A wave parameter determination method, device, apparatus and storage medium

Abstract

This invention discloses a method, apparatus, device, and storage medium for determining wave parameters. The method includes: acquiring multiple historical water surface images from multiple water areas; determining wave surface height field labels corresponding to each historical water surface image based on a teacher network; inputting the historical water surface images and their corresponding wave surface height field labels into a student network; determining a multi-constraint joint loss function based on the wave surface height field labels corresponding to each historical water surface image, the historical wave surface height fields determined by the student network, and the confidence level; updating the student network based on backpropagation until the loss function converges, resulting in a trained student network; acquiring a sequence of water surface images from the current water area; inputting the left and right image sequences included in the water surface image sequence into the trained student network to enable the student network to output a wave surface height field sequence; reconstructing a continuous wave surface geometric model based on the wave surface height field sequence; and determining wave parameters by analyzing the continuous wave surface geometric model.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of marine engineering technology, and in particular to a method, apparatus, device and storage medium for determining wave parameters. Background Technology

[0002] Ocean wave parameter measurement plays a crucial role in offshore operations. Traditional methods for measuring ocean wave parameters include wave buoy method and radar measurement method. However, these traditional wave measurement methods have disadvantages such as high cost, limited spatial coverage and difficulty in deployment.

[0003] With the rapid development of computer hardware and photogrammetry technologies, video measurement methods, which acquire live video images of the sea surface using video image sensors and then calculate ocean wave parameters using image processing and analysis techniques, are gradually gaining attention. Video measurement methods can determine wave parameters by analyzing the acquired sea surface images using wave analysis algorithms.

[0004] However, sea surface images in complex marine environments are complex, and it is difficult to analyze them using a single algorithm, resulting in poor reliability of wave parameters determined by directly analyzing sea surface images. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for determining wave parameters to address the problem of poor reliability of existing technologies in complex marine environments.

[0006] In a first aspect, embodiments of the present invention provide a method for determining wave parameters, including:

[0007] Multiple historical water surface images were acquired in multiple water areas, and wave surface height field labels corresponding to each historical water surface image were determined based on a teacher network constructed by the WaveAcquisition Stereo System (WASS).

[0008] The historical water surface images and their corresponding wavefront height field labels are input into a student network constructed by an Attention Augmented Pyramid Stereo Network (AAPSN). A multi-constraint joint loss function is determined based on the wavefront height field labels corresponding to each historical water surface image, the historical wavefront height field corresponding to each historical water surface image determined by the student network, and the confidence level. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and its labels.

[0009] Acquire a sequence of water surface images of the current water area, wherein the water surface image sequence is a binocular image sequence, including a left image sequence and a right image sequence;

[0010] The left image sequence and the right image sequence are input into the trained student network so that the student network processes the left image sequence and the right image sequence to obtain the wavefront height field sequence;

[0011] A continuous wavefront geometry model is reconstructed based on the wavefront height field sequence. Wave parameters are determined by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

[0012] The technical solution of this invention provides a wave parameter determination method, comprising: acquiring multiple historical water surface images in multiple water areas; determining wave surface height field labels corresponding to each historical water surface image based on a teacher network constructed by WASS; inputting the historical water surface images and corresponding wave surface height field labels into a student network constructed by AAPSN; determining a multi-constraint joint loss function based on the wave surface height field labels corresponding to each historical water surface image, the historical wave surface height fields corresponding to each historical water surface image determined by the student network, and the confidence level; updating the student network based on backpropagation until the multi-constraint joint loss function converges, thereby obtaining a trained student network; aligning the intermediate layer features of the student network with the corresponding layer features of the teacher network through a mapping function; and the multi-constraint joint loss function being... The wave height field sequence is determined by weighting and summing the confidence-weighted height supervision loss of historical wave height fields and wave height field labels, the logarithmic power spectral density consistency loss, the feature-level knowledge distillation loss, the slope consistency loss, the temporal continuity loss of historical wave height fields, and the dispersion relation physical constraint loss. A sequence of water surface images of the current water area is acquired; this sequence is a binocular image sequence, including a left image sequence and a right image sequence. The left and right image sequences are input into a trained student network to process them and obtain the wave height field sequence. A continuous wave surface geometric model is reconstructed based on the wave height field sequence. Wave parameters, including significant wave height, peak period, and main wave direction, are determined by analyzing the continuous wave surface geometric model. The aforementioned technical solution utilizes a teacher network and a student network trained through a multi-constraint learning mechanism that integrates confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, temporal continuity loss, and dispersion relation physical constraint loss. This student network inherits the wavefront height field determination capability of the teacher network and further enhances this capability through training with multiple losses, enabling it to determine a reliable, real-time wavefront height field. A sequence of water surface images of the current water area is acquired; this sequence consists of two binocular images, one left and one right, containing more detailed information about the current water area. The left and right image sequences are input into the trained student network, which processes them to obtain a wavefront height field sequence, achieving real-time and accurate output of the wavefront height field corresponding to the current water area. Based on the continuous wavefront height field sequence, a continuous wavefront geometric model can be reconstructed. By analyzing the reconstructed continuous wavefront geometric model, the wave parameters of the current water area can be determined, achieving real-time and accurate wave parameter determination based on the precise wavefront height field, thus improving the reliability of wave parameter determination.

[0013] Furthermore, after acquiring multiple historical water surface images from multiple water areas, the process also includes:

[0014] At least one of the following enhancement methods is applied to each of the historical water surface images: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain enhanced historical water surface images;

[0015] Accordingly, after acquiring the water surface image sequence of the current water area, the following is also included:

[0016] The water surface image sequence is enhanced by at least one of the following enhancement methods: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain an enhanced water surface image sequence.

[0017] Furthermore, the process of enhancing the historical water surface images with haze includes:

[0018] The historical water surface image is input into the teacher network, which processes the historical water surface image to obtain a historical scene depth map; based on the historical scene depth map, the historical water surface image is enhanced with haze to obtain a historical water surface haze-enhanced image.

[0019] Accordingly, the process of enhancing the haze in the water surface image sequence includes:

[0020] The water surface image sequence is input into the teacher network so that the teacher network processes the water surface image sequence to obtain a scene depth map; based on the scene depth map, the water surface image sequence is enhanced with haze to obtain a water surface haze-enhanced image sequence.

[0021] Furthermore, based on the teacher network constructed by WASS, the wavefront height field labels corresponding to each of the historical water surface images are determined, including:

[0022] Dense stereo matching and triangulation are performed on the historical water surface image to obtain the three-dimensional point cloud corresponding to the historical water surface image;

[0023] By fitting a reference plane to the 3D point cloud using a self-supervised geometric plane, each 3D point in the 3D point cloud is projected onto the reference plane to obtain the wavefront height field label corresponding to each 3D point in the 3D point cloud.

[0024] Furthermore, the confidence-weighted height supervision loss is determined based on the wavefront height field label and the historical wavefront height field; the logarithmic power spectral density consistency loss is determined based on the label power spectral density determined by the wavefront height field label and the historical power spectral density determined by the historical wavefront height field; the feature-level knowledge distillation loss is determined based on the intermediate features corresponding to the historical water surface image determined by the student network and the corresponding layer features corresponding to the historical water surface image determined by the teacher network; the slope consistency loss is determined based on the label slope determined by the wavefront height field label and the historical slope determined by the historical wavefront height field; the temporal continuity loss is determined based on the historical wavefront height field and the N wavefront height fields corresponding to the previous N frames of the historical water surface image, where N is a natural number; and the dispersion relation physical constraint loss is determined based on the energy spectrum corresponding to the historical wavefront height field.

[0025] Furthermore, the process of updating the student network based on backpropagation also includes:

[0026] The degree of anomaly of the historical wavefront height field corresponding to the historical water surface image is determined based on the confidence level corresponding to the historical water surface image, the spectral consistency error between the historical wavefront height field corresponding to the historical water surface image and the wavefront height field label, and the physical residual.

[0027] When the degree of anomaly is determined to be greater than the anomaly threshold, the wavefront height field label is updated based on the teacher network, or the weights of each loss in the multi-constraint joint loss function are updated.

[0028] Furthermore, a continuous wavefront geometry model is reconstructed based on the wavefront height field sequence, and wave parameters are determined by analyzing the continuous wavefront geometry model, including:

[0029] The continuous wavefront height field sequence is stitched together to reconstruct the continuous wavefront geometry model;

[0030] Wavefront feature analysis, peak / trough detection, and spatiotemporal spectrum feature analysis are performed on the continuous wavefront geometric model to obtain the effective wave height, peak period, and main wave direction.

[0031] Secondly, embodiments of the present invention also provide a wave parameter determination device, comprising:

[0032] The first acquisition module is used to acquire multiple historical water surface images in multiple water areas and determine the wave surface height field label corresponding to each historical water surface image based on the teacher network constructed by WASS.

[0033] The training module is used to input the historical water surface images and the corresponding wavefront height field labels into a student network constructed by AAPSN. Based on the wavefront height field labels corresponding to each historical water surface image and the historical wavefront height field and confidence level determined by the student network for each historical water surface image, a multi-constraint joint loss function is determined. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and the wavefront height field labels.

[0034] The second acquisition module is used to acquire a sequence of water surface images of the current water area, wherein the water surface image sequence is a binocular image sequence, including a left image sequence and a right image sequence;

[0035] The execution module is used to input the left image sequence and the right image sequence into the trained student network, so that the student network processes the left image sequence and the right image sequence to obtain a wavefront height field sequence;

[0036] The determination module is used to reconstruct a continuous wavefront geometry model based on the wavefront height field sequence, and to determine wave parameters by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

[0037] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0038] At least one processor; and a memory communicatively connected to said at least one processor;

[0039] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the wave parameter determination method as described in any of the first aspects.

[0040] Fourthly, embodiments of the present invention also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the wave parameter determination method as described in any of the first aspects.

[0041] Fifthly, this application provides a computer program product including computer instructions that, when executed on a computer, cause the computer to perform the wave parameter determination method as provided in the first aspect.

[0042] It should be noted that the aforementioned computer instructions may be stored, in whole or in part, on a computer-readable storage medium. This computer-readable storage medium may be packaged together with the processor of the wave parameter determination device, or it may be packaged separately from the processor of the wave parameter determination device; this application does not impose any limitations on this.

[0043] The descriptions of the second, third, fourth, and fifth aspects in this application can be referred to the detailed description of the first aspect; and the beneficial effects of the descriptions of the second, third, fourth, and fifth aspects can be referred to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0044] In this application, the name of the wave parameter determination device does not limit the device or functional module itself. In actual implementation, these devices or functional modules may appear under other names. As long as the function of each device or functional module is similar to that of this application, it falls within the scope of the appended claims and their equivalents.

[0045] These or other aspects of this application will become more readily apparent in the following description. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of a wave parameter determination method provided in an embodiment of the present invention;

[0048] Figure 2 A flowchart of another wave parameter determination method provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of a wave parameter determination device provided in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0051] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0052] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0053] The terms "first" and "second," etc., used in the specification and drawings of this application are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0054] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0055] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subroutine, etc. Moreover, embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0056] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0057] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0058] Figure 1This is a flowchart of a wave parameter determination method provided by an embodiment of the present invention. This embodiment is applicable to situations requiring the determination of wave parameters in complex aquatic environments. The method can be executed by a wave parameter determination device, such as... Figure 1 As shown, the specific steps include the following:

[0059] Step 110: Acquire multiple historical water surface images in multiple water areas, and determine the wave surface height field label corresponding to each historical water surface image based on the teacher network constructed by WASS.

[0060] A body of water can be understood as a certain spatial area covered by water, which can include inland waters and sea areas. A water surface image can be understood as an image of the water surface corresponding to a body of water.

[0061] WASS can acquire high-precision 3D sea surface data through dense stereo matching and triangulation. The teacher network constructed by WASS is an analytical reconstruction model based on binocular geometry, dense matching, and triangulation. Its inputs are synchronously acquired binocular images and binocular calibration parameters, which may include resolution, sampling frequency, observation height, baseline, and synchronization accuracy. The outputs include 3D point clouds and free surface height field labels.

[0062] Specifically, multiple historical water surface images can be acquired from multiple water areas. Specifically, at least one water surface image can be acquired from each of the multiple water areas within multiple historical time periods, resulting in multiple historical water surface images. Since WASS requires binocular images, all historical water surface images are binocular images, including left and right images. For each historical water surface image, the image is input into a teacher network constructed by WASS. The teacher network performs dense stereo matching and triangulation on the left and right images included in the historical water surface image to determine the corresponding 3D point cloud, and then determines the wavefront height field label.

[0063] The teacher network can first perform dense stereo matching on the left and right images, then triangulate based on binocular intrinsics and baselines to obtain a 3D point cloud in the camera coordinate system. Then, it fits a reference sea level on the 3D point cloud through self-supervised geometric plane fitting, and determines the signed distance from each point in the 3D point cloud to the reference sea level as the free surface height corresponding to that point, i.e., wavefront height field label.

[0064] In this embodiment of the invention, multiple historical water surface images are acquired from multiple water areas to achieve diversified acquisition of water surface images. By inputting each historical water surface image into a teacher network constructed by WASS, the teacher network can determine the wave height field corresponding to the historical water surface image. The wave height field corresponding to the historical water surface image can be used as a label to train the student network, thereby achieving the acquisition of diversified training samples required to train the student network.

[0065] Step 120: Input the historical water surface images and corresponding wavefront height field labels into the student network constructed by AAPSN. Determine the multi-constraint joint loss function based on the wavefront height field labels corresponding to each historical water surface image and the historical wavefront height field and confidence level determined by the student network for each historical water surface image. Update the student network based on backpropagation until the multi-constraint joint loss function converges, and obtain the trained student network.

[0066] In this system, the intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The student network is constructed using AAPSN, which includes a feature extraction module, a feature fusion module, an attention enhancement module, a pyramid feature aggregation module, and a decoding and output module.

[0067] The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss of historical wavefront height field and wavefront height field label, the logarithmic power spectral density consistency loss, the feature-level knowledge distillation loss and the slope consistency loss, as well as the temporal continuity loss of historical wavefront height field and the physical constraint loss of dispersion relation.

[0068] Specifically, historical water surface images and corresponding wave height field labels are input into a student network constructed by AAPSN. The student network can determine the historical wave height field and confidence level corresponding to the historical water surface image based on the feature extraction module, feature fusion module, attention enhancement module, pyramid feature aggregation module, and decoding and output module. Specifically, the feature extraction module can first extract features from the left and right images included in the historical water surface image, and then the feature fusion module can fuse the extracted features. Then, the attention enhancement module can combine channel attention and spatial attention to enhance the peaks, valleys and texture regions. Furthermore, the pyramid feature aggregation module can fuse multi-scale contextual information. Finally, the historical wave height field can be output based on the wave height field output head of the decoding and regression output module, and the confidence level can be output based on the confidence level output head of the decoding and regression output module.

[0069] It should be noted that the student network sets up an independent confidence branch on the basis of shared coding features, which is then convolved and activated by Sigmoid to output a pixel-wise confidence map with values ​​ranging from (0,1].

[0070] The multi-constraint joint loss function is determined based on the wave height field labels corresponding to historical water surface images, the historical wave height field corresponding to historical water surface images determined by the student network, and the confidence level. This multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wave height field. Therefore, the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, and slope consistency loss can be determined based on the wave height field labels corresponding to historical water surface images, the historical wave height field, and the confidence level of the historical wave height field. Furthermore, the temporal continuity loss and dispersion relation physical constraint loss can be determined based on the historical wave height field corresponding to historical water surface images. Finally, the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, temporal continuity loss, and dispersion relation physical constraint loss are weighted and summed to obtain the multi-constraint joint loss function.

[0071] During the training of the student network, the student network can be updated based on backpropagation until the multi-constraint joint loss function converges. The student network at which the multi-constraint joint loss function converges is determined as the trained student network.

[0072] During training, the confidence score does not require separate human labels, but is obtained adaptively through confidence score-weighted highly supervised loss.

[0073] In this embodiment of the invention, a multi-constraint joint loss function is determined based on the wavefront height field label corresponding to the historical water surface image and the historical wavefront height field and confidence level determined by the student network. The student network is then updated based on backpropagation until the multi-constraint joint loss function converges, thereby training the student network and obtaining a trained student network.

[0074] Step 130: Obtain the water surface image sequence of the current water area.

[0075] Specifically, the current water area can be a sea area, and the water surface image sequence can include a left image sequence and a right image sequence. Therefore, the current water area can be continuously image acquired based on the binocular image acquisition device to obtain the water surface image sequence of the current water area, which means the left image sequence and the right image sequence of the current water area can be obtained.

[0076] In this embodiment of the invention, a sequence of water surface images of the current water area is obtained, and real-time image acquisition of the current water area is achieved through the image sequence. The image accuracy is improved through binocular images.

[0077] Step 140: Input the left image sequence and the right image sequence into the trained student network so that the student network can process the left image sequence and the right image sequence to obtain the wavefront height field sequence.

[0078] Specifically, the left and right image sequences are input into the previously trained student network. The student network has a high accuracy in determining the real-time wavefront height field. Therefore, the student network can output the corresponding wavefront height fields in chronological order, and the output of the corresponding wavefront height fields in chronological order can form a wavefront height field sequence.

[0079] In this embodiment of the invention, a pre-trained student network is used to determine the wave height field sequence corresponding to the current water surface image sequence, thereby achieving real-time and accurate determination of the wave height field of the current water area and reducing the difficulty of determining the wave height field.

[0080] Step 150: Reconstruct a continuous wavefront geometric model based on the wavefront height field sequence, and determine wave parameters by analyzing the continuous wavefront geometric model.

[0081] Among them, the continuous wave surface geometric model can be understood as a continuous smooth wave surface description of water surface waves. It is a two-dimensional free surface and a time-evolving surface. The wave parameters include significant wave height, peak period and main wave direction.

[0082] Specifically, a continuous wavefront geometric model can be reconstructed based on the wavefront height field sequence. Specifically, a three-dimensional wavefront point set corresponding to each time moment can be constructed based on the wavefront height field corresponding to each time moment contained in the wavefront height field sequence. Then, adjacent grid points in the three-dimensional wavefront point set at the same time moment can be connected into a triangular network to obtain the wavefront spatial geometric model corresponding to each time moment. Then, a continuous wavefront geometric model corresponding to the continuous time moment can be constructed based on the wavefront spatial geometric model corresponding to each time moment.

[0083] After constructing a continuous wavefront geometry model, wave parameter analysis can be performed based on this model. Wave parameters can include significant wave height, peak period, and dominant wave direction. Specifically, the significant wave height can be determined by performing spectral analysis on the wavefront height field within the time window corresponding to the continuous wavefront geometry model. The peak period can also be determined by the peak frequency of the power spectral density in the time direction. This can be achieved by performing a Fourier transform on the wavefront height field within the time window corresponding to the continuous wavefront geometry model and determining the peak period based on the Fourier transform result. Alternatively, the dominant wave direction can be determined using the circular mean method of the two-dimensional wavenumber spectrum. This involves performing a two-dimensional Fourier transform on the wavefront height field within the time window corresponding to the continuous wavefront geometry model in space to obtain a two-dimensional wavenumber spectrum. After determining the direction angle based on the two-dimensional wavenumber spectrum, the dominant wave direction can be determined based on the direction angle.

[0084] In this embodiment of the invention, a continuous wave surface geometric model is reconstructed based on a continuous time wave surface height field sequence. The wave parameters of the current water area are determined by analyzing the reconstructed continuous wave surface geometric model. This enables the determination of real-time accurate wave parameters based on the real-time accurate wave surface height field, thereby improving the reliability of wave parameter determination.

[0085] The wave parameter determination method provided in this invention includes: acquiring multiple historical water surface images in multiple water areas; determining wave surface height field labels corresponding to each historical water surface image based on a teacher network constructed by WASS; inputting the historical water surface images and corresponding wave surface height field labels into a student network constructed by AAPSN; determining a multi-constraint joint loss function based on the wave surface height field labels corresponding to each historical water surface image, the historical wave surface height fields corresponding to each historical water surface image determined by the student network, and the confidence level; updating the student network based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is obtained by mapping the historical wave surface height field labels corresponding to each historical water surface image, the teacher network, and the wave surface height field labels corresponding to each historical water surface image determined by the teacher network. The wave height field sequence is determined by weighted summation of the confidence-weighted height supervision loss of the historical wave height field and wave height field labels, the logarithmic power spectral density consistency loss, the feature-level knowledge distillation loss, the slope consistency loss, the temporal continuity loss of the historical wave height field, and the dispersion relation physical constraint loss. A sequence of water surface images of the current water area is acquired; this sequence is a binocular image sequence, including a left image sequence and a right image sequence. The left and right image sequences are input into a trained student network to process them and obtain the wave height field sequence. A continuous wave surface geometric model is reconstructed based on the wave height field sequence. Wave parameters, including significant wave height, peak period, and main wave direction, are determined by analyzing the continuous wave surface geometric model. The aforementioned technical solution utilizes a teacher network and a student network trained through a multi-constraint learning mechanism that integrates confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, temporal continuity loss, and dispersion relation physical constraint loss. This student network inherits the wavefront height field determination capability of the teacher network and further enhances this capability through training with multiple losses, enabling it to determine a reliable, real-time wavefront height field. A sequence of water surface images of the current water area is acquired; this sequence consists of two binocular images, one left and one right, containing more detailed information about the current water area. The left and right image sequences are input into the trained student network, which processes them to obtain a wavefront height field sequence, achieving real-time and accurate output of the wavefront height field corresponding to the current water area. Based on the continuous wavefront height field sequence, a continuous wavefront geometric model can be reconstructed. By analyzing the reconstructed continuous wavefront geometric model, the wave parameters of the current water area can be determined, achieving real-time and accurate wave parameter determination based on the precise wavefront height field, thus improving the reliability of wave parameter determination.

[0086] Figure 2 This is a flowchart of another wave parameter determination method provided by an embodiment of the present invention. This embodiment is a specific modification based on the above embodiment. Figure 2As shown, in this embodiment, the method may further include:

[0087] Step 210: Acquire multiple historical water surface images in multiple water areas, and determine the wave surface height field label corresponding to each historical water surface image based on the teacher network constructed by WASS.

[0088] Specifically, historical water surface images can be binocular water surface images. Therefore, binocular vision devices can be installed on shore-based or offshore platforms in multiple water areas. These devices can acquire at least one water surface image in each of the multiple historical time periods, resulting in multiple historical water surface images. Furthermore, the corresponding wavefront height field label can be determined for each historical water surface image.

[0089] In one implementation, the wavefront height field label corresponding to each historical water surface image is determined based on a teacher network constructed by WASS, including: performing dense stereo matching and triangulation on the historical water surface image to obtain a three-dimensional point cloud corresponding to the historical water surface image; and projecting each three-dimensional point in the three-dimensional point cloud onto the reference plane by fitting a reference plane to the three-dimensional point cloud through self-supervised geometric plane fitting to obtain the wavefront height field label corresponding to each three-dimensional point in the three-dimensional point cloud.

[0090] The image coordinates of the left camera in the binocular vision device are: The image coordinates of the right camera are After performing binocular target calibration and epipolar correction on the binocular vision device, the disparity of each pixel in the left and right images acquired by the binocular vision device is determined. .

[0091] Specifically, for each historical water surface image, the historical water surface image can be input into the teacher network, which determines the wavefront height field label corresponding to the historical water surface image based on the WASS algorithm. Specifically, firstly, dense stereo matching can be performed on the historical water surface image, and then triangulation can be performed based on binocular intrinsic parameters and baseline to obtain the 3D points corresponding to each pixel in the historical water surface image in the camera coordinate system. , , Where (u, v) represents the coordinates of each pixel, and f x f y Indicates the camera's focal length. B represents the principal point coordinates, Z represents the binocular baseline length, and Z represents the base point coordinates. c Representing depth, this determines the corresponding 3D point cloud (x, y, z) for historical water surface images. T It can be represented as .

[0092] After determining the 3D point cloud corresponding to the historical water surface images, a reference sea level was fitted using a random sampling consensus algorithm combined with the least squares method, and the reference sea level equation was determined to be Π: n T P+b=0, where n=(n x n y n z ) T , where represents the normal vector to the reference sea level, b represents the displacement parameter of the reference sea level (which can also be understood as the signed offset of the plane relative to the origin of the camera coordinate system), and P represents each 3D point in the 3D point cloud. n determines the projection direction of the height, and b determines the zero position of the reference sea level; together, they define the sea level reference datum. The wavefront height field label can be understood as the signed normal distance of a 3D point relative to the reference sea level. After normalizing the normal vector to ||n||=1, any 3D point P i The signed distance to the reference sea level is the wavefront height field label η corresponding to that 3D point. WASS (P i )=n T P i +b. Of course, if the normal vector is not normalized, the wavefront height field label should be written as η. WASS (P i )=(n T P i +b) / ||n||.

[0093] In practical applications, by jointly fitting a reference sea level over several consecutive frames and removing wave crests, occlusion points, and outliers, more stable normal vectors and displacement parameters can be obtained.

[0094] In this embodiment of the invention, diverse acquisition of historical water surface images and diverse acquisition of training samples required for training student networks are achieved.

[0095] Step 220: Perform at least one of the following enhancements on each historical water surface image: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain enhanced historical water surface images.

[0096] The enhanced historical water surface images include enhanced left images and enhanced right images.

[0097] In complex marine environments, stereo image acquisition devices typically produce stereo water surface images of poor quality and low accuracy, resulting in poor model training performance when directly based on these images. Therefore, image enhancement of historical water surface images is necessary.

[0098] Specifically, historical water surface images can be enhanced using at least one of the following methods: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain corresponding enhanced historical water surface images and construct various complex sea surface visual degradation scenes. Furthermore, in any enhancement dimension, the left and right images included in the historical water surface image share corresponding physical or statistical parameters and are synchronously transformed in conjunction with scene depth maps or disparity relationships to maintain binocular stereo geometric consistency.

[0099] When enhancing historical water surface images with haze, haze degradation modeling can be performed on the historical water surface images to determine the corresponding enhanced historical water surface images.

[0100] In one embodiment, the process of enhancing historical water surface images with haze includes: inputting historical water surface images into a teacher network so that the teacher network processes the historical water surface image sequence to obtain a historical scene depth map; and enhancing the historical water surface images with haze based on the historical scene depth map to obtain a historical water surface haze-enhanced image.

[0101] Specifically, haze enhancement requires the use of scene depth maps. Therefore, it is first necessary to determine the scene depth map corresponding to historical water surface images. After determining the 3D point cloud, the projection matrix of the binocular image acquisition device can be used to project each 3D point in the 3D point cloud onto the corresponding image. The intrinsic parameter matrix and rotation matrix of the left camera are then used to project each 3D point in the 3D point cloud onto the left image, resulting in... By using the intrinsic parameter matrix and rotation matrix of the right camera, each 3D point in the 3D point cloud is projected onto the right image, resulting in... Where P represents a three-dimensional point, This represents the pixel point P in the left image. K represents the pixel point corresponding to the 3D point P in the right image. L K represents the intrinsic parameter matrix of the left camera in a binocular vision device. R Let I represent the intrinsic parameter matrix of the right camera in the binocular image acquisition device, where I represents the 3×3 identity matrix. This can be written in matrix form as follows: When the left camera coordinate system is chosen as the reference coordinate system, the left camera does not rotate relative to itself. The rotation matrix is ​​the identity matrix I, where 0 represents the 3×1 zero vector, which can be written as a column vector. When the origin of the left camera coordinate system is taken as the reference origin, the left camera does not translate relative to itself, and the translation vector is the zero vector. R represents the rotation matrix from the left camera coordinate system to the right camera coordinate system, and τ represents the translation vector from the left camera coordinate system to the right camera coordinate system.

[0102] The scene depth map corresponding to historical water surface images can be understood as the axial depth of each 3D point in the 3D point cloud corresponding to the historical water surface image in the corresponding camera coordinate system, which can be represented as d.L (u, v) = Z L (u, v), d R (u, v) = Z R (u, v), where Z L This represents the depth value of a 3D point in the Z-direction within the left camera coordinate system. R This represents the depth value of a 3D point in the Z direction within the right camera coordinate system. If multiple 3D points are projected onto the same pixel location, the z-buffer rule is used to retain the depth of the nearest point; if there is no valid projection point at the pixel location, a dense depth map is generated through neighborhood interpolation or hole filling.

[0103] Furthermore, based on the scene depth map d corresponding to the left image, L (u, v) and the scene depth map d corresponding to the right image R (u, v) Models haze degradation on historical water surface images. Specifically, the scene depth map d corresponding to the left image can be used. L (u, v) and the scene depth map d corresponding to the right image R (u, v) are uniformly represented as First, the transmittance can be calculated based on the scene depth map. Where β is the extinction coefficient, and secondly, historical water surface haze enhancement images can be determined based on atmospheric scattering models. The atmospheric light term is represented by A(1-T(x)), which characterizes the background light component introduced by atmospheric scattering. As the scene depth increases, the transmittance decreases, and the proportion of the atmospheric light term increases, thus forming a more obvious haze appearance. For a normalized image, A can be a scalar value in the range of (0~1), or a three-channel vector A=[Ar, Ag, Ab], which is estimated from the mean of the highlighted area of ​​the input image or the mean of the global brightness. To maintain the consistency of binocular radiative degradation, the left and right images share the parameter {A,β}. When I(x) represents the left image, I'(x) represents enhancing the left image, and when I(x) represents the right image, I'(x) represents enhancing the right image.

[0104] When enhancing historical water surface images in low light and with noise reduction, brightness attenuation and composite noise perturbation can be applied to simulate cloudy days, nighttime, low-light acquisition, and the effects of sensor noise, resulting in the enhanced historical water surface image. Where I represents historical water surface images, Indicates the luminance attenuation coefficient. This represents Gaussian additive noise. I represents Poisson multiplicative noise. When I represents the left image, I' represents enhancing the left image; when I represents the right image, I' represents enhancing the right image.

[0105] When enhancing glare and sea surface reflection in historical water surface images, to simulate solar reflection, bright sea surface flares, and local specular reflection interference, multiple Gaussian flare kernels can be superimposed on the historical water surface image to obtain the corresponding enhanced historical water surface image. The enhanced historical water surface image obtained by enhancing glare and sea surface reflection in historical water surface images can be represented as follows: Where I(x) represents the historical water surface image, and x represents the pixels in the historical water surface image. Indicates the Gaussian flare nucleus. , , These represent the weight, mean, and covariance matrices, respectively. This indicates the number of flares. The left and right images share flare locations and are offset by parallax. Specifically, the flare locations in the left and right images are determined using a joint generation method. First, flare centers are randomly generated in the left image. Then, based on the parallax correspondence between the left and right images, the flare centers are mapped to the corresponding positions in the right image, so that the same flare in the left and right images corresponds to the same sea surface area. On this basis, flare distributions with consistent parameters are superimposed on the left and right images respectively to achieve glare and sea surface reflection enhancement that satisfies binocular geometric consistency. When I(x) represents the left image, I'(x) represents the enhanced left image; when I(x) represents the right image, I'(x) represents the enhanced right image.

[0106] When dynamically enhancing historical water surface images through occlusion, to simulate scenarios such as wave splashing, hull component occlusion, water droplet adhesion, and localized obstruction of view, a random occlusion mask can be constructed and applied to the historical water surface image to obtain the corresponding enhanced historical water surface image. The enhanced historical water surface image obtained by dynamically enhancing the historical water surface image can be represented as follows: B represents the fill brightness value, and O represents a random occlusion mask. The mask is a soft mask with a binary value [0, 1]. The masks of the left and right images are offset according to the disparity relationship. Specifically, the mask of the right image is obtained by offsetting the mask of the left image according to the disparity relationship. That is, the occlusion mask is first generated in the left image, and then the occlusion mask in the left image is mapped to the corresponding position in the right image along the epipolar direction according to the disparity value corresponding to the pixel position of the occlusion mask. For occlusion masks with small internal depth changes, a uniform disparity can be used for overall offset. For occlusion masks with large internal depth changes, pixel-by-pixel disparity can be used for point-by-point mapping. After mapping, the occlusion mask of the right image is smoothed at the edges and holes are repaired to ensure that the occlusion masks in the left and right images meet the binocular geometric consistency. When I represents the left image, I' represents enhancing the left image. When I represents the right image, I' represents enhancing the right image.

[0107] When performing local blurring and water droplet defocus enhancement on historical water surface images, to simulate the spatially variable blur caused by lens-attached water droplets, local defocusing due to wet fog, and wave splashing, a spatially variable convolution kernel can be applied to the binocular water surface image to obtain the corresponding enhanced historical water surface image. The enhanced historical water surface image obtained by performing local blurring and water droplet defocus enhancement on the historical water surface image can be represented as follows: Let ξ represent the local offset variable in the convolution integral, x represent a pixel in the historical water surface image, Kx(ξ) represent the spatially variable local blur kernel at pixel x, used to describe the weighted influence of different offset positions in the neighborhood on the imaging result of pixel x, and I(x-ξ) represent the pixel at the neighborhood position x-ξ corresponding to pixel x in the historical water surface image. Since Kx(ξ) varies with pixel x, it can characterize the spatial non-uniform blur effect caused by lens-attached water droplets, local defocusing due to wet fog, and wave splashing. When I(x-ξ) represents the pixel at the neighborhood position x-ξ corresponding to pixel x in the left image of the historical water surface image, I'(x) represents enhancing the left image; when I(x-ξ) represents the pixel at the neighborhood position x-ξ corresponding to pixel x in the right image of the historical water surface image, I'(x) represents enhancing the right image.

[0108] The blurred regions in the left and right images are generated synchronously according to the projection correspondence. Specifically, the blurred region is first generated in the left image, and then the blurred region in the left image is mapped to the corresponding position in the right image along the epipolar direction according to the disparity value corresponding to the blurred region. When the depth change inside the blurred region is small, the average disparity is used for overall offset. When the depth change is large, pixel-by-pixel disparity is used for point-by-point mapping. After mapping, the blurred region in the right image is then filled with holes and the edges are smoothed to ensure that the occluded regions in the left and right images meet the binocular geometric consistency.

[0109] It should be noted that when performing image enhancement, the binocular parallax geometry constraint must be strictly maintained to avoid enhancement compromising the premise of stereo matching.

[0110] In this embodiment of the invention, historical water surface images are enhanced by image enhancement in at least one dimension, resulting in an enhanced historical water surface image including an enhanced left image and an enhanced right image. Furthermore, during the enhancement process, the left and right images share physical or statistical parameters, and are combined with depth maps, disparity relationships, and wavefront geometry quantities for synchronous transformation to maintain stereo geometric consistency.

[0111] Step 230: Input the enhanced historical water surface image and the corresponding wave height field label into the student network constructed by AAPSN. Determine the multi-constraint joint loss function based on the wave height field label corresponding to each enhanced historical water surface image and the historical wave height field and confidence level determined by the student network. Update the student network based on backpropagation until the multi-constraint joint loss function converges to obtain the trained student network.

[0112] Specifically, historical water surface images and corresponding wavefront height field labels are input into a student network constructed by AAPSN. The student network can determine the historical wavefront height field corresponding to the historical water surface image. Then, based on the wavefront height field labels corresponding to the historical water surface image, the historical wavefront height field determined by the student network, and the confidence level, a multi-constraint joint loss function is determined. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss of the historical wavefront height field and wavefront height field labels, the logarithmic power spectral density consistency loss, the feature-level knowledge distillation loss, the slope consistency loss, the temporal continuity loss of the historical wavefront height field, and the dispersion relation physical constraint loss. It can be expressed as follows: ,in, This indicates a confidence-weighted, highly supervised loss. This represents the logarithmic power spectral density uniformity loss. Indicates the loss of time continuity. Represents the physical constraint loss of the dispersion relation. This represents the loss from feature-level knowledge distillation. Indicates the loss of slope consistency, each This represents the weighting coefficient. The confidence-weighted height supervision loss is determined based on the wavefront height field labels and historical wavefront height fields, and can be expressed as... , The confidence level per pixel can be represented as: , of which F θ (x,y) represents the intermediate features of the student network, f c (·) represents the confidence regression head, and σ(·) represents the sigmoid activation function. The higher the confidence level, the more reliable the wavefront height field prediction result at that pixel location. express and The absolute value of the difference This represents the weight coefficient of the confidence regularization term, used to control the regularization strength to prevent all confidence levels from being zero. It takes a positive real number, and the specific value can be set empirically based on the training data. During training, the confidence level does not require separate manual labels; instead, it is adaptively learned through a confidence-weighted highly supervised loss. High-error regions tend to lower the confidence level. To prevent the network from compressing all pixels into low confidence levels.

[0113] The logarithmic power spectral density consistency loss is determined based on the tag power spectral density determined by the wavefront height field tag and the historical power spectral density determined by the historical wavefront height field, and can be expressed as follows: , The power spectral density represents the historical wavefront height field. The power spectral density of the wavefront height field label. This represents a one-dimensional Fourier transform along the time dimension, where f is the time frequency.

[0114] The feature-level knowledge distillation loss is determined based on the intermediate features corresponding to the historical water surface images determined by the student network and the corresponding layer features corresponding to the historical water surface images determined by the teacher network, and can be expressed as follows: , This represents the characteristics of the middle layer of the student network. This represents the corresponding layer features of the teacher network. and This represents a learningable feature dimension alignment projection layer. Since the teacher and student networks differ in the number of feature channels, spatial resolution, and representation, they must first be projected onto a common feature space using a mapping function, and then the Euclidean distance is calculated to achieve distillation.

[0115] The slope consistency loss is determined based on the labeled slope as determined by the wavefront height field label and the historical slope as determined by the historical wavefront height field, and can be expressed as follows: ,in, The historical wavefront height field η is represented by the student network prediction. θ The spatial gradient is used to characterize the local slope variation of the predicted wavefront in both horizontal directions. The wavefront height field label η represents the wavefront height field generated by the teacher network. WASS The spatial gradient is used to characterize the local slope changes of the label wavefront in both horizontal directions, and the slope consistency loss is used to enhance the student network's ability to recover local peaks, troughs and slope changes.

[0116] The temporal continuity loss is determined based on the historical wavefront height field and the N wavefront height fields corresponding to the N preceding frames of the historical water surface image, where N is a natural number. When N=2, the temporal continuity loss is calculated using the wavefront height fields corresponding to the two preceding frames of the historical water surface image. The temporal continuity loss can be expressed as... , This represents the historical wavefront height field corresponding to historical water surface images. This represents the historical wavefront height field corresponding to the previous frame image in the historical water surface image. This represents the historical wavefront height field corresponding to the two preceding frames of the historical water surface image. These are the weighting coefficients.

[0117] The dispersion relation physical constraint loss is determined based on the energy spectrum corresponding to the historical wavefront height field, and can be expressed as: , This is represented as the three-dimensional spacetime Fourier transform in the wavenumber-frequency domain. The energy spectrum, Represents the three-dimensional spacetime Fourier transform. For wavenumber modulus, Indicates the time angular frequency. Represents gravitational acceleration. Indicates water depth.

[0118] Then, the student network can be updated based on backpropagation until the multi-constraint joint loss function converges. The student network at which the multi-constraint joint loss function converges is then determined as the trained student network.

[0119] In one implementation, the process of updating the student network based on backpropagation further includes: determining the degree of anomaly of the historical wavefront height field corresponding to the historical water surface image based on the confidence level corresponding to the historical water surface image, the spectral consistency error between the historical wavefront height field corresponding to the historical water surface image and the wavefront height field label, and the physical residual; when the degree of anomaly is determined to be greater than the anomaly threshold, updating the wavefront height field label based on the teacher network, or updating the weights of each loss in the multi-constraint joint loss function.

[0120] Specifically, anomaly triggering criteria can be constructed based on the confidence level corresponding to historical water surface images, the spectral consistency error between historical wavefront height fields and wavefront height field labels, and the physical residual. This determines the corresponding anomaly level. Furthermore, the anomaly level can be compared with an anomaly threshold. If the anomaly level exceeds the threshold, the teacher network can be called again to update the wavefront height field labels, or the weights of each loss in the multi-constraint joint loss function can be updated. Based on the sparse correction mechanism where the teacher network participates on demand, the student network can be trained to obtain a student network that better meets practical needs.

[0121] In this embodiment of the invention, a multi-constraint joint loss function is determined based on the wavefront height field labels corresponding to historical water surface images and the historical wavefront height field and confidence level determined by the student network. The student network is then updated based on backpropagation until the multi-constraint joint loss function converges, thus training the student network and obtaining a trained student network. By combining feature-level distillation and output-level distillation, the network's ability to inherit the teacher network under degradation conditions is improved, enhancing the stability and real-time performance of the student network in determining the wavefront height field. Furthermore, by introducing logarithmic power spectral density consistency loss and dispersion relation physical constraint loss, constraints are applied in the three-dimensional spatiotemporal Fourier domain, improving the physical consistency of the student network.

[0122] Step 240: Obtain the water surface image sequence of the current water area.

[0123] Specifically, a sequence of water surface images of the current water area can be obtained. This sequence of water surface images can be a sequence of binocular water surface images, which can be represented as follows: ,in, Represents the left image sequence. This represents the right image sequence, and t represents the time frame number.

[0124] In this embodiment of the invention, a sequence of water surface images of the current water area is obtained, and real-time image acquisition of the current water area is achieved through the image sequence. The image accuracy is improved through binocular images.

[0125] Step 250: Perform at least one of the following enhancements on the water surface image sequence: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, and local blur and water droplet defocus enhancement, to obtain an enhanced water surface image sequence.

[0126] The enhanced water surface image sequence includes a left enhanced image sequence and a right enhanced image sequence.

[0127] In one embodiment, the process of enhancing the haze of a water surface image sequence includes: inputting the water surface image sequence into a teacher network so that the teacher network processes the water surface image sequence to obtain a scene depth map; and enhancing the haze of the water surface image sequence based on the scene depth map to obtain a water surface haze-enhanced image sequence.

[0128] Based on steps similar to step 220 above, at least one of the following enhancements can be performed on the water surface image sequence: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain the corresponding enhanced water surface image sequence.

[0129] In this embodiment of the invention, image enhancement in at least one dimension is used to enhance the water surface image sequence, thereby obtaining a corresponding enhanced water surface image sequence and optimizing the image quality of the water surface image sequence.

[0130] Step 260: Input the left enhanced image sequence and the right enhanced image sequence into the trained student network so that the student network can process the left enhanced image sequence and the right enhanced image sequence to obtain the wavefront height field sequence.

[0131] Specifically, the left and right enhanced image sequences are input into the previously trained student network, which can output a wavefront height field sequence, that is, it can output wavefront height fields arranged in chronological order. .

[0132] In this embodiment of the invention, a pre-trained student network is used to determine the wave height field sequence corresponding to the current water surface image sequence, thereby achieving real-time and accurate determination of the wave height field of the current water area and reducing the difficulty of determining the wave height field.

[0133] Step 270: Reconstruct a continuous wavefront geometric model based on the wavefront height field sequence, and determine wave parameters by analyzing the continuous wavefront geometric model.

[0134] In one embodiment, step 270 may specifically include: reconstructing a continuous wavefront geometric model by splicing together a continuous wavefront height field sequence; and performing wavefront feature analysis, wave crest / valley detection, and spatiotemporal spectrum feature analysis on the continuous wavefront geometric model to obtain the effective wave height, peak period, and main wave direction.

[0135] Specifically, firstly, a three-dimensional wavefront point set corresponding to each moment can be constructed based on the wavefront height field corresponding to each moment contained in the wavefront height field sequence. More specifically, the discrete grid coordinate set corresponding to the current water surface can be represented as Ω={(x... i y j | i=1, ..., N x j=1…N y}, x i Let y represent the coordinates of the i-th horizontal grid. j N represents the coordinate of the j-th vertical grid. x N represents the number of grid points in the horizontal direction of the discrete grid. y This indicates the number of grid points in the vertical direction of the discrete grid.

[0136] For any time t, based on the wavefront height field corresponding to time t, a three-dimensional wavefront point set corresponding to time t can be constructed, and S can be obtained. t ={(x i y j η θ (x i y j ,t))},η θ (x i y j ,t) represents grid point (xi y j The wavefront height value at time t. Specifically, the grid coordinates of each grid point can be combined with its corresponding height value to form the three-dimensional wavefront point set S at time t. t Then, adjacent grid points in the three-dimensional wavefront point set at time t can be connected to form a triangular network, yielding the wavefront spatial geometric model corresponding to time t. Specifically, adjacent grid points can be connected to form triangular patches according to the regular grid adjacency relationship, resulting in the three-dimensional wavefront mesh model Mt=Tri(S) corresponding to time t. t ), This represents a regular triangular mesh connection operation on a 3D point set. Furthermore, based on the wavefront spatial geometry model corresponding to each consecutive time step, a continuous wavefront geometry model corresponding to each consecutive time step can be constructed. That is, as time progresses, the sequence of wavefront geometry models at consecutive time steps can be represented as... N t This indicates the number of wavefront height fields contained in the wavefront height field sequence.

[0137] After constructing a continuous wavefront geometry model, wave parameter analysis can be performed based on this model. Wave parameters can include significant wave height, peak period, and dominant wave direction. Specifically, the significant wave height can be determined by performing spectral analysis on the wavefront height field within the time window corresponding to the continuous wavefront geometry model. This can be achieved using the statistical standard deviation method or the spectral moment method. When determining the significant wave height using the statistical standard deviation method, the following can be obtained: ,in, , .

[0138] The peak period can also be determined based on the peak frequency of the time-direction power spectral density of the wavefront height field sequence within the time window corresponding to the continuous wavefront geometry model. Specifically, the peak frequency S(f) of the time-direction power spectral density can be obtained by performing a Fourier transform on the wavefront height field sequence within the time window corresponding to the continuous wavefront geometry model, and then the peak frequency can be determined based on the peak frequency S(f). Therefore, the peak period can be determined based on the peak frequency. .

[0139] The dominant wave direction can also be determined using the circular mean method of two-dimensional wavenumber spectrum. Specifically, a two-dimensional Fourier transform can be performed on the wavefront height field within the time window corresponding to the continuous wavefront geometry model to obtain the two-dimensional wavenumber spectrum. Determine the direction angle based on the two-dimensional wavenumber spectrum Then, by determining the main wave direction based on the azimuth angle, the direction can be determined. .

[0140] In practical applications, wave parameters can also include ridge lines, trough lines, and spatiotemporal energy spectra. Specifically, the local gradient can be obtained by calculating the local gradient, normal vector, and curvature based on a continuous wavefront geometry model. normal vector curvature .

[0141] Furthermore, extremum detection can be performed within the local neighborhood, identifying points satisfying the local maximum condition as candidate peaks and points satisfying the local minimum condition as candidate troughs. Further, by combining curvature thresholds, matrix eigenvalues, or slope continuity, false peaks and troughs can be suppressed, and ridge and trough lines can be extracted. A three-dimensional spatiotemporal Fourier transform can be performed on the wavefront height field sequence to obtain... This allows us to determine the spacetime energy spectrum. Of course, the main propagation direction, main frequency, main wavenumber, local clustering characteristics, and propagation consistency index can also be extracted based on the spatiotemporal energy spectrum.

[0142] In addition, wave crest density, local kurtosis index, wavegroup focusing index, and suspected wave fragmentation events can be determined based on continuous wavefront geometry models, providing data support for engineering monitoring and early warning.

[0143] In this embodiment of the invention, a continuous wave surface geometric model is reconstructed based on a continuous time wave surface height field sequence. The wave parameters of the current water area are determined by analyzing the reconstructed continuous wave surface geometric model. This enables the determination of real-time accurate wave parameters based on the real-time accurate wave surface height field, thereby improving the reliability of wave parameter determination.

[0144] The wave parameter determination method provided in this invention includes: acquiring multiple historical water surface images in multiple water areas; determining wave surface height field labels corresponding to each historical water surface image based on a teacher network constructed by WASS; performing at least one enhancement on each historical water surface image, including haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain enhanced historical water surface images; inputting the enhanced historical water surface images and corresponding wave surface height field labels into a student network constructed by AAPSN; and determining the wave surface height field labels corresponding to each enhanced historical water surface image and the historical wave surface height field labels corresponding to each enhanced historical water surface image determined by the student network. The confidence level determines the multi-constraint joint loss function, and the student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. A sequence of water surface images of the current water area is acquired. At least one of the following enhancement methods is applied to the water surface image sequence: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, and local blur and water droplet defocus enhancement, resulting in an enhanced water surface image sequence. The left and right enhanced image sequences are input into the trained student network to process them and obtain a wavefront height field sequence. A continuous wavefront geometric model is reconstructed based on the wavefront height field sequence, and wave parameters are determined by analyzing the continuous wavefront geometric model. The aforementioned technical solution employs a teacher network and a student network trained using a multi-constraint learning mechanism that integrates confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, temporal continuity loss, and dispersion relation physical constraint loss. The enhanced historical water surface images used for training are obtained through image enhancement, simultaneously satisfying the complex imaging mechanisms at sea and the constraints of binocular stereo geometry. The student network inherits the wavefront height field determination capability of the teacher network and further enhances this capability through training with multiple losses, enabling it to determine a real-time and reliable wavefront height field. The solution involves acquiring a sequence of water surface images of the current water area and applying at least one of the following enhancement methods: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, and local blur and water droplet defocus enhancement. This results in an enhanced water surface image sequence, achieving image quality optimization for the water surface image sequence. The left and right enhanced image sequences included in the enhanced water surface image sequence are input into the trained student network. The student network processes the left and right enhanced image sequences to obtain the wave surface height field sequence, thereby achieving real-time and accurate output of the wave surface height field corresponding to the current water area.Based on the continuous wave surface height field sequence, a continuous wave surface geometric model can be reconstructed. By analyzing the reconstructed continuous wave surface geometric model, the wave parameters of the current water area can be determined. This enables the determination of real-time accurate wave parameters based on the real-time accurate wave surface height field, realizing a systematic process from data construction, image enhancement, continuous wave surface geometric model reconstruction to wave parameter determination, and enhancing the stability, real-time performance, and interpretability of the wave parameter determination method in complex marine environments.

[0145] Figure 3 This is a schematic diagram of a wave parameter determination device provided in an embodiment of the present invention. This device is applicable to situations requiring the determination of wave parameters in complex aquatic environments, improving the real-time performance and accuracy of wave parameters. The device can be implemented through software and / or hardware and is generally integrated into electronic devices, such as computer equipment.

[0146] like Figure 3 As shown, the device includes:

[0147] The first acquisition module 310 is used to acquire multiple historical water surface images in multiple water areas and determine the wave surface height field label corresponding to each historical water surface image based on the teacher network constructed by WASS.

[0148] Training module 320 is used to input the historical water surface images and the corresponding wavefront height field labels into a student network constructed by AAPSN. Based on the wavefront height field labels corresponding to each historical water surface image and the historical wavefront height field and confidence level determined by the student network for each historical water surface image, a multi-constraint joint loss function is determined. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and the wavefront height field labels.

[0149] The second acquisition module 330 is used to acquire a sequence of water surface images of the current water area, wherein the sequence of water surface images is a binocular image sequence, including a left image sequence and a right image sequence;

[0150] Execution module 340 is used to input the left image sequence and the right image sequence into the trained student network, so that the student network processes the left image sequence and the right image sequence to obtain a wavefront height field sequence;

[0151] The determination module 350 is used to reconstruct a continuous wavefront geometry model based on the wavefront height field sequence, and to determine wave parameters by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

[0152] The wave parameter determination device provided in this embodiment acquires multiple historical water surface images in multiple water areas. Based on a teacher network constructed using WASS, it determines the wavefront height field label corresponding to each historical water surface image. The historical water surface images and their corresponding wavefront height field labels are input into a student network constructed using AAPSN. A multi-constraint joint loss function is determined based on the wavefront height field labels corresponding to each historical water surface image, the historical wavefront height field determined by the student network, and the confidence level. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is obtained by analyzing the historical wavefront height field. The wavefront height field sequence is determined by weighted summation of the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field. A sequence of water surface images of the current water area is acquired; this sequence is a binocular image sequence, including a left image sequence and a right image sequence. The left and right image sequences are input into a trained student network to process them and obtain the wavefront height field sequence. A continuous wavefront geometric model is reconstructed based on the wavefront height field sequence. Wave parameters, including significant wave height, peak period, and main wave direction, are determined by analyzing the continuous wavefront geometric model. The aforementioned technical solution utilizes a teacher network and a student network trained through a multi-constraint learning mechanism that integrates confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, temporal continuity loss, and dispersion relation physical constraint loss. This student network inherits the wavefront height field determination capability of the teacher network and further enhances this capability through training with multiple losses, enabling it to determine a reliable, real-time wavefront height field. A sequence of water surface images of the current water area is acquired; this sequence consists of two binocular images, one left and one right, containing more detailed information about the current water area. The left and right image sequences are input into the trained student network, which processes them to obtain a wavefront height field sequence, achieving real-time and accurate output of the wavefront height field corresponding to the current water area. Based on the continuous wavefront height field sequence, a continuous wavefront geometric model can be reconstructed. By analyzing the reconstructed continuous wavefront geometric model, the wave parameters of the current water area can be determined, achieving real-time and accurate wave parameter determination based on the precise wavefront height field, thus improving the reliability of wave parameter determination.

[0153] Based on the above embodiments, the device further includes:

[0154] The enhancement module is used to perform at least one of the following enhancements on each of the historical water surface images after acquiring multiple historical water surface images in multiple water areas: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain an enhanced historical water surface image.

[0155] It is also used to perform at least one of the following enhancements on the water surface image sequence after acquiring the current water area: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain an enhanced water surface image sequence.

[0156] In one embodiment, the process of enhancing the historical water surface image with haze by the enhancement module includes:

[0157] The historical water surface image is input into the teacher network, which processes the historical water surface image to obtain a historical scene depth map; based on the historical scene depth map, the historical water surface image is enhanced with haze to obtain a historical water surface haze-enhanced image.

[0158] The process of enhancing the water surface image sequence by the enhancement module includes:

[0159] The water surface image sequence is input into the teacher network so that the teacher network processes the water surface image sequence to obtain a scene depth map; based on the scene depth map, the water surface image sequence is enhanced with haze to obtain a water surface haze-enhanced image sequence.

[0160] Based on the above embodiments, the first acquisition module 310 is specifically used for:

[0161] Multiple historical water surface images are acquired from multiple water areas; dense stereo matching and triangulation are performed on the historical water surface images to obtain a three-dimensional point cloud corresponding to the historical water surface images; a reference plane corresponding to the three-dimensional point cloud is fitted by self-supervised geometric plane fitting, and each three-dimensional point in the three-dimensional point cloud is projected onto the reference plane to obtain a wavefront height field label corresponding to each three-dimensional point in the three-dimensional point cloud.

[0162] In one implementation, the confidence-weighted height supervision loss is determined based on the wavefront height field label and the historical wavefront height field; the logarithmic power spectral density consistency loss is determined based on the label power spectral density determined by the wavefront height field label and the historical power spectral density determined by the historical wavefront height field; the feature-level knowledge distillation loss is determined based on the intermediate features corresponding to the historical water surface image determined by the student network and the corresponding layer features corresponding to the historical water surface image determined by the teacher network; the slope consistency loss is determined based on the label slope determined by the wavefront height field label and the historical slope determined by the historical wavefront height field; the temporal continuity loss is determined based on the historical wavefront height field and the N wavefront height fields corresponding to the previous N frames of the historical water surface image, where N is a natural number; and the dispersion relation physical constraint loss is determined based on the energy spectrum corresponding to the historical wavefront height field.

[0163] Based on the above embodiments, the training module 320 is specifically used for:

[0164] During the process of updating the student network based on backpropagation, the degree of anomaly of the historical wavefront height field corresponding to the historical water surface image is determined according to the confidence level corresponding to the historical water surface image, the spectral consistency error and physical residual between the historical wavefront height field corresponding to the historical water surface image and the wavefront height field label; when the degree of anomaly is determined to be greater than the anomaly threshold, the wavefront height field label is updated based on the teacher network, or the weights of each loss in the multi-constraint joint loss function are updated.

[0165] Based on the above embodiments, module 350 is specifically used for:

[0166] The continuous wavefront height field sequence is spliced ​​together to reconstruct the continuous wavefront geometric model; wavefront feature analysis, wave crest / valley detection, and spatiotemporal spectrum feature analysis are performed on the continuous wavefront geometric model to obtain the effective wave height, the peak period, and the main wave direction.

[0167] The wave parameter determination device provided in this embodiment of the invention can execute the wave parameter determination method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of executing the wave parameter determination method.

[0168] It is worth noting that in the embodiments of the wave parameter determination device described above, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0169] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 4 A block diagram of an exemplary electronic device 4 suitable for implementing embodiments of the present invention is shown. Figure 4 The electronic device 4 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0170] like Figure 4 As shown, electronic device 4 is represented in the form of a general-purpose computing electronic device. The components of electronic device 4 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0171] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0172] Electronic device 4 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 4, including volatile and non-volatile media, removable and non-removable media.

[0173] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Electronic device 4 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (… Figure 4 Not shown; usually referred to as a "hard drive"). Although Figure 4 As not shown, disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0174] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0175] Electronic device 4 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with electronic device 4, and / or with any device that enables electronic device 4 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 22. Furthermore, electronic device 4 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. Figure 4 As shown, network adapter 20 communicates with other modules of electronic device 4 via bus 18. It should be understood that, although... Figure 4 Not shown, it can be combined with electronic device 4 to use other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0176] Processing unit 16 executes various functional applications and page displays by running programs stored in system memory 28, such as implementing the wave parameter determination method provided in this embodiment of the invention, which includes:

[0177] Multiple historical water surface images were acquired in multiple water areas, and wavefront height field labels corresponding to each historical water surface image were determined based on a teacher network constructed by WASS.

[0178] The historical water surface images and their corresponding wavefront height field labels are input into a student network constructed by AAPSN. A multi-constraint joint loss function is determined based on the wavefront height field labels corresponding to each historical water surface image, the historical wavefront height field corresponding to each historical water surface image determined by the student network, and the confidence level. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and its labels.

[0179] Acquire a sequence of water surface images of the current water area, wherein the water surface image sequence is a binocular image sequence, including a left image sequence and a right image sequence;

[0180] The left image sequence and the right image sequence are input into the trained student network so that the student network processes the left image sequence and the right image sequence to obtain the wavefront height field sequence;

[0181] A continuous wavefront geometry model is reconstructed based on the wavefront height field sequence. Wave parameters are determined by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

[0182] Of course, those skilled in the art will understand that the processor can also implement the technical solution of the wave parameter determination method provided in any embodiment of the present invention.

[0183] This invention provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements, for example, the wave parameter determination method provided in this invention, which includes:

[0184] Multiple historical water surface images were acquired in multiple water areas, and wavefront height field labels corresponding to each historical water surface image were determined based on a teacher network constructed by WASS.

[0185] The historical water surface images and their corresponding wavefront height field labels are input into a student network constructed by AAPSN. A multi-constraint joint loss function is determined based on the wavefront height field labels corresponding to each historical water surface image, the historical wavefront height field corresponding to each historical water surface image determined by the student network, and the confidence level. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and its labels.

[0186] Acquire a sequence of water surface images of the current water area, wherein the water surface image sequence is a binocular image sequence, including a left image sequence and a right image sequence;

[0187] The left image sequence and the right image sequence are input into the trained student network so that the student network processes the left image sequence and the right image sequence to obtain the wavefront height field sequence;

[0188] A continuous wavefront geometry model is reconstructed based on the wavefront height field sequence. Wave parameters are determined by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

[0189] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0190] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0191] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0192] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0193] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0194] Furthermore, the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with relevant laws and regulations.

[0195] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for determining wave parameters, characterized in that, include: Multiple historical water surface images were acquired in multiple water areas, and wave surface height field labels corresponding to each historical water surface image were determined based on a teacher network constructed by the wave acquisition stereo algorithm WASS. The historical water surface images and their corresponding wavefront height field labels are input into a student network constructed by an Attention-Enhanced Pyramid Stereo Reconstruction Network (AAPSN). A multi-constraint joint loss function is determined based on the wavefront height field labels corresponding to each historical water surface image, the historical wavefront height field corresponding to each historical water surface image determined by the student network, and the confidence level. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and its labels. Acquire a sequence of water surface images of the current water area, wherein the water surface image sequence is a binocular image sequence, including a left image sequence and a right image sequence; The left image sequence and the right image sequence are input into the trained student network so that the student network processes the left image sequence and the right image sequence to obtain the wavefront height field sequence; A continuous wavefront geometry model is reconstructed based on the wavefront height field sequence. Wave parameters are determined by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

2. The wave parameter determination method according to claim 1, characterized in that, After acquiring multiple historical water surface images across multiple water areas, the process also includes: At least one of the following enhancement methods is applied to each of the historical water surface images: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain enhanced historical water surface images; Accordingly, after acquiring the water surface image sequence of the current water area, the following is also included: The water surface image sequence is enhanced by at least one of the following enhancement methods: haze enhancement, low light and noise enhancement, glare and sea surface reflection enhancement, dynamic occlusion enhancement, and local blur and water droplet defocus enhancement, to obtain an enhanced water surface image sequence.

3. The wave parameter determination method according to claim 2, characterized in that, The process of enhancing the historical water surface images with haze includes: The historical water surface image is input into the teacher network, which processes the historical water surface image to obtain a historical scene depth map; based on the historical scene depth map, the historical water surface image is enhanced with haze to obtain a historical water surface haze-enhanced image. Accordingly, the process of enhancing the haze in the water surface image sequence includes: The water surface image sequence is input into the teacher network so that the teacher network processes the water surface image sequence to obtain a scene depth map; based on the scene depth map, the water surface image sequence is enhanced with haze to obtain a water surface haze-enhanced image sequence.

4. The wave parameter determination method according to claim 1, characterized in that, Based on the teacher network constructed by WASS, the wavefront height field labels corresponding to each of the historical water surface images are determined, including: Dense stereo matching and triangulation are performed on the historical water surface image to obtain the three-dimensional point cloud corresponding to the historical water surface image; By fitting a reference plane to the 3D point cloud using a self-supervised geometric plane, each 3D point in the 3D point cloud is projected onto the reference plane to obtain a wavefront height field label corresponding to each 3D point in the 3D point cloud.

5. The wave parameter determination method according to claim 1, characterized in that, The confidence-weighted height supervision loss is determined based on the wavefront height field label and the historical wavefront height field; the logarithmic power spectral density consistency loss is determined based on the label power spectral density determined by the wavefront height field label and the historical power spectral density determined by the historical wavefront height field; the feature-level knowledge distillation loss is determined based on the intermediate features corresponding to the historical water surface image determined by the student network and the corresponding layer features corresponding to the historical water surface image determined by the teacher network; the slope consistency loss is determined based on the label slope determined by the wavefront height field label and the historical slope determined by the historical wavefront height field; the temporal continuity loss is determined based on the historical wavefront height field and the N wavefront height fields corresponding to the previous N frames of the historical water surface image, where N is a natural number; the dispersion relation physical constraint loss is determined based on the energy spectrum corresponding to the historical wavefront height field.

6. The wave parameter determination method according to claim 1, characterized in that, The process of updating the student network based on backpropagation also includes: The degree of anomaly of the historical wavefront height field corresponding to the historical water surface image is determined based on the confidence level corresponding to the historical water surface image, the spectral consistency error between the historical wavefront height field corresponding to the historical water surface image and the wavefront height field label, and the physical residual. When the degree of anomaly is determined to be greater than the anomaly threshold, the wavefront height field label is updated based on the teacher network, or the weights of each loss in the multi-constraint joint loss function are updated.

7. The wave parameter determination method according to claim 1, characterized in that, A continuous wavefront geometry model is reconstructed based on the wavefront height field sequence, and wave parameters are determined by analyzing the continuous wavefront geometry model, including: The continuous wavefront height field sequence is stitched together to reconstruct the continuous wavefront geometry model; Wavefront feature analysis, peak / trough detection, and spatiotemporal spectrum feature analysis are performed on the continuous wavefront geometric model to obtain the effective wave height, peak period, and main wave direction.

8. A wave parameter determination device, characterized in that, include: The first acquisition module is used to acquire multiple historical water surface images in multiple water areas and determine the wave surface height field label corresponding to each historical water surface image based on the teacher network constructed by WASS. The training module is used to input the historical water surface images and the corresponding wavefront height field labels into a student network constructed by AAPSN. Based on the wavefront height field labels corresponding to each historical water surface image and the historical wavefront height field and confidence level determined by the student network for each historical water surface image, a multi-constraint joint loss function is determined. The student network is updated based on backpropagation until the multi-constraint joint loss function converges, resulting in a trained student network. The intermediate layer features of the student network are aligned with the corresponding layer features of the teacher network through a mapping function. The multi-constraint joint loss function is determined by weighting and summing the confidence-weighted height supervision loss, logarithmic power spectral density consistency loss, feature-level knowledge distillation loss, slope consistency loss, and the temporal continuity loss and dispersion relation physical constraint loss of the historical wavefront height field and the wavefront height field labels. The second acquisition module is used to acquire a sequence of water surface images of the current water area, wherein the water surface image sequence is a binocular image sequence, including a left image sequence and a right image sequence; The execution module is used to input the left image sequence and the right image sequence into the trained student network, so that the student network processes the left image sequence and the right image sequence to obtain a wavefront height field sequence; The determination module is used to reconstruct a continuous wavefront geometry model based on the wavefront height field sequence, and to determine wave parameters by analyzing the continuous wavefront geometry model, wherein the wave parameters include significant wave height, peak period, and main wave direction.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the wave parameter determination method as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the wave parameter determination method as described in any one of claims 1-7.

Images