Method and system for underwater detection and simulation of water conveyance culvert section and siltation thickness

By employing techniques such as adaptive frequency optimization and fractal geometric acoustic models, the problems of multipath effects and biological interference in complex culvert environments were solved, enabling high-precision underwater detection and sediment thickness reconstruction, thus improving the reliability and interpretability of the detection results.

CN118673737BActive Publication Date: 2024-11-15NANJING HYDRAULIC RES INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411167502.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-15
Estimated Expiration
2044-08-23

AI Technical Summary

Technical Problem

Existing underwater detection technologies face challenges in complex culvert environments, including severe multipath effects, the impact of non-uniform sediment on sound wave propagation, insufficient handling of biological activity interference, misjudgment of detection results, and inadequate quantification of uncertainty. These challenges make it difficult to achieve high-resolution cross-section and sediment distribution reconstruction.

Method used

An adaptive frequency optimization algorithm is used to generate the sonar array configuration. The dynamic time-reversal mirror algorithm and fractal geometric acoustic model are combined to compensate for multipath effects and enhance data. Sediment features are extracted through multi-scale topological data analysis and Volterra series expansion method. Adaptive filters are used to remove the influence of biological activity. High-resolution reconstruction is performed by combining compressed sensing and topology-preserving manifold learning. Uncertainty quantification is also performed. Finally, visualization results are generated through ray tracing and volume rendering.

Benefits of technology

It enables high-precision analysis and characterization of the cross-section and siltation thickness of water conveyance culverts, improves the accuracy and reliability of detection, enhances the interpretability and visualization of results, and supports scientific decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118673737B_ABST
    Figure CN118673737B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for underwater detection and simulation of water conveyance culvert cross section and siltation thickness, the method comprising obtaining culvert environmental parameters, generating an optimized sonar array configuration, forming an original multi-band sonar data set, performing noise reduction processing and multipath effect compensation; constructing a fractal geometry acoustic model, enhancing and reconstructing the sonar data set; performing multi-scale topological data analysis on the enhanced sonar data set, and constructing a layered siltation model; performing time-frequency domain chaos analysis on the enhanced sonar data set, constructing an adaptive filter, and performing high-resolution image reconstruction to obtain preliminary reconstruction data; optimizing preliminary reconstruction data based on the layered siltation model; performing uncertainty quantification and generating an adaptive grid, performing high-fidelity visual rendering, and generating a culvert cross section diagram and a siltation thickness distribution diagram. The present invention not only improves the accuracy and reliability of underwater detection, but also improves the accuracy and interpretability of underwater detection results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of underwater detection, and in particular to a method and a system for underwater detection and simulation of a water conveyance culvert section and siltation thickness. Background Art

[0002] As a key infrastructure in water conservancy projects, the safe operation of water transfer culverts is directly related to the efficiency of water resource allocation and regional water safety. With the increase of service life, the deterioration of the internal structure and siltation of culverts are becoming more and more serious, which may lead to a decrease in water transfer capacity, deterioration of water quality, and even structural accidents. Therefore, accurate and efficient detection of culvert cross-sectional structure and siltation thickness distribution is crucial for timely assessment of culvert conditions, formulation of maintenance strategies and ensuring long-term safe operation. Especially in the context of the increasing number of large-scale inter-basin water transfer projects, the importance of underwater detection technology is more prominent. It is not only related to the safety of the project itself, but also affects regional economic development and ecological environmental protection.

[0003] At present, underwater detection technology has made some progress, mainly including sonar detection, video endoscopy and laser scanning. These technologies perform well under ideal conditions, but still face many challenges in complex actual culvert environments. Specifically: in culverts with complex geometric structures, the multipath effect seriously affects the quality of sonar signals. Traditional signal processing methods are difficult to effectively separate and identify useful information, resulting in insufficient cross-section reconstruction accuracy. The presence of non-uniform sediment makes the propagation characteristics of sound waves complicated. Existing acoustic models are difficult to accurately describe this complex medium, affecting the measurement accuracy of sediment thickness. The interference caused by biological activities (such as algae growth and fish activities) on sonar signals has not been fully valued and effectively processed, which may lead to misjudgment of detection results. It is still a challenge to achieve high-resolution reconstruction of cross-sections and silt distribution with limited detection data, especially in large or long-distance culverts. How to balance detection efficiency and accuracy becomes a key issue. Existing visualization methods are often difficult to intuitively display complex three-dimensional structures and silt distributions, especially in the presentation of uncertainty information, which affects the interpretability and practicality of detection results. Most existing methods lack uncertainty quantification of test results, which is a significant shortcoming in risk assessment and decision support.

[0004] Therefore, further research and innovation are needed to solve the above problems existing in the prior art. Summary of the invention

[0005] The purpose of the invention is to provide a method and system for underwater detection and simulation of water conveyance culvert cross-section and siltation thickness, so as to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution, a method for underwater detection and simulation of water conveyance culvert section and siltation thickness, comprising the following steps:

[0007] S1. Obtain culvert environmental parameters, use an adaptive frequency optimization algorithm to generate an optimized sonar array configuration; based on the sonar array configuration, simultaneously transmit and receive sonar signals of predetermined frequency bands to form an original multi-band sonar data set; perform noise reduction processing on the original multi-band sonar data set to obtain a noise-reduced multi-band sonar data set; wherein the culvert environmental parameters include estimated size data and water quality data;

[0008] S2. Based on the multi-band sonar dataset after denoising, a dynamic time reversal mirror algorithm is used to compensate for multipath effects to obtain a sonar dataset after multipath effects compensation; based on the sonar dataset after multipath effects compensation, a fractal geometry acoustic model is constructed, and the fractal geometry acoustic model is used to enhance and reconstruct the sonar dataset to obtain an enhanced sonar dataset;

[0009] S3. Perform multi-scale topological data analysis on the enhanced sonar data set to extract the topological features of the sediments. Based on the topological features of the sediments, use the Volterra series expansion method to extract the nonlinear acoustic characteristics. Combine the topological features and nonlinear acoustic characteristics to construct a layered sedimentation model.

[0010] S4. Perform time-frequency domain chaos analysis on the enhanced sonar data set to extract acoustic features caused by biological activities; construct an adaptive filter based on the acoustic features, remove the influence of biological activities based on the adaptive filter, and obtain a sonar data set compensated for biological activities;

[0011] S5. Based on the sonar data set after biological activity compensation, high-resolution image reconstruction was performed using compressed sensing theory and L1 optimization to obtain preliminary reconstruction data; based on the layered sedimentation model, the preliminary reconstruction data was optimized using the topology-preserving manifold learning algorithm to obtain optimized reconstruction data; the polynomial chaos expansion method was used to quantify the uncertainty of the optimized reconstruction data to obtain the cross-section and sedimentation thickness data with uncertainty;

[0012] S6. Generate an adaptive mesh based on the uncertain cross-section and siltation thickness data. Based on the adaptive mesh, use ray tracing and volume rendering technology to perform high-fidelity visualization rendering to generate culvert cross-section diagrams and siltation thickness distribution maps.

[0013] The system for underwater detection and simulation of water conveyance culvert section and siltation thickness includes:

[0014] at least one processor; and,

[0015] a memory communicatively connected to at least one of the processors; wherein,

[0016] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method of underwater detection and simulation of water transfer culvert cross-section and siltation thickness described in any of the above technical solutions.

[0017] Beneficial effects: The present invention not only realizes high-precision analysis and characterization of non-uniform silt in water transfer culverts, improving the accuracy and reliability of underwater detection; it also realizes high-precision reconstruction and uncertainty quantification of water transfer culvert cross-sections and silt thickness, improving the accuracy and interpretability of underwater detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the present invention.

[0019] Figure 2 This is a flow chart of step S1 of the present invention.

[0020] Figure 3 This is a flow chart of step S2 of the present invention.

[0021] Figure 4 This is a flow chart of step S3 of the present invention.

[0022] Figure 5 This is a flow chart of step S4 of the present invention.

[0023] Figure 6 This is a flow chart of step S5 of the present invention.

[0024] Figure 7 This is a flow chart of step S6 of the present invention. DETAILED DESCRIPTION

[0025] like Figure 1 As shown, the present application proposes a method for underwater detection and simulation of water conveyance culvert cross-section and siltation thickness, comprising the following steps:

[0026] S1. Obtain culvert environmental parameters, use an adaptive frequency optimization algorithm to generate an optimized sonar array configuration; based on the sonar array configuration, simultaneously transmit and receive sonar signals of predetermined frequency bands to form an original multi-band sonar data set; perform noise reduction processing on the original multi-band sonar data set to obtain a noise-reduced multi-band sonar data set; wherein the culvert environmental parameters include estimated size data and water quality data;

[0027] S2. Based on the multi-band sonar dataset after denoising, a dynamic time reversal mirror algorithm is used to compensate for multipath effects to obtain a sonar dataset after multipath effects compensation; based on the sonar dataset after multipath effects compensation, a fractal geometry acoustic model is constructed, and the fractal geometry acoustic model is used to enhance and reconstruct the sonar dataset to obtain an enhanced sonar dataset;

[0028] S3. Perform multi-scale topological data analysis on the enhanced sonar data set to extract the topological features of the sediments. Based on the topological features of the sediments, use the Volterra series expansion method to extract the nonlinear acoustic characteristics. Combine the topological features and nonlinear acoustic characteristics to construct a layered sedimentation model.

[0029] S4. Perform time-frequency domain chaos analysis on the enhanced sonar data set to extract acoustic features caused by biological activities; construct an adaptive filter based on the acoustic features, remove the influence of biological activities based on the adaptive filter, and obtain a sonar data set compensated for biological activities;

[0030] S5. Based on the sonar data set after biological activity compensation, high-resolution image reconstruction was performed using compressed sensing theory and L1 optimization to obtain preliminary reconstruction data; based on the layered sedimentation model, the preliminary reconstruction data was optimized using the topology-preserving manifold learning algorithm to obtain optimized reconstruction data; the polynomial chaos expansion method was used to quantify the uncertainty of the optimized reconstruction data to obtain the cross-section and sedimentation thickness data with uncertainty;

[0031] S6. Generate an adaptive mesh based on the uncertain cross-section and siltation thickness data. Based on the adaptive mesh, use ray tracing and volume rendering technology to perform high-fidelity visualization rendering to generate culvert cross-section diagrams and siltation thickness distribution maps.

[0032] This embodiment can obtain high-precision sonar data in a complex underwater environment by using an adaptive frequency optimization algorithm and multi-band sonar signals. Noise reduction processing and multipath effect compensation further improve the accuracy of the data and ensure the reliability of the detection results. The fractal geometry acoustic model is used to enhance and reconstruct the sonar data, so that the spatial resolution and detail performance of the data are improved, and the internal structure and siltation of the culvert can be more accurately reflected, providing a high-quality data basis for subsequent analysis. The topological features and nonlinear acoustic characteristics of the sediment can be extracted by multi-scale topological data analysis and Volterra series expansion method. The extraction of these features helps to build a more accurate layered sedimentation model, so as to better understand and predict the distribution and changes of the sediment. The use of time-frequency domain chaos analysis and adaptive filters enables the system to effectively remove the interference of biological activities on the sonar data, obtain a purer sonar data set, improve the credibility of the data, and reduce the influence of external factors on the detection results. High-resolution image reconstruction can be achieved through compressed sensing theory and L1 optimization technology. Combined with the topology-preserving manifold learning algorithm, the reconstructed data is further optimized, so that the final image has higher resolution and better visual effects; the application of the polynomial chaos expansion method enables the system to quantify the uncertainty of the reconstructed data and provide cross-sectional and siltation thickness data with uncertainty, which helps to evaluate the reliability and accuracy of the detection results and provide a scientific basis for decision-making; based on adaptive grids, ray tracing and volume rendering technology, high-fidelity culvert cross-sectional diagrams and siltation thickness distribution maps can be generated. These visualization results intuitively show the internal structure and siltation conditions of the culvert, which is convenient for engineering personnel to analyze and make decisions.

[0033] This embodiment realizes high-precision detection and simulation of the cross-section and siltation thickness of the water conveyance culvert, providing strong technical support for underwater engineering detection and maintenance.

[0034] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0035] S11, extracting culvert environmental parameters from a pre-stored database, using an adaptive frequency optimization algorithm to generate an initial frequency range and array layout scheme, and calculating the optimal sonar array configuration, including an optimized frequency set and array layout, through an iterative optimization algorithm;

[0036] S12, based on the sonar array configuration, control the sonar array to simultaneously transmit sonar signals of predetermined frequency bands, record the echo signals received in each frequency band, use synchronous sampling technology to ensure the time alignment of the signals in each frequency band, and combine the sonar signals transmitted in all frequency bands and the echo signals received into an original multi-band sonar data set;

[0037] S13, based on the original multi-band sonar data set, using wavelet packet transform to obtain time-frequency domain representation; using adaptive threshold method to perform noise reduction on the time-frequency domain representation to obtain a noise-reduced time-frequency domain representation;

[0038] S14. Using inverse wavelet packet transform, the denoised time-frequency domain representation is converted back to the time domain to obtain a denoised signal; and the denoised signals of all frequency bands are combined into a denoised multi-band sonar data set.

[0039] According to one aspect of the present application, step S12 further includes:

[0040] S12a, using a real-time signal quality evaluation algorithm to monitor the quality of the sonar signal and the echo signal, and obtaining a monitoring quality result;

[0041] S12b. Determine whether the monitoring quality result is lower than a preset threshold. If so, re-collect the signal of the frequency band; otherwise, do nothing.

[0042] In one embodiment of the present application, culvert environmental parameters E, including estimated size d and water quality parameters w, are read from a pre-stored database. These parameters are processed using an adaptive frequency optimization algorithm (AFO) to obtain an initial frequency range f range and array layout scheme L initial Using the iterative optimization algorithm (IOA), according to the initial frequency range f range and array layout scheme L initial , calculate the best sonar array configuration parameter A. Among them, the sonar array configuration parameter A contains the optimized frequency set F opt and the array layout L opt , A = IOA(AFO(E)) = {F opt , L opt}, where E = {d, w}, d is the estimated size of the culvert, and w is the water quality parameter. Specifically:

[0043] The initial frequency range f range and array layout scheme L initial As the initial input, generate the initial configuration parameters. In each iteration, adjust the frequency set and array layout to optimize the coverage and resolution of the sonar signal. Define an objective function to measure the quality of the configuration parameters. The objective function includes indicators such as signal coverage, resolution, and signal-to-noise ratio. Use optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) to adjust the frequency set and array layout so that the objective function reaches the optimal value. Set the convergence condition. When the value of the objective function does not change much in multiple iterations, stop the iteration and output the final optimized configuration parameter A.

[0044] Based on the optimized configuration parameter A, the sonar array is controlled to simultaneously transmit sonar signals S in multiple frequency bands i (i = 1, 2, ..., n, where n is the number of frequency bands). For each frequency band, record the received echo signal R i The transmitted and received signals of all frequency bands are combined into the original multi-band sonar dataset D. The structure of dataset D is: D = {(S1, R1), (S2, R2), ..., (S n , R n During the signal acquisition process, synchronous sampling technology (SST) is used to ensure the time alignment of the signals in each frequency band, and a real-time signal quality assessment algorithm (RSQA) is applied to monitor the signal quality. If the signal quality of a certain frequency band is lower than the preset threshold T q , the data of this frequency band will be collected again immediately.

[0045] Read the original multi-band sonar data set D and calculate the received signal R for each band i Apply wavelet packet transform (WPT) to get the time-frequency domain representation TF i ; Use the adaptive threshold method (ATM) to represent TF in the time-frequency domain i Perform noise reduction to obtain the time-frequency domain representation TF after noise reduction i '; Use the inverse wavelet packet transform (IWPT) to transform the denoised time-frequency domain representation TF i 'Convert back to the time domain to get the noise-reduced received signal R i '. The denoised signals of all frequency bands are combined into a denoised multi-band sonar dataset D'. Specifically: TF i = WPT(R i );TF i = ATM(TF i );R i = IWPT(TF i ');D' = {(S1, R1'), (S2, R2'),..., (S n , R n ')}, where the adaptive threshold T of the adaptive threshold method a Dynamically adjusted according to the noise level of each sub-band: T a = k * σ j , σ j represents the noise standard deviation of the j-th subband, and k is an adjustable coefficient.

[0046] This embodiment improves the adaptability and data acquisition quality of the underwater detection system in a complex water culvert environment by adopting an adaptive frequency optimization algorithm and multi-band sonar signal acquisition technology. Specifically, the adaptive frequency optimization algorithm can dynamically adjust the frequency and layout of the sonar array according to the estimated size of the culvert and water quality parameters, so that the sonar signal can better penetrate silt layers of different depths and densities, thereby improving the detection accuracy of silt thickness. Multi-band signal acquisition technology allows sonar data of different frequencies to be acquired simultaneously, which is crucial for distinguishing culvert walls, silt layers, and possible suspended matter, because different materials and structures have different sound wave reflection characteristics for different frequencies. The introduction of a real-time signal quality assessment algorithm ensures the reliability of the collected data, and reduces the error accumulation in subsequent processing by instantly re-collecting frequency band data of poor quality. The application of wavelet packet transform and adaptive threshold method achieves efficient signal noise reduction, which is crucial for extracting effective information in a complex underwater environment. Especially in the case of multiple echoes and environmental noise interference that may exist in the culvert, this noise reduction method can effectively retain the key features of the signal while suppressing background noise, laying a solid data foundation for subsequent cross-sectional imaging and silt thickness estimation. This embodiment can obtain high-quality raw data in a complex water transfer culvert environment, providing a strong guarantee for the accuracy and reliability of the entire detection process.

[0047] In another embodiment of the present application, step S13 and step S14 may also be implemented by a wavelet attention module.

[0048] Among them, the wavelet attention module includes:

[0049] The multi-scale wavelet decomposition layer is used to perform multi-scale wavelet decomposition on the input sonar signal (original sonar signal D'); output a set of sub-band signals {S1, S2, ..., Sn} in different frequency bands;

[0050] An adaptive threshold denoising module is used to perform adaptive threshold denoising on each sub-band signal based on the sub-band signal set {S1, S2, ..., Sn}; obtain a denoised sub-band signal set {S1', S2', ..., Sn'};

[0051] The subband attention module is used to learn and assign importance weights of information in different frequency bands, assign weights to the denoised subband signal set {S1', S2', ..., Sn'}, and obtain the weighted subband signal set {W1S1, W2S2,..., WnSn};

[0052] Subband CNN is used to extract specific spatial features for each subband and output the subband feature map Fi based on the weighted subband signal WiSi;

[0053] Cross-scale fusion LSTM is used to fuse the features of different sub-band scales to capture time dependency; based on all sub-band feature maps {F1, F2, ..., Fn}; the fused temporal features T are obtained;

[0054] Adaptive reconstruction layer, used to learn the optimal wavelet reconstruction strategy, fuse the model output and the original wavelet coefficients; its input data is the fused time series feature T and the original wavelet coefficients; the output is the denoised and enhanced sonar signal D'''

[0055] At work, the data processing flow is as follows:

[0056] Step S131: Receive an original multi-band sonar data set D. Perform wavelet decomposition on each frequency band signal in D to obtain a sub-band signal set {S1, S2, ..., Sn} of different frequency bands.

[0057] Step S132, processing each sub-band signal, including: calculating the noise standard deviation σi for each sub-band Si; constructing an adaptive threshold Ti = k * σi, where k is an adjustable coefficient; performing soft threshold processing on coefficients less than the threshold Ti; and outputting the denoised sub-band signal Si';

[0058] Step S133: Analyze the importance of each denoised sub-band signal, assign corresponding weights, and obtain {W1S1', W2S2', ..., WnSn'}.

[0059] Step S134: Each weighted denoised subband signal is passed through a corresponding subband-specific CNN to extract spatial features and obtain a feature map set {F1, F2, ..., Fn}.

[0060] Step S135: All feature maps are taken as input through cross-scale fusion LSTM, processed in chronological order, and the fused temporal features T are output.

[0061] Step S136: Based on the adaptive reconstruction layer, the enhanced and denoised sonar signal D' is reconstructed by combining the temporal features T and the denoised wavelet coefficients.

[0062] Step S137: convert the enhanced and denoised sonar signal D' from the time-frequency domain representation back to the time domain to obtain a denoised signal; and then combine it into a denoised and enhanced multi-band sonar data set.

[0063] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0064] S21, time-reversing the received signal in the multi-band sonar data set after noise reduction to obtain the time-reversed received signal; based on the time-reversed received signal, calculating the cross-correlation function between the received signal and the transmitted signal in the multi-band sonar data set after noise reduction;

[0065] S22, estimating the sound field propagation function based on the peak position of the cross-correlation function; and obtaining the sonar data set after multipath effect compensation through a deconvolution process based on the sound field propagation function;

[0066] S23. Based on the sonar data set after multipath effect compensation, use the multi-resolution analysis method to generate components of different scales; calculate the fractal dimension and Hurst index of each scale component, and construct a fractal model to describe the acoustic propagation characteristics in the culvert based on the fractal dimension and Hurst index;

[0067] S24. Based on the sonar dataset after multipath effect compensation, a fractal model is used to perform deconvolution to obtain a deconvolved sonar dataset; a subspace projection method is used to perform noise suppression and feature extraction on the deconvolved sonar dataset to obtain an enhanced sonar dataset.

[0068] In one embodiment of the present application, the multi-band sonar data set D' after noise reduction is read, and the signal pair (S i , R i ') Apply the dynamic time reversal mirror (D-TRM) algorithm. The received signal R i 'Time reversal to obtain the received signal R after time reversal irev . Calculate the received signal R after time reversal irev With the transmission signal S i The cross-correlation function C i Using the cross-correlation function C i The peak position of the estimated sound field propagation function H i Using the sound field propagation function H i The original received signal R i 'Perform deconvolution to obtain the signal R after multipath effect compensation i ''. The specific formula is as follows:

[0069] R irev (t) = R i '(-t);C i (τ) = ∫R irev (t)S i (t+τ)dt;H i = F{C i} / F{S i * S i}; R i '' =F -1{F{R i '} / H i};

[0070] Among them, F and F -1 They represent Fourier transform and inverse Fourier transform respectively, t represents the time variable, τ represents the time delay or time offset, and dt represents the small time increment of the integration.

[0071] The multipath-compensated signals of all frequency bands are combined to obtain the multipath-compensated sonar data set D'': D'' = {(S1, R1''), (S2, R2''), ..., (S n , R n '')}.

[0072] Based on the sonar data set D'' after multipath compensation, a fractal geometry acoustic model is constructed. The signal R of each frequency band is converted into i ''Decompose into components of different scales. Calculate the fractal dimension D of each scale component f and Hurst index H. Using these parameters, a fractal model M describing the acoustic propagation characteristics in the culvert is constructed. f Specifically: for each signal R i ''Perform wavelet decomposition to obtain the approximate coefficient a j and detail factor d j (j = 1, 2, ..., J); calculate the variance σ of each scale j j 2 ; Fitting log(σ j 2 ) vs. j, and get the slope of the straight line β, where vs means contrast or comparison; calculate the fractal dimension D f = (5-β) / 2 and Hurst exponent H = (β-1) / 2; construct fractal model M f , its transfer function is: H f (f) = k * f -β / 2 * exp(i*π*β / 4), where exp( ) represents the exponential function, f represents the frequency, k represents the proportional coefficient, and i represents the imaginary unit; the resulting fractal acoustic model parameter F contains the D of each frequency band. f , H and M f :F = {(D f1 , H1, M f1 ), (D f2 , H2, M f2 ), ..., (D fn , H n , M fn )}.

[0073] The sonar data set D'' after multipath effect compensation and the fractal acoustic model parameters F are used to enhance and reconstruct the signal. i '', apply the corresponding fractal model M fi Perform deconvolution to obtain the enhanced signal R i '''. The subspace projection method (SPM) is used to suppress noise and extract features from the enhanced signal. The specific steps are as follows: i ''Perform Fourier transform to obtain the transformed signal R i ''(f); Apply fractal model for deconvolution: R temp (f) = R i ''(f) / H fi (f); Perform inverse Fourier transform to obtain the time domain signal R temp (t); Construct the Hankel matrix H of the signal and perform singular value decomposition (SVD): H = U * S * V T , where U represents an orthogonal matrix, which contains the left singular vectors of the original matrix, S represents a diagonal matrix, which contains the singular values ​​of the original matrix, which are arranged in descending order, V represents an orthogonal matrix, which contains the right singular vectors of the original matrix, and T represents a transpose operation; select the first k singular values ​​and the corresponding singular vectors to reconstruct the signal: R i ''' = U k * S k * V k T ; Combine the enhanced signals of all frequency bands to obtain the enhanced sonar data set D''': D''' = {(S1, R1'''), (S2, R2'''), ..., (S n , R n ''')}.

[0074] This embodiment effectively solves the multipath effect problem in the complex geometric structure of the water culvert by combining the dynamic time reversal mirror (D-TRM) algorithm, the fractal geometry acoustic model and the signal enhancement technology, and improves the quality and interpretability of the sonar signal. The application of the D-TRM algorithm cleverly utilizes the time reversibility of the sound field, and successfully compensates for the multipath propagation effect caused by the culvert wall, corners and irregular structures by time reversing and deconvolving the received signal. This makes it possible to accurately restore the main propagation path of the sound wave even in narrow, curved or branched culverts, thereby improving the accuracy of subsequent cross-sectional imaging. The introduction of the fractal geometry acoustic model is a breakthrough for the complex acoustic environment in the culvert. The model accurately describes the propagation characteristics of sound waves in structures of different scales through multi-resolution analysis and fractal dimension calculation, which is crucial to understanding the interaction between sound waves and the culvert wall, silt layer and possible structural defects. In particular, when dealing with old culverts that have been in operation for a long time, it can capture subtle changes in acoustic characteristics caused by erosion, deposition or structural deformation, providing more comprehensive information for evaluating the state of the culvert. The signal enhancement and reconstruction technology based on the fractal model further improves the signal-to-noise ratio and resolution of the signal. Through the subspace projection method, not only the residual noise is effectively suppressed, but also the key features related to the culvert structure and silt distribution are enhanced. This enhancement effect is particularly significant in dealing with deep silt or high turbidity water environments, enabling the system to achieve accurate cross-sectional depiction and silt thickness measurement when traditional methods are difficult to obtain clear signals. This embodiment improves the performance of the underwater detection system in complex water transfer culvert environments, lays a solid foundation for subsequent detailed analysis and visualization, and enables engineers to more accurately evaluate the structural integrity and operating conditions of the culvert.

[0075] In another embodiment of the present application, steps S23 and S24 may also be implemented by a fractal enhancement module, wherein the fractal enhancement module includes:

[0076] A fractal feature extractor is used to extract fractal features of signals from the sonar data set D'' after multipath effect compensation using a fractal convolution layer to obtain a fractal feature atlas F;

[0077] Multi-resolution graph attention network, used to process the relationship between features of different scales, and obtain the feature atlas F' with multi-scale relationship enhancement based on the fractal feature atlas F;

[0078] Recursive fractal generator, used to generate fractal model parameters, and obtain fractal acoustic model parameter set M through feature atlas F' enhanced by multi-scale relationship;

[0079] An adversarial enhancement module is used to improve the signal enhancement quality through adversarial training. The adversarial enhancement module reads the fractal acoustic model parameter set M and the original data D'' and outputs a preliminary enhanced signal E;

[0080] The physical constraint layer is used to make the generated model conform to the physical laws of acoustic propagation, read the preliminary enhanced signal E, and output the final enhanced sonar data set D''' and fractal acoustic model parameters F.

[0081] At work, the data processing flow is as follows:

[0082] Step S231, receiving the sonar data set D'' after multipath effect compensation from step S21.

[0083] Step S232, the fractal feature extractor uses a fractal convolutional layer to process D'' and extract a fractal feature atlas F.

[0084] Step S233: The multi-resolution graph attention network processes the fractal feature atlas F to generate a multi-scale relationship enhanced feature atlas F'.

[0085] Step S234: The recursive fractal generator generates a fractal acoustic model parameter set M based on the feature atlas F' enhanced by multi-scale relationships.

[0086] Step S235: The adversarial enhancement module generates a preliminary enhanced signal E using the fractal acoustic model parameter set M and the original data D''.

[0087] Step S236: The physical constraint layer applies acoustic propagation constraints to the initially enhanced signal E, and outputs the final enhanced sonar data set D''' and fractal acoustic model parameters F.

[0088] Step S237: Use the enhanced sonar data set D''' as input for subsequent steps, and the fractal acoustic model parameter F is used to construct a fractal model for subsequent analysis.

[0089] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0090] S31. Based on the enhanced sonar data set, the persistence graph of each signal is constructed using the persistence coherence algorithm; the topological feature vector of each persistence graph is calculated, including the Betti number sequence and persistence entropy; the topological feature vectors of all frequency bands are combined using the kernel method to obtain the topological characteristics of the sediment;

[0091] S32. Based on the topological characteristics of sediment and the enhanced sonar data set, the Volterra series expansion method is used to construct a nonlinear system model and estimate the high-order kernel function. Based on the high-order kernel function, nonlinear metrics are calculated, including harmonic distortion rate and intermodulation distortion. Based on the nonlinear metrics, the principal component analysis dimensionality reduction method is used to obtain nonlinear acoustic characteristics.

[0092] S33. Based on the topological characteristics and nonlinear acoustic characteristics of the sediment, a hierarchical Bayesian network is used to construct a stratified sedimentation model framework, and the expectation maximization algorithm is used to estimate the parameters of the stratified sedimentation model, ultimately forming a stratified sedimentation model.

[0093] According to one aspect of the present application, step S31 is further:

[0094] S311, constructing a time delay embedding based on the enhanced sonar data set to obtain point cloud data;

[0095] S312, based on point cloud data, using the Vietoris-Rips complex VR algorithm to construct simple complex sequences at different scales;

[0096] S313. Based on the simplicial complex sequences at different scales, 0-dimensional and 1-dimensional persistence homology are calculated to obtain persistence graphs; at the same time, the Betti number sequence is calculated;

[0097] S314, based on the persistence graph and the Betti number sequence, the persistence entropy is calculated, and the topological feature vector of each frequency band is constructed;

[0098] S315. Use the Gaussian kernel function to fuse the topological feature vectors of all frequency bands to obtain the topological features of the sediment.

[0099] In one embodiment of the present application, the enhanced sonar data set D''' is read, and a multi-scale topological data analysis (TDA) is performed on the signal Ri''' of each frequency band; a persistence graph PD of each signal is constructed using a persistent homology algorithm (PH) i . Calculate the topological feature vector TF of each persistence graph i , including the Betti number sequence and persistence entropy; the topological feature vectors of all frequency bands are combined using the kernel method to obtain the overall topological feature representation T of the sediment. The specific steps are as follows: for each reconstructed signal R i '''Build time delay embedding to get point cloud data PC i ; Apply the Vietoris-Rips complex VR algorithm to construct simplicial complex sequences at different scales ε; calculate 0-dimensional and 1-dimensional persistent homology to obtain the persistent graph PD i ={PD i0 , P.D. i1}; Calculate the Betti number sequence β i (ε) = {β i0 (ε), β i1 (ε)}; Calculate the continuous entropy PE i = -Σ(p k * log(p k )), where p kis the relative life span of the kth persistence interval in the persistence graph; construct the topological feature vector TF i = [β i (ε), PE i ]; use Gaussian kernel function K(x, y) = exp(-||xy|| 2 / (2σ 2 )) The topological feature vectors of all frequency bands are fused to obtain the sediment topological feature T, where x and y represent two different topological feature vectors, σ 2 represents the variance parameter in the Gaussian kernel function; the final sediment topological feature T is expressed as: T=Σ(w i * K(TF i , TF j )), where w i The weight for each frequency band.

[0100] Based on the topological features T of the sediment and the enhanced sonar data set D''', the nonlinear acoustic characteristics of the sediment are extracted. The Volterra series expansion method is used to build a nonlinear system model and estimate high-order kernel functions. Nonlinear metrics such as harmonic distortion (THD) and intermodulation distortion (IMD) are calculated. The specific steps are as follows: For each frequency band, the signal R i '''Perform Volterra series expansion:

[0101] y(t) = h1(τ1)x(t-τ1)dτ1+∫∫h2(τ1,τ2)x(t-τ1)x(t-τ2)dτ1dτ2+ ...;

[0102] Where h1, h2, etc. are first-order and second-order Volterra kernels, τ1 and τ2 represent time delay variables at different time points, x( ) represents the input signal, t represents time, and d represents the integration operation;

[0103] Use the least squares method to estimate the Volterra kernel function; calculate the harmonic distortion rate THD = sqrt(Σ(A n 2 )) / A1, where A n is the nth harmonic amplitude; calculate the intermodulation distortion IMD = (A f1+f2 + A f1-f2 ) / (A f1 + A f2 ), where f1 and f2 are two fundamental frequencies; construct the nonlinear feature vector NLF i =[THD i , IMD i , h 2normi , ...], where h 2normirepresents the norm of the second-order Volterra kernel h2; principal component analysis (PCA) is used to reduce the dimension to obtain the final nonlinear acoustic feature N. The final nonlinear acoustic feature N of the sediment is expressed as: N = PCA([NLF1, NLF2, ..., NLF n ]).

[0104] Combine the topological characteristics T and nonlinear acoustic characteristics N of the sediment to construct a layered sedimentation model. A hierarchical Bayesian network (HBN) is used to represent the vertical structure and attribute distribution of the sediment. The specific steps are as follows: define the model hierarchy: bottom layer (close to the bottom plate), middle layer and surface layer; define node variables for each layer: density ρ, sound speed c, attenuation coefficient α, nonlinear parameter β; set the prior distribution P(ρ), P(c), P(α), P(β) based on historical data and expert knowledge; construct a conditional probability table (CPT) to represent the inter-layer relationship and parameter correlation; use the expectation maximization algorithm (EM) to estimate the model parameters θ combined with the topological characteristics T and nonlinear acoustic characteristics N of the sediment; calculate the posterior probability distribution P(ρ, c, α, β|T, N)∝P(T, N|ρ, c, α, β)P(ρ, c, α, β), where ∝ means proportional to; the final layered sedimentation model parameter L contains the parameter distribution of each layer and the inter-layer relationship: L = {θ bottom , θ middle , θ top , CPT}, where θ bottom ,θ middle and θ top Represent the model parameters of the bottom layer (near the bottom plate), middle layer and surface layer respectively.

[0105] This embodiment achieves high-precision analysis and characterization of non-uniform sediments in water transfer culverts by combining multi-scale topological data analysis (TDA), nonlinear acoustic characteristics extraction and layered sedimentation model construction. The application of multi-scale TDA technology is a breakthrough in dealing with complex sedimentation structures. By constructing a persistence graph and calculating topological eigenvectors, the structural characteristics of sediments at different spatial scales can be captured, which is crucial for identifying stratification, inclusions or discontinuities in sediments. In particular, when dealing with complex sedimentation layers formed over a long period of time, it can reveal subtle structural changes that are difficult to detect with traditional acoustic analysis, such as bubble distribution in silt, changes in organic matter content or tiny geological faults, thereby providing key information for assessing the stability of sediments and the difficulty of dredging. The application of nonlinear acoustic characteristics extraction technology, especially the Volterra series expansion method, enhances the system's ability to identify the physical properties of sediments. By analyzing high-order harmonics and intermodulation effects, different types of sediments (such as clay, silt, sandy sediments) can be distinguished and their compaction degree and fluidity can be evaluated. The system can not only measure the thickness of siltation, but also provide detailed information on the properties of silt, providing a scientific basis for formulating accurate siltation strategies and evaluating the long-term operation risks of culverts. The construction of the layered siltation model solves the problem that the traditional single model is difficult to describe complex siltation structures by integrating topological features and nonlinear acoustic characteristics. The model uses a layered Bayesian network, which can not only accurately represent the vertical distribution of siltation, but also reflect the interaction and evolution trend between layers. This dynamic model is particularly suitable for monitoring long-term water transfer culverts. It can predict the siltation process, evaluate the siltation cycle, and even identify potential structural risk points, such as local erosion or abnormal sedimentation areas. This embodiment improves the underwater detection system's ability to analyze complex siltation environments, allowing engineering personnel to obtain more comprehensive and accurate siltation information, thereby formulating more targeted maintenance strategies, extending the service life of culverts, and improving the efficiency of water resource management.

[0106] In another embodiment of the present application, step S31 may also be implemented by using a topology network memory module, wherein the topology network memory module specifically includes:

[0107] The continuous coherence calculation layer, the enhanced sonar data set D''', calculates the continuous coherence features of the input data and outputs the continuous coherence features H;

[0108] The topological feature memory library reads the current continuous homology feature H and historical features, stores and retrieves historical topological features, and obtains the enhanced topological feature H'.

[0109] The differential topology comparator obtains the enhanced topological feature H' and the reference feature, compares the topological features at different time points, captures the structural changes, and outputs the topological change feature V;

[0110] The topological feature fusion device reads the continuous homology feature H, the enhanced topological feature H' and the topological change feature V, integrates all the above topological information, and generates a comprehensive topological feature T;

[0111] At work, the data processing flow is as follows:

[0112] Step S311, receiving the enhanced sonar data set D''' from the previous step.

[0113] Step S312, processing the enhanced sonar data set D''' through the continuous coherence calculation layer to generate a continuous coherence feature H;

[0114] Step S313, comparing the continuous homology feature H with the stored historical features through the topology feature memory library to generate an enhanced topology feature H';

[0115] Step S314, using a differential topology comparator to analyze the enhanced topology feature H' and the reference feature, identifying topology changes, and outputting a change feature V;

[0116] Step S315: The continuous homology feature H, the enhanced topology feature H' and the variation feature V are integrated by a topology feature fuser to generate a final integrated topology feature T, and the integrated topology feature T is output as the sediment topology feature.

[0117] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0118] S41. Based on the enhanced sonar data set, the continuous wavelet transform is used to obtain the time-frequency representation of the signal; based on the time-frequency representation, the time delay method is used to reconstruct the phase space;

[0119] S42. Based on the phase space, the acoustic characteristics of biological activities are calculated, including the local Lyapunov exponent, correlation dimension and Kolmogorov entropy;

[0120] S43, based on the acoustic signatures of biological activities and the enhanced sonar dataset, nonlinear Kalman filtering is used to estimate and compensate for signal disturbances caused by biological activities;

[0121] S44. Based on signal perturbations, a state-space model is constructed to obtain a sonar dataset compensated for biological activities.

[0122] In one embodiment of the present application, the enhanced sonar data set D''' is read, and the signal R of each frequency band is i'''Perform chaos analysis in the time-frequency domain to extract the acoustic features caused by biological activities. Use continuous wavelet transform (CWT) to obtain the time-frequency representation of the signal; calculate the local Lyapunov exponent LLE and phase space reconstruction parameters on the time-frequency plane. The specific steps are as follows: For each frequency band, the signal R i '''Perform continuous wavelet transform: W i (a,b) = ∫R i '''(t)ψ((tb) / a)dt, where ψ is the wavelet function, a is the scale parameter, and b is the translation parameter; select a region of interest (ROI) on the time-frequency plane, usually a high-energy region; for each ROI, reconstruct the phase space using the time delay method: X(t) = [x(t), x(t+τ), ..., x(t+(m-1)τ)], where τ is the time delay and m is the embedding dimension; use the mutual information method (MI) to estimate the optimal time delay τ opt ; Use the false nearest neighbor method (FNN) to determine the optimal embedding dimension m opt ; Calculate the local Lyapunov exponent LLE: LLE = 1 / Δt * ln(d(t+Δt) / d(t)), where d(t) is the distance between adjacent trajectories in the phase space, Δt represents the change in time or the time interval, and ln( ) represents the natural logarithm; extract other chaotic features, such as the correlation dimension D2 and the Kolmogorov entropy K; the final biological activity acoustic feature B is expressed as: B = {(LLE1, τ opt1 , m opt1 , D 21 , K1), ..., (LLE n , τ optn , m optn , D 2n , K n )}.

[0123] Based on the acoustic signature B of biological activities and the enhanced sonar dataset D''', an adaptive filter is constructed to remove the impact of biological activities. A nonlinear Kalman filter (NLKF) is used to estimate and compensate for signal disturbances caused by biological activities. The specific steps are as follows: Construct a state space model: x(k+1) = f(x(k)) + w(k); y(k) = h(x(k)) + v(k); where x is the system state, y is the observation value, f and h are nonlinear functions, w and v are process noise and observation noise; initialize the state estimate x'(0) and the error covariance matrix P(0); for each time step k: the prediction step is: x'*(k+1) = f(x'(k)); P*(k+1) = F(k)P(k)F(k) T+ Q(k), where F(k) is the Jacobian matrix of f, Q(k) is the process noise covariance, x'*(k+1) represents the predicted state estimate at time step k+1, and P*(k+1) represents the prediction error covariance matrix at time step k+1; the update steps are:

[0124] K(k+1) = P*(k+1)H(k+1) T [H(k+1)P*(k+1)H(k+1) T + R(k+1)] -1 ;

[0125] x'(k+1) = x'*(k+1) + K(k+1)[y(k+1) - h(x'*(k+1))];

[0126] P(k+1) = [I - K(k+1)H(k+1)]P*(k+1);

[0127] where H(k) is the Jacobian matrix of h, R(k) is the observed noise covariance, and I represents the identity matrix. The estimated state x'(k) is used to reconstruct the signal to obtain the biological activity compensated signal Ri''''. Finally, the biological activity compensated sonar data set D'''' is obtained: D'''' = {(S1, R1''''), (S2, R2''''), ..., (S n , R n '''')}.

[0128] This embodiment effectively solves the problem of the impact of biological activities on sonar signals in water culverts by introducing time-frequency domain chaos analysis and nonlinear Kalman filtering technology, and improves the accuracy and reliability of underwater detection. The application of time-frequency domain chaos analysis technology is a major breakthrough in dealing with small disturbances in complex dynamic systems. Through continuous wavelet transform and local Lyapunov exponent calculation, it is possible to accurately capture the changes in acoustic characteristics caused by aquatic biological activities (such as fish, plankton or microbial communities). This highly sensitive analysis can not only identify the direct interference of large organisms, but also detect the impact of subtle water quality changes caused by microbial activities on sound wave propagation. Especially in water culverts with obvious seasonal changes or sensitive ecological environments, it provides a key basis for distinguishing between biological activities and structural changes, avoiding false alarms or omissions caused by biological factors. The application of nonlinear Kalman filtering technology enhances the system's real-time compensation capability for dynamic interference. By constructing a nonlinear state space model that takes into account the characteristics of biological activities, it is possible to effectively filter out random disturbances caused by biological activities while retaining the culvert structure and siltation information. This enables the system to maintain high-precision detection capabilities even in areas with frequent biological activities (such as near the entrance or organic-rich siltation areas), expanding the scope of application and working conditions of the underwater detection system. Especially for long-term monitoring tasks, this adaptive filtering technology can automatically adjust parameters to adapt to the biological activity patterns of different seasons or diurnal cycles to ensure the consistency and comparability of the detection results. By combining the results of time-frequency domain chaos analysis with nonlinear Kalman filtering, it can not only compensate for the impact of biological activities, but also provide valuable ecological and environmental information. For example, by analyzing the spatiotemporal distribution of biological activity characteristics, the water quality status and ecosystem health in the culvert can be inferred, and even the potential risk of biological blockage can be warned. The acquisition of this multidimensional information provides a new perspective for culvert management, allowing maintenance strategies to consider both engineering and ecological factors. This embodiment improves the robustness and information value of the underwater detection system in a complex biological environment, and provides strong technical support for the safe operation and eco-friendly management of water transfer culverts.

[0129] like Figure 6 As shown, according to one aspect of the present application, step S5 is further:

[0130] S51. Based on the sonar data set after biological activity compensation, an overcomplete dictionary and a random measurement matrix are constructed to form an L1 optimization problem; the L1 optimization problem is solved using iterative reweighted least squares method to obtain sparse coefficients; based on the sparse coefficients, the signal is reconstructed to obtain preliminary reconstructed cross-section and siltation thickness data;

[0131] S52. Based on the layered sedimentation model, the initially reconstructed cross-section and sedimentation thickness data are optimized in a low-dimensional space using a combination method of local linear embedding and Laplace eigenmapping to obtain an optimized low-dimensional representation; the optimized low-dimensional representation is mapped back to a high-dimensional space to obtain optimized cross-section and sedimentation thickness data;

[0132] S53, based on the optimized cross-section and sediment thickness data, calling the pre-configured orthogonal polynomial basis function to calculate the expansion coefficient and statistical moment;

[0133] S54. Based on the expansion coefficient and statistical moment, confidence intervals are constructed to obtain cross-section and sediment thickness data with uncertainty.

[0134] In one embodiment of the present application, the biological activity compensated sonar dataset D'''' is read, and high-resolution image reconstruction is performed using compressed sensing theory and L1 optimization. A sparse representation basis and a measurement matrix are constructed, and then the L1 optimization problem is solved using iterative reweighted least squares (IRLS). The specific steps are as follows: For each frequency band, the signal R i '''', using a combination of discrete cosine transform (DCT) basis and wavelet basis to construct an overcomplete dictionary D i ; Construct a random measurement matrix Φ that satisfies the restricted isometry property (RIP); obtain the compressed measurement value y* = Φx, where x is the original signal; construct an L1 optimization problem: min ||α||1; st ||y* - ΦD i α||2≤ε, where α is the sparse coefficient vector and ε is the noise tolerance; use the IRLS algorithm to solve the optimization problem: Initialize the weight w j 0 = 1, t = 0; solve the weighted least squares problem:

[0135] α t+1 = argminΣ(w j t α j 2 ); st || y *- ΦD i α||2≤ε;

[0136] where α t+1 represents the sparse coefficient vector obtained in the t+1th iteration, argmin represents the parameter that minimizes the objective function, α j Represents the jth element in the sparse coefficient vector; update weight: w j t+1 = 1 / (|α j t+1| +Δ), where Δ is a small positive number; if it does not converge, t = t + 1, return to the previous step and continue iterative update; reconstruct the signal: x recon = D i α*, where α* is the optimal sparse coefficient; the preliminary reconstructed cross section and sediment thickness data R are obtained: R = {x recon1 , x recon2 , ..., x reconn}.

[0137] Based on the preliminary reconstruction data R and the layered sedimentation model parameters L, a topology-preserving manifold learning algorithm is applied to optimize the reconstruction of cross-section and sedimentation thickness, using a combination of local linear embedding (LLE) and Laplace eigenmap (LE). The specific steps are as follows: Each reconstruction data x in the preliminary reconstruction of cross-section and sedimentation thickness data R is converted into reconi Treat it as a point in high-dimensional space; for each point, find its k nearest neighbors (use kd tree algorithm to speed up the search); calculate the local reconstruction weight W: min Σ||x i -Σ j (W ij * x j )|| 2 ; st Σ j (W ij ) = 1, where x i represents the i-th reconstructed data point in the preliminary reconstructed cross section and sediment thickness data R, W ij Denotes the local reconstruction weight, which is the weight between the i-th point and the j-th point in its k-nearest neighbors; constructs the Laplace matrix L = D - W, where D is the degree matrix; solves the generalized eigenvalue problem: Ly = λDy, where λ represents the eigenvalue, Ly represents the result of the Laplace matrix acting on the eigenvector y, and Dy represents the result of the degree matrix acting on the eigenvector y; selects the eigenvectors corresponding to the smallest d non-zero eigenvalues ​​to form a low-dimensional embedding Y; applies the layered sedimentation model L in the low-dimensional space for optimization: Y opt = argmin E(Y, L), where E is the energy function, considering data consistency and model constraints; the optimized low-dimensional representation Y opt Mapping back to the high-dimensional space, we can get the optimized cross-section and sedimentation thickness data R'; finally, we can get the optimized cross-section and sedimentation thickness data R': R' = {x opt1 , x opt2 , ..., x optn}.

[0138] The uncertainty of the optimized data R' is quantified using the polynomial chaos expansion (PCE) method. The specific steps are as follows: Based on the probability distribution of the input parameters (such as Hermite polynomials for Gaussian distribution), select the orthogonal polynomial basis function Ψ i (ξ); the optimized data is expressed as a polynomial expansion: x opt =Σ i (c i *Ψ i (ξ)), where c i is the expansion coefficient, ξ is a random variable; the expansion coefficient c is calculated using the projection method or regression method i :c i =E[x opt *Ψ i (ξ)] / E[Ψ i 2 (ξ)]; calculate the statistical moment: mean is μ = c0; variance is σ 2 = Σ i>0 (c i 2 * E[Ψ i 2 (ξ)]). Construct the confidence interval: CI = [μ-k*σ, μ+k*σ], where k is the coefficient corresponding to the confidence level; obtain the cross-section and sediment thickness data R'' with uncertainty estimation: R'' = {(x opt1 , CI1), (x opt2 , CI2), ..., (x optn , CI n )}.

[0139] This embodiment combines the compressed sensing reconstruction algorithm, topology-preserving manifold learning and polynomial chaos expansion method to achieve high-precision reconstruction and uncertainty quantification of the cross-section and siltation thickness of the water conveyance culvert, thereby improving the accuracy, reliability and interpretability of the underwater detection results. The application of the compressed sensing reconstruction algorithm is a major breakthrough in reconstructing high-resolution images from limited sampling data. By constructing an overcomplete dictionary and a random measurement matrix, combined with L1 optimization and iterative reweighted least squares, high-resolution culvert cross-sections and siltation thickness distribution maps can be restored from sparse sonar data. Not only does it greatly improve the accuracy of image reconstruction, it also reduces the amount of data collection required, thereby shortening the detection time and reducing the equipment load. Especially when dealing with large or long-distance water conveyance culverts, it can provide comprehensive and detailed structural information within a limited detection time, providing engineers with richer visual basis for evaluating the culvert condition. The introduction of topology-preserving manifold learning technology solves the problem of information loss in the process of dimensionality reduction and reconstruction of complex nonlinear structures. By combining local linear embedding and Laplace eigenmap, the key topological features of culvert structure and siltation distribution can be retained in low-dimensional space, so that complex geometric shapes and material distribution can be restored more accurately during the reconstruction process. It is suitable for processing water transfer culverts with irregular shapes, branches or local deformations, and can accurately capture minor structural changes and siltation anomalies, providing the possibility for early detection of potential problems. The application of polynomial chaos expansion method provides reliable uncertainty quantification for reconstruction results. By expressing system parameters and measurement errors as random variables and expanding them using orthogonal polynomial basis functions, the statistical moments and confidence intervals of the reconstruction results can be efficiently calculated. This uncertainty analysis not only improves the credibility of the detection results, but also provides a quantitative basis for risk assessment and decision making. For example, engineers can formulate targeted further inspection or maintenance plans and optimize resource allocation based on the reconstruction accuracy and uncertainty level of different areas. This embodiment improves the performance and practicality of the underwater detection system. High-precision cross-section reconstruction and siltation thickness measurement provide a more reliable basis for culvert structural assessment, siltation monitoring and maintenance planning, while uncertainty quantification provides a scientific basis for risk management and decision support, thereby comprehensively improving the management level and operational safety of water transfer culverts.

[0140] In another embodiment of the present application, step S51 may also be implemented using an adaptive CS network module, wherein the adaptive CS network module includes:

[0141] Dynamic sparse representation layer: obtain the sonar dataset D'''' after biological activity compensation, adaptively adjust the sparse representation basis according to the characteristics of the input signal, and output the adaptive sparse representation S.

[0142] Multi-agent collaborative reconstruction network, obtain adaptive sparse representation S, use multiple specialized neural network agents to collaborate to complete the reconstruction task, and obtain the preliminary reconstructed cross section and sediment thickness data R_init;

[0143] Progressive refinement module: Based on the preliminary reconstruction data R_init and the original data D'''', the reconstruction quality is gradually improved through an iterative refinement process to obtain the optimized section and siltation thickness data R.

[0144] When working, the specific workflow is as follows:

[0145] Step S511, receiving the sonar data set D'''' after biological activity compensation.

[0146] Step S512, analyzing the biological activity compensated sonar data set D'''' through a dynamic sparse representation layer to generate an adaptive sparse representation S;

[0147] Step S513, based on the multi-agent collaborative reconstruction network processing adaptive sparse representation S, each agent is responsible for each aspect of reconstruction, and jointly generates preliminary reconstruction data R_init;

[0148] Step S514, compare the preliminary reconstructed data R_init with the original data (sonar data set after biological activity compensation) D'''' through the progressive refinement module, gradually refine and optimize the reconstruction results; output the final optimized cross-section and siltation thickness data R.

[0149] like Figure 7 As shown, according to one aspect of the present application, step S6 is further:

[0150] S61, based on the uncertain cross-section and sediment thickness data, initialize a uniform grid, and calculate a data complexity index of each grid cell in the uniform grid; based on the data complexity index, subdivide or merge the grid cells to obtain a final adaptive grid;

[0151] S62. Based on the adaptive grid and the uncertain cross-section and siltation thickness data, set the virtual camera parameters and generate the main ray for each pixel of the virtual camera; calculate the intersection sequence of the main ray and the adaptive grid; based on the intersection sequence, use tone mapping and post-processing effects to generate the final culvert cross-section map and siltation thickness distribution map.

[0152] In one embodiment of the present application, based on the data R'' with uncertainty estimation, an adaptive grid is generated to optimize the visualization effect. The grid is refined using a quadtree (2D) or octree (3D) algorithm. The specific steps are as follows: Initialize a uniform grid G0 to cover the entire culvert section area; For each grid cell e i, calculate the data complexity index C i :C i = f(▽x opt , γ), where ▽x opt is the local gradient, γ is the local uncertainty; if the data complexity index C i Greater than the preset threshold T c , then subdivide the grid unit; repeat the above steps for the newly generated sub-units until the maximum refinement level or the data complexity index C of all units is reached i are all less than the preset threshold T c ; Merge adjacent low-complexity cells to reduce redundancy and obtain the adaptive grid parameters G: G = {e1, e2, ..., e m}, where e i It is a grid unit, containing position and size information.

[0153] Use adaptive mesh parameters G and data R'' with uncertainty estimation to perform high-fidelity visualization rendering. Use ray tracing and volume rendering technology combined with physical based rendering (PBR). The specific steps are as follows: Set the virtual camera parameters: position, direction, field of view (FOV); for each pixel p ij :Generate the main ray r ij ; Calculate the intersection sequence {t1, t2, ..., t k}; For each interval [t i , t i+1 ]: sampling data x opt and uncertainty CI, calculate the local scattering coefficient σ s and the absorption coefficient σ a , using the rendering equation to calculate the radiation transfer: L(x, ω) = ∫ Ω (f(x, ω' ∽ω)L i (x, ω')dω') + L e (x, ω), where Ω represents the integration domain, ω' represents the unit vector in the incident direction, x represents the position in space, ω represents the unit vector in the outgoing direction, and L i (x, ω') represents the radiant brightness incident from direction ω' to position x, L e (x, ω) represents the spontaneous radiation brightness along the direction ω at position x, and ∽ represents the change or conversion of direction; the contribution of all intervals is accumulated to obtain the pixel color C ij ; Apply tone mapping and post-processing effects (such as depth of field, ambient occlusion). Finally, the culvert cross-section map I1 and the siltation thickness distribution map I2 are generated: I1 = {C ij | i, j∈section image range}; I2 = {C ij|i, j∈deposition thickness distribution image range}.

[0154] This embodiment combines adaptive grid generation technology and high-fidelity visualization rendering methods to achieve high-quality and efficient visualization of the cross-section and siltation thickness distribution of the water culvert, and enhances the intuitiveness, readability and practicality of the detection results. The application of adaptive grid generation technology is a breakthrough in optimizing the visualization of large-scale complex data. By using quadtree or octree algorithms, the grid density can be dynamically adjusted according to the complexity of the data, while ensuring the clarity of key details, reducing the amount of redundant data processing, and is suitable for processing complex geometric structures and uneven siltation distribution in water culverts. For example, in culvert corners, joints or abnormal siltation areas, the system automatically increases the grid density to capture more details; while in areas with simple structures or uniform siltation, coarser grids are used to optimize the allocation of computing resources. Not only does it improve the efficiency of visualization, but it can also achieve real-time rendering of large-scale culverts under limited computing resources, providing strong technical support for on-site detection and remote monitoring. The introduction of high-fidelity visualization rendering technology, especially the combination of ray tracing and volume rendering technology, improves the realism and information content of the visualization results. By simulating the propagation and scattering of light in different materials, a near-photo-quality view of the interior of the culvert can be generated, including the texture of the sediment layer, thickness changes, and possible structural defects. This highly realistic visualization not only helps engineers intuitively understand the condition of the culvert, but also provides easy-to-understand visual reports for non-professionals (such as managers or decision makers). Especially when dealing with complex sediment structures accumulated over a long period of time, volume rendering technology can clearly show the internal stratification and density changes of sediment, providing an intuitive basis for assessing the difficulty of dredging and formulating dredging strategies. In addition, the application of physically based rendering (PBR) further enhances the authenticity and credibility of visualization. By accurately simulating the reflection, scattering, and absorption properties of materials, PBR technology can accurately reproduce the visual characteristics of different types of sediment (such as clay, sand, or organic matter), helping experts quickly identify and classify sediment composition. This is of great significance for assessing the source of sediment, predicting future sedimentation trends, and formulating long-term management strategies. By integrating uncertainty visualization technology, the system can intuitively display the confidence distribution of the reconstruction results through color mapping or transparency changes. This enables engineers to quickly identify areas that require further verification or focus, improving the pertinence and efficiency of inspection and maintenance. For example, in the visualization results, high uncertainty areas may indicate sensor blind spots, poor data quality, or abnormal structures, and this information is critical for optimizing future inspection plans and formulating maintenance plans. This embodiment not only improves the visualization quality of underwater inspection results, but also enhances the interpretability and practical value of the data. By providing intuitive, detailed, and informative visualization results, the system provides a powerful support tool for status assessment, risk management, and maintenance decisions of water transfer culverts, thereby promoting the management of the entire water conservancy infrastructure to a more refined and intelligent direction.This advanced visualization capability not only helps improve the efficiency of daily management, but also provides fast and accurate situation assessment in emergency situations (such as sudden leakage or structural damage), providing critical support for timely decision-making and intervention.

[0155] According to another aspect of the present application, a system for underwater detection and simulation of water conveyance culvert cross-section and siltation thickness includes:

[0156] at least one processor; and,

[0157] a memory communicatively connected to at least one of the processors; wherein,

[0158] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method of underwater detection and simulation of water conveyance culvert cross-section and siltation thickness described in any of the above embodiments.

[0159] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for underwater detection and simulation of water conveyance culvert cross-section and siltation thickness, characterized in that: The steps include: S1. Obtain culvert environmental parameters, use an adaptive frequency optimization algorithm to generate an optimized sonar array configuration; based on the sonar array configuration, simultaneously transmit and receive sonar signals of predetermined frequency bands to form an original multi-band sonar data set; The original multi-band sonar data set is subjected to denoising to obtain a denoised multi-band sonar data set; the culvert environmental parameters include estimated size data and water quality data; S2, based on the multi-band sonar dataset after noise reduction, a dynamic time reversal mirror algorithm is used to compensate for multipath effects, thereby obtaining a sonar dataset after multipath effect compensation; Based on the sonar data set after multipath effect compensation, a fractal geometry acoustic model is constructed, and the sonar data set is enhanced and reconstructed using the fractal geometry acoustic model to obtain an enhanced sonar data set; S3, perform multi-scale topological data analysis on the enhanced sonar data set to extract the topological features of the sediment; Based on the topological characteristics of sediment, the Volterra series expansion method is used to extract nonlinear acoustic characteristics. The layered sedimentation model is constructed by combining the topological characteristics and nonlinear acoustic characteristics. S4, perform time-frequency domain chaos analysis on the enhanced sonar data set to extract acoustic features caused by biological activities; Based on the acoustic features, an adaptive filter is constructed to remove the influence of biological activities and obtain a sonar data set compensated for biological activities. S5. Based on the sonar data set after biological activity compensation, high-resolution image reconstruction was performed using compressed sensing theory and L1 optimization to obtain preliminary reconstruction data; based on the layered sedimentation model, the preliminary reconstruction data was optimized using the topology-preserving manifold learning algorithm to obtain optimized reconstruction data; the polynomial chaos expansion method was used to quantify the uncertainty of the optimized reconstruction data to obtain the cross-section and sedimentation thickness data with uncertainty; S6. Generate adaptive mesh based on cross-section and sediment thickness data with uncertainty; Based on adaptive mesh, ray tracing and volume rendering technology are used to perform high-fidelity visualization rendering to generate culvert cross-section diagrams and silt thickness distribution maps; Step S1 is further as follows: S11, extracting culvert environmental parameters from a pre-stored database, using an adaptive frequency optimization algorithm to generate an initial frequency range and array layout scheme, and calculating the optimal sonar array configuration, including an optimized frequency set and array layout, through an iterative optimization algorithm; S12, based on the sonar array configuration, control the sonar array to simultaneously transmit sonar signals of predetermined frequency bands, record the echo signals received in each frequency band, use synchronous sampling technology to ensure the time alignment of the signals in each frequency band, and combine the sonar signals transmitted in all frequency bands and the echo signals received into an original multi-band sonar data set; S13, based on the original multi-band sonar data set, wavelet packet transform is used to obtain the time-frequency domain representation; Using an adaptive threshold method to perform denoising on the time-frequency domain representation to obtain a denoised time-frequency domain representation; S14, converting the denoised time-frequency domain representation back to the time domain using an inverse wavelet packet transform to obtain a denoised signal; combining the denoised signals of all frequency bands into a denoised multi-band sonar data set; Step S2 is further as follows: S21, time-reversing the received signal in the multi-band sonar data set after noise reduction to obtain the time-reversed received signal; based on the time-reversed received signal, calculating the cross-correlation function between the received signal and the transmitted signal in the multi-band sonar data set after noise reduction; S22, estimating the sound field propagation function based on the peak position of the cross-correlation function; and obtaining the sonar data set after multipath effect compensation through a deconvolution process based on the sound field propagation function; S23, based on the sonar data set after multipath effect compensation, use multi-resolution analysis method to generate components of different scales; The fractal dimension and Hurst index of each scale component are calculated, and based on the fractal dimension and Hurst index, a fractal model describing the acoustic propagation characteristics in the culvert is constructed. S24, based on the sonar data set after multipath effect compensation, using a fractal model to perform deconvolution to obtain a deconvolved sonar data set; using a subspace projection method to perform noise suppression and feature extraction on the deconvolved sonar data set to obtain an enhanced sonar data set; Step S3 is further as follows: S31. Based on the enhanced sonar data set, the persistence graph of each signal is constructed using the persistence coherence algorithm; the topological feature vector of each persistence graph is calculated, including the Betti number sequence and persistence entropy; the topological feature vectors of all frequency bands are combined using the kernel method to obtain the topological characteristics of the sediment; S32, based on the topological characteristics of sediment and the enhanced sonar data set, the Volterra series expansion method is used to construct a nonlinear system model and estimate the high-order kernel function; Based on the high-order kernel function, nonlinear metrics are calculated, including harmonic distortion and intermodulation distortion; Based on nonlinear metrics, nonlinear acoustic features are obtained through principal component analysis dimensionality reduction method; S33. Based on the topological characteristics and nonlinear acoustic characteristics of sediments, a hierarchical Bayesian network is used to construct a layered sedimentation model framework, and the expectation maximization algorithm is used to estimate the parameters of the layered sedimentation model, and finally a layered sedimentation model is formed; Step S4 is further as follows: S41. Based on the enhanced sonar data set, the continuous wavelet transform is used to obtain the time-frequency representation of the signal; based on the time-frequency representation, the time delay method is used to reconstruct the phase space; S42. Based on the phase space, the acoustic characteristics of biological activities are calculated, including the local Lyapunov exponent, correlation dimension and Kolmogorov entropy; S43, based on the acoustic signatures of biological activities and the enhanced sonar dataset, nonlinear Kalman filtering is used to estimate and compensate for signal disturbances caused by biological activities; S44. Based on the signal disturbance, a state space model is constructed to obtain a sonar data set after biological activity compensation; Step S5 is further as follows: S51, based on the biological activity compensated sonar data set, construct an overcomplete dictionary and a random measurement matrix to form an L1 optimization problem; The sparse coefficients are obtained by solving the L1 optimization problem using iterative reweighted least squares method; Based on the sparse coefficient, the signal is reconstructed to obtain the preliminary reconstructed cross section and sediment thickness data; S52. Based on the layered sedimentation model, the initially reconstructed cross-section and sedimentation thickness data are optimized in a low-dimensional space using a combination method of local linear embedding and Laplace eigenmapping to obtain an optimized low-dimensional representation; the optimized low-dimensional representation is mapped back to a high-dimensional space to obtain optimized cross-section and sedimentation thickness data; S53, based on the optimized cross-section and sediment thickness data, calling the pre-configured orthogonal polynomial basis function to calculate the expansion coefficient and statistical moment; S54. Based on the expansion coefficient and statistical moment, the confidence interval is constructed to obtain the cross-section and sediment thickness data with uncertainty; Step S6 is further as follows: S61, based on the uncertain cross-section and sediment thickness data, initialize a uniform grid, and calculate a data complexity index of each grid cell in the uniform grid; based on the data complexity index, subdivide or merge the grid cells to obtain a final adaptive grid; S62, based on the adaptive grid and the cross-section and siltation thickness data with uncertainty, setting virtual camera parameters, and generating a main ray for each pixel of the virtual camera; calculating a sequence of intersection points between the main ray and the adaptive grid; Based on the intersection sequence, tone mapping and post-processing effects are used to generate the final culvert cross-section map and silt thickness distribution map.

2. The method for underwater detection and simulation of water conveyance culvert cross-section and siltation thickness according to claim 1, characterized in that: Step S12 also includes: S12a, using a real-time signal quality evaluation algorithm to monitor the quality of the sonar signal and the echo signal, and obtaining a monitoring quality result; S12b. Determine whether the monitoring quality result is lower than a preset threshold. If so, re-collect the signal of the frequency band; otherwise, do nothing.

3. The method for underwater detection and simulation of water conveyance culvert cross section and siltation thickness according to claim 1, characterized in that: Step S31 is further as follows: S311, constructing a time delay embedding based on the enhanced sonar data set to obtain point cloud data; S312, based on point cloud data, using the Vietoris-Rips complex VR algorithm to construct simple complex sequences at different scales; S313. Based on the simplicial complex sequences at different scales, 0-dimensional and 1-dimensional persistence homology are calculated to obtain persistence graphs; at the same time, the Betti number sequence is calculated; S314, based on the persistence graph and the Betti number sequence, the persistence entropy is calculated, and the topological feature vector of each frequency band is constructed; S315. Use the Gaussian kernel function to fuse the topological feature vectors of all frequency bands to obtain the topological features of the sediment.

4. A system for underwater detection and simulation of water conveyance culvert cross-section and siltation thickness, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method for underwater detection and simulation of water transfer culvert cross-section and siltation thickness as described in any one of claims 1 to 3.