Wave breaking physical model test method based on target detection technology

Abstract

The invention relates to the technical field of ocean testing, in particular to a wave breaking physical model test method based on a target detection technology, which comprises the following steps: setting two coordinate origins and a plurality of irregular coordinate registration points in a test water tank, carrying out wave tests with different wave heights and periods, and recording wave propagation and breaking processes by using a high-definition camera. Training a deep learning model by using a marked water surface line, a white cap and a broken form, and automatically identifying wave features by using the trained model; a mapping model of image pixel coordinates and actual physical coordinates is established, the image is corrected through an interpolation method, and it is ensured that the wave feature position is accurately matched with the actual position. Segmenting a high-definition picture into sub-pictures, and identifying the characteristics of a water surface line, a spindrift surface and a fracture zone by using the training model; and processing the identification result, and extracting the space and time characteristics of the water surface line and the position and range of the fracture zone. According to the invention, positioning precision can be maintained, and observation interference can be reduced.

Description

Technical Field

[0001] The invention relates to the field of ocean testing technology, in particular to a wave breaking object model testing method based on target detection technology. Background Art

[0002] In nuclear power water intake projects, traditional open channel water intake is gradually replaced by box culvert structures due to the influence of waves and floating objects. However, when the box culvert is located on the seabed, the wave breaking energy is large and it is easy to disturb the sediment, threatening the safety of water intake. The existing physical model test locates the wave breaking position through dense grids. Although the accuracy is high, too many grid lines will block the camera observation (see Figure 1 ), and rely on manual judgment of wave breaking patterns. Numerical simulation relies on empirical parameter adjustment and lacks universality.

[0003] Therefore, there is an urgent need for an experimental method that can maintain positioning accuracy and reduce observation interference. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a wave breaking physical model test method based on target detection technology. The present invention combines target detection technology with physical model experiments, and realizes accurate identification of wave breaking positions by improving physical model layout, image feature extraction and coordinate registration methods, while reducing the interference of grids on observations, providing convenience for wave breaking mechanism research and engineering applications.

[0005] In order to achieve the above technical effects, the present invention proposes a wave breaking object model test method based on target detection technology, comprising the following steps: Step 1: Implementation of physical model test: Two coordinate origins and multiple registration points were set in the test water tank; wave conditions with different wave heights and periods were selected for testing, and image data of wave propagation and breaking process were continuously captured by high-definition cameras; Step 2: Feature annotation and extraction implementation: Using image processing software to pre-process the image data in step 1 to highlight the water surface line and wave features; using a target detection algorithm to automatically annotate the pre-processed image, identify the wave breaking type, water surface line and wave surface features, and use the annotated data to train a deep learning model; Step 3: Position feature extraction and registration implementation Based on the position information of sparse registration points, a mapping model between image pixel coordinates and actual physical coordinates is established; the image is corrected by interpolation method to achieve accurate matching between wave feature positions and actual positions; Step 4: Image segmentation and feature processing: The acquired high-definition images are evenly divided into sub-images, and the water surface line, wave surface and broken zone features in the sub-images are identified through the trained model; the identification results are processed and statistically analyzed to extract the spatial and temporal features of the water surface line and the location and range of the broken zone.

[0006] As a preferred technical solution, the arrangement of the coordinate origin and the registration points in step 1 is positioned using the compass method, and the registration points are sparsely distributed in the test water tank to reduce occlusion of camera observation.

[0007] As a preferred technical solution, the target detection algorithm in step 2 is the YOLO algorithm, the annotated wave breaking types include collapsing waves, rolling waves and breaking waves, and the annotated data is used to train the model to automatically identify wave breaking characteristics.

[0008] As a preferred technical solution, in step 3, the interpolation method is bilinear interpolation, and the specific steps are: 1) linearly interpolate adjacent registration points in the x direction to obtain the coordinate values ​​of the intermediate points R1 and R2; 2) perform secondary interpolation based on R1 and R2 in the y direction to obtain the actual physical coordinates of the target point P.

[0009] As a preferred technical solution, in step 1), the coordinate value of the middle point R1 is obtained by the following formula: ; The coordinate value of the middle point R2 is obtained by the following formula: ; Among them, Q 11 The horizontal axis is x 1 , the vertical coordinate is y 1 Q 21 The horizontal axis is x 2 , the vertical coordinate is y 1 Q 12 The horizontal axis is x 1 , the vertical coordinate is y 2 Q 22 The horizontal axis is x 2 , the vertical coordinate is y 2 point.

[0010] As a preferred technical solution, in step 2), the actual coordinates of the target point P are obtained by the following formula: ; Among them, Q 11 The horizontal axis is x 1 , the vertical coordinate is y 1 Q 21 The horizontal axis is x 2 , the vertical coordinate is y 1 Q12 The horizontal axis is x 1 , the vertical coordinate is y 2 Q 22 The horizontal axis is x 2 , the vertical coordinate is y 2 point.

[0011] As a preferred technical solution, in step 4, the threshold pixel spacing is set when the image is segmented, specifically: ; Where w is the preset image width, Dpi is the threshold pixel density, and i is the i-th image in the horizontal direction; ; Where h is the preset image height, Dpi is the threshold pixel, and j is the jth image in the vertical direction.

[0012] As a preferred technical solution, step 4 also includes: taking the average of the upper and lower layer heights of the wave surface noise data with a certain thickness as the wave height line, and using Fourier transform to filter the non-smooth wave surface curve.

[0013] As a preferred technical solution, the Fourier transform F(ω) formula is: ; The inverse Fourier transform is used to return from the frequency domain to the time domain, and its expression is:

[0014] In the above formula: f(t) is the original function; i is the imaginary unit, ω is the angular frequency, t is the time variable; the complex exponential function e −iωt is the weight.

[0015] As a preferred technical solution, step 4 also includes: obtaining time process data of a single point by extracting the water surface line; calculating the gradient change of the wave by calculating the changes of multiple different coordinate points in the same wave propagation process; the single wave gradient calculation formula is as follows:

[0016] In the formula, Represents the gradient of the function f(x,y), the partial derivative of the function f(x,y) with respect to the independent variable y.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention organically combines target detection technology, image processing technology, coordinate registration method, etc. to form a complete set of wave breaking position physical model test methods. Through the improved physical model test layout and target detection technology, it can more accurately determine the position of the wave breaking zone, track the wave gradient change, reduce the human error and camera observation interference in the traditional method, and improve the reliability of the test results. Specifically: 1) Design the origin and sparse registration points in the water tank: reduce the interference of the grid on camera observation, while maintaining positioning accuracy through relative position mapping; 2) Introducing the YOLO algorithm into wave breaking feature recognition to improve the efficiency of automatic labeling of water surface lines and wave surfaces; 3) Use bilinear interpolation optimization to solve the coordinate offset problem caused by image distortion and improve the accuracy of actual position matching; 4) Capture the breaking critical point through the change of wave propagation gradient and provide dynamic data support for engineering protection design. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of locating the wave breaking position by dense grids in a physical model test in the prior art; Figure 2 It is the physical model arrangement form in the test water tank of the present invention; Figure 3 Annotate schematic diagrams for point location features; Figure 4 It is a schematic diagram of bilinear interpolation; Figure 5 This is a schematic diagram of pixel position matching; Figure 6 Schematic diagram of cropping image segmentation; Figure 7 The water surface line and wave surface features obtained after identification; Figure 8 , Fig. 9 , Fig.10 The actual capture effect of three different wave surfaces; Fig.11 is the wave surface curves at n different frames obtained by frame-by-frame reasoning statistics, and the unit of the X-axis is relative distance; Fig.12 is the wave height process at a specified distance (time process of a single x-coordinate point); Fig.13 This is a statistical diagram of the wave-breaking zone. DETAILED DESCRIPTION

[0019] In order to facilitate the understanding of the present invention, the present invention will be described more comprehensively below, and preferred embodiments of the present invention are given. However, it should be understood that these embodiments are only used for more detailed description and should not be understood as limiting the present invention in any form, i.e., not intended to limit the scope of protection of the present invention.

[0020] like Figures 1 to 13 As shown, this embodiment discloses a wave breaking object model test method based on target detection technology, comprising the following steps: Step 1: Implementation of physical model test: like Figure 2 As shown, in the test tank, two points are marked as the origin of the coordinates, and multiple sparse points are used as the registration points. The compass method can be used to determine the location of the points, which is convenient for testing. Wave conditions with different wave heights and period combinations are selected for testing, and high-definition cameras are used for filming to ensure that clear and complete image data of wave propagation and breaking processes are obtained; Step 2: Feature annotation and extraction implementation: Image processing software is used to perform image enhancement, exposure, contrast adjustment and other pre-processing on the images obtained by the camera to highlight the features of the water surface line and breaking waves. The YOLO algorithm model is used to annotate the processed images, annotate key features such as wave breaking type, water surface line, and wave surface, and the annotated data is used to train the deep learning model, so that the model can accurately identify and annotate the following features: 1) Wave breaking characteristics There are three types of wave breaking: collapsing waves, breaking waves, and breaking waves.

[0021] Breaking wave: The first wave of the wave is white foam at the top of the wave crest. As the wave propagates forward, the foam on the top of the wave crest continues to be generated until it reaches the coast. (It occurs when the wave is steep and the bottom slope is small) Breaking wave: The leading edge of the wave crest first becomes steep, then curls into a tongue shape, the tongue-shaped wave crest gradually rolls down, then plunges into the water, breaks, and is accompanied by the entrainment of air. (Medium wave steepness, medium bottom slope) Breaking wave: The wave crest gradually becomes very asymmetrical, and then the front root of the wave crest begins to break, and then most of the front of the wave crest is in a very chaotic state of breaking and climbing up along the slope. (The wave is small and the bottom slope is large) 2) Wave stability characteristics The shapes of the crests and troughs are relatively fixed and regular, not as chaotic as when they break. The crests are usually smooth.

[0022] Step 3: Position feature extraction and registration implementation During the image recognition process, the identified pixel position must be aligned with the actual coordinate system in order to obtain its actual position. Due to the influence of visual effects, the pixels may be distorted during the camera process, resulting in the actual position and the pixel position no longer being in a fixed ratio, while the relative position relationship remains unchanged. For example, in Figure 3 In the figure, the relative position relationship between the two points (1 on the left, 2 on the bottom) and (5 on the left, 1 on the bottom) in the actual coordinate system is also maintained in the image. Using this relative position relationship, after identifying the position information of the registration point in the image, the distance between it and the actual registration point is matched to establish a mapping model between the image pixel coordinates and the actual physical coordinates: starting from the origin, the corresponding positions are placed clockwise according to the minimum pixel distance, and linear interpolation, neighboring, cubic interpolation and other interpolation methods are used to achieve accurate registration of the pixel position and the actual position, ensuring the accuracy of the wave feature position information.

[0023] Taking linear interpolation as an example, bilinear interpolation is performed in the xy direction to obtain the following calculation formula: 1) First, linearly interpolate the adjacent registration points in the x direction and obtain the coordinate value of the middle point R1 using the following formula: ; The coordinate value of the middle point R2 is obtained by the following formula: ; Among them, Q 11 The horizontal axis is x 1 , the vertical coordinate is y 1 Q 21 The horizontal axis is x 2 , the vertical coordinate is y 1 Q 12 The horizontal axis is x 1 , the vertical coordinate is y 2 Q 22 The horizontal axis is x 2 , the vertical coordinate is y 2 point.

[0024] 2) Perform secondary interpolation based on R1 and R2 in the y direction to obtain the actual physical coordinates of the target point P: ; FIG5 is a schematic diagram of bilinear interpolation.

[0025] Step 4: Image segmentation and feature processing: The acquired high-definition images are evenly segmented, and the appropriate threshold pixel distance is set to avoid the problem of low boundary recognition accuracy. The trained model is used to perform feature recognition on the segmented images to obtain surface feature information such as water surface lines and waves. The identified features are processed, such as calculating wave height lines and smoothing curves, and the spatial and temporal features of the water surface lines, as well as the location and range of the breaking zone, are statistically analyzed to accurately determine and analyze the wave breaking location.

[0026] Step 4.1: Figure 6 As shown in the figure, the acquired high-definition image is evenly divided into small-sized sub-images, and a suitable threshold pixel distance is set to reduce the GPU memory requirement; a certain threshold pixel space (width, height) is reserved between two images to avoid the problem of low boundary recognition accuracy.

[0027] ; Where w is the preset image width, Dpi is the threshold pixel density, and i is the i-th image in the horizontal direction; ; Where h is the preset image height, Dpi is the threshold pixel, and j is the jth image in the vertical direction.

[0028] Step 4.2: Use the trained model to perform feature recognition on the segmented image. The registration method is the same as step 3. Figure 7 As shown, the water surface line and wave surface features are obtained after recognition.

[0029] Process the identified features: 1) The identified wavefront is noisy and has a certain thickness; 2) Take the average of the upper and lower layer heights as the wave height line; 3) For the uneven areas, use Fourier transform to smooth the curve: ; The Fourier transform F(ω) is defined as: i-imaginary unit, ω-angular frequency, t-time variable; This integral represents the weighted sum of the original function f(t) at all time points, and the weight is the complex exponential function e −iωt The complex exponential function can be decomposed into sine and cosine functions, so the Fourier transform is essentially finding the amplitude and phase of the various frequency components in the original signal.

[0030] The inverse Fourier transform is used to return from the frequency domain to the time domain, and its expression is:

[0031] The original function can be reconstructed by integrating all frequency components.

[0032] Step 4.3: Count the spatial and temporal characteristics of the water surface line: Water surface line recognition effect Figures 8 to 10 (No registration was performed, the registration algorithm is an existing technology, and the implementation process can be found in ZL202311648914.4 "Map annotation recognition method, terminal and medium based on computer vision model").

[0033] like Fig.11 Through frame-by-frame inference statistics, n wave surface curves (spatial process) at different frames can be obtained.

[0034] like Fig.12 , the wave height process at a specified distance can be counted (single x-coordinate point time process).

[0035] Step 4.4: Breaking band calculation The breaking zone refers to the area where waves change shape as they propagate from deep water to shallow water, and eventually break near the coastline.

[0036] Wave breaking features can be directly identified by target detection technology. For objectivity and scalability, a gradient detection algorithm can be used for calculation: By extracting the water surface line, the time process data of a single point can be obtained; further, by calculating the changes of multiple different coordinate points in the same wave propagation process, the gradient change of the wave can be calculated. The calculation formula for a single wave gradient is as follows:

[0037] In the formula, Represents the gradient of the function f(x,y), the partial derivative of the function f(x,y) with respect to the independent variable y.

[0038] During the propagation of a wave, as long as it does not encounter resistance, the waveform will not change. Even if the wave encounters resistance, it will not break immediately, but will undergo a gradual change process. By calculating, we can get the uniform change process of the wave gradient under different X coordinates. Then, by aligning the peaks or troughs with different time points, we can depict the change of the wave gradient during the process of wave force or climbing.

[0039] It should be noted that the above contents are further detailed descriptions of the present invention in combination with specific implementation methods, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions; the dimensional data of this embodiment does not limit the technical solution, but only shows one of the specific working conditions. For ordinary technicians in the technical field to which the present invention belongs, several simple improvements and modifications can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A wave breaking object model test method based on target detection technology, characterized in that: The steps include: Step 1: Implementation of physical model test: Two coordinate origins and multiple irregular coordinate registration points were set in the test water tank; wave conditions with different wave heights and periods were selected for testing, and image data of wave propagation and breaking process were continuously captured by high-definition cameras; Step 2: Feature annotation and extraction implementation: Using image processing software to pre-process the image data in step 1 to highlight the water surface line and wave features; using a target detection algorithm to automatically annotate the pre-processed image, identify the wave breaking type, water surface line and wave surface features, and use the annotated data to train a deep learning model; Step 3: Position feature extraction and registration implementation Based on the position information of sparse registration points, a mapping model between image pixel coordinates and actual physical coordinates is established; the image is corrected by interpolation method to achieve accurate matching between wave feature positions and actual positions; Step 4: Image segmentation and feature processing: The acquired high-definition images are evenly divided into sub-images, and the water surface line, wave surface and broken zone features in the sub-images are identified through the trained model; The identification results are processed and statistically analyzed to extract the spatial and temporal characteristics of the water surface line and the location and range of the broken zone.

2. The wave breaking object model test method based on target detection technology according to claim 1 is characterized in that: The arrangement of the coordinate origin and the registration points in step 1 is positioned using the compass method, and the registration points are sparsely distributed in the test water tank to reduce occlusion of camera observation.

3. The wave breaking object model test method based on target detection technology according to claim 1 is characterized in that: The target detection algorithm described in step 2 is the YOLO algorithm. The annotated wave breaking types include collapsing waves, rolling waves and breaking waves. The annotated data is used to train the model to automatically identify wave breaking features.

4. The wave breaking object model test method based on target detection technology according to claim 1 is characterized in that: In step 3, the interpolation method is bilinear interpolation, and the specific steps are: 1) linearly interpolating adjacent registration points in the x direction to obtain the coordinate values ​​of the intermediate points R1 and R2; 2) performing secondary interpolation based on R1 and R2 in the y direction to obtain the actual physical coordinates of the target point P.

5. The wave breaking object model test method based on target detection technology as claimed in claim 4 is characterized in that: In step 1), the coordinate value of the middle point R1 is obtained by the following formula: ； Among them, Q 11 is a point with x1 as the horizontal coordinate and y1 as the vertical coordinate; Q 21 is a point with abscissa x2 and ordinate y1; The coordinate value of the middle point R2 is obtained by the following formula: ； Among them, Q 12 is a point with abscissa x1 and ordinate y2; Q 22 is a point with x2 as the horizontal coordinate and y2 as the vertical coordinate.

6. The wave breaking object model test method based on target detection technology according to claim 4 is characterized in that: In step 2), the actual coordinates of the target point P are obtained by the following formula: ； Among them, Q 11 is a point with x1 as the horizontal coordinate and y1 as the vertical coordinate; Q 21 is a point with abscissa x2 and ordinate y1; Q 12 is a point with abscissa x1 and ordinate y2; Q 22 is a point with x2 as the horizontal coordinate and y2 as the vertical coordinate.

7. The wave breaking object model test method based on target detection technology according to claim 1 is characterized in that: In step 4, the threshold pixel spacing is set when the image is segmented, specifically: ； Where w is the preset image width, Dpi is the threshold pixel density, and i is the i-th image in the horizontal direction; ； Where h is the preset image height, Dpi is the threshold pixel, and j is the jth image in the vertical direction.

8. The wave breaking object model test method based on target detection technology according to claim 1 is characterized in that: The step 4 also includes: taking the average of the upper and lower layer heights of the wave surface noise data with a certain thickness as the wave height line, and using Fourier transform to filter the non-smooth wave surface curve.

9. The wave breaking object model test method based on target detection technology as claimed in claim 8, characterized in that: The Fourier transform F(ω) formula is: ； The inverse Fourier transform is used to return from the frequency domain to the time domain, and its expression is: ； In the above formula: f(t) is the original function, i is the imaginary unit, ω is the angular frequency, t is the time variable; the complex exponential function e −iωt is the weight.

10. The wave breaking object model test method based on target detection technology according to claim 1, characterized in that: The step 4 also includes: obtaining time process data of a single point by extracting the water surface line; calculating the gradient change of the wave by calculating the changes of multiple different coordinate points in the same wave propagation process; the single wave gradient calculation formula is as follows: In the formula, Represents the gradient of the function f(x,y), the partial derivative of the function f(x,y) with respect to the independent variable y.

Images