A multi-channel bathymetric waveform decomposition method and system combined with morphological priors

Abstract

The present invention discloses a multi-channel depth sounding waveform decomposition method and system combined with morphological priors, the method comprising: collecting multiple groups of waveform data of shallow water channels and deep water channels; calculating the eigenvalues ​​of each group of waveform data and constructing an eigenvector matrix; standardizing the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data; considering a Gaussian kernel function, substituting the standardized data into a support vector machine algorithm to train a classification model, and selecting a pulse broadening coefficient z according to the classification model; based on the pulse broadening coefficient z, using a Gaussian kernel function to fit the water surface and water bottom; fitting the water surface and water bottom according to the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth. The advantages of the present invention are: improving the efficiency, accuracy and stability of water depth measurement.

Description

Technical Field

[0001] The present invention relates to the field of signal processing, and in particular to a multi-channel bathymetric waveform decomposition method and system combined with morphological priors. Background Art

[0002] LiDAR (Light Detection and Ranging) scanning technology is a high-tech technology that uses laser active detection to collect the position information of the target and stores the acquired data as three-dimensional coordinates, providing a high-precision data source for various subsequent applications. With the development of laser scanning technology, LiDAR has been widely used. For example, the document "Wang Ke. Research on separable nonlinear least squares calculation method and its application in LiDAR waveform decomposition [J]. Journal of Geodesy and Cartography, 2024, 53(02): 391." In the field of ocean exploration, airborne dual-frequency LiDAR is equipped with near-infrared and blue-green laser transmitting and receiving equipment, and combines different fields of view and different circuit gains to form a multi-channel depth measurement system. It integrates multiple high-tech technologies such as laser ranging, positioning and high-speed digital signal processing, and is an important means for rapid mapping of shallow waters. However, in many applications, the depth measurement method based solely on waveform decomposition cannot meet the requirements of high-precision water depth measurement with a large dynamic range. Therefore, it is often necessary to integrate waveform classification algorithms for dynamic parameter debugging and incorporate prior knowledge into the waveform decomposition algorithm.

[0003] When airborne lidar scans over land and sea targets, the full-waveform echo signal presents different forms due to the varying depths of the target and the seawater. For land targets, the radar receives a mixed wave from the ground and other obstacles within the scanning area. For ocean targets, the radar's final received echo consists of three components: a surface reflection echo, a backscattered echo from the water column, and an echo from the seafloor. For airborne bathymetric lidar systems, this echo variability means that the same depth extraction algorithm often cannot adapt to all echo conditions. Consequently, depth measurement efficiency is low, and accuracy and stability are difficult to guarantee. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that the water depth measurement efficiency of the existing airborne bathymetric lidar system is low and the accuracy and stability are difficult to guarantee.

[0005] The present invention solves the above technical problems through the following technical means: a multi-channel bathymetric waveform decomposition method combined with morphological priors, comprising the following steps:

[0006] Step 1: Collect multiple sets of waveform data of shallow water channels and deep water channels;

[0007] Step 2: Calculate the eigenvalue of each set of waveform data and construct the eigenvector matrix;

[0008] Step 3: normalize the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data;

[0009] Step 4: Considering the Gaussian kernel function, the standardized data is substituted into the support vector machine algorithm to train a classification model, and the pulse broadening coefficient z is selected according to the classification model. There are three classification models, which are used to classify land echoes and ocean echoes, nearshore shallow water echoes and other ocean echoes, and shallow water echoes and deep water echoes respectively;

[0010] Step 5: Based on the pulse broadening coefficient z, a Gaussian kernel function is used to fit the water surface and the bottom;

[0011] Step 6: Fit the water surface and bottom with the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth.

[0012] Furthermore, the step 1 includes:

[0013] The n data points of the m groups of waveform data of the shallow water channel and the deep water channel are combined into a matrix S 2m×n , extract a part of the data to form the training set S1, and the other part of the data to form the test set S2, where m is an even number, and the shallow water and deep water channels of the i-th group of data in S1 are represented by X1 respectively i1 , X1 i2 ,but The shallow and deep water channels of the i-th group of data in S2 are represented by X2 i1 , X2 i2 ,but

[0014] Furthermore, the step 2 includes:

[0015] The characteristic values ​​include the area under the curve, skewness, kurtosis, first echo width and echo time difference of shallow water channels, the area under the curve, skewness, kurtosis, first echo width and echo time difference of deep water channels, and the peak ratio of seabed and surface echoes of shallow water channels. The calculation formula of the area under the curve is:

[0016]

[0017] Where x(i) represents the i-th data point of the shallow water channel or deep water channel, and dt represents the sampling rate;

[0018] The calculation formula for skewness is: Where X represents the discrete distribution of the echo waveform of the shallow water channel or deep water channel, μ represents the mean of the distribution, σ represents the standard deviation of the echo waveform of the shallow water channel or deep water channel, and E represents the averaging operation.

[0019] The calculation formula for kurtosis is:

[0020] The calculation formula of echo time difference is: ΔT = T2-T1, where T2 represents the time when the water surface echo is received, and T1 represents the time when the bottom echo is received;

[0021] The 11 eigenvalues ​​of each scanning point of the m sets of waveform data are combined into an eigenvector matrix E m×11 , where the eigenvector of the i-th group of data is represented by V i , then E m×11 =[V1, V2, ..., V m ] T .

[0022] Furthermore, the step 4 includes:

[0023] A Gaussian kernel function representing the waveform morphology is selected and normalized using the Gaussian kernel function to obtain standardized data. This standardized data is then fed into the support vector machine algorithm to train the classification model. First, the land echo label is set to 0 and the ocean echo label is set to 1 to train the land-sea classification model, enabling the classification of land echoes from ocean echoes. Next, the nearshore shallow water echo label is set to 0 and the ocean echo label is set to 1 to train the nearshore shallow water classification model, enabling the classification of nearshore shallow water echoes from ocean echoes at other depths. Finally, the shallow water echo label is set to 0 and the deep water echo label is set to 1 to train the deep-shallow water classification model, enabling the classification of shallow water echoes from deep water echoes. The three classification models are saved separately, and different pulse stretching coefficients z are selected. The Gaussian kernel function corresponds to different waveform morphologies and is used to describe them. The standardized data obtained by normalizing the Gaussian kernel function serves as the training samples. Different classification results correspond to different pulse stretching coefficients z, so different pulse stretching coefficients z are selected based on the classification results.

[0024] Furthermore, the step five includes:

[0025] The Gaussian kernel function that represents the waveform is used to fit the water surface and bottom into

[0026]

[0027] Among them, g(t) represents the fitting result of the water surface and the bottom at time t, N represents the number of Gaussian components, and α j , β j , γ jRepresent the intensity, position, and pulse width at half maximum of the j-th Gaussian waveform component, respectively. Using a Gaussian kernel function to characterize the waveform morphology to fit the water surface and bottom primarily involves using the Gaussian kernel function to characterize the waveform morphology to obtain sample data, and then fitting the sample data using the above formula. βj represents the position of the j-th Gaussian waveform component, which is actually the time T2 at which the water surface echo is received and the time T1 at which the bottom echo is received.

[0028] Furthermore, the step six includes:

[0029] The Gaussian kernel function is used to fit the water surface and bottom to obtain the echo position of the Gaussian function. The water depth is calculated based on the echo position. The formula is as follows

[0030]

[0031] Where H represents the water depth, c represents the speed of light in water, and θ1 represents the incident angle of the laser into the water during echo detection.

[0032] The present invention also provides a multi-channel bathymetric waveform decomposition system combined with morphological priors, comprising:

[0033] A data acquisition module is used to collect multiple sets of waveform data from shallow water channels and deep water channels;

[0034] Feature extraction module, used to find the eigenvalues ​​of each set of waveform data and construct the eigenvector matrix;

[0035] The standardization module is used to standardize the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data;

[0036] A classification module is used to consider the Gaussian kernel function, substitute the standardized data into the support vector machine algorithm to train the classification model, and select the pulse broadening coefficient z according to the classification model. There are three classification models, which are used to classify land echoes and ocean echoes, nearshore shallow water echoes and other ocean echoes, and shallow water echoes and deep water echoes respectively;

[0037] Data fitting module, used to fit the water surface and bottom using Gaussian kernel function based on pulse broadening coefficient z;

[0038] The water depth acquisition module is used to fit the water surface and water bottom according to the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth.

[0039] Furthermore, the data acquisition module is also used for:

[0040] The n data points of the m groups of waveform data of the shallow water channel and the deep water channel are combined into a matrix S 2m×n, extract a part of the data to form the training set S1, and the other part of the data to form the test set S2, where m is an even number, and the shallow water and deep water channels of the i-th group of data in S1 are represented by X1 respectively i1 , X1 i2 ,but The shallow and deep water channels of the i-th group of data in S2 are represented by X2 i1 , X2 i2 ,but

[0041] Furthermore, the feature extraction module is also used to:

[0042] The characteristic values ​​include the area under the curve, skewness, kurtosis, first echo width and echo time difference of shallow water channels, the area under the curve, skewness, kurtosis, first echo width and echo time difference of deep water channels, and the peak ratio of seabed and surface echoes of shallow water channels. The calculation formula of the area under the curve is:

[0043]

[0044] Where x(i) represents the i-th data point of the shallow water channel or deep water channel, and dt represents the sampling rate;

[0045] The calculation formula for skewness is: Where X represents the discrete distribution of the echo waveform of the shallow water channel or deep water channel, μ represents the mean of the distribution, σ represents the standard deviation of the echo waveform of the shallow water channel or deep water channel, and E represents the averaging operation.

[0046] The calculation formula for kurtosis is:

[0047] The calculation formula of echo time difference is: ΔT = T2-T1, where T2 represents the time when the water surface echo is received, and T1 represents the time when the bottom echo is received;

[0048] The 11 eigenvalues ​​of each scanning point of the m sets of waveform data are combined into an eigenvector matrix E m×11 , where the eigenvector of the i-th group of data is represented by V i , then E m×11 =[V1, V2, ..., V m ] T .

[0049] Furthermore, the classification module is also used for:

[0050] A Gaussian kernel function is selected to characterize the waveform morphology. The Gaussian kernel function is normalized to obtain standardized data. The standardized data is substituted into the support vector machine algorithm to train the classification model. First, the land echo label is set to 0 and the ocean echo is set to 1 to train the land-sea classification model to realize the classification of land echo and ocean echo; then the nearshore shallow water echo label is set to 0 and the ocean echo of other depths is set to 1 to train the nearshore shallow water classification model to realize the classification of nearshore shallow water echo and ocean echo of other depths; finally, the shallow water echo label is set to 0 and the deep water echo is set to 1 to train the deep-shallow water classification model to realize the classification of shallow water echo and deep water echo. The three classification models are saved separately, and different pulse broadening coefficients z are selected.

[0051] Furthermore, the data fitting module is further used to:

[0052] The Gaussian kernel function that represents the waveform is used to fit the water surface and bottom into

[0053]

[0054] Among them, g(t) represents the fitting result of the water surface and the bottom at time t, N represents the number of Gaussian components, and α j , β j , γ j Respectively represent the intensity, position and pulse half-height width of the jth Gaussian waveform component. Among them, the Gaussian kernel function that characterizes the waveform shape is used to fit the water surface and the bottom. The main purpose is to use the Gaussian kernel function to characterize the waveform shape to obtain sample data, and then use the above formula to fit the sample data. Among them, β j It represents the position of the j-th Gaussian waveform component, which is actually the time when the water surface echo is received, T2, and the time when the water bottom echo is received, T1.

[0055] Furthermore, the water depth acquisition module is further used to:

[0056] The Gaussian kernel function is used to fit the water surface and bottom to obtain the echo position of the Gaussian function. The water depth is calculated based on the echo position. The formula is as follows

[0057]

[0058] Where H represents the water depth, c represents the speed of light in water, and θ1 represents the incident angle of the laser into the water during echo detection.

[0059] The advantages of the present invention are that: in the face of the continuous changes in echo shape when the airborne laser radar scans over the sea and land, traditional waveform decomposition algorithms generally lack adaptability to sudden changes in ocean echoes, resulting in an inability to adapt to complex and changeable sea and land environments and noise scenes. The present invention can extract representative feature vectors and implement waveform classification based on the support vector machine algorithm according to the characteristics of airborne laser radar echoes in different scenes, thereby better adapting to waveform changes caused by the continuous changes in the scene and its noise, achieving better depth solution robustness, and thus improving water depth measurement efficiency, accuracy and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flowchart of training a classification model in a multi-channel bathymetric waveform decomposition method combined with morphological priors disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with 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. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Example 1

[0063] like Figure 1 As shown, the present invention provides a multi-channel bathymetric waveform decomposition method combined with morphological priors, comprising the following steps:

[0064] S1. Collect multiple sets of waveform data from shallow and deep water channels. The specific process is as follows:

[0065] The n data points of the m groups of waveform data of the shallow water channel and the deep water channel are combined into a matrix S 2m×n , extract a part of the data to form the training set S1, and the other part of the data to form the test set S2, where m is an even number, and the shallow water and deep water channels of the i-th group of data in S1 are represented by X1 respectively i1 , X1 i2 ,but The shallow and deep water channels of the i-th group of data in S2 are represented by X2 i1 , X2 i2 ,but In this embodiment, m is 406, n is 8000, and the 8000 data points of the 406 sets of waveform data of the shallow water channel and the deep water channel form the matrix S 812×8000 , extract half of the data to form the training set S1, and the other half of the data to form the test set S2, S1 = [X1 11, X1 12 , X1 21 , X1 22 …, X1 203·1， X1 203·2 ] T , S2=[X2 11 , X2 12 , X2 21 , X2 22 …, X2 203·1， X2 203·2 ] T .

[0066] S2. Calculate the eigenvalues ​​of each set of waveform data and construct an eigenvector matrix. The specific process is as follows:

[0067] The characteristic values ​​include the area under the curve, skewness, kurtosis, first echo width W1 and echo time difference of shallow water channels, the area under the curve, skewness, kurtosis, first echo width W2 and echo time difference of deep water channels, and the peak ratio of seabed and surface echoes (A2 / A1) of shallow water channels. The calculation formula of the area under the curve is:

[0068]

[0069] Where x(i) represents the i-th data point of the shallow water channel or deep water channel, and dt represents the sampling rate;

[0070] The calculation formula for skewness is: Where X represents the discrete distribution of the echo waveform of the shallow water channel or deep water channel, μ represents the mean of the distribution, σ represents the standard deviation of the echo waveform of the shallow water channel or deep water channel, and E represents the averaging operation.

[0071] The calculation formula for kurtosis is:

[0072] The calculation formula of echo time difference is: ΔT = T2-T1, where T2 represents the time when the water surface echo is received, and T1 represents the time when the bottom echo is received;

[0073] The 11 eigenvalues ​​of each scanning point of the 203 sets of waveform data are combined into an eigenvector matrix E 203×11 , where the eigenvector of the i-th group of data is represented by V i , then E 203×11 =[V1, V2, ..., V 203 ] T .

[0074] It should be noted that the calculation formulas for the area under the curve, skewness, kurtosis, echo time difference and eigenvector matrix above are all general formulas. In practical applications, the above formulas are used to calculate the area under the curve, skewness, kurtosis, echo time difference and eigenvector matrix of shallow water channels, and the above formulas are used to calculate the area under the curve, skewness, kurtosis, echo time difference and eigenvector matrix of deep water channels.

[0075] S3. Normalize the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data. The specific process is as follows:

[0076] The 11 eigenvalues ​​of each scanning point of the above 203 sets of waveform data form an eigenvector matrix. Then each eigenvector matrix has 11 eigenvalues. Each eigenvalue is normalized. The general formula for normalization is:

[0077]

[0078] Among them, q′ is the standardized data, and q is the eigenvalue of each dimension.

[0079] S4. Considering the Gaussian kernel function, the standardized data is substituted into the support vector machine algorithm to train a classification model, and the pulse broadening coefficient z is selected according to the classification model. There are three classification models, which are used to classify land echoes and ocean echoes, nearshore shallow water echoes and other ocean echoes, and shallow water echoes and deep water echoes respectively. The specific process is as follows:

[0080] A Gaussian kernel function is selected to characterize the waveform morphology. Normalized data is obtained by applying the Gaussian kernel function to the data. This normalized data is then fed into the support vector machine algorithm to train the classification model. First, the land echo label is set to 0 and the ocean echo label is set to 1 to train the land-sea classification model, enabling classification between land and ocean echoes. Next, the nearshore shallow water echo label is set to 0 and the ocean echo label at other depths is set to 1 to train the nearshore shallow water classification model, enabling classification between nearshore shallow water echoes and ocean echoes at other depths. Finally, the shallow water echo label is set to 0 and the deep water echo label is set to 1 to train the deep-shallow water classification model, enabling classification between shallow and deep water echoes. The three classification models are saved separately, and different pulse stretching coefficients z are selected. The Gaussian kernel function corresponds to different waveform morphologies and is used to describe them. The normalized data obtained by applying the Gaussian kernel function serves as the training samples. Different classification results correspond to different pulse stretching coefficients z, so different pulse stretching coefficients z are selected based on the classification results.

[0081] S5. Based on the pulse broadening coefficient z, a Gaussian kernel function is used to fit the water surface and bottom. The specific process is as follows:

[0082] The Gaussian kernel function that represents the waveform is used to fit the water surface and bottom into

[0083]

[0084] Among them, g(t) represents the fitting result of the water surface and the bottom at time t, N represents the number of Gaussian components, and α j , β j , γ j Respectively represent the intensity, position and pulse half-height width of the jth Gaussian waveform component. Among them, the Gaussian kernel function that characterizes the waveform shape is used to fit the water surface and the bottom. The main purpose is to use the Gaussian kernel function to characterize the waveform shape to obtain sample data, and then use the above formula to fit the sample data. Among them, β j It represents the position of the j-th Gaussian waveform component, which is actually the time when the water surface echo is received, T2, and the time when the water bottom echo is received, T1.

[0085] S6. Fit the water surface and bottom with the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth. The formula is as follows

[0086]

[0087] Where H represents the water depth, c represents the speed of light in water, T2 represents the time when the water surface echo is received, T1 represents the time when the water bottom echo is received, and θ1 represents the incident angle of the laser into the water.

[0088] Through the above technical scheme, the present invention proposes a water depth solution architecture of full-waveform lidar, which obtains the eigenvalue matrix by analyzing the multi-channel depth sounding waveform, and uses the support vector machine model to classify the full waveform data of the multi-channel lidar, and then combines the classification results to perform waveform decomposition strategies of different scales, and finally obtains the relative position of the sea surface and the seabed for solving the water depth. Therefore, the present invention mainly includes two processes: waveform classification and waveform decomposition. The waveform classification process is shown in the attached figure. Figure 1As shown, the eigenvalues ​​of the full echo data of the airborne laser radar are first extracted, including multidimensional eigenvalues ​​representing waveform shape, echo time, similarity, and echo differences between channels. Then, a support vector machine algorithm is used to classify different radar echoes. The support vector machine algorithm is suitable for the classification of high-dimensional eigenvalues ​​and is one of the commonly used algorithms in the field of machine learning classification, and is widely used for waveform classification. The airborne laser radar transmits pulses that interact with land or ocean targets at different depths, generating echoes of different shapes. The support vector machine algorithm is used in the present invention to achieve offshore detection, realizing a four-class classification model for land, nearshore shallow water, shallow water, and deep water. Since the support vector machine is a binary classification algorithm, it is necessary to combine multiple classification models to implement multi-classification problems. At the same time, in order to improve the accuracy of the classification algorithm, significant eigenvalues ​​are required to distinguish different types of echo data. The full bathymetric waveform is then decomposed into sea surface, water body, and seabed, and the water depth is calculated.

[0089] Example 2

[0090] Based on Example 1, Example 2 of the present invention further provides a multi-channel bathymetric waveform decomposition system combined with morphological priors, including:

[0091] A data acquisition module is used to collect multiple sets of waveform data from shallow water channels and deep water channels;

[0092] Feature extraction module, used to find the eigenvalues ​​of each set of waveform data and construct the eigenvector matrix;

[0093] The standardization module is used to standardize the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data;

[0094] A classification module is used to consider the Gaussian kernel function, substitute the standardized data into the support vector machine algorithm to train the classification model, and select the pulse broadening coefficient z according to the classification model. There are three classification models, which are used to classify land echoes and ocean echoes, nearshore shallow water echoes and other ocean echoes, and shallow water echoes and deep water echoes respectively;

[0095] Data fitting module, used to fit the water surface and bottom using Gaussian kernel function based on pulse broadening coefficient z;

[0096] The water depth acquisition module is used to fit the water surface and water bottom according to the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth.

[0097] Specifically, the data acquisition module is further used to:

[0098] The n data points of the m groups of waveform data of the shallow water channel and the deep water channel are combined into a matrix S 2m×n, extract a part of the data to form the training set S1, and the other part of the data to form the test set S2, where m is an even number, and the shallow water and deep water channels of the i-th group of data in S1 are represented by X1 respectively i1 , X1 i2 ,but The shallow and deep water channels of the i-th group of data in S2 are represented by X2 i1 , X2 i2 ,but

[0099] Specifically, the feature extraction module is further used to:

[0100] The characteristic values ​​include the area under the curve, skewness, kurtosis, first echo width and echo time difference of shallow water channels, the area under the curve, skewness, kurtosis, first echo width and echo time difference of deep water channels, and the peak ratio of seabed and surface echoes of shallow water channels. The calculation formula of the area under the curve is:

[0101]

[0102] Where x(i) represents the i-th data point of the shallow water channel or deep water channel, and dt represents the sampling rate;

[0103] The calculation formula for skewness is: Where X represents the discrete distribution of the echo waveform of the shallow water channel or deep water channel, μ represents the mean of the distribution, σ represents the standard deviation of the echo waveform of the shallow water channel or deep water channel, and E represents the averaging operation.

[0104] The calculation formula for kurtosis is:

[0105] The calculation formula of echo time difference is: ΔT = T2-T1, where T2 represents the time when the water surface echo is received, and T1 represents the time when the bottom echo is received;

[0106] The 11 eigenvalues ​​of each scanning point of the m sets of waveform data are combined into an eigenvector matrix E m×11 , where the eigenvector of the i-th group of data is represented by V i , then E m×11 =[V1, V2, ..., V m ] T .

[0107] Specifically, the classification module is further used for:

[0108] A Gaussian kernel function representing the waveform morphology is selected and normalized using the Gaussian kernel function to obtain standardized data. This standardized data is then fed into the support vector machine algorithm to train the classification model. First, the land echo label is set to 0 and the ocean echo label is set to 1 to train the land-sea classification model, enabling the classification of land echoes from ocean echoes. Next, the nearshore shallow water echo label is set to 0 and the ocean echo label is set to 1 to train the nearshore shallow water classification model, enabling the classification of nearshore shallow water echoes from ocean echoes at other depths. Finally, the shallow water echo label is set to 0 and the deep water echo label is set to 1 to train the deep-shallow water classification model, enabling the classification of shallow water echoes from deep water echoes. The three classification models are saved separately, and different pulse stretching coefficients z are selected. The Gaussian kernel function corresponds to different waveform morphologies and is used to describe them. The standardized data obtained by normalizing the Gaussian kernel function serves as the training samples. Different classification results correspond to different pulse stretching coefficients z, so different pulse stretching coefficients z are selected based on the classification results.

[0109] More specifically, the data fitting module is further used to:

[0110] The Gaussian kernel function that represents the waveform is used to fit the water surface and bottom into

[0111]

[0112] Among them, g(t) represents the fitting result of the water surface and the bottom at time t, N represents the number of Gaussian components, and α j , β j , γ j Respectively represent the intensity, position and pulse half-height width of the jth Gaussian waveform component. Among them, the Gaussian kernel function that characterizes the waveform shape is used to fit the water surface and the bottom. The main purpose is to use the Gaussian kernel function to characterize the waveform shape to obtain sample data, and then use the above formula to fit the sample data. Among them, β j It represents the position of the j-th Gaussian waveform component, which is actually the time when the water surface echo is received, T2, and the time when the water bottom echo is received, T1.

[0113] Specifically, the water depth acquisition module is further used to:

[0114] The Gaussian kernel function is used to fit the water surface and bottom to obtain the echo position of the Gaussian function. The water depth is calculated based on the echo position. The formula is as follows

[0115]

[0116] Where H represents the water depth, c represents the speed of light in water, and θ1 represents the incident angle of the laser into the water during echo detection.

[0117] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A multi-channel bathymetric waveform decomposition method combined with morphological priors, characterized in that: The following steps are involved: Step 1: Collect multiple sets of waveform data of shallow water channels and deep water channels; Step 2: Calculate the eigenvalue of each set of waveform data and construct the eigenvector matrix; Step 3: normalize the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data; Step 4: Considering the Gaussian kernel function, the standardized data is substituted into the support vector machine algorithm to train a classification model, and the pulse broadening coefficient z is selected according to the classification model. There are three classification models, which are used to classify land echoes and ocean echoes, nearshore shallow water echoes and other ocean echoes, and shallow water echoes and deep water echoes respectively; Step 5: Based on the pulse broadening coefficient z, the Gaussian kernel function is used to fit the water surface and the bottom of the water; the Gaussian kernel function that represents the waveform is used to fit the water surface and the bottom of the water into in, represents the fitting results of the water surface and the bottom at time t, represents the number of Gaussian components, 、 、 Respectively represent The intensity, position and half-height width of the Gaussian waveform component, The position of the Gaussian waveform components Actually, it is the time when the water meter echo is received and the bottom echo reception time ; Step 6: Fit the water surface and bottom with the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth.

2. The multi-channel bathymetric waveform decomposition method combined with morphological priors according to claim 1 is characterized in that: The step one comprises: The n data points of the m groups of waveform data of the shallow water channel and the deep water channel are combined into a matrix , extract a part of the data to form a training set , and the other part of the data constitutes the test set , where m is an even number, The shallow water and deep water channels of the i-th group of data are expressed as , ,but , The shallow water and deep water channels of the i-th group of data are expressed as , ,but .

3. The multi-channel bathymetric waveform decomposition method combined with morphological priors according to claim 1 is characterized in that: The second step includes: The characteristic values ​​include the area under the curve, skewness, kurtosis, first echo width and echo time difference of shallow water channels, the area under the curve, skewness, kurtosis, first echo width and echo time difference of deep water channels, and the peak ratio of seabed and surface echoes of shallow water channels. The calculation formula of the area under the curve is: in, represents the i-th data point of the shallow water channel or deep water channel, Indicates the sampling rate; The calculation formula for skewness is: , where X represents the discrete distribution of the echo waveform of the shallow water channel or deep water channel, represents the mean of the distribution, It represents the standard deviation of the echo waveform of the shallow water channel or deep water channel, and E represents the averaging operation; The calculation formula for kurtosis is: ; The calculation formula for echo time difference is: ,in, Indicates the time when the water meter echo is received. Indicates the moment of receiving the bottom echo; The 11 eigenvalues ​​of each scanning point of the m sets of waveform data are combined into an eigenvector matrix , where the eigenvector of the i-th group of data is expressed as ,but .

4. The multi-channel bathymetric waveform decomposition method combined with morphological priors according to claim 1 is characterized in that: The fourth step includes: A Gaussian kernel function is selected to characterize the waveform morphology. The Gaussian kernel function is normalized to obtain standardized data. The standardized data is substituted into the support vector machine algorithm to train the classification model. First, the land echo label is set to 0 and the ocean echo is set to 1 to train the land-sea classification model to realize the classification of land echo and ocean echo; then the nearshore shallow water echo label is set to 0 and the ocean echo of other depths is set to 1 to train the nearshore shallow water classification model to realize the classification of nearshore shallow water echo and ocean echo of other depths; finally, the shallow water echo label is set to 0 and the deep water echo is set to 1 to train the deep-shallow water classification model to realize the classification of shallow water echo and deep water echo. The three classification models are saved separately, and different pulse broadening coefficients z are selected.

5. The multi-channel bathymetric waveform decomposition method combined with morphological priors according to claim 1 is characterized in that: The step six comprises: The Gaussian kernel function is used to fit the water surface and bottom to obtain the echo position of the Gaussian function and calculate the water depth. The formula is as follows in, Indicates water depth. represents the speed of light in water, Indicates the incident angle of the laser into the water during echo detection.

6. A multi-channel bathymetric waveform decomposition system combined with morphological priors, characterized in that: include: A data acquisition module is used to collect multiple sets of waveform data from shallow water channels and deep water channels; Feature extraction module, used to find the eigenvalues ​​of each set of waveform data and construct the eigenvector matrix; The standardization module is used to standardize the eigenvalues ​​of each dimension in the eigenvector matrix to obtain standardized data; A classification module is used to consider the Gaussian kernel function, substitute the standardized data into the support vector machine algorithm to train the classification model, and select the pulse broadening coefficient z according to the classification model. There are three classification models, which are used to classify land echoes and ocean echoes, nearshore shallow water echoes and other ocean echoes, and shallow water echoes and deep water echoes respectively; The data fitting module is used to fit the water surface and the bottom of the water using the Gaussian kernel function based on the pulse broadening coefficient z; the Gaussian kernel function that characterizes the waveform is used to fit the water surface and the bottom of the water into in, represents the fitting results of the water surface and the bottom at time t, represents the number of Gaussian components, 、 、 Respectively represent The intensity, position and half-height width of the Gaussian waveform component, The position of the Gaussian waveform components Actually, it is the time when the water meter echo is received and the bottom echo reception time ; The water depth acquisition module is used to fit the water surface and bottom according to the Gaussian kernel function to obtain the echo position of the Gaussian function and calculate the water depth.

7. The multi-channel bathymetric waveform decomposition system combined with morphological priors according to claim 6, characterized in that: The data acquisition module is also used for: The n data points of the m groups of waveform data of the shallow water channel and the deep water channel are combined into a matrix , extract a part of the data to form a training set , and the other part of the data constitutes the test set , where m is an even number, The shallow water and deep water channels of the i-th group of data are expressed as , ,but , The shallow water and deep water channels of the i-th group of data are expressed as , ,but .

8. The multi-channel bathymetric waveform decomposition system combined with morphological priors according to claim 6, characterized in that: The feature extraction module is also used to: The characteristic values ​​include the area under the curve, skewness, kurtosis, first echo width and echo time difference of shallow water channels, the area under the curve, skewness, kurtosis, first echo width and echo time difference of deep water channels, and the peak ratio of seabed and surface echoes of shallow water channels. The calculation formula of the area under the curve is: in, represents the i-th data point of the shallow water channel or deep water channel, Indicates the sampling rate; The calculation formula for skewness is: , where X represents the discrete distribution of the echo waveform of the shallow water channel or deep water channel, represents the mean of the distribution, It represents the standard deviation of the echo waveform of the shallow water channel or deep water channel, and E represents the averaging operation; The calculation formula for kurtosis is: ; The calculation formula for echo time difference is: ,in, Indicates the time when the water meter echo is received. Indicates the moment of receiving the bottom echo; The 11 eigenvalues ​​of each scanning point of the m sets of waveform data are combined into an eigenvector matrix , where the eigenvector of the i-th group of data is expressed as ,but .

9. The multi-channel bathymetric waveform decomposition system combined with morphological priors according to claim 6, characterized in that: The classification module is also used to: A Gaussian kernel function is selected to characterize the waveform morphology. The Gaussian kernel function is normalized to obtain standardized data. The standardized data is substituted into the support vector machine algorithm to train the classification model. First, the land echo label is set to 0 and the ocean echo is set to 1 to train the land-sea classification model to realize the classification of land echo and ocean echo; then the nearshore shallow water echo label is set to 0 and the ocean echo of other depths is set to 1 to train the nearshore shallow water classification model to realize the classification of nearshore shallow water echo and ocean echo of other depths; finally, the shallow water echo label is set to 0 and the deep water echo is set to 1 to train the deep-shallow water classification model to realize the classification of shallow water echo and deep water echo. The three classification models are saved separately, and different pulse broadening coefficients z are selected.

Images