A data processing method for measuring flow field velocity based on marker line characteristics

Abstract

To address the problems of existing methods for extracting intrinsic characteristic parameters of marker lines, which suffer from severe background interference, severe signal distortion, and large uncertainty in complex measurement environments, the present invention provides a data processing method for measuring flow field velocity based on marker line characteristics. The method comprises the following steps: 1. analyzing the noise characteristics of the obtained marker line fluorescence image and removing background noise using a generative adversarial neural network method based on the noise distribution law; 2. Gaussian filtering the image; 3. Image enhancement using the Hessian feature map; 4. OSTU image segmentation; 5. Extracting the basic skeleton of the marker line from the segmented image through refinement; 6. Calculating the normal direction of the marker line center using the Hessian matrix, and finding the extreme point in the normal direction through quadratic differentiation to obtain the centerline position; 7. Calculating the flow field velocity information by dividing the centerline position difference of the two marker fluorescence images by the time interval between the two images.

Description

Technical Field

[0001] The present invention relates to image processing technology, belonging to the technical field of laser spectrum diagnosis and analysis of complex physical fields, and particularly to a data processing method for measuring flow field velocity based on marker line features. Background Art

[0002] Accurately extracting the intrinsic characteristic parameters of the marker line is the key to truly using the fluorescent marker image for flow field parameter measurement. Studying its precise extraction method and the precisely extracted information can be used for complex field velocity measurement.

[0003] However, due to stray light such as fluorescence interference, particle scattering, and wall scattering in complex flow fields, the signal-to-noise ratio (SNR) of the marker line fluorescence image will be greatly reduced. The instability and unevenness in high-turbulence flow fields will also cause distortion in the spatial and morphological distribution of the marker line. The uneven distribution of velocity in the flow field caused by local turbulence will cause the marker line morphology to change, resulting in irregular bending and horizontal broadening of the marker line. At the same time, molecular diffusion and chemical reactions also affect measurement uncertainty. The properties of molecules lead to their inevitable random diffusion in the flow field, which will also cause the marker line to expand or distort. Changes in the fluorescence intensity of the marker line under the action of chemical reactions will also affect the measurement uncertainty. Therefore, the key to reducing measurement uncertainty in complex environments is to improve the noise and distortion resistance of the marker line extraction method.

[0004] Existing methods for extracting the intrinsic characteristic parameters of a marker line can be divided into the following three categories: (a) Fitting. The intensity distribution of the marker line profile is obtained through fitting, and the marker line position is determined by the peak of the fitting curve. (b) Correlation. A row-by-row correlation algorithm is used. (c) Morphological methods. The active contour method is used to find the line contour. Based on the contour extraction, the intensity distribution is then Gaussian-fitted to extract the line position information. However, these methods still suffer from significant uncertainty in complex measurement environments with severe background interference and large signal distortion. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem of large uncertainty in the existing method for extracting the intrinsic characteristic parameters of the marking line in a complex measurement environment with severe background interference and large signal distortion, and to provide a data processing method for measuring flow field velocity based on the marking line characteristics.

[0006] The design concept of the present invention is as follows: first, noise characteristics of the obtained fluorescent image of the marker line are analyzed, and a generative adversarial neural network method is used to remove background noise according to the noise distribution law, thereby achieving image enhancement and highlighting the contrast between the image marker line and the background; second, Gaussian filtering is performed on the image to ensure the stability of Hessian matrix decomposition; then, Otsu threshold image segmentation is performed based on the enhanced effect of the Hessian feature map; the basic skeleton of the marker line is extracted by refinement in the segmented image, the normal direction of the marker line center is calculated using the Hessian matrix, and the extreme point in the normal direction is obtained by quadratic differentiation to obtain the center line position; finally, the flow field velocity information is obtained by dividing the center line position difference of the two marker fluorescence images by the shooting interval of the two marker fluorescence images.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] A data processing method for measuring flow field velocity based on marker line features is characterized in that it includes the following steps:

[0009] Step 1: Use a camera to capture the original fluorescent image of the marker light spot in the flow field, analyze the characteristic parameters of the captured original fluorescent image I1, and use a generative adversarial neural network method to remove noise, thereby obtaining a preprocessed fluorescent image I2;

[0010] Step 2: Based on the characteristics of the marking line, the weighted average grayscale value of the pixels in the neighborhood of the preprocessed marking line fluorescence image I2 is calculated and used as the value of the pixel at the center of the spot. The preprocessed marking line fluorescence image I2 is Gaussian filtered to obtain the filtered spot image I3. Gaussian filtering eliminates the influence of the Hessian matrix being sensitive to isolated noise points, and the combustion background interference in the Hessian feature map can be effectively suppressed.

[0011] Step 3: Calculate the 2×2 Hessian matrix corresponding to each pixel in the filtered spot image I3, and then calculate the eigenvalues of each Hessian matrix to form a diagonal matrix D. The eigenvectors corresponding to each column in the diagonal matrix D form a new feature matrix V. The image obtained according to the feature matrix V is the Hessian feature image I4.

[0012] Step 4: Use the maximum inter-class variance to perform Otsu threshold segmentation on the Hessian feature image I4 to obtain the spot area image I5;

[0013] Step 5: Refine the spot area image I5 and extract the basic skeleton of the marking line in the image to obtain the spot image I6;

[0014] Step 6. In the skeleton of the spot image I6, the grayscale distribution function on the skeleton cross section is expanded by the second-order Taylor in the normal direction using the Hessian matrix to obtain the centerline position. The normal direction of the skeleton of the spot image I6 is then calculated using the Hessian matrix. The extreme points in the normal direction are obtained by quadratic differentiation. The extreme points are connected to obtain the spot centerline image I7. By replacing the corrosion operation with thinning, the overall structure of the marker line skeleton is retained, the ability to handle small cracks is improved, the purpose of repairing broken lines is achieved, and the ability to resist morphological distortion is improved.

[0015] Step 7: Repeat steps 1 to 6 at least once. At this time, at least two or more spot center line images I7 are obtained. The spot center point of the same pixel in any two spot center line images I7 is selected and defined as d1 and d2 respectively.

[0016] Step 8. Calculate speed

[0017] The flow field velocity v at the location of the fluorescent image of the marking spot is calculated using the spot center point difference:

[0018] v=(d1-d2) / Δt

[0019] Wherein, Δt is the time interval between capturing two fluorescent images of the marking spot.

[0020] Furthermore, step three is specifically as follows:

[0021] Define the pixel points in the filtered spot image I3 as X1,...,X j ,...X n , and its corresponding 2×2 Hessian matrix is Where 1≤j≤n, n is the number of pixels, and j and n are both integers;

[0022] 3.1、Calculate any pixel X j The corresponding 2×2 Hessian matrix

[0023] 3.1.1、X j The row coordinates of are convolved with the vector [1, -2, 1], and the column coordinates are convolved with "1", resulting in the vector

[0024] 3.1.2、X j The row coordinates of are convolved with the vector [-0.5, 0, 0.5], and the column coordinates of are convolved with the vector [-0.5, 0, 0.5] to obtain the vector

[0025] 3.1.3、X jThe row coordinates of are convolved with the vector [-0.5, 0, 0.5], and the column coordinates of are convolved with the vector [-0.5, 0, 0.5] to obtain the vector

[0026] 3.1.4、X j The row coordinates of are convolved with "1", and the column coordinates are convolved with the vector [1, -2, 1] to obtain the vector

[0027] 3.1.5. Composition of pixel point X j The Hessian matrix

[0028] 3.2. Calculate the eigenvalues of each Hessian matrix to form a diagonal matrix D;

[0029] 3.3. A new feature matrix V is constructed by the eigenvectors corresponding to each column in the diagonal matrix D, and the corresponding image obtained is the Hessian feature image I4.

[0030] Compared with the prior art, the present invention has the following beneficial technical effects:

[0031] 1. The data processing method for measuring flow field velocity based on marker line characteristics proposed in the present invention takes into account the characteristics of the marker line, has strong anti-noise and anti-distortion capabilities, and effectively improves the flow field velocity measurement accuracy in complex laser diagnostic technology.

[0032] 2. The data processing method for measuring flow field velocity based on marker line features proposed in the present invention performs Gaussian filtering on the input image to eliminate the influence of the Hessian matrix's sensitivity to isolated noise points. The corresponding 2×2 Hessian matrix is calculated for each pixel after filtering, and the largest eigenvalue in the Hessian matrix is taken as the value of the pixel in the feature map. At this time, the combustion background interference in the Hessian feature map can be effectively suppressed, and the Otsu method can be used to segment marker lines with multi-pixel widths.

[0033] 3. The data processing method proposed in the present invention for measuring flow field velocity based on marker line features combines the anti-distortion Hessian matrix and skeleton refinement method to repair details and accurately fit the centerline position of the marker signal.

[0034] 4. The present invention adopts a morphological method in view of the morphological characteristics of the marker line signal, is compatible with the integrity of the marker line, replaces the corrosion operation by refinement, retains the overall structure of the marker line skeleton, improves the processing ability of small cracks, achieves the purpose of repairing broken lines, and improves the ability to resist morphological distortion.

[0035] 5. The present invention introduces a Hessian matrix optimization algorithm to obtain the structural normal direction information, avoid uncertainty in the position of the marker line with large curvature changes, and retain rich marker line curvature information; considering that the grayscale value in the marker line cross section is Gaussian distributed along the normal direction of the line, the grayscale centroid method is used to fit the center position; the flow field velocity is obtained by dividing the center position difference of the two marker line fluorescence images by the time interval. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Schematic diagram of the flow field velocity inversion calculation process in the data processing method for measuring flow field velocity based on marker line characteristics of the present invention;

[0037] Figure 2 Schematic diagram of the process of extracting the center line of the light spot from the HTV experimental data in the combustion flow field according to an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of the center position of the corresponding marking line obtained by extracting HTV experimental data in an embodiment of the present invention, where (a), (b), and (c) are the results of HTV experimental data extraction in three different combustion flow fields. DETAILED DESCRIPTION

[0039] To further clarify the objectives, advantages, and features of the present invention, the following describes in further detail a data processing method for measuring flow field velocity based on marker line features, as proposed by the present invention, in conjunction with the accompanying drawings and specific examples. Those skilled in the art should understand that these embodiments are intended only to illustrate the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0040] like Figure 1 As shown, a data processing method for measuring flow field velocity based on marker line features includes the following steps:

[0041] Step 1: Image preprocessing

[0042] The original marker spot fluorescence image in the flow field is captured by a shooting device. The characteristic parameters of the captured original marker spot fluorescence image I1 are analyzed. The generative adversarial neural network method is used for denoising to obtain the preprocessed marker spot fluorescence image I2. The image is enhanced and the contrast between the image marker line and the background is highlighted.

[0043] Step 2: Gaussian filtering

[0044] According to the characteristics of the marking line, the weighted average gray value of the pixels in the neighborhood is calculated and used as the value of the pixel at the center of the spot. The marking line fluorescence image I2 is Gaussian filtered to obtain the filtered spot image I3;

[0045] Step 3: Get the Hessian feature map

[0046] The 2×2 Hessian matrix corresponding to each pixel in the filtered spot image I3 is calculated. The eigenvalues of each Hessian matrix are then calculated to form a diagonal matrix D. The eigenvectors corresponding to each column in the diagonal matrix D form a new feature matrix V. The resulting image is the Hessian feature image I4. The largest eigenvalue in the Hessian matrix is taken as the value of that pixel in the feature image. At this point, the combustion background interference in the Hessian feature image can be effectively suppressed.

[0047] Define the pixel points in the filtered spot image I3 as X1,...,X j ,...X n , and its corresponding 2×2 Hessian matrix is Where 1≤j≤n, n is the number of pixels, and j and n are both integers.

[0048] Take any pixel X in the filtered image I3 j For example, the process of calculating the corresponding 2×2 Hessian matrix is: j The row coordinates of are convolved with the vector [1, -2, 1], and the column coordinates are convolved with "1", resulting in the vector X j The row coordinates of are convolved with the vector [-0.5, 0, 0.5], and the column coordinates of are convolved with the vector [-0.5, 0, 0.5] to obtain the vector X j The row coordinates of are convolved with the vector [-0.5, 0, 0.5], and the column coordinates of are convolved with the vector [-0.5, 0, 0.5] to obtain the vector X j The row coordinates of are convolved with "1", and the column coordinates are convolved with the vector [1, -2, 1] to obtain the vector Pixel X j The Hessian matrix

[0049] Step 4: Image Segmentation

[0050] The maximum inter-class variance is used to perform Otsu threshold segmentation (OSTU segmentation) on the Hessian feature image I4 to obtain the image I5 of the spot region (ROI). Narrowing the range of threshold segmentation can effectively improve the computational efficiency.

[0051] Step 5: Refine

[0052] The image I5 of the spot region (ROI) is thinned, and the basic skeleton of the marking line in the image is extracted to obtain an image I6 after the skeleton is extracted.

[0053] Step 6: Find the center line

[0054] In the spot image I6 skeleton, the center line position is obtained by performing a second-order Taylor expansion of the grayscale distribution function on the skeleton cross section in the normal direction using the Hessian matrix. The normal direction of the I6 skeleton is calculated using the Hessian matrix. The extreme points in the normal direction are obtained by quadratic differentiation. The extreme points are connected to obtain the spot centerline image I7.

[0055] Considering that in certain extreme scenarios, such as the interaction between turbulence and combustion, the light intensity distribution on the OHp (photodissociated hydroxyl) marker line will be modulated and the degree of broadening will be inconsistent, which cannot strictly satisfy the Gaussian distribution. Therefore, the Hessian matrix is used to calculate the normal direction of the center of the marker line, and the extreme point of the normal direction is obtained by quadratic differentiation as the Hessian characteristic map. This fully considers the light intensity characteristics of the cross section on the marker line. To reduce uncertainty, based on the segmentation of candidate targets and background, morphological methods such as region growing, filling, expansion, and erosion are used to merge close regions. At the same time, over-segmented candidate target regions caused by the background are removed to obtain the detection target region. Curve fitting is performed on the target region to optimize the results, and the disconnected regions are repaired to obtain the precise centerline position.

[0056] like Figure 2 FIG. 4 shows the process of obtaining the center position of the OHp marker line using the above steps, taking the HTV experimental data in the combustion flow field as an example. FIG.

[0057] Figure 3 The HTV experimental data in three different combustion flow fields were extracted using the above method. The corresponding schematic diagram of the center position of the marker line is obtained. It can be seen that after extraction using this method, the significance of the central (significant) area can be ensured to be significantly higher than that of other areas, and it has better noise resistance performance.

[0058] Step 7: Repeat steps 1 to 6 at least once to obtain at least two center line images I7. Select the center point d of the same pixel in any two center line images I7 and define them as d1 and d2 respectively.

[0059] Step 8. Calculate speed

[0060] The flow field velocity v at the location of the fluorescent image of the marking spot is calculated using the spot center point difference:

[0061] v=(d1-d2) / Δt

[0062] Wherein, Δt is the time interval between capturing two fluorescent images of the marking spot.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. A data processing method for measuring flow field velocity based on marker line features, characterized in that: The following steps are involved: Step 1: Use a camera to capture the original marker line fluorescence image in the flow field, analyze the characteristic parameters of the captured original marker line fluorescence image I1, perform denoising, and obtain a preprocessed marker line fluorescence image I2; Step 2: Calculate the weighted average grayscale value of the pixels in the neighborhood of the pre-processed marker line fluorescence image I2 according to the characteristics of the marker line, and use it as the value of the center pixel of the spot. Perform Gaussian filtering on the pre-processed marker line fluorescence image I2 to obtain the filtered spot image I3; Step 3: Calculate the 2×2 Hessian matrix corresponding to each pixel in the filtered spot image I3, and then calculate the eigenvalues of each Hessian matrix to form a diagonal matrix D. The eigenvectors corresponding to each column in the diagonal matrix D form a new feature matrix V. The image obtained according to the feature matrix V is the Hessian feature image I4; specifically: Define the pixel points in the filtered spot image I3 as X1,...,X j ,...,X n , and its corresponding 2×2 Hessian matrix is Where 1≤j≤n, n is the number of pixels, and j and n are both integers; 3.1) Calculate any pixel X j The corresponding 2×2 Hessian matrix 3.1.1) X j The row coordinates of are convolved with the vector [1, -2, 1], and the column coordinates are convolved with 1 to obtain the vector 3.1.2) X j The row coordinates of are convolved with the vector [-0.5, 0, 0.5], and the column coordinates of are convolved with the vector [-0.5, 0, 0.5] to obtain the vector 3.1.3) X j The row coordinates of are convolved with the vector [-0.5, 0, 0.5], and the column coordinates of are convolved with the vector [-0.5, 0, 0.5] to obtain the vector 3.1.4) X j The row coordinates of are convolved with 1, and the column coordinates are convolved with the vector [1, -2, 1] to obtain the vector 3.1.5) Pixel X j The Hessian matrix 3.2) Calculate the eigenvalues of each Hessian matrix to form a diagonal matrix D; 3.3) The eigenvectors corresponding to each column in the diagonal matrix D form a new eigenmatrix V. The image obtained according to the eigenmatrix V is the Hessian eigenimage I4 Step 4: Use the maximum inter-class variance to perform Otsu threshold segmentation on the Hessian feature image I4 to obtain the spot area image I5; Step 5: Refine the spot area image I5 and extract the basic skeleton of the marking line in the image to obtain the spot image I6; Step 6. In the skeleton of the spot image I6, use the Hessian matrix to perform a second-order Taylor expansion of the grayscale distribution function on the skeleton cross section in the normal direction to obtain the center line position. Then use the Hessian matrix to calculate the normal direction of the spot image I6 skeleton. Use the second differential to find the extreme point in the normal direction. Connect the extreme points to obtain the spot centerline image I7. Step 7: Repeat steps 1 to 6 at least once to obtain at least two light spot center line images I7, and select the light spot center point of the same pixel in any two light spot center line images I7, and define them as d1 and d2 respectively; Step 8. Calculate speed The flow field velocity v at the location of the original marked spot line image is calculated using the spot center point difference: v=(d1-d2) / Δt in, Δt is the time interval between the two original marker line fluorescence images.

2. The data processing method for measuring flow field velocity based on marker line characteristics according to claim 1, characterized in that: In step 1, the denoising method is a generative adversarial neural network method.

Images