Method for detecting ground sound and classifying sediment based on multi-acoustic feature fusion of double-layer structure
By employing a dual-layer structure ground acoustic detection method that integrates multiple acoustic features, the problems of large inversion bias and insufficient robustness in existing sonar detection technologies have been solved. This method achieves high-precision sediment grain size classification and improves the stability and stratification resolution of sediment identification.
Patent Information
- Application Number
- CN202610641630.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-11
- Publication Date
- 2026-08-25
AI Technical Summary
Existing sonar detection technologies rely on a single acoustic feature in the classification of bottom sediments, resulting in large inversion biases and insufficient robustness. This makes it difficult to achieve high-precision and high-robustness sediment grain size classification, especially for layered sediments.
A dual-layer structure ground acoustic detection method based on multi-sound feature fusion was adopted. Multi-dimensional acoustic features such as reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrix, sub-band reflection coefficient and sub-band positive and negative reflection matrix were extracted from HF-SSBP echo information. Combined with Folk triangulation method, a multi-sound feature joint ground acoustic model dominated by sediment grain size and assisted by density and porosity was constructed. The environmental applicability was improved through laboratory calibration and field correction.
It significantly improves the stability and stratification resolution of sediment identification, enabling it to penetrate the surface layer to obtain sub-layer structural information, achieve high-precision classification of bottom sediments, and enhance the engineering applicability of the model in complex field environments.
Smart Images
Figure CN122632267A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of sonar detection technology, specifically a two-layer structure ground acoustic detection and sediment classification method based on multi-sound feature fusion. Background Technology
[0002] Sonar detection is a crucial tool for studying the properties of sediments on the seabed of rivers, lakes, and oceans. Sonar-based sediment classification is invaluable for water conservancy projects, oil and gas exploration, underwater facility safety monitoring, and river / reservoir sediment cleanup. Essentially, sonar sediment classification studies the relationship between the acoustic echo properties of sediments and their physical properties (primarily grain size). Ground acoustic models are the core foundation for inverting and identifying the acoustic characteristics and types of seabed sediments. Traditional ground acoustic models are often built based on single acoustic features, and due to the complexity of ground acoustic relationships and the variability of the field measurement environment, they suffer from large inversion biases, insufficient robustness, and weak resolution of layered sediments. While existing high-frequency shallow seismic profilers (HF-SSBPs) possess high-resolution layer structure identification and broadband echo information acquisition capabilities, they have not yet fully explored the more multidimensional acoustic features in the echoes for systematic modeling, making it difficult to achieve high-precision and robust sediment grain size classification. Therefore, it is necessary to develop a new ground acoustic detection and classification method that can fully utilize HF-SSBP echo information, integrate multi-sound features, and adapt to the structure of two-layer sediments, so as to improve the classification accuracy and engineering applicability of underwater sediments.
[0003] Disadvantages of existing technology:
[0004] Currently, there are three main methods for classifying bottom sediments using sonar detection. The first is based on multibeam backscattering intensity, which relies on backscattering intensity and angular response curves to identify the sediment. This method is susceptible to interference from seabed topography, changes in incident angle, and ambient noise, resulting in poor classification accuracy and stability. The second is based on side-scan sonar image features, which primarily extracts features from image dimensions such as grayscale and texture to identify the sediment. This method is easily affected by external factors such as water flow, shadows, benthic organisms, and water turbidity. The first category has several drawbacks. First, there's a lack of direct quantitative correlation between image features and sediment physical parameters, allowing only macroscopic type identification and failing to retrieve key physical properties required for underwater engineering, such as grain size, density, and porosity. Second, there are sediment classification methods based on empirical physical models. These rely on Biot-Stoll theory, Hamilton's empirical formula, and other methods, depending on single acoustic parameters like sound velocity and attenuation coefficients to retrieve sediment properties. However, these models contain many physical parameters difficult to measure in the field, leading to complex parameter calibration, limited applicability, and insufficient robustness and adaptability to field environments. Furthermore, all of these methods only reflect the surface acoustic characteristics of sediments and cannot penetrate the surface to obtain sub-layer structural information, resulting in insufficient ability to distinguish between two layers of sediments. Summary of the Invention
[0005] To address the shortcomings of existing sonar sediment classification methods, which rely on single acoustic features, suffer from large model biases, weak robustness, and insufficient ability to distinguish layered sediments, this invention aims to provide a two-layer structure ground acoustic detection and sediment classification method based on multi-acoustic feature fusion. This method is adapted to HF-SSBP profile detection, integrates multi-dimensional acoustic features with a two-layer structure, and utilizes HF-SSBP acoustic echo information to extract multi-dimensional acoustic features such as reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrices, sub-band reflection coefficients, and sub-band positive and negative reflection matrices. Based on the Folk triangulation method, a multi-acoustic feature joint ground acoustic model is constructed, dominated by sediment grain size and supplemented by density and porosity. Contribution fusion and regularized regression are employed to suppress collinearity and overfitting. Laboratory calibration and field correction enhance environmental applicability, achieving effective classification of seabed sediments. This significantly improves the stability, layer resolution, and classification accuracy of seabed sediment identification, and enhances the model's engineering applicability in complex field environments.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure, comprising the following steps:
[0007] S1. In a controlled laboratory environment, relying on the high-resolution layer structure identification capability of HF-SSBP, the traditional single-layer sediment geosonic model is refined into a two-layer structure model, the contribution of the secondary layer reflection signal to sediment classification is quantified, a certain amount of HF-SSBP acoustic echo data is collected, and the particle size, density and porosity of sediment samples are measured and recorded simultaneously.
[0008] S2. Extract six types of acoustic characteristic parameters from the echo data: reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrix, sub-band reflection coefficient, and sub-band positive and negative reflection matrix.
[0009] S3. The above 6 types of acoustic feature parameters are comprehensively modeled, parametrically fitted, and regularized regression is used to obtain the contribution coefficient to suppress the risk of feature collinearity and overfitting.
[0010] S4. Select a sedimentary area with good natural sorting and record actual ground acoustic echo data under the field navigation mode. Simultaneously, use sediment samples for grain size analysis and verify the independent characteristic parameters and the parameter model dominated by sediment grain size and the fused ground acoustic model through measured data.
[0011] S5. Based on the actual field measurement results, revise the ground acoustic model based on the laboratory standardized measurement to improve the model's applicability to the real subsurface environment.
[0012] Preferably, the linear frequency modulated signal transmitted by the HF-SSBP is
[0013] (1)
[0014] in For carrier frequency, The bandwidth is [missing value]; the echo received by the double-layer sediment medium is [missing value].
[0015] (2)
[0016] in Let represent the echo signal amplitude of the i-th layer, where i=0 represents the first layer and i=1 represents the second layer. The distance between the HF-SSBP emission surface and the i-th layer of sediment is given by pulse compression processing.
[0017] (3)
[0018] in After pulse compression processing, the amplitude of the echo signal of the i-th layer, as shown in equation (3), will appear at the interface layer. Pulse spikes, through continuous ping echoes, can form a two-dimensional shallow profile image. The Canny algorithm is used to extract the boundary layer structure, and its core Sobel operator is...
[0019] (4)
[0020] in These represent the magnitude and phase of the gradient intensity at each pixel. The layer structure of the shallow subsurface profile image is extracted using the same phase axis. The constraint is to find vertical local maxima. Assuming the short-term measurement layer structure remains unchanged and ignoring the influence of the x-direction, the shallow seismic profile layer structure and the reflection amplitudes of the first and second layers can be expressed as follows:
[0021] (5)
[0022] in These are the local extrema of the first and second layers of the shallow profile image found using the Canny algorithm.
[0023] Preferably, the reflection coefficient is calculated using the reference interface ratio method, with the water-air interface as the reference, and the amplitude of the echo received at the transducer during water-air reflection is recorded across the entire shallow probing range. Therefore, the reflectance of the first layer of sediment is
[0024] (6)
[0025] Under the condition that the amplitudes of both are at the same distance, the transmission and reflection patterns of the two sediment layers can be obtained.
[0026] (7)
[0027] in These are the reflection coefficients of the secondary layer, Attenuation occurs in water and sediment layers, respectively. The emitted wave at the first-layer interface is pre-calibrated based on the total internal reflection experiment at the water-air interface. According to equation (7), the autoregressive formula for the separation of the reflection coefficients of the first and second layers is as follows:
[0028] (8)
[0029] in Multiple sets of acoustic echo data at different distances need to be collected and then fitted and optimized using the above formula.
[0030] Preferably, regarding the extraction of the layer structure, it is based on the two-dimensional image formed by pulse compression of HF-SSBP echo. Combining equations (1) to (5), the Canny algorithm and Sobel operator are used for edge detection, and the vertical local maxima are extracted to locate the in-phase axis. The interface position and reflection amplitude of the first and second layers of the double sediments are obtained, and the accurate identification of the layer structure is completed. Regarding the extraction of the reflection coefficient, the total reflection signal of the water-air interface is used as a reference. Combining equation (6), the reference interface ratio method is used to calculate the reflection coefficient of the first layer of sediments. This illustrates the propagation path of sound waves in a two-layer sediment structure, where the total propagation path of the first layer is... The total path of secondary propagation is Combining the transmission-reflection law and attenuation characteristics of the double-layer medium, and combining equations (7) and (8), the reflection coefficients of the first and second layers are separated and calculated through the autoregressive formula, and the reflection coefficients of the first and second layers are obtained.
[0031] Preferably, the waveform of the first wave is quantitatively described, i.e., using the wavefront funnel shape factor. and waveform fluctuation factor The description of the initial waveform shape is as follows:
[0032] (9)
[0033] Where k is the slope of the wavefront funnel. The resulting curve is the envelope of the first wave waveform. The parameters are fitted using the least squares method.
[0034] Preferably, based on the fundamental principles of acoustics and the property of sound wave reflection in different propagation media, the echo phase at the reflection interface propagating from a low-impedance region to a high-impedance region remains consistent with the emitted waveform. In the case of a reflection wave propagating from a high-impedance region to a low-impedance region, the phase of the reflected wave at the interface will differ from the transmitted waveform. By flipping the HF-SSBP, which transmits a coherent wave, the phase information of the received echo is completely preserved, defining the positive and negative reflection matrices.
[0035] (10)
[0036] As one of the parameters for identifying sediment types.
[0037] Preferably, a dispersion characteristic analysis method based on subband decomposition is used to divide the broadband signal into subband signals with overlap. In this case, the subband signals are short linear frequency modulated signals with a certain bandwidth and duration, possessing coherence and a certain duration. A dual filtering method of traditional filtering + subband matched filtering is used to reduce mutual interference between subbands. The theoretical basis is...
[0038] (11)
[0039] in For the sub-band received signal set, The number of sub-bands, For the defined subband bandwidth, and The center frequency and received amplitude of the i-th sub-band are respectively given. Pre-filtering separates the frequencies of each sub-band. Sub-band matched filtering is then performed on top of this, suppressing frequencies outside the defined sub-band range that contribute no to sub-band matching.
[0040] (12)
[0041] in This is a collection of pulse compressions from each frequency band channel after filtering. Since the pulse compression result reflects the properties of the center frequency, the pulse compression result within each frequency band channel is the amplitude and phase characteristics at the center frequency of each sub-band.
[0042] Preferably, based on equation (12), sub-band-based reflection coefficient calibration is performed. The total internal reflection signal of the water surface is processed in the same sub-band manner, and the sediment echo and the total internal reflection echo of the water surface can be subjected to reflection coefficient calibration on their respective sub-bands. The resulting set of reflection coefficients... for
[0043] (13)
[0044] Similarly, the positive and negative reflection matrices of the sub-bands can be obtained. ,for
[0045] (14)
[0046] in It is a three-dimensional matrix, with the two rows representing the positive and negative values of the reflections of the first and second layers, respectively. M is the number of Ping echoes, and the upper right subscript of z represents the sub-band number, which corresponds to the frequency of the sub-band.
[0047] Preferably, regarding the extraction of the first wave waveform, the first wave waveforms of sedimentary echoes of different particle sizes are different. Combining equation (9), the least squares method is used to fit the slope of the wavefront flare, calculate the wavefront flare shape factor and waveform fluctuation factor, and quantitatively describe the shape of the first wave envelope. Regarding the extraction of the positive and negative reflection matrices, the phase information of the coherent echoes is deeply mined. Combining equation (10), the positive and negative reflection matrices are constructed according to the phase reversal law of acoustic impedance interface reflection, and the interface type of sound wave propagating from low impedance to high impedance or from high impedance to low impedance is determined. Regarding the sub-band reflection coefficient and sub-band positive and negative reflection moments... Array extraction, broadband echo is divided into overlapping sub-bands, and after pre-filtering and sub-band matching filtering, the water surface total reflection signal is used for calibration in each sub-band. Combined with equations (10) to (13), the sub-band level reflection coefficient is calculated, making full use of the broadband signal dispersion characteristics to improve the detection efficiency. On the basis of sub-band decomposition and amplitude and phase extraction, phase discrimination and matrix construction are performed on each sub-band echo. Combined with equation (14), a three-dimensional sub-band positive and negative reflection matrix containing sub-band number, layer number and Ping number is formed, realizing the refined characterization of sediment interface impedance characteristics under multiple frequency points.
[0048] Preferably, based on the multi-feature fusion criterion of contribution, a comprehensive model is performed on feature parameters such as reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrices, sub-band reflection coefficient, and sub-band positive and negative reflection matrices, and its expression is:
[0049] (15)
[0050] in
[0051] (16)
[0052] Where Y is the target output, i.e., the sediment grain size classification result. The parameterized model for the above six features, The particle size, density, and porosity corresponding to various types of sediments are as follows: For each feature, a single-parameterized model is provided. This represents the contribution coefficient of each model to the overall model. This represents a systematic bias.
[0053] The beneficial effects of this invention are as follows:
[0054] 1. For the first time, six types of acoustic features are integrated, including reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrix, sub-band reflection coefficient, and sub-band positive and negative reflection matrix. The acoustic characteristics of sediments are characterized from multiple dimensions such as amplitude, phase, time domain waveform, frequency domain, layer structure, and broadband dispersion. This method overcomes the shortcomings of traditional methods that rely on a single acoustic parameter, resulting in large inversion bias and susceptibility to environmental interference. It significantly improves classification stability and environmental adaptability.
[0055] 2. Upgrade the traditional single-layer geosonic model to a double-layer sediment structure model, which can penetrate the surface layer to obtain the reflection information of the second layer, realize the separation calculation of the reflection coefficients of the first and second layers, solve the technical shortcomings of existing methods that can only identify surface sediments and cannot distinguish the two-layer bottom sediments, and improve the resolution of complex sedimentary structures.
[0056] 3. This invention explores the phase reversal characteristics of coherent echoes and constructs positive and negative reflection matrices, which can directly determine the acoustic impedance interface type (low impedance → high impedance / high impedance → low impedance), making up for the shortcomings of traditional reflection coefficients that only represent amplitude and lack phase information, thus making the determination of sediment type more rigorous and the physical meaning clearer.
[0057] 4. Based on the Folk triangulation method, a quantitative correlation is established between acoustic characteristics and particle size (dominant) + density + porosity. This can invert the core physical parameters required for water conservancy, exploration, and dredging projects, breaking through the limitation that side scan / multibeam scanning can only make macroscopic judgments and cannot quantitatively invert the parameters, thus having higher practical engineering value.
[0058] 5. The process of “controlled laboratory calibration – field verification – model calibration” is adopted to ensure that the model maintains high classification accuracy and reliability in real water environment.
[0059] 6. This invention is specifically designed for high-resolution shallow seismic profilers such as HF-SSBP. No new hardware is required; performance can be significantly improved simply by upgrading the algorithm. It has low deployment costs, strong compatibility, and is easy to promote in engineering. Attached Figure Description
[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0061] Figure 1 This is a schematic diagram of the six types of acoustic feature extraction methods in this invention;
[0062] Figure 2 This is a schematic diagram illustrating the construction and classification of a multi-feature fusion ground acoustic model based on contribution in this invention. Detailed Implementation
[0063] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0064] Example 1
[0065] This embodiment discloses a two-layer structure ground acoustic detection and sediment classification method based on multi-sound feature fusion. By fully mining the acoustic echo information of HF-SSBP, multi-dimensional acoustic features such as reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrix, sub-band reflection coefficient and sub-band positive and negative reflection matrix are extracted. Based on the Folk triangulation method, a multi-sound feature joint ground acoustic model is constructed with sediment grain size as the main factor and density and porosity as auxiliary factors. Contribution fusion and regularized regression are used to suppress collinearity and overfitting. The environmental applicability is improved through laboratory calibration and field correction, so as to achieve effective classification of bottom sediments.
[0066] like Figure 1 As shown, firstly, six types of acoustic features are derived and extracted, including reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrices, sub-band reflection coefficient, and sub-band positive and negative reflection matrices. Relying on the high-resolution layer structure identification capability of HF-SSBP, this invention refines the traditional single-layer sediment geoacoustic model into a two-layer structure model, quantifying the contribution of secondary layer reflection signals to sediment classification. The linear frequency modulated signal transmitted by HF-SSBP is...
[0067] (1)
[0068] in For carrier frequency, The bandwidth is [missing value]. The received echo in the double-layer sediment medium is [missing value].
[0069] (2)
[0070] in The amplitude of the echo signal of the i-th layer (i=0 is the first layer, i=1 is the second layer, and so on). This represents the distance from the HF-SSBP emission surface to the i-th layer of sediment. The result after pulse compression processing is...
[0071] (3)
[0072] in The amplitude of the echo signal at layer i after pulse compression processing. The above formula will appear at the interface layer. Pulse spikes, through continuous ping echoes, can form a two-dimensional shallow profile image. The Canny algorithm is considered for extracting the boundary layer structure, with its core Sobel operator being...
[0073] (4)
[0074] in These represent the magnitude and phase of the gradient intensity for each pixel. Considering that the layer structure of shallow subsurface profile images is extracted along the same phase axis, The constraint is to find vertical local maxima. Assuming the short-term measurement layer structure remains unchanged (ignoring the influence of the x-direction), the shallow seismic profile layer structure and the reflection amplitudes of the first and second layers can be expressed as follows:
[0075] (5)
[0076] in These are the local extrema of the first and second layers of the shallow profile image found using the Canny algorithm.
[0077] Next, the reflection coefficient was calculated. The reflection coefficient of the first sediment layer was calculated using the reference interface ratio method, with the water-air interface as the reference. The amplitude of the echo received at the transducer during water-air reflection was recorded in advance at the entire range of the shallow probing exploration. Therefore, the reflectance of the first layer of sediment is
[0078] (6)
[0079] The amplitudes of both must be at the same distance; combined Figure 1 From the transmission and reflection patterns of the two sediment layers in (b) we can obtain...
[0080] (7)
[0081] in These are the reflection coefficients of the secondary layer, The attenuation occurs in water and in sediment layers, respectively. The emitted wave at the first-layer interface is pre-calibrated based on the total internal reflection experiment at the water-air interface. According to equation (7), the autoregressive formula for separating the reflection coefficients of the first and second layers is:
[0082] (8)
[0083] in Multiple sets of acoustic echo data at different distances need to be collected and then fitted and optimized using the above formula.
[0084] The reflection coefficient reflects the overall contribution of acoustic properties at the sediment interface. However, due to the influence of saturated sediment-water structure, different sediment types exhibit varying acoustic penetration and reflection capabilities. For example, in some cases, sonar can penetrate sediment more easily, resulting in a distinct flattened conical echo waveform; conversely, in cases where sonar cannot easily penetrate sediment, the echo waveform will be steeper. Related studies use the first-wave waveform to describe these phenomena, enabling the differentiation of different acoustic reflection types. The first-wave waveform reflects the evolution of the acoustic echo signal over time. Quantitative description of the first-wave waveform is achieved using the wavefront horn shape factor. and waveform fluctuation factor The description of the initial waveform shape is as follows:
[0085] (9)
[0086] Where k is the slope of the wavefront funnel. The resulting curve is the envelope of the first wave waveform. The parameters are fitted using the least squares method.
[0087] Acoustic reflection from sediments occurs because different layers of sediment have different acoustic impedances. The reflection coefficient reflects the amplitude of the acoustic impedance but does not take the phase into account. According to basic acoustic principles, sound waves reflect in different propagation media. At the reflection interface from a low-impedance region to a high-impedance region, the phase of the echo remains consistent with the emitted waveform. In the case of a reflection wave propagating from a high-impedance region to a low-impedance region, the phase of the reflected wave at the interface will differ from the transmitted waveform. The phase information of the received echo is completely preserved by the HF-SSBP, which transmits a coherent wave. The positive and negative reflection matrices are defined.
[0088] (10)
[0089] As one of the parameters for identifying sediment types.
[0090] The acoustic response at different frequencies will also vary (consider citing a reference). The HF-SSBP itself has a 30kHz bandwidth. If its broadband characteristics can be utilized to simultaneously acquire sediment information at multiple frequency points in a single echo, and to extract acoustic parameters at each frequency, the detection efficiency will be greatly improved. Therefore, a dispersion characteristic analysis method based on subband decomposition is considered, dividing the broadband signal into overlapping subband signals. These subband signals are short linear frequency-modulated signals with a certain bandwidth and duration, possessing coherence and a certain duration. For filtering, a dual filtering method of traditional filtering + subband matched filtering is considered to reduce mutual interference between subbands. The theoretical basis is...
[0091] (11)
[0092] For the sub-band received signal set, The number of sub-bands, For the defined subband bandwidth, and These are the center frequency and received amplitude of the i-th sub-band, respectively. Pre-filtering roughly separates the frequencies of each sub-band. Sub-band matched filtering is then performed on this basis; frequencies outside the specified sub-band range contribute little to sub-band matching and are thus greatly suppressed.
[0093] (12)
[0094] This is a collection of pulse compressions from each frequency band channel after filtering. Since the pulse compression result reflects the properties of the center frequency, the pulse compression result within each frequency band channel is the amplitude and phase characteristics at the center frequency of each sub-band.
[0095] Based on equation (12), sub-band-based reflection coefficient calibration is performed. The total internal reflection signal from the water surface is processed using the same sub-band method. The sediment echo and the total internal reflection echo from the water surface can then be calibrated using reflection coefficient technology on their respective sub-bands, thus obtaining a set of reflection coefficients. for
[0096] (13)
[0097] Similarly, the positive and negative reflection matrices of the sub-bands can be obtained. ,for
[0098] (14)
[0099] It is a three-dimensional matrix, with the two rows representing the positive and negative values of the reflections of the first and second layers, respectively. M is the number of Ping echoes, and the upper right subscript of z represents the sub-band number, which corresponds to the frequency of the sub-band.
[0100] Based on the extraction of the above six types of acoustic feature parameters, the following section describes the contribution-based multi-feature fusion ground acoustic model construction and classification method of this invention (e.g., Figure 2 (As shown). Sedimentary physical properties are primarily grain size-dependent, with density and porosity as secondary factors. A geoacoustic model is established to correlate the aforementioned six types of acoustic features with sediment grain size characteristics (supplemented by density and porosity). Since different acoustic features differ in their characterization of sediments, correlation significance, and quantification accuracy, and the inferences of geoacoustic relationships corresponding to each feature may be contradictory, this invention proposes a multi-feature fusion criterion based on contribution. This criterion comprehensively models feature parameters such as reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrices, sub-band reflection coefficients, and sub-band positive and negative reflection matrices. Its expression is:
[0101] (15)
[0102] in
[0103] (16)
[0104] Y represents the target output, which is the sediment grain size classification result (typically sand, silty sand, sandy silt, silty clay, mud). The parameterized model for the above six features, The particle size, density, and porosity corresponding to various types of sediments are as follows: For each feature, a single-parameterized model is provided. This represents the contribution coefficient of each model to the overall model. This represents a systematic bias.
[0105] The implementation process of this classification method is as follows: In a controlled laboratory environment, HF-SSBP acoustic echo data of a certain number of samples were collected, and the particle size, density and porosity of the sediment samples were measured and recorded simultaneously. Extract the above six types of acoustic parameters from the echo data; Parametric fitting is performed according to Equation (15), and regularized regression is used to obtain the contribution coefficient to suppress the risk of feature collinearity and overfitting. We selected a sedimentary area with good natural sorting and recorded actual ground acoustic echo data under field navigation mode. We also used sediment samples for grain size analysis and verified the independent characteristic parameters, the sediment grain size-dominated parameter model, and the fused ground acoustic model through measured data. Based on the actual field measurement results, the ground acoustic model equations (15) and (16) based on the laboratory standardized measurement were revised to improve the applicability of the model to the real subsurface environment.
[0106] Example 2
[0107] Regarding the extraction of the six types of acoustic features in this invention, refer to... Figure 1 Explain it. Reference for layer structure extraction Figure 1 a) Based on the two-dimensional image formed by pulse compression of HF-SSBP echo, combined with equations (1) to (5), edge detection is performed using the Canny algorithm and Sobel operator, and vertical local maxima are extracted to locate the in-phase axis, thereby obtaining the interface position and reflection amplitude of the first and second layers of the double-layer sediment, and completing the accurate identification of the layer structure. Regarding the extraction of the reflection coefficient, the total internal reflection signal at the water-air interface is used as a reference, and the first-layer reflection coefficient of the sediment is calculated using the reference interface ratio method in conjunction with equation (6); Figure 1 b), This illustrates the propagation path of sound waves in a two-layer sediment structure, where the total propagation path of the first layer is... The total path of secondary propagation is Combining the transmission-reflection law and attenuation characteristics of the double-layer medium, and combining equations (7) and (8), the reflection coefficients of the first and second layers are separated and calculated through the autoregressive formula, and the reflection coefficients of the first and second layers are obtained. Regarding the extraction of the first wave waveform Figure 1c) The first wave waveform of sedimentary echoes with different particle sizes is different. Combining equation (9), the slope of the wavefront horn is fitted by the least squares method, and the wavefront horn shape factor and waveform fluctuation factor are calculated to quantitatively describe the shape of the first wave envelope. Exam on the extraction of positive and negative reflection matrices Figure 1 d) Deeply mine the phase information of the coherent echo, combine with equation (10), construct positive and negative reflection matrices according to the phase reversal law of acoustic impedance interface reflection, and determine the interface type of sound wave propagating from low impedance to high impedance or from high impedance to low impedance. Examining subband reflection coefficients and subband positive and negative reflection matrices Figure 1 e) The broadband echo is divided into overlapping sub-bands. After pre-filtering and sub-band matched filtering, the total internal reflection signal of the water surface is used for calibration in each sub-band. Combined with equations (10) to (13), the sub-band level reflection coefficient is calculated, making full use of the broadband signal dispersion characteristics to improve the detection efficiency. Based on the sub-band decomposition and amplitude-phase extraction, phase discrimination and matrix construction are performed on the echoes of each sub-band. Combined with equation (14), a three-dimensional sub-band positive and negative reflection matrix containing sub-band number, layer number, and ping count is formed to realize the refined characterization of sediment interface impedance characteristics at multiple frequency points.
[0108] Regarding the contribution-based multi-feature fusion ground acoustic model construction and classification method in this invention, please refer to... Figure 2 Explain it.
[0109] Based on the bottom sediment echo signal obtained by HF-SSBP, the reflection coefficient of the first layer was first obtained primarily through direct measurement. and secondary reflection coefficient The system obtains information on the structure of the two sedimentary layers, and then, through feature extraction and dispersion characteristic analysis, six types of acoustic features are obtained: first wave waveform feature parameters, positive and negative reflection matrices, sub-band reflection coefficients, and sub-band positive and negative reflection matrices. These six types of features are then used to construct corresponding parameterized models, which are coupled with geoacoustic parameters that use sediment grain size as the main parameter and density and porosity as auxiliary parameters. A contribution coefficient-based multi-feature fusion criterion is used, and a regularized regression method is employed to solve for the contribution coefficients of each feature, thus constructing a multi-feature joint geoacoustic model. After the model is constructed, it is validated using field measurement data, and the parameters are corrected based on the validation results. Finally, a highly robust classification of underwater bottom sediment types is achieved.
[0110] In this invention, the sediment classification types refer to the internationally accepted Folk triangulation method. Based on the grain size of the sediments, they can be divided into sand, silty sand, sandy silt, silty clay, mud, etc. The number of sediment types can also be increased or decreased according to actual needs, but the basis is always the grain size of the sediments.
[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A two-layer structure ground acoustic detection and sediment classification method based on multi-sound feature fusion, characterized in that: Includes the following steps: S1. In a controlled laboratory environment, relying on the high-resolution layer structure identification capability of HF-SSBP, the traditional single-layer sediment geosonic model is refined into a two-layer structure model, the contribution of the secondary layer reflection signal to sediment classification is quantified, a certain amount of HF-SSBP acoustic echo data is collected, and the particle size, density and porosity of sediment samples are measured and recorded simultaneously. S2. Extract six types of acoustic characteristic parameters from the echo data: reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrix, sub-band reflection coefficient, and sub-band positive and negative reflection matrix. S3. The above 6 types of acoustic feature parameters are comprehensively modeled, parametrically fitted, and regularized regression is used to obtain the contribution coefficient to suppress the risk of feature collinearity and overfitting. S4. Select a sedimentary area with good natural sorting and record actual ground acoustic echo data under the field navigation mode. Simultaneously, use sediment samples for grain size analysis and verify the independent characteristic parameters and the parameter model dominated by sediment grain size and the fused ground acoustic model through measured data. S5. Based on the actual field measurement results, revise the ground acoustic model based on the laboratory standardized measurement to improve the model's applicability to the real subsurface environment.
2. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 1, characterized in that: The linear frequency modulated signal transmitted by the HF-SSBP is (1) in For carrier frequency, For bandwidth; the echo received by the double-layer sediment medium is (2) in Let represent the echo signal amplitude of the i-th layer, where i=0 represents the first layer and i=1 represents the second layer. The distance between the HF-SSBP emission surface and the i-th layer of sediment is given by pulse compression processing. (3) in After pulse compression processing, the amplitude of the echo signal of the i-th layer, as shown in equation (3), will appear at the interface layer. Pulse spikes, through continuous ping echoes, can form a two-dimensional shallow profile image. The Canny algorithm is used to extract the boundary layer structure, and its core Sobel operator is... (4) in These represent the magnitude and phase of the gradient intensity at each pixel. The layer structure of the shallow subsurface profile image is extracted using the same phase axis. The constraint is to find vertical local maxima. Assuming the short-term measurement layer structure remains unchanged and ignoring the influence of the x-direction, the shallow seismic profile layer structure and the reflection amplitudes of the first and second layers can be expressed as follows: (5) in These are the local extrema of the first and second layers of the shallow profile image found using the Canny algorithm.
3. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 2, characterized in that: The reflection coefficient of the first sediment layer was calculated using the reference interface ratio method, with the water-air interface as the reference. The amplitude of the echo received at the transducer was recorded during water-air reflection across the entire shallow probing range. Therefore, the reflectance of the first layer of sediment is (6) Under the condition that the amplitudes of both are at the same distance, the transmission and reflection patterns of the two sediment layers can be obtained. (7) in These are the reflection coefficients of the secondary layer, Attenuation occurs in water and sediment layers, respectively. The emitted wave at the first-layer interface is pre-calibrated based on the total internal reflection experiment at the water-air interface. According to equation (7), the autoregressive formula for the separation of the reflection coefficients of the first and second layers is as follows: (8) in Multiple sets of acoustic echo data at different distances need to be collected and then fitted and optimized using the above formula.
4. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 3, characterized in that: Regarding the extraction of the layer structure, it is based on the two-dimensional image formed by pulse compression of HF-SSBP echo. Combining equations (1) to (5), the Canny algorithm and Sobel operator are used for edge detection, and the vertical local maxima are extracted to locate the in-phase axis. The interface position and reflection amplitude of the first and second layers of the double sediments are obtained, and the accurate identification of the layer structure is completed. Regarding the extraction of the reflection coefficient, the total reflection signal of the water-air interface is used as a reference. Combining equation (6), the reference interface ratio method is used to calculate the reflection coefficient of the first layer of sediments. This illustrates the propagation path of sound waves in a two-layer sediment structure, where the total propagation path of the first layer is... The total path of secondary propagation is Combining the transmission-reflection law and attenuation characteristics of the double-layer medium, and combining equations (7) and (8), the reflection coefficients of the first and second layers are separated and calculated through the autoregressive formula, and the reflection coefficients of the first and second layers are obtained.
5. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 1, characterized in that: Quantitative description of the first wave waveform is achieved by using the wavefront funnel shape factor. and waveform fluctuation factor The description of the initial waveform shape is as follows: (9) Where k is the slope of the wavefront funnel. The resulting curve is the envelope of the first wave waveform. The parameters are fitted using the least squares method.
6. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 5, characterized in that: According to basic acoustic principles, sound waves reflect in different propagation media. At the reflection interface from a low-impedance region to a high-impedance region, the echo phase remains consistent with the emitted waveform. In the case of a reflection wave propagating from a high-impedance region to a low-impedance region, the phase of the reflected wave at the interface will differ from the transmitted waveform. By flipping the HF-SSBP, which transmits a coherent wave, the phase information of the received echo is completely preserved, defining the positive and negative reflection matrices. (10) As one of the parameters for identifying sediment types.
7. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 6, characterized in that: A dispersion characteristic analysis method based on subband decomposition is employed to divide the broadband signal into subband signals with overlap. These subband signals are short linear frequency-modulated signals with a certain bandwidth and duration, possessing coherence and a certain duration. A dual filtering method combining traditional filtering and subband matched filtering is used to reduce mutual interference between subbands. The theoretical basis is... (11) in For sub-band received signals, The number of sub-bands, For the defined subband bandwidth, and The center frequency and received amplitude of the i-th sub-band are respectively given. Pre-filtering separates the frequencies of each sub-band. Sub-band matched filtering is then performed on top of this, suppressing frequencies outside the defined sub-band range that contribute no to sub-band matching. (12) in This is a collection of pulse compressions from each frequency band channel after filtering. Since the pulse compression result reflects the properties of the center frequency, the pulse compression result within each frequency band channel is the amplitude and phase characteristics at the center frequency of each sub-band.
8. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 7, characterized in that: Based on equation (12), sub-band-based reflection coefficient calibration is performed. The total internal reflection signal from the water surface is processed using the same sub-band method. The sediment echo and the total internal reflection echo from the water surface can then be calibrated using reflection coefficient technology on their respective sub-bands, thus obtaining a set of reflection coefficients. for (13) Similarly, the positive and negative reflection matrices of the sub-bands can be obtained. ,for (14) in It is a three-dimensional matrix, with the two rows representing the positive and negative values of the reflections of the first and second layers, respectively. M is the number of Ping echoes, and the upper right subscript of z represents the sub-band number, which corresponds to the frequency of the sub-band.
9. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 8, characterized in that: Regarding the extraction of the first wave waveform, the first wave waveform differs for sedimentary echoes of different particle sizes. Combining equation (9), the least squares method is used to fit the wavefront flare slope, calculate the wavefront flare shape factor and waveform undulation factor, and quantitatively describe the first wave envelope morphology. Regarding the extraction of the positive and negative reflection matrices, the phase information of the coherent echoes is deeply mined. Combining equation (10), the positive and negative reflection matrices are constructed based on the acoustic impedance interface reflection phase reversal law to determine the interface type from low impedance to high impedance or from high impedance to low impedance. Regarding the extraction of subband reflection coefficients and subband positive and negative reflection matrices... The broadband echo is divided into overlapping sub-bands. After pre-filtering and sub-band matching filtering, the water surface total reflection signal is used for calibration in each sub-band. Combined with Equations (10) to (13), the sub-band level reflection coefficient is calculated. The broadband signal dispersion characteristics are fully utilized to improve the detection efficiency. Based on the sub-band decomposition and amplitude and phase extraction, the phase discrimination and matrix construction of each sub-band echo are performed. Combined with Equation (14), a three-dimensional sub-band positive and negative reflection matrix containing sub-band number, layer number and Ping number is formed to realize the refined characterization of sediment interface impedance characteristics under multiple frequency points.
10. The method for ground acoustic detection and sediment classification based on multi-sound feature fusion in a two-layer structure according to claim 6, characterized in that: Based on the multi-feature fusion criterion of contribution, a comprehensive model is performed on feature parameters such as reflection coefficient, layer structure, first wave waveform, positive and negative reflection matrices, sub-band reflection coefficient, and sub-band positive and negative reflection matrices. The expression is as follows: (15) in (16) Where Y is the target output, i.e., the sediment grain size classification result. The parameterized model for the above six features, The particle size, density, and porosity corresponding to various types of sediments are as follows: For each feature, a single-parameterized model is provided. This represents the contribution coefficient of each model to the overall model. This represents a systematic bias.