Acoustic monitoring method, device and equipment for underwater target and storage medium

Abstract

The application belongs to the technical field of data processing, and discloses an underwater target acoustic monitoring method, device, equipment and storage medium, the method comprising: receiving an underwater collected acoustic signal; performing adaptive noise reduction processing based on an environmental reference signal on the acoustic signal to filter out an environmental noise component in the acoustic signal; extracting a voiceprint feature of the acoustic signal after filtering out the environmental noise component; and determining behavior situation information of the underwater target based on the voiceprint feature. The application can be applied to a financial technology business management system, and solves the technical problem that it is difficult to effectively extract weak acoustic signals of an underwater target in an underwater environment in the prior art, which leads to the problem that the state of the underwater target cannot be accurately monitored.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology and is applied in the field of financial technology. In particular, it relates to an acoustic monitoring method, device, equipment and storage medium for underwater targets. Background Technology

[0002] Acoustic monitoring of underwater targets is a technical means to detect, identify and assess the status of underwater targets by receiving and analyzing acoustic signals radiated or reflected by them. It is widely used in marine resource development, aquaculture monitoring, and underwater equipment status monitoring.

[0003] In existing technologies, acoustic monitoring of underwater targets mainly faces the following difficulties and challenges.

[0004] I. Strong interference from underwater environmental noise makes target signal identification difficult. The underwater environment is a typical high-noise environment, with widespread natural noise such as wind, waves, and currents, as well as noise from human activities such as aerators and water pumps. The sound pressure level of these environmental noises is often much higher than the weak acoustic signals emitted by underwater targets (such as the sound of fish feeding), making target signals easily submerged. Existing simple methods such as energy threshold detection cannot effectively distinguish target signals from environmental noise, resulting in a high false alarm rate.

[0005] Taking aquaculture insurance as an example, the mechanical noise generated by aerators can reach over 100dB, while the sound of fish feeding is usually only about 40dB to 60dB, completely drowning out key biological signals such as feeding sounds. Insurance institutions find it difficult to remotely verify the survival status of aquaculture targets through acoustic monitoring.

[0006] Second, the non-stationarity and complexity of underwater target acoustic signals increase the difficulty of extraction. Underwater target acoustic signals are characterized by non-stationarity and strong time-varying properties. The feeding sounds of fish schools change dynamically with their behavior and physiological state. Using fixed-parameter filtering methods is difficult to adapt to the dynamic changes in target signals—parameters that are too wide result in insufficient noise suppression, while parameters that are too narrow may filter out the target signal altogether.

[0007] Taking aquaculture insurance as an example, the feeding behavior of fish varies greatly between feeding and non-feeding periods, and the acoustic characteristics of the same fish group are significantly different at different times, making it difficult to take into account the characteristics of fish in a traditional fixed-parameter method.

[0008] Third, traditional noise reduction methods are insufficiently adaptable to non-stationary, broadband noise. Existing noise reduction methods are mostly designed for specific types and frequency bands of noise, and their effectiveness is limited for non-stationary, broadband noise commonly found in underwater environments (such as broadband mechanical noise from aerators and variable frequency noise from water circulation pumps). When the spectrum or intensity of environmental noise changes over time, the performance of traditional methods deteriorates significantly, and some methods may even destroy the original characteristics of the target signal while suppressing noise.

[0009] Taking aquaculture insurance as an example, the noise spectrum of the aerator changes dynamically during startup, operation and shutdown. Traditional noise reduction methods with fixed filter coefficients cannot adaptively track the noise changes. Moreover, when mechanical noise and fish feeding sounds overlap in the frequency band, frequency domain filtering is difficult to effectively separate noise without damaging the target signal.

[0010] Fourth, existing underwater monitoring methods have inherent limitations. Manual pond patrols rely on subjective experience, resulting in poor timeliness and the inability to conduct continuous monitoring; physicochemical index sensors only monitor environmental parameters rather than the acoustic characteristics of the target itself, and water quality changes often lag behind changes in biological behavior; underwater optical cameras produce extremely poor imaging results in turbid waters and have limited field of view coverage. These limitations make acoustic monitoring the only feasible technical approach in many scenarios, but acoustic monitoring itself faces the severe challenge of the aforementioned noise interference.

[0011] In summary, how to effectively extract weak acoustic signals from underwater targets in an underwater environment and achieve accurate monitoring of the underwater target's status is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0012] This invention provides an acoustic monitoring method, apparatus, device, and storage medium for underwater targets, which can solve the technical problem in the prior art that it is difficult to effectively extract the weak acoustic signals of underwater targets in the underwater environment, resulting in the inability to accurately monitor the state of underwater targets.

[0013] In a first aspect, the present invention provides an acoustic monitoring method for underwater targets, comprising: Receives acoustic signals collected underwater; Adaptive noise reduction processing based on an environmental reference signal is performed on the acoustic signal to filter out the environmental noise component in the acoustic signal; Extract the acoustic signature features of the acoustic signal after filtering out the environmental noise component; Based on the aforementioned acoustic signature features, the behavioral and situational information of the underwater target is determined.

[0014] Secondly, the present invention provides an acoustic monitoring device for underwater targets, comprising: The signal receiving module is used to receive acoustic signals collected underwater. The signal noise reduction module is used to perform adaptive noise reduction processing on the acoustic signal based on the environmental reference signal to filter out the environmental noise component in the acoustic signal; The feature extraction module is used to extract the acoustic signature features of the acoustic signal after filtering out the environmental noise component; The behavior determination module is used to determine the behavior and situation information of the underwater target based on the voiceprint features.

[0015] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described acoustic monitoring method for underwater targets.

[0016] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described acoustic monitoring method for underwater targets.

[0017] Understandably, this invention performs adaptive noise reduction processing on acoustic signals based on environmental reference signals, enabling the noise reduction parameters to be adjusted in real time according to changes in environmental noise. This is a capability that traditional fixed filtering methods in the background technology lack. Specifically, the environmental reference signal provides real-time characteristics of the current noise, and the adaptive mechanism ensures that the noise reduction parameters can follow the dynamic changes of non-stationary, broadband noise. Thus, without prior knowledge of the noise statistical characteristics, it achieves dynamic suppression of strong background noise, allowing previously submerged weak target acoustic signals to emerge, thereby effectively extracting weak acoustic signals from underwater targets.

[0018] Because the present invention effectively preserves weak acoustic signals after filtering out environmental noise components from the acoustic signals, the technical means of further processing the acoustic signals based on the present invention to determine the behavioral situation information of underwater targets effectively improves the accuracy of the final state determination of underwater targets. This solves the technical problem in the prior art that it is difficult to effectively extract weak acoustic signals of underwater targets in the underwater environment, resulting in the inability to accurately monitor the state of underwater targets. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart of an acoustic monitoring method for underwater targets in one embodiment of the present invention.

[0021] Figure 2 yes Figure 1 A flowchart of step S130.

[0022] Figure 3 yes Figure 1 A flowchart of step S140.

[0023] Figure 4 yes Figure 3 A flowchart of step S1423.

[0024] Figure 5 This is a schematic diagram of an acoustic monitoring device for underwater targets in one embodiment of the present invention.

[0025] Figure 6 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention.

[0026] Figure 7 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Figure 1 A flowchart of an acoustic monitoring method for underwater targets provided in an embodiment of the present invention is shown below. Figure 1 As shown, the acoustic monitoring method for underwater targets provided in this embodiment of the invention includes the following steps.

[0029] Step S110: Receive the acoustic signals collected underwater; In one specific implementation, acoustic signals collected underwater can be received using acoustic sensors deployed in the underwater monitoring environment. The acoustic sensors can be hydrophones or other transducers capable of converting underwater sound pressure changes into electrical signals. The acoustic sensors can be fixed to the seabed, suspended in the water, or mounted on an underwater mobile platform; there can be one or more of them. The received acoustic signals are amplified, filtered, and converted from analog to digital to form digitized acoustic data for subsequent processing. Acoustic signal acquisition can be continuous or intermittent at preset time intervals.

[0030] Understandably, this step completes the acquisition of raw data on the underwater acoustic environment, providing a data source for subsequent signal processing and information extraction. If underwater acoustic signals cannot be effectively received, the entire monitoring process will fail to start due to a lack of input. In one optional implementation, the received acoustic signal can be real-time streaming data or a recorded acoustic data file read from a storage medium; the specific source does not affect the implementation of this method.

[0031] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, hydrophones are deployed around the aquaculture cages to receive acoustic signals from the water in real time. These signals include the sounds of the farmed fish activity and the surrounding water sounds, providing raw data for subsequent analysis of the fish's survival status and helping insurance companies to remotely monitor and assess the risks of the farmed assets.

[0032] Step S120: Perform adaptive noise reduction processing based on environmental reference signal on the acoustic signal to filter out the environmental noise component in the acoustic signal; In one specific implementation, after receiving the underwater acoustic signal, adaptive noise reduction processing based on an environmental reference signal is performed on the acoustic signal. This adaptive noise reduction processing refers to using a reference signal related to environmental noise to estimate and remove environmental noise components from the received acoustic signal through adaptive filtering, thereby preserving the target acoustic signal. The environmental reference signal can be pre-acquired or acquired synchronously with the target signal, and its content should reflect the noise characteristics of the current underwater environment. The core of adaptive noise reduction processing lies in dynamically adjusting the noise reduction parameters based on the statistical characteristics of the acoustic signal and the environmental reference signal, so that the filtered noise components are as close as possible to the actual environmental noise, achieving the best noise reduction effect.

[0033] In one optional implementation, the adaptive noise reduction process can employ an Adaptive Noise Cancellation (ANC) framework. This framework uses the received acoustic signal as the main input and the environmental reference signal as the reference input, and an adaptive filter is used to estimate and eliminate noise components. The update algorithm for the adaptive filter can be any iterative algorithm based on the minimum mean square error criterion; the specific algorithm type can be selected according to actual needs.

[0034] In a specific example within the property insurance sector, continuing with the aforementioned scenario of marine ranching insurance, the received acoustic signals are mixed with environmental noise such as water flow and waves. By simultaneously acquiring environmental reference signals (e.g., acoustic signals from a distance from the aquaculture cages), and using adaptive noise reduction processing to filter out environmental noise, a clearer acoustic signal of fish activity is obtained, providing a clean signal basis for accurately judging the behavior and health status of the fish.

[0035] In some embodiments of the present invention, step S110 includes the following steps.

[0036] Step S111: Receive the underwater mixed signal acquired by the main hydrophone, wherein the mixed signal includes the target acoustic signal and the ambient noise component; In one specific implementation, a main hydrophone collects acoustic signals from the area where the underwater target is located. The main hydrophone can be deployed near the area where the underwater target is active, such as near the underwater equipment, aquaculture cages, or underwater structures to be monitored. The main hydrophone converts the received sound pressure signal into an electrical signal, which is a mixed signal. This mixed signal includes both the target's acoustic signals (e.g., radiated noise, motion sounds, or communication signals from the underwater target) and various background noise components present in the underwater environment, such as water flow noise, wave noise, and other biological activity noise. The mixed signal collected by the main hydrophone is amplified and converted from analog to digital to become raw data suitable for digital signal processing.

[0037] In one alternative implementation, the number of main hydrophones can be one or more. When multiple main hydrophones are used, the signals from each channel can be processed separately, or the multi-channel signals can be beamformed and then noise-reduced. The specific configuration can be flexibly selected according to the geometry of the monitoring area and the target signal strength.

[0038] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, a main hydrophone is deployed inside the aquaculture cage near the area where the fish are active. This hydrophone collects a mixed signal containing sounds of the fish swimming and feeding, as well as environmental noise such as water flow and waves. This mixed signal serves as the foundational data source for subsequently assessing the health status of the fish.

[0039] Step S112: Receive the ambient noise signal acquired by the reference hydrophone; In one specific implementation, a reference hydrophone is used to acquire noise signals from the underwater environment. The reference hydrophone can be deployed in the same aquatic environment as the main hydrophone, but its location should be as far away as possible from the source direction or area of ​​the target acoustic signal to ensure that the acquired signal is dominated by environmental noise components, with the target acoustic signal components being as weak or absent as possible. The environmental noise signal acquired by the reference hydrophone is correlated with the environmental noise components in the mixed signal received by the main hydrophone (e.g., originating from the same water flow noise source or the same far-field interference source), thus serving as a reference input for subsequent adaptive noise reduction processing.

[0040] In one alternative implementation, the reference hydrophone and the main hydrophone can maintain a certain spatial distance. This distance should not be too close (to avoid the target signal being mixed into the reference signal) nor too far (to avoid reducing the correlation of environmental noise). The specific placement distance can be optimized according to the frequency characteristics of the target signal and the underwater sound propagation conditions.

[0041] In a specific example within the property insurance sector, continuing the scenario of marine ranching insurance, reference hydrophones are deployed outside the aquaculture cages, away from the fish's activity area, to collect environmental noise signals, primarily consisting of water flow and waves. The sound of fish activity is extremely weak in this reference signal, which mainly reflects the environmental sound field characteristics of the aquaculture area.

[0042] Step S120 includes the following steps.

[0043] Step S121: The environmental noise signal is filtered using filtering coefficients to generate an estimated signal of the environmental noise component in the mixed signal; In one specific implementation, the ambient noise signal collected by the reference hydrophone is input to an adaptive filter. The adaptive filter filters the ambient noise signal according to the current filtering coefficients. The filtering coefficients are a set of parameters used by the filter to weight each tap of the input signal. After filtering, the input reference signal is linearly transformed and output as an estimated signal. This estimated signal is an approximate estimate of the true ambient noise component in the mixed signal from the main hydrophone. The filtering process can employ a finite impulse response (FIR) filter structure or an infinite impulse response (IIR) filter structure; an FIR filter is preferred to ensure system stability.

[0044] In one alternative implementation, the order of the filter and the initial values ​​of the filter coefficients can be set based on prior knowledge or previous calibration data. The filter coefficients can be a set of real or complex numbers, and their values ​​determine the amount of amplitude and phase adjustment of the reference signal by the filter, thereby affecting how closely the estimated signal approximates the real environmental noise components.

[0045] In a specific example in the property insurance field, continuing the scenario of marine ranching insurance mentioned above, the environmental noise signal (mainly water flow noise) collected by the reference hydrophone is input into an adaptive filter. The filter uses the current filtering coefficients to perform a weighted summation of the reference signal to generate an estimate of the water flow noise component in the mixed signal inside the aquaculture cage.

[0046] Step S122: Subtract the estimated signal from the mixed signal to obtain the denoised target acoustic signal, and use the denoised target acoustic signal as the error signal; In one specific implementation, the mixed signal acquired by the main hydrophone is subtracted from the estimated signal generated in step S121, that is, the estimated value of the environmental noise component is subtracted from the mixed signal. Since the mixed signal contains both the target acoustic signal and the environmental noise component, and the estimated signal is an approximation of the environmental noise component, the environmental noise component is suppressed and the target acoustic signal is enhanced in the result of the subtraction. This result is the noise-reduced target acoustic signal.

[0047] Simultaneously, the denoised target acoustic signal is fed back to the adaptive filter as an error signal. The error signal reflects the residual error between the estimated signal and the real environmental noise component in the mixed signal. When the residual error is large, it indicates that the current filter coefficients are not accurately estimating the noise; when the residual error approaches zero, it indicates that the estimated signal has approximated the real environmental noise component well.

[0048] In one alternative implementation, the subtraction operation can be performed in the numeric field by subtracting samples one by one. If a multi-channel configuration is used, each channel performs the subtraction operation independently.

[0049] In a specific example within the property insurance sector, continuing with the aforementioned marine ranching insurance scenario, the mixed signal collected by the main hydrophone, containing both fish sounds and water flow noise, is subtracted from the water flow noise estimation signal generated by adaptive filtering to obtain the denoised acoustic signal of fish activity. This denoised signal also serves as an error signal, used to evaluate the accuracy of the water flow noise estimation.

[0050] Step S123: Adjust the filtering coefficients according to the error signal so that the subsequently generated estimated signal approximates the environmental noise component in the mixed signal.

[0051] In one specific implementation, the adaptive filter adjusts and updates the current filter coefficients according to a preset adaptive algorithm based on the error signal fed back in step S122. The goal of the adjustment is to minimize the statistics of the error signal (such as the mean square value), so that the difference between the output estimated signal of the filter and the real environmental noise component in the mixed signal continuously decreases. As the filter coefficients are iteratively updated, the estimated signal gradually approaches the real environmental noise component, and the residual noise in the denoised target acoustic signal becomes smaller and smaller, thereby achieving effective noise suppression.

[0052] In one optional implementation, the adaptive algorithm for adjusting the filter coefficients can employ the Least Mean Square (LMS) algorithm, which has low computational complexity and is easy to implement. Alternatively, the Normalized Least Mean Square (NLMS) algorithm can be used, which improves the convergence stability and speed by normalizing the step size factor over the input signal energy. In another optional implementation, the Recursive Least Square (RLS) algorithm can be used, which has a faster convergence speed but requires more computation, making it suitable for scenarios with high convergence speed requirements. The above algorithms are merely illustrative examples; the specific algorithm type used can be selected and replaced according to the statistical characteristics of the noise and real-time requirements in the actual application scenario.

[0053] In a specific example within the property insurance sector, continuing with the aforementioned marine ranching insurance scenario, the adaptive filter adjusts its coefficients using the LMS algorithm based on the magnitude of residual noise in the denoised acoustic signal of fish activity (i.e., the error signal). When significant water flow noise is still audible in the fish sound signal, it indicates that the filter coefficients have not yet converged, and the algorithm continues to adjust the coefficients. As the coefficients iterate, the water flow noise in the output signal gradually weakens, ultimately achieving effective filtering of water flow noise and making the fish activity sound clearly identifiable.

[0054] It is understood that the underwater target acoustic monitoring method described in this embodiment achieves noise cancellation with closed-loop adaptive capability by setting a main hydrophone to collect a mixed signal and a reference hydrophone to collect an environmental noise signal in step S110, and by using filtering coefficients to filter the environmental noise signal to generate an estimated signal in step S120, subtracting the estimated signal from the mixed signal to obtain the denoised target acoustic signal and using it as an error signal, and adjusting the filtering coefficients according to the error signal to make the subsequent estimated signal approximate the real environmental noise component. This method, through the dual-hydrophone configuration, acquires a mixed signal containing the target signal and noise, as well as a pure environmental noise reference signal, avoiding the problem of difficulty in distinguishing between the target and noise in single-channel noise reduction methods. By using the denoised signal as an error signal as feedback and dynamically adjusting the filtering coefficients in a closed-loop mechanism, the filter can track the time-varying changes in the noise characteristics of the underwater environment. Even if the spectrum or intensity of the environmental noise changes over time, the filtering coefficients can adaptively adjust accordingly, always maintaining a good noise reduction effect, thereby continuously outputting a high-quality target acoustic signal.

[0055] Step S130: Extract the acoustic signature features of the acoustic signal after filtering out the ambient noise component; In one specific implementation, after adaptive noise reduction, acoustic signature features are extracted from the acoustic signal after filtering out environmental noise components. These acoustic signature features are quantifiable parameters that characterize the essential properties of the target acoustic signal, typically represented as a set of feature vectors or feature matrices. The acoustic signature features should possess inter-class separability and intra-class stability; that is, the acoustic signature features of different underwater targets or the same target under different behavioral states should have significant differences, while features collected multiple times under the same state should remain relatively stable. Specific methods for extracting acoustic signature features can employ time-domain features, frequency-domain features, or joint time-frequency features, such as spectral envelope, cepstral coefficients, linear prediction coefficients, wavelet transform coefficients, etc. This step does not limit the specific feature type, as long as it extracts feature quantities from the denoised acoustic signal that can be used for subsequent discrimination.

[0056] In one alternative implementation, the acoustic signal can be first processed by framing and windowing, then feature parameters can be extracted from each frame, and then the features of each frame can be statistically aggregated (such as calculating the mean, variance, or temporal splicing) to form a fixed-length voiceprint feature vector.

[0057] Understandably, while the signal-to-noise ratio is improved after noise reduction, the signal is still a one-dimensional time-domain waveform that cannot be directly used for state determination. This step extracts acoustic signature features, condensing the high-dimensional, redundant original acoustic signal into a discriminative feature vector, thus achieving a quantitative characterization of the essential attributes of the target signal. This transformation is crucial; the quantifiability of the features makes subsequent automated determination possible, avoiding the drawbacks of "subjectivity and lag in human experience-based judgment" in the background technology. In a specific example in the property insurance field, continuing the scenario of marine ranching insurance, acoustic signature features, such as spectral envelope features, are extracted from the acoustic signal of fish activity after noise reduction. Fish in different health states have different spectral distributions of swimming and feeding sounds; extracting these features can provide quantitative indicators for subsequent judgment of fish behavior and status, assisting insurance companies in assessing aquaculture risks.

[0058] In some embodiments of the present invention, such as Figure 2 As shown, step S130 includes the following steps.

[0059] Step S131: Transform the one-dimensional time-domain waveform of the acoustic signal into a two-dimensional time-spectrum diagram; In one specific implementation, after adaptive noise reduction, the denoised target acoustic signal undergoes time-frequency transformation. The denoised target acoustic signal is represented in the time domain as a one-dimensional waveform that varies with time, directly reflecting the change in sound pressure over time. To simultaneously acquire the signal's distribution characteristics in both time and frequency dimensions, the one-dimensional time-domain waveform is transformed into a two-dimensional time-frequency spectrum.

[0060] Specifically, the denoised acoustic signal can be segmented into frames, dividing the continuous long-time signal into multiple short-time frames. Each frame can be considered as an approximately stationary signal segment. A window function (such as a Hamming or Hanning window) is applied to each frame to reduce spectral leakage. Then, a Fourier transform is performed on each windowed frame to obtain its spectrum. Arranging the spectra of each frame in chronological order creates a two-dimensional time-spectrum graph. The horizontal axis of this time-spectrum graph represents time, and the vertical axis represents frequency. The color or grayscale value of each pixel represents the energy or amplitude value at that frequency at that moment. Through this transformation, frequency evolution patterns that are difficult to observe in the time domain can be visually presented on the time-spectrum graph.

[0061] In an optional implementation, the Fourier transform can employ a Fast Fourier Transform (FFT) to improve computational efficiency. The temporal and frequency resolutions of the time-spectrum are affected by the frame length and window function. Appropriate framing parameters can be selected based on the characteristics of the target acoustic signal, for example, a frame length of 20 to 50 milliseconds and a frame shift of half or a quarter of the frame length.

[0062] In a specific example within the property insurance sector, continuing with the aforementioned marine ranching insurance scenario, the one-dimensional time-domain waveform of the denoised acoustic signal of fish activity is framed and subjected to short-time Fourier transform to generate a two-dimensional time-spectrum diagram of the fish activity sound. This spectrum clearly displays the frequency distribution characteristics of the fish feeding and swimming sounds at different time points; for example, the high-frequency pulses generated during feeding appear as broadband energy distributions within specific time regions on the spectrum.

[0063] Step S132: Extract the Mel frequency cepstral coefficients from the two-dimensional time-spectrum graph as the acoustic signature features of the acoustic signal.

[0064] In one specific implementation, after obtaining the two-dimensional time-spectrum diagram of the acoustic signal, Mel-Frequency Cepstral Coefficients (MFCCs) are extracted from the time-spectrum diagram as the voiceprint feature of the acoustic signal. MFCCs are acoustic feature parameters based on the characteristics of human auditory perception, effectively characterizing the spectral envelope information of acoustic signals and having wide applications in the field of acoustic recognition. The extraction process may include: mapping the linear frequencies in the time-spectrum diagram to a Mel-frequency scale; arranging several triangular filter banks on the Mel-frequency scale; filtering and weighting the spectral energy to obtain the output energy value of each filter; taking the logarithm of the energy value and performing a Discrete Cosine Transform (DCT) to obtain the Mel-Frequency Cepstral Coefficients. Typically, the first few dimensions of the coefficients (e.g., 12 or 13 dimensions) are taken as the feature vector, which is the voiceprint feature.

[0065] It should be noted that the Mel-scale mapping and filter bank design involved in the specific extraction process of Mel frequency cepstral coefficients are all conventional techniques known in the field. This step only requires the extraction of the coefficients as features from the time-spectrum diagram, and does not limit the specific parameters in the extraction process (such as the number of filters, the order of DCT, etc.), as long as the Mel frequency cepstral coefficients can be obtained.

[0066] In an optional implementation, in addition to extracting the Mel frequency cepstral coefficients, first-order and second-order difference operations can be performed on the extracted coefficients. The difference coefficients are then combined with the static coefficients to form a feature vector containing dynamic information, thereby better characterizing the time-varying properties of the acoustic signal. This method can be used as a supplement to this step but does not affect the core protection scope of this step.

[0067] In a specific example within the property insurance sector, continuing the scenario of marine ranching insurance, Mel-frequency cepstral coefficients (MFCCs) are extracted from the two-dimensional time-spectrum image of the acoustic signals of fish activity, serving as the acoustic signature of the fish school. Different fish species and fish schools in different health states have different acoustic spectral envelopes. The extracted Mel-frequency cepstral coefficients can quantify these differences into numerical features. For example, the feeding sounds of healthy fish schools have abundant high-frequency components, resulting in larger values ​​for higher-order coefficients in their MFCCs, while the feeding sounds of sick or weak fish schools have attenuated high-frequency components, showing significant differences in MFCC characteristics.

[0068] Step S140: Based on the acoustic signature features, determine the behavioral situation information of the underwater target; In one specific implementation, after extracting the voiceprint features, the behavioral situation information of the underwater target is determined based on these features. The behavioral situation information refers to information characterizing the current behavioral state and development trend of the underwater target, such as the target's activity level, direction of movement, feeding behavior, habitat status, and abnormal behavior. The determination process can employ methods such as template matching, classifier discrimination, regression prediction, or rule-based reasoning to compare the extracted voiceprint features with preset reference features or determination criteria for various behavioral situations, thereby determining the behavioral situation category or situation parameters corresponding to the current voiceprint features. This determination can be static (based on a single feature) or dynamic (based on the changing trends of features across multiple frames).

[0069] In one optional implementation, a behavioral posture database can be pre-built to store typical voiceprint feature ranges or feature patterns corresponding to different behavioral postures. During the determination, the similarity between the currently extracted voiceprint features and the patterns in the database is calculated, and the most matching behavioral posture is selected as the determination result.

[0070] Understandably, this step leverages the inherent correlation between extracted acoustic signature features and the behavioral state of underwater targets to automatically identify the target's current behavioral state through pre-defined judgment logic. This automated judgment replaces traditional monitoring methods that rely on manual observation or simple threshold detection, achieving objective, real-time, and continuous perception of the underwater target's state and solving the fundamental problem in the background technology of "inability to accurately monitor the state of underwater targets." In a specific example in the property insurance field, continuing the scenario of marine ranching insurance, the current behavioral state of the fish school is determined based on the extracted acoustic signature features of the fish school. For example, if the acoustic signature features show that the fish school's activity sounds are active and regular, it is judged as a "normal feeding state"; if the activity sounds are significantly weakened or show abnormal intermittent periods, it is judged as an "abnormal state." This judgment result can directly reflect the survival status of the farmed fish school, providing a basis for insurance underwriting and claims.

[0071] It is understood that the embodiments of the present invention, by performing adaptive noise reduction processing on acoustic signals based on environmental reference signals, enable the noise reduction parameters to be adjusted in real time according to changes in environmental noise. This is a capability that traditional fixed filtering methods in the background art do not possess. Specifically, the environmental reference signal provides the real-time characteristics of the current noise, and the adaptive mechanism ensures that the noise reduction parameters can follow the dynamic changes of non-stationary, broadband noise. Thus, without prior knowledge of the noise statistical characteristics, dynamic suppression of strong background noise is achieved, allowing previously submerged weak target acoustic signals to emerge, thereby effectively extracting weak acoustic signals of underwater targets.

[0072] Because the embodiments of the present invention effectively retain weak acoustic signals after filtering out environmental noise components from the acoustic signals, the technical means of further processing the acoustic signals based on the embodiments of the present invention to determine the behavioral situation information of underwater targets effectively improves the accuracy of the final state determination of underwater targets. This solves the technical problem in the prior art that it is difficult to effectively extract weak acoustic signals of underwater targets in the underwater environment, resulting in the inability to accurately monitor the state of underwater targets.

[0073] In some embodiments of the present invention, step S140 includes the following steps.

[0074] Step S1411: Generate the acoustic activity index of the underwater target based on the acoustic features; In one specific implementation, after extracting the acoustic signature features of the acoustic signal, an acoustic signature activity index for the underwater target is generated based on these features. The acoustic signature activity index is a numerical parameter used to quantify the intensity and frequency of the underwater target's acoustic activity within the current time period. This index can be calculated based on one or more dimensions of the acoustic signature features, such as energy, frequency distribution, and temporal rate of change. The value of the acoustic signature activity index reflects the activity level of the underwater target—a higher value indicates more frequent and intense target activity, while a lower value indicates less frequent and weaker target activity.

[0075] In one alternative implementation, the voiceprint activity index can be generated by weighted summation or nonlinear mapping of the voiceprint feature vector, or by using a pre-trained regression model to map voiceprint features into activity scores. The activity index can be normalized to a preset numerical range (e.g., 0 to 100) to facilitate the uniform setting and comparison of subsequent thresholds.

[0076] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, an acoustic activity index for fish schools is generated based on extracted acoustic signature characteristics (such as Mel-frequency cepstral coefficients and energy in each frequency band). This index comprehensively reflects the feeding activity and swimming intensity of the fish school at the current moment. When the fish school is feeding normally, the activity index remains at a high level; when the fish school stops feeding or its activity weakens, the activity index drops significantly. Insurance companies can use this activity index as an objective basis for assessing the survival status of the farmed animals.

[0077] Step S1412: When the voiceprint activity index is continuously lower than the first preset threshold within a preset time window, it is determined that the underwater target is in a continuous silent state. In one specific implementation, the generated acoustic activity index is compared with a preset first threshold. The first preset threshold is a lower limit threshold used to determine whether an underwater target is in a "continuous silence" state, and its value can be set according to the normal activity level of different monitored objects. When the acoustic activity index remains below the first preset threshold within a preset time window, it indicates that the underwater target lacks normal acoustic activity for a period of time, and it is determined that it is in a continuous silence state.

[0078] The preset time window can be flexibly set according to the application scenario, for example, it can be several minutes, tens of minutes, or several hours. The specific duration can be adjusted according to the biological characteristics of the monitored target and the timeliness requirements of risk response. Continuous silence indicates an abnormal cessation of the underwater target's behavior, which may be due to target death, serious disease, drastic environmental changes, etc. Continuous silence usually indicates a serious abnormal event that has occurred or is occurring, requiring immediate attention and intervention.

[0079] In one optional implementation, the first preset threshold can be set based on historical activity statistics under normal conditions (such as the mean minus two standard deviations), or it can be calibrated based on expert experience. By using the persistence of the activity index within the time window as the criterion, false triggering caused by transient noise interference or occasional signal loss can be effectively avoided.

[0080] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, a first preset threshold is set at 30% of the average normal activity level, with a preset time window of 30 minutes. When the vocalization activity index of the fish school remains below 30% of the average normal activity level for 30 consecutive minutes, the fish school is determined to be in a state of continuous inactivity. This situation may indicate a large-scale fish mortality or severe oxygen depletion, constituting an emergency red-line event, allowing insurance claims personnel to promptly initiate on-site verification procedures.

[0081] Step S1413: When the voiceprint activity index exceeds the second preset threshold within a preset time window, it is determined that the underwater target is in an explosive agitation state. In one specific implementation, the generated acoustic activity index is compared with a preset second preset threshold. The second preset threshold is an upper limit threshold used to determine whether an underwater target is in a state of "explosive agitation." Its value can also be set according to the normal activity level of different monitored objects, and is usually higher than the upper limit of normal activity. When the acoustic activity index exceeds the second preset threshold within a preset time window, it indicates that the underwater target has exhibited abnormally intense acoustic activity, and is determined to be in an explosive agitation state.

[0082] Explosive agitation indicates a dramatic and abnormal fluctuation in the behavior of an underwater target, possibly caused by fright, stress, illness, pain, or struggle due to lack of oxygen. In contrast to continuous silence, explosive agitation is also an extreme anomaly requiring immediate attention and intervention. By simultaneously setting a first preset threshold (too low) and a second preset threshold (too high), a complete monitoring boundary for the acoustic activity of underwater targets is established.

[0083] In one alternative implementation, the second preset threshold can be set based on historical activity statistics under normal conditions (such as the mean plus two or three standard deviations), or it can be calibrated based on expert experience. The second preset threshold is usually much higher than the first preset threshold, and the two monitor the abnormal behavior of underwater targets bidirectionally from the two directions of "too still" and "too active".

[0084] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, a second preset threshold is set at three times the average normal activity level, with a preset time window of 10 minutes. When the vocalization activity index of the fish school exceeds three times the average normal activity level for 10 consecutive minutes, the fish school is determined to be in an explosively agitated state. This situation may indicate that the fish school has been frightened (e.g., by predators or human disturbance) or is experiencing an emergency such as struggling due to lack of oxygen, requiring immediate triggering of an alert and dispatch of personnel for investigation.

[0085] Following step S140, the following steps are also included.

[0086] Step S160: Generate corresponding red line warning information based on the continuous silent state or the explosive turbulent state.

[0087] In one specific implementation, when it is determined that an underwater target is in a continuous silent state or an explosively agitated state, a corresponding red line warning message is generated. This red line warning message is an emergency-level warning message, distinct from general prompts or ordinary warnings, indicating that an extreme abnormal event has occurred or is occurring at the underwater target, requiring immediate intervention or response measures. The generation of the red line warning message is an independent output step executed after the behavioral situation determination is completed.

[0088] Red line warning messages can include one or more of the following: warning type (continuous silence or explosive agitation), target identifier, occurrence time, activity index value, preset threshold, and suggested response measures. Depending on the warning type, red line warning messages can trigger different response procedures—continuous silence warnings may trigger a "death verification" procedure, while explosive agitation warnings may trigger a "stress intervention" or "emergency rescue" procedure.

[0089] In one optional implementation, red-line warning information can be sent to the terminal devices of relevant personnel in real time via various methods such as SMS, application push notifications, and audible and visual alarms. Red-line warning information has a higher priority than ordinary warning information to ensure that relevant personnel can be informed and respond immediately. In another optional implementation, the red-line warning information may also include a timestamp and corresponding voiceprint feature data as objective evidence for subsequent verification and tracing.

[0090] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, when the system determines that the fish population is in a state of continuous inactivity, it generates a "continuous inactivity red line warning" message, which is pushed to the mobile phones of fish farmers and insurance claims personnel via SMS and application, prompting, "The fish population in the XX cage has been inactive for more than 30 minutes; please go to the site immediately for verification." When the system determines that the fish population is in a state of explosive agitation, it generates an "explosive agitation red line warning" message, prompting, "The fish population in the XX cage is exhibiting abnormal agitation; please immediately investigate for oxygen deficiency or invasive risks." Insurance companies can use the red line warning records as objective evidence for claims verification, effectively identifying and preventing insurance fraud risks.

[0091] It is understood that the underwater target acoustic monitoring method described in this embodiment generates an acoustic activity index based on acoustic characteristics, compares this index with a first preset threshold and a second preset threshold, determines continuous silent states and explosive agitation states, and generates corresponding red line warning information. This achieves two-way, real-time monitoring and emergency warning of extreme abnormal events of underwater targets. By setting thresholds for "too low" and "too high" directions, this method forms a complete monitoring boundary for the acoustic activity of underwater targets. Whether the target's activity suddenly decreases or becomes abnormally agitated, it can be captured in a timely manner, avoiding missed reports that may occur with one-sided monitoring. Through a continuous judgment mechanism within a preset time window, it effectively avoids false triggers caused by instantaneous noise interference or occasional signal loss, improving the reliability and accuracy of the warning. The generated red line warning information, as an emergency level warning output distinct from ordinary situation information, ensures that relevant personnel are informed of extreme abnormal events immediately and can respond promptly, buying valuable time to reduce losses.

[0092] In some embodiments of the present invention, such as Figure 3 As shown, step S140 includes the following steps.

[0093] Step S1421: Feed the voiceprint features into the temporal behavior analysis model, and perform time series modeling on the voiceprint features based on the self-attention mechanism through the temporal behavior analysis model to capture the change pattern of the voiceprint features in the temporal dimension. In one specific implementation, after extracting the acoustic signature features from the acoustic signal, these features are fed into a time-series behavior analysis model for processing. The time-series behavior analysis model is a computational model capable of processing sequential data, used to model and analyze the evolution of acoustic signature features over time. The values ​​of the acoustic signature features at different times constitute a time series, which reflects the dynamic changes in the behavior of underwater targets over time.

[0094] Specifically, the temporal behavior analysis model uses a self-attention mechanism to model the time series of voiceprint features. The self-attention mechanism can capture the dependencies between any two time steps in the sequence, without being limited by local windows, thus effectively identifying the changing patterns and periodic regularities of voiceprint features over long periods. Through the calculation of the self-attention mechanism, the model can output the changing patterns of voiceprint features in the temporal dimension, which can manifest as trends, periodic fluctuations, abrupt changes, or pattern transitions in feature values. Compared to traditional temporal models that can only process local information, the self-attention mechanism can establish correlations between features globally, giving it an advantage in capturing the long-term evolutionary trends of underwater target behavior.

[0095] In one alternative implementation, the temporal behavior analysis model can employ a Transformer architecture, which is entirely built on a self-attention mechanism and can process sequential data in parallel, offering advantages in capturing long-range dependencies. Alternatively, the model can also employ other variant network structures based on the self-attention mechanism; the specific network type can be selected based on the real-time requirements and computational resource conditions of the actual application scenario.

[0096] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, fish vocalization features (such as Mel-frequency cepstral coefficient vectors) collected over multiple consecutive days are arranged chronologically to form a time series of vocalization features. This series is then fed into a time-series behavior analysis model based on a self-attention mechanism. The model performs global correlation analysis on the features at each time step in the series to capture the changing trends of fish activity over continuous time periods, such as whether the time of the daily feeding peak shifts or whether the feeding intensity shows a continuous downward trend.

[0097] Step S1422: Compare the change pattern with preset multiple behavioral state patterns; In one specific implementation, after the temporal behavior analysis model captures the changing patterns of acoustic signature features, these patterns are compared with a set of preset behavioral situation patterns. These preset behavioral situation patterns are predefined reference patterns that characterize different behavioral states of underwater targets, with each pattern corresponding to a set of typical acoustic signature change patterns. By matching or calculating the similarity between the currently observed change patterns and these preset patterns, the type of the current behavioral situation can be determined.

[0098] In one optional implementation, multiple preset behavioral patterns can be obtained through cluster analysis or machine learning training on a large amount of historical voiceprint feature data and its corresponding known behavioral states. Different types of patterns can be set for different application scenarios. For example, in aquaculture monitoring scenarios, active feeding mode, reduced feeding mode, and silent mode can be set; in equipment monitoring scenarios, normal operation mode and abnormal vibration mode can be set. The number and specific types of preset patterns can be flexibly configured according to actual monitoring needs.

[0099] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, preset behavioral patterns include, but are not limited to: normal feeding mode (active feeding sounds, stable frequency, and moderate intensity), reduced feeding mode (sparse feeding sounds, decreased frequency, and weakened intensity), silent mode (no obvious bioacoustic activity), and agitated mode (abnormally intense acoustic activity with a surge in intensity). The changing patterns of the fish school's acoustic signatures are compared one by one with these preset patterns to determine which pattern's characteristics the current trend of the fish school most closely resembles.

[0100] Step S1423: Determine the behavioral situation information of the underwater target based on the comparison results.

[0101] In one specific implementation, the underwater target's behavioral situation information is determined based on the comparison results between the changing patterns in step S1422 and multiple preset behavioral situation patterns. This behavioral situation information characterizes the underwater target's behavioral state type and its changing trend within the current time period. The comparison results reflect the similarity between the current changing patterns and each preset pattern; the final behavioral situation information is determined based on the pattern with the highest similarity or that meets specific matching conditions.

[0102] In one optional implementation, when the similarity between the change pattern and a preset pattern exceeds a preset matching threshold, the behavioral state corresponding to that preset pattern is determined as the current behavioral situation information. In another optional implementation, when the similarity between the change pattern and multiple preset patterns does not reach a matching threshold, the current behavioral situation can be determined as an "unknown situation" or a "transitional situation," triggering a data review or manual confirmation process. In yet another optional implementation, the determined behavioral situation information may include auxiliary parameters such as a timestamp, duration, and confidence level to enrich the descriptive dimensions of the situation information.

[0103] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, the current behavioral status of a fish population is determined based on comparison results. For instance, when the model captures a pattern of change with a similarity of over 85% to a "reduced feeding pattern," it is determined that the fish population is currently in a reduced feeding state, and the duration and confidence level of this state are output. The insurance company can then use this behavioral status information to assess the health risk level of the fish population and determine whether to initiate an early warning or verification process.

[0104] It is understood that the underwater target acoustic monitoring method described in this embodiment, by feeding acoustic features into a time-series behavior analysis model based on a self-attention mechanism for time-series modeling and capturing change patterns, compares these change patterns with multiple preset behavior posture patterns, and determines the underwater target's behavior posture information based on the comparison results, thus achieving intelligent and automated judgment of the underwater target's behavior posture. This method utilizes the characteristic of the self-attention mechanism to capture the dependency relationship between any two time steps in a sequence, effectively overcoming the limitation of traditional time-series models that can only process local information. It can identify the change patterns and pattern transitions of acoustic features in the long-term dimension globally. By comparing the change patterns with multiple preset behavior posture patterns, the judgment process has clear reference standards and interpretability, avoiding the problem of lacking reference benchmarks when judging based on a single threshold or simple classifier. Determining the behavior posture information based on the comparison results achieves automatic conversion from raw acoustic features to high-level semantic information, reducing reliance on manual interpretation and subjectivity, and improving the efficiency and accuracy of underwater target behavior posture judgment.

[0105] In some embodiments of the present invention, the preset multiple behavioral posture patterns include a first mode and a second mode of feeding rhythm; such as Figure 4 As shown, step S1423 includes: Step S14231: When the change pattern indicates that the feeding rhythm of the voiceprint feature within a preset time window has shifted from the first mode to the second mode, it is determined that the underwater target is in an abnormal precursor state.

[0106] In one specific implementation, the preset multiple behavioral patterns include a first mode and a second mode of feeding rhythm. The first mode characterizes the feeding rhythm characteristics of the underwater target under normal conditions, while the second mode characterizes the feeding rhythm characteristics of the underwater target under abnormal conditions. The first mode and the second mode represent two different stages in the transition of the feeding rhythm from normal to abnormal.

[0107] After the temporal behavior analysis model captures the changing patterns of voiceprint features, it compares these patterns with multiple preset behavioral state patterns. When the changing pattern indicates that the feeding rhythm of the voiceprint features has shifted from the first pattern to the second pattern within a preset time window, the underwater target is determined to be in an abnormal precursor state.

[0108] The term "migration" refers to the process by which the feeding rhythm of an underwater target gradually or suddenly shifts from a normal state (first mode) to an abnormal state (second mode) over a continuous period of time. This pattern migration reflects a substantial change in the behavior of the underwater target and serves as an early signal of an impending abnormal event. The preset time window refers to the time range used to determine pattern migration, which can be flexibly set according to different monitoring targets and application scenarios; for example, it can be several hours, several days, or longer. Within the preset time window, if the characteristic parameters of the feeding rhythm are observed to continuously deviate from the typical range of the first mode and gradually enter the characteristic range of the second mode, then pattern migration is considered to have occurred.

[0109] The aforementioned anomaly precursors refer to the early signs that an underwater target is about to experience an abnormal state. During this stage, certain behavioral indicators of the underwater target show deviations from normal trends, but have not yet fully deteriorated to an irreversible abnormal state. Identifying these anomaly precursors means that preventative interventions can be taken before the anomaly fully occurs, thereby avoiding or mitigating losses.

[0110] In one optional implementation, the determination of anomaly precursors can also incorporate parameters such as the speed, amplitude, and duration of pattern migration. For example, the faster the migration speed, the greater the amplitude, and the longer the duration, the higher the confidence level of the anomaly precursor, and the corresponding warning level is also increased. In another optional implementation, the specific characteristic parameters of the first and second modes can be obtained through cluster analysis or expert annotation of historical voiceprint data. The mode parameters may differ for different aquaculture areas and different fish species, and can be adaptively configured according to specific application scenarios.

[0111] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, the pre-defined behavioral patterns include: Mode 1 (normal feeding mode - active feeding sounds, stable frequency, and moderate intensity) and Mode 2 (reduced feeding mode - sparse feeding sounds, significantly lower feeding frequency than normal, and weakened intensity). The temporal behavioral analysis model analyzes the changes in the fish's vocal characteristics over several consecutive days. When it identifies a shift in the fish's feeding rhythm from Mode 1 to Mode 2—for example, a continuous decline in feeding activity over three consecutive days, or a decrease in feeding frequency of more than 40% from normal levels—it determines that the fish are in an abnormal premonition state. This determination indicates that the fish may be facing health problems or mortality risks. Based on this, insurance companies can issue early warnings to fish farmers before large-scale mortality events occur, advising them to promptly investigate factors such as water quality and disease, thereby reducing the risk of insurance payouts. Simultaneously, the record of this abnormal premonition state determination can also serve as objective data for insurance companies' risk assessment and underwriting decisions.

[0112] It is understood that the acoustic monitoring method for underwater targets described in this embodiment achieves early prediction of abnormal underwater target states by setting multiple preset behavioral state modes, including at least a first mode and a second mode of feeding rhythm, and determining that the underwater target is in an abnormal precursor state when the feeding rhythm migrates from the first mode to the second mode. This method, by setting a first mode representing a normal state and a second mode representing an abnormal state, forms a quantitative monitoring of the process of feeding rhythm transitioning from normal to abnormal, making the previously difficult-to-quantify "behavioral degradation" process detectable. By identifying the mode migration of feeding rhythm within a preset time window rather than simply identifying the current state, it can capture early signals of behavioral changes before the anomaly fully manifests, achieving a leap from "post-event monitoring" to "pre-event warning." The determined abnormal precursor state provides a valuable time window for subsequent intervention measures, helping to reduce the probability and severity of losses.

[0113] Optionally, the following steps may be included after step S140.

[0114] Step S150: Output the behavior and situation information of the underwater target.

[0115] In one specific implementation, after determining the behavioral situation information of an underwater target, the behavioral situation information is output. The output format may include, but is not limited to: displaying it on a local display device as text, graphics, or an alarm; sending it to a remote monitoring center or user terminal via a network; storing it in a database for historical records and trend analysis; or using it as a trigger signal to activate other linked devices. The output content should at least include the type or quantitative indicators of the behavioral situation, and may also include auxiliary information such as target identification, timestamp, and confidence level to facilitate understanding and use by the recipient.

[0116] In one optional implementation, if the determined behavioral state meets preset abnormal conditions, an early warning prompt can be attached to the output so that relevant personnel can respond in a timely manner.

[0117] In a specific example within the property insurance sector, continuing the scenario of marine ranching insurance, the assessed fish behavior and situation information is output to the insurance company's monitoring platform. The platform interface displays the fish situation in each aquaculture cage in a visual manner, such as "normal" or "abnormal." When an abnormal situation occurs, the platform automatically pops up an early warning notification, reminding claims personnel and aquaculture farmers to promptly verify and handle the situation, thereby reducing losses.

[0118] In some embodiments of the present invention, the acoustic monitoring method for underwater targets further includes the following steps.

[0119] Step S210: Acquire the raw acoustic signal through the edge end; In one specific implementation, raw acoustic signals are acquired using edge devices deployed in an underwater monitoring environment. The edge device refers to a computing node close to the data source, which may include acoustic sensors (e.g., hydrophones) and edge computing devices connected to the acoustic sensors. The edge device is deployed in the monitoring area where the underwater target is located, such as inside aquaculture cages, near underwater equipment, or in a patient rehabilitation training pool, to acquire raw underwater acoustic signals in real time or on demand.

[0120] The raw acoustic signal refers to the digitized acoustic data formed by the analog-to-digital conversion of the unprocessed electrical signal directly acquired by the acoustic sensor. This data may include target acoustic signals from underwater targets, as well as various sound components such as environmental noise, water flow sounds, and equipment operation sounds. Edge devices can digitize the analog acoustic signal at a preset sampling frequency (e.g., 10kHz to 100kHz) and quantization bit depth to obtain high-quality digital acoustic signal data.

[0121] In one optional implementation, the edge device may include a data acquisition array consisting of multiple acoustic sensors to acquire acoustic signals over a wider range or in a specific direction. In another optional implementation, the edge device may perform intermittent acquisition according to a preset acquisition cycle, or it may initiate a continuous acquisition mode upon receiving a trigger signal. The specific acquisition strategy can be configured according to the application scenario requirements and power consumption limitations.

[0122] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, edge devices, including hydrophones and edge computing boxes, are deployed inside or around the aquaculture cages. The hydrophones collect raw acoustic signals from the aquaculture area in real time at a preset sampling frequency. These signals include target sounds such as fish swimming and feeding, as well as environmental noise generated by aquaculture equipment such as aerators and water pumps. The edge devices continuously collect these raw acoustic signals, providing data for subsequent preprocessing.

[0123] Step S220: Perform preprocessing on the original acoustic signal at the edge end. The preprocessing includes: filtering out high-frequency noise in the original acoustic signal, and / or removing silent segments in the original acoustic signal. In one specific implementation, after the raw acoustic signal is acquired at the edge, preprocessing operations are performed directly on the edge device without uploading all the raw data to the cloud for further processing. The preprocessing includes filtering out high-frequency noise from the raw acoustic signal and / or removing silent segments. By performing these preprocessing operations at the edge, the raw acoustic signal can be streamlined and purified at the data source.

[0124] Specifically, filtering out high-frequency noise from the original acoustic signal refers to using a low-pass or band-pass filter to filter the original acoustic signal and remove high-frequency noise components that are above the target acoustic signal frequency range. Various high-frequency interference sources exist in the underwater environment, such as bubble bursting sounds and high-frequency vibrations from equipment. These high-frequency noise components may interfere with subsequent acoustic signature analysis. By setting an appropriate cutoff frequency at the edge (e.g., setting the cutoff frequency to 2kHz to 8kHz based on the target acoustic signal's frequency band), high-frequency noise can be effectively suppressed without losing the target signal.

[0125] Specifically, removing silent segments from the original acoustic signal involves performing Voice Activity Detection (VAD) on the original acoustic signal to identify silent or low-energy time periods that do not contain valid acoustic information. These silent segments are then removed from the signal, retaining only signal segments containing potentially valid acoustic information. Silent segment removal can be achieved through methods such as energy threshold detection and spectral feature detection. For example, time periods with signal energy below a preset energy threshold can be identified as silent segments and removed from the original signal. By removing silent segments, the amount of data that needs to be stored or uploaded can be significantly reduced.

[0126] In one optional implementation, preprocessing may simultaneously include two operations: filtering high-frequency noise and removing silent segments. The execution order can be filtering first and then removing silent segments, removing silent segments first and then filtering, or both operations can be performed in parallel. In another optional implementation, preprocessing may further include other signal conditioning operations such as signal amplification, DC offset removal, and anti-aliasing filtering to further improve signal quality.

[0127] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, edge devices perform preprocessing on the acquired raw acoustic signals. First, a low-pass filter with a cutoff frequency of 8kHz is used to filter out high-frequency noise (such as the high-frequency components of bubble bursting sounds and equipment electromagnetic interference) from the raw signal. Then, energy detection is performed on the filtered signal, and silent segments with signal energy consistently below a preset threshold (e.g., low-energy segments acquired at night when there is no fish activity and the aerator is not operating) are discarded. After this preprocessing, only acoustic segments containing valid information such as fish feeding sounds and swimming sounds are retained.

[0128] Step S230: Upload the pre-processed and retained valid acoustic segments to the cloud as underwater acoustic signals.

[0129] In one specific implementation, after preprocessing is completed at the edge, the retained valid acoustic segments are uploaded to the cloud. The cloud refers to a server or cloud computing platform deployed at a remote network endpoint, used to receive data from the edge and perform subsequent in-depth analysis. The edge endpoint sends the retained valid acoustic segments to the cloud via wired or wireless networks (e.g., 4G, 5G, Wi-Fi, IoT communication protocols, etc.).

[0130] The effective acoustic segments retained after filtering out high-frequency noise and / or removing silent segments are high-quality acoustic data segments that have been streamlined in terms of data volume and purified in terms of signal quality. Their data volume is much smaller than the total amount of original acquired data. After these effective acoustic segments are uploaded to the cloud, they serve as underwater acquired acoustic signals for the cloud to perform subsequent advanced analysis steps such as adaptive noise reduction processing, voiceprint feature extraction, and behavior situation determination.

[0131] In one optional implementation, before uploading the valid acoustic segment, the edge device can further compress the valid acoustic segment (e.g., using lossless or lossy compression algorithms) and encrypt it to further reduce the amount of data transmitted and ensure security during data transmission. In another optional implementation, data upload between the edge device and the cloud can be real-time streaming or batch upload at preset time intervals; the specific method can be flexibly selected based on network conditions and real-time requirements.

[0132] In a specific example within the property insurance sector, specifically in the application scenario of marine ranching insurance, the edge device uploads the valid acoustic segments—after filtering out high-frequency noise and removing silent segments—to a cloud-based voiceprint analysis platform in real time via a 4G / 5G network. Because silent segments are removed, the uploaded data volume is only 10% to 20% of the original data, significantly reducing the limitations imposed by insufficient network bandwidth in remote aquaculture areas on data transmission. The valid acoustic segments uploaded to the cloud serve as input signals for subsequent fish school voiceprint feature extraction and behavioral situation determination. The cloud platform no longer needs to store and preprocess large-scale raw data; it can directly proceed to the feature analysis process.

[0133] It is understood that the underwater target acoustic monitoring method described in this embodiment achieves simplification and purification of acoustic data at the source by acquiring raw acoustic signals at the edge and performing preprocessing at the edge, including filtering out high-frequency noise and / or eliminating silent segments, and then uploading the retained effective acoustic segments after preprocessing to the cloud. This method removes high-frequency interference components from the raw signal before data upload by performing preprocessing to filter out high-frequency noise at the edge close to the data source, reducing the computational burden on the cloud for processing high-frequency noise; by eliminating silent segments, it significantly reduces the amount of data that needs to be stored and uploaded, lowering network bandwidth consumption and cloud storage pressure, making large-scale distributed deployment possible in remote areas or weak network environments; and by uploading only segments containing effective acoustic information to the cloud, it improves the efficiency and real-time performance of subsequent cloud processing, making the overall link from acoustic signal acquisition to behavioral situation determination more efficient and reliable.

[0134] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0135] In one embodiment, an acoustic monitoring device for underwater targets is provided, which corresponds one-to-one with the acoustic monitoring methods for underwater targets described in the above embodiments. For example... Figure 5 As shown, the acoustic monitoring device for the underwater target includes a signal receiving module 510, a signal noise reduction module 520, a feature extraction module 530, and a behavior determination module 540. Detailed descriptions of each functional module are as follows: The signal receiving module 510 is used to receive acoustic signals collected underwater; The signal noise reduction module 520 is used to perform adaptive noise reduction processing based on an environmental reference signal on the acoustic signal to filter out the environmental noise component in the acoustic signal. The feature extraction module 530 is used to extract the acoustic signature features of the acoustic signal after filtering out the environmental noise component; The behavior determination module 540 is used to determine the behavior and situation information of the underwater target based on the voiceprint features.

[0136] In one embodiment, the signal receiving module 510 is specifically used for: Receives a mixed underwater signal acquired by the main hydrophone, wherein the mixed signal includes a target acoustic signal and an ambient noise component; Receives environmental noise signals acquired via a reference hydrophone; The adaptive noise reduction processing based on the environmental reference signal for the acoustic signal includes: The environmental noise signal is filtered using filter coefficients to generate an estimated signal of the environmental noise component in the mixed signal; The estimated signal is subtracted from the mixed signal to obtain the denoised target acoustic signal, and the denoised target acoustic signal is used as the error signal. The filter coefficients are adjusted based on the error signal so that the subsequently generated estimated signal approximates the environmental noise component in the mixed signal.

[0137] In one embodiment, the feature extraction module 530 is specifically used for: The one-dimensional time-domain waveform of the acoustic signal is transformed into a two-dimensional time-spectrum diagram. Mel frequency cepstral coefficients are extracted from the two-dimensional time-spectrum graph and used as the acoustic signature features of the acoustic signal.

[0138] In one embodiment, the behavior determination module 540 is specifically used for: Based on the acoustic signature characteristics, an acoustic signature activity index for the underwater target is generated; When the voiceprint activity index remains below the first preset threshold for an extended period of time, the underwater target is determined to be in a state of continuous silence. When the voiceprint activity index exceeds the second preset threshold within a preset time window, it is determined that the underwater target is in an explosive agitation state. After determining the underwater target's behavioral situation information based on the acoustic signature features, the method further includes: Based on the continuous silent state or the explosive turbulent state, a corresponding red line warning message is generated.

[0139] In one embodiment, the behavior determination module 540 is specifically used for: The voiceprint features are fed into a temporal behavior analysis model, which performs time-series modeling of the voiceprint features based on a self-attention mechanism to capture the changing patterns of the voiceprint features in the temporal dimension. The aforementioned pattern of change is compared with multiple preset behavioral patterns; Based on the comparison results, the behavioral situation information of the underwater target is determined.

[0140] In one embodiment, the behavior determination module 540 is further configured to: When the change pattern indicates that the feeding rhythm of the voiceprint feature has shifted from the first mode to the second mode within a preset time window, the underwater target is determined to be in an abnormal precursor state.

[0141] In one embodiment, the signal receiving module 510 is further configured to: Raw acoustic signals are acquired through the edge end; Preprocessing is performed on the original acoustic signal at the edge end, the preprocessing including: filtering out high-frequency noise in the original acoustic signal, and / or, removing silent segments in the original acoustic signal; The valid acoustic segments retained after preprocessing are uploaded to the cloud as acoustic signals collected underwater.

[0142] It is understood that the embodiments of the present invention, by performing adaptive noise reduction processing on acoustic signals based on environmental reference signals, enable the noise reduction parameters to be adjusted in real time according to changes in environmental noise. This is a capability that traditional fixed filtering methods in the background art do not possess. Specifically, the environmental reference signal provides the real-time characteristics of the current noise, and the adaptive mechanism ensures that the noise reduction parameters can follow the dynamic changes of non-stationary, broadband noise. Thus, without prior knowledge of the noise statistical characteristics, dynamic suppression of strong background noise is achieved, allowing previously submerged weak target acoustic signals to emerge, thereby effectively extracting weak acoustic signals of underwater targets.

[0143] Because the embodiments of the present invention effectively retain weak acoustic signals after filtering out environmental noise components from the acoustic signals, the technical means of further processing the acoustic signals based on the embodiments of the present invention to determine the behavioral situation information of underwater targets effectively improves the accuracy of the final state determination of underwater targets. This solves the technical problem in the prior art that it is difficult to effectively extract weak acoustic signals of underwater targets in the underwater environment, resulting in the inability to accurately monitor the state of underwater targets.

[0144] Based on the above-mentioned acoustic monitoring methods for underwater targets, such as Figure 6 As shown in the diagram, this embodiment of the invention also provides a structural schematic of an apparatus for an acoustic monitoring method for underwater targets. The apparatus includes a processor 61 and a memory 62 coupled to the processor 61. The memory 62 stores a computer program, which, when executed by the processor 61, causes the processor 61 to perform the steps of the acoustic monitoring method for underwater targets described in the above embodiment.

[0145] For further details regarding the implementation of the above-mentioned technical solution by the processor 61 in the device for the acoustic monitoring method of underwater targets, please refer to the description in the above-mentioned embodiment of the invention for the acoustic monitoring method of underwater targets, which will not be repeated here.

[0146] The processor 61 can also be called a CPU (Central Processing Unit). The processor 61 may be an integrated circuit chip with signal processing capabilities. The processor 61 can also be a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor, or the processor 61 can be any conventional processor.

[0147] like Figure 7 As shown in the diagram, this embodiment of the invention also provides a schematic diagram of a computer-readable storage medium, on which a readable computer program 71 is stored. The computer program 71 can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in various embodiments of the invention. The aforementioned storage medium includes various storage media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks or optical disks, ROM (Read-Only Memory), RAM (Random Access Memory), or terminal devices such as computers, servers, mobile phones, and tablets.

[0148] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or modules, and may be electrical, mechanical, or other forms.

[0149] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0150] Furthermore, the functional modules in the various embodiments of this invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0151] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0152] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available storage medium that a computer can store or a data storage device such as a server or data center that integrates one or more available storage media. The available storage medium may be a magnetic storage medium (e.g., floppy disk, hard disk, magnetic tape, optical storage medium) or a semiconductor storage medium (e.g., SSD (solid state disk)).

[0153] The technical solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

[0154] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0155] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0156] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0157] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0158] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0159] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An acoustic monitoring method for underwater targets, characterized in that, include: Receives acoustic signals collected underwater; Adaptive noise reduction processing based on an environmental reference signal is performed on the acoustic signal to filter out the environmental noise component in the acoustic signal; Extract the acoustic signature features of the acoustic signal after filtering out the environmental noise component; Based on the aforementioned acoustic signature features, the behavioral and situational information of the underwater target is determined.

2. The acoustic monitoring method for underwater targets according to claim 1, characterized in that, The receiving of underwater acoustic signals includes: Receives a mixed underwater signal acquired by the main hydrophone, wherein the mixed signal includes a target acoustic signal and an ambient noise component; Receives environmental noise signals acquired via a reference hydrophone; The adaptive noise reduction processing based on the environmental reference signal for the acoustic signal includes: The environmental noise signal is filtered using filter coefficients to generate an estimated signal of the environmental noise component in the mixed signal; The estimated signal is subtracted from the mixed signal to obtain the denoised target acoustic signal, and the denoised target acoustic signal is used as the error signal. The filter coefficients are adjusted based on the error signal so that the subsequently generated estimated signal approximates the environmental noise component in the mixed signal.

3. The acoustic monitoring method for underwater targets according to claim 1, characterized in that, The extraction of the acoustic signature features of the acoustic signal after filtering out the environmental noise component includes: The one-dimensional time-domain waveform of the acoustic signal is transformed into a two-dimensional time-spectrum diagram. Mel frequency cepstral coefficients are extracted from the two-dimensional time-spectrum graph and used as the acoustic signature features of the acoustic signal.

4. The acoustic monitoring method for underwater targets according to claim 1, characterized in that, The determination of the underwater target's behavioral situation information based on the acoustic signature features includes: Based on the acoustic signature characteristics, an acoustic signature activity index for the underwater target is generated; When the voiceprint activity index remains below the first preset threshold for an extended period of time, the underwater target is determined to be in a state of continuous silence. When the voiceprint activity index exceeds the second preset threshold within a preset time window, it is determined that the underwater target is in an explosive agitation state. After determining the underwater target's behavioral situation information based on the acoustic signature features, the method further includes: Based on the continuous silent state or the explosive turbulent state, a corresponding red line warning message is generated.

5. The acoustic monitoring method for underwater targets according to claim 1, characterized in that, The determination of the underwater target's behavioral situation information based on the acoustic signature features includes: The voiceprint features are fed into a temporal behavior analysis model, which performs time-series modeling of the voiceprint features based on a self-attention mechanism to capture the changing patterns of the voiceprint features in the temporal dimension. The aforementioned pattern of change is compared with multiple preset behavioral patterns; Based on the comparison results, the behavioral situation information of the underwater target is determined.

6. The acoustic monitoring method for underwater targets according to claim 5, characterized in that, The preset multiple behavioral posture patterns include a first mode and a second mode of feeding rhythm; determining the behavioral posture information of the underwater target based on the comparison results includes: When the change pattern indicates that the feeding rhythm of the voiceprint feature has shifted from the first mode to the second mode within a preset time window, the underwater target is determined to be in an abnormal precursor state.

7. The acoustic monitoring method for underwater targets according to claim 1, characterized in that, Also includes: Raw acoustic signals are acquired through the edge end; Preprocessing is performed on the original acoustic signal at the edge end, the preprocessing including: filtering out high-frequency noise in the original acoustic signal, and / or, removing silent segments in the original acoustic signal; The valid acoustic segments retained after preprocessing are uploaded to the cloud as acoustic signals collected underwater.

8. An acoustic monitoring device for underwater targets, characterized in that, include: The signal receiving module is used to receive acoustic signals collected underwater. The signal noise reduction module is used to perform adaptive noise reduction processing on the acoustic signal based on the environmental reference signal to filter out the environmental noise component in the acoustic signal; The feature extraction module is used to extract the acoustic signature features of the acoustic signal after filtering out the environmental noise component; The behavior determination module is used to determine the behavior and situation information of the underwater target based on the voiceprint features.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the acoustic monitoring method for underwater targets as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the acoustic monitoring method for underwater targets as described in any one of claims 1 to 7.

Images