved, the analysis accuracy and automation are improved, and the diversified conditions are adapted to achieve The precise identification and reconstruction of complex morphological bubbles is achieved, reducing computing resource consumption.

CN118365626BActive Publication Date: 2024-11-26INSTITUTE OF PROCESS ENGINEERING CHINESE ACADEMY OF SCIENCES +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410573382.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-10
Publication Date
2024-11-26
Estimated Expiration
2044-05-10

AI Technical Summary

Technical Problem

The prior art has limitations in dealing with complex multiphase flow dispersed phase images, especially when facing highly overlapping bubbles, edge blur and light uneven problems, a lot of manual intervention and parameter adjustment are required, and deep learning methods require data. and computing resource consumption have limitations, making it difficult to adapt to unseen data types and new tasks, especially when bubble shapes are irregular or affected by adjacent objects.

Method used

Using multiphase flow dispersed phase recognition and completion methods based on SAM neural network, including image preprocessing, identification and segmentation using SAM neural network, post-processing and output accurate bubble masks, and shape reconstruction through bubble reconstruction algorithm to reduce the Reliance on large-scale labeled data and computing resources to adapt to diverse experimental conditions and image types.

Benefits of technology

Improve image analysis accuracy and processing efficiency, achieve higher degree of automation, reduce the needs of manual intervention and parameter adjustment, can adapt to diverse experimental conditions and image types, and provide accurate identification and effective reconstruction of complex morphological bubbles. Reduces computing resource consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118365626B_ABST
    Figure CN118365626B_ABST
Patent Text Reader

Abstract

The present invention provides a method for identifying and completing the dispersed phase of a multiphase flow based on a SAM neural network, the method comprising: preprocessing an image to filter out noise and non-critical frequency components; using a SAM neural network to identify and segment the preprocessed image to obtain a segmentation mask; post-processing the segmentation mask to output an accurate bubble mask; and reconstructing the shape of the bubble using a bubble reconstruction algorithm. The method improves the accuracy of image analysis, improves processing efficiency and automation, and reduces dependence on large-scale labeled data and computing resources; and is adaptable to a variety of experimental conditions and image types, and can be used for accurate identification and effective reconstruction of bubbles with complex morphologies, and is suitable for measurement and analysis of multiphase flow systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of multiphase flow dispersed phase image processing, and relates to a multiphase flow dispersed phase identification and completion method based on a SAM neural network. Background Art

[0002] Image processing and analysis of the dispersed phase of multiphase flows are critical in industrial applications, especially in the petroleum and chemical industries. Typical applications in these industries include, but are not limited to, performance evaluation of gas-liquid reactors, research on fluid dynamics, and process monitoring and optimization. Traditional image processing methods, such as high-speed photography combined with edge detection, watershed algorithms, and threshold segmentation, although effective in certain applications, show obvious limitations when dealing with images in complex situations, such as highly overlapping bubbles, blurred edges, and uneven illumination. Traditional image processing methods often require a lot of manual intervention and parameter adjustments, which is impractical in a rapidly changing industrial environment.

[0003] In contrast, deep learning-based methods, especially convolutional neural networks (CNNs), have shown excellent performance in the field of image processing. These methods are able to learn and recognize complex image patterns and have better robustness to noise and illumination changes. However, they still have limitations in terms of data requirements, task-specific dependencies, and computing resource consumption.

[0004] Moreover, the above existing methods often require retraining or extensive adjustments when faced with unseen data types or new tasks; and they also have difficulties in reconstructing bubbles of complex shapes, especially when the bubbles are irregular in shape or affected by neighboring objects.

[0005] With the advancement of AI technology, especially the development of general artificial intelligence, the basic model has become the key to solving these problems. The basic model is a large artificial intelligence neural network pre-trained on a large-scale data set with strong generalization capabilities, which can achieve strong zero-sample generalization and adapt to a wide range of downstream tasks. As an advanced basic model, the Segment Anything Model (SAM) launched by Meta AI Research has shown excellent performance in image segmentation tasks and can effectively process images without any human intervention. Therefore, the SAM model has broad prospects in multiphase flow dispersed phase image processing. Summary of the invention

[0006] In order to solve the technical problems existing in the prior art, the present invention provides a multiphase flow dispersed phase identification and completion method based on the SAM neural network. The method improves the image analysis accuracy, improves the processing efficiency and the degree of automation, reduces the dependence on large-scale labeled data and computing resources; and is adaptable to a variety of experimental conditions and image types, and can be used for the accurate identification and effective reconstruction of bubbles with complex morphologies.

[0007] In order to achieve the above technical effects, the present invention adopts the following technical solutions:

[0008] The present invention provides a method for multiphase flow dispersed phase identification and completion based on SAM neural network, the method comprising the following steps:

[0009] Preprocess the image to filter out noise and non-critical frequency components;

[0010] Using a SAM neural network to identify and segment the preprocessed image to obtain a segmentation mask;

[0011] Post-processing the segmentation mask to output an accurate bubble mask;

[0012] The bubble shape is reconstructed using the bubble reconstruction algorithm.

[0013] As a preferred technical solution of the present invention, the image preprocessing is to convert the image from the spatial domain to the frequency domain, filter out noise and non-critical frequency components in the frequency domain, and convert the image from the frequency domain back to the spatial domain.

[0014] As a preferred technical solution of the present invention, fast Fourier transform is used to convert the image from the spatial domain to the frequency domain.

[0015] Preferably, the formula for converting the image from the spatial domain to the frequency domain using fast Fourier transform is:

[0016] I(f x ,f y ) = FFT(I(x,y))

[0017] Among them, I(x,y) is the original image, x,y is the pixel position on the image, and f x ,f y are the coordinates of the image in the frequency domain.

[0018] As a preferred technical solution of the present invention, a threshold T is set in the frequency domain coordinates to filter out noise and non-critical frequency components.

[0019] Preferably, the formula for filtering out noise and non-critical frequency components by setting the threshold T is:

[0020]

[0021] Among them, I′(f x ,f y ) is the image after filtering out noise and non-critical frequency components.

[0022] As a preferred technical solution of the present invention, an inverse Fourier transform is used to transform the image from the frequency domain back to the spatial domain.

[0023] As a preferred technical solution of the present invention, a method for identifying and segmenting the preprocessed image using a SAM neural network includes:

[0024] The SAM model receives the preprocessed image as input;

[0025] The SAM model uses an image encoder to extract features of the image;

[0026] The SAM model uses a prompt encoder to process the input image to identify and segment bubbles;

[0027] The SAM model converts the encoded image and hint data into a segmentation mask using a mask decoder.

[0028] As a preferred technical solution of the present invention, post-processing the segmentation mask includes:

[0029] Identify and exclude masks that are too large and have very low fill rates, representing background;

[0030] Remove masks that cause under-segmentation and masks that are nested inside other masks.

[0031] As a preferred technical solution of the present invention, the bubble compensation algorithm includes:

[0032] Extracting boundary contour points of the precise bubble mask;

[0033] Using global information, the relationship between the target and the surrounding is extracted, the shared boundary is identified and the boundary ownership is determined, and the identified bubble contour is mapped to the polar coordinate system;

[0034] Use cubic spline curve to fit the missing part and complete the overall outline of the bubble.

[0035] As a preferred technical solution of the present invention, an edge extraction algorithm is used to extract boundary contour points from the bubble mask.

[0036] As a preferred technical solution of the present invention, the identifying shared boundary is identifying the shared boundary between the blocking bubble and the blocked bubble;

[0037] Preferably, the shared boundary is determined by analyzing the mask intersection of the blocking bubble and the blocked bubble;

[0038] Preferably, the boundary contour points of the shared boundary are attributed to the occluding bubble, and the boundary contour points are removed from the mask of the occluded bubble.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects:

[0040] (1) Improved analysis accuracy: Through advanced image preprocessing and deep learning models, this method can more accurately identify and analyze complex bubble flow images. This improvement in accuracy, especially in the identification and quantification of bubble morphology, provides important technical support for multiphase flow system research and process monitoring in industries such as the petroleum and chemical industries.

[0041] (2) Improved processing efficiency and automation: Compared with traditional image processing methods, the present invention significantly reduces the need for manual intervention and parameter adjustment, and achieves a higher degree of automation and processing efficiency. This is particularly important in a dynamically changing multiphase flow system environment.

[0042] (3) Reduced reliance on large-scale labeled data and computing resources: Due to the powerful generalization ability of the SAM model, the present invention reduces the reliance on large amounts of labeled data and is more efficient than traditional deep learning methods, reducing the consumption of computing resources.

[0043] (4) Stronger generalization ability: The present invention can adapt to a variety of experimental conditions and image types, and can show good performance even for unseen data and new tasks, which is difficult to achieve with traditional methods.

[0044] (5) Effective reconstruction of bubbles with complex shapes: The BSR algorithm is particularly suitable for reconstructing bubbles that are obscured or have irregular shapes. It provides a more accurate and effective method for reconstructing bubble shapes, which is of great significance for a deeper understanding of bubble dynamics and optimizing the bubble control process. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic diagram of a process for identifying and completing a dispersed phase in a multiphase flow based on a SAM neural network provided in a specific embodiment of the present application;

[0046] Figure 2a The original image of the dispersed phase of the multiphase flow of Application Example 1 of this application;

[0047] Figure 2b This is the multiphase flow dispersed phase identification diagram of Application Example 1 of this application;

[0048] Figure 2c This is the multiphase flow dispersed phase completion diagram of Application Example 1 of this application;

[0049] Figure 3aThe original image of the dispersed phase of the multiphase flow of Application Example 2 of this application;

[0050] Figure 3b This is the multiphase flow dispersed phase identification diagram of Application Example 2 of this application;

[0051] Figure 3c This is the multiphase flow dispersed phase completion diagram of Application Example 2 of this application;

[0052] Figure 4a The original image of the dispersed phase of the multiphase flow of Application Example 3 of this application;

[0053] Figure 4b This is the multiphase flow dispersed phase identification diagram of Application Example 3 of this application;

[0054] Figure 4c This is the multiphase flow dispersed phase completion diagram of Application Example 3 of this application;

[0055] Figure 5a The original image of the dispersed phase of the multiphase flow of Application Example 4 of this application;

[0056] Figure 5b This is the multiphase flow dispersed phase identification diagram of Application Example 4 of this application;

[0057] Figure 5c This is the multiphase flow dispersed phase completion diagram of Application Example 4 of this application.

[0058] The present invention is further described in detail below. However, the following examples are only simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims. DETAILED DESCRIPTION

[0059] The technical solution of the present application is further described below through specific implementation methods.

[0060] The specific implementation of the present application provides a method for multiphase flow dispersed phase identification and completion based on SAM neural network, and its process is as follows: Figure 1 As shown, the method comprises the following steps:

[0061] Preprocess the image to filter out noise and non-critical frequency components;

[0062] Using a SAM neural network to identify and segment the preprocessed image to obtain a segmentation mask;

[0063] Post-processing the segmentation mask to output an accurate bubble mask;

[0064] The bubble shape is reconstructed using the bubble compensation algorithm.

[0065] The shapes of the bubbles in the segmented image are reconstructed using a bubble compensation algorithm.

[0066] In the present invention, the SAM neural network is combined with the frequency domain method to remove interference factors in the image, improve the clarity of the image, and use the powerful generalization ability of SAM to perform effective image segmentation, and further post-process through the segmentation mask to identify and exclude those masks that represent the background and are too large, as well as masks with extremely low filling rates, while removing those masks that cause under-segmentation and masks nested inside other masks, and finally output an accurate bubble mask. In addition, the present invention also introduces an innovative bubble completion algorithm, BSR (Bubble Shape Reconstruction), to improve the reconstruction accuracy of bubbles.

[0067] In the present invention, the SAM model used is an advanced image segmentation technology based on deep learning. Its core advantage is that it can adapt to various image types and segmentation tasks, even scenes that have not been encountered during training.

[0068] In a specific implementation of the present invention, before preprocessing the image, the image needs to be collected and prepared.

[0069] In a multiphase flow experimental environment, a high-resolution camera with a high dynamic range and optimized optical properties is used for image acquisition. In a specific embodiment of the present invention, a camera capable of capturing at least 1080p resolution is selected to ensure that the image quality can meet the requirements of subsequent processing.

[0070] In a specific embodiment of the present invention, the exposure time, ISO setting and focal length of the camera are adjusted according to the specific needs of the experiment, and the specific values ​​of the above parameters are not specifically limited here. As for the distance, in a fast-flowing scene, a shorter exposure time is used to reduce motion blur.

[0071] In one embodiment of the present invention, when collecting images, it is ensured that various possible operating conditions are covered, such as different ventilation rates, different types of liquid media (considering differences in viscosity and surface tension), and different rotation speeds. This ensures that sufficiently diverse data is obtained, so that the analysis results are more universal.

[0072] In a specific implementation of the present invention, the acquired image is preprocessed.

[0073] In a specific implementation of the present invention, the image is enhanced before being converted from the spatial domain to the frequency domain.

[0074] In one embodiment of the present invention, a fast Fourier transform (FFT) is applied to convert the captured image from the spatial domain to the frequency domain. This is achieved by representing the image as a combination of waveforms of different frequencies, where each waveform represents a different feature of the image. FFT can reveal the key frequency components in the image, which is crucial for subsequent filtering and feature extraction.

[0075] In a specific implementation of the present invention, the image is filtered in the frequency domain.

[0076] In one embodiment of the present invention, filtering includes removing image noise and non-critical frequency components to highlight important features of the image, such as the edges and shapes of bubbles, wherein low-frequency components are usually associated with the background and smooth areas of the image, while high-frequency components are associated with the details of the image.

[0077] In a specific implementation of the present invention, a suitable threshold T is set to filter out noise and non-critical frequency components.

[0078] In the present invention, image preprocessing can enhance key features in the image, thereby making the edge and shape of the bubble more prominent visually.

[0079] In a specific embodiment of the present invention, the image is converted from the frequency domain back to the spatial domain by inverse FFT, ensuring that the effects of image enhancement and filtering can be reflected in the final image: processed (x,y)=IFFT(I′(f x ,f y )).

[0080] In the present invention, the inverse FFT makes the processed image clearer visually and provides a more accurate information basis for image analysis and processing in subsequent steps.

[0081] In a specific embodiment of the present invention, the SAM model can recognize and segment images by learning a large amount of image data.

[0082] In a specific embodiment of the present invention, the SAM model processes the input image through its flexible prompt encoder to accurately identify bubbles in the image.

[0083] The SAM model has the ability to handle multiple types of prompts, such as points, boxes, texts, and fully automatic. In one embodiment of the present invention, the fully automatic segmentation mode is used to identify and segment the dispersed phase of the multiphase flow.

[0084] In a specific embodiment of the present invention, the SAM model mask decoder part is responsible for converting the encoded image and hint data into the final segmentation mask.

[0085] In a specific implementation of the present invention, the segmentation mask is further post-processed.

[0086] In a specific embodiment of the present invention, post-processing mainly includes: identifying and excluding masks that represent the background and are too large, and masks with extremely low filling rates, while removing masks that cause under-segmentation and masks nested inside other masks, and finally outputting an accurate bubble mask.

[0087] In the present invention, the SAM model, combined with pre- and post-processing steps, can accurately distinguish bubbles from other elements in an image and generate detailed bubble outlines. The key advantage is the model's adaptability and flexibility to different images and cues. Even in complex or challenging image conditions, bubbles can be effectively processed and identified.

[0088] In the present invention, the SAM neural network can effectively process and analyze the pre-processed image through the above steps, and with the post-processing step, the bubbles in the image can be accurately identified and segmented. This efficient and accurate image processing capability provides strong support for subsequent bubble behavior analysis and multiphase flow research.

[0089] In multiphase flow analysis, correctly identifying and reconstructing the shape of bubbles is crucial for accurately understanding the fluid dynamics. Although the above methods can effectively identify and segment bubbles in images, in some cases, due to occlusion or the complex morphology of bubbles, the generated masks may require further processing to restore the true shape of the bubbles. Therefore, this application introduces a bubble shape reconstruction (BSR) algorithm.

[0090] In practical applications, bubbles may be partially obscured or have irregular shapes due to the influence of fluid dynamics, which makes the bubble contours directly obtained from image segmentation incomplete or inaccurate. In a specific embodiment of the present invention, an edge extraction algorithm is used to extract the required boundary contour points from a specified mask.

[0091] In a specific embodiment of the present invention, the method for identifying the shared boundary between the occluding bubble and the occluded bubble is as follows: Assume that there are two bubbles B1 and B2, where B1 partially occludes B2, and define the shared boundary between the two as C 12 This step is achieved by analyzing the intersection of the masks where the two bubbles overlap. Mathematically, the intersection C 12 The calculation is as follows:

[0092]

[0093] in, and Represent the boundaries of B1 and B2 respectively.

[0094] In one embodiment of the present invention, C12 After that, the ownership of the boundary is determined. The shared boundary is considered to belong to the occluding bubble. Therefore, C 12 It is retained as part of the boundary of B1 and removed from the data set of B2.

[0095] In the present invention, the cubic spline interpolation method is a method of interpolating values ​​in each interval through a series of cubic polynomials. We use the cubic spline fitting method to continuously and smoothly reconstruct the bubble contour.

[0096] In a specific implementation of the present invention, a spline function is constructed to fit the contour of the occlusion interval. In order to ensure that the curve is smooth at the junction, under the constraint of continuous function value, the spline function of adjacent nodes also needs to satisfy the following constraint, that is, the first-order derivative and the second-order derivative at the endpoint are continuous:

[0097] For each inner node θ i = 1, 2, ..., N-1 i , we can get:

[0098]

[0099] At the same time, since the spline curve does not need to pass through all sample data points accurately, but allows a certain error to obtain a smoother curve, the following optimization objective function is used in the implementation:

[0100]

[0101] Among them, r i and θ i is the polar coordinate representation of the curve, r i Represents the radial distance from the pole to a point in polar coordinates, θ i Represents the angle between the line connecting the point to the pole and the positive direction of the x-axis; λ is a smoothing parameter that controls the smoothness of the fit; the second-order derivative value of the spline function S(θ) is included in the optimization function.

[0102] At this point, the contour reconstruction problem can be equivalent to the optimization problem of the optimization function. The solution that achieves the minimum value of the optimization function is the polar coordinate representation of the desired reconstructed contour. The reconstruction of the missing contour can be completed by converting the polar coordinate representation back to the Cartesian coordinate system.

[0103] In the present invention, the BSR algorithm is very important for the reconstruction and completion of the bubble shape, because the bubble is deformed by the fluid dynamics, thereby forming a bubble with a very complex shape. Therefore, the BSR algorithm is not limited to circular or elliptical bubbles, but is also applicable to more complex shapes, such as distorted or elongated bubbles.

[0104] In the present invention, the BSR algorithm has good adaptability to complex environments. In actual multiphase flow, there may be other bubbles, solid particles or fluid structures around the bubbles. The BSR algorithm can work effectively under these complex conditions and accurately reconstruct the shape of the bubbles. This ability is very important for analyzing and understanding the gas-liquid interaction in multiphase flow, because the shape and size of the bubble directly affect the mass transfer and kinetic behavior between the gas and the liquid.

[0105] In a specific embodiment of the present invention, the image data processed by the SAM model and reconstructed by the BSR algorithm is analyzed in detail, including quantifying key parameters such as bubble size, shape, and distribution, and analyzing how these parameters change with experimental conditions.

[0106] In one embodiment of the present invention, the accuracy and reliability of the method of the present invention are verified by comparing the experimental data with the model prediction results. The experiment is repeated under different operating conditions to evaluate the generalization ability and adaptability of the method to ensure that the proposed method can work effectively under a variety of flow conditions.

[0107] To better illustrate the present invention and facilitate understanding of the technical solution of the present invention, typical but non-limiting embodiments of the present invention are as follows:

[0108] Example

[0109] This embodiment provides a method for multiphase flow dispersed phase identification and completion based on SAM neural network, the method comprising the following steps:

[0110] (1) Image acquisition:

[0111] Use a 1080P resolution camera with high dynamic range and optimized optical properties for image acquisition, and adjust the camera exposure time, ISO setting and focal length according to the fluid properties;

[0112] (2) Image preprocessing:

[0113] ① Use fast Fourier transform to convert the image from the spatial domain to the frequency domain, the formula is:

[0114] I(f x ,f y ) = FFT(I(x,y))

[0115] Among them, I(x,y) is the original image, x,y is the pixel position on the image, and f x ,f y are the coordinates of the image in the frequency domain;

[0116] ② In the frequency domain coordinates, the threshold T is set to filter out noise and non-critical frequency components. The formula is:

[0117]

[0118] Among them, I′(f x ,f y ) is the image after filtering out noise and non-critical frequency components;

[0119] ③ Use inverse fast Fourier transform to convert the image from frequency domain back to spatial domain to ensure that the effects of image enhancement and filtering can be reflected in the final image: I processed (x,y)=IFFT(I′(f x ,f y )).

[0120] (3) Using the SAM neural network to identify and segment the preprocessed image:

[0121] The SAM model receives the preprocessed image as input;

[0122] The SAM model uses an image encoder to extract features of the image;

[0123] The SAM model uses a prompt encoder to process the input image to identify and segment bubbles;

[0124] The SAM model converts the encoded image and hint data into a segmentation mask using a mask decoder;

[0125] The segmentation mask is post-processed to output an accurate bubble mask, wherein the post-processing includes: identifying and excluding masks that are too large and represent backgrounds and masks with very low fill rates, and removing masks that cause under-segmentation and masks that are nested inside other masks.

[0126] (4) Reconstruct the shape of the bubble using the bubble reconstruction algorithm

[0127] Use edge extraction algorithm to extract boundary contour points of accurate bubble mask;

[0128] Using global information, the relationship between the target and the surrounding is extracted, the shared boundary is identified and the boundary ownership is determined, and the identified bubble contour is mapped to the polar coordinate system;

[0129] Assume there is an occluding bubble B1 and an occluded bubble B2, where B1 partially occludes B2, and the shared boundary between the two is defined as C 12 ; By analyzing the mask intersection of the two bubbles overlapping, the shared boundary is determined as C 12 , intersection C 12 The calculation is as follows:

[0130]

[0131] in, and represent the boundaries of B1 and B2 respectively;

[0132] Identify C 12 After that, the ownership of the boundary is determined; the shared boundary is considered to belong to the occluding bubble, so C 12 It is retained as part of the B1 boundary and removed from the B2 mask.

[0133] The bubble contour is reconstructed continuously and smoothly using cubic spline fitting;

[0134] Use the spline function to fit the contour of the occlusion interval. To ensure that the curve is smooth at the junction, under the constraint of continuous function value, the spline function of adjacent nodes must also satisfy the following constraints, that is, the first-order and second-order derivatives at the endpoints are continuous:

[0135] For each inner node θ i = 1, 2, ..., N-1 i , we can get:

[0136]

[0137] At the same time, since the spline curve does not need to pass through all sample data points accurately, but allows a certain error to obtain a smoother curve, the following optimization objective function is used in the implementation:

[0138]

[0139] At this point, the contour reconstruction problem can be equivalent to the optimization problem of the optimization function. The solution that achieves the minimum value of the optimization function is the polar coordinate representation of the desired reconstructed contour. The reconstruction of the missing contour can be completed by converting the polar coordinate representation back to the Cartesian coordinate system.

[0140] Application Example 1

[0141] This application example uses the multiphase flow dispersed phase identification and completion method based on the SAM neural network provided in the embodiment to identify and complete the multiphase flow dispersed phase;

[0142] Among them, the continuous phase of the multiphase flow is water phase, the dispersed phase is air, and the ventilation volume is 0.18L / min ventilation volume;

[0143] The original image is Figure 2a As shown, the recognition image is Figure 2b As shown, the completed image is Figure 2c shown.

[0144] Application Example 2

[0145] This application example uses the multiphase flow dispersed phase identification and completion method based on the SAM neural network provided in the embodiment to identify and complete the multiphase flow dispersed phase;

[0146] The dispersed phase of the multiphase flow is the same as that in Application Example 1, and the ventilation volume is 0.27 L / min ventilation volume;

[0147] The original image is Figure 3a As shown, the recognition image is Figure 3b As shown, the completed image is Figure 3c shown.

[0148] Application Example 3

[0149] This application example uses the multiphase flow dispersed phase identification and completion method based on the SAM neural network provided in the embodiment to identify and complete the multiphase flow dispersed phase;

[0150] Among them, the multiphase dispersed phase is the same as that in Application Example 1, and the ventilation volume is 0.63L / min ventilation volume;

[0151] The original image is Figure 4a As shown, the recognition image is Figure 4b As shown, the completed image is Figure 4c shown.

[0152] Application Example 4

[0153] This application example uses the multiphase flow dispersed phase identification and completion method based on the SAM neural network provided in the embodiment to identify and complete the multiphase flow dispersed phase;

[0154] The continuous phase of the multiphase flow is water, and the dispersed phase is air and solid particles.

[0155] The original image is Figure 5a As shown, the recognition image is Figure 5b As shown, the completed image is Figure 5c shown.

[0156] It can be seen from the recognition and completion images of Application Examples 1-4 that the multiphase flow dispersed phase recognition and completion method based on the SAM neural network of the present invention has good recognition and completion performance, performs well in different multiphase flow systems, adapts to diverse experimental conditions and image types, can be used for accurate recognition and effective reconstruction of bubbles with complex morphologies, and is suitable for the measurement and analysis of multiphase flow systems.

[0157] In order to solve the technical problems existing in the prior art, the present invention provides a multiphase flow dispersed phase identification and completion method based on the SAM neural network. The method improves the image analysis accuracy, improves the processing efficiency and the degree of automation, reduces the dependence on large-scale labeled data and computing resources; and is adaptable to a variety of experimental conditions and image types, and can be used for the effective reconstruction of bubbles with complex morphologies.

[0158] The applicant declares that the present invention illustrates the detailed structural features of the present invention through the above-mentioned embodiments, but the present invention is not limited to the above-mentioned detailed structural features, that is, it does not mean that the present invention must rely on the above-mentioned detailed structural features to be implemented. Those skilled in the art should understand that any improvement of the present invention, equivalent replacement of the components selected by the present invention, addition of auxiliary components, selection of specific methods, etc., all fall within the protection scope and disclosure scope of the present invention.

[0159] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0160] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0161] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for multiphase flow dispersed phase identification and completion based on SAM neural network, characterized in that: The method comprises the following steps: Preprocess the image to filter out noise and non-critical frequency components; Use the SAM neural network to identify and segment the preprocessed image to obtain a segmentation mask; Post-processing the segmentation mask to output an accurate bubble mask; The bubble shape is reconstructed using a bubble reconstruction algorithm; The image post-processing step comprises: Identify and exclude masks that are too large and have very low fill rates, representing background; Remove masks that cause under-segmentation and masks that are nested inside other masks; The bubble reconstruction algorithm includes: Extracting boundary contour points of the precise bubble mask; Using global information, the relationship between the target and the surrounding is extracted, the shared boundary is identified and the boundary ownership is determined, and the identified bubble contour is mapped to the polar coordinate system; Use cubic spline curve to fit the missing part and complete the overall outline of the bubble; Extracting boundary contour points from the bubble mask using an edge extraction algorithm; The identifying of the shared boundary is identifying the shared boundary between the blocking bubble and the blocked bubble; Determine a shared boundary by analyzing the mask intersection of the blocking bubble and the blocked bubble; Boundary contour points of the shared boundary are attributed to occluding bubbles, and the boundary contour points are removed from the mask of the occluded bubble.

2. The method according to claim 1, characterized in that The image preprocessing comprises: Converting the image from the spatial domain to the frequency domain; filtering out noise and non-critical frequency components in the frequency domain; The image is converted from the frequency domain back to the spatial domain.

3. The method according to claim 2, characterized in that converting the image from the spatial domain to the frequency domain using a fast Fourier transform; The formula for converting the image from the spatial domain to the frequency domain using fast Fourier transform is: ; in, is the original image, is the pixel position on the image, are the coordinates of the image in the frequency domain.

4. The method according to claim 3, characterized in that In the frequency domain coordinates, by setting the threshold To filter out noise and non-critical frequency components; By setting the threshold The formula to filter out noise and non-critical frequency components is: ; in, This is the image after filtering out noise and non-critical frequency components.

5. The method according to claim 2, characterized in that: The image is converted from the frequency domain back to the spatial domain using an inverse fast Fourier transform.

6. The method according to claim 1, characterized in that The method of using the SAM neural network to identify and segment the preprocessed image includes: The SAM model receives the preprocessed image as input; The SAM model uses an image encoder to extract features of the image; The SAM model uses a prompt encoder to process the input image to identify and segment bubbles; The SAM model converts the encoded image and hint data into a segmentation mask using a mask decoder.