A method and system for underwater object detection and imaging based on SBA

Abstract

The application provides an SBA-based underwater object detection and imaging method and system, wherein the method comprises the following steps: acquiring original sonar data of an underwater array element; constructing an underwater multi-array element model based on the original sonar data and performing filtering and denoising; constructing an underwater beam signal model through signal minimum variance beamforming and acoustic wave beamforming signal processing; optimizing the beam signal model, wherein the optimization comprises signal beamforming convolutional neural network model optimization and underwater acoustic wave particle filtering operation mask optimization, so as to obtain an optimized underwater sonar signal; constructing a two-dimensional image of an underwater object based on the optimized underwater sonar signal by using an SBA algorithm; and generating a three-dimensional image of the underwater object according to the two-dimensional image. The application realizes the improvement of the detection capability and efficiency of the ocean underwater sonar by constructing an acoustic wave beam mathematical model and function operation through innovation on the basis of the beam signal of the acoustic wave emitted and fed back by the ocean sonar equipment.

Description

Technical Field

[0001] This invention relates to the field of underwater detection and imaging technology, specifically to an underwater object detection and imaging method and system based on SBA. Background Technology

[0002] With the rapid development of marine resource development, scientific research, and fisheries management, the demand for high-precision detection and clear imaging of underwater targets is becoming increasingly urgent. Sonar technology, as a primary means of underwater detection, relies on utilizing the propagation characteristics of sound waves in water to detect, locate, and image underwater targets.

[0003] Traditional sonar signal processing methods mostly employ delay techniques based on fixed beamforming. While simple to implement, these methods have significant limitations in complex underwater environments: First, for weak target signals under multipath effects and strong noise interference, traditional methods struggle to achieve effective focusing and noise suppression. Second, conventional filtering algorithms lack adaptability in scenarios with dynamically changing target numbers, making it impossible to accurately track the motion of multiple targets. Furthermore, existing imaging algorithms often suffer from low resolution, poor contrast, and loss of detail when processing underwater images, making it difficult to meet the demands of high-precision detection. Summary of the Invention

[0004] This invention provides an underwater object detection and imaging method and system based on SBA (Sound Absorption Array), which uses an innovatively constructed mathematical model and functional calculations to enhance the detection capability and efficiency of marine underwater sonar by focusing the acoustic wave signals emitted and retrieved by marine sonar equipment.

[0005] Firstly, this application provides an underwater object detection and imaging method based on SBA, including the following steps:

[0006] Acquire raw sonar data from underwater array elements;

[0007] An underwater multi-element model is constructed based on the original sonar data, and filtering and denoising are performed.

[0008] An underwater bundled signal model is constructed by signal minimum variance bundle and acoustic wave bundle signal processing.

[0009] The clustered signal model is optimized, including optimization of the signal clustered convolutional neural network model and optimization of the underwater acoustic particle filter manipulation mask, to obtain the optimized underwater sonar signal.

[0010] Two-dimensional images of underwater objects are constructed using the SBA algorithm based on optimized underwater sonar signals;

[0011] Based on the two-dimensional image, a three-dimensional image of the underwater object is generated by constructing an SBA underwater three-dimensional image space and optimizing the SBA underwater three-dimensional image structure.

[0012] Secondly, embodiments of this specification provide an underwater object detection and imaging system based on SBA, comprising:

[0013] The data acquisition module acquires the raw sonar data of the underwater array elements;

[0014] The preprocessing module constructs an underwater multi-element model based on the original sonar data and performs filtering and noise reduction.

[0015] The model building module constructs an underwater bundled signal model through minimum variance bundled signal processing and acoustic bundled signal processing.

[0016] The model optimization module optimizes the clustered signal model, including signal clustered convolutional neural network model optimization and underwater acoustic particle filter manipulation mask optimization, to obtain the optimized underwater sonar signal.

[0017] The first image generation module uses the SBA algorithm to construct a two-dimensional image of an underwater object based on the optimized underwater sonar signal.

[0018] The second image generation module generates a three-dimensional image of an underwater object based on the two-dimensional image by constructing an SBA underwater three-dimensional image space and optimizing the SBA underwater three-dimensional image structure.

[0019] The beneficial effects of the technical solutions provided in this specification include at least the following:

[0020] This specification presents an underwater object detection and imaging method based on SBA, constructing a complete and collaborative processing scheme from raw data to 3D images. By constructing an underwater multi-element model and performing filtering and denoising, the quality of the raw data is effectively improved, laying a solid foundation for subsequent processing. Through signal minimum variance beamforming and acoustic wave beamforming signal processing steps, a beamform signal with excellent directionality and strong anti-interference capability is formed, ensuring the accuracy of detection from the source. By optimizing the signal beamforming convolutional neural network model and optimizing the underwater acoustic wave particle filter manipulation mask, the feature extraction capability of deep learning is combined with the parameter search capability of intelligent optimization, enabling the system to automatically adapt to complex acoustic environments. This effectively overcomes the limitations of traditional algorithms in target recognition and noise suppression, achieving a leap from fixed algorithm processing to intelligent adaptive optimization. The resolution and fidelity of the 2D and 3D images generated by the SBA algorithm from the optimized signal are fundamentally improved. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of an application scenario for marine sonar detection and imaging provided in the embodiments of this specification.

[0023] Figure 2 This is a schematic diagram of the sonar ocean detection acoustic wave clustering algorithm model provided in the embodiments of this specification.

[0024] Figure 3 This is a schematic diagram illustrating the relationship between SBA time shift and acoustic beam oscillation direction provided in the embodiments of this specification.

[0025] Figure 4 This is a schematic flowchart of an underwater object detection and imaging method based on SBA provided in the embodiments of this specification.

[0026] Figure 5 This is a schematic diagram of an underwater object detection and imaging system based on SBA, provided as an embodiment of this specification. Detailed Implementation

[0027] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of this specification and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of this specification.

[0028] The terms "first," "second," "third," etc., in the description, claims, and accompanying drawings are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0029] Before this specification describes in detail an SBA-based underwater object detection and imaging method with reference to one or more embodiments, it first introduces some application scenarios of this SBA-based underwater object detection and imaging method.

[0030] Please see Figure 1 and Figure 2 Sonar detection is a technology that utilizes the underwater propagation characteristics of sound waves to detect the location, shape, and motion of targets. It is widely used in marine exploration and underwater navigation. Its core principle is based on the emission, propagation, reflection, and reception of sound waves. Sonar equipment, whether side-mounted or otherwise, is easily obstructed by waves and other factors when emitting waves underwater, scattering the sonar detection ripples and reducing the detection range and imaging capability. The Sound Wave Beamforming Algorithm (SBA) is constructed through mathematical design and functional formulas. It uses SBA to capture and converge the scattered signals generated by the resistance of ocean waves and other factors during the sonar's detection process, thus better concentrating the sonar's sound wave ripples and forming a stronger sound wave beam penetration capability. This increases the detection range, improves the imaging clarity, and greatly enhances the detection capability of sonar equipment, which is especially important when large sonar equipment cannot be installed on ships. Therefore, an innovative Sound Wave Beamforming Algorithm (SBA) for improving the marine sonar detection capability has been designed and invented. Its application in marine exploration improves the quality of underwater sonar detection signals and image formation. Please refer to [link / reference]. Figure 3 The expression for the acoustic beam model is as follows: Sn(t) = An(t) + Nn(t) describes the total signal received by the sonar system at time t, which consists of two parts: the useful signal An(t) reflected by the underwater object and the background noise and interference signal Nn(t). This model is the foundation for sonar system design and signal processing algorithm development and plays an important role.

[0031] Please see Figure 4 , Figure 4 This is a flowchart illustrating an underwater object detection and imaging method based on SBA provided in an embodiment of the present invention. This SBA-based underwater object detection and imaging method may include at least the following steps:

[0032] S101: Acquire raw sonar data from N underwater array elements, wherein the raw sonar data includes at least one of audio data and signal parameters;

[0033] S102: Preprocess the raw sonar data, the preprocessing including: constructing an underwater multi-element model based on the raw sonar data and performing filtering and noise reduction;

[0034] In this embodiment, constructing an underwater multi-element model based on the original sonar data includes:

[0035] In this embodiment, the multi-element linear array is composed of a series of uniformly spaced elements. The arrangement of the multi-element linear array is based on the geometric arrangement of equally spaced linear arrays and signal processing requirements. The relationship between element spacing and array length is as follows: in underwater multi-element linear array sonar, the arrangement is centered on a linear distribution. Assuming the spacing between each underwater element is d, there are a total of N underwater elements. The total length L from the first underwater element to the last underwater element is the sum of N-1 spacings. Therefore, the total length of the multi-element linear array... In the formula, N represents the total number of array elements, and d represents the preset spacing between adjacent array elements. Equally spaced linear arrays simplify the calculation of the acoustic beamforming function, making it easier to control the directionality of the acoustic beamforming signal. (Calculation formula) This ensures a fixed relative positional relationship between underwater array elements, facilitating precise pointing and directional control of the acoustic beamforming signal. It's important to note that in sonar signal processing, to achieve the formation of an underwater acoustic beamforming signal, the phase difference of the beamforming signal across different underwater array elements must be calculated. For equally spaced linear arrays, since the phase difference between adjacent underwater array elements is fixed, the signals can achieve in-phase superposition in a specific direction. Furthermore, in sonar detection, the phase and amplitude of the acoustic beamforming signal from each underwater array element are calculated and adjusted to form an underwater acoustic beamforming signal pointing in a specific direction. The equally spaced arrangement ensures that the propagation delay and phase difference of the signal across different underwater array elements are regular, which helps improve the efficiency of signal processing.

[0036] The signals received by the underwater array elements are weighted and summed, and the integrated signal after weighted summation is output. The calculation formula is as follows: This calculation formula adjusts the weights of each underwater array element to coherently superimpose the signals received by different underwater array elements in a specific direction, thereby enhancing the signal in that direction. This weighted summation can achieve spatial filtering, improving the signal-to-noise ratio and resolution for underwater object detection. In the formula, Output(t) represents the integrated output signal at time t, which is the acoustic signal reflected from the underwater object. n represents the total number of array elements. This represents the weighting coefficient of the i-th underwater array element, used to adjust the contribution ratio of the underwater array element's signal to the total output. The weight can be positive, negative, or zero, depending on the design of the calculation function. Let represent the signal received by the i-th underwater array element at time t. Specifically, the underwater acoustic beam signal is first acquired. The underwater sonar system consists of N underwater array elements, and each of the N underwater array elements independently receives the acoustic beam signal from underwater. The signal received by the i-th underwater array element at time t is denoted as . The signal of each underwater array element is assigned a weight during integration. The weights can be determined based on factors such as the sensitivity, position, and orientation of the underwater array elements, and can also be dynamically adjusted according to specific circumstances to suit different application scenarios of various underwater objects, thereby further improving the signal-to-noise ratio. This applies to the signals of N underwater array elements. Multiply by its corresponding weight The weighted signal is obtained. The entire weighted acoustic beam signal is summed to obtain the integrated output signal Output(t). This process involves weighted averaging of signals from different underwater array elements. The summed integrated signal has a higher signal-to-noise ratio than the signal of a single underwater array element, which can further improve the effective detection and confirmation of various underwater objects and enhance the effective signal of underwater objects.

[0037] In this embodiment, the filtering and denoising steps include: filtering based on the probability density function of the target quantity, calculated as follows: In the formula, represents Predict the probability density function of the number of n targets at time k. This represents the filtered survival probability of the target at time k-1; This represents the filtered probability density function indicating the existence of n-1 targets at time k-1. Specifically, it is calculated by summing... The filter integrates the contributions of all objectives to the current time step, where each weight... This represents the impact of the i-th target on the number of targets at time k. It should be noted that the implicit assumption includes the survival of n-1 targets at time k-1 through filtering, evolving into n targets at time k. In some embodiments, k-1(n-1) represents the probability of having n-1 targets at the previous time, and a multiplication operation is performed, multiplying the weighted sum by the historical potential distribution to reflect the filtering evolution relationship from n-1 targets at the previous time to n targets at the current time. In the filtering survival scenario, if each target has a probability of survival... If they continue to exist, the number of targets n at time k consists of the n-1 surviving targets at time k-1, plus one newly added target. This newly added target is formed by target splitting or the appearance of a new target. In the filtered multiplication scenario, if target i is weighted... If multiple sub-objectives are generated, where these sub-objectives are achieved through division or reproduction, then the summation operation integrates all filtered proliferation paths. It should be noted that this calculation does not explicitly include the new object birth model; it is assumed that the new object has already been filtered and integrated into the weights. This method only supports filter increments from n-1 to n, making it suitable for application scenarios where the number of underwater objects changes gradually. The calculation formula, through weighted summation and correlation with historical potential distribution, achieves recursive prediction of the number of multiple targets. Its core is to utilize weighted compensation based on the survival or proliferation of target filters to integrate the dispersed target contributions into a global probability distribution. This approach offers advantages such as computational efficiency and ease of implementation in low-dynamic, low-target scenarios.

[0038] The signals from N underwater array elements are weighted and summed for denoising to output the denoised underwater signal. The calculation formula is as follows: In the formula: y(t) is the underwater denoised signal, wi represents the weight of the i-th underwater array element, s(t) represents the underwater object signal received by the i-th underwater array element at time t, and n(t) represents noise, which includes background noise and other interference. Specifically, first, the total number of underwater array elements N is determined. At time t, the signal of the underwater object received by the i-th underwater array element is denoted as s(t). Each underwater array element's signal is assigned a weight wi during integration. It should be noted that the weight selection is determined based on factors such as the sensitivity, position, and orientation of the underwater array element. It also possesses dynamic and predictive properties, dynamically calculating and adjusting values ​​based on real-time signal data. Multiplying the signal s(t) of each underwater array element by its corresponding weight wi yields the weighted signal. The entire weighted signal is summed to obtain the integrated output signal. This integrated signal typically has a higher signal-to-noise ratio than the signal from a single underwater element, further enhancing the ability to identify underwater objects. The integrated signal can also be used for further processing and noise reduction.

[0039] Furthermore, the underwater object detection and imaging method based on SBA also includes analysis and optimization of the underwater post-signal time domain. The core function of underwater post-signal time domain signal analysis and optimization is to integrate the independent signal processing time-frequency analysis and detection steps into a coherent process. The specific steps are as follows:

[0040] The time-domain signal is mapped to a new domain containing azimuth or time delay information through kernel function transformation, resulting in the post-transformed signal. The calculation formula is as follows: In the formula, The physical meaning of the post-transformed signal is determined by the kernel function, such as beam output, time-frequency distribution, and azimuth spectrum, which can serve as the target for further underwater object detection using sonar. The kernel function represents the transformation weights of the signal from the time domain t to the new domain u, including time delay and weighting. It determines the weights of the original signal x(t) at different times t for the output signal. The contribution weight. x(t) is a time-domain window function, which is the back-time domain acoustic beam signal of the underwater object received by the sonar equipment. It includes the echo signal reflected by the underwater object and information such as environmental noise. Due to the complexity of the underwater environment, the back-time acoustic beam signal is also affected by attenuation, scattering, multipath effects, etc. during propagation. It needs to be processed by SBA calculation formula to obtain effective data information. x(t) is processed by SBA function calculation, which reduces the dependence on high-performance sonar hardware, thereby reducing the loss of data resources for function calculation formula. dt represents the time-frequency joint distribution and is used to detect transient signals. It should be noted that in sonar detection, u represents different physical quantities, such as time delay, spatial position, angle, etc. Different u values ​​correspond to different processing modes or parameter settings. For example, when u represents time delay, Reflecting the signal strength under different delays, through analysis The peak position can be used to assess the runtime of the underwater sonar's transmitted feedback acoustic beam signal. Furthermore, the kernel function design can consider factors such as signal propagation characteristics, noise characteristics, and underwater object characteristics to achieve the best analysis and processing results.

[0041] The optimal weight vector is calculated based on the function of optimizing the underwater object target to maximize the signal response in the desired direction. The calculation formula is as follows: In the formula, J(w) is the function for optimizing the underwater target. The goal is to find the optimal weight vector w, which, given the weight vector w, maximizes the response of the output acoustic beamline signal in the possible directions of feedback from the underwater target, while satisfying certain constraints, such as a fixed beamwidth or sidelobe level. The weight vector w is adjusted to weight the received acoustic beamline signal. The aim is to maximize the separation of the acoustic beamline signal from the underwater target and the noise near the underwater target, minimizing interference and maximizing the signal-to-noise ratio of the acoustic beamline signal, thus optimizing the underwater target to obtain the optimal weight vector. The conjugate transpose of w also maximizes the reflection of the power and energy of the received acoustic beamforming signal and shows the effect of the acoustic beamforming signal on a specified direction of an underwater object target. J(w) is a cost function representing the performance index of acoustic beamforming signal processing, outputting optimized acoustic beamforming signal flow data. Throughout the process, iterative methods are used to optimize parameters, such as weighting coefficients and delays, to improve system performance. R represents the covariance matrix of the input acoustic beamforming signal, reflecting the statistical characteristics of the acoustic beamforming signal and noise. The power of the weighted acoustic beam signal is represented by... Constrain the orientation response of the underwater object marker. It should be noted that w must guarantee sensitivity to the orientation of the underwater object marker. Where R is... The core of the optimization is its eigenvalues ​​and eigenvectors, which influence the solution for optimal w. R acts as a whitening agent during the optimization process, suppressing noise and interference direction weights. d represents the steering vector, the desired acoustic beam signal vector, which is the direction or characteristic of the acoustic beam signal of the underwater object target. d and R together determine the orientation pattern of optimal w. λ represents the iterative control parameters, used to balance the relationship between the parameters controlling the acoustic beam signal during optimization. This is a penalty applied to the desired acoustic beamwidth signal, reflecting its correlation with the desired acoustic beamwidth signal. λ transforms the constraint type into an unconstrained optimization type, and changes the constraint conditions... An embedded function for underwater object targeting is used to optimize the solution by reasonably matching the specified direction of the underwater object target with the noise environment. Furthermore, by solving for the condition that the first derivative is zero, the optimal solution and the synergistic effect of R,d,λ are obtained, thus constructing an SBA function for calculation.

[0042] To handle constraint types, accurately identify and precisely locate underwater objects, and improve the clarity of acoustic beam signals.

[0043] S103: Constructing an underwater bundled signal model through minimum variance bundled signal processing and acoustic wave bundled signal processing includes:

[0044] Minimum variance beamforming is performed to calculate the minimum variance beamforming signal. Based on the covariance matrix R of the received signal and the steering vector d constraining the beamforming direction, the optimal weight vector w is calculated using the minimum variance distortionless response criterion. The optimal weight vector w is obtained by solving the objective function J(w), which is calculated as follows: In the formula, λ represents the regularization fraction parameter, used to control the strength of the constraint condition. Let w represent the conjugate transpose of the weight vector w; R is the covariance matrix of the received signal, characterizing the statistical properties of the array's received data, including signal and noise. This involves inverting the covariance matrix to whiten noise and interference, suppressing components from non-target directions. `d` represents the steering vector, used to constrain the acoustic beam direction, representing the phase relationship of the underwater target's direction, which is related to the array geometry and the target's arrival direction. This ensures a constant response in the underwater target's direction. . This is to prevent the weight vector from scaling out of control due to constraints. J(w) is the output objective function, which obtains the minimum variance bundled signal to optimize and minimize the underwater object target's function J(w). In sonar array processing, the SBA minimum variance bundled calculation of the underwater object target obtains the optimal weight vector w, minimizing the power of the output acoustic bundled signal, achieving distortion-free response of the underwater object target's direction minimum variance bundled signal, and ensuring the signal gain of the underwater object target. In this embodiment, wR -1d reflects the interaction between the weight vector w and the steering vector d after transformation by the inverse of the covariance matrix. It embodies the response characteristics of the acoustic beamformer in a specific direction and is related to signal focusing and gain. By adjusting the weight vector w, changing the value of this part, the performance of the acoustic beamformer in the desired direction can be affected, thus enhancing the signal strength in the direction of the underwater target. Furthermore,

[0045] This represents a constraint condition. Similar to the standard MVDR calculation formula , representing the weight vector w and another guiding vector The inner product between them. Subtracting 1 indicates that the inner product is desired to be equal to 1, thereby satisfying a certain beam directivity or gain requirement. This is a further adjustment to the SBA calculation formula, related to the strength and importance of the constraints. d, as the steering vector, is related to the desired beam shape or direction. The entire section is used to correct the weight vector w, optimizing the performance of the acoustic beamformer and ensuring effective passage of the acoustic beam signal distortion in the direction of the underwater object. J(w) is a function of the underwater object formed by the MVDR acoustic beamformer and is also the optimization objective of the MVDR acoustic beamformer, achieving minimum output power and noise suppression optimization. The MVDR acoustic beamformer can achieve more efficient and accurate underwater object detection in complex underwater environments, providing effective support for improving the performance of underwater detection sonar equipment.

[0046] The acoustic beam signal received by the underwater sonar is processed to synthesize an acoustic beam signal with a specific direction. The calculation formula is as follows: ,in Let represent the direction signal of the underwater object received by the nth sensor, t represent the weighting coefficient, and s(t) represent the raw target signal received by the sonar. This raw signal contains information about the sound waves reflected or emitted by the target and is a crucial signal that needs to be focused on and extracted during sonar detection. As the starting signal for the entire computation process, it provides the basic data for subsequent signal processing. In the computation, it directly participates in subsequent addition and subtraction operations, and its amplitude, phase, and other characteristics affect the final output signal. Its characteristics. This represents the time delay relative to the acoustic beam source and is a compensation term related to system calibration, propagation loss, or other error factors. The negative sign indicates that adjustments are made to the signal to eliminate or reduce the influence of these factors on the target signal. During the calculation, a constant is directly subtracted from the signal s(t). This step is mainly to remove the bias. Adding it to s(t) corrects the original target signal s(t). The signal amplitude is reduced due to propagation path loss, then This can compensate for the loss to some extent, making the signal closer to the true target signal strength. The underwater noise signal represents an unavoidable interference factor in sonar detection, which degrades signal quality and signal-to-noise ratio. R represents the covariance matrix of the received underwater acoustic beam signal, and a is the conjugate transpose of the steering vector. θ represents the orientation angle of the underwater object. The directional vector represents the direction in which underwater sound waves reach an underwater target, and is related to the directionality of the noise. θ represents the orientation angle of the underwater target. Describe the distribution characteristics of noise in different directions. Calculate. and The effect of surrounding noise on the acoustic beam signal of the underwater object is obtained by multiplying the functions of these factors. It should be noted that in the specific calculations... Adding this to the previously corrected acoustic beamwidth signal is, from a computational perspective, part of the overall operation. In practice, it's equivalent to superimposing direction-dependent noise onto the acoustic beamwidth signal. Subsequent acoustic beamwidth signal processing steps need to suppress this noise as much as possible to improve the detection capability of the acoustic beamwidth signal for underwater targets. R represents the covariance matrix, and the inverse of matrix R is calculated. Multiplying this by vector 'a' yields the product that adjusts the covariance structure of the acoustic beamforming signal. In acoustic beamforming signal processing, the covariance matrix contains the statistical characteristics of the acoustic beamforming signal and the surrounding noise of the underwater object. It is its inverse matrix, used for whitening or decorrelation processing of the acoustic beam signal to eliminate the correlation between the acoustic beam signal and the surrounding noise of the underwater target, thereby improving the separability of the acoustic beam signal. 'a' is related to the steering vector, which describes the response characteristics of the sonar array to acoustic beam signals from different directions from the underwater target. This represents an optimization process applied to the acoustic beamforming signal using the inverse of the covariance matrix and the steering vector. The aim is to enhance the acoustic beamforming signal in the direction indicated by the underwater object and eliminate noise interference from other directions. It should be noted that during the calculation process... This is added to the previously noise-superimposed acoustic beamforming signal. In fact, this is part of signal processing algorithms such as acoustic beamforming. The calculation results are used to adjust the weight vector, and then the array receives the acoustic beam signal and performs a weighted summation to enhance the directionality of the acoustic beam signal. This can also be understood as... This represents the amount of optimization adjustment to the acoustic beamforming signal, which, together with the preceding acoustic beamforming signal components, constitutes the final processing result. Compared to the original received acoustic beamforming signal s(t), this beamforming signal exhibits improved directionality; specifically, the beamforming signal in the direction of the underwater object is enhanced, while interference and noise in other directions are suppressed. Subsequent acoustic beamforming signal processing algorithms can be based on... Further underwater object detection, localization, and identification operations are then performed. Overall, the computational formula and its components form a complete acoustic beamforming signal processing model in sonar detection. This model aims to improve the quality and directionality of the acoustic beamforming signal from underwater objects by comprehensively analyzing data from N underwater sonar sensors using SBA functions. By constructing an underwater object beamforming technology through a reasonable function calculation formula, sonar equipment can more clearly and accurately identify and locate underwater objects.

[0047] S104: Optimize the clustered signal model, the optimization including signal clustered convolutional neural network model optimization and underwater acoustic particle filter manipulation mask optimization, to obtain the optimized underwater sonar signal;

[0048] The optimization of the signal bundle convolutional neural network model includes:

[0049] The input signal is subjected to depth-bundling convolution processing using the impulse response function, and the processed underwater acoustic beam signal is output: ,in, It is the signal after SBA depth bundle separation convolution processing, representing the impulse response of the i-th bundle. The amplitude of the signal is represented by , and the amplitude of the i-th beam represents the signal strength. ð(t-τi) is the impulse function, representing the signal delay characteristic, indicating the impulse response at time (t-τi). It takes a value of 1 at t=τi and a value of 0 at other times. τi is the signal delay time, reflecting the propagation time of the sound wave from the source to the detector. Convolution involves convolving the input underwater acoustic beam signal with the impulse response of the underwater object to obtain the final effective output underwater acoustic beam signal, reflecting the temporal variation and delay of the acoustic beam signal. In this way, the sonar system can effectively process and separate acoustic beam signals from different directions and at different times, achieving efficient detection and identification of underwater objects.

[0050] Specifically, the amplitude of each bundle needs to be determined. and time delay τi. For each bundle i, at time t=τi, the impulse response... The value is To enhance or reduce the effect of a specific acoustic beam signal, an amplitude factor needs to be introduced. This factor can be adjusted according to the characteristics of the underwater target or detection requirements to improve the identifiability of the acoustic beam signal. In this embodiment, This can be the gain of the acoustic beamforming signal or the attenuation coefficient, with the aim of optimizing the reception of the acoustic beamforming signal. At other times t≠τi, the impulse response... The value is 0.

[0051] When performing convolution calculations, it's important to note that convolution is a type of matched filtering, enhancing the signal in a specific direction while suppressing noise in other directions. In sonar signal processing, it's necessary to convert the impulse response... The value is In sonar signal processing, it is necessary to process the impulse response h. i (t) is used to perform calculations with the received acoustic beam signal: , where x(t) is the received acoustic beam signal and y(t) is the convolution result.

[0052] Deep beamforming separation refers to the process where, in multiple beams, the impulse response of each beam has a non-zero value at different times. By forming acoustic beams, echoes from different directions are separated into independent channels, thus achieving separation between beams. Specifically, a unit signal t-τi is formed, where t=τi is 1 and at other times is 0. This is equivalent to marking a specific time τi on the time axis and using it as the response separation point of the acoustic beamforming signal.

[0053] Perform coefficient multiplication, and multiply the generated impulse function δ(t-τi) with the coefficient a. i Multiply, we get Amplify the intensity of the impulse function at that moment. This process, involving spatial filtering and depth-based decoupling convolution, extracts information about underwater objects, improving the resolution of sonar systems and the accuracy of underwater object identification.

[0054] Furthermore, the optimization of the signal bundle convolutional neural network model also includes SBA neural network structure optimization, calculated as follows: Here, w(t) represents the output of the neural network at time t. It should be noted that t here is the current input sample index; if processing time-series data, it is the time step. The symbol representing the activation function is used to indicate the error term of the output layer, not to directly refer to the activation function itself. Its function is to introduce non-linearity into the neural network, enabling it to form a structural pattern. Its mathematical form is: a = f(z), where z is the weighted input of the neuron (z = wTx + b), f is the activation function, and a is the output after activation. Activation functions include...

[0055] . It can also be the error term of the output layer, or gradient, used to calculate the gradient during backpropagation. It depends on the derivative of the activation function. Therefore, the choice of activation function directly affects gradient updates. It should be noted that... Activation functions are non-linear functions that enable the output of a neural network to be no longer a simple linear combination of its inputs, thereby enhancing the expressive power of the model.

[0056] Here is the weight matrix of the fully connected layer. The dimension is ,

[0057] It is the number of output neurons. It represents the number of input neurons. is the input feature vector of the sonar signal, is the input of the fully connected layer, is the output from the previous layer, and is the feature representation of the preprocessed sonar image. Its dimension is . . The bias vector of the fully connected layer is a neural network parameter used to map input features to the output target, such as object classification or localization. The core parameters of the fully connected layer are the weight matrix W and the bias vector b. The bias vector (denoted as bb) is one of the key parameters of the model, providing an independent offset for each output neuron, thereby enhancing the model's expressive power. The bias vector provides an independent offset for each output neuron, helping the model to fit the data distribution more flexibly. Together with the weight matrix, the bias vector expands the model's parameter space. When there are differences in the feature distribution of the input data, the bias can compensate for differences in the means of different features, improving model stability. Specifically, the input vector... With weight matrix Perform matrix multiplication to obtain the result of a linear combination. The vector dimension of this result is... Input features Mapped to the output feature space. After... and The product of these two factors is the optimized weight value. The bias vector is then used to calculate the weight. Added to the result of the linear combination, i.e. Modify the output offsets of N neurons. Then... Applying the above results to obtain That is, w(t), where w(t) represents the target's classification result, location coordinates, and feature vector, depending on the neural network design and task requirements. This can be achieved by optimizing the neural network structure, such as adjusting the weight matrix. Size, selection of appropriate activation function This improves the accuracy and efficiency of sonar detection.

[0058] The clustered signal model is optimized, and the optimization also includes underwater acoustic particle filter manipulation mask optimization. The specific steps are as follows:

[0059] The positions of the acoustic wave cluster particles are updated through acoustic wave cluster particle filtering and acoustic wave cluster particle swarm optimization. The position of the i-th acoustic wave cluster particle at time k is updated by the following formula: In the formula, This represents the new position of the i-th acoustic wave cluster particle at time K in the acoustic wave cluster particle filtering result. This represents the current position of the i-th sound wave cluster particle at time K; This represents the historical best position of the i-th acoustic wave cluster particle, which is the best position achieved by the acoustic wave cluster particle in all previous moments, representing the historical best state. This represents the global optimal position, which is the optimal state achieved by all sound wave cluster particles in all previous moments. and The inertia factor is used to control the speed at which the sound wave cluster particles move towards the historical best position and the global best position; the values ​​of these two inertia factors range from 0 to 2.

[0060] Specifically, the first step is to calculate the move towards the historical best position. This indicates that the sound wave cluster particles are moving towards their historical best position. The amount of movement, for the i-th iteration with N acoustic wave cluster particles k, is determined by the learning factor. The difference between the historical best position and the current position of the sound wave cluster particles. The decision is made. The dimension of the weight vector for each acoustic beamparticle is equal to the number of array elements; for example, η1=1.2, η2=2.0, and the inertia factor η=0.8. This step allows the acoustic beamparticles to reference their historical best positions when updating their positions, further maintaining effective search and detection capabilities for underwater targets.

[0061] Next, calculate the movement towards the global optimum: This indicates that the sound wave cluster particles are directed to the global optimal position. The amount of movement. This amount is determined by the learning factor. The difference between the global best position and the current position The decision is made. This step allows the sound wave cluster particles to reference the globally optimal position when updating their positions, thereby maintaining global search capability.

[0062] Finally, update the positions of the sound wave cluster particles: update the current positions of the sound wave cluster particles. The new position of the sound wave cluster particle is determined by the distance moved from the historical best position to the global best position, and the sum of these two distances is the new position of the particle. This calculation integrates acoustic beamwidth data from both local and global searches of underwater sonar, allowing the acoustic beamwidth particles to update their positions by utilizing both their own historical information and referencing the globally optimal signal. The calculation formula... This is a hybrid method combining acoustic beamforming particle filtering and acoustic beamforming particle swarm optimization for updating the positions of acoustic beamforming particles in sonar detection. By moving towards historical and global best positions, this method can further improve the accuracy and efficiency of underwater target localization.

[0063] The underwater acoustic particle filtering manipulation mask optimization also includes optimizing the acoustic beaming signal manipulation mask through sparse regularization. The optimized mask M is calculated using the following formula: In the formula, the optimized acoustic beamforming control mask is usually a matrix used to perform weighting or filtering operations on the input acoustic beamforming signal. This indicates the goal of finding a parameter M that minimizes the underwater object's function; specifically, it returns the value of M that minimizes L(M). Y is the observed acoustic beamwidth matrix, representing the raw acoustic beamwidth signal received from the sonar sensor. X is the underwater object's acoustic beamwidth matrix, representing the desired acoustic beamwidth signal or the known acoustic beamwidth pattern of the underwater object. λ is a regularization parameter used to balance the sparsity of the mask and reconstruction error; a larger λ increases sparsity, while a smaller λ reduces reconstruction error. ║M║1 represents the mask's... The norm is the sum of the absolute values ​​of all elements in a mask. Norms are used to promote the sparsity of masks, that is, to make as many elements in the mask as possible zero, thereby reducing computational complexity and enhancing the interpretability of the signal. The reconstruction error represents the difference between the observed acoustic beamwidth Y and the acoustic beamwidth X generated by the mask M acting on the underwater object. This error is expressed as follows: The norm (Euclidean norm) is used to measure the square of the acoustic beam pattern. Its purpose is to ensure that the mask M can effectively reconstruct the observed acoustic beam pattern signal Y by converting the acoustic beam pattern signal X of the underwater object as accurately as possible.

[0064] Specifically, the function for underwater objects consists of two parts; Normative terms are The goal is to achieve sparsity in the solution. The norm is calculated by summing the absolute values ​​of all elements, which causes most elements to tend to zero, thus retaining only the important components that contribute to the acoustic beam signal of underwater objects in signal processing. Normative terms are The purpose is to measure the fitting error. Here, Y is the received signal, and MX is the acoustic beam signal processed through mask M. This term evaluates the model's fit by calculating the squared difference between Y and MX. The underwater object optimization goal is to find a mask M that minimizes the weighted sum of sparsity constraints and reconstruction errors. Initialization is performed as follows: It is a random matrix or a zero matrix with reasonable initial values. The optimization process begins with initial values ​​obtained heuristically. The loss function is constructed by substituting the current value of M into the objective function and calculating the loss value. The gradient data is obtained by taking the derivative of the loss function L(M) with respect to M. The derivative is discontinuous where M is zero, so a subgradient is used to handle this. Then, a suitable SBA-manipulated mask optimization formula is used to update M. In sonar detection, the acoustic beamwidth signal of the underwater target is extracted. Through iterative optimization, the gradient is calculated to adjust M, and the final M is the optimized mask. M is then applied to the acoustic beamwidth signal X. Alternating optimization with M fixed or X fixed yields the optimal mask, enhancing the optimized acoustic beamwidth signal result.

[0065] S105: Construct a two-dimensional image of an underwater object using the SBA algorithm based on the optimized underwater sonar signal;

[0066] Two-dimensional image models can simultaneously present the orientation and distance information of underwater objects, making their localization more accurate and intuitive. The shape, volume, and state of the underwater objects provide valuable data for classification and identification. Constructing a two-dimensional image of an underwater object involves SBA (Surface Area Modeling) underwater two-dimensional image detail synthesis, decomposition, and reconstruction.

[0067] Based on existing image information, SBA underwater 2D image detail synthesis is performed to generate new 2D image information. The calculation formula is as follows: Here, I(x,y) represents the generated two-dimensional image, and A(r,θ) represents the echo intensity at a distance r and an angle θ from the underwater object. This value is usually measured by a sonar sensor and reflects the reflection of sound waves on the underwater object. δ(x-rcosθ,y-rsinθ) is the application function used to map the echo intensity A(r,θ) of the underwater object to a specific location on the two-dimensional plane, mapping A(r,θ) to the coordinates (rcosθ,rsinθ). The method involves performing two-dimensional image detail synthesis on all angles θ and distances r of underwater objects, and accumulating the echo intensity values ​​A(r,θ) of all underwater objects into the two-dimensional image I(x,y) of the underwater objects according to their corresponding positions (rcosθ,rsinθ).

[0068] Specifically, SBA underwater 2D image detail synthesis first requires polar coordinate transformation. Polar coordinate transformation of an underwater object's 2D image is a function technique that converts the image from a traditional Cartesian coordinate system to a polar coordinate system. This transformation, by remapping the pixel positions of the underwater object's 2D image, provides a new perspective and tool for the analysis, processing, and enhancement of underwater object 2D images. In the polar coordinate system, the coordinates of any point are represented as: x = rcosθ, y = rsinθ. In this calculation, each point in the polar coordinate system of the underwater object's 2D image is processed to map these points to the 2D coordinate system.

[0069] Next, the amplitude A(r,θ) is calculated: For each (r,θ), the amplitude A(r,θ) is calculated. This amplitude is obtained by the sonar signal after detailed synthesis processing, and represents the intensity of the acoustic beam signal at a specific distance and direction from the underwater object.

[0070] Next, the δ application function is used to locate specific points on the signal. The characteristic of the δ application function is that it takes a non-zero value when its parameter is zero, so δ(x-rcosθ,y-rsinθ) means that it takes a value of 1 when (x,y) equals (rcosθ,rsinθ), and a value of 0 in other cases.

[0071] Finally, the summation is performed for all r and θ: This means that all amplitudes A(r,θ) calculated in polar coordinates are weighted and applied to the corresponding (x,y) coordinates. Only when (x,y) corresponds to a certain (rcosθ,y-rsinθ) will the amplitude A(r,θ) produce an output result for the final image I(x,y). After all steps, the synthesized detail image I(x,y) ultimately represents the spatial distribution of targets detected by the sonar underwater. By summing over different r and θ values, all information from the sonar signal can be integrated into a complete two-dimensional image. Through the transformation from polar coordinates to two-dimensional coordinates, amplitude calculation, application of the δ function, and summation over all directions and distances, a visual two-dimensional image of the underwater object's application scenario is finally constructed. This detail synthesis calculation can effectively analyze the two-dimensional image distribution of underwater objects.

[0072] SBA underwater 2D image detail synthesis decomposition and reconstruction also includes SBA underwater 2D image decomposition and reconstruction, outputting a reconstructed 2D image of the underwater object target. The calculation formula is: In the formula, I(x,y) represents the reconstructed two-dimensional image of the underwater object. A(x,y) represents the A-th decomposition, and N(x,y) represents the noise or error term. R(I) represents the error between the reconstructed 2D image and its decomposed components. It is a regularization term applied to the 2D image of the underwater object, typically used to suppress noise or ensure the smoothness of the 2D image. λ represents the regularization parameter, used to balance the fitting error and the weight of the regularization term. R(I) represents the regularization term, used to constrain the reconstructed 2D image of the underwater object, controlling its smoothness or sparsity.

[0073] Specifically, in sonar detection, the final two-dimensional image I(x,y) initialized from the received two-dimensional image signal of an underwater object is composed of an ideal two-dimensional image A(x,y) and noise N(x,y). The ideal two-dimensional image is then extracted. After optimizing and solving to obtain the reconstructed two-dimensional image I(x,y), I(x,y) can be considered an approximation of the structural parts in the final two-dimensional image. In some embodiments, further post-processing calculations can be performed on the filtering and image enhancement of the two-dimensional image I(x,y) of the underwater object to improve the quality of the two-dimensional image of the underwater object. Minimize the objective function: The first item This represents the difference between the decomposed and reconstructed 2D image and the final 2D image (data fidelity), and we want this term to be as small as possible. λR(I) represents the 2D image regularization term for the underwater object target, which aims to prevent overfitting and improve restoration stability through iterative optimization. Sparsity or smoothness is introduced. This process makes the decomposed and reconstructed 2D image more consistent with reality, and then the minimum value of the objective function is found. The objective function is calculated as: J(k,I)=k║A(x,y)-I(x,y)║2+λR(I), and A(x,y) is updated to minimize J(k,I). This operation is performed using gradient descent optimization, and the final 2D image I(x,y) is updated according to the optimization objective. The iteration terminates when the change in the objective function is less than a set threshold or the maximum number of iterations is reached. This formula is a key calculation formula used for the decomposition and reconstruction of 2D images of underwater objects in sonar detection. SBA optimization is used to remove noise from the original data, enhancing the clarity of the 2D image of underwater objects.

[0074] Furthermore, the construction of underwater object two-dimensional images also includes SBA underwater two-dimensional image histogram comparison, which is mainly used to enhance the contrast of the two-dimensional images of underwater objects and further improve the accuracy of underwater object testing and judgment.

[0075] SBA underwater 2D image histogram comparison includes SBA underwater 2D image histogram equalization, outputting the equalized grayscale values ​​of the underwater object target's 2D image. This corresponds to the cumulative probability of the original gray level. The calculation formula is: In the formula Let i represent the probability density function of the gray value i of the two-dimensional image of the input underwater object, where k is the index of the current gray value.

[0076] Specifically, the cumulative distribution function of the underwater object is calculated, the grayscale value is mapped, and the frequency N(i) of each grayscale level i in the two-dimensional image of the underwater object is counted. Let represent the new grayscale value corresponding to grayscale level k in the equalized two-dimensional image of the underwater object. The value of k is typically from 0 to L-1, where L is the total number of grayscale levels in the two-dimensional image of the underwater object. The cumulative distribution function is then calculated. For each gray level k, calculate For each gray level k, according to the calculation formula... Calculate the cumulative distribution function value. Specifically, sum the probabilities p_r(i) of gray levels from 0 to k sequentially. The result is a decimal between 0 and 1, which is mapped to a gray level range of 0 to L-1. The mapping is then rounded down. Each pixel of the original 2D image of the underwater object is iterated over, and its gray value r is replaced with the corresponding equalized gray value S_r, thus obtaining the equalized 2D image.

[0077] To calculate the probability density function of a two-dimensional image of an underwater object, first, count the number of pixels for each probability density value in the input two-dimensional image of the underwater object. Then, calculate the probability density function of the two-dimensional image of the underwater object. The probability density levels of the frequency histogram in the two-dimensional image of underwater objects are redistributed, resulting in a more refined and uniform histogram distribution in the two-dimensional image of underwater objects, further enhancing the overall contrast of the two-dimensional image of underwater objects. (Calculation) , where N is the total number of pixels in the two-dimensional image of the underwater object. It is the probability of gray level i appearing in the two-dimensional image of the original underwater object, and the calculation formula is: ,in is the number of pixels with gray level i in the original 2D image of the underwater object, and n is the total number of pixels in the 2D image of the underwater object. Through PDF transformation, the gray levels of the histogram of the input 2D image of the underwater object are mapped to a new gray level range. For each gray level k, its mapped gray level... pass Calculate, where L is the total number of gray levels. For example, when there are 256 gray levels, L = 256.

[0078] Mapping to the new grayscale level will Map to a new grayscale range, using Discretize the data. Use the cumulative distribution function to map the input grayscale values ​​to the output grayscale values. This step generates a contrast-enhanced 2D image s(x,y) of the underwater object. Through this process, histogram equalization enhances the contrast of the 2D image of the underwater object, resulting in a clearer image resolution. (Application of histogram equalization for underwater objects) To avoid generating excessively enhanced noise in the two-dimensional images of underwater objects. Using The system recovers the scene's illumination information and performs gain processing on the 2D image of the underwater object. This preserves the color fidelity of the 2D image, expands its dynamic range, and improves the contrast of the underwater object image.

[0079] SBA underwater 2D image histogram comparison also includes SBA underwater 2D image contrast gain, outputting the contrast value of the original underwater object's 2D image at each pixel, calculated as follows: In the formula, f(i,j) represents the contrast value of the original 2D image of the underwater object at pixel (i,j). m(i,j) represents the mean contrast within the local window, obtained by averaging the contrast values ​​of all pixels within the local window. K(i,j) represents the local average contrast value at position (i,j), used to maintain the brightness level of the 2D image of the underwater object. It is a function of the gain coefficient related to the contrast value at position (i,j), used to control the degree of contrast enhancement. The adjustment coefficient is the mean gray value of the entire 2D image of the underwater object, and the standard deviation of the gray values ​​within the local window. ð(i,j) represents the gain coefficient, which is equal to the sum of the contrast value and the gray value. The gain coefficient represents the contrast value of the enhanced 2D image of the underwater object at position (i,j).

[0080] Specifically, first, the mean contrast value m(i,j) within the local window is calculated. A local window is defined, centered at pixel (i,j). The window size can be set according to the actual situation, such as 3×3, 5×5, etc. The average contrast value of all pixels within this local window is then calculated. , where N is the total number of pixels in the local window, and W is the set of pixels in the local window.

[0081] Secondly, the standard deviation ð(i,j) of the gain coefficient within the local window is calculated. This standard deviation is determined based on the local characteristics of the 2D image of the underwater object. The gain coefficient is calculated using the local variance or gradient of the 2D image of the underwater object to accommodate different contrast enhancement requirements. The m(i,j) term in the formula ensures that the brightness level of the 2D image of the underwater object is maintained while enhancing contrast. This helps avoid the problem of the underwater object image being too bright or too dark. (Calculation formula follows) The calculation process includes calculating the local average contrast value and gain coefficient. Through these steps, the contrast gain of the two-dimensional image of underwater objects is effectively enhanced, thereby improving the accuracy of underwater object detection and judgment.

[0082] Constructing two-dimensional images of underwater objects also includes SBA (Single-Scale Basis) for multi-scale dehazing of underwater two-dimensional images;

[0083] SBA underwater 2D image multi-scale dehazing discrete convolution includes:

[0084] Perform multi-scale dehazing of SBA underwater 2D images, outputting a multi-scale σ-dehazed and smoothed 2D image of the underwater object target. The calculation formula is: In the formula, This is a two-dimensional image of an underwater object after multi-scale dehazing and smoothing with σ. G(x,y,σ) is the value of the dehazing smoothing filter at its location, (x,y) is the kernel coordinate of the dehazing smoothing filter in the original two-dimensional image of the underwater object, representing the dehazing operation, and σ is the standard deviation used to control the scale of the filter; the choice of σ value determines the degree of blurring. In the processing of two-dimensional images of underwater objects, the choice of σ needs to be determined based on the fog level of the two-dimensional image of the underwater object and the required scale spatial resolution. By selecting different σ values, smoothed two-dimensional images of underwater objects at different scales are obtained, thereby extracting multi-scale features of the two-dimensional images of underwater objects. I(x,y) represents the original two-dimensional image of the underwater object; in some embodiments, multi-scale dehazing processing is to smooth the two-dimensional image of the underwater object under different σ values, determining the size of the dehazing smoothing kernel and the σ value: the size of the dehazing smoothing kernel is usually chosen to be an odd number, such as 3×3, 5×5, etc., to facilitate the dehazing operation. The size of the smoothing kernel (such as the radius r) determines the number of neighboring pixels to be considered in each processing. The larger the smoothing kernel, the larger the neighborhood range of each pixel. A large smoothing kernel can capture a wider range of application scene structure, reduce block artifacts, and is suitable for handling scenes with uniform lighting or large-scale fog distribution.

[0085] Specifically, a discretized (2k+1)×(2k+1) matrix is ​​generated based on the σ value. For example, a 3×3 kernel with k=3σ and σ=1 is used. The edges of the two-dimensional image of the underwater object are symmetrically filled. For example, a 5×5 kernel needs to be extended by 2 pixels. The kernel and the pixel are multiplied and then summed.

[0086] A dehazing smoothing kernel G(x,y,σ) is generated. Based on the selected σ value and the kernel size, the value of (x,y) at each position within the kernel is calculated. For example, for a 3×3 kernel, the kernel coordinates (x,y) of the underwater object's 2D image range from (−1,0,1). The calculated kernel values ​​are normalized so that the sum of all elements in the kernel is 1, ensuring that the brightness of the underwater object's 2D image remains unchanged after dehazing. Using three scales (σ=15,80,250), the kernel is multiplied with the original underwater object's 2D image I(x,y) to obtain the dehazing-smoothed 2D image of the underwater object. Then, the desired underwater object features are extracted from the two-dimensional image of the underwater object after dehazing and smoothing kernel, resulting in two-dimensional images of the dehazed underwater object at different scales, thus obtaining multi-scale features of the two-dimensional image of the dehazed underwater object.

[0087] The dehazing smoothing kernel G(x,y,σ) of the 2D image of the underwater object is processed in conjunction with the original 2D image I(x,y). The process involves sliding the smoothing kernel across the 2D image, and for each position (x,y), calculating a weighted sum of the smoothing kernel and the local region of the 2D image. For each pixel (x,y) in the 2D image of the underwater object, the result changes the value of σ, resulting in a representation of the 2D image of the underwater object at different scales, thus constructing the scale space of the 2D image of the underwater object. In the dehazing processing of 2D images of underwater objects, multiple scales are used for tasks such as feature extraction and 2D image enhancement. Detecting features such as edges and corners of the 2D image of the underwater object at different scales and fusing data from 2D images of the underwater object at different scales can continuously enhance the contrast and clarity of the 2D image of the underwater object. Finally, a clear image of the underwater object after dehazing is obtained using a restoration calculation formula. The multi-scale dehazing calculation formula described above effectively removes the fogging phenomenon in the two-dimensional image of underwater objects, improving the contrast and detail visibility of the image, thereby enhancing the detectability of underwater objects. This plays an important role in the detection and localization of underwater objects in sonar detection.

[0088] SBA underwater 2D image multi-scale dehazing discrete convolution also includes performing SBA underwater 2D image discrete convolution, outputting a 2D image of the underwater object target after convolution, calculated as follows: In the formula, o(x,y) is the output two-dimensional image of the underwater object target after convolution. I(x+m,y+n) is the input two-dimensional image of the underwater object target, K(m,n) is the value of the convolution kernel of the underwater object target, and M and N are the half-size of the convolution kernel of the underwater object target. For example, for a 3x3 convolution kernel, M=N=1; This represents the values ​​of two discrete convolution functions. The range of m is specified from -M to M, representing the size of the convolution kernel, and the range of n is specified from -N to N, representing the shape of the convolution kernel.

[0089] Specifically, in the divergent convolution operation, for each pixel (x,y) of the input two-dimensional image I(x,y) of the underwater object, the convolution kernel K(m,n) of the underwater object will perform a weighted summation calculation within a region around that point. The calculation formula indicates that the output value o(x,y) at position (x,y) is a weighted sum calculated from the value of the input 2D image I of the underwater object at (x+m,y+n) and the value of the convolution kernel K(m,n) of the underwater object, resulting in M=1 and N=1. The values ​​of m and n are in the range of -1, 0, 1 and -1, 0, 1, respectively. This allows the convolution to slide across the entire 2D image of the underwater object to extract features at different locations.

[0090] Calculate the convolution kernel for the sliding underwater object: Place the center of the underwater object's convolution kernel K sequentially at each pixel position (x, y) of the input 2D image I of the underwater object. The underwater object's convolution kernel K(m, n) covers a local region of the input 2D image. Multiply the value of the underwater object's convolution kernel K(m, n) with the value at the corresponding position in the input 2D image, and then sum all the calculated products to obtain the pixel value O(x, y) of the underwater object's output 2D image at that position. For example, if the input 2D image I of the underwater object is a 5×5 matrix, and the underwater object's convolution kernel K is a 3×3 matrix, where M=N=1. Calculate the pixel values ​​of the underwater object's output 2D image at other positions sequentially to finally obtain the complete output 2D image of the underwater object.

[0091] Determine a suitable convolution kernel for underwater objects, such as edge detection or blurring, and determine its size. Perform a convolution operation on each pixel (x, y) of the input 2D image I. Use two nested loop functions (signal values) to iterate through each position K(m, n) of the convolution kernel, where m ranges from -M to M and n ranges from -N to N.

[0092] Calculate the convolution sum: For each pixel (x, y) in the 2D image of the underwater object, perform the following convolution operation: Initialize o(x, y) to 0. For each element in the convolution kernel of the underwater object, perform the following operation: The calculation results are accumulated into o(x,y). In this operation, I(x+m,y+n) are the neighboring pixel values ​​around the current pixel, and K(m,n) are the corresponding values ​​of the underwater object's convolution kernel. Finally, each calculated value is stored in the output 2D image of the underwater object, resulting in a new output 2D image o(x,y) of the underwater object after discrete convolution processing, which displays the specific features of the input 2D image of the underwater object. This further enhances the detection level of underwater objects through SBA, effectively improving the quality and readability of underwater images.

[0093] S106: Based on the two-dimensional image, a three-dimensional image of the underwater object is generated by constructing an SBA underwater three-dimensional image space and optimizing the SBA underwater three-dimensional image structure. In some embodiments, the fusion of the three-dimensional image spatial structure of the underwater object can be derived from data from multiple acoustic beam signals, thereby improving the integrity and reliability of the beam signal data and providing more intuitive three-dimensional visual information for calculating the underwater object. During data acquisition, a sonar system is used to collect echo data of the underwater object, and at the same time, multi-angle, multi-directional, and multi-depth stereo data of the surrounding application scene of the underwater object is also collected to obtain comprehensive three-dimensional information of the underwater object.

[0094] In this embodiment, constructing the SBA underwater three-dimensional image space based on the two-dimensional image includes:

[0095] Constructing the SBA underwater 3D image space and generating the actual spatial coordinates of underwater objects is achieved using the following formula: Specifically, the first step is to perform spatial positioning calibration of the three-dimensional image: This is used to convert the pixel coordinates of an underwater object's sonar image into its actual spatial coordinates. It imports the pixel coordinates of the underwater object's sonar image into its actual three-dimensional spatial location in a two-dimensional plane. This represents the transformation matrix from pixel coordinates to spatial coordinates, obtained through the intrinsic and extrinsic parameters of the sonar image coordinates of the underwater object. Its function is to map the raw signal or feature vector received by the sonar to the spatial coordinates of the underwater object. It is determined based on the parameters of the sonar equipment used to detect the underwater object, such as beam angle, sound velocity, and equipment position, and the transformation matrix is ​​calculated using a function model. It should be noted that the matrix can be a 3×2 or 3×3 matrix, depending on the requirements and methods of linear and constraint transformations. The product obtained after matrix multiplication becomes... Each element in the array is a vector of length 3, representing the coordinates of that pixel in three-dimensional space. Using this 3D positional information, a 3D spatial model of the underwater object is constructed for subsequent analysis and visualization, such as underwater target localization and shape reconstruction. The intrinsic parameters of the underwater object's sonar image include focal length and principal point coordinates, while the extrinsic parameters include the positional location of N sonar sensors. Represents the pixel coordinates of an underwater object on a sonar image; [the pixel coordinates are then...] By setting the coordinates to homogeneous coordinates and then converting them to actual coordinates, the three-dimensional spatial position of the underwater object in the world coordinate system can be obtained. .

[0096] Constructing a three-dimensional image space for underwater objects, assuming It represents the positional information of each pixel in a two-dimensional image of an underwater object. It consists of N two-dimensional arrays, where each element is a vector containing the relative position of the pixel on the image plane or some positional encoding information related to the sonar device.

[0097] conduct The preparation involves acquiring two-dimensional image data of underwater objects from sonar equipment, extracting the position data information of each pixel, and organizing it into a suitable data structure, where each element is a vector of length 2 or more, used to store the position data information of underwater objects. It is the final result, and its data type and dimensions are usually related to Related, used to store the process The processed new location information. Through this calculation, the three-dimensional spatial data of the underwater object is recovered from the two-dimensional sonar image, generating a three-dimensional spatial model of the underwater object's surrounding environment. The spatial location calibration calculation formula is used to convert the pixel coordinates of the underwater object's sonar image into spatial coordinates, achieving precise positioning of the underwater object.

[0098] Constructing the SBA underwater 3D image space also involves processing sonar data from multiple time points to generate SBA underwater 3D image point clouds, calculated as follows: Where PointCloud is a collection of images of the underwater object, used to represent the generated 3D point cloud data. N is the total number of points in the point cloud of the underwater object, which is obtained by processing sonar data at multiple time points to obtain multiple (x,y,z) points, and finally generating the point cloud. is the three-dimensional coordinate point of the underwater object marker, used to represent the spatial position of each point in the point cloud of the underwater object marker. N is the total number of points in the point cloud of the underwater object marker, representing the number of points generated during a specific sonar detection process. Each i represents a unique point, and as i increases, it represents a different three-dimensional coordinate point.

[0099] In this embodiment, when the underwater sonar device is operating, it continuously emits sound waves into the surrounding environment of the underwater object and continuously receives the reflected sound wave signal data from the underwater object. By measuring data such as the propagation time and angle of the sound waves from the underwater object, and based on the position and location data detected by the underwater sonar device, the coordinates of the reflection point in the three-dimensional space of the underwater object are calculated. The sonar device collects information from N reflection points, each corresponding to a three-dimensional coordinate. The total distance traveled by the sound wave is calculated by measuring the time t it takes for the sound wave to travel from emission to reception at an underwater object. The formula is: In this equation, d represents the distance from the object to the sonar transmitter, u represents the speed of sound in water (usually taken as 1500 m / s), and t represents the round-trip time of the sound wave, i.e., the total time from transmission to reception. It should be noted that dividing by 2 is because t is the round-trip time of the sound wave, and the actual distance is half of that time.

[0100] After determining the distance *d* to each reflection point, the distance data is then converted into coordinates in three-dimensional space. Assuming the sonar transmitter's coordinates are (x0, y0, z0), and given the elevation θ and azimuth ϕ, the coordinates of each point can be obtained, where... ; ; , Let (x0, y0, z0) represent the distance to the i-th point, and (x0, y0, z0) represent the known position of the sonar transmitter. θ is the elevation angle, used to represent the vertical tilt angle of the sonar. denoted as azimuth, used to represent the left and right rotation angle of the sonar. Using trigonometric functions, the distance d is... i Projecting the coordinates of each point on the underwater object into its three-dimensional space yields the coordinates of that point. Finally, all the calculated points (x, y, z) are combined. i ,y i ,z i The data are collected into a set to form a complete point cloud of underwater object targets.

[0101] In this embodiment, constructing the SBA underwater 3D image space further includes performing SBA underwater 3D image texture mapping, mapping the 2D texture information of the underwater object to the 3D model of the underwater object, and the calculation formula is: Here, T(x,y) represents the result after texture mapping, which is the visual attribute corresponding to a point (x,y) on the 3D point cloud surface of the underwater object, such as color intensity and texture brightness. g is the texture fusion function, which includes the lighting model and color mapping rules, and is used to integrate height, texture features, and distance information into the final texture value. c represents the sound velocity profile. Texture feature parameters are calculated through the gray-level co-occurrence matrix of the underwater object, reflecting the roughness and contrast of the underwater object's texture. d represents the sound wave propagation distance, which is the nearest neighbor distance between the sonar ray and the 3D point cloud of the underwater object during the simulation of sonar beam sonar data. This value is used to simulate the degree of matching between the reflection intensity of the underwater object detected by the sonar and the point cloud. This is a depth map, used to represent the three-dimensional spatial distribution of underwater objects and the height values ​​of the three-dimensional point cloud. It is obtained by mapping the grayscale values ​​of the sonar image of the underwater object, and the calculation formula is: Its function is to convert the grayscale information of a two-dimensional sonar image into three-dimensional height information, forming a basic model of the terrain's undulations. Here, I(x,y) is the grayscale value of the sonar image of the underwater object at pixel (x,y), representing the reflection intensity of the underwater target or terrain. The linear mapping from grayscale to height is adjusted by a scaling factor n and an offset m. For example, if the grayscale range of the sonar image of the underwater object is [0,255], and n=0.1 and m=0, then... The range is [0, 25.5]. Linear interpolation is performed on the sparse point cloud of the underwater object to improve density and smoothness. Point cloud interpolation optimization is performed to fill the gaps in the point cloud and enhance the detail of the 3D model of the underwater object. The 3D point cloud coordinates of the underwater object are output. This forms the spatial distribution of underwater terrain or targets. Further, the texture color g(z,C,d) is calculated; the specific form of the function g depends on the needs of the actual application and the characteristics of the scene. Specifically: Where f(z,d) is a function used to adjust the color value C, which depends on the depth z and other parameters. d. For example, f(z,d) is used to simulate lighting effects, making points with greater depth darker, adjusting the brightness of the color according to the light intensity. In this embodiment, the selected f function includes, but is not limited to, linear functions and exponential decay functions, to adapt to different rendering needs. The texture values ​​T(x,y) of the underwater object are constructed: after obtaining the depth value, the color value is then obtained, and... Calculate the corresponding texture color value. Finally, output the texture, filling in the calculated color value with the (x,y) position of the image; for example, a depth map. Generated from sonar data, with values ​​ranging from [0, 10]. Color C represents the RGB color value, for example... C=(255,0,0) (red). d is the light intensity, where d = 0.5. Calculate the depth value: Assuming... Texture color: Assuming Then f(5,0.5)=1-105×0.5=1-0.25=0.75, and the final texture value is T(10,20)=g(5,(255,0,0),0.5)=(255,0).

[0102] Through the above steps By mapping the two-dimensional texture information of underwater objects onto their three-dimensional models, sonar data is mapped to 3D texture images of underwater objects, balancing geometric accuracy, texture detail, and data reliability. This provides richer visual information for sonar detection of underwater objects, enhancing the accuracy of target identification and increasing the realism of environmental modeling.

[0103] Furthermore, based on the two-dimensional image, optimizing the SBA underwater three-dimensional image structure includes:

[0104] Based on the geometric shape and texture features of the surrounding environment of the underwater object, the SBA (Structured Base Allocation) processing function is used to combine and calculate the structure, fusing multiple sonar echoes and depth information to convert the 3D data of the underwater object into a visualized elevation image. This generates a high-precision 3D reconstruction model of the underwater object, achieving elevation reshaping. The output is the 3D image intensity of the underwater object, which serves to identify and analyze underwater objects and structures, achieving data visualization, enhancing effective signals, and presenting complex 3D data in an easily observable form for subsequent analysis and processing, ultimately achieving the goal of identifying the 3D structure of the underwater object. The calculation formula is: Where I(x,y,z) represents the 3D image intensity of the underwater object; x,y,z are the spatial structural coordinates of the underwater object. This represents the frequencies that form different characteristics of the sonar echo signal of the underwater object and the reconstructed 3D image of the underwater object. k is a constant coefficient used to adjust the signal strength of the underwater object. represents the distribution of the sonar echo signal of the underwater object in 3D space. N(x,y,z) is the noise or error term. The projection function represents the propagation characteristics of the pulse signal; the incident angle and azimuth angle determine the scattering intensity. The two-dimensional coordinates (u,v) are converted into three-dimensional coordinates (x,y,z), where dx, dy, and dz represent minute volume elements in the three-dimensional space of the underwater object.

[0105] The calculation result of I(x,y,z) is the image intensity and signal intensity of a point (x,y,z) in the three-dimensional space of the underwater object. Specifically, it is first divided into a three-dimensional grid through calculation: each point in the three-dimensional space has (x,y,z), and the sonar echo signal is superimposed. Noise N(x,y,z) and pulse signal propagation characteristics The image intensity I(x,y,z) is calculated. The continuous integral summation process is discretized, and the image intensity is calculated iteratively at each spatial point. I(x,y,z) is then calculated using a function. During this process, the noise term N(x,y,z) needs to be estimated and compensated using statistical methods.

[0106] Secondly, the outer loop iterates through the x-axis coordinates. The middle loop iterates through the y-axis coordinates. The inner loop iterates through the z-axis coordinates. During the calculation, the contribution of each differential volume element dx, dy, dz to the overall image needs to be considered. This is done by multiplying dx, dy, dz as a constant factor. To process the product, an angle correction term is applied. Calculation. The corrected numerical value needs to be adjusted based on the specific sonar layout and the relative position signal value of the target object. The calculation formula is as follows: Where P0 is the initial sound source intensity, the initial sound pressure level or intensity emitted by the sonar (unit: Pa or W / m²). Sound pressure level (SPL): P0 = 1P a (Reference sound pressure level, corresponding to 0 dB SPL). Actual sonar: P0 is 10³~10⁶ P. a (Depending on the transmission power). This is the distance attenuation term, which represents the effect of sound wave energy attenuation as the propagation distance increases. It quantifies the energy loss of sound waves due to the propagation distance. Angle attenuation term: The effect of the incident angle θ of the sound wave on the reflection / scattering intensity, correcting for the change in reflected energy caused by the incident angle.

[0107] Next, the calculated image intensities of each point are combined to form a complete 3D image dataset. This 3D visualization dataset is then converted into a visualized 3D image for easier observation, analysis, and reconstruction of the final image.

[0108] Calculation results The product of the two signals, which reflects the interaction between the sonar signal and the target object, is the basis for generating a 3D image.

[0109] The product of the calculated results N(x,y,z) represents the noise removal model values, including background noise, random noise, etc. This term is used to compensate for and remove interference to the final image, thereby improving the clarity and accuracy of the 3D image of underwater objects.

[0110] Calculation results The product: This value represents the sonar emission angle and its propagation characteristics at different angles. A geometric calculation and sound wave propagation model are constructed within the three-dimensional elevation space. This value is used to adjust the image to accurately reflect the sound wave response at different angles, ensuring the image's realism and consistency.

[0111] Constant coefficient k: The value of k is determined through experiments or data fitting and is used to adjust image intensity and weights. In 3D image facade reconstruction, this factor affects the brightness and contrast of the final image, ensuring that the image has acceptable visual quality.

[0112] Finally, the 3D image facade is reconstructed: the various calculation results are combined. , , Combined, a complete image intensity expression is formed. Using the SBA 3D elevation reconstruction function, I(x,y,z) is converted into a 3D image. This includes, but is not limited to, the following methods: volume rendering, elevation reconstruction, post-processing, etc., thereby further improving the clarity of the 3D image of the underwater object. The final 3D image is then generated. I(x,y,z) will accurately reflect the surrounding environment and characteristics of the underwater object, facilitating subsequent analysis and application.

[0113] Furthermore, optimizing the SBA underwater 3D image structure also includes performing SBA underwater 3D image spectrum conversion to output the frequency domain signal of the underwater object target. The calculation formula is as follows: Where x(f) is the time-domain signal of the underwater object, representing the complex value of the signal at the frequency f of the underwater object. The result of the spectral transform of the underwater object is a complex number containing amplitude and phase information. This calculation formula is a function calculation formula for performing frequency domain analysis of the acoustic signal to extract and enhance the features of underwater 3D images. Through the SBA spectral transform function calculation formula, the time-domain signal x(t) of the underwater object can be projected onto different frequencies f of the underwater object, thereby obtaining the spectrum X(f) of the underwater object signal. This allows for the calculation and analysis of the signal's effect at various frequencies. The result X(f) of the SBA spectral transform and the frequency distribution of the underwater object tells us the frequency components in the underwater object signal, as well as their intensity (amplitude) and phase.

[0114] x(t) is the time-domain signal of the underwater object. It represents the signal of the underwater object changing with time t, and can be any continuous signal, such as sound or image signals. This is a complex exponential function containing frequency *f* and time variable *t*. This term multiplies the time-domain signal by sine waves (complex form) of different frequencies to achieve frequency domain analysis. *dt* is the integration variable, representing the integration over time *t* on the underwater object, and *f* is the frequency. The frequency to be analyzed is a range of frequency values. *j* is the imaginary unit. It should be noted that in actual calculations, multiple frequencies *f* need to be calculated. The frequency is discretized to calculate *X(f)* at different frequencies. For example, choosing to start from... arrive The frequency is determined by a step size of Δf. For each frequency f, the following steps can be taken:

[0115] Substitute the selected f into the SBA spectral transform calculation formula: .

[0116] Calculate the piecewise integral on the right: If x(t) is a piecewise function, the integral needs to be calculated separately in different intervals, and then the complex exponent needs to be solved. Expand using OL formulas

[0117] .

[0118] Decompose the integral into real and imaginary parts: .

[0119] Calculate each component: real part calculation Imaginary part calculation The results of these two integrals will form the real and imaginary parts of X(f), respectively.

[0120] In this embodiment, once the amplitude and phase X(f) are calculated, the characteristics of the underwater object's signal at frequency f can be obtained, where the amplitude is: Phase: .

[0121] The X(f) value of the underwater object is calculated using SBA, and its spectrum is plotted to display the amplitude and phase data of its frequency components. This is crucial for the calculation, analysis, processing, and identification of underwater object features using acoustic beam signals. These steps transform the acoustic beam signals of the underwater object from the time domain to the frequency domain, providing an important foundation for subsequent 3D image signal processing, feature extraction, and 3D image spectral change reconstruction.

[0122] Optimizing the SBA underwater 3D image structure further includes underwater 3D image sharpening processing, outputting the pixel values ​​of the underwater object target after SBA sharpening in the 3D image, calculated as follows: The core of the SBA sharpening calculation formula for 3D images of underwater objects is to enhance the edge details of the underwater object's 3D image through calculations such as frequency domain enhancement, acoustic beamforming, and image sharpening. Among these, L(x,y) represents the pixel value at position (x,y) in the 3D image of the underwater object after SBA sharpening. L(x,y) represents the low-frequency components of the original 3D image of the underwater object, obtained through Gaussian filtering or other low-pass filtering methods to preserve the background information of the image. α is the sharpening intensity factor, used to control the degree of sharpening; increasing α can enhance the sharpening effect, but can also introduce more noise. F represents the Fourier transform, used to transform the image from the spatial domain to the frequency domain; H(u,v) is the sharpening filter function in the frequency domain, used to enhance the high-frequency components of the image. This represents the inverse acoustic beamforming of an underwater object, which converts the frequency-domain processed 3D image of the underwater object back into the spatial domain. This means that after converting the original image of the underwater object to the frequency domain, multiplying it with the sharpening filter H(u,v), and then converting it back to the spatial domain, the high-frequency enhanced part of the three-dimensional image of the underwater object is obtained.

[0123] In this embodiment, the original image beam transform (F{L(x,y)}) aims to convert the spatial domain image L(x,y) of the underwater object into the frequency domain image of the underwater object, represented as H(u,v), which facilitates the analysis of the frequency components of the underwater object. The application employs Discrete Beam Transform (DFT), calculated using a fast beam transform algorithm. After conversion, the low-frequency components correspond to the overall brightness of the underwater object's 3D image, while the high-frequency components correspond to edges and details. B. Frequency domain filtering H(u,v). Its function is to enhance the high-frequency components of the underwater object by filtering H(u,v), suppress low-frequency noise, and improve the details of the underwater object's 3D image. Sharpening high-frequency enhancement filtering... Filtering achieves sharpening by amplifying high-frequency signals (such as edge gradients), but over-enhancing should be avoided to prevent noise amplification and image blurring. Inverse beam transform F -1 The goal is to transform the filtered frequency domain image H(u,v)⋅F{L(x,y)} of the underwater object back into the spatial domain. The inverse bundle transform yields the enhanced high-frequency detail image of the underwater object. Furthermore, the original image and the enhanced image of the underwater object are fused (weighted overlay) to balance the smoothness of the original image and the sharpness of the enhanced image, avoiding over-sharpening. The calculation formula is... The sharpening intensity coefficient α (0 < α ≤ 1) controls the contribution ratio of the enhanced image. Too large an α may cause edge ringing, while too small an α will result in insignificant sharpening. It should be noted that the sharpening algorithm optimization can perform a low-pass filter on H(u,v) before filtering to reduce high-frequency noise interference. Using a nonlinear filter (bilateral filter) instead of a linear filter can avoid edge blurring. α is dynamically adjusted based on the local contrast of the underwater object in the image; for example, increasing α in edge regions and decreasing α in smooth regions. Assuming the input image size is M × N, the calculation is as follows: Perform an FFT on L(x,y) to obtain H(u,v). Design a high-frequency enhancement filter H(u,v) and calculate... Performing an inverse bundle FFT on the product result yields... Final calculation The algorithm's sharpening processing function effectively enhances high-frequency information, further improving the display of edges, terrain, and details in sonar images of underwater objects, thereby improving the quality of 3D image sharpening and reconstruction and the clarity of 3D images.

[0124] Furthermore, this invention also provides an underwater object detection and imaging system based on SBA (Surface Mount Technology). Please refer to [link to relevant documentation]. Figure 5 ,include:

[0125] The data acquisition module acquires the raw sonar data of the underwater array elements;

[0126] The preprocessing module constructs an underwater multi-element model based on the original sonar data and performs filtering and noise reduction.

[0127] The model building module constructs an underwater bundled signal model through minimum variance bundled signal processing and acoustic bundled signal processing.

[0128] The model optimization module optimizes the clustered signal model, including signal clustered convolutional neural network model optimization and underwater acoustic particle filter manipulation mask optimization, to obtain the optimized underwater sonar signal.

[0129] The first image generation module uses the SBA algorithm to construct a two-dimensional image of an underwater object based on the optimized underwater sonar signal.

[0130] The second image generation module generates a three-dimensional image of an underwater object based on the two-dimensional image by constructing an SBA underwater three-dimensional image space and optimizing the SBA underwater three-dimensional image structure.

[0131] The embodiments described above are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for underwater object detection and imaging based on the acoustic beamforming algorithm (SBA), characterized in that, The method includes the following steps: Acquire raw sonar data from underwater array elements; An underwater multi-element model is constructed based on the original sonar data, and filtering and denoising are performed. An underwater bundled signal model is constructed by signal minimum variance bundle and acoustic wave bundle signal processing. The clustered signal model is optimized, including optimization of the signal clustered convolutional neural network model and optimization of the underwater acoustic particle filter manipulation mask, to obtain the optimized underwater sonar signal. The optimization of the signal beamforming convolutional neural network model includes: performing deep beamforming convolution on the input signal using the impulse response function, and outputting the processed underwater acoustic beamforming signal. ,in, Indicates the amplitude of the signal. Let be the impulse function, representing the delay characteristic of the signal, indicating the time delay. Immediate impact response The signal delay time reflects the propagation time of the sound wave from the source to the detector; The SBA neural network structure optimization is performed using the following formula: Where w(t) represents the output of the neural network at time t. The symbol representing the activation function is used to represent the error term of the output layer. This is the weight matrix of the fully connected layer. This is the bias vector of the fully connected layer. This is the feature vector of the sonar signal; The underwater acoustic particle filtering manipulation mask optimization includes updating the position of the acoustic cluster particles through acoustic cluster particle filtering and acoustic cluster particle swarm optimization. The position of the i-th acoustic cluster particle at time k is updated using the following formula: In the formula, This represents the new position of the i-th acoustic wave cluster particle at time K in the acoustic wave cluster particle filtering result. This represents the current position of the i-th sound wave cluster particle at time K; This represents the historical best position of the i-th sound wave cluster particle. Indicates the globally optimal position. and The inertia factor is used to control the speed at which sound wave cluster particles move towards their historical best and global best positions. The optimized mask M, obtained by optimizing the acoustic beamforming mask through sparse regularization, is calculated using the following formula: In the formula, This indicates the goal of finding a parameter M that minimizes the function of the underwater object target. Y represents the observed acoustic beamforming matrix; X represents the acoustic beamforming matrix of the underwater object target; λ is a regularization parameter used to balance the sparsity of the mask and reconstruction error; and ║M║1 represents the mask's... The norm is the sum of the absolute values ​​of all elements in a mask. Norms are used to improve the sparsity of masks. The reconstruction error represents the difference between the observed acoustic beam signal Y and the acoustic beam signal X obtained after the mask M acts on the underwater object target. Two-dimensional images of underwater objects are constructed using the SBA algorithm based on optimized underwater sonar signals; Based on the two-dimensional image, a three-dimensional image of the underwater object is generated by constructing an SBA underwater three-dimensional image space and optimizing the SBA underwater three-dimensional image structure.

2. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Constructing an underwater multi-element model based on the original sonar data includes: The formula for arranging a multi-element linear array and calculating the total array length is as follows: In the formula, N represents the total number of array elements, and d represents the preset spacing between adjacent array elements; The signals received by the underwater array elements are weighted and summed to output an integrated signal. The calculation formula is as follows: In the formula, n represents the total number of array elements, and w i This represents the weight coefficient of the i-th underwater array element. This represents the signal received by the i-th underwater array element at time t.

3. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, The filtering and denoising steps include: Filtering is performed using the probability density function of the target quantity, and the calculation formula is: In the formula, This represents the probability density function for predicting the number of n targets at time k. This represents the filtered survival probability of the target at time k-1; This represents the filtered probability density function indicating the existence of n-1 targets at time k-1; The underwater array element signals are weighted and summed for denoising to output the denoised underwater signal. The calculation formula is as follows: In the formula: wi represents the weight of the i-th underwater array element, s(t) represents the underwater object signal received by the i-th underwater array element at time t, and n(t) represents noise.

4. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, It also includes underwater post-signal time-domain analysis and optimization, the specific steps of which are as follows: The time-domain signal is mapped to a new domain containing azimuth or time delay information through kernel function transformation, resulting in the post-transformed signal. The calculation formula is as follows: In the formula, Let x(t) be the kernel function representing the transformation weight of the signal from the time domain t to the new domain u, x(t) be the time-domain window function, and dt be the time-frequency joint distribution used to detect transient signals. The optimal weight vector is calculated based on the function of optimizing the underwater object target to maximize the signal response in the desired direction. The calculation formula is as follows: In the formula, w represents the optimized weight vector. Let w be the conjugate transpose of w, and R be the covariance matrix of the input acoustic beam signal. λ represents the power of the weighted acoustic beam signal, λ represents the iterative control parameter, and d represents the steering vector.

5. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Constructing an underwater bundled signal model through minimum variance bundled signal processing and acoustic wave bundled signal processing includes: Minimum variance beamforming is performed to calculate the minimum variance beamforming signal. Based on the covariance matrix R of the received signal and the steering vector d constraining the beamforming direction, the optimal weight vector w is calculated using the minimum variance distortionless response criterion. The optimal weight vector w is obtained by solving the objective function J(w), which is calculated as follows: In the formula, λ represents the regularization fraction parameter, used to control the strength of the constraint condition, and w h This represents the conjugate transpose of the weight vector w; The acoustic beam signal received by the underwater sonar is processed to synthesize an acoustic beam signal with a specific direction. The calculation formula is as follows: ,in Let represent the direction signal of the underwater object target received by the nth sensor, t represent the weighting coefficient, and s(t) represent the original target signal received by the sonar. This represents the time delay relative to the acoustic beam signal source. Let R represent the underwater noise signal, R represent the covariance matrix of the received underwater acoustic beam signal, and θ represent the orientation angle of the underwater object. The directional vector represents the direction in which underwater sound waves reach an underwater object, and 'a' is the conjugate transpose of the directional vector.

6. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Constructing 2D images of underwater objects using the SBA algorithm includes SBA underwater 2D image detail synthesis, decomposition, and reconstruction. Based on existing image information, SBA underwater 2D image detail synthesis is performed to generate new 2D image information. The calculation formula is as follows: Where A(r,θ) represents the echo intensity at a distance r and an angle θ from the underwater object, and δ(x-rcosθ,y-rsinθ) is the application function used to map the echo intensity A(r,θ) of the underwater object to a specific location on a two-dimensional plane. Perform SBA underwater 2D image decomposition and reconstruction, and output the reconstructed 2D image of the underwater object target. The calculation formula is as follows: In the formula, A(x,y) represents the A-th decomposition, N(x,y) represents the noise or error term, and ║A(x,y)-I(x,y)║ 2 R(I) represents the error between the reconstructed 2D image and the decomposed components, R(I) represents the regularization term for the 2D image of the underwater object, λ represents the regularization parameter, and R(I) represents the regularization term.

7. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Constructing 2D images of underwater objects using the SBA algorithm also includes SBA underwater 2D image histogram comparison: SBA underwater 2D image histogram equalization outputs the equalized grayscale value S of the underwater object target's 2D image. k The calculation formula is: In the formula P r (i) represents the probability density function of the gray value i of the two-dimensional image of the input underwater object, and k is the index of the current gray value; Perform SBA underwater 2D image contrast gain, outputting the contrast value of the original underwater object's 2D image at each pixel. The calculation formula is as follows: In the formula, m(i,j) represents the mean contrast within the local window, K(i,j) represents the local average contrast value at position (i,j), and ð(i,j) represents the gain coefficient.

8. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Constructing 2D images of underwater objects using the SBA algorithm also includes SBA underwater 2D image multi-scale dehazing discrete convolution: Perform multi-scale dehazing of SBA underwater 2D images, outputting a multi-scale σ-dehazed and smoothed 2D image of underwater objects. The calculation formula is: I sooth (x,y,σ)=G(x,y,σ).I(x,y), where G(x,y,σ) is the value of the dehazing smoothing filter at the position, (x,y) is the kernel coordinate of the dehazing smoothing filter in the original two-dimensional image of the underwater object, representing the dehazing operation, and σ is the standard deviation, used to control the scale of the filter; I(x,y) represents the original two-dimensional image of the underwater object; Perform SBA discrete convolution on the underwater 2D image, and output the 2D image of the underwater object target after convolution. The calculation formula is: In the formula, I(x+m,y+n) is the two-dimensional image of the input underwater object, K(m,n) is the value of the convolution kernel of the underwater object, and M and N are the half-sizes of the convolution kernel of the underwater object. This represents the magnitude of two discrete convolution functions, where m represents the size of the convolution kernel and n represents the shape of the convolution kernel.

9. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Based on the two-dimensional image, constructing the SBA underwater three-dimensional image space includes: Constructing the SBA underwater 3D image space and generating the actual spatial coordinates of underwater objects is achieved using the following formula: , where w t This represents the transformation matrix from pixel coordinates to spatial coordinates. Represents the pixel coordinates of an underwater object in the sonar image; Sonar data from multiple time points are processed to generate SBA underwater 3D image point clouds. The calculation formula is as follows: ,in, These are the three-dimensional coordinates of an underwater object marker, used to represent the spatial position of each point in the point cloud of the underwater object marker. Let represent the coordinates of the i-th point along the X-axis in the three-dimensional space of the underwater object. Let represent the coordinates of the i-th point along the Y-axis in the three-dimensional space of the underwater object. Let N represent the coordinates of the i-th point along the Z-axis in the three-dimensional space of the underwater object, and let N represent the number of points. Perform SBA underwater 3D image texture mapping to map the 2D texture information of the underwater object to its 3D model. The calculation formula is as follows: Where g is the texture blending function, which includes the lighting model and color mapping rules. This is a depth map, used to represent the three-dimensional spatial distribution of underwater objects and the height values ​​of the three-dimensional point cloud. The depth map is obtained by mapping the grayscale values ​​of the sonar image of the underwater object, and the calculation formula is: , where I(x,y) is the gray value of the sonar image of the underwater object at pixel (x,y), n and m are adjustment coefficients used to convert the gray value into the actual height scale, and c represents the sound speed profile.

10. The underwater object detection and imaging method based on the acoustic beamforming algorithm (SBA) according to claim 1, characterized in that, Optimizing the SBA underwater 3D image structure based on the two-dimensional image includes: Perform SBA underwater 3D image elevation reconstruction and output the 3D image intensity of the underwater object target. The calculation formula is as follows: Where x, y, z are the spatial structural coordinates of the underwater object. This represents the sonar echo signal of the underwater object, where k is a constant coefficient used to adjust the signal strength of the underwater object. For noise, To represent the propagation characteristics of pulse signals, the two-dimensional coordinates (u,v), dx, dy, and dz represent the tiny volume elements in the three-dimensional space of the underwater object. Perform SBA underwater 3D image spectral conversion to output the frequency domain signal of the underwater object target. The calculation formula is as follows: Where: x(t) is the time-domain signal of the underwater object, representing the signal of the underwater object changing with time t. dt is a complex exponential function containing frequency f and time variable t, dt is the integration variable, representing the integration over time t of the underwater object, and j is the imaginary unit; Perform underwater 3D image sharpening processing, and output the pixel values ​​of the underwater object target after SBA sharpening 3D image. The calculation formula is: Where L(x,y) represents the low-frequency components of the original 3D image; α is the sharpening intensity factor, used to control the degree of sharpening; F represents the Fourier transform, used to convert the image from the spatial domain to the frequency domain; H(u,v) is the sharpening filter function in the frequency domain, used to enhance the high-frequency components of the image; This represents the inverse Fourier transform, used to convert an image from the frequency domain back to the spatial domain.

11. An underwater object detection and imaging system based on the acoustic beamforming algorithm (SBA), characterized in that, include: The data acquisition module acquires the raw sonar data of the underwater array elements; The preprocessing module constructs an underwater multi-element model based on the original sonar data and performs filtering and noise reduction. The model building module constructs an underwater bundled signal model through minimum variance bundled signal processing and acoustic bundled signal processing. The model optimization module optimizes the clustered signal model, including signal clustered convolutional neural network model optimization and underwater acoustic particle filter manipulation mask optimization, to obtain the optimized underwater sonar signal. The optimization of the signal beamforming convolutional neural network model includes: performing deep beamforming convolution on the input signal using the impulse response function, and outputting the processed underwater acoustic beamforming signal. ,in, Indicates the amplitude of the signal. Let be the impulse function, representing the delay characteristic of the signal, indicating the time delay. Immediate impact response The signal delay time reflects the propagation time of the sound wave from the source to the detector; The SBA neural network structure optimization is performed using the following formula: Where w(t) represents the output of the neural network at time t. The symbol representing the activation function is used to represent the error term of the output layer. This is the weight matrix of the fully connected layer. This is the bias vector of the fully connected layer. This is the feature vector of the sonar signal; The underwater acoustic particle filtering manipulation mask optimization includes updating the position of the acoustic cluster particles through acoustic cluster particle filtering and acoustic cluster particle swarm optimization. The position of the i-th acoustic cluster particle at time k is updated using the following formula: In the formula, This represents the new position of the i-th acoustic wave cluster particle at time K in the acoustic wave cluster particle filtering result. This represents the current position of the i-th sound wave cluster particle at time K; This represents the historical best position of the i-th sound wave cluster particle. Indicates the globally optimal position. and The inertia factor is used to control the speed at which sound wave cluster particles move towards their historical best and global best positions. The optimized mask M, obtained by optimizing the acoustic beamforming mask through sparse regularization, is calculated using the following formula: In the formula, This indicates the goal of finding a parameter M that minimizes the function of the underwater object target. Y represents the observed acoustic beamforming matrix; X represents the acoustic beamforming matrix of the underwater object target; λ is a regularization parameter used to balance the sparsity of the mask and reconstruction error; and ║M║1 represents the mask's... The norm is the sum of the absolute values ​​of all elements in a mask. Norms are used to improve the sparsity of masks. The reconstruction error represents the difference between the observed acoustic beam signal Y and the acoustic beam signal X obtained after the mask M acts on the underwater object target. The first image generation module uses the SBA algorithm to construct a two-dimensional image of an underwater object based on the optimized underwater sonar signal. The second image generation module generates a three-dimensional image of an underwater object based on the two-dimensional image by constructing an SBA underwater three-dimensional image space and optimizing the SBA underwater three-dimensional image structure.

Images