A non-contact wave measurement method and device based on machine vision

Abstract

This invention discloses a non-contact wave measurement method and apparatus based on machine vision. The method includes: simultaneously capturing images from multiple visions using a multi-view imaging system; standardizing the captured images to obtain wavefront images, wherein the standardization processing includes color adjustment, noise reduction, data augmentation, texture analysis, edge detection, and frequency domain analysis, and the wavefront images include at least two complete wave cycles; and performing image analysis on the wavefront images using a pre-trained convolutional neural network to obtain predicted wave height values. This invention overcomes the problems of traditional monitoring methods, such as requiring contact with the object being measured and the environment, complex deployment, and susceptibility to marine environmental influences.

Description

Technical Field

[0001] This invention belongs to the technical field of marine engineering, specifically relating to a non-contact wave measurement method and device based on machine vision. Background Technology

[0002] Currently, wave monitoring is crucial for understanding ocean dynamic processes and developing marine resources in the field of marine monitoring. However, traditional monitoring methods inevitably require contact with the monitored object and environment, such as wave height meters, which are easily affected by surrounding structures and the environment. In recent years, the widespread application of computer vision systems in various fields has also brought many conveniences to marine monitoring. This invention proposes a non-contact wave measurement method and system based on machine vision, which utilizes a multi-view imaging system and deep learning technology to achieve automatic wave height recognition. It aims to provide a non-contact, easier-to-operate and deploy, lower-cost, and more efficient solution suitable for the rapid and accurate determination of wave parameters in various marine environments. Summary of the Invention

[0003] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a non-contact wave measurement method and device based on machine vision. It combines a binocular / trinocular imaging system with advanced image processing technology and uses a convolutional neural network model to accurately extract wave height data, thereby overcoming the problems of traditional monitoring methods that require contact with the object being measured and the environment, are complex to deploy, and are easily affected by the marine environment.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] In a first aspect, the present invention provides a non-contact wave measurement method based on machine vision, comprising the following steps:

[0006] Images are acquired by simultaneously capturing images from multiple viewpoints using a multi-view imaging system.

[0007] The captured image is standardized to obtain a wavefront image. The standardization process includes color adjustment, noise reduction, data enhancement, texture analysis, edge detection, and frequency domain analysis. The wavefront image includes at least two complete wave cycles.

[0008] A pre-trained convolutional neural network is used to perform image analysis on wavefront images to obtain predicted wave height values.

[0009] As a preferred technical solution, the multi-view imaging system is a binocular imaging system or a triocular imaging system.

[0010] As a preferred technical solution, the simultaneous capture from multiple visual perspectives specifically includes:

[0011] Several multi-view cameras are installed on an ocean observation platform. The installation position of the multi-view cameras is set above the sea surface at a certain height. The imaging of the multi-view cameras covers a certain area of ​​the sea surface in the near field. When the sea surface is active, the multi-view cameras take pictures of the sea surface area at their respective set angles to acquire the captured images. The captured images include at least two complete wave cycles.

[0012] As a preferred technical solution, the color adjustment reduces the color difference of images in different scenes by adjusting brightness, contrast and color balance, and converts the captured image into a single-channel grayscale image using a grayscale function;

[0013] The noise reduction process removes random noise from the image by applying a filter.

[0014] The data augmentation includes random cropping, rotation, and translation.

[0015] As a preferred technical solution, the texture analysis includes:

[0016] The co-occurrence probability of pixel values ​​at specific distances and directions is calculated using the gray-level co-occurrence matrix, and the texture characteristics of the image are quantified; the texture features include contrast, correlation, energy, and entropy.

[0017] Local binary mode is used to capture local texture details. By comparing the gray values ​​of the center pixel with its surrounding neighbors, a binary code is formed, and the local binary mode histogram of the entire image is counted as the texture feature vector.

[0018] As a preferred technical solution, the edge detection and frequency domain analysis include:

[0019] Edge detection: The Sobel operator is used to calculate the horizontal and vertical gradients of the image to obtain the gradient information of each pixel. The gradient information includes the gradient magnitude and direction. Based on the gradient information, the Canny operator is used to perform edge detection on the image.

[0020] The process of edge detection using the Canny operator based on gradient information specifically involves: reducing image noise using Gaussian filtering; performing non-maximum suppression on the image based on the gradient information of each pixel, refining edges by retaining pixels with local maxima and removing redundant responses; determining edge points by setting high and low thresholds using a dual-threshold processing method, where edge points include strong edges and weak edges, marking pixels above the high threshold as strong edges and pixels between the two thresholds as weak edges; and concatenating all edge points to obtain the edge detection result.

[0021] Frequency domain analysis: The image is converted to the frequency domain using Fast Fourier Transform to obtain a spectrum. The spectrum shows the intensity of different frequency components, with the high-frequency part corresponding to subtle changes and the low-frequency part reflecting the overall trend. Significant peaks in the spectrum are identified to determine the main wavelength frequencies, and the effectiveness of the frequency domain analysis is verified by inverse transform.

[0022] As a preferred technical solution, the convolutional neural network includes an input layer, five convolutional layers and three fully connected layers. Each convolutional layer includes a filter, and the number of filters increases layer by layer. Each convolutional layer is followed by a ReLU activation function and a max pooling layer. The first two fully connected layers are used to integrate high-level features, and the last fully connected layer outputs the wave height prediction value.

[0023] As a preferred technical solution, the pre-training of the convolutional neural network includes:

[0024] We collected wavefront images taken from different heights and angles, and used the corresponding ground truth wave heights as labels to obtain a dataset. We then used K-fold cross-validation to split the dataset and input it into a convolutional neural network for training. We set several epochs and used early stopping to avoid overfitting.

[0025] The process of inputting the dataset into the convolutional neural network for training specifically involves:

[0026] Features of a single-channel grayscale image are extracted using convolutional layers and filtered to obtain wave height features. The first two fully connected layers integrate the wave height features, and the last fully connected layer outputs the predicted value. The mean square error is used as the loss function to minimize the difference between the predicted and true values. The difference is backpropagated to calculate the gradient, and the Adam optimizer is used to update the network parameters based on the gradient. The convolutional neural network is evaluated using evaluation metrics, and a residual plot is plotted to visually check the difference between the predicted and true values. The evaluation metrics include root mean square error, mean absolute error, and R² coefficient of determination.

[0027] Secondly, the present invention also provides a non-contact wave measurement system based on machine vision, applied to the aforementioned non-contact wave measurement method based on machine vision, including a hardware platform, a basic operating system, and a communication module.

[0028] A hardware platform for acquiring images by simultaneously capturing images from multiple visions using a multi-view imaging system;

[0029] The basic operating system is used to standardize the captured images to obtain wavefront images. The standardization process includes color adjustment, noise reduction, data enhancement, texture analysis, edge detection, and frequency domain analysis. The wavefront images include at least two complete wave cycles. A pre-trained convolutional neural network is used to perform image analysis on the wavefront images to obtain predicted wave height values.

[0030] The communication module is used to send the processed wave height data to the cloud server or local monitoring center. The system also has remote management and firmware update functions, which facilitates operation by maintenance personnel.

[0031] Thirdly, the present invention provides an electronic device, the electronic device comprising:

[0032] At least one processor; and,

[0033] A memory communicatively connected to the at least one processor; wherein,

[0034] The memory stores computer program instructions that can be executed by the at least one processor to enable the at least one processor to perform the machine vision-based non-contact wave measurement method.

[0035] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0036] (1) The present invention uses a dual-view / tri-view imaging system to simultaneously capture the sea surface wave situation from multiple angles. By using the multi-view geometric principle to reconstruct three-dimensional information, the image resolution error caused by the change of shooting angle and distance is overcome, and the absolute value of wave height is extracted as accurately as possible.

[0037] (2) By standardizing a single static image, this invention ensures that the image contains sufficient information on the spatial distribution of the wave surface, making it easy to identify and calculate the wave height, minimizing the impact of ship motion on the measurement results, and ensuring the accuracy and reliability of the wave height data.

[0038] (3) This invention integrates a camera system and a machine learning model into a simple and efficient wave height monitoring product, which can directly output wave height data based on image information. The system is designed to simplify operation and improve efficiency, ensuring that users can easily deploy it. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a non-contact wave measurement method based on machine vision, as described in an embodiment of the present invention.

[0041] Figure 2This is a schematic diagram of the structure of a convolutional neural network according to an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the structure of the non-contact wave measurement system based on machine vision according to an embodiment of the present invention;

[0043] Figure 4 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0045] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0046] Please see Figure 1 This embodiment provides a non-contact wave measurement method based on machine vision, including the following steps:

[0047] S1. Use a multi-view imaging system to capture images simultaneously from multiple perspectives.

[0048] This embodiment uses a binocular or tri-lens camera as the imaging system. Compared to a monocular camera, binocular / tri-lens cameras have significant advantages in depth perception and anti-interference capabilities when capturing wavefront images. Especially when the camera is shooting at a small angle and in poor lighting conditions, binocular / tri-lens cameras extract richer effective wavefront information. When shooting the wavefront head-on, i.e., when the camera's optical axis is perpendicular to the object's surface, a monocular camera is insufficient to provide rich spatial information.

[0049] To adapt to the complex marine environment, the camera is waterproof, dustproof, and corrosion-resistant, and is equipped with an automatic exposure control system to cope with changing light conditions. In addition, the camera housing integrates a solar panel to provide continuous power support.

[0050] S2. Standardize the captured image to obtain a wavefront image. The standardization process includes color adjustment, noise reduction, data enhancement, texture analysis, edge detection, and frequency domain analysis. The wavefront image includes at least two complete wave cycles.

[0051] When shooting, ensuring that images are captured containing at least two complete wave cycles is crucial for accurately extracting wave height data. This is because waves are periodic natural phenomena, and a single wave cycle is insufficient to provide enough information to train a model or make reliable predictions. Capturing multiple wave cycles improves the representativeness of the data and the generalization ability of the model. In practical applications, cameras are typically mounted on platforms capable of capturing multiple wave height cycles.

[0052] Subsequent analysis also utilizes convolutional neural networks to extract deeper spatial features from the preprocessed images. Specifically, a deep learning model extracts key wave features from the preprocessed images, including but not limited to the positions of wave crests and troughs and the relationships between them. These features are then integrated using fully connected layers to ultimately output accurate wave height predictions.

[0053] Next, the standardization process not only improves image quality but also provides high-quality data support for subsequent deep learning model training. Specifically, the standardization process for captured images in this embodiment includes the following steps: color adjustment, noise reduction, data augmentation, texture analysis, edge detection, and frequency domain analysis of the input image.

[0054] The color adjustment reduces color differences in images from different scenes by adjusting brightness, contrast, and color balance. To reduce computation, a grayscale function is used to convert the captured image into a single-channel grayscale image. The initial image acquired by the camera is a color image (RGB three-channel image). The grayscale function built into OpenCV is used to convert the three-channel image into a single-channel grayscale image to reduce computation.

[0055] The noise reduction process removes random noise from the image by applying filters, thereby improving image quality.

[0056] The data augmentation utilizes existing multi-angle and distance datasets, further expanding the datasets through methods such as random cropping, rotation, and translation, thereby increasing the model's generalization ability.

[0057] The texture analysis described employs methods such as Gray-Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP) to extract texture features from the wave surface, describing changes in the wave surface and the location of wave crests and troughs. It's important to explain that texture analysis extracts information describing surface structure features from images; for wave images, it helps distinguish waves from the background and enhances the model's adaptability to complex environments. First, the GLCM calculates the co-occurrence probability of pixel values ​​at specific distances and directions, quantifying the image's texture characteristics. After constructing the GLCM, four key features are calculated: contrast, correlation, energy, and entropy. Furthermore, the LBP captures local texture details by comparing the gray values ​​of the central pixel with its surrounding neighbors, forming a binary code, and statistically analyzing the LBP histogram of the entire image as a texture feature vector. These methods work together to provide rich wave surface features, facilitating subsequent edge detection.

[0058] The edge detection and frequency domain analysis described herein utilize methods such as the Canny or Sobel operators to perform edge detection on the image, helping to identify the locations of wave peaks and troughs. It also allows the image to be converted to the frequency domain, through which the periodicity and spatial distribution characteristics of the waves can be extracted. Edge detection further highlights the contour information of the waves, while frequency domain analysis reveals their periodicity and spatial distribution features.

[0059] Edge detection involves two steps: first, the Sobel operator calculates the gradient information of the image; second, the Canny operator further optimizes the edge detection results. Specifically, the steps include:

[0060] S21. The Sobel operator is used to calculate the horizontal and vertical gradients, and the gradient information of each pixel is obtained by combining them. The gradient information reflects the intensity of the brightness change of each pixel in the image. For example, the magnitude of the gradient indicates that the pixel with a significant brightness change may be an edge location. However, whether it is an edge location is determined by subsequent operations. The direction of the gradient helps to determine the directionality of the edge, but it is not a direct edge detection result. This gradient information is the basis for subsequent processing. Through steps such as Gaussian filtering to reduce noise, non-maximum suppression to refine edges, and double thresholding to distinguish between strong and weak edges, the Canny operator finally uses this gradient information to generate accurate and clear edge detection results.

[0061] S22. The Canny operator further optimizes the edge detection results through Gaussian filtering, non-maximum suppression, and double thresholding, ensuring that the subsequent convolutional neural network can gradually capture higher-level structural features in the image, such as the positions of peaks and valleys and the relationships between them, during feature extraction. The specific steps are as follows:

[0062] (1) Apply Gaussian filtering to the image to reduce noise;

[0063] (2) Non-maximum suppression is performed based on gradient information, and pixels with local maximum values ​​are retained to refine the edges and remove redundant responses;

[0064] (3) Use the dual threshold processing method to set high and low thresholds to determine edge points, that is, pixels above the high threshold are marked as strong edges, and pixels between the two thresholds are marked as weak edges;

[0065] (4) Connect all edge points to obtain the edge detection results.

[0066] Connecting these confirmed edge points forms a complete edge contour, ensuring accurate, clear, and noise-resistant edge detection results. This process utilizes gradient information to progressively optimize the quality of edge detection.

[0067] In this embodiment, frequency domain analysis is used to confirm the consistency and integrity of the wave cycle. The image is transformed to the frequency domain using a Fast Fourier Transform (FFT), and the spectrum plot shows the intensity of different frequency components, with high-frequency components corresponding to subtle variations and low-frequency components reflecting the overall trend. Significant peaks in the spectrum plot are identified to determine the main wavelength frequencies, and the effectiveness of the frequency domain analysis is verified through inverse transform. The frequency domain analysis results serve as an auxiliary verification method for spatial domain texture analysis and edge detection, ensuring the reliability of both results.

[0068] S3. Use a pre-trained convolutional neural network to perform image analysis on the wavefront image and obtain the predicted wave height value.

[0069] like Figure 2 As shown, the architecture of the convolutional neural network model in this embodiment includes: an input layer, convolutional layers, fully connected layers, a loss function, an optimization algorithm, a training strategy, and an evaluation metric. It includes five convolutional layers and three fully connected layers. Each convolutional layer contains filters, with the number of filters increasing layer by layer. Each convolutional layer is followed by a ReLU activation function and a max-pooling layer. The first two fully connected layers are used to integrate high-level features, and the last fully connected layer outputs the predicted wave height value. For example, the first convolutional layer may have 32 filters, the second convolutional layer may have 64 filters, and so on.

[0070] Specifically, during the pre-training of the convolutional neural network, wavefront images taken from different heights and angles are collected, and the corresponding ground truth wave heights are used as labels to obtain a dataset. K-fold cross-validation is used to segment the dataset, and the dataset is input into the model for training. The training duration is set to 10,000 epochs, and early stopping is used to avoid overfitting. Specifically, the training process includes: converting the images in the dataset into single-channel grayscale images using the input layer; extracting and filtering the features of the single-channel grayscale images using convolutional layers to obtain wave height features; integrating the wave height features using the first two fully connected layers; and outputting the predicted wave height using the last fully connected layer. The mean squared error (MSE) is used as the loss function to minimize the difference between the predicted and ground truth values. The difference is backpropagated to calculate the gradient, and the Adam optimizer is used to update the network parameters based on the gradient. The convolutional neural network is evaluated using evaluation metrics, and a residual plot is plotted to visually check the difference between the predicted and ground truth values. The evaluation metrics include root mean square error (RMSE), mean absolute error (MAE), and the R² coefficient of determination. K-fold cross-validation is used to make full use of limited data and ensure that the model has good generalization performance.

[0071] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0072] Based on the same idea as the machine vision-based non-contact wave measurement method in the above embodiments, the present invention also provides a machine vision-based non-contact wave measurement system, which can be used to perform the above-described machine vision-based non-contact wave measurement method. For ease of explanation, the structural schematic diagram of the machine vision-based non-contact wave measurement system embodiment only shows the parts related to the embodiments of the present invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0073] Please see Figure 3 In another embodiment of this application, a non-contact wave measurement system 10 based on machine vision is provided. The system includes a hardware platform 11, a basic operating system 12, and a communication module 13.

[0074] Hardware platform 11 is used to acquire images by simultaneously capturing images from multiple visions using a multi-view imaging system; the hardware platform 11 integrates the camera system with an embedded computing unit in a compact and durable waterproof housing, and has a built-in high-performance ARM processor to support real-time image processing and model inference tasks.

[0075] The base operating system 12 uses a lightweight Linux distribution (such as Ubuntu Core) as the base operating system to provide a stable running environment. The programming language is Python, which is used to develop the system by leveraging its rich image processing libraries (such as OpenCV) and machine learning frameworks (such as TensorFlow Lite). This system is used to standardize the captured images to obtain wavefront images. The standardization process includes color adjustment, noise reduction, data augmentation, texture analysis, edge detection, and frequency domain analysis. The wavefront images include at least two complete wave cycles. A pre-trained convolutional neural network is used to perform image analysis on the wavefront images to obtain predicted wave height values.

[0076] The communication module 13 is used to send the processed wave height data to the cloud server or local monitoring center. The system also has remote management and firmware update functions, which facilitates operation by maintenance personnel.

[0077] It should be noted that the non-contact wave measurement system based on machine vision of the present invention corresponds one-to-one with the non-contact wave measurement method based on machine vision of the present invention. The technical features and beneficial effects described in the embodiments of the non-contact wave measurement method based on machine vision described above are applicable to the embodiments of the non-contact wave measurement method based on machine vision. For details, please refer to the description in the embodiments of the method of the present invention, which will not be repeated here.

[0078] Furthermore, in the embodiments of the non-contact wave measurement system based on machine vision described above, the logical division of each program module is merely an example. In actual applications, the functions described above can be assigned to different program modules as needed, for example, for the sake of hardware configuration requirements or software implementation convenience. That is, the internal structure of the non-contact wave measurement system based on machine vision can be divided into different program modules to complete all or part of the functions described above.

[0079] Please see Figure 4 In one embodiment, an electronic device is provided for implementing a machine vision-based non-contact wave measurement method. The electronic device 20 may include a first processor 21, a first memory 22, and a bus, and may also include a computer program stored in the first memory 22 and executable on the first processor 21, such as a machine vision-based non-contact wave measurement program 23.

[0080] The first memory 22 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the first memory 22 can be an internal storage unit of the electronic device 20, such as the portable hard drive of the electronic device 20. In other embodiments, the first memory 22 can be an external storage device of the electronic device 20, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 20. Furthermore, the first memory 22 can include both internal and external storage units of the electronic device 20. The first memory 22 can be used not only to store application software and various types of data installed on the electronic device 20, such as the code of a machine vision-based non-contact wave measurement program 23, but also to temporarily store data that has been output or will be output.

[0081] In some embodiments, the first processor 21 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The first processor 21 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the first memory 22 and calls data stored in the first memory 22 to perform various functions of the electronic device 20 and process data.

[0082] Figure 4 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 4 The structure shown does not constitute a limitation on the electronic device 20, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0083] The machine vision-based non-contact wave measurement program 23 stored in the first memory 22 of the electronic device 20 is a combination of multiple instructions. When run in the first processor 21, it can achieve the following:

[0084] Images are acquired by simultaneously capturing images from multiple viewpoints using a multi-view imaging system.

[0085] The captured image is standardized to obtain a wavefront image. The standardization process includes color adjustment, noise reduction, data enhancement, texture analysis, edge detection, and frequency domain analysis. The wavefront image includes at least two complete wave cycles.

[0086] A pre-trained convolutional neural network is used to perform image analysis on wavefront images to obtain predicted wave height values.

[0087] Furthermore, if the modules / units integrated in the electronic device 20 are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0088] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0089] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A non-contact wave measurement method based on machine vision, characterized in that, Includes the following steps: The system utilizes a multi-view imaging system to simultaneously capture images from multiple perspectives; The captured image is standardized to obtain a wavefront image. The standardization process includes color adjustment, noise reduction, data enhancement, texture analysis, edge detection, and frequency domain analysis. The wavefront image includes at least two complete wave cycles. A pre-trained convolutional neural network is used to perform image analysis on wavefront images to obtain predicted wave height values; The convolutional neural network includes an input layer, five convolutional layers and three fully connected layers. Each convolutional layer includes a filter, and the number of filters increases layer by layer. Each convolutional layer is followed by a ReLU activation function and a max pooling layer. The first two fully connected layers are used to integrate high-level features, and the last fully connected layer outputs the wave height prediction value. The pre-training of the convolutional neural network includes: We collected wavefront images taken from different heights and angles, and used the corresponding ground truth wave heights as labels to obtain a dataset. We then used K-fold cross-validation to split the dataset and input it into a convolutional neural network for training. We set several epochs and used early stopping to avoid overfitting. The process of inputting the dataset into the convolutional neural network for training specifically involves: Features of a single-channel grayscale image are extracted using convolutional layers and filtered to obtain wave height features. The first two fully connected layers integrate the wave height features, and the last fully connected layer outputs the predicted value. The mean square error is used as the loss function to minimize the difference between the predicted and true values. The difference is backpropagated to calculate the gradient, and the Adam optimizer is used to update the network parameters based on the gradient. The convolutional neural network is evaluated using evaluation metrics, and a residual plot is plotted to visually check the difference between the predicted and true values. The evaluation metrics include root mean square error, mean absolute error, and R² coefficient of determination.

2. The non-contact wave measurement method based on machine vision according to claim 1, characterized in that, The multi-view imaging system is either a binocular imaging system or a triocular imaging system.

3. The non-contact wave measurement method based on machine vision according to claim 1, characterized in that, The simultaneous capture from multiple visions specifically refers to: Several multi-view cameras are installed on an ocean observation platform. The installation position of the multi-view cameras is set above the sea surface at a certain height. The imaging of the multi-view cameras covers a certain area of ​​the sea surface in the near field. When the sea surface is active, the multi-view cameras take pictures of the sea surface area at their respective set angles to acquire the captured images. The captured images include at least two complete wave cycles.

4. The non-contact wave measurement method based on machine vision according to claim 1, characterized in that, The color adjustment reduces color differences in images from different scenes by adjusting brightness, contrast, and color balance, and uses a grayscale function to convert the captured image into a single-channel grayscale image. The noise reduction process removes random noise from the image by applying a filter. The data augmentation includes random cropping, rotation, and translation.

5. The non-contact wave measurement method based on machine vision according to claim 1, characterized in that, The texture analysis includes: The co-occurrence probability of pixel values ​​at specific distances and directions is calculated using the gray-level co-occurrence matrix, and the texture characteristics of the image are quantified; the texture features include contrast, correlation, energy, and entropy. Local binary mode is used to capture local texture details. By comparing the gray values ​​of the center pixel with its surrounding neighbors, a binary code is formed, and the local binary mode histogram of the entire image is counted as the texture feature vector.

6. The non-contact wave measurement method based on machine vision according to claim 1, characterized in that, The edge detection and frequency domain analysis include: Edge detection: The Sobel operator is used to calculate the horizontal and vertical gradients of the image to obtain the gradient information of each pixel. The gradient information includes the gradient magnitude and direction. Based on the gradient information, the Canny operator is used to perform edge detection on the image. The process of edge detection using the Canny operator based on gradient information specifically involves: reducing image noise using Gaussian filtering; performing non-maximum suppression on the image based on the gradient information of each pixel, refining edges by retaining pixels with local maxima and removing redundant responses; determining edge points by setting high and low thresholds using a dual-threshold processing method, where edge points include strong edges and weak edges, marking pixels above the high threshold as strong edges and pixels between the two thresholds as weak edges; and concatenating all edge points to obtain the edge detection result. Frequency domain analysis: The image is converted to the frequency domain using Fast Fourier Transform to obtain a spectrum. The spectrum shows the intensity of different frequency components, with the high-frequency part corresponding to subtle changes and the low-frequency part reflecting the overall trend. Significant peaks in the spectrum are identified to determine the main wavelength frequencies, and the effectiveness of the frequency domain analysis is verified by inverse transform.

7. A non-contact wave measurement system based on machine vision, characterized in that, The non-contact wave measurement method based on machine vision, applicable to any one of claims 1-6, comprises: a hardware platform, a basic operating system, and a communication module; A hardware platform for acquiring images by simultaneously capturing images from multiple visions using a multi-view imaging system; The basic operating system is used to standardize the captured images to obtain wavefront images. The standardization process includes color adjustment, noise reduction, data enhancement, texture analysis, edge detection, and frequency domain analysis. The wavefront images include at least two complete wave cycles. A pre-trained convolutional neural network is used to perform image analysis on the wavefront images to obtain predicted wave height values. The communication module is used to send the processed wave height data to the cloud server or local monitoring center. The system also has remote management and firmware update functions, which facilitates operation by maintenance personnel.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor to enable the at least one processor to perform the machine vision-based non-contact wave measurement method as described in any one of claims 1-6.

Images