Multi-beam sonar seabed terrain reconstruction method, system, device, product and medium

Abstract

The present application relates to the field of seabed terrain reconstruction, and provides a multi-beam sonar seabed terrain reconstruction method, system, device, product and medium, comprising obtaining an approximation coefficient and an approximation coefficient variance, thereby obtaining a state estimation point, obtaining an initial filtering data point according to the state estimation point, and performing adaptive filtering on the filtering data point to obtain a filtering data point; calculating attention weights and spatial similarity parameters of multi-source data, and then calculating spatial similarity; calculating a message vector according to the spatial similarity, obtaining a final node representation based on the message vector, calculating a data fidelity term, a smoothing constraint term and a terrain regularization term, and obtaining fused terrain data; inputting the fused terrain data into an encoder, decoding through a decoder, performing dense skip connection, and obtaining decoding output; calculating a terrain reconstruction total loss, adjusting model parameters according to the terrain reconstruction total loss, outputting reconstructed terrain data, and completing the reconstruction of the seabed terrain.

Description

Technical Field

[0001] This invention relates to the field of seabed topography reconstruction technology, and in particular to multibeam sonar seabed topography reconstruction methods, systems, equipment, products and media. Background Technology

[0002] Seafloor topography mapping is a crucial foundation for marine scientific research. Multibeam bathymetry (MBB) systems, as the primary technology for acquiring high-precision, full-coverage seafloor topography data, directly determine the accuracy and reliability of the final data. However, in actual measurements, the raw data often contains various types of noise, outliers, and inconsistencies due to factors such as complex underwater acoustic environments, inherent sensor noise, variations in sound velocity profiles, vessel attitude disturbances, and insufficient fusion of multi-source data. This results in reconstructed topography exhibiting problems such as low resolution, blurred details, artifacts, and geometric distortion.

[0003] Currently, the processing of multibeam sonar data typically employs a step-by-step process, including noise filtering, data fusion, and grid reconstruction. Regarding noise suppression, traditional methods such as mean filtering and median filtering, while simple and easy to implement, often excessively smooth out realistic terrain details while denoising. Advanced methods based on wavelet transform or Kalman filtering, when dealing with non-stationary, multi-scale ocean noise, typically have fixed parameters, making it difficult to adapt to complex and changing data characteristics, and they also have limitations in distinguishing weak terrain signals from noise. In terms of data fusion, effectively utilizing multi-source heterogeneous data such as multibeam bathymetry point clouds, side-scan sonar images, and sound velocity profiles is a key challenge. Existing methods are mostly based on simple interpolation or weighted averaging, failing to deeply explore the spatial correlations and semantic complementarities between different data sources, resulting in limited accuracy of the fusion results and difficulty in recovering subtle features of complex terrain. In the terrain reconstruction process, conventional interpolation algorithms or methods based on a single neural network often cannot simultaneously guarantee the large-scale terrain fidelity and local detail richness of the reconstruction results, and are prone to over-smoothing or artificial artifacts, especially in areas with sparse data or severe noise. Summary of the Invention

[0004] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a multibeam sonar seabed topography reconstruction method, system, equipment, product, and medium, which achieves high-precision and robust reconstruction of seabed topography, and enhances the ability to handle noise and outliers during the reconstruction process. While smoothing the seabed topography, it also preserves details and avoids excessive smoothing.

[0005] This invention provides a method for reconstructing seabed topography using multibeam sonar, comprising:

[0006] S1: Acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points.

[0007] S2: Obtain multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity using the spatial similarity parameters and attention weights;

[0008] S3: Calculate message vectors based on spatial similarity, obtain final node representations based on message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and final node representations, and obtain fused terrain data based on data fidelity terms, smoothing constraint terms, and terrain regularization terms;

[0009] S4: Construct an encoder and decoder. Input the fused terrain data into the encoder, and decode it through the decoder. Perform dense skip connections to obtain the decoded output.

[0010] S5: Acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0011] According to the multibeam sonar seabed topography reconstruction method provided by the present invention, step S1 further includes:

[0012] S11: Obtain the initial sonar data, construct a high-pass filter and a low-pass filter, and perform discrete wavelet transform on the elevation of the initial sonar data through the high-pass filter and the low-pass filter to obtain the approximation coefficients and the variance of the approximation coefficients.

[0013] S12: Calculate the wavelet detail coefficient weighting parameter using the approximation coefficient variance and the approximation coefficient, and obtain the state estimation point based on the wavelet detail coefficient weighting parameter;

[0014] S13: Calculate the Kalman gain, and obtain the initial filtered data points using the Kalman gain and the state estimation points;

[0015] S14: Calculate the local standard deviation and median absolute deviation of the initial filtered data points, and use the local standard deviation and median absolute deviation to remove outliers and obtain the filtered data points.

[0016] According to the multibeam sonar seabed topography reconstruction method provided by the present invention, step S2 further includes:

[0017] S21: Obtain the multi-source data and construct a multi-source data graph model of the multi-source data;

[0018] S22: Calculate the spatial similarity parameter through the multi-source data graph model, use a multi-head attention mechanism and obtain the attention weight through the multi-source data graph model, and calculate the spatial similarity through the spatial similarity parameter and the attention weight.

[0019] According to the multibeam sonar seabed topography reconstruction method provided by the present invention, step S3 further includes:

[0020] S31: Calculate the message vector based on the spatial similarity, calculate the intermediate node representation through the message vector, and iterate the intermediate node representation layer by layer to obtain the final node representation;

[0021] S32: Calculate the data fidelity term using the filtered data points and the final node representation, calculate the smoothing constraint term and terrain regularization term represented by the final node, and perform a weighted summation of the data fidelity term, smoothing constraint term and terrain regularization term to obtain the fused terrain data.

[0022] According to the multibeam sonar seabed topography reconstruction method provided by the present invention, in step S4, the output of the dense skip connections is convolved and activated to obtain the decoded output.

[0023] According to the multibeam sonar seabed topography reconstruction method provided by the present invention, step S5 further includes:

[0024] S51: Obtain the real terrain data, and interpolate the real terrain data to obtain interpolated terrain data;

[0025] S52: Construct a discriminator, score the interpolated terrain data using the discriminator to obtain an interpolated terrain scalar score, score the decoded output using the discriminator to obtain a decoded output scalar score, score the real terrain data using the discriminator to obtain a real terrain score, and obtain an adversarial loss based on the interpolated terrain scalar score, the real terrain score, and the decoded output scalar score.

[0026] S53: Calculate the L1 norm and L2 norm between the real terrain data and the decoded output to obtain the terrain reconstruction loss, and calculate the perceptual loss and structural similarity loss between the real terrain data and the decoded output;

[0027] S54: The total terrain reconstruction loss is obtained through the adversarial loss, terrain reconstruction loss, perception loss and structural similarity loss. The model parameters are adjusted according to the total terrain reconstruction loss to obtain the target model. The reconstructed terrain data is output through the target model to complete the reconstruction of the seabed terrain.

[0028] This invention also provides a multibeam sonar seabed topography reconstruction system, comprising:

[0029] Filtered data point module: used to acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points;

[0030] Spatial similarity module: used to acquire multi-source data, calculate attention weights and spatial similarity parameters of multi-source data, and calculate spatial similarity through spatial similarity parameters and attention weights;

[0031] The terrain data fusion module is used to calculate message vectors based on spatial similarity, obtain the final node representation based on the message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and the final node representations, and obtain fused terrain data based on the data fidelity terms, smoothing constraint terms, and terrain regularization terms.

[0032] Encoding / decoding module: Used to build encoders and decoders. It takes fused terrain data as input to the encoder, decodes it through the decoder, performs dense skip connections, and obtains the decoded output.

[0033] Terrain Reconstruction Module: This module is used to acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and the decoded output, adjust the model parameters based on the total terrain reconstruction loss to obtain the target model, and output the reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0034] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multibeam sonar seabed topography reconstruction method as described above.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multibeam sonar seabed topography reconstruction method as described above.

[0036] The present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the computer is able to execute any of the multibeam sonar seabed topography reconstruction methods described above.

[0037] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0038] The multibeam sonar seabed topography reconstruction method, system, equipment, product, and medium provided by this invention can effectively improve the accuracy and robustness of seabed topography reconstruction through adaptive hybrid filtering and topography reconstruction. It can also enhance the ability to handle noise and outliers, thereby avoiding the influence of noise and outliers on the reconstruction results. While smoothing the seabed topography, it also takes into account the preservation of details, avoiding excessive smoothing of details, and can effectively combat data loss.

[0039] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the multibeam sonar seabed topography reconstruction method provided by the present invention.

[0042] Figure 2 This is a map showing the seabed topography measurement results of the multibeam sonar seabed topography reconstruction method provided by the present invention.

[0043] Figure 3 This is a schematic diagram of the structure of the multibeam sonar seabed topography reconstruction system provided by the present invention.

[0044] Figure 4 This is a schematic diagram of the structure of the multibeam sonar seabed topography reconstruction device provided by the present invention.

[0045] Figure label:

[0046] 100. Filtered data point module; 200. Spatial similarity module; 300. Fusion terrain data module; 400. Encoding and decoding module; 500. Terrain reconstruction module; 810. Processor; 820. Communication interface; 830. Memory; 840. Communication bus. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0048] In the description of the embodiments of the present invention, it should be noted that the terms "first", "second" and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0049] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention based on the specific circumstances.

[0050] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0051] The following is combined with Figures 1 to 4 Specific embodiments of the present invention are described below. Figure 1 This is a flowchart illustrating the multibeam sonar seabed topography reconstruction method provided by the present invention, including:

[0052] S1: Acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points.

[0053] Furthermore, the objective of this stage is to perform discrete wavelet transform to obtain approximation coefficients and their variance, thereby enabling improved Kalman filtering and adaptive filtering to obtain filtered data points. Specifically, step S1 further includes:

[0054] S11: Obtain the initial sonar data, construct a high-pass filter and a low-pass filter, and perform discrete wavelet transform on the elevation of the initial sonar data through the high-pass filter and the low-pass filter to obtain the approximation coefficients and the variance of the approximation coefficients.

[0055] S12: Calculate the wavelet detail coefficient weighting parameter using the approximation coefficient variance and the approximation coefficient, and obtain the state estimation point based on the wavelet detail coefficient weighting parameter;

[0056] S13: Calculate the Kalman gain, and obtain the initial filtered data points using the Kalman gain and the state estimation points;

[0057] S14: Calculate the local standard deviation and median absolute deviation of the initial filtered data points, and use the local standard deviation and median absolute deviation to remove outliers and obtain the filtered data points.

[0058] The specific implementation method for the above steps in this embodiment is as follows:

[0059] First, underwater topography needs to be detected using multibeam sonar to obtain initial sonar data. Then, due to interference from the underwater acoustic environment, the initial sonar data contains a large amount of nonlinear noise and outliers. If this noise is not removed, subsequent fusion and reconstruction will produce severe artifacts. Therefore, discrete wavelet transform needs to be performed first to obtain approximation coefficients and their variance. To this end, a series of high-pass and low-pass filters need to be constructed according to a predetermined method. Specifically, the q-th high-pass filter g[q] and the q-th low-pass filter h[q] constructed according to the predetermined filter construction method need to satisfy the orthogonality constraint condition:

[0060]

[0061]

[0062]

[0063] in, The first is obtained by performing discrete convolution through a filter. Unit impulse function, The first filter constructed according to the predetermined filter construction method A high-pass filter, The first filter constructed according to the predetermined filter construction method A low-pass filter.

[0064] This allows wavelet decomposition using low-pass and high-pass filters, thus obtaining the approximate coefficients of the j-th layer of the initial sonar data at the i-th data point. The variance of the approximation coefficients of each data point in the j-th level wavelet decomposition is taken as the variance of the approximation coefficients of that level. The approximation coefficient can reflect the topographic trend of the seabed. Furthermore, it can also obtain the detail coefficients of the j-th layer in the i-th data point of the initial sonar data. The detail coefficients can reflect the details of the seabed topography and noise. After wavelet decomposition, the elevation data of the i-th data point in the initial sonar data... It can be represented as:

[0065]

[0066] Where J is the total number of wavelet decomposition layers. For the time stamp of the i-th data point, Let i be the residual error of the i-th data point. Let j be the low-frequency scaling function of the j-th layer under the time stamp of the i-th data point. Let be the high-frequency scaling function of the j-th layer under the time stamp of the i-th data point.

[0067] Next, calculate the wavelet detail coefficient weighting parameters for the i-th data point. :

[0068]

[0069] Wherein, the wavelet coefficients of the j-th layer are adapted to the weights. The calculation method is as follows:

[0070]

[0071] Traditional Kalman filtering achieves optimal estimation performance in linear Gaussian systems, but seabed topography data exhibits nonlinearity and non-Gaussian properties. Therefore, an improved Kalman filtering algorithm is used here to obtain the estimated state point at time k predicted from the posterior state at time k-1. :

[0072]

[0073] in, The state at time k-1 is obtained based on the data at time k-1, which is the elevation coordinate of the seabed topography data. is the pre-determined wavelet detail coefficient weighting coefficient, and f() is the state equation for the Kalman filter.

[0074] Subsequently, the Kalman gain was obtained. :

[0075]

[0076] in, Let be the prior error covariance matrix. The Jacobian matrix of the observation equation is obtained by taking the first-order partial derivative of the observation function h() of the Kalman filter at the state estimation point, where T denotes the transpose. To observe the noise covariance matrix, To obtain the adaptive adjustment term, we first need to form a diagonal matrix from the variances of the approximation coefficients of each layer, and then weight the diagonal matrix.

[0077] This allows us to obtain the initial filtered data points for predicting time k based on the posterior state at time k. :

[0078]

[0079] in, Let be the observation vector of the elevation of the data point at time k.

[0080] Next, the other data points surrounding the i-th data point in the initial filtered data points can be considered as its neighborhood, and the standard deviation of the elevation of all data points in the neighborhood can be calculated to obtain the local standard deviation of the i-th data point. Similarly, the median absolute deviation of the elevation of all data points in the neighborhood is calculated to obtain the median absolute deviation of the i-th data point. Finally, outlier removal is achieved using local standard deviation and median absolute deviation, which involves first calculating the first adaptive coefficient for the i-th data point. :

[0081]

[0082] in, Let exp() be the standard deviation of the elevation of all detected data points, exp() be the exponential function, and exp() be the second adaptive coefficient for the i-th data point. The calculation method is as follows:

[0083]

[0084] This allows us to calculate the adaptive threshold for the i-th data point. :

[0085]

[0086] Outliers are identified and removed based on an adaptive threshold, meaning data points that meet the following criteria are considered outliers:

[0087]

[0088] in, This is the sensitivity coefficient. Let be the average elevation of all data points in the neighborhood of the i-th data point. This gives us the filtered data points.

[0089] S2: Obtain multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity using the spatial similarity parameters and attention weights;

[0090] Furthermore, the objective of this stage is to calculate the attention weights and spatial similarity parameters of the multi-source data, thereby calculating spatial similarity. Specifically, step S2 further includes:

[0091] S21: Obtain the multi-source data and construct a multi-source data graph model of the multi-source data;

[0092] S22: Calculate the spatial similarity parameter through the multi-source data graph model, use a multi-head attention mechanism and obtain the attention weight through the multi-source data graph model, and calculate the spatial similarity through the spatial similarity parameter and the attention weight.

[0093] The specific implementation method for the above steps in this embodiment is as follows:

[0094] First, multiple beamside scan sonars are used to detect the seabed topography, thereby obtaining multi-source data. Then, a deep attention graph neural network is used to process the multi-source data to obtain a multi-source data graph model including multiple data nodes.

[0095] Subsequently, the spatial similarity parameters between the m-th node and the n-th node were calculated using a multi-source data graph model. :

[0096]

[0097] in, To find the norm, For spatial scale parameters, Let be the coordinates of the m-th node. Let be the coordinates of the nth node.

[0098] Subsequently, a multi-head attention mechanism is employed to process the m-th node and the n-th node, obtaining their respective query vectors and key vectors, thereby determining the attention weight between the m-th node and the n-th node. :

[0099]

[0100] Where softmax[ ] represents performing a softmax transformation. Let m be the query vector of the m-th node at the h-th head. Let H be the key vector of the nth node in the hth head, and H be the total number of heads in the multi-head attention mechanism. Let h be the dimension of the h-th head. This allows us to calculate the spatial similarity between the m-th and n-th nodes. :

[0101]

[0102] Here, the nth node is located in the neighborhood of the mth node.

[0103] S3: Calculate message vectors based on spatial similarity, obtain final node representations based on message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and final node representations, and obtain fused terrain data based on data fidelity terms, smoothing constraint terms, and terrain regularization terms;

[0104] Furthermore, the objective of this stage is to calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms to obtain fused terrain data.

[0105] The specific implementation method for the above steps in this embodiment is as follows:

[0106] First, it is necessary to calculate the first step of the graph neural network based on spatial similarity. The message vector of the m-th node in the layer :

[0107]

[0108] in, For graph neural networks Layer message functions, For the nth node in the neighborhood, in the graph neural network... The message function is obtained by weighting the feature representations of nodes within a layer using a weight matrix and then biasing them using a bias vector. By iterating through the message vector layer by layer, the message vector of the last layer can be used as the final node representation of the m-th node. .

[0109] Subsequently, the data fidelity item is calculated by representing filtered data points and final nodes. :

[0110]

[0111] Where M is the total number of nodes. Let be the node weight of the m-th node. Let m be the m-th final fused terrain data point obtained by optimizing the final node representation. This refers to the filtered data point corresponding to the m-th final fused terrain data point. The node weights are calculated as follows:

[0112]

[0113] in, Let m be the elevation of the m-th final fused terrain data point.

[0114] We also need to calculate the smoothing constraint terms represented by the final nodes. and topographic regularization terms :

[0115]

[0116]

[0117] in, The attention weights between the m-th and n-th nodes learned by the multi-head attention mechanism. For the nth final fused terrain data point in the neighborhood of the mth final fused terrain data point, Let z be the elevation coordinate of the m-th final fused terrain data point, x be the x-coordinate of the m-th final fused terrain data point, and y be the y-coordinate of the m-th final fused terrain data point. Finally, a weighted sum of the data fidelity term, smoothing constraint term, and terrain regularization term is performed to obtain the fused terrain data. For the fused terrain data, it can also be substituted into the multi-source data from the beginning of this step, thereby applying gradient descent iteratively to the fused terrain data to improve its accuracy.

[0118] S4: Construct an encoder and decoder. Input the fused terrain data into the encoder, and decode it through the decoder. Perform dense skip connections to obtain the decoded output.

[0119] Furthermore, the purpose of step S4 is to input the fused terrain data into the encoder and perform dense connection skipping to obtain the decoded output. Specifically, in step S4, the output of the dense skipping connections is convolved and activated to obtain the decoded output.

[0120] The specific implementation method for the above steps in this embodiment is as follows:

[0121] First, we need to construct an encoder consisting of multiple downsampling blocks and a decoder consisting of the same number of upsampling blocks. The r-th downsampling block... and the r-th upsampling block The structure is as follows:

[0122]

[0123]

[0124] Where u is the input of the downsampling block and v is the input of the upsampling block. This is a 3×3 convolution operation, BN() is batch normalization, and LeakyReLU() is the LeakyReLU activation function. For the (r-1)th downsampling block, This is a 3×3 deconvolution operation. For the (r-1)th upsampling block, ReLU() is the ReLU activation function.

[0125] The fused terrain data is then input into the encoder and decoded step-by-step by the decoder. During this step-by-step decoding process, the decoder output needs to be densely connected and skipped to obtain a densely skipped output. .

[0126]

[0127] Where c is the ordinate in the decoder output and d is the abscissa in the decoder output. The coordinates directly below the current coordinate in the decoder output. The coordinates that are directly to the left of the current coordinate in the decoder output. For decoder, For an encoder, when the horizontal coordinate is 0, its output is the output of a pure encoder. When the ordinate is 0, its output is This is the pure decoder output of the coordinates directly to the left of the current coordinate, where U() represents the upsampling operation. Finally, convolution and activation are applied to the output of the dense skip connections to obtain the decoded output. .

[0128] S5: Acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0129] Furthermore, the objective of this stage is to obtain the target model, and then output reconstructed terrain data through the target model to complete the reconstruction of the seabed topography. Specifically, step S5 further includes:

[0130] S51: Obtain the real terrain data, and interpolate the real terrain data to obtain interpolated terrain data;

[0131] S52: Construct a discriminator, score the interpolated terrain data using the discriminator to obtain an interpolated terrain scalar score, score the decoded output using the discriminator to obtain a decoded output scalar score, score the real terrain data using the discriminator to obtain a real terrain score, and obtain an adversarial loss based on the interpolated terrain scalar score, the real terrain score, and the decoded output scalar score.

[0132] S53: Calculate the L1 norm and L2 norm between the real terrain data and the decoded output to obtain the terrain reconstruction loss, and calculate the perceptual loss and structural similarity loss between the real terrain data and the decoded output;

[0133] S54: The total terrain reconstruction loss is obtained through the adversarial loss, terrain reconstruction loss, perception loss and structural similarity loss. The model parameters are adjusted according to the total terrain reconstruction loss to obtain the target model. The reconstructed terrain data is output through the target model to complete the reconstruction of the seabed terrain.

[0134] The specific implementation method for the above steps in this embodiment is as follows:

[0135] To improve the accuracy of the decoding output obtained in step S4, it is necessary to obtain real seabed topographic data as the real topographic data Y. Subsequently, the decoding output will be linearly interpolated using the real topographic data to obtain the interpolated topographic data. Subsequently, a multi-scale discriminator is constructed. This discriminator sequentially performs multi-scale feature extraction, convolution, and activation on its input to obtain a score for that scale. By changing the discriminator's scale, it outputs scores for the interpolated terrain data, decoded output, and real terrain data at different scales, thus obtaining the interpolated terrain scalar score, decoded output scalar score, and real terrain score for each scale. Then, the WGAN-GP loss is calculated using the interpolated terrain scalar score, decoded output scalar score, and real terrain score at each scale, yielding the adversarial loss for each scale. Finally, the adversarial losses at each scale are weighted and summed using learnable scale weights to obtain the final adversarial loss. .

[0136] Subsequently, the L1 norm and L2 norm between the real terrain data and the decoded output are calculated to obtain the terrain reconstruction loss. :

[0137]

[0138] in, To find the L1 norm, To find the L2 norm, For L1 loss weights, The weights are L2 loss weights.

[0139] The real terrain data and the decoded output also need to be fed into the visual geometry group model to calculate the perceptual loss between the real terrain data and the decoded output. :

[0140]

[0141] in, The weights of the p-th layer in the visual geometry group model. This is the p-th layer of the visual geometry group model.

[0142] Structural similarity loss The calculation method is as follows:

[0143]

[0144] Where SSIM() is used to calculate the structural similarity parameter within the parentheses, and we have:

[0145]

[0146] in, This is the mean vector of the actual terrain data. This is the mean vector of the decoded output. The covariance between the actual terrain data and the decoded output. It is the first stability constant. It is the second stability constant. The variance of the actual terrain data. This represents the variance of the decoded output.

[0147] Subsequently, the adversarial loss, terrain reconstruction loss, perception loss, and structural similarity loss are weighted and summed to obtain the total terrain reconstruction loss. To minimize the total terrain reconstruction loss, the model parameters in steps S1 to S4 are adjusted to obtain a new decoding output. The total terrain reconstruction loss of the new decoding output is calculated, and the model parameters are adjusted again. This process is repeated iteratively from steps S1 to S4. The weight of the terrain reconstruction loss in each iteration is the terrain reconstruction loss weight in the s-th iteration. The calculation method is as follows:

[0148]

[0149] in, The terrain reconstruction loss is for the (s-1)th iteration. The perceptual loss for the (s-1)th iteration is... The structural similarity loss is for the (s-1)th iteration. Thus, when the total terrain reconstruction loss is less than a preset loss threshold or reaches a predetermined number of iterations, steps S1 to S4 with adjusted parameters can be used as the target model, and the decoded output of the target model can be used as terrain data to complete the reconstruction of the seabed terrain.

[0150] Here, the effectiveness of the multibeam sonar seabed topography reconstruction method was also verified, and its reconstruction results for the seabed topography are as follows: Figure 2 As shown, compared with traditional methods, this method improves terrain reconstruction accuracy by 20% to 40%, enhances the ability to preserve detailed features by 30% to 50%, improves processing efficiency by 20% to 35%, and significantly enhances robustness to noise and missing data.

[0151] The following describes the multibeam sonar seabed topography reconstruction device provided by the present invention. The multibeam sonar seabed topography reconstruction device described below and the multibeam sonar seabed topography reconstruction method described above can be referred to in correspondence.

[0152] Figure 3 A schematic diagram of a multibeam sonar seabed topography reconstruction system is shown, such as... Figure 3 As shown, the method for performing the multibeam sonar seabed topography reconstruction method described above includes:

[0153] Filtered data point module 100: used to acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on the state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points;

[0154] Spatial similarity module 200: used to acquire multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity through the spatial similarity parameters and attention weights;

[0155] The fused terrain data module 300 is used to calculate message vectors based on spatial similarity, obtain the final node representation based on the message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and the final node representations, and obtain fused terrain data based on the data fidelity terms, smoothing constraint terms, and terrain regularization terms.

[0156] Encoding / decoding module 400: Used to build encoders and decoders, inputting fused terrain data into the encoder, and decoding it through the decoder, performing dense jump connections to obtain the decoded output;

[0157] Terrain Reconstruction Module 500: Used to acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0158] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call a computer program in the memory 830 to execute a multibeam sonar seabed topography reconstruction method, which includes:

[0159] S1: Acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points.

[0160] S2: Obtain multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity using the spatial similarity parameters and attention weights;

[0161] S3: Calculate message vectors based on spatial similarity, obtain final node representations based on message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and final node representations, and obtain fused terrain data based on data fidelity terms, smoothing constraint terms, and terrain regularization terms;

[0162] S4: Construct an encoder and decoder. Input the fused terrain data into the encoder, and decode it through the decoder. Perform dense skip connections to obtain the decoded output.

[0163] S5: Acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0164] Furthermore, when the computer program in the aforementioned memory 830 can be implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0165] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the multibeam sonar seabed topography reconstruction method provided by the above methods, the method comprising:

[0166] S1: Acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points.

[0167] S2: Obtain multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity using the spatial similarity parameters and attention weights;

[0168] S3: Calculate message vectors based on spatial similarity, obtain final node representations based on message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and final node representations, and obtain fused terrain data based on data fidelity terms, smoothing constraint terms, and terrain regularization terms;

[0169] S4: Construct an encoder and decoder. Input the fused terrain data into the encoder, and decode it through the decoder. Perform dense skip connections to obtain the decoded output.

[0170] S5: Acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0171] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the aforementioned multibeam sonar seabed topography reconstruction methods, the method comprising:

[0172] S1: Acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points.

[0173] S2: Obtain multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity using the spatial similarity parameters and attention weights;

[0174] S3: Calculate message vectors based on spatial similarity, obtain final node representations based on message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and final node representations, and obtain fused terrain data based on data fidelity terms, smoothing constraint terms, and terrain regularization terms;

[0175] S4: Construct an encoder and decoder. Input the fused terrain data into the encoder, and decode it through the decoder. Perform dense skip connections to obtain the decoded output.

[0176] S5: Acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

[0177] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0178] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reconstructing seabed topography using multibeam sonar, characterized in that, include: S1: Acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points. S2: Obtain multi-source data, calculate the attention weights and spatial similarity parameters of the multi-source data, and calculate spatial similarity using the spatial similarity parameters and attention weights; S3: Calculate message vectors based on spatial similarity, obtain final node representations based on message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and final node representations, and obtain fused terrain data based on data fidelity terms, smoothing constraint terms, and terrain regularization terms; S4: Construct an encoder and decoder. Input the fused terrain data into the encoder, and decode it through the decoder. Perform dense skip connections to obtain the decoded output. S5: Acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and decoded output, adjust model parameters based on the total terrain reconstruction loss to obtain the target model, and output reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

2. The multibeam sonar seabed topography reconstruction method according to claim 1, characterized in that, Step S1 further includes: S11: Obtain the initial sonar data, construct a high-pass filter and a low-pass filter, and perform discrete wavelet transform on the elevation of the initial sonar data through the high-pass filter and the low-pass filter to obtain the approximation coefficients and the variance of the approximation coefficients. S12: Calculate the wavelet detail coefficient weighting parameter using the approximation coefficient variance and the approximation coefficient, and obtain the state estimation point based on the wavelet detail coefficient weighting parameter; S13: Calculate the Kalman gain, and obtain the initial filtered data points using the Kalman gain and the state estimation points; S14: Calculate the local standard deviation and median absolute deviation of the initial filtered data points, and use the local standard deviation and median absolute deviation to screen out outliers and obtain the filtered data points.

3. The multibeam sonar seabed topography reconstruction method according to claim 1, characterized in that, Step S2 further includes: S21: Obtain the multi-source data and construct a multi-source data graph model of the multi-source data; S22: Calculate the spatial similarity parameter through the multi-source data graph model, use a multi-head attention mechanism and obtain the attention weight through the multi-source data graph model, and calculate the spatial similarity through the spatial similarity parameter and the attention weight.

4. The multibeam sonar seabed topography reconstruction method according to claim 1, characterized in that, Step S3 further includes: S31: Calculate the message vector based on the spatial similarity, calculate the intermediate node representation through the message vector, and iterate the intermediate node representation layer by layer to obtain the final node representation; S32: Calculate the data fidelity term using the filtered data points and the final node representation, calculate the smoothing constraint term and terrain regularization term represented by the final node, and perform a weighted summation of the data fidelity term, smoothing constraint term and terrain regularization term to obtain the fused terrain data.

5. The multibeam sonar seabed topography reconstruction method according to claim 1, characterized in that, In step S4, the output of the dense skip connections is convolved and activated to obtain the decoded output.

6. The multibeam sonar seabed topography reconstruction method according to claim 1, characterized in that, Step S5 further includes: S51: Obtain the real terrain data, and interpolate the real terrain data to obtain interpolated terrain data; S52: Construct a discriminator, score the interpolated terrain data using the discriminator to obtain an interpolated terrain scalar score, score the decoded output using the discriminator to obtain a decoded output scalar score, score the real terrain data using the discriminator to obtain a real terrain score, and obtain an adversarial loss based on the interpolated terrain scalar score, the real terrain score, and the decoded output scalar score. S53: Calculate the L1 norm and L2 norm between the real terrain data and the decoded output to obtain the terrain reconstruction loss, and calculate the perceptual loss and structural similarity loss between the real terrain data and the decoded output; S54: The total terrain reconstruction loss is obtained through the adversarial loss, terrain reconstruction loss, perception loss and structural similarity loss. The model parameters are adjusted according to the total terrain reconstruction loss to obtain the target model. The reconstructed terrain data is output through the target model to complete the reconstruction of the seabed terrain.

7. A multibeam sonar seabed topography reconstruction system, used to perform the multibeam sonar seabed topography reconstruction method as described in any one of claims 1 to 6, characterized in that, include: Filtered data point module: used to acquire initial sonar data, perform discrete wavelet transform on the initial sonar data to obtain approximation coefficients and approximation coefficient variance, obtain state estimation points through approximation coefficients and approximation coefficient variance, obtain initial filtered data points based on state estimation points, and perform adaptive filtering on the initial filtered data points to obtain filtered data points; Spatial similarity module: used to acquire multi-source data, calculate attention weights and spatial similarity parameters of multi-source data, and calculate spatial similarity through spatial similarity parameters and attention weights; The terrain data fusion module is used to calculate message vectors based on spatial similarity, obtain the final node representation based on the message vectors, calculate data fidelity terms, smoothing constraint terms, and terrain regularization terms by filtering data points and the final node representations, and obtain fused terrain data based on the data fidelity terms, smoothing constraint terms, and terrain regularization terms. Encoding / decoding module: Used to build encoders and decoders. It inputs fused terrain data into the encoder, decodes it through the decoder, performs dense skip connections, and obtains the decoded output. Terrain Reconstruction Module: This module is used to acquire real terrain data, calculate the total terrain reconstruction loss using the real terrain data and the decoded output, adjust the model parameters based on the total terrain reconstruction loss to obtain the target model, and output the reconstructed terrain data using the target model to complete the reconstruction of the seabed terrain.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multibeam sonar seabed topography reconstruction method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multibeam sonar seabed topography reconstruction method as described in any one of claims 1 to 6.

10. A computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, characterized in that, When the program instructions are executed by a computer, the computer is able to perform the multibeam sonar seabed topography reconstruction method as described in any one of claims 1 to 6.

Images