High-precision particle image velocity measurement method

Abstract

The invention discloses a high-precision particle image velocity measurement method, and relates to the technical field of particle image velocity measurement, and the method specifically comprises the steps: 1, building a high-precision particle image measurement experiment platform, capturing continuous particle images through a laser-camera system with a synchronous time sequence, achieving the precise synchronization of laser pulses and camera exposure through an FPGA synchronous controller, and carrying out the measurement of the continuous particle images; the method comprises the following steps of: 1, acquiring a displacement image captured by a high-speed camera, and verifying synchronization precision, 2, processing the displacement image captured by the high-speed camera by using a technology combining a cross-correlation method and an optical flow method, and 3, introducing singular value decomposition in a PIV data processing process, decomposing flow field data, and separating main signals and noise. The sub-pixel-level precision of particle displacement measurement is remarkably improved, the signal-to-noise ratio of data is enhanced, and real-time error correction is achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of particle image velocimetry, and in particular relates to a high-precision particle image velocimetry method. Background Art

[0002] Particle Image Velocimetry (PIV), as the preferred experimental technique for measuring velocity fields in fluid mechanics, has made significant progress in image acquisition and data processing algorithms. PIV measures the movement of tracer particles in the fluid and captures continuous images to calculate the flow velocity. The PIV system uses high-frame-rate cameras and higher-intensity lasers to provide sufficient time resolution to capture rapidly changing flow fields. In addition, some advanced PIV systems also integrate functions such as real-time data processing and automated flow field identification to improve the convenience and efficiency of measurement. The technology has been widely used in the study of various complex flow fields. Its maturity and wide recognition make it a powerful tool for analyzing various flow conditions.

[0003] However, despite the progress made in improving frame rate and image quality, these systems still have limitations in dealing with image noise, improving the accuracy of sub-pixel displacement measurement, and extending the measurement range. In particular, in applications that require high sensitivity and high precision measurement, such as microfluidics and biomedical flow field measurement, existing solutions often have difficulty meeting these requirements. Summary of the invention

[0004] The purpose of the present invention is to provide a high-precision particle image velocimetry method to solve the technical problems in the prior art that are still limited in processing image noise, improving the accuracy of sub-pixel displacement measurement and expanding the measurement range.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A high-precision particle image velocimetry method, comprising: Step 1: Build a high-precision particle image measurement experimental platform and use a time-synchronized laser-camera system to capture continuous particle images; The FPGA synchronization controller is used to achieve precise synchronization between laser pulses and camera exposure, and the synchronization accuracy is verified; Step 2: Process the displacement image captured by the high-speed camera using a combination of cross-correlation method and optical flow method; The displacement image is preliminarily processed, the normalized cross-correlation function is calculated, the velocity vector is solved iteratively through the optical flow equation, and the sub-pixel interpolation technology is introduced to improve the displacement resolution; Step 3: Introduce singular value decomposition in the PIV data processing process to decompose the flow field data and separate the main signal from the noise; The refined velocity field data is decomposed by singular value decomposition, the low-order singular values ​​are truncated, and the denoised velocity field matrix is ​​reconstructed using the retained singular values ​​and the corresponding singular vectors.

[0006] Furthermore, a high-precision particle image measurement experimental platform was built, and a time-synchronized laser-camera system was used to capture continuous particle images. The specific method is as follows: A high-precision particle image measurement experimental platform with a time-synchronized laser-camera system was built. The platform used a three-dimensional coordinate frame with a positioning uncertainty in the range of (-a, a) for high-precision positioning, and a CMOS camera with a resolution of A×B pixels and a frame rate ≥h as the image capture device. A high-energy dual-pulse laser was selected, and a synchronization signal was generated through an FPGA synchronization controller to achieve precise synchronization between the laser pulse trigger signal and the CMOS camera exposure signal, collect dynamic particle images, and use a high-speed camera to capture displacement images.

[0007] Furthermore, the synchronization signal is generated by the FPGA synchronization controller to achieve accurate synchronization between the laser pulse trigger signal and the CMOS camera exposure signal. The specific method is as follows: A TTL pulse signal was sent to simultaneously trigger the laser double pulse emission and camera exposure. The laser pulse interval Δt was dynamically adjusted according to the preset flow rate, and the synchronization error was monitored by an oscilloscope to be ≤1 microsecond. When it was detected that the particle displacement between adjacent frames exceeded 50% of the window size, Δt was shortened to 50% of the original value and the camera frame rate was increased synchronously.

[0008] Furthermore, the displacement image is preliminarily processed, and the specific method is as follows: The quality of the collected image is verified by calculating the modulation transfer function of the particles in the displacement image and the grayscale standard deviation of the particle-free area of ​​the image. The thresholds of the modulation transfer function and the grayscale standard deviation are preset. If the modulation transfer function value is less than or equal to the modulation transfer function threshold, it is judged that there is blur, otherwise, it is judged that there is no blur. If the grayscale standard deviation is greater than or equal to the grayscale standard deviation threshold, it is judged that the noise is too high, otherwise, it is judged that the noise is normal. Multi-frame particle images with no blur and normal noise are collected, and the denoising autoencoder is used to remove background noise from the image.

[0009] Furthermore, the normalized cross-correlation function is calculated, and the specific method is: After the preliminary processing, two consecutive particle images are divided into multiple cross-correlation windows of n*n pixels. The normalized cross-correlation function between the two frames of images is calculated for each window. The formula is used represents the normalized cross-correlation function between two frames of images, where represents the candidate displacement value between two frames of images, (x, y) represents the coordinates of the image pixels in the window, Represents the grayscale value of the image pixel at the (x, y) coordinate in the first frame image, Indicates the corresponding coordinate translation in the second frame image The gray value after Represents the grayscale mean of the first frame image in the selected window, Represents the grayscale mean of the second frame image in the selected window; By calculating the normalized cross-correlation function value under different displacements , find The candidate displacement value that reaches the maximum value is recorded as the best estimated displacement , output the best estimated displacement.

[0010] Furthermore, the velocity vector is solved iteratively by the optical flow equation, and the sub-pixel interpolation technology is introduced to improve the displacement resolution. The specific method is as follows: The best estimated displacement output by the cross-correlation method is used as the initial value of the iterative optimization of the optical flow method. For each pixel in the image, a window of a fixed size is selected. Within the selected window, the gradients of the image in the x and y directions are calculated. The optical flow equation of the Lucas-Kanade algorithm is used in combination with the initial value to iteratively solve the velocity vector pixel by pixel. During the iterative process, the velocity vector estimate is continuously updated until the preset maximum number of iterations is reached. After obtaining the velocity vector at the integer pixel level, the sub-pixel interpolation technology is introduced to improve the displacement resolution. Through bicubic interpolation, the displacement resolution is increased from the integer pixel level to the m pixel level.

[0011] Furthermore, the refined velocity field data is decomposed by singular value decomposition. The specific method is as follows: Receive the refined velocity field data generated by the optical flow method, including the horizontal velocity component u and the vertical velocity component v of each pixel point, and arrange the original velocity field data obtained by the optical flow method into a two-dimensional matrix according to the time and space dimensions, denoted as matrix M, where the row vector of the M matrix is ​​defined by the spatial coordinate point, that is, each row corresponds to a spatial coordinate point, the total number of rows is the number of all spatial coordinate points, and the total number of rows is represented by N. The column vector of the M matrix corresponds to the velocity vector of different time frames, the total number of columns is the length of the time series, and the total number of columns is represented by Q. Zero mean processing is performed on the matrix M to eliminate the influence of dimensional differences on the decomposition results. The formula is used Perform singular value decomposition on the matrix M, where U is the left singular vector matrix of the M matrix, with an order of N*N. Each column of the left singular vector matrix U represents a spatial modal distribution of the velocity field, reflecting the different characteristics or modes of the velocity field in space. It represents the singular value diagonal matrix of the M matrix, with an order of N*Q. Its diagonal elements are the singular values ​​of the M matrix. Larger singular values ​​correspond to modes with larger energy contributions. V is the right singular vector matrix of the M matrix, with an order of Q*Q. Each column represents the changing characteristics of the velocity field in the time dimension.

[0012] Furthermore, the low-order singular values ​​are truncated, and the retained singular values ​​and corresponding singular vectors are used to reconstruct the velocity field matrix after noise reduction. The specific method is as follows: Using the formula represents the matrix after truncated low-order singular values ​​reconstruction and denoising, where To retain the submatrix of the left singular vector matrix after retaining the first k principal components, To retain the submatrix of the singular value diagonal matrix after the first k principal components, In order to retain the submatrix of the right singular vector matrix after the first k principal components, the velocity field data obtained after denoising by singular value decomposition is the matrix , based on the first k principal components retained, the horizontal velocity component u and vertical velocity component v of each pixel are reconstructed.

[0013] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. The present invention can effectively verify the quality of the collected image by calculating the modulation transfer function of the particles in the displacement image and the grayscale standard deviation of the particle-free area of ​​the image. By combining the coordinated optimization of the cross-correlation method and the optical flow method and introducing the bicubic sub-pixel interpolation technology, it is helpful to improve the pixel level of the displacement resolution. The best estimated displacement output by the cross-correlation method is used as the initial value of the iterative optimization of the optical flow method, which can accelerate the iterative process and improve the convergence speed. In combination with the optical flow equation, the velocity vector is iteratively solved pixel by pixel, which is helpful to achieve high-precision displacement measurement; 2. The present invention introduces singular value decomposition in the process of particle image velocimetry data processing, which can accurately decompose the flow field data into main signal and noise components, help identify and separate the components that contribute most to the flow field characteristics, thereby improving the signal-to-noise ratio of the data, and reconstruct a denoised velocity field data matrix by truncating low-order singular values, which helps to improve the quality and reliability of the original data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1A method step diagram of high-precision particle image velocimetry is shown; Figure 2 A method step diagram for introducing singular value decomposition in the PIV data processing process is shown. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Embodiment 1: Figure 1 , Figure 2 A high-precision particle image velocimetry method is shown, which specifically includes the following steps: Step 1: Build a high-precision particle image measurement experimental platform and capture continuous particle images through a time-synchronized laser-camera system.

[0018] A three-dimensional coordinate frame with a positioning uncertainty in the range of (-a, a) is used as an experimental platform for high-precision particle image measurement. The three-dimensional coordinate frame is used to support PIV (particle image velocimetry) images and accurately calibrate their positions. In this embodiment, a is equal to 0.01 mm. A CMOS camera with a resolution of A×B pixels and a frame rate ≥ h is used. The camera needs to have 16-bit grayscale image capture. In this embodiment, A is set to be equal to B and equal to 2000, and h is set to be 1000fps. A high-energy dual-pulse laser is selected, and the emission wavelength, pulse energy and pulse width of the laser are limited to ensure that the particle scattering light intensity meets the high-speed imaging requirements. An FPGA synchronous controller is used to generate a synchronization signal to control the trigger signal of the laser pulse and the exposure signal of the CMOS camera. The FPGA synchronous controller sends a TTL pulse signal to simultaneously trigger the laser to emit a dual pulse and the camera to expose. The pulse interval is determined by the preset flow rate of the laser. The camera captures a frame of image after each laser pulse, and monitors the alignment of the rising edge of the laser pulse and the camera exposure signal through an oscilloscope to determine the synchronization error between the trigger signal and the exposure signal. In this embodiment, the synchronization error is required to be less than or equal to 1 microsecond. The dynamic particle image is collected, and the quality of the collected image is verified. The synchronization accuracy is verified by checking the particle position offset between adjacent frames. The smaller the particle position offset between adjacent frames, the higher the synchronization accuracy. When it is detected that the particle displacement exceeds 50% of the window size, the pulse interval Δt is shortened by half, and the camera frame rate is increased by half synchronously to avoid the displacement exceeding the tracking range of the algorithm. The collected image is recorded as the image to be processed, and the image to be processed is translated r mm along the y-axis direction. After the image to be processed is translated, the high-speed camera is used to capture these displacement images. When each image is collected, the camera is facing the synthetic image and ensures that the displacement image is captured in full.

[0019] Step 2: Use the cross-correlation method and optical flow method to process the displacement image captured by the high-speed camera.

[0020] The quality of the collected image is verified by calculating the modulation transfer function of the particles in the displacement image and the grayscale standard deviation of the particle-free area of ​​the image. The thresholds of the modulation transfer function and the grayscale standard deviation are preset. If the modulation transfer function value is less than or equal to the modulation transfer function threshold, it is judged that there is blur and the synchronization parameters need to be readjusted. Otherwise, it is judged that there is no blur. If the grayscale standard deviation is greater than or equal to the grayscale standard deviation threshold, it is judged that the noise is too high and noise reduction processing is required. Otherwise, it is judged that the noise is normal, thereby improving the image quality in low-light or high-noise environments, collecting multi-frame particle images without blur and normal noise, and applying denoising autoencoders to remove background noise from the images. The two consecutive frames of particle images after preprocessing are divided into multiple n*n pixel cross-correlation windows. For each window, the normalized cross-correlation function between the two frames of images is calculated. The specific calculation method is as follows: ; in, represents the candidate displacement value between two frames of images, (x, y) represents the coordinates of the image pixels in the window, Represents the grayscale value of the image pixel at the (x, y) coordinate in the first frame image, Indicates the corresponding coordinate translation in the second frame image The gray value after Represents the grayscale mean of the first frame image in the selected window, Represents the grayscale mean of the second frame image in the selected window; By calculating the normalized cross-correlation function value under different displacements , find The candidate displacement value that reaches the maximum value is recorded as the best estimated displacement , output the best estimated displacement.

[0021] The best estimated displacement output by the cross-correlation method is used as the initial value of the iterative optimization of the optical flow method. For each pixel in the image, a window of a fixed size is selected. Within the selected window, the gradient of the image in the x and y directions is calculated. The optical flow equation of the Lucas-Kanade algorithm is used in combination with the initial value to iteratively solve the velocity vector pixel by pixel. During the iteration, the velocity vector estimate is continuously updated until the preset maximum number of iterations is reached. After obtaining the velocity vector at the integer pixel level, the sub-pixel interpolation technology is introduced to improve the displacement resolution. Through bicubic interpolation, the displacement resolution is increased from the integer pixel level to the m or higher pixel level. In this embodiment, m is set equal to 0.1 to achieve high-precision flow field reconstruction.

[0022] Step 3: Introduce singular value decomposition in the PIV data processing process to decompose the flow field data and separate the main signal from the noise.

[0023] Receive the refined velocity field data generated by the optical flow method, including the horizontal velocity component u and the vertical velocity component v of each pixel point, and arrange the original velocity field data obtained by the optical flow method into a two-dimensional matrix according to the time and space dimensions, denoted as matrix M, where the row vector of the M matrix is ​​defined by the spatial coordinate point, that is, each row corresponds to a spatial coordinate point, the total number of rows is the number of all spatial coordinate points, and the total number of rows is represented by N. The column vector of the M matrix corresponds to the velocity vector of different time frames, the total number of columns is the length of the time series, and the total number of columns is represented by Q. Zero mean processing is performed on the matrix M to eliminate the influence of dimensional differences on the decomposition results, and singular value decomposition is performed on the matrix M. The specific decomposition method is as follows: ; Where U is the left singular vector matrix of the M matrix, with an order of N*N. Each column of the left singular vector matrix U represents a spatial modal distribution of the velocity field, reflecting the different characteristics or modes of the velocity field in space. It represents the singular value diagonal matrix of the M matrix, with an order of N*Q. Its diagonal elements are the singular values ​​of the M matrix. Larger singular values ​​correspond to modes with larger energy contributions. V is the right singular vector matrix of the M matrix, with an order of Q*Q. Each column represents the changing characteristics of the velocity field in the time dimension.

[0024] The singular value reflects the energy contribution of each mode. Larger singular values ​​correspond to modes with larger energy contributions, while smaller singular values ​​correspond to modes with smaller energy contributions. By retaining the first k largest singular values, that is, truncating low-order singular values, data denoising and signal enhancement can be achieved. The matrix reconstructed by truncating low-order singular values ​​after denoising is recorded as , It is expressed as: ; in, To retain the submatrix of the left singular vector matrix after the first k principal components, To retain the submatrix of the singular value diagonal matrix after the first k principal components, It is a submatrix of the right singular vector matrix after retaining the first k principal components. According to the pulse interval, the k value is automatically expanded to retain more high-frequency components. For example, when the flow velocity is low as determined by the pulse interval, k can be taken in the range of (h, 2h) to achieve the purpose of retaining the mainstream velocity component. When the flow velocity is high, k can be taken in the range of (4h, 6h) to achieve the purpose of retaining the high-frequency turbulent structure. h is a constant value.

[0025] The velocity field data obtained after denoising using singular value decomposition is the matrix , according to the first k principal components retained, the horizontal velocity component u and vertical velocity component v of each pixel are reconstructed, and the discrete velocity data points are converted into a continuous velocity field distribution map through interpolation. The reconstructed velocity field is compared with the expected fluid dynamics behavior. The effect of singular value decomposition denoising on velocity field reconstruction is evaluated, and the difference in velocity field before and after denoising is compared, including the calculation of absolute and relative errors of velocity field reconstruction. The experimental environment is optimized by adjusting the laser intensity, camera frame rate and pulse interval, and the data processing algorithm is optimized by adjusting the window size of the cross-correlation method, thereby reducing the error.

[0026] Embodiment 2: Collecting multiple frames of particle images without blur and with normal noise, the specific steps are as follows: A high-speed camera is used to collect a sequence of moving particle images. Each frame of the image is divided into an area of ​​interest containing particles and a background area without particles. An image segmentation algorithm is used to automatically identify the particle aggregation area, and a 10-15% border area around the image is set as the background analysis area. The modulation transfer function of the particle aggregation area is calculated, and 20-50 typical particles are selected. The modulation transfer function value is calculated by the edge diffusion function. The modulation transfer function curve is constructed using the cubic spline interpolation method. The modulation transfer function value at a spatial frequency of 1 / 2 of the Nyquist frequency is taken as the evaluation index, and the modulation transfer function threshold is set to 0.2. When the measured modulation transfer function is ≤0.2, a fuzzy warning is triggered. An 8-bit grayscale statistical analysis is performed in the particle-free background area, and its grayscale standard deviation is calculated. The grayscale standard deviation threshold is set to 10 (corresponding to a signal-to-noise ratio SNR≥30dB). When the grayscale standard deviation is greater than or equal to 10, it is judged that the noise is too high and noise reduction processing is required. After iterative optimization, 50-100 frames of image data with a modulation transfer function>0.2 and a grayscale standard deviation<10 are continuously collected.

[0027] Example 3: The experimental environment is optimized by adjusting the laser intensity, camera frame rate and pulse interval, and the data processing algorithm is optimized by adjusting the window size of the cross-correlation method. The specific steps are as follows: The reconstructed matrix after denoising , converted into a continuous velocity field distribution map through bicubic spline interpolation, and the absolute error AE and relative error RE are defined to evaluate the noise reduction effect. The specific formula of absolute error AE is as follows: ; Among them, i represents the i-th pixel, N represents a total of N pixels, u(i) represents the horizontal velocity component of the i-th pixel, v(i) represents the vertical velocity component of the i-th pixel, ur represents the horizontal velocity component of the theoretical model, and vr represents the vertical velocity component of the theoretical model.

[0028] ; Among them, Mr represents the reference matrix of the theoretical model.

[0029] The experimental area was divided into high-resolution grids (2048×2048 pixels). For each grid node, the interpolation weight was calculated based on the velocity values ​​of the four adjacent discrete points to generate smooth streamlines and vector diagrams, and the velocity field differences before and after noise reduction were compared. The above operation was repeated after adjusting the original experimental laser intensity, camera frame rate and pulse interval to reduce the absolute error and relative error before and after the velocity field reconstruction. In this embodiment, the pulse laser energy was increased from 150 mJ to 300 mJ, the particle image signal-to-noise ratio (SNR) was increased by 20%, the frame rate was adjusted from 2 kHz to 8 kHz, and the pulse interval was shortened to 25 μs to ensure that the particle displacement did not exceed 1 / 4 of the cross-correlation window. The experimental results showed that the absolute error of the velocity field after adjustment was reduced by about 38%, and the relative error was reduced from 9.5% to 2.8%, thereby optimizing the experimental environment. By comparing the errors of 16×16, 24×24, and 32×32 pixel windows, it is determined that the 16×16 window can improve the spatial resolution by 15% in turbulent boundary layer measurements, while the cross-correlation peak signal-to-noise ratio reaches 42 dB.

[0030] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0031] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A high-precision particle image velocimetry method, characterized in that: include: Step 1: Build a high-precision particle image measurement experimental platform and use a time-synchronized laser-camera system to capture continuous particle images; The FPGA synchronization controller is used to achieve precise synchronization between laser pulses and camera exposure, and the synchronization accuracy is verified; Step 2: Process the displacement image captured by the high-speed camera using a combination of cross-correlation method and optical flow method; The displacement image is preliminarily processed, the normalized cross-correlation function is calculated, the velocity vector is solved iteratively through the optical flow equation, and the sub-pixel interpolation technology is introduced to improve the displacement resolution; Step 3: Introduce singular value decomposition in the PIV data processing process to decompose the flow field data and separate the main signal from the noise; The refined velocity field data is decomposed by singular value decomposition, the low-order singular values ​​are truncated, and the denoised velocity field matrix is ​​reconstructed using the retained singular values ​​and the corresponding singular vectors.

2. A high-precision particle image velocimetry method according to claim 1, characterized in that: A high-precision particle image measurement experimental platform was built, and a time-synchronized laser-camera system was used to capture continuous particle images. The specific method is as follows: A high-precision particle image measurement experimental platform with a time-synchronized laser-camera system was built. The platform used a three-dimensional coordinate frame with a positioning uncertainty in the range of (-a, a) for high-precision positioning, and a CMOS camera with a resolution of A×B pixels and a frame rate ≥h as the image capture device. A high-energy dual-pulse laser was selected, and a synchronization signal was generated through an FPGA synchronization controller to achieve precise synchronization between the laser pulse trigger signal and the CMOS camera exposure signal, collect dynamic particle images, and use a high-speed camera to capture displacement images.

3. A high-precision particle image velocimetry method according to claim 2, characterized in that: The synchronization signal is generated by the FPGA synchronization controller to achieve precise synchronization between the laser pulse trigger signal and the CMOS camera exposure signal. The specific method is as follows: A TTL pulse signal was sent to simultaneously trigger the laser double pulse emission and camera exposure. The laser pulse interval Δt was dynamically adjusted according to the preset flow rate, and the synchronization error was monitored by an oscilloscope to be ≤1 microsecond. When it was detected that the particle displacement between adjacent frames exceeded 50% of the window size, Δt was shortened to 50% of the original value and the camera frame rate was increased synchronously.

4. A high-precision particle image velocimetry method according to claim 1, characterized in that: The displacement image is preliminarily processed as follows: The quality of the collected image is verified by calculating the modulation transfer function of the particles in the displacement image and the grayscale standard deviation of the particle-free area of ​​the image. The thresholds of the modulation transfer function and the grayscale standard deviation are preset. If the modulation transfer function value is less than or equal to the modulation transfer function threshold, it is judged that there is blur, otherwise, it is judged that there is no blur. If the grayscale standard deviation is greater than or equal to the grayscale standard deviation threshold, it is judged that the noise is too high, otherwise, it is judged that the noise is normal. Multi-frame particle images with no blur and normal noise are collected, and the denoising autoencoder is used to remove background noise from the image.

5. A high-precision particle image velocimetry method according to claim 1, characterized in that: Calculate the normalized cross-correlation function as follows: After the preliminary processing, two consecutive particle images are divided into multiple cross-correlation windows of n*n pixels. The normalized cross-correlation function between the two frames of images is calculated for each window. The formula is used represents the normalized cross-correlation function between two frames of images, where represents the candidate displacement value between two frames of images, (x, y) represents the coordinates of the image pixels in the window, Represents the grayscale value of the image pixel at the (x, y) coordinate in the first frame image, Indicates the corresponding coordinate translation in the second frame image The gray value after Represents the grayscale mean of the first frame image in the selected window, Represents the grayscale mean of the second frame image in the selected window; By calculating the normalized cross-correlation function value under different displacements , find The candidate displacement value that reaches the maximum value is recorded as the best estimated displacement , output the best estimated displacement.

6. A high-precision particle image velocimetry method according to claim 1, characterized in that: The velocity vector is solved by iteratively solving the optical flow equation, and the sub-pixel interpolation technology is introduced to improve the displacement resolution. The specific method is as follows: The best estimated displacement output by the cross-correlation method is used as the initial value of the iterative optimization of the optical flow method. For each pixel in the image, a window of a fixed size is selected. Within the selected window, the gradients of the image in the x and y directions are calculated. The optical flow equation of the Lucas-Kanade algorithm is used in combination with the initial value to iteratively solve the velocity vector pixel by pixel. During the iterative process, the velocity vector estimate is continuously updated until the preset maximum number of iterations is reached. After obtaining the velocity vector at the integer pixel level, the sub-pixel interpolation technology is introduced to improve the displacement resolution. Through bicubic interpolation, the displacement resolution is increased from the integer pixel level to the m pixel level.

7. A high-precision particle image velocimetry method according to claim 1, characterized in that: The refined velocity field data is decomposed by singular value decomposition. The specific method is as follows: Receive the refined velocity field data generated by the optical flow method, including the horizontal velocity component u and the vertical velocity component v of each pixel point, and arrange the original velocity field data obtained by the optical flow method into a two-dimensional matrix according to the time and space dimensions, denoted as matrix M, where the row vector of the M matrix is ​​defined by the spatial coordinate point, that is, each row corresponds to a spatial coordinate point, the total number of rows is the number of all spatial coordinate points, and the total number of rows is represented by N. The column vector of the M matrix corresponds to the velocity vector of different time frames, the total number of columns is the length of the time series, and the total number of columns is represented by Q. Zero mean processing is performed on the matrix M to eliminate the influence of dimensional differences on the decomposition results. The formula is used Perform singular value decomposition on the matrix M, where U is the left singular vector matrix of the M matrix, with an order of N*N. Each column of the left singular vector matrix U represents a spatial modal distribution of the velocity field, reflecting the different characteristics or modes of the velocity field in space. It represents the singular value diagonal matrix of the M matrix, with an order of N*Q. Its diagonal elements are the singular values ​​of the M matrix. Larger singular values ​​correspond to modes with larger energy contributions. V is the right singular vector matrix of the M matrix, with an order of Q*Q. Each column represents the changing characteristics of the velocity field in the time dimension.

8. A high-precision particle image velocimetry method according to claim 1, characterized in that: The low-order singular values ​​are truncated, and the denoised velocity field matrix is ​​reconstructed using the retained singular values ​​and the corresponding singular vectors. The specific method is: Using the formula represents the matrix after truncated low-order singular values ​​reconstruction and denoising, where To retain the submatrix of the left singular vector matrix after the first k principal components, To retain the submatrix of the singular value diagonal matrix after the first k principal components, In order to retain the submatrix of the right singular vector matrix after the first k principal components, the velocity field data obtained after denoising by singular value decomposition is the matrix , based on the first k principal components retained, the horizontal velocity component u and vertical velocity component v of each pixel are reconstructed.

Images