Seabed sediment classification method and product based on acoustic characteristics of seabed surface sediments

Abstract

The present invention discloses a method and product for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments, and relates to the field of seabed acoustic detection. The present invention first obtains the measured data of seabed acoustic reflection in the measurement area in the marine environment, and inverts to obtain the characteristic parameters of seabed surface sediments; the acoustic characteristic parameters of seabed surface sediments are screened out by analyzing the uncertainty and correlation of the characteristic parameters of seabed surface sediments; a deep neural network model is constructed based on the acoustic characteristic parameters of seabed surface sediments, including a seabed acoustic-physical information neural network model and a sediment attribute prediction neural network model; the training sample set is expanded according to the measured data of seabed acoustic reflection and the acoustic characteristic parameters of seabed surface sediments, and the deep neural network model is trained and tested; the seabed sediment classification model is used to classify the seabed sediments in the measurement area, so as to realize the refined classification of the seabed sediments and improve the continuity and classification accuracy of the classification.

Description

Technical Field

[0001] The present invention relates to the technical field of seabed acoustic detection, and in particular to a seabed sediment classification method and product based on the acoustic characteristics of seabed surface sediments. Background Art

[0002] The research on constructing the method of characterizing the acoustic characteristics of the seafloor based on acoustic models and seafloor data has been the research frontier and hotspot in the field of seafloor acoustics in recent years, and has important application value in acoustic modeling of shallow sea environments and remote sensing of bottom sediments. Seafloor surface sediments (SSS) have a multi-phase, multi-particle, and multi-morphological composition structure, which leads to complex and diverse acoustic characteristics. Seafloor environmental parameters such as the composition and structure of seafloor surface sediments have an important influence on the physical and acoustic properties of sediments. However, the current acoustic model based on traditional theories is difficult to realize the acoustic characteristics analysis of seafloor surface sediments in complex environments such as different sedimentary environments and hydrodynamics, and thus cannot perform fine classification of seafloor sediments. Summary of the invention

[0003] The purpose of the present invention is to provide a method and product for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments, which can achieve refined classification of seabed sediments and improve the continuity and classification accuracy of the classification.

[0004] To achieve the above object, the present invention provides the following solution.

[0005] In one aspect, the present invention provides a method for classifying seafloor sediments based on acoustic characteristics of seafloor surface sediments, comprising:

[0006] Acquiring the measured data of seabed acoustic reflection in the measurement area in the marine environment by using an acoustic in-situ measurement technology based on the acoustic reflection method; the measured data of seabed acoustic reflection includes the reflected signal waveform, propagation time and reflected power at different grazing angles;

[0007] Based on the acquired seabed acoustic reflection measured data, the seabed surface sediment characteristic parameters are obtained by using the seabed surface sediment characteristic parameter inversion method; the seabed surface sediment characteristic parameters include: sediment type, particle size, density, porosity, impedance, absorption coefficient, sound velocity, sound attenuation coefficient and sound reflection coefficient, as well as the number of sediment layers, layer thickness, sound velocity distribution, sound velocity depth profile, sound attenuation coefficient depth profile, average particle size, average density and average porosity;

[0008] By analyzing the uncertainty and correlation of the characteristic parameters of the seafloor surface sediments, the acoustic characteristic parameters of the seafloor surface sediments for building the deep neural network model are screened out;

[0009] Construct a deep neural network model based on the acoustic characteristic parameters of seafloor surface sediments, including a seafloor acoustic-physical information neural network model and a sediment property prediction neural network model;

[0010] Expand the training sample set based on the measured data of seabed acoustic reflection and the acoustic characteristic parameters of seabed surface sediments;

[0011] The deep neural network model is trained and tested using the training sample set, and a seabed sediment classification model is obtained after the training is completed;

[0012] The seabed sediment classification model is used to classify the seabed sediments in the survey area.

[0013] Optionally, the method of obtaining the measured data of seabed acoustic reflection in the measurement area in the marine environment by using an acoustic in-situ measurement technology based on an acoustic reflection method specifically includes:

[0014] An acoustic in-situ measurement device for seafloor surface sediments is constructed by using an acoustic in-situ measurement technology based on an acoustic reflection method; the acoustic in-situ measurement device for seafloor surface sediments comprises a sound source system, a receiving system, a data acquisition and processing system, and a synchronous control system; the sound source system comprises a signal generator, a power amplifier, and a transducer array, which are used to generate an acoustic beam with a preset frequency, waveform, and grazing angle; the receiving system comprises a hydrophone array and a preamplifier, which are used to receive and amplify the seafloor acoustic reflection echo signal; the data acquisition and processing system is used to perform digital conversion, filtering, gain compensation, beamforming, and data analysis and processing on the amplified echo signal, and extract the measured data of the seafloor acoustic reflection; the synchronous control system is used to provide a unified clock and trigger signal for the sound source system and the receiving system, so as to ensure the synchronization of transmission and reception;

[0015] Using the acoustic in-situ measurement device for seabed surface sediments, the measurement tasks of sound wave emission, reception, collection, storage and processing are automatically completed according to the preset emission waveform, emission voltage, emission width and measurement position parameters, and the reflected signal waveform, propagation time and reflection power at different grazing angles in the measurement area are obtained as the measured data of seabed acoustic reflection; during the measurement process, the mechanical or electronic scanning of the transducer array is controlled to achieve sound wave incidence at different grazing angles.

[0016] Optionally, the method of obtaining the characteristic parameters of the seafloor surface sediments based on the acquired seafloor acoustic reflection measured data by using a seafloor surface sediment characteristic parameter inversion method specifically includes:

[0017] By using the inversion method of the characteristic parameters of the seabed surface sediments, the type, particle size, density and porosity of the sediments are inferred by analyzing the change law of the reflected signal waveform; the inversion method of the characteristic parameters of the seabed surface sediments includes the least squares method, particle swarm optimization algorithm, genetic algorithm, Bayesian inversion method, back propagation neural network inversion method and neural network inversion method;

[0018] By measuring the propagation time difference at different grazing angles, the number of sediment layers, layer thickness and sound velocity distribution are calculated using the inversion method of seafloor surface sediment characteristic parameters.

[0019] By measuring the reflected power spectrum at different grazing angles, the seabed acoustic reflection coefficient is calculated using the seabed surface sediment characteristic parameter inversion method, and then the impedance, absorption coefficient and acoustic attenuation coefficient of the sediment are inverted;

[0020] The average grain size, average density and average porosity of the sediment layer are further calculated based on the grain size, density and porosity of the sediment;

[0021] The sound velocity depth profile and the sound attenuation coefficient depth profile of the sediment layer are further calculated based on the sound velocity distribution and the sound attenuation coefficient of the sediment layer.

[0022] Optionally, the acoustic characteristic parameters of the seafloor surface sediments are screened out for constructing a deep neural network model by analyzing the uncertainty and correlation of the characteristic parameters of the seafloor surface sediments, specifically including:

[0023] The variance method was used to analyze the uncertainty of the characteristic parameters of the seafloor surface sediments, and the number of sediment layers, layer thickness, average grain size, average density, average porosity, sound velocity depth profile, and sound attenuation coefficient depth profile were selected as the acoustic characteristic parameters of the seafloor surface sediments for constructing the deep neural network model.

[0024] The normalized covariance matrix method is used to analyze the correlation between the acoustic characteristic parameters of the seafloor surface sediments and the measured data of seafloor acoustic reflection, and the number of sediment layers, layer thickness, average density, sound velocity depth profile, and sound attenuation coefficient depth profile are selected as the output parameters of the first sub-model;

[0025] The normalized covariance matrix method was used to analyze the correlation between the acoustic characteristic parameters of the seafloor surface sediments and the output parameters of the first sub-model, and the average grain size, average density and average porosity of the entire measurement area and each sediment layer were selected as the output parameters of the second sub-model.

[0026] Optionally, the deep neural network model constructed based on the acoustic characteristic parameters of the seafloor surface sediments includes a seafloor acoustic-physical information neural network model and a sediment property prediction neural network model, specifically including:

[0027] A seabed acoustic-physical information neural network model is constructed by integrating a classic seabed acoustic model and a physical information neural network, wherein the input of the seabed acoustic-physical information neural network model is the measured data of seabed acoustic reflection in the measurement area, and the output is the output parameter of the first sub-model;

[0028] The first sub-model output parameter is used as the input of the sediment property prediction neural network model, and the output of the sediment property prediction neural network model is the second sub-model output parameter.

[0029] Optionally, the expanding the training sample set according to the measured seabed acoustic reflection data and the acoustic characteristic parameters of the seabed surface sediments specifically includes:

[0030] A large amount of measured seabed acoustic reflection data is generated as samples using sound field simulation software, and the samples are paired with labels consisting of corresponding acoustic characteristic parameters of seabed surface sediments to form a training sample set; the training sample set covers different seabed types, terrain features and frequency ranges.

[0031] Optionally, during the training of the deep neural network model, the loss function used is ;in is the value of the weighted mean square error loss function of the seabed acoustic characteristic parameters; is the sample size; is the number of types of acoustic characteristic parameters of seafloor surface sediments; For the In the sample The true values ​​of acoustic characteristic parameters of the seafloor surface sediments; The deep neural network model is In the sample Prediction of acoustic characteristic parameters of seafloor surface sediments; For the The weight coefficients of acoustic characteristic parameters of the seafloor surface sediments.

[0032] On the other hand, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the seabed sediment classification method based on the acoustic characteristics of seabed surface sediments.

[0033] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments.

[0034] On the other hand, the present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for classifying seabed sediments based on acoustic characteristics of seabed surface sediments.

[0035] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0036] The present invention utilizes an acoustic in-situ measurement technology based on an acoustic reflection method to obtain measured data of seabed acoustic reflection in a marine environment, and obtains acoustic reflection characteristics of seabed surface sediments at different grazing angles by controlling the range of grazing angles of incident sound waves, thereby providing high-quality measured data support for the inversion and classification of characteristic parameters of seabed surface sediments; on this basis, the distribution characteristics of the characteristic parameters of seabed surface sediments can be inverted using an inversion method for characteristic parameters of seabed surface sediments; that is, the present invention combines the acoustic in-situ measurement technology with an inversion algorithm to realize rapid, real-time, in-situ, and precise observation of the characteristic parameters of seabed surface sediments; further, by analyzing the uncertainty and correlation of the characteristic parameters of seabed surface sediments, the optimal model input and output parameters are screened out, thereby improving the performance and efficiency of the deep neural network model. The deep neural network model proposed in the present invention includes a seabed acoustic-physical information neural network model and a sediment property prediction neural network model, which are used for the inversion of acoustic characteristic parameters and physical property prediction of seabed surface sediments. Through the collaborative work of the two sub-models, end-to-end mapping from seabed acoustic reflection measured data to acoustic characteristic parameters, and from acoustic characteristic parameters to sediment physical properties is realized; by training the deep neural network model with an extended training sample set, accurate estimation and prediction of seabed acoustic characteristic parameters and sediment physical properties can be achieved. By using the trained seabed sediment classification model for prediction output and obtaining the physical parameters of the sediment, high-precision and high-resolution sediment identification and characterization can be achieved, and the continuity and classification accuracy of the classification can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for classifying seafloor sediments based on acoustic characteristics of seafloor surface sediments provided by the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] The present invention aims to provide a method and product for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments, which can achieve refined classification of seabed sediments and improve the continuity and classification accuracy of the classification.

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] See also Figure 1 The present invention provides a method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments, comprising the following steps 1 to 7.

[0043] Step 1: Use the acoustic in-situ measurement technology based on the acoustic reflection method to obtain the measured data of seabed acoustic reflection in the measurement area in the marine environment.

[0044] As an emerging technology that can quickly obtain seabed information, the acoustic in-situ measurement technology based on the acoustic reflection method is very suitable for the measurement of characteristic parameters of seabed surface sediments. The present invention uses the acoustic in-situ measurement technology based on the acoustic reflection method to build an acoustic in-situ measurement device for seabed surface sediments (hereinafter referred to as sediments). The acoustic in-situ measurement device for seabed surface sediments mainly consists of a sound source system, a receiving system, a data acquisition and processing system, and a synchronous control system. The sound source system includes a signal generator, a power amplifier, and a transducer array, which are used to generate a sound beam with a preset frequency, waveform, and grazing angle. The receiving system includes a hydrophone array and a preamplifier, which are used to receive and amplify the seabed acoustic reflection echo signal. The data acquisition and processing system is used to perform digital conversion, filtering, gain compensation, beamforming, and data analysis and processing on the amplified echo signal to extract the actual measured data of seabed acoustic reflection. The synchronous control system is used to provide a unified clock and trigger signal for the sound source system and the receiving system to ensure the synchronization of transmission and reception.

[0045] When conducting acoustic in-situ measurements, the acoustic in-situ measurement device for seafloor surface sediments is used to automatically complete measurement tasks such as acoustic wave emission, reception, collection, storage and processing according to preset parameters such as emission waveform, emission voltage, emission width and measurement position, and obtain the reflected signal waveform, propagation time and reflection power at different grazing angles in the measurement area as the actual measured data of seafloor acoustic reflection. During the measurement process, the incidence of sound waves at different grazing angles can be achieved by controlling the mechanical or electronic scanning of the transducer array. The grazing angle range of the incident sound wave is related to the measurement purpose and the seafloor environmental conditions. Generally, the following typical angles are selected: 90° (vertical incidence), 90~75° (interval 2°), 60°, 50°, 30°, 20°, 15°, etc. The measurement data at different grazing angles can provide angle-dependent information on the acoustic characteristics of seafloor surface sediments, which is helpful for analyzing the anisotropy and roughness characteristics of sediments.

[0046] The measured data of seabed acoustic reflection obtained by the in-situ acoustic measurement device for seabed surface sediments include data such as the reflection signal waveform, propagation time and reflection power at different grazing angles. These data are closely related to the grazing angle range of the incident sound wave, as described below.

[0047] Reflection signal waveform: The reflection signal waveforms at different grazing angles are significantly different, reflecting the acoustic response characteristics of the seafloor surface sediments. When the incidence is vertical (90°), the reflection signal mainly includes two parts: seafloor surface reflection and seafloor volume scattering; the smaller the grazing angle, the weaker the seafloor surface reflection and the stronger the volume scattering. By analyzing the changing pattern of the reflection signal waveform, the physical properties of the sediment, such as type, density, particle size, and porosity, can be inferred.

[0048] Propagation time: The propagation time of the reflected signal is related to the thickness of the sediment layer and the speed of sound. When the incidence is vertical, the propagation time is the shortest; the smaller the grazing angle, the longer the propagation time. By measuring the propagation time difference at different grazing angles, the number of sediment layers, layer thickness and sound speed distribution can be estimated.

[0049] Reflection power: The power of the reflected signal reflects the reflection loss characteristics of the seafloor surface sediments. When the incident light is vertical, the reflected power is the largest; the smaller the grazing angle, the smaller the reflected power. By measuring the reflected power spectrum at different grazing angles, the seafloor acoustic reflection coefficient can be calculated, and then the acoustic characteristic parameters such as the impedance, absorption coefficient and acoustic attenuation coefficient of the sediment can be inverted.

[0050] Seabed acoustic reflection measurement is an effective method for measuring sediment characteristic parameters. The acoustic reflection signal carries information about the structure and physical properties of the seabed surface sediments. Effective classification features can be extracted from the reflection signal. Combined with the three-dimensional acoustic reflection model, the distribution characteristics of layered sediment characteristics can be comprehensively analyzed. The three-dimensional acoustic reflection model describes the pulse energy, propagation loss and received echo energy of the seabed reflection model. The present invention extracts the reflection signal from the record of the acoustic in-situ measurement device, calculates the energy value received from the time window to obtain the acoustic reflection coefficient, and then compares it with the modeling value to invert the sediment characteristic parameters.

[0051] In the transmitting and receiving system in a given direction, the sound source transmits a certain pulse width of transmitting pulse energy and receives echo energy. The formula for the received signal power that is only applicable to the high frequency band is:

[0052] (1)

[0053] in, is the received signal power; is the transmitted signal power; is the distance from the sound source to the seabed interface, calculated from the sound propagation time and speed in seawater; is the acoustic reflection coefficient of the seabed interface; is the water absorption rate; is the spherical expansion factor. Considering that the layered sediment signal is composed of multiple reflections, it is necessary to further consider physical processes such as absorption, transmission, and reflection. Due to water absorption and spherical diffusion, the pulse energy of the sound wave will weaken during propagation in the water. After that, the pulse will be reflected for the first time at the seabed interface (the highest component). The remaining energy will penetrate the first layer of sediment and respond to the secondary spherical wave propagation loss of the layer thickness. Therefore, the received signal power is: (2)

[0054] in, is the transmission coefficient at the interface between water and sediment; is the acoustic reflection coefficient at the interface between water and sediment; is the sound propagation loss of the first layer of sediment; is the thickness of the first sediment layer (i.e. the thickness of the first sediment layer); is the acoustic reflection coefficient at the boundary between the first layer and the second layer; is the transmission coefficient at the sediment-water interface. The number of parameters required in equation (2) depends on the number of sediment layers. The energy distributed in each sediment layer during propagation Described by the following formula:

[0055] (3)

[0056] in, , , is the sound wave propagation time in the vertical direction, For the Sample windows, is the number of sample windows, Represents a discrete time point The signal amplitude. The size of the sample window is very important for the reasonable calculation of the local acoustic reflection coefficient. If the sample window is too short, the correct value of the desired local layer energy will basically not be captured. If the sample window is too large, it will overlap with the energy of the next layer and calculate the reflected energy of multiple layers, which will overestimate the acoustic reflection coefficient, resulting in inappropriate and unreasonable calculation results. It is necessary to select an appropriate sample window, and the size of the sample window can be selected by applying the reflection power spectrum analysis method. It is generally one or two times the length of the transmission pulse. The present invention selects the sample window size to be twice the pulse length.

[0057] Based on the acquired seabed acoustic reflection measured data, the sound velocity depth profile and acoustic attenuation coefficient depth profile of each sediment layer of the seabed surface sediment can be further analyzed. Based on the waveform, propagation time and power of the received signal, the three-dimensional acoustic reflection model and inversion algorithm can be used to invert the characteristic parameter distribution of the seabed surface sediments by minimizing the difference between the measured and modeled acoustic signals, and realize rapid, real-time, in-situ and precise observation of the seabed surface sediments.

[0058] The present invention adopts a seabed surface sediment acoustic in-situ measurement device to perform acoustic in-situ measurement to obtain the measured data of seabed acoustic reflection in the marine environment. By controlling the grazing angle range of the incident sound wave, the acoustic reflection characteristics of the seabed surface sediments at different angles are obtained, providing high-quality measured data support for the inversion and classification of seabed surface sediment parameters.

[0059] Step 2: Based on the acquired seabed acoustic reflection measured data, the seabed surface sediment characteristic parameters are obtained using the seabed surface sediment characteristic parameter inversion method.

[0060] The acoustic in-situ measurement device for seafloor surface sediments of the present invention can be used to obtain the measured data of seafloor acoustic reflection with frequency, grazing angle and position information (horizontal and vertical, or X-axis and Y-axis). On this basis, the three-dimensional acoustic reflection model and the inversion method of the characteristic parameters of the seafloor surface sediments, such as the least square method, particle swarm optimization algorithm, genetic algorithm, Bayesian inversion method, back propagation neural network inversion method and neural network inversion method, can be used to invert the distribution characteristics of the characteristic parameters of the seafloor surface sediments. The characteristic parameters of the seafloor surface sediments can include: the type, particle size, density, porosity, impedance, absorption coefficient, sound velocity, sound attenuation coefficient and sound reflection coefficient of the sediment, as well as the number of layers, layer thickness, sound velocity distribution, sound velocity depth profile, sound attenuation coefficient depth profile, average particle size, average density and average porosity of the sediment layer.

[0061] Acoustic in-situ measurement technology is used to obtain acoustic parameters of seafloor surface sediments, such as acoustic reflection coefficient, sound velocity, and acoustic attenuation coefficient, through acoustic reflection method, acoustic transmission method, etc. On this basis, acoustic theoretical models (such as Biot-Stoll model, Buckingham model, etc.) or empirical relationships (such as Hamilton empirical formula, Holmes empirical formula, etc.) are used, combined with optimization algorithms such as least squares method, gradient descent method, back propagation neural network, particle swarm optimization algorithm, genetic algorithm, Bayesian inversion, etc., to invert and estimate physical parameters such as particle size, density, and porosity of sediments. Further, using the empirical relationship between acoustic parameters and sediment types, a classification model between acoustic parameters and sediment categories is established (such as linear discriminant analysis, support vector machine, K-means clustering, support vector machine, deep learning, etc.), and the category of sediments is directly predicted according to their acoustic characteristics, so as to achieve fast and high-resolution refined acoustic classification of seafloor sediments, providing important technical support and scientific basis for marine geophysical exploration, seafloor engineering construction, and environmental monitoring.

[0062] Specifically, by using the inversion method of the characteristic parameters of the seabed surface sediments, the type, particle size, density and porosity of the sediments can be inferred by analyzing the changing pattern of the reflected signal waveform. By measuring the propagation time difference at different grazing angles, the number of layers, layer thickness and sound velocity distribution of the sediment layer can be estimated using the inversion method of the characteristic parameters of the seabed surface sediments. By measuring the reflection power spectrum at different grazing angles, the seabed sound reflection coefficient can be calculated using the inversion method of the characteristic parameters of the seabed surface sediments, and then the impedance, absorption coefficient and sound attenuation coefficient of the sediment can be inverted. Furthermore, based on the particle size, density and porosity of the sediment, the average particle size, average density and average porosity of the sediment layer can also be calculated. Based on the sound velocity distribution and sound attenuation coefficient of the sediment layer, the sound velocity depth profile and sound attenuation coefficient depth profile of the sediment layer can also be further calculated.

[0063] The present invention combines acoustic in-situ measurement technology and inversion algorithm to achieve rapid, real-time, in-situ and precise observation of characteristic parameters of seabed surface sediments, providing important technical support and data guarantee for marine geophysical exploration, seabed engineering construction, environmental monitoring, etc.

[0064] Step 3: By analyzing the uncertainty and correlation of the characteristic parameters of the seabed surface sediments, the acoustic characteristic parameters of the seabed surface sediments are screened out for building the deep neural network model.

[0065] The deep neural network model constructed by the present invention includes a seabed acoustic-physical information neural network model (also called the first sub-model) and a sediment property prediction neural network model (also called the second sub-model); by analyzing the uncertainty and correlation of characteristic parameters of seabed surface sediments (referred to as characteristic parameters), the input and output parameters of the first sub-model and the second sub-model are screened out to optimize the performance of the deep neural network model.

[0066] In order to analyze the uncertainty of the characteristic parameters of the seafloor surface sediments, the present invention adopts the variance method. The variance calculation formula is as follows:

[0067] (4)

[0068] in, represents the number of samples, Indicates Characteristic parameters in samples The value of Indicates characteristic parameters By calculating each characteristic parameter Variance , the uncertainty of the characteristic parameters of the seafloor surface sediments can be obtained, and then the characteristic parameters with smaller uncertainty can be selected as the acoustic characteristic parameters of the seafloor surface sediments (referred to as acoustic characteristic parameters) for constructing the deep neural network model. In some embodiments, the selected acoustic characteristic parameters are the number of layers, layer thickness, average grain size, average density, average porosity, sound velocity depth profile, and sound attenuation coefficient depth profile of the sediment layer.

[0069] Furthermore, it is necessary to analyze the correlation between the above acoustic characteristic parameters and the measured data of the seabed acoustic reflection, and select the acoustic characteristic parameters with high correlation with the measured data of the seabed acoustic reflection as the output parameters of the first sub-model. In order to analyze the correlation, the present invention adopts the method of normalized covariance matrix, and the calculation formula of the normalized covariance matrix is ​​as follows:

[0070] (5)

[0071] in, Representation parameters and The correlation coefficient between them, that is, the normalized covariance matrix; Selected from various acoustic characteristic parameters; Selected from measured data of seabed acoustic reflection, such as reflected signal waveform, propagation time and reflected power at different grazing angles; and Respectively represent parameters and The variance of Representation parameters and By calculating the normalized covariance matrix, the correlation between each acoustic characteristic parameter and the measured data of seabed acoustic reflection can be obtained, and then the acoustic characteristic parameters with higher correlation can be selected as the output parameters of the first sub-model, while the measured data of seabed acoustic reflection are the input parameters of the first sub-model.

[0072] Similar to the first sub-model, the present invention uses the normalized covariance matrix method to analyze the correlation between each acoustic feature parameter and the output parameter of the first sub-model. The calculation formula also uses formula (5). The difference is is selected from various acoustic characteristic parameters, and Selected from each first sub-model output parameter. This is because the first sub-model output parameter is the second sub-model input parameter. By calculating the normalized covariance matrix, the correlation between each acoustic feature parameter and the second sub-model input parameter can be obtained, and then the acoustic feature parameters with higher correlation are screened out as the second sub-model output parameters. In some embodiments, the screened second sub-model output parameters are average particle size, average density and average porosity.

[0073] In summary, the present invention provides a method for screening acoustic characteristic parameters of seabed surface sediments for constructing a deep neural network model. By analyzing the correlation between parameters, the optimal input and output parameters are screened out, which can improve the performance and efficiency of the deep neural network model.

[0074] Step 4: Construct a deep neural network model based on the acoustic characteristic parameters of the seabed surface sediments, including a seabed acoustic-physical information neural network model and a sediment property prediction neural network model.

[0075] The present invention proposes a deep neural network model for inversion of acoustic characteristic parameters and prediction of physical properties of seabed surface sediments. The deep neural network model consists of a first sub-model and a second sub-model. Through the collaborative work of the two sub-models, end-to-end mapping from seabed acoustic reflection measured data to acoustic characteristic parameters, and from acoustic characteristic parameters to physical properties of sediments is achieved.

[0076] Specifically, the seabed acoustic-physical information neural network (SAPINN) model constructed by the present invention is a hybrid model that integrates the classic seabed acoustic model and the physical information neural network (PINN). It uses known acoustic theories and equations, such as the Biot-Stoll model and the Buckingham model, to establish an analytical relationship between acoustic characteristic parameters and the physical properties of sediments. At the same time, the SAPINN model introduces a neural network module for learning and compensating for uncertainties and approximate errors in theoretical models. The output parameters of the SAPINN model include: 1) the number of sedimentary layers, which represents the layered structure of the seabed surface sediments; 2) the thickness of each layer, which describes the vertical scale of each sedimentary layer; 3) the average density of each layer, which reflects the material composition and compactness of each sedimentary layer; 4) the sound velocity depth profile and the sound attenuation coefficient depth profile of each layer, which reflect the propagation and attenuation characteristics of sound waves in the seabed surface sediments.

[0077] The measured data of seabed acoustic reflection in the measurement area are used as input parameters of the SAPINN model. Based on these input parameters, the SAPINN model predicts and outputs the output parameters of the first sub-model as the input of the second sub-model through the forward propagation process, combined with the acoustic theory model and the neural network module. In the training process of the SAPINN model, the idea of ​​PINN is adopted, and the acoustic theory equation is used as the regularization term of the neural network to guide the optimization of the network parameters in a physically reasonable direction. At the same time, through a data-driven approach, the SAPINN model can adapt to the changes in acoustic characteristics under different sea areas and environmental conditions and improve the inversion accuracy.

[0078] The Sediment Property Prediction Neural Network (SPPNN) model is a prediction model based on deep learning, which is used to estimate the physical properties of seabed surface sediments, such as average grain size, average density and average porosity. It takes the sound velocity depth profile, attenuation coefficient depth profile, number of layers, layer thickness and average density predicted by the SAPINN model as input, and learns the nonlinear mapping relationship between acoustic characteristic parameters and physical properties through a multilayer perceptron (MLP) network.

[0079] The input parameters of the SPPNN model are: 1) the sound velocity depth profile and the sound attenuation coefficient depth profile, which come from the prediction results of the SAPINN model; 2) the number of sedimentary layers, the thickness of each layer, and the average density of each layer, which also come from the prediction results of the SAPINN model. Based on these input parameters, the SPPNN model optimizes the network weights and biases through the forward propagation and back propagation processes to generate the following output parameters: 1) the average grain size, average density, and average porosity of the sediments in the entire measurement area, reflecting the overall physical properties of the seafloor surface sediments; 2) the average grain size, average density, and average porosity of each sedimentary layer, providing more detailed vertical stratification characteristics. The output parameters of the SPPNN model characterize the physical properties of the sediments, so they are also called physical parameters of the sediments.

[0080] In the training process of the SPPNN model, a large amount of measured data and prior knowledge, such as drilling sampling analysis and core measurement, are used to build the correspondence between acoustic feature parameters and physical properties. Through end-to-end learning, the SPPNN model can automatically extract and learn the physical property information contained in the input parameters, avoiding the tediousness and subjectivity of manual feature engineering.

[0081] In summary, the deep neural network model proposed in the present invention realizes an integrated solution for the inversion of acoustic characteristic parameters and prediction of physical properties of seafloor surface sediments through the collaborative work of the two sub-models of the SAPINN model and the SPPNN model. The SAPINN model integrates the classical acoustic model and the physical information neural network to improve the accuracy and adaptability of acoustic inversion; the SPPNN model uses deep learning technology to directly predict the physical properties of sediments from acoustic characteristic parameters, simplifying the traditional inversion and interpretation process. This deep neural network model provides new technical means and application ideas for the fields of marine geophysical exploration, seabed engineering construction, environmental monitoring, etc.

[0082] Step 5: Expand the training sample set based on the measured data of seabed acoustic reflection and the acoustic characteristic parameters of seabed surface sediments.

[0083] A large amount of measured seabed acoustic reflection data is generated using sound field simulation software, and paired with labels consisting of corresponding acoustic characteristic parameters of seabed surface sediments to form a training sample set; the training sample set should cover different seabed types, terrain features and frequency ranges.

[0084] The sample generation process based on the forward model is as follows: Based on the measured data of seabed acoustic reflection, for typical sea areas (sandy mud shallow sea areas and muddy shallow sea areas), based on the seabed acoustic model of the VGS (the Viscous Grain Shearing theory) model and Hamilton's geoacoustic model, it is assumed that the SSS is composed of a stack of multiple sedimentary layers with flat interfaces, and the vertical SSS acoustic characteristics are studied. The acoustic characteristic parameters studied include the number of sedimentary layers, layer thickness, average grain size, average density, average porosity, sound speed depth profile (sound speed profile), sound attenuation coefficient depth profile (sound attenuation profile) and other parameters. The vertical changes of SSS sound speed and sound attenuation coefficient under the spatial sedimentary characteristics of typical seabed environments are simulated at frequencies of 200Hz, 500Hz, 1000Hz, 2000Hz and 10000Hz. From the perspective of the physical and mechanical properties of SSS, the theoretical prediction results of the acoustic characteristic parameters of the more complex structure SSS are analyzed. Using the SSS data of the double-layer structure in the sandy mud shallow sea area, the depth profiles of the compression wave and shear wave velocity and the acoustic attenuation coefficient under different frequency conditions are calculated. According to the physical parameters and environmental parameters of the bottom data in the muddy shallow sea area, the acoustic characteristic parameters of the three-layer structure SSS are calculated. The simulation results show that with the change of water depth, the change of SSS acoustic characteristics is consistent with the change of physical parameters.

[0085] Physics-informed Neural Network (PINN) is a type of neural network used to solve supervised learning tasks while respecting any given physical laws described by general nonlinear partial differential equations. Its principle is that it can learn the distribution law of training data samples like traditional neural networks, and can learn the physical laws described by mathematical equations; its advantage is that compared with purely data-driven neural network learning, PINN imposes physical information constraints during the training process, so it can learn a more generalized model with fewer data samples.

[0086] In some embodiments, the present invention is based on the sound field model of the PINN network, and uses the marine environment parameters to more accurately complete the sound field simulation in the marine environment. The theoretical prediction accuracy of the marine environment parameters and the authenticity of the observation data are crucial to the influence of sound propagation in shallow water environments. The present invention analyzes the influence of factors such as water depth, water body sound velocity and acoustic characteristic parameters of seabed surface sediments on sound propagation. Generally speaking, the sound velocity profile is the most important analysis of the water body sound velocity. The sound velocity profile in the shallow sea environment of the present invention has a positive slope and a linear relationship, with a speed range of 1469m / s to 1472m / s, and a water depth range of 50m to 100m. Referring to the previous research results, the following four typical seabed surface sediments were selected for ocean acoustic field simulation: 1) mud, referring to the 2017 Seabed Characterization Experiment (SBCEXP) sea area; 2) muddy sand; 3) sandy mud, generally distributed in the shelf break in the bay, with the top sediment composed of fine-grained sediments and fine or medium-grained sand below; 4) sand, generally distributed in shallow seabed. It is assumed that each sediment layer can be described by six parameters: compression wave sound velocity and sound attenuation coefficient, sediment layer thickness, constant density, sound velocity gradient and sound attenuation gradient. Assuming that the water body has a constant sound velocity and density, each sediment layer can be simplified by the compression wave sound velocity.

[0087] Step 6: Use the training sample set to train and test the deep neural network model. After the training is completed, the seabed sediment classification model is obtained.

[0088] By training the deep neural network model with a training sample set, accurate estimation and prediction of the seabed acoustic characteristic parameters can be achieved. Aiming at the limitations of traditional inversion methods, such as high computational complexity and strong dependence on prior knowledge, the present invention introduces the data-driven idea of ​​deep learning, and automatically learns and extracts the intrinsic mapping relationship between the seabed acoustic characteristic parameters and the acoustic observation data (i.e., the measured data of seabed acoustic reflection) through a large number of training samples, thereby improving the accuracy and efficiency of the inversion.

[0089] In the inversion of high-frequency seabed acoustic characteristic parameters, the sample training method of the SAPINN model mainly includes the following steps:

[0090] 6.1.1) Construction of training samples: Based on the submarine acoustic theoretical model and known submarine parameters, a large amount of acoustic observation data is generated using acoustic field simulation software, and paired with the corresponding submarine acoustic characteristic parameter labels to form a training sample set. These samples should cover different submarine types, terrain features, frequency ranges, etc. to ensure the generalization ability of the model.

[0091] 6.1.2) Data preprocessing: Preprocess the original acoustic observation data, such as denoising, normalization, feature extraction, etc., to improve data quality and feature representation capabilities. At the same time, encode and convert the seabed acoustic feature parameter labels to adapt them to the output format of the deep neural network model.

[0092] 6.1.3) Model architecture selection: According to the characteristics and complexity of the inversion task, a suitable deep neural network model architecture is selected. The present invention adopts the SAPINN model network structure that integrates the classic seabed acoustic model and PINN, which can effectively learn and represent the local and global features in the acoustic observation data, and capture its spatial correlation and multi-scale changes with the seabed acoustic characteristic parameters.

[0093] 6.1.4) Training strategy design: Use appropriate training strategies, such as batch training, learning rate scheduling, regularization methods, etc., to optimize the convergence speed and generalization performance of the model. At the same time, introduce cross-validation, early stopping and other techniques to prevent the model from overfitting the training data.

[0094] 6.1.5) Loss function definition: According to the objectives and evaluation indicators of the inversion task, a suitable loss function is designed to measure the difference between the model prediction value and the true value, and to guide the optimization direction of the model parameters.

[0095] The PINN in the SAPINN model of the present invention is a neural network that uses a network constraint that approximates a physical system and a loss function, which can be expressed as , whose loss function includes boundary conditions and basic equations, as shown in the following formula (6):

[0096] (6)

[0097] in, is the loss function of PINN; is the initial sound field; For time, is the spatial position, is the sound source position, subscript Indicates sound source, subscript Indicates sound source, ; , , and are the symbolic upper bounds of their respective summations; is a hyperparameter; is the network parameter of PINN; is the velocity model; is a differential operator; Represents the norm.

[0098] In order to further improve the performance of deep neural network models in the inversion of high-frequency seabed acoustic characteristic parameters, the present invention proposes a new loss function for the entire SAPINN model, called the Weighted Mean Squared Error Loss for Seabed Acoustic Parameter (WMSE-SAP). This loss function combines the physical properties of seabed acoustic characteristic parameters and the importance of the inversion task, and weights the prediction errors of different parameters, thereby optimizing the model training process in a more targeted manner.

[0099] The WMSE-SAP loss function is defined as follows:

[0100] (7)

[0101] in, is the value of the weighted mean square error loss function of the seabed acoustic characteristic parameters; is the sample size, is the number of types of acoustic characteristic parameters; For the In the sample The true value of the acoustic characteristic parameter; The SAPINN model is In the sample The predicted values ​​of acoustic characteristic parameters; For the The weight coefficient of the acoustic characteristic parameter. The value of is determined according to the physical meaning of different acoustic characteristic parameters and the importance of the inversion task. For example, for parameters such as seabed sediment thickness, sound velocity, density, etc. that have a greater impact on the propagation and scattering of the sound field, a higher weight can be given; while for some minor parameters or parameters with greater uncertainty, a lower weight can be given. The setting of the weight coefficient can be determined by expert experience, prior knowledge or data analysis, or it can be adaptively adjusted through learning.

[0102] After the introduction of the WMSE-SAP loss function, the SAPINN model will pay more attention to the seafloor acoustic characteristic parameters that have a greater impact on the inversion task during the training process, and adjust the model parameters accordingly to reduce the prediction errors of these parameters. This can effectively improve the accuracy and reliability of the inversion results, especially when the importance of parameters is uneven or the data quality is inconsistent. In addition to the WMSE-SAP loss function, other loss functions can be developed and optimized according to specific inversion tasks and data characteristics, such as weighted mean absolute error loss, segmented loss function, multi-task loss function, etc., to further improve the inversion performance of the SAPINN model.

[0103] 6.1.6) Model training and tuning: Using the constructed training sample set, the weight and bias parameters of the model are continuously updated and optimized through forward propagation and back propagation algorithms, so that the model can learn the mapping relationship between acoustic observation data and seabed acoustic characteristic parameters while minimizing the loss function. During the training process, the convergence speed and generalization ability of the model are improved by adjusting hyperparameters and introducing regularization terms.

[0104] 6.1.7) Model evaluation and application: Evaluate the performance of the trained SAPINN model on an independent test sample set, calculate inversion accuracy, computational efficiency and other indicators, and compare with traditional inversion methods. Apply the trained model to the actual seafloor acoustic characteristic parameter inversion task, and combine prior knowledge and physical constraints to further improve the reliability and interpretability of the inversion.

[0105] In summary, the present invention proposes a method for training a SAPINN model using training samples in the inversion of high-frequency seabed acoustic characteristic parameters. By constructing a large number of training samples, selecting a suitable model architecture and loss function, and adopting an effective training strategy and tuning method, high-precision and high-efficiency inversion of seabed acoustic characteristic parameters is achieved. Among them, the specially developed WMSE-SAP loss function further improves the accuracy and reliability of the inversion results by weighted processing of the prediction errors of different seabed acoustic characteristic parameters. The trained SAPINN model of the present invention makes full use of the data-driven and feature learning capabilities of deep learning, overcomes the limitations of traditional inversion methods, and provides new technical means and application ideas for the fields of marine geophysical exploration, seabed engineering construction, and environmental monitoring.

[0106] Similar to the SAPINN model, the training process of the SPPNN model of the present invention is described as follows:

[0107] 6.2.1) Data preparation: Collect and annotate a large amount of high-frequency seafloor acoustic observation data as training samples, covering different types of seafloor substrates, geographical locations, and water depth conditions. Preprocess these acoustic observation data and extract acoustic feature parameters.

[0108] 6.2.2) Model selection: The SPPNN model is used as the basic architecture of the deep neural network model. The SPPNN model can effectively extract and learn local and global features in acoustic feature parameters and is suitable for seabed sediment classification tasks.

[0109] 6.2.3) Model design: In the SPPNN model, a multi-scale feature extraction and fusion strategy is adopted. By setting convolution kernels of different sizes and numbers, the multi-scale features of acoustic feature parameters are extracted and combined and integrated through the feature fusion layer.

[0110] 6.2.4) Training strategy: An end-to-end learning method is used to directly input the preprocessed acoustic feature parameters into the SPPNN model, and the model parameters are optimized through forward propagation and back propagation. To prevent overfitting, data enhancement and regularization techniques are used.

[0111] 6.2.5) Hyperparameter tuning: Use grid search, random search and other methods to tune hyperparameters such as learning rate, batch size, regularization coefficient, and select the optimal parameter combination.

[0112] 6.2.6) Model evaluation: Use methods such as cross-validation to evaluate the trained SPPNN model, calculate indicators such as classification accuracy, precision, recall, and perform error analysis and visualization.

[0113] 6.2.7) Model deployment: The trained SAPINN model and the trained SPPNN model are deployed together as the seabed sediment classification model into the actual seabed acoustic sediment classification system to perform real-time classification and prediction of new acoustic observation data.

[0114] In the high-frequency seafloor acoustic bottom classification task, choosing a suitable loss function is crucial to training a high-performance deep neural network model. The following loss functions can be used in the deep neural network model training process of the present invention:

[0115] 6.3.1) Cross-entropy Loss: Cross-entropy loss is one of the most commonly used loss functions in multi-classification tasks. It measures the difference between the model prediction distribution and the true label distribution. The formula is as follows:

[0116] (8)

[0117] in, is the sample size; is the number of categories, which in the present invention is the number of types of acoustic characteristic parameters; For sample In category The true label (also called true value) on For models in the category The predicted probability (also called predicted value) on It refers to the Acoustic characteristic parameters of seafloor surface sediments.

[0118] 6.3.2) Center Loss: Center loss is an auxiliary loss function used to enhance the discriminability and clustering of features. It minimizes the internal variance of features of the same sample and maximizes the spacing between features of different samples to improve classification accuracy. The formula is as follows:

[0119] (9)

[0120] in, For sample The characteristic representation of For sample Category feature center.

[0121] 6.3.3) Total loss function: In order to balance classification accuracy and feature discriminability, this paper adopts a weighted combination of cross entropy loss and center loss as the total loss function :

[0122] (10)

[0123] in, is the weight coefficient of the center loss, which is used to balance the contribution of the two losses.

[0124] Through the design and optimization of the above loss functions, the present invention can effectively train a high-performance deep neural network model to improve the accuracy and reliability of high-frequency seabed acoustic bottom classification. At the same time, the selection and combination strategies of these loss functions also provide references and inspirations for other similar acoustic classification tasks.

[0125] Step 7: Use the seabed sediment classification model to classify the seabed sediments in the survey area.

[0126] The present invention proposes a method for fine classification of seafloor surface sediments using a deep neural network, and obtains physical parameters of sediments by predicting and outputting using a trained seafloor sediment classification model. According to standards such as the "Marine Survey Specification" (GB / T 12763-2007) and the "Marine Geological Survey Specification" (GB / T 13909-2006), the physical parameters of seafloor surface sediments obtained at the station, including particle size, density, porosity, etc., are used, and the internationally recognized seafloor sediment classification method applicable to the area, such as the Folk classification method or the Shepard classification method, is applied to classify and describe the seafloor surface sediments at the station in detail, so as to achieve high-precision and high-resolution bottom identification and characterization.

[0127] After completing the training of the deep neural network model, the measured data of the seabed acoustic reflection of the seabed surface sediments to be classified are input into the seabed sediment classification model for parameter inversion calculation. The model learns and extracts the physical property information of the sediments contained in the input data through layer-by-layer feature extraction and nonlinear transformation, such as average particle size, average density, average porosity and other physical parameters. In the output layer, the model generates a vector containing multiple predicted values ​​of physical parameters. Assume that the output layer has neurons, corresponding to predefined sediment physical parameters. For a given input data , the output of the model for:

[0128] (11)

[0129] in, Represents the mapping function of the seabed sediment classification model, Indicates The predicted values ​​of physical parameters such as average grain size, average density and average porosity can be obtained.

[0130] According to the physical parameters of sediments predicted by the seabed classification model, combined with the standards such as the "Marine Survey Code" (GB / T12763-2007) and the "Marine Geological Survey Code" (GB / T 13909-2006), the seabed surface sediments are finely classified. These standards stipulate the principles, methods and naming rules for the classification of seabed surface sediments, and provide a unified classification system and terminology. According to the "Marine Survey Code" (GB / T 12763-2007), seabed surface sediments can be divided into the following types: rock, gravel, sand, silt, clay, mud, volcanic ash, bioclastic debris and others. Among them, types such as sand, silt, and clay can be further divided into different subclasses based on parameters such as particle size and sorting coefficient. For example, sand can be divided into very coarse sand, coarse sand, medium sand, fine sand, and very fine sand.

[0131] According to the "Marine Geological Survey Specification" (GB / T 13909-2006), the particle size parameters of seafloor surface sediments can be described by the following indicators: mean grain size, median grain size, sorting coefficient, skewness, and kurtosis. The mean grain size and median grain size reflect the average grain size of the sediment, the sorting coefficient reflects the uniformity of the sediment particle size, the skewness reflects the symmetry of the grain size distribution, and the kurtosis reflects the sharpness of the grain size distribution curve. According to the predicted physical parameters of the sediment, each sample point is classified according to national standards. For example, according to the mean grain size and sorting coefficient, the sediment can be divided into the following categories:

[0132] 1) Sand: Very coarse sand: average particle size 1-2mm, sorting coefficient <0.35; coarse sand: average particle size 0.5-1mm, sorting coefficient 0.35-0.50; medium sand: average particle size 0.25-0.5mm, sorting coefficient 0.50-0.70; fine sand: average particle size 0.125-0.25mm, sorting coefficient 0.70-1.00; very fine sand: average particle size 0.0625-0.125mm, sorting coefficient 1.00-2.00;

[0133] 2) Silt: average particle size 0.004-0.0625mm, sorting coefficient>2.00;

[0134] 3) Clay: average particle size <0.004mm.

[0135] In addition to the particle size parameters, other physical parameters such as density and porosity can also be considered to make a more detailed sub-classification and description of sediments. For example, according to density, it can be divided into loose, medium-dense, and dense, etc.; according to porosity, it can be divided into high porosity, medium porosity, and low porosity, etc. By classifying and describing the physical parameters of each sample point, a refined classification result of the seafloor surface sediments is obtained. In order to improve the continuity and consistency of the classification, the spatial correlation between sample points can be further considered, and the classification results can be optimized by using spatial interpolation, smoothing and other methods.

[0136] Based on the prediction output and standardized classification of the deep neural network model, the present invention generates high-precision and high-resolution classification maps and attribute tables of seafloor surface sediments. These results can intuitively display the spatial distribution characteristics and physical property changes of seafloor surface sediments, and provide important basic data and decision support for marine geological surveys, resource evaluation, engineering construction, etc. The classification results can be superimposed and analyzed with other marine geological data (such as topography, landforms, structures, etc.) to reveal the relationship between seafloor surface sediments and environmental conditions, and explain the dynamic mechanism of sediment transportation, deposition and evolution. At the same time, refined bottom classification also provides important basic information for marine ecological environment evaluation, benthic biological surveys, etc. Through three-dimensional visualization technology, three-dimensional models and scenes of seafloor surface sediment classification can be generated to intuitively display the distribution characteristics of sediments in the vertical and horizontal directions, and improve the readability and application value of classification results.

[0137] In summary, the present invention utilizes the prediction output of the deep neural network model to improve the robustness and accuracy of the seabed sediment classification method and the environmental adaptability of the model in the shallow sea acoustic environment, thereby realizing the qualitative and quantitative classification of the seabed sediment. Compared with traditional manual interpretation and simple interpolation methods, the method of the present invention makes full use of the feature extraction and prediction capabilities of deep learning, takes into account the various physical properties and spatial correlations of sediments, and provides more accurate, detailed, and continuous sediment classification results, which provides important technical support and data guarantee for marine geological surveys, resource exploration, engineering construction and other fields, and promotes the development of marine scientific research and practical applications.

[0138] In some embodiments, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the method for classifying seabed sediments based on acoustic characteristics of seabed surface sediments.

[0139] In some embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments.

[0140] In some embodiments, the present invention also provides a computer device, which includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it can implement the seabed sediment classification method based on the acoustic characteristics of seabed surface sediments.

[0141] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for classifying seafloor sediments based on acoustic characteristics of seafloor surface sediments, characterized in that: include: Acquiring the measured data of seabed acoustic reflection in the measurement area in the marine environment by using an acoustic in-situ measurement technology based on the acoustic reflection method; the measured data of seabed acoustic reflection includes the reflected signal waveform, propagation time and reflected power at different grazing angles; Based on the acquired seabed acoustic reflection measured data, the seabed surface sediment characteristic parameters are obtained by using the seabed surface sediment characteristic parameter inversion method; the seabed surface sediment characteristic parameters include: sediment type, particle size, density, porosity, impedance, absorption coefficient, sound velocity, sound attenuation coefficient and sound reflection coefficient, as well as the number of sediment layers, layer thickness, sound velocity distribution, sound velocity depth profile, sound attenuation coefficient depth profile, average particle size, average density and average porosity; By analyzing the uncertainty and correlation of the characteristic parameters of the seabed surface sediments, the acoustic characteristic parameters of the seabed surface sediments for constructing the deep neural network model are screened out; the acoustic characteristic parameters of the seabed surface sediments are the number of sediment layers, layer thickness, average grain size, average density, average porosity, sound velocity depth profile, and sound attenuation coefficient depth profile; Construct a deep neural network model based on the acoustic characteristic parameters of seafloor surface sediments, including a seafloor acoustic-physical information neural network model and a sediment property prediction neural network model; Expand the training sample set based on the measured data of seabed acoustic reflection and the acoustic characteristic parameters of seabed surface sediments; The deep neural network model is trained and tested using the training sample set, and a seabed sediment classification model is obtained after the training is completed; The seabed sediment classification model is used to classify the seabed sediments in the survey area.

2. The method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments according to claim 1, characterized in that: The method of using the acoustic in-situ measurement technology based on the acoustic reflection method to obtain the measured data of the seabed acoustic reflection in the measurement area in the marine environment specifically includes: An acoustic in-situ measurement device for seafloor surface sediments is constructed by using an acoustic in-situ measurement technology based on an acoustic reflection method; the acoustic in-situ measurement device for seafloor surface sediments comprises a sound source system, a receiving system, a data acquisition and processing system, and a synchronous control system; the sound source system comprises a signal generator, a power amplifier, and a transducer array, which are used to generate an acoustic beam with a preset frequency, waveform, and grazing angle; the receiving system comprises a hydrophone array and a preamplifier, which are used to receive and amplify the seafloor acoustic reflection echo signal; the data acquisition and processing system is used to perform digital conversion, filtering, gain compensation, beamforming, and data analysis and processing on the amplified echo signal, and extract the measured data of the seafloor acoustic reflection; the synchronous control system is used to provide a unified clock and trigger signal for the sound source system and the receiving system, so as to ensure the synchronization of transmission and reception; Using the acoustic in-situ measurement device for seabed surface sediments, the measurement tasks of sound wave emission, reception, collection, storage and processing are automatically completed according to the preset emission waveform, emission voltage, emission width and measurement position parameters, and the reflected signal waveform, propagation time and reflection power at different grazing angles in the measurement area are obtained as the measured data of seabed acoustic reflection; during the measurement process, the mechanical or electronic scanning of the transducer array is controlled to achieve sound wave incidence at different grazing angles.

3. The method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments according to claim 2, characterized in that: The method of obtaining the characteristic parameters of the seafloor surface sediments based on the acquired seafloor acoustic reflection measured data and using the seafloor surface sediment characteristic parameter inversion method specifically includes: By using the inversion method of the characteristic parameters of the seabed surface sediments, the type, particle size, density and porosity of the sediments are inferred by analyzing the change law of the reflected signal waveform; the inversion method of the characteristic parameters of the seabed surface sediments includes the least squares method, particle swarm optimization algorithm, genetic algorithm, Bayesian inversion method, back propagation neural network inversion method and neural network inversion method; By measuring the propagation time difference at different grazing angles, the number of sediment layers, layer thickness and sound velocity distribution are calculated using the inversion method of seafloor surface sediment characteristic parameters. By measuring the reflected power spectrum at different grazing angles, the seabed acoustic reflection coefficient is calculated using the seabed surface sediment characteristic parameter inversion method, and then the impedance, absorption coefficient and acoustic attenuation coefficient of the sediment are inverted; The average grain size, average density and average porosity of the sediment layer are further calculated based on the grain size, density and porosity of the sediment; The sound velocity depth profile and the sound attenuation coefficient depth profile of the sediment layer are further calculated based on the sound velocity distribution and the sound attenuation coefficient of the sediment layer.

4. The method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments according to claim 3, characterized in that: By analyzing the uncertainty and correlation of the characteristic parameters of the seafloor surface sediments, the acoustic characteristic parameters of the seafloor surface sediments used to construct the deep neural network model are screened out, specifically including: The variance method was used to analyze the uncertainty of the characteristic parameters of the seafloor surface sediments, and the number of sediment layers, layer thickness, average grain size, average density, average porosity, sound velocity depth profile, and sound attenuation coefficient depth profile were selected as the acoustic characteristic parameters of the seafloor surface sediments for constructing the deep neural network model. The normalized covariance matrix method is used to analyze the correlation between the acoustic characteristic parameters of the seafloor surface sediments and the measured data of seafloor acoustic reflection, and the number of sediment layers, layer thickness, average density, sound velocity depth profile, and sound attenuation coefficient depth profile are selected as the output parameters of the first sub-model; The normalized covariance matrix method was used to analyze the correlation between the acoustic characteristic parameters of the seafloor surface sediments and the output parameters of the first sub-model, and the average grain size, average density and average porosity of the entire measurement area and each sediment layer were selected as the output parameters of the second sub-model.

5. The method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments according to claim 4, characterized in that: The deep neural network model is constructed based on the acoustic characteristic parameters of the seafloor surface sediments, including a seafloor acoustic-physical information neural network model and a sediment property prediction neural network model, specifically including: A seabed acoustic-physical information neural network model is constructed by integrating a classic seabed acoustic model and a physical information neural network, wherein the input of the seabed acoustic-physical information neural network model is the measured data of seabed acoustic reflection in the measurement area, and the output is the output parameter of the first sub-model; The first sub-model output parameter is used as the input of the sediment property prediction neural network model, and the output of the sediment property prediction neural network model is the second sub-model output parameter.

6. The method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments according to claim 5, characterized in that: The method of expanding the training sample set based on the measured seabed acoustic reflection data and the acoustic characteristic parameters of the seabed surface sediments specifically includes: A large amount of measured seabed acoustic reflection data is generated as samples using sound field simulation software, and the samples are paired with labels consisting of corresponding acoustic characteristic parameters of seabed surface sediments to form a training sample set; the training sample set covers different seabed types, terrain features and frequency ranges.

7. The method for classifying seabed sediments based on the acoustic characteristics of seabed surface sediments according to claim 6, characterized in that: During the training process of the deep neural network model, the loss function used is ;in is the value of the weighted mean square error loss function of the seabed acoustic characteristic parameters; is the sample size; is the number of types of acoustic characteristic parameters of seafloor surface sediments; For the In the sample The true values ​​of acoustic characteristic parameters of the seafloor surface sediments; The deep neural network model is In the sample Prediction of acoustic characteristic parameters of seafloor surface sediments; For the The weight coefficients of acoustic characteristic parameters of the seafloor surface sediments.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the seabed sediment classification method based on the acoustic characteristics of seabed surface sediments described in any one of claims 1 to 7.

Images