A system for breaking and leaking nets in marine aquaculture cages

Abstract

The present invention relates to the technical field of marine aquaculture detection, and in particular to a system for detecting damage and leakage in marine aquaculture cages. The system comprises: a plurality of fixed hydrophone units for acquiring passive acoustic signals; an underwater detection robot for emitting high-frequency sonar pulses and collecting reflected echo signals; a control unit for issuing drive instructions to the underwater detection robot and for receiving reflected echo signals and passive acoustic signals; a signal processing unit for extracting time delay and spectrum characteristics from the reflected echo signals and determining the position information of the damage noise source based on the passive acoustic signal; an analysis module for extracting a joint acoustic feature vector from the spectrum characteristics through a convolutional neural network, and for generating a node feature vector representing the acoustic correlation between cages through a graph neural network, and fusing the two to perform damage detection; and a position optimization unit for iteratively optimizing the position information of the damage noise source and outputting the final cage damage position.

Description

Technical Field

[0001] The present invention relates to the technical field of marine aquaculture detection, in particular to a network breaking and leak detection system for marine aquaculture cages. Background Art

[0002] my country boasts abundant fishery resources, with the total volume of aquatic products and aquaculture output exceeding one trillion yuan. Deep-sea aquaculture equipment and smart fisheries are currently key areas of fishery development. Marine aquaculture encompasses harbor ponds, ponds, net enclosures, tanks, rafts, and cages. Cage aquaculture is a crucial component of deep-sea aquaculture, and underwater cage aquaculture has gradually become a major development direction for my country's marine aquaculture industry.

[0003] Existing cage aquaculture systems include conventional cages and deep-sea gravity cages. Conventional cages use a suspended mesh structure to enclose fish. Compared to deep-sea gravity cages, conventional cages offer lower construction costs, easier maintenance, and greater flexibility. They are often deployed alongside deep-sea gravity cages, primarily for hatching fry. However, their structural strength is relatively low, making them susceptible to escape risks. Deep-sea gravity cages, on the other hand, utilize a rigid frame and mesh structure to enclose fish. While they offer high structural strength and yield, they also come with higher construction costs and technical requirements.

[0004] According to statistics, approximately two-thirds of losses in cage aquaculture are caused by damaged nets. Damaged nets not only result in unforeseen economic losses but also pose a serious ecological risk. Currently, to mitigate losses from damaged nets, regular net replacement or frequent cage inspections by divers are common methods to check net integrity. However, these methods are not only inefficient but also pose significant safety risks. Therefore, a more reliable and secure cage inspection solution is needed to address this critical issue that hinders the healthy development of underwater cage aquaculture. Summary of the Invention

[0005] In view of the above-mentioned prior art, the present invention provides a system for breaking and detecting leaks in seawater aquaculture cages, which mainly solves the technical problems existing in the above-mentioned background technology.

[0006] To achieve the above-mentioned purpose, the technical solution of the embodiment of the present invention is implemented as follows:

[0007] A system for breaking and detecting leaks in a marine aquaculture cage, comprising:

[0008] A plurality of fixed hydroacoustic units for acquiring passive acoustic signals, wherein the passive acoustic signals include ambient noise signals and turbulent noise signals from different sources;

[0009] An underwater inspection robot equipped with active sonar, used to emit high-frequency sonar pulses and collect reflected echo signals;

[0010] A control unit, configured to issue a drive instruction to the underwater detection robot and to receive the reflected echo signal and the passive acoustic signal;

[0011] a signal processing unit configured to extract time delay and spectrum characteristics from the reflected echo signal and determine position information of the damage noise source according to the passive acoustic signal;

[0012] An analysis module, which is equipped with a convolutional neural network and a graph neural network. The convolutional neural network is used to extract joint acoustic feature vectors from spectral features, and the graph neural network is used to generate node feature vectors representing the acoustic correlation between cages. The two are combined to perform damage detection;

[0013] The position optimization unit is configured to iteratively optimize the position information of the damage noise source and output the final cage damage position.

[0014] Optionally, extract delay and spectrum characteristics from the reflected echo signal, including:

[0015] Preprocessing the reflected echo signal, including bandpass filtering and noise reduction;

[0016] Perform cross-correlation analysis on the transmitted signal and the pre-processed reflected echo signal to obtain the delay value;

[0017] The pre-processed reflected echo signal is subjected to short-time Fourier transform to generate a time-frequency distribution matrix, and the spectrum center of gravity and the frequency band energy ratio are extracted from the time-frequency distribution matrix as spectrum features.

[0018] Optionally, determining the location information of the damage noise source according to the passive acoustic signal specifically includes:

[0019] M hydrophone units acquire M passive acoustic signals, perform band-stop filtering on the M passive acoustic signals to suppress the frequency bands of ship noise and biological noise, and obtain a target acoustic signal containing a specific frequency band;

[0020] Determine whether there is corruption noise in the target acoustic signal. If so, cross-correlate multiple passive acoustic signals to obtain arrival time differences, establish a TDOA positioning equation group based on the multiple arrival time differences, and use the least squares method to solve the TDOA positioning equation group to obtain the position coordinates of the corruption noise source.

[0021] Optionally, when the position coordinates of the damaged noise source are obtained, the cage closest to the position coordinates of the damaged noise source is selected, and the control unit sends a driving instruction to the underwater detection robot to control the underwater detection robot to move to the cage to emit high-frequency sonar pulses and collect reflected echo signals.

[0022] Optionally, determining whether there is damage noise in the target acoustic signal specifically includes: calculating the signal power spectrum density in the target acoustic signal, and if the signal power spectrum density exceeds a threshold , it means that there is corruption noise in the target acoustic signal.

[0023] Optionally, multiple passive acoustic signals are cross-correlated to obtain arrival time differences, and a TDOA positioning equation group is established based on the multiple arrival time differences, specifically including: taking the hydroacoustic unit that receives the target acoustic signal earliest as the reference hydroacoustic unit, cross-correlating the target acoustic signal received by the reference hydroacoustic unit with the target acoustic signals received by other M-1 hydroacoustic units, respectively, to obtain the arrival time differences between the target acoustic signal of the damage noise source reaching the other M-1 hydroacoustic units and the reference hydroacoustic unit, and forming a group of positioning information groups based on the multiple arrival time differences, forming a total of M-1 positioning information groups.

[0024] Optionally, a joint acoustic feature vector is extracted from the spectral features, specifically including: establishing and training a CNN model, inputting the time-frequency distribution matrix into the pre-trained CNN model to obtain CNN high-order features, and splicing the time delay features, spectral center of gravity features, frequency band energy ratio features, and CNN high-order features into a joint acoustic feature vector.

[0025] Optionally, a node feature vector is generated to characterize the acoustic correlation between cages, specifically including: establishing and training a GNN model, where the pre-established GNN model includes nodes and edges, where each node represents a breeding cage, and the edge represents the sound wave propagation path loss between nodes, and a node feature vector of each node is generated by encoding and aggregating the features of the nodes and edges.

[0026] Optionally, the node feature vector corresponding to the box closest to the position coordinates of the damage noise source is selected and dynamically weighted fused with the joint acoustic feature vector based on the attention mechanism to obtain a cross-modal feature vector, and the cross-modal feature is input into the SVM classifier to output the probability value of damage / non-damage, where the activation function of the SVM classifier is the Softmax function.

[0027] Optionally, if the SVM classifier outputs a high probability of damage to the cage, a particle filter algorithm is used to further optimize the position coordinates of the damage noise source and output damage position information of the target cage.

[0028] The beneficial effects of the present invention are as follows: by continuously collecting passive acoustic signals through the hydrophone unit and combining them with the high-frequency sonar pulse echoes emitted directionally by the underwater inspection robot, a dual detection mechanism of passive noise positioning and active sonar scanning is realized. The passive acoustic signal quickly locates the potential damage area through the TDOA algorithm, and the time-frequency analysis of the active sonar accurately extracts local features. The collaborative work of the two overcomes the limitations of traditional single detection methods.

[0029] By introducing multimodal feature fusion of convolutional neural networks and graph neural networks, the robustness of damage identification is further enhanced. The high-order acoustic features extracted by CNN from the time-frequency distribution matrix can capture microscopic patterns such as spectral distortion and texture anomalies caused by damage. The GNN-modeled path loss correlation between acoustic wave propagation and other grid boxes reflects the propagation characteristics of the damage signal at a topological level. The two types of features are dynamically weighted and fused through the attention mechanism to form a cross-modal feature vector, solving the problem of feature separation between the time, frequency, and spatial domains in traditional methods.

[0030] A particle filter algorithm was used to iteratively optimize the initial positioning results, achieving sub-meter accuracy in the damage location output. By establishing a state model to describe the spatial constraints of the damage point and combining time-delayed observation data from multiple hydrophones with the spectral characteristics of active sonar, the particle filter dynamically adjusts particle weights, gradually converging to the true damage coordinates. It remains stable in the presence of multipath effects or environmental noise interference, reducing the positioning error from the initial TDOA of 3-5 meters to less than 0.5 meters.

[0031] This application changes the traditional high-risk operation mode that relies on manual diving inspections. It can complete damage location and verification without human intervention, eliminate the safety hazards of divers operating in deep, turbid or dangerous waters, and realize continuous monitoring of large-scale aquaculture areas, providing reliable technical guarantees for the sustainable development of deep-sea aquaculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a schematic diagram of the module connections of a seawater aquaculture cage net breaking and leak detection system in an embodiment of the present application. DETAILED DESCRIPTION

[0033] The technical solution of the present invention is further elaborated in detail below in conjunction with the drawings and specific embodiments of the specification. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In the following description, reference is made to "some embodiments", which describes a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0034] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art are not described to avoid confusion with the present invention.

[0035] It should be understood that the present invention can be implemented in different forms and should not be interpreted as being limited to the embodiments proposed herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the present invention to those skilled in the art. And the purpose of the terms used herein is only to describe specific embodiments and is not intended to limit the present invention. When used herein, the singular forms "one", "an" and "said / the" are also intended to include plural forms, unless the context clearly indicates another way. It should also be understood that the terms "comprising" and / or "comprising" when used in this specification determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0036] It should also be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. The terms "vertical," "horizontal," "inner," "outer," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only implementation methods.

[0037] In order to fully understand the present invention, a detailed structure will be provided in the following description to illustrate the technical solution proposed by the present invention. Optional embodiments of the present invention are described in detail below. However, in addition to these detailed descriptions, the present invention may also have other implementations.

[0038] Please refer to the attached Figure 1 The present application provides a system for breaking and detecting leaks in marine aquaculture cages, the system comprising:

[0039] A plurality of fixed hydroacoustic units for acquiring passive acoustic signals, wherein the passive acoustic signals include ambient noise signals and turbulent noise signals from different sources;

[0040] An underwater inspection robot equipped with active sonar, used to emit high-frequency sonar pulses and collect reflected echo signals;

[0041] A control unit, configured to issue a drive instruction to the underwater detection robot and to receive the reflected echo signal and the passive acoustic signal;

[0042] a signal processing unit configured to extract time delay and spectrum characteristics from the reflected echo signal and determine position information of the damage noise source according to the passive acoustic signal;

[0043] An analysis module, which is equipped with a convolutional neural network and a graph neural network. The convolutional neural network is used to extract joint acoustic feature vectors from spectral features, and the graph neural network is used to generate node feature vectors representing the acoustic correlation between cages. The two are combined to perform damage detection;

[0044] The position optimization unit is configured to iteratively optimize the position information of the damage noise source and output the final cage damage position.

[0045] Specifically, passive acoustic signals are continuously collected by a fixed hydrophone unit arranged under the deep-sea aquaculture cage. The fixed hydrophone unit can be suspended and fixed under the deep-sea aquaculture cage or fixed on the seabed. The passive acoustic signals collected are input into the signal processing unit for analysis to determine whether there is a damage source noise signal. When a damage source noise signal exists, the position information of the damage noise source is solved and the position information is mapped to the geometric space range of the target cage. The control unit drives the underwater detection robot to approach the target cage according to the positioning result, and transmits a pulse signal through high-frequency active sonar and collects the reflected echo. The signal processing unit extracts the time-frequency features of the echo signal and obtains the time delay value of the reflection path based on the cross-correlation analysis. The analysis module uses a convolutional neural network to perform multi-level feature extraction on the time-frequency graph to capture the damage. The resulting local spectral distortion and texture anomalies form acoustic feature vectors. A graph neural network is then used to construct an acoustic correlation model between cage nodes. Sound wave propagation losses are quantified as edge weights. Graph convolution is then used to aggregate the acoustic state features of adjacent cages to generate node feature vectors representing spatial relationships. These two features are dynamically weighted and fused using an attention mechanism before being input into a classification model. The probability of damage is determined by integrating information from the time, frequency, and spatial domains. After determining that a cage is damaged, the location optimization unit iteratively optimizes the initial location of the damage noise source using a particle filter algorithm. The spatial constraints of the damage point are described using a state equation, and multi-source acoustic data is integrated using an observation equation. A resampling strategy is then used to gradually narrow the damage location until convergence to the true damage coordinates. The system ultimately outputs the precise damage location of the target cage with sub-meter accuracy. This system guides active sonar directional scanning through passive noise localization, combining data-driven and physical models to enhance detection robustness, achieving fully automated detection from wide-area anomaly detection to localized, precise diagnosis.

[0046] It should be noted that the selection of the fixed hydroacoustic unit and the underwater detection robot and the arrangement of the fixed hydroacoustic unit underwater are conventional technical means in this field and will not be described in detail in this embodiment.

[0047] In some embodiments, extracting time delay and spectrum characteristics from the reflected echo signal specifically includes:

[0048] Reflected echo signal Preprocessing is performed, which includes bandpass filtering and noise reduction. Specifically, the digital bandpass filter of the original reflected echo signal retains the effective frequency band signal from 20kHz to 200kHz, eliminates low-frequency ship noise and high-frequency interference, and suppresses random noise through the wavelet threshold noise reduction algorithm to improve the signal-to-noise ratio. The wavelet threshold noise reduction algorithm expression is:

[0049]

[0050] Where WT represents wavelet transform, The threshold parameter is used to perform cross-correlation analysis on the transmitted signal and the pre-processed reflected echo signal to obtain the delay value. The calculation formula for the cross-correlation analysis is:

[0051]

[0052] To transmit the signal, For the preprocessed signal, the delay value is determined by finding the peak position of the cross-correlation function:

[0053]

[0054] This value corresponds to the round-trip time difference of the sound wave, which is converted into physical distance , then short-time Fourier transform is performed on the preprocessed reflected echo signal to generate a time-frequency distribution matrix, and the spectrum center of gravity and the frequency band energy ratio are extracted from the time-frequency distribution matrix as spectrum features, wherein short-time Fourier transform is performed on the preprocessed reflected echo signal and the spectrum center of gravity and the frequency band energy ratio are extracted from the time-frequency distribution matrix. These are conventional technical means in the field and will not be described in detail in this embodiment.

[0055] Breakage noise is essentially turbulence noise, caused by pressure pulsations in a turbulent flow field. When turbulent flow passes over obstacles or boundaries, these pressure pulsations generate sound waves. Turbulence is the irregular, random motion of a fluid. Its instability causes collisions and vibrations between fluid molecules, generating noise. When aquaculture cages are damaged, the flow velocity and direction of water passing through the breach changes, leading to unstable flow and the formation of turbulence. This turbulence triggers pressure pulsations, which in turn generate turbulent noise. This noise may be mixed with other environmental noise and eddy current noise generated by other sources.

[0056] Therefore, in some embodiments, determining the location information of the damage noise source based on the passive acoustic signal specifically includes: M hydrophone units acquiring M passive acoustic signals, performing band-stop filtering on the M passive acoustic signals to suppress the ship noise frequency band and the biological noise frequency band, and obtaining a target acoustic signal containing a specific frequency band;

[0057] Determine whether there is corruption noise in the target acoustic signal. If so, cross-correlate multiple passive acoustic signals to obtain arrival time differences, establish a TDOA positioning equation group based on the multiple arrival time differences, and use the least squares method to solve the TDOA positioning equation group to obtain the position coordinates of the corruption noise source.

[0058] Furthermore, multiple passive acoustic signals are cross-correlated to obtain arrival time differences, and a TDOA positioning equation group is established based on the multiple arrival time differences, specifically including: taking the hydroacoustic unit that receives the target acoustic signal earliest as the reference hydroacoustic unit, cross-correlating the target acoustic signal received by the reference hydroacoustic unit with the target acoustic signals received by other M-1 hydroacoustic units, respectively, to obtain the arrival time differences between the target acoustic signal of the damage noise source reaching the other M-1 hydroacoustic units and the reference hydroacoustic unit, and forming a group of positioning information groups based on the multiple arrival time differences, forming a total of M-1 positioning information groups.

[0059] Specifically, the original signals collected by M hydrophone units are subjected to band-stop filtering, and the stopband range is designed to be the ship noise band and the biological noise band. After suppressing the interference, the target acoustic signal is obtained. If it is determined that there is damage noise in the target acoustic signal, select the reference hydroacoustic unit M=1, and calculate the cross-correlation function between the remaining hydroacoustic units and the reference hydroacoustic unit:

[0060]

[0061] Improving delay resolution through PHAT weighting:

[0062]

[0063] Extract the peak position to get the delay difference:

[0064]

[0065] Construct TDOA positioning equations based on time delay difference: Assume the coordinates of the noise source are , the reference hydroacoustic unit coordinates are , then the equation of the reference hydroacoustic unit and another hydroacoustic unit is expressed as:

[0066]

[0067] in is the coordinate of another hydroacoustic unit, and so on. The target acoustic signal received by the reference hydroacoustic unit is cross-correlated with the target acoustic signals received by other M-1 hydroacoustic units to obtain multiple equations. The multiple equations constitute a positioning information group, and the LM algorithm is used to transform the positioning information group into a nonlinear least squares problem and solve it to minimize the objective function and obtain the initial position coordinates of the damage noise source.

[0068] Furthermore, the frequency band of turbulent noise is closely related to the characteristics of turbulence, including the intensity and scale of turbulence and the motion characteristics of the fluid. Different turbulent structures will produce noise of different frequencies. For example, small-scale turbulence will produce high-frequency noise, while large-scale turbulence will produce low-frequency noise. Therefore, after the aquaculture cage is damaged, the turbulent noise generated by the water flowing through the breach may have a specific frequency band and signal power spectrum density. Therefore, determining whether there is damage noise in the target acoustic signal specifically includes: calculating the signal power spectrum density in the target acoustic signal, and the calculation formula is:

[0069]

[0070] in, Target acoustic signal The Fourier transform of is the acquisition period, if the signal power spectrum density exceeds the threshold , it means that there is corruption noise in the target acoustic signal.

[0071] When the position coordinates of the damaged noise source are obtained, the net cage closest to the position coordinates of the damaged noise source is selected, and the control unit sends a driving instruction to the underwater detection robot to control the underwater detection robot to move to the net cage to emit high-frequency sonar pulses and collect reflected echo signals.

[0072] In some embodiments, after obtaining the coordinates of the damage noise source and the echo signal, and obtaining the time delay and time-frequency distribution matrix from the echo signal, a convolutional neural network in the analysis module is used to extract a joint acoustic feature vector from the spectral features. The process of extracting the joint acoustic feature vector from the spectral features aims to integrate multi-dimensional acoustic information and improve the robustness of damage detection.

[0073] Furthermore, a joint acoustic feature vector is extracted from the spectral features, specifically including: establishing and training a CNN model, inputting the time-frequency distribution matrix into the pre-trained CNN model to obtain CNN high-order features, and splicing the time delay features, spectral center of gravity features, frequency band energy ratio features, and CNN high-order features into a joint acoustic feature vector.

[0074] The core principle of extracting joint acoustic feature vectors is to extract high-order abstract features from the time-frequency distribution matrix through convolutional neural networks, and splice them with delay features, spectral center of gravity features, and frequency band energy ratio features to form a cross-domain joint representation. Specifically, the time-frequency distribution matrix is ​​generated by short-time Fourier transform, which represents the energy distribution of the signal in time window t and frequency f. The time-frequency distribution matrix is ​​input into the pre-trained CNN model, and the model extracts spatial features through multi-layer convolution and pooling operations, and finally outputs a high-order feature vector. , where D is the feature dimension, and the final delay feature, spectrum center feature, band energy ratio feature, and high-order feature vector are fused into a joint acoustic feature vector through splicing operation ,Right now This approach leverages the complementary information in the time domain (delay), frequency domain (center of gravity and energy ratio), and spatial domain (CNN features) to enhance the model's ability to distinguish damage signatures. For example, delay features can locate the spatial distance of the damage point, the spectral center of gravity reflects the dominant frequency characteristics of the noise, and CNN features capture localized abnormal patterns in the time-frequency graph. Through joint optimization, the system can more accurately identify minor damage or abnormal signals in complex interference environments, providing highly discriminative input features for subsequent classification and location.

[0075] Furthermore, a node feature vector is generated to characterize the acoustic correlation between cages, specifically including: establishing and training a GNN model, where the pre-established GNN model includes nodes and edges, where each node represents a breeding cage, and the edge represents the sound wave propagation path loss between nodes. By encoding and aggregating the features of the nodes and edges, a node feature vector of each node is generated.

[0076] The process of generating node feature vectors representing the acoustic correlation between cages relies on a graph neural network (GNN) that jointly models the topological structure and acoustic propagation characteristics of aquaculture cages. Its core is to abstract each cage as a node in the graph. The acoustic path loss between nodes is quantified as edge weights. Through message passing and feature aggregation through multiple layers of graph convolutional layers, node feature vectors reflecting global acoustic correlation are gradually generated. After multiple rounds of iteration, the node features ultimately encode the acoustic path loss, topological correlation, and local acoustic state between cages. For example, edges with large path loss can suppress the strength of feature transfer between adjacent nodes, while high-frequency resonance characteristics can be diffused to neighboring nodes through multi-layer aggregation, thus reflecting the coordinated acoustic behavior of the cage group in the feature vector. Ultimately, the node feature vector not only captures local acoustic characteristics but also reflects abnormal propagation paths in the global network, thereby improving the robustness of damage detection. For example, when a cage is damaged, its abnormal acoustic signal is weighted by the path loss of adjacent cages, ultimately forming a significant difference in the node features, making it easier for the classifier to identify.

[0077] In an optional embodiment, the node feature vector corresponding to the box closest to the position coordinates of the damage noise source is selected and dynamically weighted fused with the joint acoustic feature vector based on the attention mechanism to obtain a cross-modal feature vector. The cross-modal feature is input into the SVM classifier, and the probability value of damage / non-damage is output, where the activation function of the SVM classifier is the Softmax function.

[0078] Specifically, the core of the attention mechanism is to adaptively assign weights to highlight key information by calculating the correlation between two feature vectors. Specifically, the two feature vectors are first mapped to query vectors and key vectors respectively, and the attention score is calculated by dot product. The softmax function normalizes the score to a weight distribution. Subsequently, the value vector is weighted and summed with the attention weight to obtain the fused cross-modal feature vector. ,in ,in is the attention weight, is a value vector, This process dynamically balances the contribution of local acoustic features (such as spectral distortion) and global topological associations (such as path loss). For example, when the damage noise feature is significant, the attention weight The joint acoustic features are preferred; when the acoustic correlation between cages is complex, the weight of node features is enhanced.

[0079] Cross-modal feature vector Then the data is input into the support vector machine (SVM) classifier for damage probability prediction. SVM divides the feature space by a hyperplane, and its output is the output probability value. The Softmax function is used to map the decision value to the probability distribution. The SVM parameters are optimized by maximizing the interval loss function to make the classification boundary as far away from the two types of samples as possible. For example, when the cross-modal feature vector contains high frequency energy anomalies, such as the band energy ratio Sudden increase and path loss mutation, i.e. node characteristics In the event of anomalies, the probability of Softmax output corruption increases significantly.

[0080] The attention mechanism strengthens key modal information through adaptive weight allocation, while SVM combined with Softmax maps high-dimensional features into interpretable probabilistic outputs. The two work together to achieve efficient conversion from multi-source data to decision results. In the complex ocean noise background, if the node feature of a certain cage has a reduced weight due to abnormal path loss, but the joint acoustic feature shows obvious delay and spectral distortion, the attention mechanism can still ensure accurate identification of the damaged signal through dynamic weighting, thereby improving the system's environmental adaptability and detection accuracy.

[0081] In some embodiments, if the SVM classifier outputs a high probability of cage damage, a particle filter algorithm is used to further optimize the position coordinates of the damage noise source and output damage location information of the target cage.

[0082] Specifically, when the SVM classifier determines that the probability of cage damage is high, the system uses a particle filter algorithm to iteratively optimize the initial damage noise source position to improve positioning accuracy. The particle filter is based on the Monte Carlo method, which represents the posterior probability distribution through a set of weighted particles and dynamically updates the particle state using observation data. The details are as follows:

[0083] Based on the initial position coordinates of the damage noise source obtained by the TDOA positioning equation, N particles are assigned to the possible state space, and the covariance matrix is ​​set based on the positioning error. Each particle has the same initial weight. For each particle, the state at the next moment is predicted based on the motion model:

[0084] The state space model describes the dynamic characteristics of the damage location. Assuming that the damage location is stationary for a short period of time, the state equation is:

[0085]

[0086] in is the coordinate of time step k, The observation model maps the state to the sensor data, which includes the time delay measurement of multiple hydrophones and the spectrum characteristics of the reflected echo of active sonar, such as the sound wave from the damage location. The actual time delay of propagation to the mth hydrophone is:

[0087]

[0088] Use observation data to adjust particle weights. For each particle Calculate the degree of match with the observed data, generate a new particle set based on the weight distribution to avoid particle degradation, adopt a systematic resampling strategy, copy high-weight particles according to the weight ratio, remove low-weight particles, ensure particle diversity, and obtain the optimized position through weighted average: Through multiple rounds of iteration, the particles gradually converge to the high probability area and finally output the optimized position In complex, noisy ocean environments, initial TDOA positioning can have errors of several meters due to multipath effects. However, particle filtering, after fusing multi-sensor data, can reduce positioning errors to sub-meter levels. This process effectively combines SVM probabilistic triggering with the acoustic wave propagation equation, balancing efficiency and accuracy to provide reliable spatial positioning results for cage damage detection.

[0089] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. The scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A system for breaking and detecting leaks in marine aquaculture cages, characterized in that: The system comprises: A plurality of fixed hydroacoustic units for acquiring passive acoustic signals, wherein the passive acoustic signals include ambient noise signals and turbulent noise signals from different sources; An underwater inspection robot equipped with active sonar, used to emit high-frequency sonar pulses and collect reflected echo signals; A control unit, configured to issue a drive instruction to the underwater detection robot and to receive the reflected echo signal and the passive acoustic signal; a signal processing unit configured to extract time delay and spectrum characteristics from the reflected echo signal and determine position information of the damage noise source according to the passive acoustic signal; An analysis module, which is equipped with a convolutional neural network and a graph neural network. The convolutional neural network is used to extract joint acoustic feature vectors from spectral features, and the graph neural network is used to generate node feature vectors representing the acoustic correlation between cages. The two are combined to perform damage detection; a position optimization unit configured to iteratively optimize the position information of the damage noise source and output a final damage position of the cage; When the location coordinates of the damage noise source are obtained, a net cage closest to the location coordinates of the damage noise source is selected, and the control unit sends a drive instruction to the underwater detection robot to control the underwater detection robot to move to the net cage to emit high-frequency sonar pulses and collect reflected echo signals; Generate a node feature vector that characterizes the acoustic correlation between cages, specifically including: establishing and training a GNN model, where the pre-established GNN model includes nodes and edges, where each node represents a breeding cage, and the edge represents the sound wave propagation path loss between nodes. By encoding and aggregating the features of the nodes and edges, a node feature vector for each node is generated.

2. A system for breaking and detecting leaks in seawater aquaculture cages according to claim 1, characterized in that: Extracting time delay and spectrum characteristics from the reflected echo signal, including: Preprocessing the reflected echo signal, including bandpass filtering and noise reduction; Perform cross-correlation analysis on the transmitted signal and the pre-processed reflected echo signal to obtain the delay value; The pre-processed reflected echo signal is subjected to short-time Fourier transform to generate a time-frequency distribution matrix, and the spectrum center of gravity and the frequency band energy ratio are extracted from the time-frequency distribution matrix as spectrum features.

3. A system for breaking and detecting leaks in seawater aquaculture cages according to claim 2, characterized in that: Determining the location information of the damage noise source according to the passive acoustic signal specifically includes: M hydrophone units acquire M passive acoustic signals, perform band-stop filtering on the M passive acoustic signals to suppress the frequency bands of ship noise and biological noise, and obtain a target acoustic signal containing a specific frequency band; Determine whether there is corruption noise in the target acoustic signal. If so, cross-correlate multiple passive acoustic signals to obtain arrival time differences, establish a TDOA positioning equation group based on the multiple arrival time differences, and use the least squares method to solve the TDOA positioning equation group to obtain the position coordinates of the corruption noise source.

4. A system for breaking and leak detection of seawater aquaculture cages according to claim 3, characterized in that: Determining whether there is damage noise in the target acoustic signal specifically includes: calculating the signal power spectrum density in the target acoustic signal, and if the signal power spectrum density exceeds a threshold , it means that there is corruption noise in the target acoustic signal.

5. A system for breaking and detecting leaks in seawater aquaculture cages according to claim 4, characterized in that: After cross-correlating multiple passive acoustic signals, the arrival time difference is obtained, and a TDOA positioning equation group is established based on the multiple arrival time differences, specifically including: taking the hydroacoustic unit that receives the target acoustic signal earliest as the reference hydroacoustic unit, cross-correlating the target acoustic signal received by the reference hydroacoustic unit with the target acoustic signals received by the other M-1 hydroacoustic units, respectively, to obtain the arrival time difference between the target acoustic signal of the damage noise source reaching the other M-1 hydroacoustic units and the reference hydroacoustic unit, and forming a group of positioning information groups based on the multiple arrival time differences, forming a total of M-1 positioning information groups.

6. A system for breaking and leak detection of seawater aquaculture cages according to claim 5, characterized in that: Extracting a joint acoustic feature vector from spectral features specifically includes: establishing and training a CNN model, inputting the time-frequency distribution matrix into the pre-trained CNN model to obtain CNN high-order features, and splicing the time delay features, spectral center of gravity features, frequency band energy ratio features, and CNN high-order features into a joint acoustic feature vector.

7. A system for breaking and detecting leaks in seawater aquaculture cages according to claim 6, characterized in that: The node feature vector corresponding to the box closest to the location coordinates of the damage noise source is selected and dynamically weighted fused with the joint acoustic feature vector based on the attention mechanism to obtain a cross-modal feature vector. The cross-modal feature is input into the SVM classifier and the probability value of damage / non-damage is output, where the activation function of the SVM classifier is the Softmax function.

8. The system for breaking and detecting leakage of seawater aquaculture cages according to claim 7, characterized in that: If the SVM classifier outputs a high probability of damage to the cage, a particle filter algorithm is used to further optimize the position coordinates of the damage noise source and output the damage position information of the target cage.

Images