Seabed sediment classification mapping method based on shallow stratum profile data

Abstract

The invention relates to the field of seabed sediment classification mapping, and provides a seabed sediment classification mapping method based on shallow stratum profile data, which comprises the following steps: acquiring sensor data of a target seabed, including multi-beam topographic data, side-scan image data, seismic data and shallow stratum profile data; multi-dimensional acoustic-terrain-seismic features are constructed, acoustic feature matching is carried out on multi-beam terrain data and side-scan image data, seismic data are introduced, and an extended hierarchical model is constructed; constructing a preliminary submarine topography model diagram by using a spatial interpolation algorithm, repairing a preliminary missing region by using a generative adversarial network, generating a complete submarine topography model diagram, embedding multi-dimensional acoustic-topography-seismic features, and predicting a sediment migration trend; and generating a seabed substrate classification planar graph and a three-dimensional geological profile map, and performing seabed substrate classification mapping in combination with a sediment migration trend. According to the invention, the resolution and integrity of the submarine topography model can be significantly improved, and the dynamic classification map reflecting the current and future patterns is generated.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of seabed sediment classification mapping, in particular to a seabed sediment classification mapping method based on shallow stratigraphic profile data. BACKGROUND

[0002] With the development of acoustic technology, especially the application of sonar equipment, humans can detect seabed topography and sediment type through sound wave reflection, which makes the detection of seabed topography and sediment type change from traditional physical sampling identification to acoustic detection. Using the high-resolution water depth data provided by the multi-beam sounding system to obtain rich sound intensity data, and using artificial excitation seismic waves to record their propagation path and time, through the combination of sound intensity data and seismic wave propagation path and time for deep geological exploration and research, it plays an important role in oil and gas resource exploration. At present, through high-precision water depth measurement, collection of backscattering intensity data, combined with topographic features for sediment classification, high-resolution seabed geomorphology images are generated to identify bedrock, round gravel, sand, silt and other geological features, which has become an important research direction of marine scientific research.

[0003] As for the prior art, the acoustic parameters such as sound velocity and amplitude used for seabed sediment classification are affected by many factors such as sediment composition, porosity and water content, and the traditional inversion method relies on artificial experience, with low classification accuracy and poor consistency; the collaborative analysis of shallow stratigraphic profile data and multi-beam topography, side scan sonar and other data lacks systematic methods, and it is difficult to construct a high-precision three-dimensional geological model. Moreover, the existing technology mostly focuses on static geological stratification, and cannot effectively simulate the influence of sediment transport under the action of marine dynamics (such as sand wave migration, liquefaction) on engineering structures. SUMMARY

[0004] The present disclosure aims to at least solve one of the problems existing in the prior art, and provides a seabed sediment classification mapping method based on shallow stratigraphic profile data.

[0005] The present disclosure provides a seabed sediment classification mapping method based on shallow stratigraphic profile data, which comprises: Collecting sensor data of the target seabed, the sensor data including multi-beam topography data, side scan image data, seismic data and shallow stratigraphic profile data; Using the sensor data, constructing multi-dimensional acoustic-topographic-seismic features, matching acoustic features of the multi-beam topography data and the side scan image data, introducing the seismic data into the matched acoustic features, and constructing an extended stratification model; Based on the extended hierarchical model, a preliminary seafloor topography model map is constructed using a spatial interpolation algorithm. A generative adversarial network is then used to repair the missing areas in the preliminary seafloor topography model map, generating a complete seafloor topography model map. The multidimensional acoustic-topographic-seismic features are then embedded into the complete seafloor topography model map to predict sediment migration trends. Based on the complete seabed topography model, a seabed sediment classification plan map and a three-dimensional geological profile map are generated. According to the predicted sediment migration trend and the seabed sediment classification plan map and the three-dimensional geological profile map, the seabed sediment is classified and mapped.

[0006] Optionally, the step of constructing multidimensional acoustic-topographic-seismic features using the sensor data includes: Based on the multibeam topographic data, the side-scan image data, and the seismic data, the characteristics of the multibeam topographic data, the acoustic characteristics of the side-scan image, and the characteristics of the seismic data are determined respectively. Based on the contribution of the multibeam topographic data features, the side-scan image acoustic features, and the seismic data features to the seabed sediment classification, the weights corresponding to the multibeam topographic data features, the side-scan image acoustic features, and the seismic data features are calculated using the information gain ratio. Based on the multibeam topographic data features, the side-scan image acoustic features, the seismic data features, and their respective weights, a multidimensional acoustic-topographic-seismic feature is constructed through element-level weighted fusion.

[0007] Optionally, the acoustic feature matching of the multibeam terrain data and the side-scan image data includes: The multibeam terrain data and the side-scan image data are projected onto the same coordinate system, and acoustic feature matching is performed on the features of the multibeam terrain data and the acoustic features of the side-scan image according to the following formula: ; in, Represents the acoustic feature matching coefficient. Indicates the first Feature values ​​of multibeam terrain data, Indicates the first Feature values ​​of each side scan image data Indicates the current acoustic feature matching point and , This represents the total number of acoustic feature matching points. This represents the mean of the feature values ​​of all multibeam terrain data. This represents the mean of the feature values ​​of all side-scan image data.

[0008] Optionally, the acoustic characteristics after matching are introduced into the seismic data to construct an extended hierarchical model, including: The stratum is divided into Mt hierarchical layers according to the hierarchical structure, and the acoustic characteristic vector and the seismic characteristic vector of the mt-th stratum are respectively , ; The prior distribution is assigned to the parameters of the extended hierarchical model, and the model parameter vector corresponding to the mt-th stratum obeys the normal distribution; The likelihood function is defined to evaluate the interpretation of the extended hierarchical model, and the Markov Chain Monte Carlo method is used to sample the posterior distribution based on the Bayesian theorem, and the extended hierarchical model is solved and expressed as , wherein is the optimal parameter set of the posterior distribution.

[0009] Optionally, the extended hierarchical model is used to construct a preliminary seabed topography model map by using a spatial interpolation algorithm, including: The optimal parameter set of the posterior distribution in the extended hierarchical model is coupled and integrated with the discrete seabed topography sampling point data in the spatial dimension; The features that are inherently related to the topographic values are extracted from the sensor data as related features, and the variogram is used to quantify the data spatial correlation and mine the potential relationship between the topographic values and the related features; The related features are used as auxiliary variables of the co-Kriging interpolation algorithm, and the co-Kriging interpolation algorithm is used to predict the interpolation points in the seabed area based on the known topographic values of the sampling points and the potential relationship between the topographic values and the related features, to generate a preliminary seabed topography model map .

[0010] Optionally, the missing area in the preliminary seabed topography model map is repaired by using a generative adversarial network to generate a complete seabed topography model map, including: The generative adversarial network model is trained by a network structure generator, so that the generative adversarial network model learns the complete extended hierarchical model and distinguishes between real topography and repaired areas; The preliminary seabed topography model and the multi-dimensional acoustic-topographic-seismic characteristics are input into the trained generative adversarial network model to repair the missing area, and the repaired continuous topography model output by the trained generative adversarial network model is obtained; The repaired continuous topography model is verified and optimized to generate the complete seabed topography model map.

[0011] ​Optionally, embedding the multidimensional acoustic-topographic-seismic features into the complete seafloor topography model map to predict sediment migration trends includes: By using the multidimensional acoustic-topographic-seismic features, acoustic features, topographic features, and seismic features are obtained respectively, and the acoustic features, topographic features, and seismic features are spatially correlated to construct a mapping relationship of "feature-sedimentary environment-migration potential". Based on the complete seabed topography model, the acoustic features, topographic features, and seismic features are embedded as constraints to build a numerical simulation framework for sediment transport as a transport numerical model. The transport numerical model is used to simulate the sediment transport process and predict sediment transport trends. The sediment migration prediction results of the migration numerical model are optimized using a machine learning model trained based on historical sediment migration data.

[0012] Optionally, the transport numerical model includes: ; in, Indicating sediment transport index, Indicates the first Weights corresponding to multibeam terrain data features Indicates the first The sediment deposition thickness corresponding to each matching point Indicates the weight of ocean current movement. Indicates the first The weights corresponding to the acoustic features of each side scan image Indicates the structural stability assessment value and , Indicates the thickness of the sedimentary layer. Indicates factors influencing seabed sediment. This indicates the average number of reflected signal points after the seismic source is excited. Indicates the acoustic echo intensity of the side scan image. Indicates ocean dynamic factors and , Indicates the sand wave mobility coefficient. Indicates pressure, Represents the velocity vector of the water flow. Indicates the bottom shear stress. Indicates the external force term.

[0013] Optionally, the step of generating a seabed sediment classification plan and a three-dimensional geological profile based on the complete seabed topography model includes: Take the complete seabed topographic model map as a spatial carrier, extract information related to the seabed bottom material in the multi-dimensional acoustic-topographic-seismic features, combine the extracted information with the topography, and construct a seabed bottom material classification rule; Using a spatial analysis algorithm, the grid cells of the complete seabed topographic model map are valued according to the seabed bottom material classification rule, and the seabed bottom material classification plan is generated. Along the preset profile line, the three-dimensional data of the complete seabed topographic model map is extracted, combined with the seabed bottom material classification result and the seismic feature, and the vertical structure of the stratum is reflected to construct the three-dimensional geological profile map.

[0014] Optionally, according to the predicted sediment transport trend and the seabed bottom material classification plan and the three-dimensional geological profile map, the seabed bottom material classification mapping is performed, including: Based on the seabed bottom material classification plan and the three-dimensional geological profile map, the "source-sink" relationship of sediment transport is superimposed, the sediment erosion and transport starting place are marked as the source area on the seabed bottom material classification plan, and the sediment accumulation destination is marked as the sink area, the seabed bottom material classification change is predicted, the seabed bottom material classification boundary is corrected, and the sediment transport to the vertical bottom material layer is reflected through the three-dimensional geological profile map, and finally the dynamic classification mapping integrating the sediment transport trend prediction is generated.

[0015] The present disclosure, compared with the prior art, constructs a multi-dimensional feature system through collaborative analysis of multi-source sensor data (multi-beam topography, side scan image, seismic and shallow stratigraphic profile data); uses a spatial interpolation algorithm to integrate model parameters and sampling data, introduces a generative adversarial network to repair the missing area of the topography, significantly improves the resolution and integrity of the seabed topographic model, and provides high-precision basic data for bottom material classification; by constructing a "feature-sedimentary environment-transport potential" mapping relationship, the multi-dimensional features are used to accurately predict the sediment transport trend; the transport trend and the bottom material classification are integrated to generate a dynamic classification map reflecting the current and future pattern, providing a scientific basis for seabed bottom material classification mapping, thereby solving the problem of insufficient accuracy of traditional methods in seabed topography and bottom material classification, especially in complex topography and variable bottom material conditions, it is difficult to obtain high-resolution and accurate data, and different types of sensor data may not be consistent, and the data quality is easily affected by environmental noise during the acquisition process, resulting in a decline in data quality, and it is difficult to fully consider the complex physical processes involved in sediment transport trend. BRIEF DESCRIPTION OF DRAWINGS

[0016] One or more embodiments are illustrated by way of example in the drawings and are described herein in connection with the embodiments described. These embodiments should not be construed as limiting the scope of the embodiments, as these embodiments are presented solely for illustrative purposes. The drawings are not drawn to scale, and the elements in the drawings are not necessarily to scale with each other.

[0017] Figure 1 A flowchart of a seabed bottom classification mapping method based on shallow stratigraphic profile data is provided for an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the embodiments of the present disclosure will be described in detail below with reference to the drawings. However, those skilled in the art can understand that, in the embodiments of the present disclosure, many technical details are presented in order to make the readers better understand the present disclosure. However, the technical solutions claimed by the present disclosure can be implemented even without these technical details and based on various changes and modifications of the following embodiments. The division of the following embodiments is for the convenience of description, and should not constitute any limitation on the specific embodiments of the present disclosure, and the embodiments can be combined with each other and cited to each other without contradiction.

[0019] One embodiment of the present disclosure relates to a seabed bottom classification mapping method based on shallow stratigraphic profile data, the flowchart of which is shown as Figure 1 including steps S110 to S140.

[0020] In step S110, sensor data of the target seabed is collected, and the sensor data includes multi-beam topographic data, side scan image data, seismic data and shallow stratigraphic profile data.

[0021] Specifically, the multi-beam topographic data can be collected by a multi-beam depth sounder Teledyne SeaBat T50, and at the same time, a Global Navigation Satellite System (GNSS), an Inertial Navigation System (INS), a sound velocity profiler and a surface sound velocity meter are used in cooperation to provide high-precision positioning and attitude compensation, and to monitor the change of water surface sound velocity, measure the vertical distribution of sound velocity in real time, correct the error of sound wave propagation, and provide high-precision seabed topographic information for marine shallow stratigraphic profile research and engineering survey.

[0022] Among them, the multi-beam bathymeter Teledyne SeaBat T50 can be applied to high-precision shallow water (0.3m-200m) detection, and the resolution can reach centimeter level. When in use, the GNSS receiver is used to provide high-precision positioning, and the motion sensor is used to supplement the seabed attitude data. The INS is fixed near the center of gravity of the ship to ensure rigid connection with the multi-beam bathymeter for static calibration and dynamic calibration; the INS and the GNSS are started to synchronize the time stamp, and the position, attitude and motion data are recorded in real time; the GNSS signal quality is checked (the number of satellites is greater than or equal to 6, and the position accuracy decay factor is less than or equal to 2); the stability of the attitude data is monitored (the roll / pitch is less than 5°); at the same time, the GNSS signal shielding is avoided, and in the strong magnetic field area, such as near the large metal structure, the magnetic interference compensation is needed, and the gyro and accelerometer are calibrated regularly, and the cable connection is checked to avoid loosening or wear.

[0023] The sound velocity profiler can use the Valeport MiniSVP type sound velocity profiler, which has the characteristics of high precision and low power consumption, and is suitable for shallow and deep water detection. The sound velocity profile curve with water depth can be obtained, the measurement range is 0m-2000m, the sound velocity measurement error is less than or equal to 0.1m / s, and the sampling frequency is greater than or equal to 1Hz. Before use, check the probe battery and sensor state of the sound velocity profiler to ensure that the probe battery and sensor state are good. When in use, the probe of the sound velocity profiler is lowered to the target depth at a uniform speed to avoid data distortion caused by rapid lowering, and the sound velocity, temperature, salinity and pressure data are recorded in real time. At the same time, check the data continuity, compare the historical sound velocity profile, and verify the data rationality; at the same time, avoid operation in strong current area to prevent probe drift, calibrate sensor, and record measurement time, position and water depth range.

[0024] The surface sound velocity meter can measure the seabed surface area (0m-10m), and the sound velocity measurement error is less than or equal to 0.1m / s, and the sampling frequency is greater than or equal to 1Hz (when measuring the area with rapid water temperature change, higher frequency is needed). Before use, first check the sensor state and cable connection of the surface sound velocity meter to ensure that the sensor state and cable connection are good. When in use, set the sampling frequency and storage mode of the surface sound velocity meter, fix the surface sound velocity meter on the ship body or buoy, ensure that the probe of the surface sound velocity meter is below the water surface, record the data collected by the surface sound velocity meter in real time, compare the data collected by the sound velocity profiler, and verify the consistency. In particular, when using the surface sound velocity meter, it is necessary to avoid operation in strong wind and wave area to prevent probe damage, and when measuring the area with obvious water temperature change, the measurement frequency also needs to be increased.

[0025] When collecting side scan sonar data, a side scan sonar system towed by a ship can be used to emit sound waves to the seabed and receive the reflected signals to form seabed topography and geomorphology images. Specifically, a high-resolution side scan sonar is used to sail along a predetermined survey line, and the side scan sonar system works continuously to record seabed echo signals in real time. Based on these real-time seabed echo signals, seabed topography and geomorphology images can be formed as side scan sonar data.

[0026] Seismic data can be collected by geophones. Specifically, a plurality of geophones are arranged along a survey line, and then a seismic wave is artificially excited, wherein the position and intensity of the seismic source need to be accurately controlled. After the excitation of the seismic source, the reflected or refracted waves generated when the seismic wave propagates to different depth interfaces are recorded to obtain clear reflected or refracted signals and record their propagation paths and times, thereby realizing the detection of the geological structure of the underground seabed and obtaining seismic data of the target seabed.

[0027] Shallow seismic data can be collected by a shallow seismic profiler. The shallow seismic profiler can detect the structure of the stratum several meters to tens of meters below the seabed, image based on the reflection characteristics of sound waves penetrating the stratum, and obtain shallow seismic data based on the reflection characteristics between different strata.

[0028] In step S120, a multi-dimensional acoustic-topography-seismic feature is constructed using the sensor data, acoustic feature matching is performed on the multi-beam topography data and the side scan sonar data, seismic data is introduced into the matched acoustic features, and an extended layered model is constructed.

[0029] In particular, after collecting the sensor data, the sensor data can also be preprocessed to improve the data quality. The preprocessing of the sensor data can specifically include: preprocessing the multi-beam topography data and the side scan sonar data, removing abnormal beams and speckle noise by Lee filtering, correcting the errors of the multi-beam topography data in the seabed propagation process and the effects of slant range distortion and ship motion in the collection process using the sound velocity profiler data. The abnormal beam refers to a data point or beam that deviates from the normal range during the collection process of the multi-beam sonar system, and the causes include equipment failure, environmental interference, data transmission error and reflection anomaly, etc. The speckle noise refers to small area noise randomly distributed in the side scan sonar data, which is usually manifested as bright spots or dark spots in the image, and the causes include scattering effect, instrument noise and environmental factors, etc.

[0030] For example, in step S120, multidimensional acoustic-topographic-seismic features are constructed using sensor data, including: determining multibeam topographic data features, side-scan image acoustic features, and seismic data features respectively based on multibeam topographic data, side-scan image data, and seismic data; calculating the weights corresponding to multibeam topographic data features, side-scan image acoustic features, and seismic data features respectively based on the contribution of multibeam topographic data features, side-scan image acoustic features, and seismic data features to seabed classification using information gain ratio; and constructing multidimensional acoustic-topographic-seismic features through element-level weighted fusion based on multibeam topographic data features, side-scan image acoustic features, seismic data features, and their respective weights.

[0031] Specifically, in the field of seabed sediment classification research, traditional methods are often limited to simple data overlay, failing to fully explore the synergistic potential between multi-source data. This implementation method incorporates three core types of sensor data: multibeam topographic data, side-scan image data, and seismic data. These data are analyzed collaboratively, and corresponding features are extracted for each. The weight of each feature is calculated based on the information gain ratio, and a weighted fusion multidimensional feature system is constructed through element-level weighted fusion to obtain multidimensional acoustic-topographic-seismic features. This overcomes the limitations of traditional methods and achieves deep integration at the feature level.

[0032] Feature extraction is performed on multibeam topographic data to obtain multibeam topographic data features. These features can specifically include reflection intensity, topographic features, and spectral features. Reflection intensity reflects the hardness of the seabed; for example, sand and gravel reflect strongly, while soft algae reflect weakly. Topographic features include slope, roughness, and micro-topography, such as seabed sand waves and channels. Spectral features include the spectral distribution of the beam echo; different seabed substrates typically correspond to different beam echo spectra, for example, sandy substrates correspond to a wider beam echo spectrum, while soft algae correspond to a narrower beam echo spectrum. The multibeam topographic data features are denoted as... , It can be represented as ,in, These are the first features in multibeam terrain data. Each element corresponds to a reflection intensity and topographic features such as topographic height, slope, and aspect, which can reflect the three-dimensional spatial structure of the seabed.

[0033] Feature extraction is performed on side-scan image data to obtain its acoustic features. These features can specifically include texture features, shadow features, and echo intensity. Texture features can be extracted using the gray-level co-occurrence matrix, including contrast, entropy, and correlation. Shadow features mainly include the length and shape of obstacle shadows, reflecting seabed undulations and obstacle height. Echo intensity reflects the surface reflectivity of the seabed; for example, a rough seabed surface has a high echo intensity. The acoustic features of side-scan images can be denoted as... , It can be represented as ,in, The first of the acoustic features of the side scan image is... The system comprises several elements, including texture features, shadow features, echo intensity, and acoustic properties such as grayscale mean, texture roughness, and edge gradient. These features can capture the differences in the material composition of the seabed surface.

[0034] Seismic data features are obtained by feature extraction from seismic data. These seismic data features can be denoted as... , It can be represented as ,in, These are the first features of earthquake data. Each element contains seismic wave propagation parameters such as reflection coefficient, layer velocity, and waveform attenuation, which are used to analyze the geological structure information inside the seabed.

[0035] Before constructing multidimensional acoustic-topographic-seismic features, determine the characteristics of multibeam topographic data. Corresponding weights Acoustic characteristics of side scan images Corresponding weights Earthquake data characteristics Corresponding weights The weights were determined based on the contribution of each feature to seabed sediment classification, using the information gain ratio. As a quantitative indicator, the information gain ratio... Through formula The calculation yielded the following result. Features Information gain is used to measure the amount of effective information provided by a feature during the classification process. Features The entropy represents the uncertainty of the features. It is calculated by analyzing multiple sets of field-collected sample data, allowing... These are multibeam terrain data features Acoustic characteristics of side scan images Earthquake data characteristics each element in the feature vector, i.e. the information gain ratio of each feature is obtained as the corresponding weight, and then normalized to ensure that the sum of the weights satisfies , so as to ensure that the relative importance of each data source in the fusion process is reasonably reflected.

[0036] After obtaining the multi-beam terrain data feature corresponding weight , side scan image acoustic feature corresponding weight , seismic data feature corresponding weight , and then through the element-level weighted fusion operator construct a multi-dimensional acoustic-terrain-seismic feature , and the calculation formula is: This fusion process is not a simple linear superposition of each feature, but a differentiated integration of different types of data according to the weight of each feature, which can effectively suppress redundant information and highlight the key features with higher discrimination for seabed bottom classification, thereby forming a multi-dimensional feature system that is more representative and can better reflect the real properties of seabed bottom, and providing high-quality input data for subsequent classification models.

[0037] For example, in step S120, acoustic feature matching is performed on the multi-beam terrain data and the side scan image data, including: projecting the multi-beam terrain data and the side scan image data into the same coordinate system, and performing acoustic feature matching on the multi-beam terrain data feature and the side scan image acoustic feature according to the following formula: .

[0038] wherein, represents the acoustic feature matching coefficient, and the value range is -1 to 1. High correlation area |r|>0.7 indicates that the acoustic features are consistent, and is used for seabed bottom classification verification. represents the feature value of the i-th multi-beam terrain data (such as reflection intensity). represents the feature value of the i-th side scan image data (such as echo intensity). represents the current acoustic feature matching point and , represents the total number of acoustic feature matching points. represents the mean value of the feature values of all multi-beam terrain data, represents the mean value of the feature values of all side scan image data.

[0039] ​​In particular, by projecting the multi-beam terrain data and the side scan image data into the same coordinate system, spatial consistency can be ensured. In the multi-beam terrain, a high reflection intensity area corresponds to a high echo intensity area in the side scan image. A flat terrain in the multi-beam terrain corresponds to a uniform texture area in the side scan image. A sand wave terrain in the multi-beam terrain corresponds to a striped texture area in the side scan image.

[0040] For example, in step S120, the seismic data is introduced into the matched acoustic features to construct an extended layered model, including: dividing the stratum into Mt levels according to the layered structure, and letting the acoustic feature vector and the seismic feature vector of the mt-th stratum be , The prior distribution is assigned to the parameters of the extended layered model, and the model parameter vector corresponding to the mt-th stratum obeys a normal distribution. The likelihood function is defined to evaluate the explanatory power of the extended layered model. Based on the Bayesian theorem, the Markov chain Monte Carlo method is used to sample the posterior distribution, and the extended layered model is solved and expressed as , wherein is the optimal parameter set of the posterior distribution.

[0041] Specifically, the seismic data is introduced into the matched acoustic features to construct an extended layered model. The extended layered model is based on a layered structure, fully considers the difference and correlation of the stratum in the vertical direction, and divides the entire stratum system into levels for analysis.

[0042] Let the acoustic feature vector of the mt-th stratum in the extended layered model be , which contains key acoustic information such as echo intensity and frequency response collected by the sonar device at the depth of the stratum; and let the seismic feature vector of the mt-th stratum in the extended layered model be , which records important parameters such as the velocity and attenuation coefficient of the seismic wave propagating in the layer.

[0043] Before constructing the extended layered model, the prior distribution needs to be assigned to the model parameters. Here, the prior distribution of the extended layered model is set as , is the parameter of the extended layered model, and . Among them, is the model parameter vector corresponding to the mt-th stratum, which specifically covers key parameters such as weight coefficients and biases that describe the acoustic-seismic feature correlation of the layer. is the local prior probability distribution corresponding to the mt-th stratum, which is a probability assumption of the model parameter vector corresponding to the mt-th stratum. It is assumed that obeys a normal distribution , is the mean of the parameter corresponding to the mt-th formation, representing the initial guess of the parameter when there is no data support, is the variance of the parameter corresponding to the mt-th formation, reflecting the uncertainty of the initial guess.

[0044] Likelihood function is one of the core parts of the extended hierarchical model, which represents the joint probability of observing the acoustic feature vector and the seismic feature vector under the given parameters of the extended hierarchical model . This likelihood function evaluates the model's ability to explain the data by quantifying the degree of matching between the model parameters and the actual observed data.

[0045] Based on Bayes' theorem, parameter estimation is performed through Bayesian inference, and the formula is: .

[0046] After observing the data and , the posterior probability distribution of the parameters of the extended hierarchical model is proportional to the product of the likelihood function and the prior distribution. In specific solving, numerical calculation methods such as Markov Chain Monte Carlo (MCMC) are used to sample the posterior distribution, and through a large number of iterative calculations, the true form of the posterior distribution is gradually approached, and the posterior distribution is finally solved.

[0047] After the above process, the parameters of the extended hierarchical model obtained can be used to construct the extended hierarchical model , which can be represented as , where is the optimal parameter set of the posterior distribution; is a loop variable, which takes integer values from 1 to Mt in turn, traversing all levels of the formation division, representing all level parameters of the extended hierarchical model covering from the 1st to the formation; is the total number of levels of the formation division, which is the number of levels obtained after dividing the entire formation system, used to define the level range. The extended hierarchical model includes the correlation between the acoustic features and the seismic features in each formation, and reflects the hierarchical differences of the formation in the vertical direction based on the statistical characteristics of the seismic data, providing accurate and reliable basic model support for seabed bottom classification mapping.

[0048] Step S130: Based on the extended hierarchical model, a preliminary seabed topography model map is constructed using a spatial interpolation algorithm. A generative adversarial network is used to repair the missing areas in the preliminary seabed topography model map to generate a complete seabed topography model map. Multidimensional acoustic-topographic-seismic features are embedded into the complete seabed topography model map to predict sediment migration trends.

[0049] Specifically, spatial interpolation algorithms can integrate the advantages of heterogeneous data to construct high-precision, high-reliability seabed topography models, providing a solid foundation for sediment transport simulation and pipeline safety assessment. Heterogeneous data includes surface, shallow, and deep data. Surface data can be obtained by interpolating multibeam topography data using side-scan imagery (e.g., correcting topographic details through texture features). Shallow (0m-300m) data is primarily based on multibeam topography data, with borehole data constraining sediment type distribution. Deep (>300m) data is interpolated using seismic data, with borehole data verifying the accuracy of layering.

[0050] For example, in step S130, based on the extended layered model, a preliminary seabed topographic model map is constructed using a spatial interpolation algorithm, including: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] The optimal set of parameters for the middle posterior distribution Spatial coupling and integration with discrete seabed topography sampling point data is performed; features intrinsically related to topography values ​​are extracted from sensor data as correlation features, and the spatial correlation of data is quantified using a variogram function to explore the potential relationship between topography values ​​and correlation features; the correlation features are used as auxiliary variables in the co-kriging interpolation algorithm. Based on the topography values ​​of known sampling points and combined with the potential relationship between topography values ​​and correlation features, the co-kriging interpolation algorithm is used to predict the interpolation points in the seabed area and generate a preliminary seabed topography model map. .

[0051] Specifically, in obtaining the extended hierarchical model Subsequently, a preliminary seabed topographic model can be constructed using spatial interpolation algorithms, which will then be used to expand the layered model. The optimal set of parameters for the middle posterior distribution This model integrates with existing discrete seabed topographic sampling point data, such as multibeam bathymetry data points. It acquires topographic values ​​from known sampling points, whose coordinates represent their specific locations in seabed space, while the topographic values ​​record the corresponding topographic height information. Simultaneously, it introduces correlation features, such as multibeam slope, which often have an intrinsic relationship with topographic values. A co-kriging algorithm is used to uncover this intrinsic relationship to assist in the interpolation of topographic values. Specifically, the co-kriging algorithm quantifies the spatial correlation between data points using a variogram function. Compared to traditional kriging algorithms, the co-kriging algorithm, by introducing correlation features, fully utilizes the potential connections between topography and other relevant factors, improving the accuracy of the preliminary topographic model and resulting in a preliminary seabed topographic model. It can be represented as: .in, Two-dimensional plane coordinates The predicted terrain value at that location. Thus, in two-dimensional plane coordinates... At any location, the preliminary seabed topography model can provide corresponding topographic predictions. This laid a solid foundation for subsequent seabed sediment classification and mapping work.

[0052] Based on the selected interpolation algorithm, namely the co-kriging interpolation algorithm, the feature association information provided by the extended hierarchical model is used to calculate all the points to be interpolated in the seabed area. The two-dimensional coordinates of all the calculated points are imported into the geographic information system (GIS) software or professional modeling software. Through operations such as gridding and rendering, a preliminary seabed topographic model map is generated, which can intuitively display the undulation shape and stratigraphic structure characteristics of the seabed topography.

[0053] In the process of seabed sediment classification and mapping, the accuracy of spatial interpolation algorithms directly affects the accuracy of the seabed topography model. To improve model quality, this implementation method uses an improved co-kriging interpolation algorithm to construct a preliminary seabed topography model. The core advantage of the co-kriging algorithm lies in its ability to comprehensively utilize the correlations between various data sources, thereby making more accurate predictions for unknown areas.

[0054] Multi-source data integration: extending the hierarchical model The optimal set of parameters for the middle posterior distribution (It covers acoustic-seismic characteristic correlation parameters, reflecting the differences in vertical stratigraphy) and spatially couples and integrates discrete seafloor topographic sampling point data (such as multibeam bathymetry data, which contains topographic height information of known locations).

[0055] Correlation feature introduction: features with inherent correlation with terrain values are extracted from sensor data as correlation features, such as multi-beam terrain slope, side-scan image texture roughness, etc., and the correlation features are used as auxiliary variables in the collaborative kriging interpolation algorithm. The correlation between terrain values and correlation features is quantified by using the variogram function to mine the potential relationship between terrain values and correlation features.

[0056] Collaborative interpolation calculation: the collaborative kriging interpolation algorithm is used to predict the terrain values of the interpolation points in the seabed area based on the known terrain values of the sampling points, and the potential relationship between the terrain values and the correlation features is used to generate a preliminary seabed terrain model. .

[0057] For example, in step S130, the missing areas in the preliminary seabed terrain model are repaired using a generative adversarial network to generate a complete seabed terrain model, including: training a generative adversarial network (GAN) model through a network structure generator to make the GAN model learn a complete extended hierarchical model and distinguish between real terrain and repaired areas; inputting the preliminary seabed terrain model and the multi-dimensional acoustic-terrain-seismic features into the trained GAN model to repair the missing areas, obtaining the repaired continuous terrain model output by the trained GAN model; verifying and optimizing the repaired continuous terrain model to generate a complete seabed terrain model.

[0058] Specifically, the process of repairing the missing areas in the preliminary seabed terrain model is as follows: training a GAN model through a network structure generator to make the GAN model learn a complete extended hierarchical model and distinguish between real terrain and repaired areas; inputting the preliminary seabed terrain model and the multi-dimensional acoustic-terrain-seismic features into the trained GAN model to repair the missing areas, improve the quality and practicality, and reconstruct high-precision geospatial data, obtaining the repaired continuous terrain model output by the trained GAN model, and verifying and optimizing the repaired continuous terrain model to generate a complete seabed terrain model.

[0059] The network structure generator is used to identify the missing areas in the preliminary seabed terrain model and determine the areas where the terrain data in the preliminary seabed terrain model is incomplete or missing. The generative adversarial network attempts to generate realistic terrain patches based on the input incomplete terrain data to fill in the missing parts, and evaluates whether the patches generated by the generator are real, i.e., distinguishes between the differences between the generated terrain patches and the actual existing terrain data, where the terrain data refers to the multi-dimensional acoustic-terrain-seismic features.

[0060] In training the GAN model, first, a training data set is prepared, and a region containing high-quality complete terrain data is selected as a training sample; ideally, it has similar geological features to the target repair area; and a loss function is set, defining an appropriate loss function to guide the generator to produce results closer to the actual situation, and optimize the discrimination ability of the discriminator; at the same time, in the iterative training process, the parameters of the generator and the discriminator are repeatedly adjusted until the generator can produce terrain patches that the discriminator cannot distinguish between true and false.

[0061] In applying the GAN model to fill the missing area, the preliminary seabed terrain model containing the missing part is input into the trained generator by inputting the missing terrain data, and the terrain patch is generated to fill the missing area to obtain the filled terrain map; the generated terrain patch is smoothed as necessary to ensure a natural transition with the surrounding existing data, while maintaining the consistency and reasonableness of the terrain, thereby obtaining the repaired continuous terrain model.

[0062] In verifying the repaired continuous terrain model, the terrain map before and after filling is compared, and the repaired area is compared with the actual measured data of the adjacent area to test the repair effect. If the repair effect of some areas is not ideal, more relevant data can be collected, the GAN model parameters can be adjusted, or other supplementary techniques can be used to further optimize the repair results.

[0063] After completing the repair of the missing area, the repaired continuous terrain model is integrated into the overall seabed terrain model to finally generate a complete seabed terrain model map.

[0064] By using the collaborative kriging interpolation algorithm to integrate the extended layered model parameters and the sampling data, and introducing the generative adversarial network to repair the missing terrain area, the resolution and completeness of the seabed terrain model can be significantly improved, providing high-precision basic data for seabed sediment classification.

[0065] For example, in step S130, multi-dimensional acoustic-terrain-seismic features are embedded in the complete seabed terrain model map to predict sediment transport trends, including: obtaining acoustic features, terrain features, and seismic features through multi-dimensional acoustic-terrain-seismic features, and correlating the acoustic features, terrain features, and seismic features in spatial dimensions to construct a mapping relationship of "feature-sedimentary environment-transport potential"; based on the complete seabed terrain model map, embedding acoustic features, terrain features, and seismic features as constraint conditions to build a numerical simulation framework of sediment transport as a transport numerical model, and using the transport numerical model to simulate the transport process of sediments to predict the transport trend of sediments; using a machine learning model trained based on historical sediment transport data to optimize the sediment transport prediction results of the transport numerical model.

[0066] Specifically, the acoustic features can reflect the different reflection / scattering characteristics of different sediments such as sand, mud, gravel, etc. due to differences in particle size, density, porosity, etc. For example, the acoustic features of sandy sediments are characterized by strong acoustic reflection and clear echoes, while the acoustic features of muddy sediments are characterized by weak acoustic reflection and blurred echoes. By analyzing the spatial distribution of multi-dimensional acoustic features, the type and boundary of the sediment can be identified, and the basic environment of migration can be preliminarily framed.

[0067] The topographic features can reflect the slope, strike, and valley / ridge morphology of the seafloor topography, which directly affects the migration direction of sediments under the action of gravity flow and tidal current. For example, in the steep slope area, sediments are prone to sliding and transporting under the influence of gravity, while in the gentle slope area, the accumulation and diffusion of sediments are dominated by hydrodynamic forces such as tidal current.

[0068] The seismic features in the seismic data reflect the structure and sequence relationship of the underground strata. For example, the unconformity surface and faults in the strata can become barriers or channels for sediment transport. If there is a fault in the strata, it can block the lateral transport of sediments, causing them to accumulate on one side of the fault. If the strata are continuous and inclined in sequence, sediments can migrate along the sequence direction and down the slope.

[0069] The above-mentioned acoustic features, topographic features, and seismic features are correlated in the spatial dimension to construct a mapping relationship between "features-sedimentary environment-migration potential". For example, the above-mentioned mapping relationship can be expressed as: a specific acoustic reflection pattern + gentle slope topography + continuous seismic sequence, which corresponds to the long-distance and slow migration of sediments down the slope. Alternatively, the above-mentioned mapping relationship can also be expressed as: strong acoustic reflection difference + steep slope topography + fault development, which may correspond to the rapid sliding and short-distance accumulation of sediments.

[0070] In simulating the migration trend of sediments, a model-driven trend simulation is used. First, a migration numerical model is constructed based on a complete seafloor topographic model, and acoustic, topographic, and seismic features are embedded as constraint conditions to build a numerical simulation framework for sediment transport as the migration numerical model. For example, the migration numerical model can be a sediment transport model based on fluid mechanics. Multi-dimensional features provide initial boundary conditions and parameters for the migration numerical model. For example, acoustic features can be used to invert the initial distribution thickness of sediments, topographic features can be used to define the flow path and velocity field of hydrodynamic forces such as tidal current, and seismic features can be used to constrain the permeability and transportability of the strata.

[0071] For example, the migration numerical model includes: .

[0072] wherein, represents a sediment transport index, represents the weight corresponding to the th multi-beam topographic data feature, Indicates the first The sediment deposition thickness corresponding to each matching point Indicates the weight of ocean current movement. Indicates the first The weights corresponding to the acoustic features of each side scan image Indicates the structural stability assessment value and , Indicates the thickness of the sedimentary layer. Indicates factors influencing seabed sediment. This indicates the average number of reflected signal points after the seismic source is excited. Indicates the acoustic echo intensity of the side scan image. Indicates ocean dynamic factors and , Indicates the sand wave mobility coefficient. Indicates pressure, Represents the velocity vector of the water flow. Indicates the bottom shear stress. Indicates the external force term.

[0073] Numerical models of sediment transport can simulate the erosion, transport, and deposition of sediment particles under the influence of hydrodynamic forces (tidal currents, ocean currents, etc.), gravity, and tectonic stress—that is, the sediment transport process. Combined with characteristic data, the flux of sediments in different regions can be calculated, i.e., the amount transported per unit time and per unit area, predicting their spatial transport direction, such as transport from high-energy topographic areas to low-energy areas, and deposition locations, such as topographic depressions and tectonically stable areas. This allows for the prediction of sediment transport trends using numerical models of sediment transport.

[0074] To improve the accuracy of predicted sediment transport trends, machine learning models can be used to assist in refining the predictions of numerical transport models. Specifically, historical sediment transport data, including changes in seafloor topography and sediment thickness over different periods, can be used as samples to train machine learning models, such as neural networks. This allows the machine learning model to learn the correspondence between feature combinations and sediment transport trends, and then use the trained model to correct the simulation results of the numerical transport model. For example, when the numerical transport model predicts a significant deviation from historical data regarding sediment transport rates in a certain area, the trained machine learning model can be used to adjust parameters such as water flow resistance coefficients and sediment erosion thresholds based on similar cases of feature matching, thereby optimizing the prediction results of the numerical transport model.

[0075] By constructing a mapping relationship between "features-sedimentary environment-migration potential", embedding multidimensional features into the migration numerical model, and combining machine learning to correct the simulation results, accurate prediction of sediment migration trends was achieved.

[0076] Step S140, based on the complete seabed topography model map, generate seabed sediment classification plan and three-dimensional geological profile map, according to the predicted sediment transport trend and seabed sediment classification plan and three-dimensional geological profile map, seabed sediment classification mapping.

[0077] For example, in step S140, based on the complete seabed topography model map, generate seabed sediment classification plan and three-dimensional geological profile map, including: taking the complete seabed topography model map as a spatial carrier, extracting the information related to seabed sediment in the multi-dimensional acoustic-topography-seismic feature, combining the extracted information with the topography, and constructing the seabed sediment classification rule; using spatial analysis algorithm, assigning value to the grid cell of the complete seabed topography model map according to the seabed sediment classification rule, and generating the seabed sediment classification plan; along the preset profile line, extracting the three-dimensional data of the complete seabed topography model map, combining the seabed sediment classification result and the seismic feature, reflecting the vertical structure of the stratum, and constructing the three-dimensional geological profile map.

[0078] Specifically, when constructing the seabed sediment classification rule, first taking the complete seabed topography model map as a spatial carrier, extracting the information related to seabed sediment in the multi-dimensional acoustic-topography-seismic feature, and then combining the extracted information with the topography, thereby constructing the seabed sediment classification rule. For example, the acoustic reflection characteristics of different seabed sediments (such as sand, mud, gravel, etc.) are obviously different, combined with the topographic features, such as more gravel in steep slope area, more mud in gentle area, seismic data can reflect the seabed sediment deposition age, sequence, etc., and the corresponding seabed sediment classification rule can be constructed.

[0079] When generating the seabed sediment classification plan, spatial analysis algorithms such as feature threshold-based classification and clustering algorithm can be used to assign values to the grid cells of the complete seabed topography model map according to the constructed seabed sediment classification rule, and generate the seabed sediment classification plan. Each grid cell in the seabed sediment classification plan can be labeled with the corresponding seabed sediment type, such as sandy, muddy and mixed, showing the distribution pattern of seabed sediment on the plane, such as sandy in nearshore area and muddy in deep-sea plain.

[0080] When constructing the three-dimensional geological profile map, along the preset specific profile line, such as vertical to the coastline and along the structure trend, extract the three-dimensional data of the complete seabed topography model map, combine the seabed sediment classification result and the seismic feature, reflect the vertical structure of the stratum, and construct the three-dimensional geological profile map. The three-dimensional geological profile map can show the distribution of seabed sediment at different vertical depths, including the thickness of the sediment layer, the contact relationship with the underlying stratum such as bedrock and old sediment layer, and present the "plane-vertical" three-dimensional sediment distribution pattern.

[0081] For example, in step S140, according to the predicted sediment transport trend and the seabed sediment classification plan and three-dimensional geological profile, the seabed sediment classification mapping is carried out, including: based on the seabed sediment classification plan and three-dimensional geological profile, superimposing the "source-sink" relationship of sediment transport, marking the sediment erosion, transport starting place as the source area on the seabed sediment classification plan, and marking the sediment accumulation destination as the sink area, predicting the seabed sediment classification change, correcting the seabed sediment classification boundary, and reflecting the sediment transport on the vertical sediment layer through the three-dimensional geological profile, and finally generating a dynamic classification map integrating the sediment transport trend prediction.

[0082] Specifically, the sediment transport trend predicted in step S130, such as the transport direction and the accumulation area, is combined with the seabed sediment classification, so that the predicted sediment transport trend constrains the seabed sediment classification. For example, if the measured sediment transport trend indicates that a certain area is currently dominated by sandy sediment, but in the future a large amount of muddy sediment will be transported and accumulated, the seabed sediment classification change trend needs to be predicted. Or, according to the sediment transport path indicated by the predicted sediment transport trend, the seabed sediment classification boundary is corrected, such as the transport channel area being classified as "unstable sediment area" due to the continuous transport of seabed sediment.

[0083] In the seabed sediment classification mapping, based on the seabed sediment classification plan and three-dimensional geological profile, the "source-sink" relationship of sediment transport is superimposed, wherein the source area refers to the sediment erosion and transport starting place, and the sink area refers to the sediment accumulation destination. The source area and the sink area are marked on the seabed sediment classification plan, and the seabed sediment classification change is predicted and the seabed sediment classification boundary is corrected. Among them, the seabed sediment corresponding to the source area is easy to be transformed and classified as "erosion sediment". The seabed sediment corresponding to the sink area is rapidly accumulated and classified as "accumulation sediment". In the three-dimensional geological profile, the transformation of the vertical sediment layer by sediment transport is embodied, such as the vertical sediment layer of the source area being thinned and the vertical sediment layer of the sink area being thickened, and finally a dynamic classification map integrating the sediment transport trend prediction is generated. The dynamic classification map can reflect the current and future classification pattern of the seabed sediment based on the sediment transport trend prediction.

[0084] By integrating the transport trend and the seabed sediment classification rule, superimposing the "source-sink" relationship to mark the corresponding source area / sink area, a dynamic classification map reflecting the current and future pattern is generated, which can provide a scientific basis for seabed sediment classification mapping.

[0085] The seabed bottom classification mapping method based on shallow stratigraphic profile data provided by the embodiments of the present disclosure, relative to the prior art, through the collaborative analysis of multi-source sensor data (multi-beam topography, side scan image, seismic and shallow stratigraphic profile data), a multi-dimensional feature system is constructed; the spatial interpolation algorithm is used to integrate the model parameters and the sampling data, the generative adversarial network is introduced to repair the missing area of the terrain, and the resolution and integrity of the seabed terrain model are significantly improved, thereby providing high-precision basic data for the bottom classification; by constructing the mapping relationship of "feature-sedimentary environment-migration potential", the multi-dimensional features are used to accurately predict the sediment migration trend; the migration trend and the bottom classification are fused to generate a dynamic classification mapping reflecting the current and future pattern, thereby providing a scientific basis for the seabed bottom classification mapping, thereby solving the problem of insufficient precision in the traditional method in the seabed terrain and bottom classification, especially under the condition of complex terrain and variable bottom, it is difficult to obtain high-resolution and accurate data, and different types of sensor data may not be consistent, and the collection process is easily disturbed by environmental noise, resulting in a decline in data quality, and it is difficult to fully consider the complex physical process involved in the sediment migration trend.

[0086] It should be noted that each formula involved in the above embodiments is calculated by removing the dimension of each parameter and taking its numerical value, and each formula is the formula closest to the actual situation obtained by simulating a large amount of collected data using software, and the specific values of the parameters in the formula can be determined according to the actual situation.

[0087] Those skilled in the art can understand that the above embodiments are specific embodiments for implementing the present disclosure, and in actual application, various changes can be made in form and details without departing from the spirit and scope of the present disclosure.

Claims

1. A seabed bottom classification mapping method based on shallow sub-bottom profile data, characterized by, The seabed bottom classification mapping method based on shallow stratigraphic profile data comprises: Collecting sensor data of a target seabed, wherein the sensor data comprises multi-beam topographic data, side scan image data, seismic data and shallow stratigraphic profile data; Using the sensor data, constructing a multi-dimensional acoustic-terrain-seismic feature, performing acoustic feature matching on the multi-beam topographic data and the side scan image data, introducing the seismic data into the matched acoustic feature, and constructing an extended layered model; Based on the extended layered model, using a spatial interpolation algorithm to construct a preliminary seabed topographic model map, using a generative adversarial network to repair missing areas in the preliminary seabed topographic model map, generating a complete seabed topographic model map, embedding the multi-dimensional acoustic-terrain-seismic feature into the complete seabed topographic model map, and predicting sediment transport trends; Based on the complete seabed topographic model map, generating a seabed bottom classification plan and a three-dimensional geological profile map, and performing seabed bottom classification mapping according to the predicted sediment transport trends and the seabed bottom classification plan and the three-dimensional geological profile map.

2. The method of mapping seabed sediment based on shallow sub-bottom profile data according to claim 1, wherein, The use of the sensor data to construct a multi-dimensional acoustic-terrain-seismic feature comprises: According to the multi-beam topographic data, the side scan image data and the seismic data, respectively determining multi-beam topographic data features, side scan image acoustic features and seismic data features; Based on the contribution of the multi-beam topographic data features, the side scan image acoustic features and the seismic data features to seabed bottom classification, using information gain ratio to calculate the weights corresponding to the multi-beam topographic data features, the side scan image acoustic features and the seismic data features respectively; Based on the multi-beam topographic data features, the side scan image acoustic features, the seismic data features and the weights corresponding thereto respectively, constructing a multi-dimensional acoustic-terrain-seismic feature through element-level weighted fusion.

3. The method of mapping seabed sediment based on shallow sub-bottom profile data according to claim 2, wherein, The acoustic feature matching on the multi-beam topographic data and the side scan image data comprises: Projecting the multi-beam topographic data and the side scan image data into the same coordinate system, and performing acoustic feature matching on the multi-beam topographic data features and the side scan image acoustic features according to the following formula: ； wherein, represents an acoustic feature matching coefficient, represents a feature value of the th multi-beam terrain data, represents a feature value of the th side-scan image data, represents a current acoustic feature matching point and , represents a total number of acoustic feature matching points, represents a mean value of feature values of all multi-beam terrain data, represents a mean value of feature values of all side-scan image data.

4. The method of mapping seabed sediment based on shallow sub-bottom profile data according to claim 3, wherein, The introduction of the seismic data into the matched acoustic feature to construct an extended layered model comprises: The stratum is divided into Mt layers according to the hierarchical structure, and the acoustic feature vector and the seismic feature vector of the mt-th stratum are respectively denoted as , ; The prior distribution is given to the parameters of the extended layered model, and the model parameter vector corresponding to the mtth stratum is Subject to normal distribution; The likelihood function defines the explanatory power of the extended hierarchical model, and the posterior distribution is sampled based on the Bayes theorem and numerical calculation methods such as Markov chain Monte Carlo, so that the extended hierarchical model is solved is represented as is represented as is the optimal parameter set of the posterior distribution.

5. The method of mapping seabed sediment based on shallow stratigraphic profile data according to claim 4, wherein, Based on the extended layered model, using a spatial interpolation algorithm to construct a preliminary seabed topographic model map comprises: Extending hierarchical models Optimal parameter set of the posterior distribution spatially dimensionally coupled integration with discrete seafloor topography sample point data; Extracting features that have an inherent correlation with terrain values from the sensor data as correlation features, quantifying data spatial correlation using a variogram, and mining potential relationships between terrain values and correlation features; The correlation feature is taken as an auxiliary variable of the co-Kriging interpolation algorithm, and the co-Kriging interpolation algorithm is used to predict the to-be-interpolated points of the seabed area based on the terrain values of the known sampling points, and a preliminary seabed terrain model map is generated by combining the potential correlation between the terrain values and the correlation features .

6. The method of mapping seabed sediment based on shallow stratigraphic profile data according to claim 5, wherein, The use of a generative adversarial network to repair missing areas in the preliminary seabed topographic model map to generate a complete seabed topographic model map comprises: Training a generative adversarial network model through a network structure generator, so that the generative adversarial network model learns a complete extended layered model and distinguishes between real terrain and repaired areas; a preliminary seabed topography model is obtained and the multi-dimensional acoustic-topography-seismic feature is input into the trained generative adversarial network model to repair the missing area, to obtain a repaired continuous topography model output by the trained generative adversarial network model; Verifying and optimizing the repaired continuous terrain model to generate the complete seabed topographic model map.

7. The method of mapping seabed sediment based on shallow stratigraphic profile data according to claim 6, wherein, The embedding the multi-dimensional acoustic-terrain-seismic features into the complete seabed terrain model map, predicting sediment transport trend, comprises: Through the multi-dimensional acoustic-terrain-seismic features, respectively acquiring acoustic features, terrain features and seismic features, and correlating the acoustic features, the terrain features and the seismic features in spatial dimensions, constructing a mapping relationship of "feature-sedimentary environment-transport potential"; Based on the complete seabed terrain model map, embedding the acoustic features, the terrain features and the seismic features as constraint conditions, building a numerical simulation framework of sediment transport as a transport numerical model, simulating the sediment transport process by using the transport numerical model, and predicting the sediment transport trend; Using a machine learning model trained based on sediment transport historical data, optimizing the sediment transport prediction result of the transport numerical model.

8. The method of mapping seabed sediment based on shallow sub-bottom profile data according to claim 7, wherein, The transport numerical model comprises: ； wherein, represents a sediment transport index, represents a weight corresponding to the th multi-beam terrain data feature, represents a sedimentation thickness corresponding to the th matching point, represents a weight of ocean current movement, represents a weight corresponding to the th side scan sonar acoustic feature, represents a construction structure stability evaluation value and , represents a sediment layer thickness, represents a seabed bottom material impact factor, represents an average reflection signal point number after excitation of a seismic source, represents a side scan sonar acoustic feature echo intensity, represents a marine dynamic factor and , represents a sand wave transport coefficient, represents pressure, represents a water flow velocity vector, represents a bottom shear stress, represents an external force term.

9. The method of claim 7, wherein the shallow stratigraphic profile data is obtained from a shallow seismic survey. Based on the complete seabed terrain model map, generating a seabed bottom classification plan and a three-dimensional geological profile map, comprises: Taking the complete seabed terrain model map as a spatial carrier, extracting information related to seabed bottom in the multi-dimensional acoustic-terrain-seismic features, combining the extracted information with the terrain, and constructing seabed bottom classification rules; Using a spatial analysis algorithm, assigning values to the grid cells of the complete seabed terrain model map according to the seabed bottom classification rules, and generating the seabed bottom classification plan; Along a preset profile line, extracting three-dimensional data of the complete seabed terrain model map, combining seabed bottom classification results and the seismic features, reflecting the vertical structure of the stratum, and constructing the three-dimensional geological profile map.

10. The method of mapping seabed sediment classes based on shallow stratigraphic profile data according to claim 9, wherein, According to the predicted sediment transport trend and the seabed bottom classification plan and the three-dimensional geological profile map, seabed bottom classification mapping, comprises: Based on the seabed bottom classification plan and the three-dimensional geological profile map, superimposing the "source-sink" relationship of sediment transport, marking the sediment erosion, transport departure place as the source area and the sediment accumulation destination as the sink area on the seabed bottom classification plan, predicting the seabed bottom classification change, correcting the seabed bottom classification boundary, and through the three-dimensional geological profile map, reflecting the reconstruction of the vertical bottom layer by sediment transport, finally generating a dynamic classification map integrating the sediment transport trend prediction.

Images