A method for constructing and degrading modeling of multi-source data in cross-medium imaging under complex sea conditions

Abstract

This invention discloses a method for constructing and modeling degradation of multi-source data for cross-medium imaging under complex sea conditions, belonging to the field of underwater optical imaging. Addressing the difficulty of existing technologies in uniformly describing geometric distortion caused by sea surface fluctuations and radiation degradation resulting from water absorption and scattering, this invention integrates multi-source data from real-sea observations, controlled laboratory tests, and auxiliary references. It constructs a wave interface function and establishes a refractive distortion displacement field based on Snell's law, while simultaneously building a water radiation degradation model to form a unified degradation expression. Furthermore, it expands the sample space through physically guided data enhancement, generating a dataset with multi-dimensional degradation labels. This invention achieves joint modeling of geometric and radiation degradation in cross-medium imaging, providing a high-fidelity, high-generalization data foundation for image restoration model training.

Description

Technical Field

[0001] This invention relates to the fields of underwater optical imaging and computational imaging, specifically to a method for constructing and modeling multi-source data for cross-medium imaging under complex sea conditions. Background Technology

[0002] Cross-medium imaging is an important means of acquiring information between underwater targets and surface or airborne observation platforms. Its imaging process involves the propagation of light through multiple media: water, wave interface, and air. During this process, random refraction caused by sea surface waves leads to non-rigid distortion of the target's geometry. Furthermore, suspended particles, colored soluble organic matter, and plankton in the water absorb and scatter light, resulting in radiation degradation phenomena such as image brightness attenuation, reduced contrast, and color distortion.

[0003] Existing techniques typically model refractive distortion or absorption scattering separately, lacking a unified data construction and degradation modeling mechanism, making it difficult to accurately reflect the degradation process in cross-medium imaging under complex sea conditions. Furthermore, the high cost and difficulty in acquiring and labeling real-world sea data result in insufficient training data, limiting the generalization ability of data-driven methods. Therefore, it is necessary to propose a data construction and modeling method that can integrate multi-source data and uniformly describe the cross-medium degradation process. Summary of the Invention

[0004] To address the issues of insufficient cross-medium imaging data and inconsistent degradation modeling under complex sea conditions, this invention proposes a method for constructing and modeling multi-source cross-medium imaging data under complex sea conditions. By integrating physical simulation data, laboratory data, and real sea data, and establishing a unified degradation model based on the cross-medium light propagation mechanism, a consistent representation of multi-source data is achieved.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A method for constructing and modeling degradation of multi-source imaging data across media under complex sea conditions includes the following steps:

[0007] Step 1), Multi-source data construction and preprocessing

[0008] First, multi-source cross-medium imaging data were acquired, including real-sea observation data, controlled laboratory data, and auxiliary reference data. The real-sea observation data consisted of cross-medium image sequences collected in the target sea area, while recording environmental parameters, including wave height, wave period, wind speed, water depth, turbidity, and light intensity. The controlled laboratory data consisted of image data collected in a flume environment simulating different wave interfaces and water conditions. The auxiliary reference data consisted of reference images under non-degradable or weakly degradable scenarios.

[0009] Secondly, construct a unified dataset.

[0010] ,

[0011] in Indicates time The acquired degraded image, For pixel coordinates, Indicates a reference image. Rm represents the environment parameter vector, and Rm represents the m-dimensional real space.

[0012] Finally, quality control, missing value handling, and normalization were performed on the multi-source data to eliminate scale differences between different data sources.

[0013] Step 2), Fluctuation Interface Modeling

[0014] To describe dynamic water surface changes under complex sea conditions, a wave interface spatial coordinate system is introduced. Construct the fluctuation interface function:

[0015] ,

[0016] in For spatial coordinates, For amplitude, For wavenumber vectors, Angular frequency, For the initial phase, The number of wave components (spectral components) participating in the superposition;

[0017] Further, calculate the wavefront normal vector:

[0018] ,

[0019] Used to describe the effect of a wave interface on the refraction of incident light rays, where The gradient of the wavefront;

[0020] Step 3), refraction distortion modeling

[0021] Establishing the cross-medium refraction relationship based on Snell's law:

[0022] ,

[0023] in The refractive index of air, Let be the refractive index of water. The angle of incidence on the air side. The angle of refraction on the water side;

[0024] Furthermore, pixel coordinates are defined on the imaging plane. Construct the geometric distortion displacement field caused by refraction:

[0025] ,

[0026] in , Indicates the image position after refraction;

[0027] This displacement field is used to describe the non-rigid spatial distortion of images under complex sea conditions;

[0028] Step 4), Water body radiation degradation modeling

[0029] Based on the absorption and scattering of light by water bodies, a water body radiation degradation model is constructed, which includes a transmittance function and a background light component.

[0030] In water, light propagation is affected by absorption and scattering, and the imaging process can be represented as follows:

[0031] ,

[0032] in Indicates a reference image; Let be the water transmittance function. The water body attenuation coefficient, The length of the light propagation path; For background light component;

[0033] Step 5), Unified modeling of cross-media degradation

[0034] Constructing a unified image degradation model:

[0035] ,

[0036] in, For cross-media degradation images, Represents the geometric distortion displacement field. Indicates reference image Applying a geometric distortion displacement field The resulting distorted image, Let be the water transmittance function. For background light component; This represents the distortion operator induced by wavefront refraction, whose execution logic is as follows: first, apply geometric distortion to the reference image, and then apply radiation degradation to the distorted image;

[0037] This model achieves a unified description of the refractive distortion and absorption scattering degradation processes;

[0038] Step 6), Multi-source data fusion and consistency processing

[0039] Perform unified modeling and alignment of multi-source data to construct a fused dataset:

[0040] ,

[0041] in For cross-media degradation images, Indicates a reference image. Represents the geometric distortion displacement field. Let be the water transmittance function. Let p(t) be the background light component, and p(t) be the environmental parameter.

[0042] And through time alignment, spatial calibration and scale normalization, consistent representation of data from different sources is achieved;

[0043] Step 7), Physical Guided Data Enhancement

[0044] Constructing parametric enhancement methods based on physical models:

[0045] ,

[0046] in For physical parameter modulation function, A set of physical parameters;

[0047] By adjusting the above parameters, data expansion can be achieved under different sea conditions, water quality, and lighting conditions;

[0048] Step 8), Degradation Tag Generation

[0049] Construct the final labeled dataset:

[0050] ,

[0051] This dataset contains observed images, reference images, displacement fields, transmittance maps, background light, and environmental parameters, and can be used for training and evaluating cross-media image restoration models.

[0052] Advantages or beneficial effects of the present invention:

[0053] 1. Unified Representation of Multi-Source Data: This invention constructs a multi-source data fusion framework, including real-sea observations, controlled laboratory data, and auxiliary reference data, to achieve a unified representation of cross-media imaging data, thereby improving data integrity and authenticity;

[0054] 2. Accurate characterization of geometric distortion: By introducing wave interface modeling and refraction distortion modeling methods, based on Snell's law and ray tracing, the accurate characterization of non-rigid geometric distortion under complex sea conditions is achieved;

[0055] 3. Physical modeling of radiation degradation: By constructing a water body radiation degradation model (absorption and scattering), physical modeling of brightness attenuation, contrast reduction and color distortion was achieved, improving the ability to express degradation;

[0056] 4. Unified Degradation Modeling: Construct a unified degradation model to achieve joint expression of geometric distortion and radiation degradation, thereby improving data consistency;

[0057] 5. Physically Guided Data Augmentation: By modulating physical parameters such as water optical parameters, wave parameters, and imaging geometric parameters, the physically guided data augmentation method effectively expands the sample space of complex sea states and improves the data coverage.

[0058] 6. Supporting Deep Learning: A multi-dimensional degradation labeling system was constructed, including observed images, reference images, displacement fields, and transmittance maps, which improved the data's ability to support and generalize the training of deep learning models. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method for constructing and modeling multi-source data for cross-media imaging according to the present invention;

[0060] Figure 2 This is a schematic diagram of the cross-medium imaging degradation modeling framework described in this invention. Detailed Implementation

[0061] The invention will now be described in detail with reference to the accompanying drawings and embodiments. This embodiment uses cross-medium observation of underwater targets in nearshore shallow sea areas as an example to specifically illustrate the method of the invention, but the application scope of the invention is not limited thereto.

[0062] In this specification, "shallow sea area" refers to a nearshore sea area with relatively shallow water depth and significantly affected by wind, waves, and tides. In the fields of marine engineering and marine observation, nearshore sea areas with a water depth of less than approximately 200m are generally referred to as shallow sea. This embodiment selects a nearshore shallow sea area with an average water depth of approximately 10m for data collection, which is only for illustrative purposes and does not constitute a limitation on the scope of application of this invention.

[0063] Example:

[0064] like Figure 1 As shown, the method proposed in this invention includes eight main steps, which are described below in conjunction with... Figure 2 The cross-medium imaging degradation modeling framework shown provides a detailed explanation of each step.

[0065] Step (1), Multi-source data construction and preprocessing

[0066] First, acquire multi-source data for cross-media imaging, specifically including:

[0067] ① Real-world observation data: Underwater color cameras and a surface synchronous imaging system were deployed in the nearshore waters of the target area with an average water depth of 10m and an average wave height of 0.8m. The camera resolution was 1920×1080, the frame rate was 30fps, and continuous data acquisition was carried out for 7 days, obtaining 500 sets of cross-media image sequences. Simultaneously recorded environmental parameters: wave height 0.3m~1.5m, wave period 3s~7s, wind speed 2m / s~10m / s, water depth 10m, turbidity 3NTU~25NTU, and light intensity 5000lux~80000lux.

[0068] ② Controlled laboratory data: In a water tank environment, regular and irregular waves (JONSWAP spectrum) were generated using a wave generator. The height of the regular waves was 0.1m to 0.5m. Different concentrations of Mg(OH)2 particles were added to the water to simulate turbidity of 0 NTU to 30 NTU, and 300 sets of controlled images were collected.

[0069] ③ Auxiliary reference data: Collect 50 sets of clear reference images (scenes with no degradation or slight degradation) under clear, windless / still water conditions.

[0070] Secondly, construct a unified dataset:

[0071] (1)

[0072] in, Represents the observed image, Indicates a reference image. This is a vector of environmental parameters, including wave height, wave period, wind speed, water depth, turbidity, and light intensity. These are pixel coordinates;

[0073] Finally, the multi-source data is preprocessed to eliminate scale differences between different data sources, specifically including:

[0074] ① Quality control: Remove blurry, overexposed / underexposed, or images with severe motion artifacts;

[0075] ② Missing value handling: Environmental parameters are filled in by linear interpolation, such as missing frames for wave height and turbidity;

[0076] ③ Normalization: Image pixel values ​​are normalized to [0,1], and environmental parameters are normalized by maximum-minimum, such as light intensity 5000 lux→0, 80000 lux→1;

[0077] ④ Uniform size: All images and derived data, including displacement field and transmittance maps, are interpolated to a resolution of 1920×1080;

[0078] Preprocessed data such as Figure 2 As shown, enter the geometric distortion modeling layer;

[0079] Step (2), Fluctuation Interface Modeling

[0080] like Figure 2 As shown, the geometric distortion modeling layer first models the wave interface; to describe the dynamic water surface under complex sea conditions, wave interface spatial coordinates are introduced. Construct the fluctuation interface function:

[0081] (2)

[0082] in, For amplitude, The wavenumber vector describes the spatial frequency of the wave. Angular frequency, describing the time frequency of the wave; For the initial phase, The number of wave components (spectral components) participating in the superposition; in this embodiment Component parameters were determined based on the JONSWAP spectrum;

[0083] The wavenumber vector and angular frequency satisfy the deep-water dispersion relation:

[0084] (3)

[0085] Further calculation of wavefront normal vector This is used to describe the effect of a wave interface on the refraction of incident light;

[0086] (4)

[0087] in, The gradient of the wavefront is calculated using the central difference method, and the spatial derivative is calculated. The three-dimensional form of the normal vector has the z-axis pointing vertically upward, ensuring the physical consistency of the direction of light refraction.

[0088] Step (3), refraction distortion modeling

[0089] Establishing the cross-medium refraction relationship based on Snell's law:

[0090] (5)

[0091] Among them, air refractive index The refractive index of water , The angle of incidence on the air side. The angle of refraction on the water side;

[0092] Angle of incidence Determined by the angle between the direction of the ray and the normal vector of the wavefront:

[0093] (6)

[0094] in, The unit vector representing the direction of the incident ray; The wavefront normal vector calculated in step (2);

[0095] Secondly, define pixel coordinates on the imaging plane. Construct the geometric distortion displacement field caused by refraction:

[0096] (7)

[0097] in, The image position after refraction is calculated using the "ray tracing method":

[0098] Starting from the optical center of the camera, passing through the wave interface point After refraction, it intersects with the target plane at an underwater depth of h=2.5m. The displacement field This is used to describe the non-rigid spatial distortion of images under complex sea conditions. In this example, the displacement at the wave crest can reach ±15 pixels, resulting in local distortion.

[0099] Step (4), Water body radiation degradation modeling

[0100] In water, light propagation is affected by absorption and scattering, and the imaging process can be represented as follows:

[0101] (8)

[0102] in, It is a function of transmittance; The background light component is estimated by combining "underwater dark channel prior" with water surface illuminance sensor data to simulate ambient light scattered by the water body; The turbidity attenuation coefficient is 10 NTU, which corresponds to the water body attenuation coefficient in this embodiment. The scope can be extended to ; Given the path length of light propagation, the path of the light ray from the refraction point on the water surface to the underwater target point is solved using geometric relationships.

[0103] Step (5), Unified modeling of cross-media degradation

[0104] Based on the mechanisms of geometric distortion and radiometric degradation, a unified image degradation expression is constructed:

[0105] (9)

[0106] in, For cross-media degradation images, Represents the geometric distortion displacement field. Indicates reference image Applying a geometric distortion displacement field The resulting distorted image, Let be the water transmittance function. For background light component; The twist operator induced by wavefront refraction has the following execution logic:

[0107] ① For the reference image Apply geometric distortion: Here, bilinear interpolation resampling is used to ensure pixel continuity;

[0108] ② For distorted images Applying radiation degradation: ;

[0109] This model achieves a unified description of the degradation processes of "refractive distortion (geometry)" and "absorption scattering (radiation)," ensuring physical consistency.

[0110] Step (6), Multi-source data fusion and consistency processing

[0111] A unified modeling and alignment were performed on 500 sets of real-sea data, 300 sets of data from water tanks, and 50 sets of reference data to construct a fused dataset.

[0112] (10)

[0113] The standardization process includes:

[0114] ① Time alignment: The time resolution of different data sources, such as 30fps for real sea and 25fps for water tank, is aligned through "linear interpolation" to ensure that the environmental parameter p(t) is synchronized with the image frame;

[0115] ② Spatial calibration: By using camera intrinsic parameters (focal length, principal point) and extrinsic parameters (relative pose of water surface and camera), data from different coordinate systems (water surface photography, underwater camera) are unified into the imaging plane coordinate system;

[0116] ③ Scale normalization: All images, displacement fields and transmittance maps are interpolated to 1920×1080 resolution to eliminate spatial scale differences;

[0117] Step (7), Physical Guided Data Enhancement

[0118] A parameterized enhancement method based on a physical model is constructed for the fused dataset. Perform multi-dimensional, multi-level physical guidance data augmentation to generate an augmented dataset:

[0119] (11)

[0120] in For physical parameter modulation function, A set of physical parameters;

[0121] The enhancement methods specifically include the following four levels:

[0122] (7.1) Enhancement of optical parameters of water bodies

[0123] By modulating the water body attenuation coefficient and background light Simulate radiation degradation changes under different water quality conditions:

[0124] ① Turbidity Enhancement: Preserving the original dataset's turbidity Value (e.g.) ) as a benchmark, in Uniform sampling within the range, i.e. The corresponding turbidity range is 5-30 NTU;

[0125] ② Background light enhancement: Enhance the original background light map Perform a scale transformation. ;in The simulation simulated the changes in background scattered light under conditions ranging from low light (approximately 1500 lux) on a cloudy day to strong sunlight (approximately 90000 lux).

[0126] (7.2) Wave parameter enhancement

[0127] By modulating wave component parameters, the geometric distortion characteristics of different sea state levels are simulated:

[0128] ① Wave amplitude enhancement: Enhancement of the original wave amplitude Apply global scaling factor The corresponding sea state grades range from 1 (wave height < 0.1m) to 5 (wave height 1.25~2.5m).

[0129] ② Wave frequency enhancement: Keeping the JONSWAP spectrum unchanged, the characteristic frequencies are enhanced. exist Adjustments within the range, including The original characteristic frequency was used to simulate the changes in the wave spectrum under different wind speeds.

[0130] ③ Wave direction enhancement: Introducing a direction distribution function ,exist The dominant wave direction is randomly adjusted within a range to simulate the effect of wind direction changes on wave propagation direction.

[0131] (7.3) Imaging geometric parameter enhancement

[0132] By modulating the relative positional relationship between the camera and the target, the observation angle and distance range can be expanded.

[0133] ① Enhanced observation distance: Keep the camera height H=4m constant, adjust the target depth h∈[1.0,5.0]m, and calculate the corresponding displacement field d(x);

[0134] ② Enhanced observation angle: Randomly adjust the relative azimuth angle between the camera and the target within the horizontal plane. Recalculate the ray tracing path;

[0135] ③ Imaging scale enhancement: For reference image Perform random scaling This simulates the imaging of targets of different sizes within the field of view.

[0136] (7.4) Enhanced composite degradation

[0137] Based on the combined modulation of parameters (7.1)-(7.3), composite degradation samples with physical consistency are generated:

[0138] ① Parameter combination strategy: From Three to five parameters are randomly selected and modulated simultaneously to generate correlated degradation changes;

[0139] ② Hierarchical enhancement: First, perform mild enhancement on a single parameter, with a variation range of ±20%; then, perform composite enhancement on multiple parameters, with a variation range of ±50%; finally, perform enhancement under extreme conditions, with a variation range of ±100%.

[0140] ③ Physical constraints are maintained: Physical constraints are maintained during the enhancement process, such as background light enhancement under high turbidity conditions and wave amplitude increase under high wind and wave conditions;

[0141] (7.5) Enhanced Sample Generation

[0142] For each set of samples in the original fusion dataset, one or more combinations of the above enhancement strategies are randomly applied, generating 8 to 12 enhanced samples for each set of original samples. The final dataset size is expanded from the original 850 sets to about 10,000 sets, covering a variety of cross-media imaging conditions from calm water to sea state 5, from clear water to high turbidity water, and from low light to strong light.

[0143] Step (8), Degradation Tag Generation

[0144] Building a labeled dataset:

[0145] (12)

[0146] This dataset can be directly used to train cross-media image restoration models, such as U-Net, GAN, and diffusion models, supporting the joint restoration of "geometric distortion + radiometric degradation".

[0147] In summary, this embodiment achieves this through the following... Figure 1 The process shown and Figure 2 The framework shown, by integrating multi-source data including real-sea observations, controlled laboratory data, and auxiliary references, constructs a cross-media imaging data construction and degradation modeling framework covering wave interface modeling, refraction distortion modeling, water body radiation degradation modeling, and a unified degradation representation. Simultaneously, a hierarchical physical-guided data augmentation strategy is employed to effectively expand the sample space under complex sea conditions, ultimately generating a high-quality dataset with multi-dimensional degradation labels. This method provides a physically consistent and comprehensive data foundation for the training and evaluation of cross-media image restoration models, demonstrating good practicality and generalization ability.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing and modeling degradation of multi-source imaging data across media under complex sea conditions, characterized in that, Includes the following steps: Step 1), Multi-source data construction and preprocessing: Acquire multi-source data for cross-media imaging, including real-sea observation data, laboratory controlled data, and auxiliary reference data, and perform quality control, missing value processing, and normalization processing to construct a unified dataset, which includes observation images, reference images, environmental parameter vectors, and pixel coordinates. Step 2), Fluctuation Interface Modeling: Construct a wave interface function and calculate the wave surface normal vector to realize wave interface modeling; Step 3), refraction distortion modeling: Based on Snell's law, a cross-medium refraction relationship is established, and a displacement field of geometric distortion caused by refraction is constructed to realize the modeling of refraction distortion. Step 4), Water body radiation degradation modeling: Based on the absorption and scattering of light by water bodies, a water body radiation degradation model is constructed, which includes a transmittance function and a background light component. Step 5), Unified modeling of cross-media degradation: Based on the geometric distortion displacement field and the water body radiation degradation model, a unified image degradation model is constructed to achieve unified modeling of cross-media degradation. The unified degradation expression includes a distortion operator induced by wavefront refraction. Step 6), Multi-source data fusion and consistency processing: We perform unified modeling and alignment of multi-source data, construct a fused dataset, and achieve consistent representation of data from different sources through time alignment, spatial calibration, and scale normalization. Step 7), Physical boot data augmentation: A parameterized enhancement method is constructed based on a physical model, which expands data by modulating a set of physical parameters. Step 8), Degradation tag generation: We construct a labeled dataset to support the training and evaluation of cross-media image restoration models.

2. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, The fluctuation interface function described in step 2) is constructed as follows: ， in For spatial coordinates, For amplitude, For wavenumber vectors, Angular frequency, For the initial phase, The number of wave components (spectral components) participating in the superposition; Calculate the wavefront normal vector: ， Used to describe the effect of a wave interface on the refraction of incident light rays, where The gradient of the wavefront.

3. The method for constructing and modeling multi-source imaging data under complex sea conditions according to claim 1, characterized in that, The geometric distortion displacement field described in step 3) is constructed as follows: ， in These are the pixel coordinates on the imaging plane. Indicates the corresponding image position after refraction. ; This displacement field is used to describe the non-rigid spatial distortion of images under complex sea conditions; The refractive relationship is determined by Snell's law: ， in The refractive index of air, Let be the refractive index of water. The angle of incidence on the air side. The angle of refraction is the water body side.

4. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, Step 4): In water, light propagation is affected by absorption and scattering, and the imaging process is represented as follows: ， in Indicates a reference image; Let be the water transmittance function. The water body attenuation coefficient, The length of the light propagation path; This is the background light component.

5. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, The unified image degradation model described in step 5) is constructed as follows: ， in For cross-media degradation images, This represents the torsion operator induced by wavefront refraction. Represents the geometric distortion displacement field. Indicates reference image Applying a geometric distortion displacement field The resulting distorted image, Let be the water transmittance function. This is the background light component.

6. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, Step 6) involves unified modeling and alignment of the multi-source data to construct a fused dataset: ， in For cross-media degradation images, Indicates a reference image. Represents the geometric distortion displacement field. Let be the water transmittance function. denoted as the background light component, and p(t) as the environmental parameter.

7. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, The time alignment described in step 6) is achieved through linear interpolation, which unifies the time resolution of different data sources; the spatial calibration unifies the data in different coordinate systems to the imaging plane coordinate system through camera intrinsic and extrinsic parameters; the scale normalization interpolates all images, displacement fields and transmittance maps to a uniform resolution.

8. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, In step 7), a parameterized enhancement method is constructed based on the physical model to enhance the fused dataset. Perform multi-dimensional, multi-level physical guidance data augmentation to generate an augmented dataset: in For physical parameter modulation function, It is a set of physical parameters.

9. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, The physical guidance data enhancement mentioned in step 7) includes enhancement of water optical parameters, wave parameters, imaging geometry parameters, and composite degradation enhancement; (7-1) The enhancement of the optical parameters of the water body is achieved by modulating the water body attenuation coefficient and background light to simulate the changes in radiation degradation under different water quality conditions: Turbidity enhancement: Preserving the original dataset The value is used as a benchmark, in Uniform sampling within the range, i.e. The corresponding turbidity range is 5-30 NTU; Background light enhancement: Enhances the original background light map. Perform a scale transformation. ;in Simulates the change in background scattered light from cloudy, low-light conditions to strong sunlight conditions; (7-2) Wave parameter enhancement simulates the geometric distortion characteristics of different sea state levels by modulating wave component parameters: Wave amplitude enhancement: Enhancement of the original wave amplitude Apply global scaling factor The corresponding sea state levels range from 1 to 5; Wave frequency enhancement: Keeping the JONSWAP spectrum unchanged, the characteristic frequencies are... exist Adjustments within the range, including The original characteristic frequency was used to simulate the changes in the wave spectrum under different wind speeds. Wave direction enhancement: Introducing a direction distribution function ,exist The dominant wave direction is randomly adjusted within a range to simulate the effect of wind direction changes on wave propagation direction. (7-3) The imaging geometry enhancement described above expands the observation angle and distance range by modulating the relative positional relationship between the camera and the target: Enhanced observation distance: Keeping the camera height H constant, adjust the target depth h∈[1.0,5.0]m and calculate the corresponding displacement field d(x); Enhanced observation angle: Randomly adjust the relative azimuth angle between the camera and the target within the horizontal plane. Recalculate the ray tracing path; Imaging scale enhancement: for reference image Perform random scaling This simulates the imaging of targets of different sizes within the field of view. The composite degradation enhancement is achieved by randomly selecting 3 to 5 parameters from (7-1) to (7-3) and simultaneously modulating them to generate composite degradation samples with physical consistency.

10. The method for constructing and modeling multi-source imaging data across media under complex sea conditions according to claim 1, characterized in that, The labeled dataset described in step 8) is constructed as follows: ， To merge datasets, To enhance the dataset; The labeled dataset includes observed images, reference images, displacement fields, transmittance maps, background light, and environmental parameters.

Images