High flow rate cloud cavitation bubble recognition method based on adaptive multi-threshold segmentation

Abstract

The application discloses a high flow rate cloud cavitation bubble recognition method based on adaptive multi-threshold segmentation, and belongs to the technical field of cavitation bubble image recognition and fluid measurement. The method first acquires evolution images of dense bubble groups in a high flow rate cloud cavitation flow field by using a high-speed camera; based on image gray distribution characteristics, an OTSU adaptive global threshold method is used to determine a background noise threshold and a scattered light noise threshold; on this basis, an adaptive multi-threshold segmentation algorithm is constructed, multi-round recognition is carried out in combination with a gray threshold sequence and a bubble diameter threshold sequence, and through connected domain analysis and area compensation, the separation of adhered bubbles and the accurate extraction of small-scale bubbles are realized. The application can adapt to high flow rate, non-uniform illumination and non-uniform bubble distribution environment, and realize high-precision, non-invasive recognition and statistical analysis of internal micro bubbles of cloud cavitation.

Description

Technical Field

[0001] This invention belongs to the field of image processing and fluid measurement technology. Specifically, it relates to a method for identifying cavitation bubbles in high-velocity clouds based on adaptive multi-threshold segmentation. Background Technology

[0002] Cavitation refers to the phase change phenomenon in which a liquid vaporizes and forms bubbles when the local static pressure drops below the liquid's saturated vapor pressure at that temperature during flow. In engineering practice, cavitation is widely present in fluid machinery such as ship propellers, pumps, turbomachinery, hydrofoils, and underwater propulsion devices. The generation and development of cavitation are often accompanied by strong pressure pulsations, noise radiation, and material surface erosion. In severe cases, it can lead to structural damage or even equipment failure, making it one of the key issues affecting the performance and reliability of fluid machinery.

[0003] Cloud cavitation specifically refers to a cloud-like cavitation structure composed of numerous tiny bubbles, exhibiting periodic shedding characteristics. Cloud cavitation is characterized by high flow velocities, unsteady behavior, and highly uneven bubble density distribution. The vapor phase structure formed by its shedding consists of numerous tiny bubbles, and the dynamic behavior of the macroscopic structure cannot fully reflect the evolutionary characteristics of the internal bubble swarms. Therefore, accurate identification and quantitative analysis of bubbles within cloud cavitation are crucial for revealing its formation and collapse mechanisms. However, under high flow velocities and strong scattered light backgrounds, bubbles easily adhere to each other, resulting in significant differences in local grayscale distribution. Traditional image segmentation methods struggle to achieve stable identification, leaving a technological gap for the accurate identification of bubble swarms within cloud cavitation.

[0004] Patent publication number CN106226205A discloses a transient bubble plume observation device and method, which uses sapphire optical fiber to measure the bubble diameter. This method is an invasive measurement method; the insertion of the probe into the flow field will disturb the original flow structure, and it is easily damaged by impact in a high-speed cavitation environment. It is also costly, makes it difficult to achieve overall identification of large-scale bubble swarms, and therefore cannot perform time-based analysis of microbubble characteristics.

[0005] Patent publication number CN110427817A discloses a method for extracting cavitation features from hydrofoils based on cavitation image localization and acoustic texture analysis. This method primarily acquires macroscopic dynamic features such as cavitation shedding and convection. While it can capture the overall changes in the cavitation structure, its analysis mainly focuses on macroscopic vapor phase structures. For the complex structure within cloud cavitation, composed of numerous tiny bubbles, this method is insufficient to meet the needs for precise identification and statistical analysis of microscopic bubbles within cloud cavitation.

[0006] Patent publication number CN114463653B discloses a method for identifying the morphology and tracking the trajectory of high-concentration microbubbles. This method is suitable for identifying high-concentration microbubbles under uniform illumination conditions. However, cloud cavitation flows are prone to scattering noise and bubble adhesion due to their significant non-uniform bubble density distribution. Therefore, image segmentation methods based on single thresholds or fixed parameters struggle to adapt to local grayscale variations, easily leading to missegmentation, misidentification, and the inability to separate adhered bubbles. Consequently, the adaptability of this method in high-velocity cloud cavitation environments remains insufficient.

[0007] High-speed imaging technology offers advantages such as non-invasiveness and a wide measurement range. Developing a bubble identification method based on a high-speed camera, capable of adapting to complex lighting conditions and non-uniform bubble density distributions while possessing strong noise resistance, remains feasible for accurate identification and quantitative analysis of bubble swarms within high-velocity cloud cavitation. Furthermore, this method enables time-series analysis of the dynamic behavior of microbubbles in cloud cavitation, potentially providing technical support for the simultaneous measurement of bubble and flow fields. Summary of the Invention

[0008] In view of the problems described above, this invention aims to develop a method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation. This invention acquires the cloud cavitation bubble field using high-speed imaging technology and obtains the statistical distribution of bubble diameters and temporal variations of bubble parameters through a bubble identification algorithm.

[0009] The high-velocity cloud cavitation bubble identification method based on adaptive multi-threshold segmentation provided by this invention includes the following steps:

[0010] 1) Acquire high-speed images of dense bubble clusters in the cloud cavitation flow field to be identified using a high-speed camera;

[0011] 2) Based on the grayscale distribution characteristics of the high-speed image, the adaptive global thresholding algorithm OTSU is used to determine the background noise threshold and the scattered light noise threshold respectively;

[0012] 3) Based on the background noise threshold, the scattered light noise threshold, and the image storage bit depth, construct a grayscale threshold sequence and set a bubble diameter threshold sequence;

[0013] 4) Based on the grayscale threshold sequence and the bubble diameter threshold sequence, perform multiple rounds of recognition on the high-speed image; in each round of recognition, divide each connected component into large targets and small targets through connected component analysis, obtain the bubble diameter and centroid of the current round of recognition based on the small targets, and obtain the Mask region for the next round of recognition based on all large targets;

[0014] Starting from the third round of recognition, the diameter of each connected region obtained in this round is compensated first, and then the connected regions are divided and the bubble diameter and centroid are output.

[0015] When the number of recognition rounds reaches the preset number of recognitions or the Mask area is zero, the bubble diameter sequence and centroid sequence obtained from the recognition are output.

[0016] 5) Based on the calibration results of the high-speed camera, the actual bubble diameter is calculated to complete the identification of high-velocity cloud cavitation bubbles.

[0017] Compared with the prior art, the beneficial effects of the present invention include:

[0018] Based on the grayscale distribution characteristics of instantaneous high-speed images, this invention adaptively determines instantaneous background noise and instantaneous scattered light noise, effectively overcoming the problem of uneven image light intensity distribution caused by non-uniform bubble distribution, and laying the foundation for achieving accurate bubble segmentation and area compensation.

[0019] This invention can adaptively set the input grayscale threshold and bubble diameter threshold for multi-stage recognition, improving the recognition accuracy of image sequences and reducing human intervention. Through adaptive adjustment of multiple thresholds, it achieves effective separation of adhering bubbles in dense bubble regions, improving segmentation accuracy under non-uniform bubble distribution conditions.

[0020] This invention enables the acquisition of bubble-scale distribution characteristics of bubble swarms within cloud cavitation, extending the study of cloud cavitation dynamics from the macroscopic to the microscopic scale. Combined with high-speed image acquisition technology, it achieves time-resolved analysis of bubble field characteristic parameters, providing a data foundation for research on the coupling mechanism between bubble fields and flow fields. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the experimental setup for generating cavitation of twisted hydrofoil clouds and acquiring high-speed images, as used in an example of the present invention.

[0022] Figure 2 This is a basic flowchart of the method of the present invention;

[0023] Figure 3 This is a specific identification diagram showing that the number of identification attempts is set to 4 in this invention;

[0024] Figure 4 This is a snapshot of four recognition results in an example of the present invention.

[0025] Figure 5 The results are presented as examples of the present invention. Detailed Implementation

[0026] The technical method of the present invention is described in detail here through specific examples. It should be noted that the examples described below are intended to facilitate the understanding of the present invention, and the method is not limited to processing cloud cavitation images involved in this example.

[0027] This embodiment takes the cavitation of a twisted hydrofoil as an example. The cloud cavitation generation device and the overall experimental layout are as follows: Figure 1 As shown, the system includes a cavitation water tunnel 1-1, a NACA 16012 twisted hydrofoil 1-2, a fiber optic probe 1-3, a CMOS high-speed camera 1-4, a continuous laser 1-5, an optical mirror group 1-6, and a laser sheet beam 1-7. The continuous laser 1-5, after optical path adjustment via the optical mirror group 1-6, forms a 1.5 mm thick laser sheet beam 1-7. The cavitation number is known. ,in The static pressure at the center of the cavitation water tunnel, It is the saturated vapor pressure. For the density of water, This is the incoming flow velocity. This example uses a fixed incoming flow velocity. =7 m / s, by adjusting the static pressure at the center of the water tunnel The cavitation flow field is acquired. The high-speed image capture window is set near the fiber optic probe, whose position can be adjusted. After stable cloud cavitation bubbles are generated, the cloud cavitation bubble field is recorded using high-speed images, providing raw bubble attribute information for subsequent processing.

[0028] like Figure 2 The diagram shown is a basic flowchart illustrating the implementation of the method of the present invention in a specific embodiment. In this embodiment, the steps of the method of the present invention are further broken down into the following specific steps for description:

[0029] 1) Based on Figure 1 The experimental setup shown generates a cloud cavitation flow field, and a high-speed image of the original dense bubble group is captured by a CMOS high-speed camera at a shooting frequency of 2kHz.

[0030] 2) To facilitate subsequent identification, this embodiment preprocesses the original high-speed image of dense bubble clusters, including linearly stretching the grayscale values ​​of the image to enhance image contrast, and adding regions of "probes" and "distorted hydrofoil walls", setting the pixels of these regions to 0 to eliminate the interference of these regions on target detection.

[0031] 3) Based on the grayscale distribution characteristics of the high-speed image, the Adaptive Global Thresholding Algorithm (OTSU) is used to determine the background noise threshold and the scattered light noise threshold, respectively; according to the background noise threshold, the scattered light noise threshold, and the image storage bits, a grayscale threshold sequence is constructed, and a bubble diameter threshold sequence is set; and the desired number of recognition attempts is set. ;

[0032] 4) Based on the grayscale threshold sequence and the bubble diameter threshold sequence, perform multiple rounds of recognition on the high-speed image; in each round of recognition, divide each connected component into large targets and small targets through connected component analysis, obtain the bubble diameter and centroid of the current round of recognition based on the small targets, and obtain the Mask region for the next round of recognition based on all large targets;

[0033] Starting from the third round of recognition, the diameter of each connected region obtained in this round is compensated first, and then the connected regions are divided and the bubble diameter and centroid are output.

[0034] When the number of recognition rounds reaches the preset number of recognitions or the Mask area is zero, the bubble diameter sequence and centroid sequence obtained from the recognition are output.

[0035] 5) Based on high-speed camera calibration parameters Calculate the actual bubble diameter .

[0036] In this embodiment, the identification results of the present invention are further compared with the results of the fiber optic probe.

[0037] The following provides a detailed explanation of some of the steps.

[0038] For the grayscale threshold sequence in step 3), the OTSU method achieves optimal segmentation by finding the threshold that maximizes the inter-class variance. This invention first performs an initial OTSU threshold calculation on the global input image to obtain the background noise threshold. This is used to identify free single bubbles. Then, a second OTSU thresholding calculation is performed on the foreground portion of the image to obtain the scattered light noise threshold. .

[0039] Subsequent grayscale thresholds are calculated in ascending order of an arithmetic sequence:

[0040]

[0041] in 65535 represents the maximum grayscale value of a 16-bit stored image.

[0042] The bubble diameter threshold in the bubble diameter threshold sequence is used to determine whether each connected component obtained in the current recognition round is a large or small target. As the recognition round increases, the bubble diameter threshold for the corresponding recognition round gradually increases.

[0043] The bubble diameter threshold sequence is preset manually, and the set diameter threshold does not affect the final recognition result. In this embodiment, the minimum diameter threshold is determined first. and maximum diameter threshold To reduce computational costs, the bubble diameter threshold for each round of identification is then calculated using an arithmetic sequence formula:

[0044] .

[0045] like Figure 2 As shown, in the first round of recognition in step 4), a background noise threshold is used. Pixel values ​​below a threshold are set to 0, and pixel values ​​above the threshold are set to 1, performing binarization to distinguish foreground from background. The foreground region identified in this round is the bubble region (containing scattered light noise). Connectivity analysis is performed on the foreground region to obtain the equivalent diameter and centroid coordinates of each connected component. The equivalent diameter of each connected component is then compared with the diameter threshold input in this round. For comparison, the equivalent diameter is greater than The connected components are labeled as "large targets" in this round, smaller than... The connected components are labeled as "small targets" for this round. The purpose of distinguishing between large and small targets is that if a target is set as a "large target" in the current round, it means that this "large target" may contain multiple small bubbles, but the grayscale threshold of this round cannot segment them, so they are not recognized in this round and are input into the next round of recognition. "Small targets" are used for the bubble diameter / centroid recognition output of this round. The bubble recognition result of the first round is a free small-diameter single bubble. All the "large targets" labeled in this round together form the Mask1 region of the output of this round, which is used for recognition in the next round.

[0046] In the second round of identification, the effective elements within the Mask1 region are first extracted, and a scattered light noise threshold is applied to this region. Pixel values ​​below a threshold are set to 0, and pixel values ​​above the threshold are set to 1, performing binarization to further distinguish the foreground from the background in this round. The foreground region identified in this round is the bubble region (excluding scattered light noise). Connectivity analysis is performed on the foreground region to obtain the equivalent diameter and centroid coordinates of each connected component. The equivalent diameter of each connected component is then compared with the diameter threshold input in this round. For comparison, the equivalent diameter is greater than The connected components are labeled as the "big goal" for this round, smaller than... The connected components are labeled as the "small targets" of this round. The "small targets" are used for the bubble diameter / centimeter recognition output of this round, and the bubble recognition result of the second round is the single bubble in the scattered light noise interference area. All the "large targets" of this round together form the Mask2 region of the output of this round, which is used for the next round of recognition.

[0047] The first two identifications were affected by both background noise and scattered light noise, and the results of the first two identifications hardly involved any adhered air bubbles. Therefore, diameter compensation was not enabled for the first two identifications.

[0048] From the third round (or the wheel, The recognition process begins, and each round of recognition includes:

[0049] S1: Obtain the total input area and input region for this round of recognition. The input region and total input area for the third round of recognition are the Mask region and its total area for the second round of recognition, respectively. The total input area is dynamically updated in each subsequent round based on the recognition results of the previous round.

[0050] S2: Based on the grayscale threshold identified in this round, binarize the effective information of the grayscale image within the input area for this round, and perform connected component analysis (the binarization and connected component analysis operations are the same as in the previous two rounds of identification; binarization sets pixel values ​​below the threshold to 0 and pixel values ​​above the threshold to 1 to distinguish between foreground and background in this round), and calculate the area ratio of each connected component relative to all connected components. ;

[0051] S3: Calculate the equivalent area of ​​each connected component as Based on this, the equivalent diameter of the circle for each connected component is calculated. ,in, This represents the total input area for this round of identification;

[0052] S4: Divide large and small targets according to the equivalent diameter of the circle: Connected regions with an equivalent diameter greater than the bubble diameter threshold of this round are retained (i.e., large targets) and used as the input region (mask region) for the next round of recognition; Connected regions with an equivalent diameter less than the bubble diameter threshold of this round are used as small targets (i.e., output bubbles of this round), and the equivalent diameter and centroid of each small target are output as the bubble diameter and centroid of this round of recognition;

[0053] S5: Calculate the percentage of the total area of ​​all smaller objectives relative to the total area of ​​all connected components in this round. The total input area for the next round of recognition is updated as follows:

[0054]

[0055] S6: Repeat S1 to S5 until the set recognition round is reached, or =0.

[0056] Figure 3 The diagram illustrates the algorithm implementation process of setting four recognition rounds using the above embodiment. The diagram shows the Mask input and output after setting the threshold for each round. The grayscale threshold is set based on the first round of recognition. and diameter threshold A red bubble is identified, and Mask1 is output. A second identification within the Mask1 area is performed based on the grayscale threshold set for the second round of identification. and diameter threshold The blue bubble is identified, and Mask2 is output simultaneously; the third identification occurs within the Mask2 area, and the grayscale threshold set in the third identification is applied. and diameter threshold Yellow bubbles (with area compensation) were identified, and Mask3 was output. Finally, within the Mask3 area, the grayscale threshold set in the fourth round of recognition was applied. and diameter threshold The cyan bubbles were identified (area compensation has been performed), and the identification result was output as follows: Figure 4 As shown. Figure 4 The recognition results of local areas were magnified, and the results showed that the recognition process with 4 recognition rounds could effectively identify bubbles in high-speed images and effectively segment adhering bubbles. Bubbles identified in different recognition rounds were represented by different colors, including red bubbles in the first recognition round, blue bubbles in the second recognition round, yellow bubbles in the third recognition round, and cyan bubbles in the fourth recognition round.

[0057] The identification results obtained by the method of this invention are compared with the results of the fiber optic probe. The accuracy of the algorithm is verified by fitting the bubble size with a log-normal distribution function, thereby providing a statistical description. Figure 5 The cavitation number is shown The distribution of bubbles obtained by the two measurement methods at a distance of 5 mm from the wall was almost identical, indicating that the method of the present invention achieves effective separation and accurate segmentation of bubbles adhering to each other in dense bubble areas through adaptive adjustment of multiple thresholds.

[0058] The embodiments described above merely illustrate the development process of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for identifying cavitation bubbles in high-velocity clouds based on adaptive multi-threshold segmentation, characterized in that, The steps include the following: 1) Acquire high-speed images of dense bubble clusters in the cloud cavitation flow field to be identified using a high-speed camera; 2) Based on the grayscale distribution characteristics of the high-speed image, the adaptive global thresholding algorithm OTSU is used to determine the background noise threshold and the scattered light noise threshold respectively; 3) Based on the background noise threshold, the scattered light noise threshold, and the image storage bit depth, construct a grayscale threshold sequence and set a bubble diameter threshold sequence; 4) Based on the grayscale threshold sequence and the bubble diameter threshold sequence, perform multiple rounds of recognition on the high-speed image; in each round of recognition, divide each connected component into large targets and small targets through connected component analysis, obtain the bubble diameter and centroid of the current round of recognition based on the small targets, and obtain the Mask region for the next round of recognition based on all large targets; Starting from the third round of recognition, the diameter of each connected region obtained in this round is compensated first, and then the connected regions are divided and the bubble diameter and centroid are output. When the number of recognition rounds reaches the preset number of recognitions or the Mask area is zero, the bubble diameter sequence and centroid sequence obtained from the recognition are output. 5) Based on the calibration results of the high-speed camera, the actual bubble diameter is calculated to complete the identification of high-velocity cloud cavitation bubbles.

2. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1, characterized in that, Step 1) specifically refers to: A laser sheet is emitted using a continuous laser and an optical mirror assembly, and a high-speed image of the cloud cavitation flow field to be identified is acquired at high frequency using a CMOS high-speed camera.

3. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1, characterized in that, In step 2), the Adaptive Global Thresholding Algorithm (OTSU) determines the image threshold by maximizing the inter-class variance. In this process, the first OTSU is performed on the high-speed image to obtain the grayscale threshold. As a background noise threshold, it is used to identify free single bubbles; Then, by performing a second OTSU on the foreground portion of the high-speed image, the grayscale threshold is obtained. As a threshold for scattered light noise.

4. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1 or 3, characterized in that, O The process of TSU obtaining the grayscale threshold is as follows: Obtain the total average grayscale value of a high-speed image; Based on each possible grayscale threshold The image is divided into two parts, where the grayscale value is less than or equal to... The portion that is the foreground has a grayscale value greater than [missing information]. The first part serves as the background; the class mean of the two parts is calculated, and the inter-class variance is further obtained. ; OTSU's goal is to maximize the inter-class variance. The grayscale threshold is determined by maximizing the inter-class variance.

5. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1, characterized in that, The first term in the grayscale threshold sequence is the background noise threshold. The second item is the scattered light noise threshold. Starting with the third item, the grayscale threshold... Calculate according to the increasing arithmetic sequence: ； in 65535 represents the maximum grayscale value of a 16-bit stored image; The bubble diameter threshold in the bubble diameter threshold sequence is used to determine whether each connected region obtained in the current recognition round is a large target or a small target, and the bubble diameter threshold of the corresponding recognition round gradually increases as the recognition round increases.

6. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1, characterized in that, In step 4), during the first round of recognition, a background noise threshold is used. Binarization is performed on the high-speed image to distinguish between foreground and background; connected component analysis is performed on the foreground portion to obtain the equivalent diameter and centroid coordinates of each connected component; the equivalent diameter of each connected component is then compared with the bubble diameter threshold input in this round. For comparison, the equivalent diameter is greater than The connected components are marked as large targets, smaller than 0. The connected components are marked as small targets; based on the small targets, the bubble diameter and centroid outputs of this round are obtained, and all large targets are combined into the Mask region of this round's output for the next round of recognition.

7. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1, characterized in that, In the second round of identification, valid elements within the Mask region identified in the first round are extracted, using a scattered light noise threshold. Binarization is performed to distinguish between foreground and background; connected component analysis is performed on the foreground to obtain the equivalent diameter and centroid coordinates of each connected component; the bubble diameter threshold of the current round is used to mark the large and small targets of the current round, and the small targets are used for the bubble diameter and centroid output of the current round. All large targets are combined into the Mask2 region of the current round output for the next round of recognition.

8. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 6, characterized in that, Starting from the third round of recognition, the effective elements within the Mask region from the previous round are first extracted. The grayscale threshold input in this round is used for binarization to distinguish the foreground from the background. Connectivity analysis is performed on the foreground to obtain the equivalent diameter and centroid coordinates of each connected component. Large and small targets are marked according to the bubble diameter threshold input in this round. Diameter compensation is performed on the small targets, and the compensated bubble diameter and centroid of the small targets are output. All the large targets in this round together constitute the Mask region output in this round.

9. The method for identifying high-velocity cloud cavitation bubbles based on adaptive multi-threshold segmentation according to claim 1 or 8, characterized in that, From the third round onwards, each round of identification includes: S1: Obtain the total input area and input region for this round of recognition. The input region and total input area for the third round of recognition are the Mask region and its total area for the second round of recognition, respectively. The total input area is dynamically updated in each subsequent round based on the recognition results of the previous round. S2: Based on the grayscale threshold identified in this round, binarize the effective information of the grayscale image within the input region of this round, and perform connected component analysis to calculate the area ratio of each connected component relative to all connected components. ; S3: Calculate the equivalent area of ​​each connected component as And based on this, convert it to the equivalent diameter of a circle, where, This represents the total input area for this round of identification; S4: Divide large and small targets according to the equivalent diameter of the circle: Connected regions with an equivalent diameter greater than the bubble diameter threshold of the current round are retained as the input region for the next round of recognition; connected regions with an equivalent diameter less than the bubble diameter threshold of the current round are regarded as small targets, and the equivalent diameter and centroid of each small target are output. S5: Calculate the percentage of the total area of ​​all smaller objectives relative to the total area of ​​all connected components in this round. The total input area for the next round of recognition is updated as follows: ； S6: Repeat S1 to S5 until the set recognition round is reached, or =0.

Images