A method for extracting parameters of mist flow droplets

Abstract

The present invention relates to a method for extracting parameters of mist droplets, comprising the following steps: performing color inversion processing on an original image to obtain an inverted image; obtaining a background homogenization image I2; obtaining a composite enhanced image I3; performing binarization processing on a filtered image I4 to obtain a binarization image I5; identifying droplets using an image recognition algorithm for waisted circle fitting to obtain a contour image and contour parameters of the droplets after fitting; and calculating an average radius and an average velocity of the droplets.

Description

Technical Field

[0001] The present invention relates to the field of gas-liquid two-phase flow parameter measurement, and in particular to a method for extracting mist flow droplet parameters. Background Art

[0002] In the field of moisture detection, it is crucial to obtain the parameters of droplets. In recent years, with the rapid development of image processing technology, many researchers have applied image methods to the field of moisture detection. By extracting and computing effective information from the collected images, they can calibrate and improve the measurement equipment and enhance the measurement performance. For example, image processing technology is applied to the design and evaluation of microfluidic devices in micro-electromechanical systems. [1] ;Measure the contact angle of liquid droplets using image analysis [2] . However, this method is rarely used in the field of wet gas flow measurement such as oil and natural gas. And when the fluid flow rate is high, the mist annular flow is the main two-phase flow type, which puts higher requirements on the application of the image method. The droplet images obtained by the existing image processing methods contain a lot of out-of-focus invalid information. At the same time, there are trailing images due to the movement of the droplets. When obtaining the flow parameters, on the one hand, it will be interfered by the invalid information. On the other hand, the accuracy of the obtained droplet parameters is not high. There is a lot of room for improvement in the image processing measurement algorithm. References:

[0003] [1] Li Xiaohua. MEMS micro-droplet volume measurement method based on image processing[J]. China Testing, 2014, 40(04): 37-41.

[0004] [2] Xiong Yan, Jia Zhihai, Cai Xiaoshu. Measurement of droplet contact angle using image analysis[J]. Metrology and Measurement Technology, 2010, 30(02): 9-11. Summary of the Invention

[0005] The present invention provides a method for extracting parameters of mist droplets, which can reduce the interference effect of out-of-focus droplets in the parameter measurement of high-speed mist droplets and improve the accuracy of droplet parameter measurement. The present invention first obtains the original image of the moving droplets through an imaging method, then transforms the image through image processing technology, and finally obtains a binary image containing only valid information. The image recognition algorithm of the waist circle fitting is used to mark the valid objects on the binarized image and obtain the corresponding parameters. The technical solution is as follows:

[0006] A method for extracting parameters of mist droplets comprises the following steps:

[0007] (1) Perform inverse color processing on the original image to obtain an inverse color image;

[0008] (2) Using the white hat transformation method, according to the size of the droplet object, the transformation kernel of the pixels occupied by it is determined to obtain the background homogenization image I2;

[0009] (3) A composite enhancement method is used to achieve image enhancement and obtain a composite enhanced image I3. The method is as follows:

[0010] 1) For the background homogenized image I2, use the Sobel operator to obtain the Sobel contour image, and then perform normalization processing on it;

[0011] 2) Using Gamma transform, nonlinear stretching of the normalized image is achieved to improve its contrast and obtain a nonlinearly stretched image;

[0012] 3) Using the Butterworth low-pass filtering method to filter the image after the nonlinear stretching in the previous step, select the filter cutoff frequency D0 and the filter power n to obtain the filtered Sobel enhanced image ΔI2;

[0013] 4) superimposing the obtained Sobel enhanced image ΔI2 onto the background homogenized image I2 to obtain a composite enhanced image I3;

[0014] (4) Using a filtering method combining frequency domain and spatial domain, the composite enhanced image I3 is filtered and denoised to obtain a filtered image I4;

[0015] (5) Binarizing the filtered image I4 to obtain a binary image I5;

[0016] (6) The droplet is identified using the image recognition algorithm of the waist circle fitting. The identification steps for a droplet are as follows:

[0017] 1) Perform rectangular bounding box fitting, and set the coordinates of the center point O of the fitted rectangular bounding box to be (x O ,y O ), the width and length of the rectangle are w and h respectively, and the rotation angle is a; when the fitted waist circle is within the rectangular bounding box, the fitted waist circle includes the upper circular arc, the middle rectangle and the lower circular arc, (x M ,y M ) is the coordinate of the center of the upper circle, (x N ,y N ) are the coordinates of the center of the circle below,

[0018] 2) Calculate the center of the upper circle and draw it:

[0019]

[0020] Where (x M ,y M ) are the coordinates of the center of the upper circle, and b is the complementary angle of a;

[0021] 3) Calculate the center position of the lower circle and draw it:

[0022]

[0023] Where (x N ,y N ) are the coordinates of the center of the circle below;

[0024] 4) Draw two parallel common tangents to the upper and lower circles.

[0025] 5) Coordinates of the points of tangency between the two common tangent lines and the upper circle arc (x E ,y E ), (x F ,y F ) is calculated as:

[0026]

[0027]

[0028] 6) The point of tangency between the two common tangent lines and the arc below the circle (x G ,y G ), (x H ,y H ) is calculated as:

[0029]

[0030]

[0031] 7) Make two parallel tangent segments EG and FH, with the waist circle radius r being half the width w of the rectangular frame, and obtain the contour image and contour parameters after droplet fitting;

[0032] (7) Calculate the average radius and average velocity of the droplets.

[0033] This method effectively removes interference from invalid background information. It also uses a waist-circle fitting method to accurately capture the droplet's contour. By calculating parameters such as the enclosing area and deflection angle from the obtained contour, the droplet's motion characteristics can be accurately determined, improving measurement accuracy. It also provides reasonable data references for calibrating and improving measurement equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a schematic diagram of the principle of waist-circular contour fitting in the present invention.

[0035] Figure 2 is the droplet image to be processed.

[0036] Figure 3 Flowchart of the image processing steps of the present invention.

[0037] Figure 4 This is the image processing result of the present invention.

[0038] Figure 5 This is a flow chart of the image composite enhancement steps of the present invention. DETAILED DESCRIPTION

[0039] All technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0040] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0041] Traditional image enhancement methods, such as adaptive histogram equalization and linear transformation, cannot effectively remove a large amount of invalid information from an image, resulting in a significant amount of interference in the binary image. Therefore, this patent employs a composite image enhancement method to improve image accuracy, enhance the contrast of the image after background homogenization, and better distinguish particles from the background. For the identification of droplet objects, their shape is approximately circular. Therefore, a rectangular bounding box for the droplet can be fitted first, and then the circular shape can be drawn based on this.

[0042] The present invention's method for extracting parameters from mist droplets combines traditional image processing methods with composite enhancement and circular fitting to extract droplet parameters. This method is implemented by the following steps: first, obtaining an original image of the moving droplets using an image method, then transforming the image using image processing techniques to obtain a binary image containing only valid information; and finally, using a circular fitting image recognition algorithm, marking valid objects in the binarized image to obtain corresponding parameters.

[0043] Figure 2 is the droplet image to be processed, Figure 3 This is a flowchart of the steps of the image processing method involved in this patent. Figure 2 , the image is processed using the processing method provided in Figure 3, as detailed below:

[0044] 1. Invert color processing

[0045] The original image obtained by the image reading method, that is, the original image, is set as I0, its gray value function is I0(x,y), and the gray value of the pixel with coordinates (i,j) is I0(i,j). In the following description, similar representation methods will not be specifically explained. In order to make the droplet more obvious compared to the background, the image is first inverted to obtain an inverted image, which is set as I1. As shown in formula (1). The processing result is as follows Figure 4 As shown in (a).

[0046] I1(i,j)=255-I0(i,j) (1)

[0047] 2. Background homogenization

[0048] Read I1, use the white hat transformation method, and adopt a transformation kernel slightly larger than the pixels occupied by the droplet object to eliminate the uneven illumination problem of the image and obtain a background uniform image, which is set as I2. The processing result is as follows Figure 4 (b) shown.

[0049] 3. Composite reinforcement

[0050] The composite enhancement method is used to achieve image enhancement. The specific process is as follows: Figure 5 shown.

[0051] First, load I2 and use the Sobel operator to directly obtain the Sobel contour image, which is then normalized as shown in formula (2);

[0052]

[0053] in, is the normalized value of the corresponding point.

[0054] Then, the Gamma transform is used to realize the nonlinear stretching of the normalized image to improve its contrast and obtain the transformed image I * 2, as shown in formula (3);

[0055]

[0056] Where c is the magnification, I * 2(i,j) is the grayscale value of the corresponding point in the transformed image.

[0057] In this embodiment, c=1 and γ=0.7 are selected.

[0058] After that, the Butterworth low-pass filter method is used to filter the previous step to remove high-frequency noise, select the appropriate filter cutoff frequency and filter power, and save the filtered Sobel enhanced image ΔI2. The specific process is as follows:

[0059] Algorithm 1: Butterworth low-pass filtering

[0060] Step 1: Now we give the discrete Fourier transform method to transform the image from the spatial domain to the frequency domain, as shown in formula (4).

[0061]

[0062] Where I is the spatial domain image to be transformed, the image size is M×N, and I(x,y) is its grayscale value function; FI(u,v) is the transformation result, u and v are both frequency components, and u=0,…,M-1; v=0,…,N-1.

[0063] Let I=I * 2 is substituted into the formula, and the spatial domain image I * 2Convert to frequency domain image FI * 2.

[0064] Step 2: Use Butterworth filter to low-pass filter it. The Butterworth filter function is shown in formula (5);

[0065]

[0066] Where D(u,v) and H(u,v) are the input and output of the filter function respectively, D0 is the cutoff frequency, and n is the filter power.

[0067] D(u,v)=FI * Substitute 2(u,v) into the formula, select the appropriate cutoff frequency and filter power, and you can get the filtered frequency domain image H_FI * 2(u,v).

[0068] Step 3: Use the inverse discrete Fourier transform to transform the image from the frequency domain back to the spatial domain. As shown in formula (6);

[0069]

[0070] This is the inverse transform form of the formula, where FI is the frequency domain image to be transformed and I is the transformation result.

[0071] Set FI(u,v)=H_FI * Substituting 2(u,v) into the formula, we can get the Sobel enhanced image ΔI2(x,y).

[0072] In this embodiment, D0=100 and n=1 are selected.

[0073] Finally, the obtained Sobel enhanced image ΔI2 is superimposed on the background homogenized image I2. The superposition process is shown in formula (7) to obtain the final composite enhanced image I3.

[0074] I3(i,j)=I2(i,j)+ΔI2(i,j) (7)

[0075] The processing results are as follows Figure 4 (c) shown.

[0076] 4. Filtering

[0077] For I3, a low-pass filter is performed using a filtering method that combines spatial and frequency domains to remove high-frequency interference in the image. The specific process is as follows.

[0078] First, in the spatial domain, a combination of Gaussian filtering and median filtering is used to reduce the noise of the image. The specific process is as follows:

[0079] Algorithm 2: Gaussian filtering

[0080] Step 1: Construct the Gaussian function as shown in formula (8);

[0081]

[0082] where σ is the standard deviation.

[0083] Step 2: Generate a Gaussian filter template based on the Gaussian function. If the sum of the template coefficients is required to be 1, the filter template can be expressed by formula (9);

[0084]

[0085] The template size is m×n. For the convenience of indexing, m=2×a+1, n=2×b+1, where a and b are both positive integers.

[0086] In this embodiment, a 3*3 template with σ=1 is selected for Gaussian filtering.

[0087] Step 3: Select the point at position (m, n) in I3, denoted by I3(m, n). With it as the center, select the surrounding "8-neighborhood points," namely, the 8 points above, below, left, right, upper left, lower left, upper right, and lower right of the point. From top to bottom and from left to right, the grayscale values of each point can be expressed as P1, P2, ..., P9. Correspondingly, the 9 values of the 3*3 Gaussian filter template can be expressed as W1, W2, ..., W9, and the filtering result can be given by formula (10).

[0088]

[0089] Among them I 31 (m,n) is the result of Gaussian filtering of I3(m,n).

[0090] Step 4: Determine whether all points in I3 have been traversed. If so, the algorithm ends; otherwise, change the position of the current point and return to step 3.

[0091] By the above algorithm, we can get the Gaussian filtered image I 31 .

[0092] Algorithm 3: Median Filter

[0093] Step 1: Select I 31 Points at the (m,n) position in the middle, use I 31 (m,n) represents that, with it as the center, the grayscale values of the "8-neighborhood points" around it are selected from top to bottom and from left to right, and can be expressed as Q1, Q2, ..., Q9 in sequence.

[0094] Step 2: Sort Q1 to Q9 by size and find the middle value Q mid , if the filtered image is I 32 , then I 32 (m,n)=Q mid .

[0095] Step 3: Determine whether I has been traversed 31 If so, the algorithm ends; otherwise, the position of the current point is changed and the algorithm returns to step 1.

[0096] By the above algorithm, we can get the median filtered image I 32 , which is the spatial domain filtered image.

[0097] Afterwards, the obtained spatial domain filtered image I 32 Perform frequency domain filtering, the process of which is given in Algorithm 1. In this embodiment, a Butterworth low-pass filter is selected, and D0=100 and n=1 are selected.

[0098] After the above processing, the filtered image I4 can be obtained.

[0099] The processing results are as follows Figure 4 (d) shown.

[0100] 5. Binarization

[0101] I4 is binarized to obtain a binarized image I5. The binarized image will be used as the basic data for droplet parameter extraction. This embodiment uses the Otsu algorithm to obtain a better threshold for image binarization and uses white hat transformation to further improve the accuracy of the binarized image. The processing results are shown in Figure 2. Figure 4 (e) shown.

[0102] 6. Waist-round fitting

[0103] Now combined Figure 1 , the image recognition algorithm based on the waist circle fitting is used to identify the droplet in I5. The direction of the reference coordinate system for this process has been marked in the figure. The specific steps are as follows:

[0104] First, the rectangular bounding box is calculated and a loop is constructed to process each droplet object to obtain the droplet contour set. Then, each element in the set, that is, the contour of each droplet object, is processed and the minimum rectangular bounding box is obtained using mathematical methods. The length, width, height and rotation angle of the vertex position are returned.

[0105] After that, based on the rectangular bounding box, the waist circle can be drawn. For a droplet (contour) object, its fitted waist circle cross section is as follows Figure 1 As shown, it includes the upper circular arc, the middle rectangle and the lower circular arc. Let the center coordinate of the rectangular bounding box object returned after the operation be (x O ,y O ), length is h, width is w, and rotation angle is a. Figure 1 As shown, if (x O ,y O ) is the origin of the coordinate system, and the given direction is the positive direction of the coordinate system. Calculate the coordinates of the required points. First, calculate the center of the upper circle and draw it. The calculation is shown in formula (11).

[0106]

[0107] Where (x M ,y M ) are the coordinates of the center of the upper circle, and b is the complementary angle of a.

[0108] Then, the lower circle can be drawn in the same way and the center of the lower circle (x N ,y N ). As shown in formula (12)

[0109]

[0110] Then, find the point of tangency between the two parallel tangent lines and the upper circle. The calculation is shown in formula (13).

[0111]

[0112] Where (x E ,y E ), (x F ,y F ) are the coordinates of the points of tangency between the two common tangent lines and the upper circle arc.

[0113] After that, we can get the tangent points of the two common tangent lines and the arc below the circle (x G ,y G ) and (x H ,y H ), as shown in formula (14).

[0114]

[0115] Draw line segments EG and FH, which are two parallel tangent segments. The radius r of the waist circle is actually half of the width w of the rectangular frame, that is,

[0116] Finally, the contour image and contour parameters of the droplet after fitting are obtained, including the center coordinates (x O ,y O ), waist radius r and waist length h.

[0117] The processing results are as follows Figure 4 (f) shown.

[0118] 7. Parameter calculation

[0119] Finally, the droplet parameters are calculated based on the obtained data. Assume that the exposure time of the shooting process is Δt, the number of valid droplet objects contained after the binarization process is T, and the waist circle radius and waist circle length returned by the waist circle fitting of the i-th droplet object are r i With h i , then the average radius and average velocity of the droplet object can be calculated using formula (15).

[0120]

[0121] in are the average radius, average waist length and average speed of the effective droplets contained in the data image, respectively.

[0122] In summary, the present invention proposes a method for extracting parameters of mist droplets, which aims to weaken the interference effect of out-of-focus droplets in the field of parameter measurement of high-speed mist droplets and improve the accuracy of droplet parameter measurement.

[0123] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the claims.

Claims

1. A method for extracting parameters of mist droplets, comprising the following steps: (1) Perform inverse color processing on the original image to obtain an inverse color image; (2) Using the white hat transformation method, according to the size of the droplet object, the transformation kernel of the pixels occupied by it is determined to obtain the background homogenization image I2; (3) A composite enhancement method is used to achieve image enhancement and obtain a composite enhanced image I3. The method is as follows: 1) For the background homogenized image I2, use the Sobel operator to obtain the Sobel contour image, and then perform normalization processing on it; 2) Using Gamma transform, nonlinear stretching of the normalized image is achieved to improve its contrast and obtain a nonlinearly stretched image; 3) Using the Butterworth low-pass filtering method to filter the image after the nonlinear stretching in the previous step, select the filter cutoff frequency D0 and the filter power n to obtain the filtered Sobel enhanced image ΔI2; 4) superimposing the obtained Sobel enhanced image ΔI2 onto the background homogenized image I2 to obtain a composite enhanced image I3; (4) Using a filtering method combining frequency domain and spatial domain, the composite enhanced image I3 is filtered and denoised to obtain a filtered image I4. The method is: (5) Binarizing the filtered image I4 to obtain a binary image I5; (6) The droplet is identified using the image recognition algorithm of the waist circle fitting. The identification steps for a droplet are as follows: 1) Perform rectangular bounding box fitting, and set the coordinates of the center point O of the fitted rectangular bounding box to be (x O ,y O ), the width and length of the rectangle are w and h respectively, and the rotation angle is a; when the fitted waist circle is within the rectangular bounding box, the fitted waist circle includes the upper circular arc, the middle rectangle and the lower circular arc, (x M ,y M ) is the coordinate of the center of the upper circle, (x N ,y N ) are the coordinates of the center of the circle below, 2) Calculate the center of the upper circle and draw it: Where (x M ,y M ) are the coordinates of the center of the upper circle, and b is the complementary angle of a; 3) Calculate the center position of the lower circle and draw it: Where (x N ,y N ) are the coordinates of the center of the circle below; 4) Draw two parallel common tangents to the upper and lower circles. 5) Coordinates of the points of tangency between the two common tangent lines and the upper circle arc (x E ,y E ), (x F ,y F ) is calculated as: 6) The point of tangency between the two common tangent lines and the arc below the circle (x G ,y G ), (x H ,y H ) is calculated as: 7) Make two parallel tangent segments EG and FH, with the waist circle radius r being half the width w of the rectangular frame, and obtain the contour image and contour parameters after droplet fitting; (7) Calculate the average radius and average velocity of the droplets.

2. The method for extracting mist droplet parameters according to claim 1, characterized in that: In step (3) 3), select D0=100, n=1.

3. The method for extracting mist droplet parameters according to claim 1, characterized in that: Step (4) is as follows: 1) In the spatial domain, a method combining Gaussian filtering and median filtering is used to perform noise reduction to obtain the spatial domain filtered image I 32 ; 2) The obtained spatial domain filtered image I 32 Perform frequency domain filtering and filter again using a Butterworth low-pass filter to obtain a filtered image I4.

4. The method for extracting mist droplet parameters according to claim 1, characterized in that: Step (5) is specifically as follows: using the Otsu algorithm to obtain the threshold value for image binarization, and using white hat transformation to improve the accuracy of the binarized image.

Images