A method for extracting wave surface profile intelligently in the evolution process of breaking wave

Abstract

The application discloses a wave surface profile intelligent extraction method for broken wave evolution process and belongs to the technical field of image processing of hydraulic model tests. A target frame positioning area is arranged in front of a structure, a target frame is determined by using image sequence normalized weighted gray features and a baseline threshold value, and a representative image before breaking is selected according to a fixed frame interval. The wave surface identification area is subjected to median filtering, fixed threshold value binarization and open operation, a water body candidate area graph is generated, a connected air area is reconstructed based on a top starting area, a wave surface boundary is extracted according to the four-neighborhood relationship between the air area and the water body candidate area, the boundary is restored to original image coordinates, the coordinates are calibrated and converted into unified actual coordinates through piecewise cubic Hermite shape-preserving interpolation, and the identified images, wave surface coordinates and representative frame wave surface graphs are saved in batches, so that the time reference of repeated tests is unified, and boundary misidentification caused by particle highlights, reflections, foam holes and dark areas is reduced.

Description

Technical Field

[0001] This invention relates to the field of image processing technology for hydraulic model tests, and more specifically, to an intelligent method for extracting wave surface contours for the evolution of breaking waves. Background Technology

[0002] In physical model experiments of breaking waves, high-speed cameras are typically used to continuously record wave propagation, steepening, and changes in the free surface before impacting the structure, and wavefront profiles are obtained through image processing. For studies involving numerous repetitive experiments, manually searching for target times and depicting the wavefront frame by frame is not only labor-intensive, but also prone to inconsistencies in the time references and frame intervals used in different experiments, affecting the comparability of wavefront morphologies.

[0003] Existing thresholding methods typically identify water regions based on brightness differences and then directly extract the upper boundary of the bright area or the boundary of a larger connected region. However, broken wave images simultaneously contain tracer particle bright spots, local reflections, foam, water droplets, and dark pores within the water body. When these local features appear simultaneously with the actual water-air interface, directly extracting the bright area boundary can easily lead to isolated bright spots, foam pores, or internal dark areas being mistaken for wavefronts.

[0004] Therefore, a wavefront contour extraction method corresponding to the actual high-speed image processing flow is needed: first, automatically determine the unified target frame in each test where the wave is about to impact the structure, and then extract the image before the breakage according to a fixed relative frame sequence; during the wavefront recognition process, use the air region connected to the upper boundary of the image to constrain the water-air interface, and complete coordinate calibration, interpolation and batch saving, thereby reducing the differences caused by manual frame selection and parameter adjustment each time.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To overcome the problems in existing technologies, such as inconsistent target times during repeated experiments, misidentification of wavefronts caused by local bright spots and internal dark areas, and difficulty in batch processing identification results, this invention provides an intelligent wavefront contour extraction method for the evolution process of broken waves. The method consists of a continuous processing flow, which includes target frame positioning, fixed relative frame taking, extraction of top connected air boundaries, coordinate transformation, and batch output.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent method for extracting wavefront profiles for the evolution of breaking waves, comprising the following steps:

[0008] Step S1: Read the high-speed image sequence, set the target frame localization region in front of the structure, and calculate the normalized weighted grayscale features of each frame. The baseline average is calculated from the baseline features of the first 10% of the image sequence, with a minimum of 5 frames. and fluctuation range ,according to Establish candidate thresholds, identify the first consecutive threshold-exceeding segments, and determine the target frame;

[0009] Step S2: Using the target frame as the time reference, select representative images before fragmentation at fixed frame intervals; extract the wavefront to identify the target region, and sequentially perform median filtering, fixed threshold binarization, and opening operations to obtain the water body candidate region map W; starting from the top of the target region, ... Morphological reconstruction was performed to obtain the top connected air region. And extract the set of wavefront boundary points F based on the four-neighbor adjacency relationship;

[0010] Step S3: Restore F to the original image pixel coordinates, take the minimum vertical coordinate for the boundary points with the same horizontal pixel coordinates to form a single-value wavefront coordinate sequence; establish a linear transformation relationship between pixel coordinates and actual coordinates based on the calibration points, and use piecewise cubic Hermite conformal interpolation to obtain the wavefront curve on the unified horizontal coordinates.

[0011] Step S4: Process the image sequence in batches for each experiment, and save the recognition overlay image, wavefront coordinates of each frame, wavefront scatter plot at a fixed relative time, and summary data.

[0012] In a preferred embodiment, the column weights in step S1 increase as the pixel position approaches the structure, and the 75th percentile is taken for the weighted grayscale response of each row and column; after the image sequence features are normalized, 5-frame sliding median processing is used.

[0013] In a preferred embodiment, the threshold amplification factor k is 4.0, and the number of consecutive over-threshold confirmation frames is 2; the target frame is obtained by compensating 4 frames backward from the feature surge point and shifting 1 frame forward.

[0014] In a preferred embodiment, six representative moments are selected at 5-frame intervals from the target frame, and wavefront contour extraction is performed on all images from the earliest representative frame to the target frame.

[0015] In a preferred embodiment, the median filtering uses a 21×21 pixel window, the fixed pixel intensity threshold is 25, the opening operation uses a 5×5 pixel square structuring element, and the starting region of the top connected air region is the top 5 rows of the wavefront recognition target region.

[0016] In a preferred embodiment, at least two calibration points with known actual coordinates are selected during calibration, and the horizontal and vertical scaling factors and translation amounts are calculated respectively; the unit of the wave surface curve after coordinate transformation is mm.

[0017] The technical effects and advantages of this invention are as follows:

[0018] This invention utilizes a sudden increase in local grayscale in front of a structure to automatically determine the target frame and extracts the image before breakage according to a fixed relative frame sequence, providing a unified time reference for different repeated experiments. Median filtering, fixed threshold binarization, and opening operations reduce interference from particle bright spots, water droplets, and local reflections. The four-neighbor relationship between the top connected air region and the water candidate region is used to determine the water-air interface, eliminating internal holes and local dark areas in the water body that are not usually connected to the top of the image. Coordinate calibration, segmented cubic Hermite shape-preserving interpolation, and batch output convert the recognition results into directly comparable and statistically significant actual coordinate data, reducing processing differences caused by manual frame-by-frame selection and repeated parameter tuning. For extreme wave conditions such as high wave height and dense foam, the entire set of processing parameters can be flexibly adjusted. When there is excessive foam, the filter window size can be appropriately enlarged, the binarization brightness threshold increased, and the morphological opening operation enhanced to remove interference from fine foam fragments. Under high wave height conditions, the grayscale threshold amplification coefficient can be increased to improve the target frame recognition sensitivity, ensuring stable and reliable wavefront extraction in various experimental scenarios. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the implementation of an intelligent wavefront profile extraction method for the evolution of breaking waves according to the present invention.

[0020] Figure 2 This is a schematic diagram illustrating the steps of an intelligent wavefront profile extraction method for the evolution process of breaking waves according to the present invention.

[0021] Figure 3 This is a diagram showing the wavefront identification process and results. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention combines target frame localization, fixed relative frame extraction, top connected air boundary extraction, coordinate transformation, and batch output into a complete processing flow, as follows: Figure 1 and Figure 2 As shown, the single-frame wavefront identification process and results are as follows: Figure 3 As shown.

[0024] Example 1

[0025] Please see Figures 1 to 3A method for intelligent extraction of wavefront profiles for the evolution of breaking waves is presented, and the specific operation process is as follows.

[0026] Step S1: Automatically locate the target frame and establish a unified relative time reference.

[0027] High-speed image sequences from the same test folder are read, and a narrow strip-shaped target frame localization region is set in front of the structure facing the wave. This region is used to sense the local brightness increase caused by waves, foam, and splash entering the vicinity of the structure, and does not participate in the subsequent boundary extraction of the entire wavefront.

[0028] Perform grayscale conversion on the t-th frame image, converting the grayscale values ​​to the range of 0 to 1. Assume the target frame location region has a total of... The column-weighted grayscale response of the v-th row is:

[0029]

[0030] in, Locate pixels within the target frame in frame t. Normalized grayscale value; Let be the weight of column u. When the structure is located to the right of the target frame's localization region, the column weight is:

[0031]

[0032] This weighting makes pixels closer to the structure contribute more to the target frame localization. For each row within the target frame localization region... The 75th percentile is used to obtain the original grayscale features of the frame. This highlights rows with significant brightness variations and reduces the impact of a few dark rows on the results.

[0033] Image sequence Normalization is performed using the minimum and maximum values, followed by sliding median processing with a window length of 5 frames to obtain normalized weighted grayscale features. :

[0034]

[0035] in, This indicates a 5-frame sliding median processing method. This processing is used to suppress short-term abrupt changes caused by bright spots in a single frame, while preserving the continuous brightness increase that occurs when the wave enters the target frame's positioning area.

[0036] The baseline frame count is determined from the first 10% of the image sequence, and must be no less than 5 frames:

[0037]

[0038] In the formula, N is the total number of frames in the image sequence. Baseline frame count. (Using the previous...) Frames Calculate the baseline average and baseline fluctuation range :

[0039]

[0040] Candidate threshold for:

[0041]

[0042] Where k is the threshold amplification factor, which is 4.0 in this embodiment. The program starts searching after the baseline segment, when... More than 2 consecutive frames The first frame of the continuous segment is identified as the starting point of the feature surge. The search range is limited to the first 70 frames of the image sequence to avoid further brightening caused by splashes and reflections at later times.

[0043] It should be noted that the search for the starting point of the sudden increase in grayscale features is limited to the first 70 frames of the image sequence to avoid misjudgment caused by repeated brightening of water splashes and reflections in the later stages of the experiment. This search frame range can be flexibly adjusted according to the total duration of the waves and the camera's frame rate, and only needs to match the normal occurrence period of wave impact on the structure.

[0044] The target frame is obtained by compensating for 4 frames backward from the characteristic rise point and then shifting forward by 1 frame. This target frame is shifted 3 frames backward relative to the characteristic rise point. This target frame corresponds to the moment when the wave is about to contact the structure and the local brightness has stabilized and increased.

[0045] By target frame number Based on this, m representative moments are selected at fixed frame intervals:

[0046]

[0047] In this embodiment, m is set to 6. Taking 5 frames, the 6 representative moments are as follows: , , , , and The camera sampling frequency was 100Hz, and the time interval between adjacent representative moments was 0.05s. The same frame interval and number of representative moments were used in different repeated experiments.

[0048] This method is compatible with both grayscale single-channel image and color multi-channel image acquisition scenarios. When acquiring grayscale images, the entire set of filtering, binarization, and morphological operation parameters can be directly reused without additional modifications. When acquiring color images, image processing can be performed on any color channel individually, and the channel with the best effect can be selected to construct water body candidate regions. Alternatively, the preprocessing results of multiple channels can be fused and superimposed to comprehensively determine the water body region, adapting to different camera imaging output formats.

[0049] Step S2: Extract the wavefront boundaries of each processing frame.

[0050] In the first set of valid experiments, a wavefront identification target region containing the main free surface changes is selected on the target frame image, and the same region is reused in subsequent experiments. The program performs wavefront identification on all images from the earliest representative frame to the target frame and extracts the above 6 fixed relative times for wavefront evolution comparison.

[0051] In this embodiment, the image is saved in RGB three-channel format, with the same pixel value at the corresponding pixel position in each of the three channels. The program processes each channel with the same parameters and selects the third channel after the opening operation to form a candidate water area map; since the values ​​of the three channels are the same, the selection of this channel does not change the binary result.

[0052] First, median filtering is applied to the wavefront identification target region. The program specifies a window side length of 20, but to meet the requirement of using an odd-numbered side length for the median filtering window, it is automatically adjusted to 21 during actual execution. Therefore, the actual filtering window is 21×21 pixels.

[0053]

[0054] in, and , representing the pixel intensity before and after filtering; u and v are the horizontal and vertical pixel coordinates within the target area for wavefront recognition; c is the color channel; In pixels A 21×21 pixel neighborhood centered on the center; This represents any pixel location within the neighborhood. Median filtering is used to reduce tracer particle bright spots, small water droplets, and isolated noise, while preserving the water-air interface location.

[0055] The image after median filtering is binarized using a fixed threshold:

[0056]

[0057] Where B(u,v,c) is the binarization result. For indicator functions; when The value is 1 if the condition is met, and 0 otherwise. In this embodiment, the pixel intensity range is 0 to 255, and the fixed threshold τ is 25. This value is determined by the pixel intensity difference test between the low-brightness air background and the bright areas of water and foam in this batch of images.

[0058] Then, an opening operation is performed on the binarized result using a 5×5 pixel square structuring element:

[0059]

[0060] Here, Open represents the opening operation of erosion followed by dilation. It is a square structural element with a side length of 5 pixels. This is the result of the opening operation on the c-th color channel. The opening operation is used to remove particle bright spots, small water droplets, local reflections, and scattered foam fragments that cannot form a main area.

[0061] The water body candidate region map W is formed by selecting processing channels from the opening operation results. This indicates that pixel (u,v) belongs to the candidate region of bright water or foam. This indicates that the pixel belongs to a non-water body candidate region.

[0062] To preserve the air above the true wavefront from the non-water candidate region, W is inverted, and morphological reconstruction is performed using the non-water candidate pixels in the top 5 rows of the wavefront recognition target region as the starting region:

[0063]

[0064] in, The result of inverting the binary value of W; The non-water body starting region in the top 5 rows; Recon indicates from Departure, at Retain pixels connected to it within the range; This represents the top connected air region. The connected path stops when it encounters a water candidate region with W=1, therefore dark holes and foam gaps inside the water that are not connected to the top of the image will not be entered. .

[0065] Finally, extract the pixels in the top connected air region that are adjacent to the four neighbors of the water candidate region:

[0066]

[0067] in, W represents the set of adjacent pixels in the top, bottom, left, and right directions for pixels with a value of 1, excluding diagonal directions; F is the set of wavefront boundary points. This boundary is located on one side of the air region connected to the top of the image and adjacent to the water candidate region, corresponding to... Figure 3The wavefront boundary and wavefront identification results in the image.

[0068] Step S3: Convert the pixel wavefront into an actual coordinate wavefront.

[0069] Add the initial offset of the wavefront-identified target region in the original image to the local pixel coordinates in F to restore the original image pixel coordinates. For cases where the same horizontal pixel coordinate u may have multiple boundary points, take the minimum vertical pixel coordinate for each column.

[0070]

[0071] This yields a single-value wavefront sequence where each horizontal pixel coordinate corresponds to a vertical pixel coordinate. The program retains the original boundary scatter points in F for identification of the result image and representative frame scatter plot, while using the single-value wavefront sequence for subsequent interpolation.

[0072] During coordinate calibration, select at least two calibration points with known actual coordinates in the reference image, and input the actual horizontal and vertical coordinates of each calibration point. The program then establishes the linear transformation relationship between the horizontal and vertical axes:

[0073]

[0074] Where u and v are pixel coordinates, and X and Y are actual coordinates. , These are the horizontal and vertical scaling factors. , This represents the translation amount. When using two calibration points, the scaling factor can be directly determined from the coordinate difference between the two points; when using more than two calibration points, the transformation parameter is calculated using linear least squares. In this embodiment, mm is used as the actual coordinate unit.

[0075] After converting the single-value wavefront points to actual coordinates, piecewise cubic Hermite conformal interpolation is performed using MATLAB's pchip function to map the wavefront in each frame to the actual horizontal coordinates corresponding to all horizontal pixel columns in the wavefront identification target region. This interpolation is used to obtain the wavefront height on a uniform horizontal coordinate without changing the local variation trend of the original wavefront.

[0076] Step S4: Save the processing results in batches.

[0077] The program processes all images from the earliest representative frame to the target frame experimentally, and saves the identification and inspection map of the superimposed wavefront boundary points for each processed frame. The wavefront curves of each frame on a unified horizontal coordinate are written to the AllWaves worksheet in an Excel file, and the coordinates of the original wavefront scatter points at six representative times are written to the Selected_Frames_Coords worksheet.

[0078] The program simultaneously generates wavefront scatter plots at six fixed representative times and copies these plots to a summary folder. If an image file is missing, it is recorded and skipped without replacing it with preceding or following frames or altering the fixed relative frame sequence of other repeated experiments, thus maintaining temporal comparability between different experiments.

[0079] Through the above processing, we can obtain wavefront results with uniform relative time, uniform actual coordinates and uniform data format, which can be used for wavefront superposition, average wavefront and dispersion analysis in subsequent repeated experiments.

[0080] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0081] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0082] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0083] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0084] The above description of the disclosed embodiments will enable those skilled in the art to make or use various modifications to these embodiments. It will be readily apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent extraction of wavefront profiles for broken wave evolution, characterized in that, Includes the following steps: Step S1: Read the high-speed image sequence of the same experiment, set the target frame localization region in front of the structure, and calculate the normalized weighted grayscale features of the target frame localization region in each frame. Candidate thresholds are established based on the baseline features of the preceding segment of the image sequence. Feature segments that first continuously exceed the candidate thresholds are located, and the target frame is determined accordingly. Step S2: Using the target frame as a unified time reference, select representative images before fragmentation according to fixed frame intervals; Median filtering, fixed-threshold binarization, and opening operations are sequentially performed on the wavefront recognition target regions in each representative image to form a water body candidate region map W. After inverting W, morphological reconstruction is performed from the top starting region of the wavefront recognition target region to obtain the top connected air region. ;extract The pixels that are adjacent to W in their four neighborhoods form the set of wavefront boundary points F; In step S2, each color channel of the wavefront recognition target region is filtered using a 21×21 pixel median window, binarized using a pixel intensity threshold τ, and opened using a 5×5 pixel square structuring element. in, and , where c represents the pixel intensity before and after filtering, τ is set to 25, and B is the binarization result. It is a square structural element with a side length of 5 pixels. The c-th color channel is the binary result after the opening operation; the processed channels are selected to form the water body candidate region map W; Represents pixels This area is a candidate region for bright areas in water or foam. This indicates that the pixel belongs to a non-water body candidate region; the top is connected to the air region. The set of wavefront boundary points F is as follows: in, Recon indicates the non-water body starting region in the preset row at the top of the target area for wavefront identification. Morphological reconstruction within the scope, This represents the set of adjacent pixels in the four directions of top, bottom, left, and right for the candidate water body region; the top preset row is the top 5 rows; Step S3: Restore the set of wavefront boundary points F to the original image pixel coordinates, take the minimum vertical pixel coordinate for boundary points with the same horizontal pixel coordinates to form a single-value wavefront coordinate sequence; establish a coordinate transformation relationship between the pixel coordinates of the calibration points and the actual coordinates, convert the single-value wavefront coordinate sequence into the actual coordinates, and use piecewise cubic Hermite conformal interpolation to obtain the wavefront curve on the unified horizontal coordinates. In step S3, boundary points with the same horizontal pixel coordinates in F are grouped together, and the smallest vertical pixel coordinate in each group is taken as the single-value wavefront point of the corresponding column: Add the horizontal and vertical initial offsets of the wavefront recognition target region in the original image to the single-value wavefront point to restore it to the pixel coordinates of the original image; Step S4: Process the image sequence in batches according to steps S1 to S3, and save the wavefront boundary overlay map, the wavefront coordinates of each processed frame, the wavefront scatter plot of the representative frame, and the summary results. In step S4, the wavefront curves on the unified horizontal coordinates of each processing frame are written into the first data table, the original wavefront scatter coordinates at a fixed relative time are written into the second data table, and the wavefront boundary overlay map and the wavefront scatter map at a fixed relative time are generated for each processing frame; missing images are only recorded and skipped without changing the fixed relative frame sequence of other experiments.

2. The intelligent wavefront contour extraction method according to claim 1, characterized in that, In step S1, the grayscale values ​​of the target frame positioning area are normalized to 0 to 1; when the structure is located to the right of the target frame positioning area, the column-weighted grayscale response of the v-th row of the t-th frame is... and the original grayscale features of the frame for: in, The number of columns for locating the target frame region. For the t-th frame pixels The normalized gray value, This indicates that the 75th percentile of the weighted grayscale response for each row and column is used, and the column weights are... for: Original grayscale features of each frame Normalization is performed based on the minimum and maximum values ​​of the image sequence, and then 5-frame moving median processing is used to obtain normalized weighted grayscale features. .

3. The intelligent wavefront contour extraction method according to claim 2, characterized in that, In step S1, the number of baseline frames Baseline average Baseline fluctuation range and candidate threshold They are respectively: Where N is the number of frames in the image sequence, and k is the threshold magnification factor, which is set to 4.0; from the 1st... Starting from frame +1, the condition first appears after at least two consecutive frames. When a segment is defined, the first frame of that segment is recorded as the starting point of the feature surge; based on the starting point of the feature surge, 4 frames are compensated backward and 1 frame is shifted forward to obtain the target frame.

4. The intelligent wavefront contour extraction method according to claim 1, characterized in that, In step S2, starting from the target frame and proceeding forward at fixed frame intervals... Select m representative times, and the frame number of the i-th representative time. for: in, Number the target frame; m is 6. Take 5 frames and perform wavefront boundary extraction on all images from the earliest representative frame to the target frame.

5. The intelligent wavefront contour extraction method according to claim 1, characterized in that, In step S3, at least two calibration points with known actual coordinates are selected, and horizontal and vertical linear coordinate transformation relationships are established respectively: Where u and v are the image pixel coordinates, and X and Y are the actual coordinates. , This is the proportionality coefficient. , The translation amount is calculated based on the actual coordinates obtained from the calibration points. The single-value wavefront coordinates are then converted into actual coordinates and mapped to a unified transverse coordinate sequence using piecewise cubic Hermite conformal interpolation.

Images