Wave period method and device based on deep learning

Abstract

The embodiment of the invention discloses a wave period identification method and device based on deep learning, and the method comprises the steps: collecting wave data in a preset time length at preset time length intervals, carrying out the image processing of the wave data, obtaining a target sea wave image, and obtaining the white-head wave value data; and performing calculation based on the target sea wave picture and the white head wave value data to obtain a wave period estimation value, and constructing a wave period identification model of the wave truth value and the wave period estimation value. A target sea wave picture is automatically extracted in a deep learning mode, and errors and cost of manual intervention are reduced; based on calculation of a target sea wave picture and white head wave value data, an accurate prediction basis is provided for wave period estimation, the accuracy and stability of the wave period estimation are further improved in combination with application of a deep learning algorithm, and a wave period recognition model constructed by the embodiment can automatically learn and optimize the relationship between sea wave features and wave periods. Through continuous learning and iteration, the model can adaptively adjust parameters, and the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a wave cycle identification method and device based on deep learning. Background Art

[0002] Nearshore waves are closely related to coastal topography, engineering design, and shallow coastal production operations. Wave elements such as wave height, period, and wavelength are important indicators that reflect the characteristics of waves and the state of the sea surface. Therefore, monitoring the movement and change trend of nearshore waves requires high-precision detection of wave elements. The wave period is the time required for two consecutive waves to pass through the same point, reflecting the movement law of waves. It plays a vital role in surfing forecasts, marine navigation positioning, and changes in the geographical environment along the coast.

[0003] The existing nearshore wave period detection methods are mainly divided into two categories: direct and indirect detection methods. The direct detection method mainly relies on ocean buoys. The wave period measured by ocean buoys is basically consistent with the real wave period and is used as a reference for evaluating other measurement methods. However, the placement points of ocean buoys are limited and the cost of buoy placement is high, so it is impossible to achieve coverage detection of the wave period. The indirect detection method is mainly based on wave spectrum analysis. This method requires artificially designed extraction algorithms to extract elements such as wave period from the wave spectrum. The calculation amount is large and the requirements for computing equipment are high. Summary of the invention

[0004] In response to the above-mentioned defects, an embodiment of the present invention discloses a wave cycle identification method and device based on deep learning, which reduces data collection costs and has an accurate detection method.

[0005] The first aspect of the embodiment of the present invention discloses a wave cycle identification method based on deep learning, comprising:

[0006] Collecting wave data within a preset time period at intervals of the preset time period, wherein the wave data includes ocean wave video data and wave cycle true value data; the ocean wave video data and the wave cycle true value data are correspondingly associated according to a time series relationship;

[0007] Performing image processing on the wave data to obtain a target wave image and obtain white-cap wave value data;

[0008] The wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, and a wave cycle recognition model of the wave true value and the wave cycle estimation is constructed.

[0009] As an optional implementation, in the first aspect of the embodiment of the present invention, the preset duration is 60 seconds.

[0010] As an optional implementation manner, in the first aspect of the embodiment of the present invention, performing image processing on the wave data to obtain a target wave picture includes:

[0011] A number of target wave images are captured from the wave video data, and each target wave image is matched with the corresponding wave period true value data;

[0012] The target wave images are divided into regions, and the region where the white-capped waves are located in the wave image is selected as the target region.

[0013] As an optional implementation, in the first aspect of the embodiment of the present invention, a plurality of target wave images are captured from the wave video data, and each target wave image is associated with corresponding wave period true value data, including:

[0014] Capture a frame of ocean wave picture at every set time interval from the ocean wave video data;

[0015] Each frame of the wave image is matched with the true value data of the wave cycle according to the time sequence relationship.

[0016] As an optional implementation manner, in the first aspect of the embodiment of the present invention, obtaining white-headed wave value data includes:

[0017] Based on the Segnet algorithm, the white-headed spray value in the target area is identified and the pixel point value of the white-headed spray value is obtained.

[0018] As an optional implementation manner, in the first aspect of the embodiment of the present invention, selecting the area where the white-headed waves in the target wave picture are located as the target area includes:

[0019] Identify white-capped waves from all target wave images, and select target wave images with white-capped waves as screening images;

[0020] Select at least one distribution area with white-headed waves in each of the screened pictures, and calculate whether the density value of the white-headed waves in the distribution area is greater than a set density value;

[0021] The distribution area with density values ​​greater than the set density value is defined as the target area.

[0022] As an optional implementation, in the first aspect of the embodiment of the present invention, the wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, including:

[0023] The wave true value and white-cap wave value data corresponding to each target wave image within a preset time period are obtained through fast Fourier transform to obtain the main frequency;

[0024] The reciprocal of the main frequency is calculated and used as the wave period estimation.

[0025] A second aspect of an embodiment of the present invention discloses a wave cycle identification device based on deep learning, comprising:

[0026] Data collection module: used for collecting wave data within a preset time period at intervals of the preset time period, wherein the wave data includes ocean wave video data and wave cycle true value data; the ocean wave video data and the wave cycle true value data are correspondingly associated in a time series relationship;

[0027] Image processing module: used for performing image processing on the wave data to obtain target wave images and obtain white-cap wave value data;

[0028] Model building module: used to calculate and obtain the wave cycle estimation based on the target wave image and the white-cap wave value data, and to build a wave cycle recognition model of the wave true value and the wave cycle estimation.

[0029] As an optional implementation, in the second aspect of the embodiment of the present invention, the preset duration is 60 seconds.

[0030] As an optional implementation, in the second aspect of the embodiment of the present invention, performing image processing on the wave data to obtain a target wave picture includes:

[0031] A number of target wave images are captured from the wave video data, and each target wave image is matched with the corresponding wave period true value data;

[0032] The target wave images are divided into regions, and the region where the white-capped waves are located in the wave image is selected as the target region.

[0033] As an optional implementation, in the second aspect of the embodiment of the present invention, a plurality of target wave images are captured from the wave video data, and each target wave image is associated with corresponding wave period true value data, including:

[0034] Capture a frame of ocean wave picture at every set time interval from the ocean wave video data;

[0035] Each frame of the wave image is matched with the true value data of the wave cycle according to the time sequence relationship.

[0036] As an optional implementation manner, in the second aspect of the embodiment of the present invention, obtaining white head wave value data includes:

[0037] Based on the Segnet algorithm, the white-headed spray value in the target area is identified and the pixel point value of the white-headed spray value is obtained.

[0038] As an optional implementation, in the second aspect of the embodiment of the present invention, selecting the area where the white-headed waves in the target wave picture are located as the target area includes:

[0039] Identify white-capped waves from all target wave images, and select target wave images with white-capped waves as screening images;

[0040] Select at least one distribution area with white-headed waves in each of the screened pictures, and calculate whether the density value of the white-headed waves in the distribution area is greater than a set density value;

[0041] The distribution area with density values ​​greater than the set density value is defined as the target area.

[0042] As an optional implementation, in the second aspect of the embodiment of the present invention, the wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, including:

[0043] The wave true value and white-cap wave value data corresponding to each target wave image within a preset time period are obtained through fast Fourier transform to obtain the main frequency;

[0044] The reciprocal of the main frequency is calculated and used as the wave period estimation.

[0045] The third aspect of an embodiment of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the wave cycle identification method based on deep learning disclosed in the first aspect of the embodiment of the present invention.

[0046] The fourth aspect of an embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the wave cycle identification method based on deep learning disclosed in the first aspect of an embodiment of the present invention.

[0047] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0048] In the embodiment of the present invention, the wave data within the interval time is automatically collected at regular intervals, that is, the embodiment reduces the deployment cost of sensors such as buoys, and the wave cycle data can be obtained directly through the camera; the embodiment performs image processing on the wolf data to obtain white-headed wave value data, extracts the image features of white-headed waves in the wave cycle based on a deep learning method, calculates the wave cycle estimation value in combination with the wave cycle true value data obtained at the same time, and models the wave cycle estimation value and the wave cycle true value through the BP neural network to obtain a wave cycle recognition model; since the time length for collecting wave data is set, the continuity and timeliness of the data are ensured, and the wave video data and the wave cycle true value data are temporally correlated. The corresponding lines provide an accurate data basis for subsequent model training; the embodiment automatically extracts target wave images by deep learning, reducing the error and cost of manual intervention, and the extraction of white-cap wave value data also provides important reference information for the calculation of wave cycle valuation; the calculation based on target wave images and white-cap wave value data provides an accurate prediction basis for wave cycle valuation, and the application of deep learning algorithms further improves the accuracy and stability of wave cycle valuation, and the wave cycle recognition model constructed in the embodiment can automatically learn and optimize the relationship between wave characteristics and wave cycle. Through continuous learning and iteration, the model can adaptively adjust parameters to improve prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 It is a flowchart of a wave cycle identification method based on deep learning provided by an embodiment of the present invention;

[0051] Figure 2 is a convolution operation graph of the Segnet algorithm provided in an embodiment of the present invention;

[0052] Figure 3 It is a structural schematic diagram of a wave cycle identification device based on deep learning provided by an embodiment of the present invention;

[0053] Figure 4 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] It should be noted that the terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present invention and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0056] The embodiments of the present invention disclose a wave cycle recognition method, device, electronic device and storage medium based on deep learning. In the embodiments, wave data within the interval time is automatically collected at regular intervals, that is, the embodiments reduce the deployment cost of sensors such as buoys, and the wave cycle data can be obtained directly through the camera; the embodiments process the wolf data to obtain white-headed wave value data, extract the image features of white-headed waves in the wave cycle based on deep learning, calculate the wave cycle estimation value in combination with the wave cycle true value data obtained at the same time, and model the wave cycle estimation value and the wave cycle true value through the BP neural network to obtain a wave cycle recognition model; since the time length for collecting wave data is set, the continuity and timeliness of the data are ensured, and the sea The wave video data and the wave cycle true value data correspond in time, providing an accurate data basis for subsequent model training; the embodiment automatically extracts target wave images through deep learning, reducing the error and cost of manual intervention, and the extraction of white-cap wave value data also provides important reference information for the calculation of wave cycle valuation; the calculation based on target wave images and white-cap wave value data provides an accurate prediction basis for wave cycle valuation, and the application of deep learning algorithms further improves the accuracy and stability of wave cycle valuation, and the wave cycle recognition model constructed in the embodiment can automatically learn and optimize the relationship between wave characteristics and wave cycles. Through continuous learning and iteration, the model can adaptively adjust parameters to improve prediction accuracy.

[0057] Embodiment 1

[0058] See also Figure 1 , Figure 1It is a flow chart of a wave cycle identification method based on deep learning disclosed in an embodiment of the present invention. Among them, the execution subject of the method described in the embodiment of the present invention is an execution subject composed of software and / or hardware, and the execution subject can receive relevant information by wired or / and wireless means, and can send certain instructions. Of course, it can also have certain processing functions and storage functions. The execution subject can control multiple devices, such as a remote physical server or cloud server and related software, or it can be a local host or server and related software that performs related operations on a device placed somewhere. In some scenarios, multiple storage devices can also be controlled, and the storage devices can be placed in the same place or different places as the devices. For example Figure 1 As shown, the wave cycle identification method based on deep learning includes the following steps:

[0059] 101. Collect wave data within a preset time period at intervals of the preset time period, wherein the wave data includes ocean wave video data and wave cycle true value data; the ocean wave video data and the wave cycle true value data are correspondingly associated in a time series relationship.

[0060] This step is to collect data, and collect wave data within the time period at preset time intervals. The preset time can be every minute, every hour, etc., and is set according to actual conditions. As a selected example, the embodiment adopts the preset time of 60 seconds. The time length of 60 seconds is usually enough to capture one or more complete cycles of waves, especially for those waves with longer cycles, which means that the data collected within 60 seconds can better reflect the periodic characteristics of waves, thereby providing a reliable basis for the calculation of wave cycles. Shorter data collection time (such as 60 seconds) can reduce the time for data processing and calculation, and improve the response speed of the overall system. Selecting 60 seconds as the preset time can reasonably control the resource consumption of data collection, storage and processing while ensuring data quality, which helps to reduce system costs and improve resource utilization efficiency. The embodiment selects 60 seconds as the preset duration, which is relatively short, but sufficient to capture the basic characteristics of the waves, so that the system can flexibly adapt to the changes in waves in different sea areas, different seasons and different weather conditions. The shorter acquisition duration means that the system can update the wave cycle information more frequently, thereby improving the real-time nature of prediction and monitoring, which is particularly important for scenarios that require rapid response, such as marine engineering safety monitoring and marine sports training. By collecting data of multiple wave cycles within 60 seconds, the system can calculate a more accurate wave cycle estimate, which helps to improve the accuracy and reliability of the wave cycle identification model and provide more reliable decision support for related applications. The preset duration of 60 seconds is relatively easy to implement, because most data acquisition devices and systems can collect and store data at this time interval, which helps to reduce the complexity and cost of system implementation. Therefore, selecting 60 seconds as the preset duration for collecting wave data brings benefits and advantages in terms of data representativeness, computing efficiency, resource optimization, flexibility, real-time, reliability and ease of implementation.

[0061] The wave data collected in the embodiment includes ocean wave video data and wave period true value data, wherein the ocean wave video data can be directly collected through a camera, and the wave true value data can still be collected through sensors such as buoys.

[0062] Among them, the wave video data and the wave cycle true value data are associated in a time series relationship, which means that the wave video data and the wave cycle true value data will be accompanied by timestamp information when they are collected. By aligning the timestamps of the two, the temporal consistency of the video data and the true value data can be ensured. The aligned wave video data are arranged in chronological order to form a video data time series. At the same time, the wave cycle true value data are also arranged in chronological order to form a true value data time series. The two time series are one-to-one corresponding in time, that is, each video data frame corresponds to one or more wave cycle true values. Furthermore, key features such as crests, troughs, wavelengths, and wave heights can be extracted from the wave video data, and these features can be associated with the wave cycle true value data to establish a mapping relationship between feature values ​​and wave cycles. This correspondence realizes the fusion of wave video data and wave cycle true value data, provides richer information for subsequent model training and analysis, and ensures the consistency of data in time and space through timestamp alignment and time series matching, thereby improving the accuracy and reliability of data analysis. The feature extraction and association process establishes a direct connection between wave features and wave cycles, providing a basis for the prediction and identification of wave cycles. The fused data contains richer information, which helps the deep learning model to better learn the relationship between wave features and wave cycles, thereby improving the prediction accuracy. By introducing a variety of wave features and true value data, the model can learn a wider feature space, thereby enhancing the generalization ability of the model, enabling it to adapt to the changes in waves in different sea areas, different seasons and different weather conditions. The data after the temporal relationship is associated is easier to analyze and visualize, which helps researchers to understand the changing laws and characteristics of waves more intuitively. By establishing a mapping relationship between feature values ​​and wave cycles, the data processing process can be simplified and the data processing efficiency can be improved. The wave video data and the wave cycle true value data are associated in accordance with the temporal relationship, which not only realizes the fusion and spatiotemporal consistency of the data, but also improves the prediction accuracy, enhances the generalization ability of the model, facilitates data analysis and visualization, and improves the data processing efficiency.

[0063] 102. Perform image processing on the wave data to obtain a target wave image, and obtain white-cap wave value data.

[0064] The embodiment uses a deep learning algorithm or other mathematical methods to calculate the wave cycle estimation based on the target wave image and the white-cap wave value data. The wave cycle estimation is a preliminary prediction or estimation of the wave cycle. In the embodiment, a deep learning image processing algorithm is used to improve the extraction accuracy of the target wave image, and deep learning technology such as convolutional neural network is used to advance image features to obtain richer wave information.

[0065] Specifically, image processing is performed on wave data to obtain target wave images, including: intercepting a number of target wave images from wave video data, and making each target wave image correspond to corresponding wave period true value data; and dividing the target wave images into regions, and selecting the region where the white-capped waves are located in the wave image as the target region.

[0066] To intercept several target wave images from the wave video data, technology is required to be able to process continuous video data and select representative or conditional wave images from them. The intercepted images correspond to the corresponding wave cycle true value data, ensuring that each image has corresponding physical parameters as a reference. Regional division of the target wave image is an effective image segmentation method that can simplify complex wave images into multiple regions that are easier to process and analyze. The area where the white-capped waves are located in the wave image is selected as the target area based on the significance and importance of the white-capped waves in the waves. The white-capped waves usually represent the active and high-energy parts of the waves.

[0067] The embodiment significantly improves the speed and efficiency of wave data processing by automatically capturing images from video data and quickly locating the white-capped wave area. The automated method reduces the workload of manual screening and marking, allowing researchers to focus more on subsequent analysis and modeling. The correspondence with the true value data of the wave cycle enables each target wave image to have a clear physical background, providing a basis for accurate analysis of wave characteristics. By focusing on the white-capped wave area, key characteristics of the waves such as energy distribution and morphological changes can be studied more accurately. Through this method, more reliable and detailed data support can be provided for subsequent wave forecasting, marine engineering, marine resource development and other fields, which helps to reveal the physical mechanism of wave movement and provide a scientific basis for early warning and protection of wave disasters.

[0068] Furthermore, a number of target wave images are captured from the wave video data, and each target wave image is matched with the corresponding wave cycle true value data, including: capturing a frame of wave image at every set time interval from the wave video data; and matching each frame of wave image with the wave cycle true value data according to the time sequence relationship.

[0069] By capturing a frame of wave images at set time intervals, it is possible to ensure uniform sampling over a period of time and capture various forms and changes of waves. This makes the captured wave images more representative and can more comprehensively reflect the periodicity, form, and energy distribution of waves. Each frame of wave images is matched with the true value data of the wave cycle according to the time series relationship, ensuring the consistency of the data in time. This correspondence is helpful for subsequent data analysis and modeling, especially when studying the law of wave changes over time, it can provide an accurate time reference. The method of capturing images by setting time intervals can significantly reduce the amount of data to be processed and improve data processing efficiency. At the same time, since each frame of the image corresponds to the true value data of the wave cycle, it is possible to quickly build a data set for training deep learning models or other analysis work while ensuring data quality. When training a deep learning model, using this data set with time series consistency can help the model better learn the relationship between wave characteristics and wave cycles. It helps to improve the prediction accuracy and generalization ability of the model, so that it can more accurately identify the wave cycle in different sea areas, different seasons, and different weather conditions. The images of waves captured at set time intervals can be arranged and displayed in time series to form a dynamic wave change process, which helps researchers to more intuitively observe and analyze the periodicity, morphological changes and other characteristics of waves, and provide strong support for fields such as wave forecasting, marine engineering design and marine environmental monitoring. This step captures a frame of wave images at set time intervals from the wave video data, and corresponds it to the true value data of the wave period according to the time series relationship. This additional technical feature not only enhances the representativeness of the data and ensures the consistency of the time series, but also improves the efficiency of data processing, optimizes the model training effect, and facilitates data visualization and analysis.

[0070] The above-mentioned acquisition of white-capped wave value data includes: identifying the white-capped wave value in the target area based on the Segnet algorithm, and acquiring the pixel point value of the white-capped wave value.

[0071] Among them, the SegNet algorithm is a deep convolutional neural network structure for image segmentation, which is particularly suitable for semantic segmentation tasks. It follows the segmentation idea of ​​the fully convolutional network (FCN) and has been improved and optimized. The core of SegNet lies in its encoder network, the corresponding decoder network, and a pixel-level classification layer. Its network structure includes an encoder, a decoder, and a pixel-level classification layer. The encoder part extracts features through the convolution layer and records the pooling index in the maximum pooling step. The VGG16 network with the fully connected layer removed is usually used as the basis for feature extraction. It contains 5 stages and a total of 13 convolution layers. Each stage consists of a convolution layer and a pooling layer. After each stage, the size of the feature map is halved. The decoder part is symmetrical with the encoder part, and also contains 5 stages and a total of 13 convolution layers (or a combination of upsampling and convolution). The maximum pooling index recorded by the encoder part is used to perform nonlinear upsampling to restore image information. Each stage contains an upsampling layer and a convolution layer. After each stage, the size of the feature map doubles. The pixel-level classification layer is located after the decoder and is used to output the probability of each pixel in all categories, where the highest probability category is the predicted value of the pixel, thereby achieving semantic segmentation of the image. In the information transmission between the encoder and the decoder, SegNet effectively restores the resolution through upsampling and deconvolution operations, retains the spatial position and context information, and helps to improve the accuracy of segmentation, especially when dealing with complex and changing image scenes. SegNet uses the storage of the maximum pooling position index to achieve fast upsampling of the decoder, reducing memory usage and improving the inference speed. The lightweight upsampling method can not only improve the image boundary segmentation effect, but also perform well in end-to-end real-time segmentation projects. In addition, SegNet can automatically learn the appearance and shape changes of objects in various scenarios through training on a large amount of data. It performs well in dealing with complex and changing image scenes and has a wide range of applicability.

[0072] In the implementation example, when applying SegNet, the conv in segnet is changed to group convolution. Compared with ordinary convolution, although the number of parameters and operations are the same, group convolution can generate g times the number of feature maps. Therefore, group conv can generate a large number of feature maps with a small number of parameters and operations, and obtain more feature information. Figure 2As shown in the figure, g is used as a control parameter for the number of groups, with a minimum value of 1 and a maximum value of the number of channels of the input feature map. When g = 1, it is a normal convolution, and when g = the number of channels of the input feature map, it is a depthwise sepereable convolution. And an attention mechanism is added after the conv+batch normalisation+relu module, namely the CBAM module. The CBAM module consists of two attention modules: the Channel Attention Module and the Spatial Attention Module. The input of channel attention is a feature F of H*W*C (H*W is the pixel size, C is the number of channels). First, perform a global maximum pooling and average pooling in a space to obtain two 1*1*C channel descriptions, and then send them to a two-layer neural network. The number of neurons in the first layer is C / r, where r is the dimension reduction coefficient, the activation function is Relu, and the number of neurons in the second layer is C. The two layers of the neural network are shared. Finally, add the two features and pass through a Sigmoid activation function to get the weight coefficient Mc. Finally, Mc*input feature F can get the scaled new feature. Similar to channel attention, given a feature F1 of H*W*C, first perform a channel dimension maximum pooling and average pooling to get two H*W*1 channel descriptions, and splice these two descriptions together according to the channel, and then pass through a 7*7 convolution layer, the activation function is sigmoid, get the weight coefficient MS, and finally MS*input feature F1 to get the scaled new feature.

[0073] Furthermore, the embodiment selects the area where the white-headed waves are located in the target wave picture as the target area, including: identifying the white-headed waves from all the target wave pictures, and screening out the target wave pictures with the white-headed waves as screening pictures; respectively selecting at least one distribution area with the white-headed waves in each screening picture, and calculating whether the density value of the white-headed waves in the distribution area is greater than the set density value; and defining the distribution area with a density value greater than the set density value as the target area.

[0074] In the above, by identifying white-capped waves from all target wave images and screening out images with white-capped waves, it is possible to ensure that the data basis for subsequent analysis is accurate and relevant. This step effectively excludes images that do not contain white-capped waves, reduces data noise, and improves the purity and representativeness of the data set. At least one distribution area with white-capped waves in each screened image is framed separately. This step achieves accurate capture of the location of white-capped waves. The framed area provides a clear spatial range for subsequent calculation of density values ​​and further analysis, which helps to more accurately evaluate the characteristics and distribution of white-capped waves. By calculating the density value of white-capped waves in the distribution area and comparing it with the set density value, areas with dense white-capped waves can be further screened out, which helps to identify areas with the most significant white-capped waves and the most obvious characteristics, providing more valuable data points for subsequent analysis.

[0075] This step can significantly reduce the amount of data that needs to be processed and improve analysis efficiency by accurately selecting the target area, focusing on areas with dense whitecaps, helping to quickly identify key features and accelerate the research process. Accurate target area selection reduces data noise and interference factors, improves the reliability of analysis results, and the density-based screening criteria ensure that the selected areas are representative and can more accurately reflect the characteristics and distribution patterns of whitecaps. Accurate target areas provide strong support for in-depth research on the formation mechanism, influencing factors, and relationship with other marine phenomena of whitecaps, and help reveal the intrinsic connection between whitecaps and environmental factors such as wave cycle, wind speed, and water depth, providing a scientific basis for fields such as wave forecasting, marine engineering design, and marine environmental monitoring. And by accurately selecting the target area, the utilization of computing resources and storage resources can be optimized, reducing research costs.

[0076] 103. Calculate and obtain wave cycle estimates based on the target wave image and white-cap wave value data, and construct a wave cycle recognition model of wave true values ​​and wave cycle estimates.

[0077] Try different deep learning models to calculate the wave cycle valuation, and combine physical models and statistical data to improve the accuracy and stability of the wave cycle valuation. A sufficiently large data set can be used for model training to improve the generalization ability of the model, and cross-validation, regularization and other techniques can be used to prevent model overfitting, monitor model performance in real time, and adjust and optimize the model based on feedback. The wave cycle identification method based on deep learning has broad application prospects and can provide strong technical support for fields such as marine scientific research, marine engineering design and marine environmental monitoring. By continuously optimizing data acquisition, image processing, wave cycle valuation calculation and model training, the prediction accuracy of the wave cycle can be further improved.

[0078] In this step, the wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, including: obtaining the main frequency by fast Fourier transform of the wave true value and the white-cap wave value data corresponding to each target wave image within a preset time length; calculating the inverse of the main frequency, and using the inverse of the main frequency as the wave cycle estimation.

[0079] The main frequency is obtained by fast Fourier transform (FFT) of the wave true value and white-cap wave value data corresponding to each target wave image within the preset time. This step utilizes the advantages of FFT in the field of signal processing and can efficiently analyze the spectral characteristics of wave signals; the main frequency reflects the main frequency component of wave fluctuations, and its reciprocal is the wave period estimation. This conversion process is based on the basic physical principle of wave fluctuations, ensuring the accuracy of wave period estimation. This step combines the target wave image (image data) and white-cap wave value data (numerical data) to achieve the fusion analysis of image and numerical data. Image data provides intuitive information about wave fluctuations, while numerical data provides more specific quantitative information. The combination of the two helps to understand the characteristics of waves more comprehensively. The calculation process from image data to wave period estimation is automatically completed through the algorithm, avoiding manual intervention and cumbersome data processing steps, which not only improves the calculation efficiency, but also reduces human errors, making the acquisition of wave period estimation faster and more accurate. Accurate wave period estimation helps to improve the accuracy of wave forecasts, especially in the fields of marine engineering and shipping safety. Accurate wave period information is crucial for formulating countermeasures and ensuring safety. The analysis method combining images and numerical data enhances the reliability of data, avoids the errors and uncertainties that may be caused by a single data source, helps to improve the scientificity and credibility of wave research, and provides strong support for relevant policy formulation and decision-making. Through automated and efficient calculation methods, this technical feature optimizes the use of computing resources and reduces research costs and time costs.

[0080] Embodiment 2

[0081] See also Figure 3 , Figure 3 Schematic diagram of a wave cycle identification device based on deep learning disclosed in an embodiment of the present invention. Figure 3As shown, the wave cycle recognition device based on deep learning may include: a data acquisition module 301, an image processing module 302 and a model construction module 303, wherein the data acquisition module 301 is used to collect wave data within a preset time period at intervals of the preset time period, and the wave data includes wave video data and wave cycle true value data; the wave video data and the wave cycle true value data are correspondingly associated according to a time series relationship; the image processing module 302 is used to perform image processing on the wave data to obtain a target wave image, and obtain white-cap wave value data; the model construction module 303 is used to calculate based on the target wave image and the white-cap wave value data to obtain a wave cycle valuation, and construct a wave cycle recognition model of wave true value and wave cycle valuation.

[0082] In the embodiment, the preset time of the data acquisition module 301 can be preferably 60 seconds. By collecting data of multiple wave cycles within 60 seconds, the system can calculate a more accurate wave cycle estimate, which helps to improve the accuracy and reliability of the wave cycle identification model and provide more reliable decision support for related applications. The preset duration of 60 seconds is relatively easy to implement, because most data acquisition devices and systems can collect and store data at this time interval, which helps to reduce the complexity and cost of system implementation. Therefore, choosing 60 seconds as the preset duration for collecting wave data brings benefits and advantages in terms of data representativeness, computing efficiency, resource optimization, flexibility, real-time, reliability and ease of implementation.

[0083] In the image processing module 302, the wave data is image processed to obtain the target wave picture, including: intercepting a number of target wave pictures from the wave video data, and corresponding each target wave picture to the corresponding wave cycle true value data; performing regional division on the target wave pictures, and selecting the area where the white-headed waves in the wave picture are located as the target area. To intercept a number of target wave pictures from the wave video data, it is necessary to have technology that can process continuous video data and select representative or specific conditional wave pictures from them. The intercepted pictures correspond to the corresponding wave cycle true value data, ensuring that each picture has corresponding physical parameters as a reference. Regional division of the target wave picture is an effective image segmentation method that can simplify complex wave images into multiple areas that are easier to process and analyze. The area where the white-headed waves in the wave picture are located is selected as the target area based on the significance and importance of the white-headed waves in the waves. The white-headed waves usually represent the active and high-energy parts of the waves.

[0084] Furthermore, the above-mentioned method of intercepting a number of target wave pictures from the wave video data and making each target wave picture correspond to the corresponding wave period true value data includes: intercepting a frame of wave picture from the wave video data at every set time interval; and making each frame of wave picture correspond to the wave period true value data according to the time sequence relationship.

[0085] Acquiring the white-headed wave value data includes: identifying the white-headed wave value in the target area based on the Segnet algorithm, and obtaining the pixel point value of the white-headed wave value. Selecting the area where the white-headed waves are located in the target wave picture as the target area includes: identifying the white-headed waves from all the target wave pictures, and screening the target wave pictures with white-headed waves as the screening pictures; respectively selecting at least one distribution area with white-headed waves in each screening picture, and calculating whether the density value of the white-headed waves in the distribution area is greater than the set density value; defining the distribution area with a density value greater than the set density value as the target area. By accurately selecting the target area, the amount of data to be processed can be significantly reduced, the analysis efficiency can be improved, and the focus can be placed on areas with dense white-headed waves, which helps to quickly identify key features and accelerate the research process. Accurate target area selection reduces data noise and interference factors, improves the reliability of the analysis results, and the density-based screening criteria ensure that the selected area is representative and can more accurately reflect the characteristics and distribution patterns of white-headed waves. The precise target area provides strong support for in-depth research on the formation mechanism, influencing factors and relationship between whitecaps and other marine phenomena, helps to reveal the intrinsic relationship between whitecaps and environmental factors such as wave cycle, wind speed and water depth, and provides a scientific basis for fields such as wave forecasting, marine engineering design and marine environmental monitoring. And by accurately selecting the target area, the use of computing resources and storage resources can be optimized, reducing research costs.

[0086] In the model building module 303, the wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, including: obtaining the main frequency through fast Fourier transform of the wave true value and the white-cap wave value data corresponding to each target wave image within the preset time; calculating the reciprocal of the main frequency, and using the reciprocal of the main frequency as the wave cycle estimation. The analysis method combining image and numerical data enhances the reliability of the data, avoids the errors and uncertainties that may be caused by a single data source, helps to improve the scientificity and credibility of wave research, and provides strong support for relevant policy formulation and decision-making. Through automated and efficient calculation methods, this technical feature optimizes the utilization of computing resources and reduces research costs and time costs.

[0087] Embodiment 3

[0088] See also Figure 4 , Figure 4 Schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device may be a computer, a server, etc. Of course, in certain circumstances, it may also be a smart device such as a mobile phone, a tablet computer, a monitoring terminal, and an image acquisition device with processing functions. Figure 4 As shown, the electronic device may include:

[0089] A memory 401 storing executable program codes;

[0090] a processor 402 coupled to the memory 401;

[0091] Among them, the processor 402 calls the executable program code stored in the memory 401 to execute part or all of the steps in the wave cycle identification method based on deep learning in Example 1.

[0092] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute part or all of the steps in the wave cycle identification method based on deep learning in Embodiment 1.

[0093] An embodiment of the present invention also discloses a computer program product, wherein when the computer program product runs on a computer, the computer executes part or all of the steps in the wave cycle identification method based on deep learning in Example 1.

[0094] An embodiment of the present invention also discloses an application publishing platform, wherein the application publishing platform is used to publish a computer program product, wherein when the computer program product runs on a computer, the computer executes part or all of the steps in the wave cycle identification method based on deep learning in embodiment one.

[0095] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the processes does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0096] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0097] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0098] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several requests for a computer device (which can be a personal computer, a server or a network device, etc., specifically a processor in a computer device) to perform some or all of the steps of the method described in each embodiment of the present invention.

[0099] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0100] A person skilled in the art may understand that part or all of the steps in the various methods of the embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, wherein the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0101] The wave cycle identification method, device, electronic device and storage medium based on deep learning disclosed in the embodiments of the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A wave cycle identification method based on deep learning, characterized in that: include: Collecting wave data within a preset time period at intervals of the preset time period, wherein the wave data includes ocean wave video data and wave cycle true value data; The ocean wave video data and the wave period true value data are correspondingly associated according to a time series relationship; Performing image processing on the wave data to obtain a target wave image and obtain white-cap wave value data; The wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, and a wave cycle recognition model of the wave true value and the wave cycle estimation is constructed.

2. The wave cycle identification method according to claim 1, characterized in that: The preset duration is 60 seconds.

3. The wave cycle identification method according to claim 2, characterized in that: Performing image processing on the wave data to obtain a target wave image includes: A number of target wave images are captured from the wave video data, and each target wave image is matched with the corresponding wave period true value data; The target wave images are divided into regions, and the region where the white-capped waves are located in the wave image is selected as the target region.

4. The wave cycle identification method according to claim 3, characterized in that: Several target wave images are captured from the wave video data, and each target wave image is matched with the corresponding wave cycle true value data, including: Capture a frame of ocean wave picture at every set time interval from the ocean wave video data; Each frame of the wave image is matched with the true value data of the wave cycle according to the time sequence relationship.

5. The wave cycle identification method according to claim 3, characterized in that: Get white head wave value data, including: Based on the Segnet algorithm, the white-headed spray value in the target area is identified and the pixel point value of the white-headed spray value is obtained.

6. The wave cycle identification method according to claim 3, characterized in that: The area where the white-capped waves are located in the target wave image is selected as the target area, including: Identify white-capped waves from all target wave images, and select target wave images with white-capped waves as screening images; Select at least one distribution area with white-headed waves in each of the screened pictures, and calculate whether the density value of the white-headed waves in the distribution area is greater than a set density value; The distribution area with density values ​​greater than the set density value is defined as the target area.

7. The wave cycle identification method according to claim 1, characterized in that: The wave cycle estimation is obtained by calculation based on the target wave image and the white-cap wave value data, including: The wave true value and white-cap wave value data corresponding to each target wave image within a preset time period are obtained through fast Fourier transform to obtain the main frequency; The reciprocal of the main frequency is calculated and used as the wave period estimation.

8. A wave cycle identification device based on deep learning, characterized in that: include: Data collection module: used for collecting wave data within a preset time period at intervals of the preset time period, wherein the wave data includes ocean wave video data and wave cycle true value data; the ocean wave video data and the wave cycle true value data are correspondingly associated in a time series relationship; Image processing module: used for performing image processing on the wave data to obtain target wave images and obtain white-cap wave value data; Model building module: used to calculate and obtain wave cycle estimation based on the target wave image and white-cap wave value data, and to build a wave cycle recognition model of wave true value and wave cycle estimation.

9. An electronic device, characterized in that: include: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the wave cycle identification method based on deep learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the wave cycle identification method based on deep learning as described in any one of claims 1 to 7.

Images