Marine ranch fish school monitoring method based on acoustic imaging

Abstract

The invention discloses a marine ranch fish school monitoring method based on acoustic imaging, and belongs to the field of fish school intelligent monitoring, and the method comprises the steps: obtaining a noise interference component from compensated coverage range data, correcting signal fusion through employing Kalman filtering, and obtaining a fusion signal with noise suppression; acquiring a trajectory tracking sequence from the three-dimensional motion vector, and predicting motion complexity by adopting particle filtering to obtain centimeter-level accurate positioning coordinates; extracting a density model and body length estimation parameters from the reliable monitoring data set, and fusing biomass measurement logic to obtain a final fish school biomass value; aiming at the final fish school biomass value analysis distribution mode, correcting deviation by adopting a statistical model, and determining an optimized biomass distribution diagram; and generating an alarm signal according to the optimized biomass distribution diagram, and if the biomass distribution is judged to be abnormal, triggering real-time notification to obtain monitoring response data. According to the invention, the precision and real-time performance of fish school monitoring are obviously improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent fish monitoring, and particularly relates to a method for monitoring fish schools in marine ranches based on acoustic imaging. Background Technology

[0002] Marine ranching, as an important development direction of modern fisheries, plays a crucial role in ensuring marine ecological balance and food supply by scientifically managing aquatic resources to achieve sustainable fish farming. With the increasing global demand for marine resources, accurate monitoring of fish population dynamics has become core to improving ranching efficiency and ecological protection. However, the complexity of the marine environment presents many challenges to fish monitoring, urgently requiring innovative technologies to overcome existing bottlenecks.

[0003] Traditional fish monitoring methods often rely on a single signal source or fixed equipment, making them ill-suited for the large-scale, multi-target monitoring needs of marine ranches. For example, traditional sonar equipment is limited by signal attenuation, making it impossible to clearly capture fish distribution over a range of several kilometers, while optical monitoring is limited by seawater turbidity and light scattering. In complex marine environments, these methods often suffer from signal interference or insufficient resolution, leading to inaccurate judgments of fish location and movement, and failing to meet the needs of refined ranch management.

[0004] In fish shoal monitoring, signal processing and target localization are two core technical challenges. In the marine environment, sound waves and radio signals are affected by factors such as ocean currents and thermoclines, resulting in complex propagation paths and signal distortion or attenuation. This makes it difficult to determine the precise location of fish shoals, especially at long distances, where insufficient spatiotemporal resolution makes it difficult to distinguish fish shoals from seabed topography or other targets. Furthermore, the dynamic characteristics of fish shoals, such as rapid changes in swimming speed and direction, make it difficult for existing technologies to analyze the movement trajectories of multiple targets in real time. This contradiction between signal processing and dynamic tracking directly affects the accuracy and real-time performance of the monitoring system.

[0005] Therefore, how to integrate multiple signal sources in a complex marine environment, overcome signal attenuation and interference, and achieve real-time high-precision capture of the location and movement direction of fish schools has become a key issue in fish school monitoring in marine ranches. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for monitoring fish schools in marine ranches based on acoustic imaging, comprising:

[0007] By collecting reflected signals and radio propagation data through an acoustic sensor array, and integrating multi-dimensional algorithms to process the initial fish density and motion state parameters, a preliminary judgment result is obtained.

[0008] Based on the preliminary judgment results, attenuation compensation features in long-distance coverage are extracted, and adaptive beamforming is used to adjust signal transmission parameters to determine the compensated coverage range data.

[0009] The noise interference component is obtained from the compensated coverage data, and the signal fusion is corrected by Kalman filtering to obtain a noise-suppressed fused signal.

[0010] Based on the noise-suppressed fused signal, directional vector analysis is performed, and the radial velocity component is calculated using the Doppler effect. If the radial velocity component exceeds a preset threshold, the phase difference information is fused to determine the three-dimensional motion vector.

[0011] The trajectory tracking sequence is obtained from the three-dimensional motion vector, and the motion trajectory is predicted by particle filtering to obtain the positioning coordinates;

[0012] Based on the positioning coordinates, the data is transmitted in a hybrid mode using the integrated communication protocol. If the transmission integrity is lower than a preset threshold, error detection and retransmission are activated to obtain the monitoring dataset.

[0013] Based on the monitoring dataset, density model and body length estimation parameters are extracted, and biomass calculation logic is integrated to obtain the final fish biomass value.

[0014] Based on the analysis of the distribution pattern of the final fish biomass value, a statistical model was used to correct the deviation and determine the optimized biomass distribution map.

[0015] An alarm signal is generated based on the optimized biomass distribution map. If the biomass distribution is abnormal, a real-time notification is triggered, and monitoring response data is obtained.

[0016] Compared with the prior art, the present invention has the following advantages and technical effects:

[0017] This invention discloses a method for precise monitoring of fish biomass based on multi-sensor fusion and adaptive signal processing. Addressing the challenge of obtaining real-time and accurate data on fish density, movement trajectories, and biomass distribution in marine fisheries resource monitoring, this method fuses acoustic sensor arrays and radio propagation data. It then processes initial fish swarm parameters using multi-dimensional algorithms, extracts attenuation compensation features, and employs adaptive beamforming to optimize signal coverage. Furthermore, Kalman filtering and Doppler effect analysis are used to suppress noise and calculate three-dimensional motion vectors, generating centimeter-level precise positioning coordinates. A hybrid-mode communication protocol is integrated to ensure reliable data transmission. Finally, biomass measurement logic and statistical models are fused to correct deviations, generating an optimized biomass distribution map and triggering real-time anomaly notifications. This invention significantly improves the accuracy and real-time performance of fish swarm monitoring through multi-sensor data fusion, adaptive signal processing, and precise trajectory tracking, providing reliable data support for marine resource management.

[0018] The monitoring method of this invention can clearly distinguish the distribution of fish schools within a range of several kilometers and track their movement trajectory in real time, providing accurate data support for ranch managers, thereby optimizing resource allocation and ecological protection strategies. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] like Figure 1 As shown, this embodiment provides a method for monitoring fish schools in marine ranches based on acoustic imaging, including:

[0024] By collecting reflected signals and radio propagation data through an acoustic sensor array, and integrating multi-dimensional algorithms to process the initial fish density and motion state parameters, a preliminary judgment result is obtained.

[0025] Based on the preliminary judgment results, attenuation compensation features in long-distance coverage are extracted, and adaptive beamforming is used to adjust signal transmission parameters to determine the compensation coverage range data.

[0026] The noise interference component is obtained from the compensated coverage data, and the signal fusion is corrected by Kalman filtering to obtain the noise-suppressed fused signal.

[0027] Directional vector analysis is performed on the fused signal with noise suppression. The radial velocity component is calculated using the Doppler effect. If the radial velocity component exceeds a preset threshold, the phase difference information is fused to determine the three-dimensional motion vector.

[0028] The trajectory tracking sequence is obtained from the three-dimensional motion vector, and the motion trajectory is predicted by particle filtering to obtain the positioning coordinates;

[0029] Based on the location coordinates, the data is transmitted in a hybrid mode using the integrated communication protocol. If the transmission integrity is lower than a preset threshold, error detection and retransmission are activated to obtain the monitoring dataset.

[0030] Based on the monitoring dataset, density model and body length estimation parameters are extracted, and biomass calculation logic is integrated to obtain the final fish biomass value.

[0031] Based on the final fish biomass values, the distribution pattern was analyzed, and a statistical model was used to correct the bias, thus determining the optimized biomass distribution map.

[0032] An alarm signal is generated based on the optimized biomass distribution map. If the biomass distribution is abnormal, a real-time notification is triggered, and monitoring response data is obtained.

[0033] Furthermore, the process of obtaining preliminary judgment results includes:

[0034] Reflected signal data and radio propagation data are acquired through an acoustic sensor array. Signal preprocessing methods are used to denoise and standardize the data to obtain a processed signal dataset.

[0035] Fish density parameters and motion state parameters are extracted from the processed signal dataset. Principal component analysis algorithm is used to reduce the dimensionality of the data to obtain the dimensionality-reduced feature parameter set.

[0036] For the reduced feature parameter set, the k-means clustering algorithm is applied to estimate the fish density distribution, and a fish density distribution model is obtained.

[0037] Based on the fish density distribution model and combined with motion state parameters, the Kalman filter algorithm is used to predict the movement trajectory of the fish and obtain the movement trajectory estimation results.

[0038] If the trajectory estimation result exceeds the preset trajectory deviation threshold, a secondary analysis of the feature parameter set is performed, the clustering parameters are adjusted, and an optimized density distribution model is obtained.

[0039] By combining the optimized density distribution model and motion trajectory estimation results with multi-dimensional algorithms for comprehensive analysis, the fish behavior determination results are obtained.

[0040] The results of fish behavior determination were smoothed using a weighted average method, and preliminary results of fish density and movement status determination were obtained.

[0041] Specifically, by acquiring reflected signals and radio propagation data through an acoustic sensor array, the sonar device emits sound waves and receives the echoes reflected by the fish, combining this with the propagation characteristics of the radio signals to obtain information on the location and density of the fish.

[0042] In one possible implementation, an acoustic sensor array is deployed in the ocean monitoring area, emitting sound waves 10 times per second in a frequency range of 50-200 kHz. The acquired raw data includes fish school reflection intensity and propagation delay. Signal preprocessing involves wavelet transform denoising to filter out ocean background noise such as wave interference and normalizing the data to the 0-1 range for a uniform scale. This preprocessing significantly improves data quality and reduces errors in subsequent analysis. Extracting fish school density and motion parameters from the processed signal dataset is a crucial step.

[0043] For example, density parameters are calculated using the mean echo intensity; assuming a mean echo intensity of 0.8 in a certain area indicates dense fish populations, motion parameters are analyzed using Doppler frequency shift to determine fish speed; a frequency shift of 0.5 Hz corresponds to a fish movement of 0.2 m / s. Principal component analysis (PCA) is used to reduce the dimensionality of these parameters, retaining 90% of the principal components of the variance, thus reducing the data dimensionality to three dimensions. This dimensionality reduction reduces computational complexity while preserving core information, improving the efficiency of subsequent clustering. For the dimensionality-reduced feature parameter set, k-means clustering is used to estimate the fish density distribution.

[0044] In one embodiment, the value of k is set to 3, and the fish population is divided into high, medium and low density regions. The initial cluster centers are randomly selected based on the data distribution, and a density distribution model is formed after iteration.

[0045] For example, high-density areas correspond to fish populations with echo intensities greater than 0.7. This model intuitively reflects the distribution patterns of fish schools, providing a foundation for subsequent trajectory prediction. The Kalman filter algorithm combines the density distribution model and motion state parameters to predict the movement trajectory of the fish school.

[0046] For example, based on the speed and position data of a school of fish in the previous 10 seconds, the trajectory for the next 5 seconds is predicted. Assuming the initial position is (100, 200) meters and the speed is 0.2 m / s, a predicted trajectory deviation of less than 0.5 meters is considered accurate. If the trajectory deviation exceeds a preset threshold of 1 meter, a secondary analysis of the feature parameter set is performed, the k value is adjusted to 4, and re-clustering is conducted to obtain an optimized density model. This optimization improves the accuracy of the distribution model and reduces trajectory prediction errors. The optimized density distribution model and trajectory estimation results are then combined with multi-dimensional algorithms for comprehensive analysis.

[0047] For example, by combining density distribution and trajectory velocity, it can be determined whether a fish school is exhibiting foraging behavior. Areas with a velocity below 0.1 m / s and high density are considered foraging zones. A weighted average method smooths the behavior determination results, with weights allocated based on data reliability, such as 60% for siltation intensity and 40% for velocity data, resulting in the final determination of fish school density and movement status.

[0048] For example, a certain area was ultimately identified as a high-density foraging zone, with a density value of 0.85 and a speed of 0.15 m / s. This smoothing process reduces noise interference, improves the stability of the determination, and provides a reliable basis for fisheries management and ecological protection.

[0049] Furthermore, the process of determining the compensated coverage data includes:

[0050] The attenuation compensation features are obtained from the preliminary judgment results. Signal processing techniques are used to extract the signal attenuation degree and environmental interference factors to obtain the attenuation compensation feature set.

[0051] Based on the attenuation compensation feature set, adaptive beamforming technology is used to calculate the beamforming angle and transmit power adjustment parameters, and determine the adjusted signal transmission parameters.

[0052] By adjusting the signal transmission parameters and combining them with the signal propagation model, the propagation characteristics of the signal over a long coverage area are simulated to obtain data on the change of the coverage boundary.

[0053] If the coverage boundary change data exceeds the preset coverage range threshold, the beamforming angle and transmit power parameters are adjusted, the propagation characteristics are re-simulated, and new coverage boundary change data are determined.

[0054] Based on the new coverage boundary change data, the minimum mean square error algorithm is used to optimize the signal transmission parameters, and the optimized signal transmission parameter set is obtained.

[0055] Based on the optimized signal transmission parameter set, the signal strength distribution within the long-distance coverage area is calculated, and the compensated coverage area data is determined.

[0056] For example, in this embodiment, when obtaining attenuation compensation characteristics based on the preliminary judgment results, the signal attenuation degree and environmental interference factors are extracted by analyzing the reflected signal collected by the acoustic sensor array. The acoustic sensor is deployed in the deep sea environment, where the signal is attenuated due to water flow, temperature gradient and seabed topography reflection. The original signal is subjected to spectrum analysis by Fourier transform to separate the low-frequency noise and high-frequency attenuation components caused by environmental interference.

[0057] For example, the signal strength attenuates from an initial 100dB to 60dB, with 20dB attributable to water flow interference and 10dB to the effect of the temperature gradient. Through normalization, an attenuation compensation feature set is obtained, containing attenuation coefficients and interference weights, facilitating subsequent analysis.

[0058] Specifically, when employing adaptive beamforming technology, the beam direction and transmit power are adjusted based on an attenuation compensation feature set. In an underwater fish monitoring scenario, assuming the sensor array covers a range of 1000 meters, the initial beam angle is 30 degrees, and the transmit power is 200W. By analyzing the attenuation characteristics, it is calculated that the beam angle needs to be adjusted to 45 degrees and the power increased to 250W to compensate for signal attenuation. This adjustment is based on the degree of attenuation reflected in the feature set, ensuring signal coverage of areas with dense fish populations. The adjusted signal transmission parameters improve signal penetration in complex environments.

[0059] In one embodiment, when simulating using a signal propagation model, a sound wave propagation characteristic model is used to simulate signal attenuation and scattering over a long coverage area. Assuming the simulation scenario is a 2000-meter-deep sea area with an initial coverage boundary of 800 meters, the simulation results show that due to seabed topographic reflection, the coverage boundary shrinks to 600 meters, exceeding the preset threshold of 50 meters. At this point, the beam angle needs to be readjusted to 50 degrees, and the power increased to 300W. After resimulating, the coverage boundary recovers to 780 meters, close to the expected range. This iterative adjustment ensures the stability of the coverage area.

[0060] For example, when optimizing signal transmission parameters using the minimum mean square error algorithm, the optimal transmission parameters are calculated based on simulated coverage boundary change data. Assume the initial signal strength distribution is uneven, with the boundary region strength at only 30 dB, below the quality threshold of 50 dB. Through algorithm optimization, the transmission power allocation is adjusted, increasing the boundary region strength to 55 dB. The optimized signal transmission parameter set can balance the signal strength within the coverage area, improving detection accuracy.

[0061] Specifically, when calculating the signal strength distribution over a long coverage area, an optimized parameter set is used to generate a signal strength map within the coverage area. Assuming a final coverage area of ​​850 meters, with the boundary region strength stabilizing at 52 dB, this meets the coverage quality threshold. This uniform signal strength distribution helps in accurately monitoring fish density and movement. The final output coverage data provides a reliable basis for subsequent fish behavior analysis, contributing to improved robustness and adaptability of the detection system.

[0062] Furthermore, the process of obtaining the noise-suppressed fused signal includes:

[0063] Noise interference components are separated from coverage data, and signal decomposition techniques are used to obtain preliminary noise components;

[0064] Based on the initial noise components, Kalman filtering is applied to iteratively correct the coverage data to obtain the corrected signal data.

[0065] By extracting residual noise features from the corrected signal data, the intermediate signal for noise suppression is determined.

[0066] If the noise characteristics of the intermediate signal are lower than the preset threshold, a weighted fusion method is used to generate a fused signal.

[0067] Analyze the frequency domain characteristics of the fused signal to determine whether there are any abnormal interference components;

[0068] If abnormal interference components exist, the fused signal is processed a second time through adaptive filtering to obtain the final noise-suppressed fused signal.

[0069] For example, in this embodiment, wavelet transform is used to decompose the signal when separating noise interference components from coverage data. Wavelet transform decomposes the signal into different frequency bands, extracting high-frequency noise components while retaining low-frequency effective signals. Assuming that in a long-distance coverage scenario, the received signal data contains environmental noise and multipath interference, the signal is decomposed into 10 frequency bands, with high-frequency bands 5-8 primarily containing noise components. By analyzing the energy distribution of these frequency bands, the noise components are initially separated, resulting in a preliminary noise component set. This method effectively isolates noise, ensuring cleaner data for subsequent processing.

[0070] In one possible implementation, when applying Kalman filtering to iteratively correct coverage data, signal changes are predicted based on a state-space model. Kalman filtering gradually corrects for noise by estimating the dynamic changes of the signal and combining this with observed data.

[0071] For example, when extracting residual noise features, statistical analysis methods are used to calculate the variance and skewness of the signal to determine the noise distribution characteristics. Assuming the variance of residual noise in the corrected signal data is 0.02, which is lower than the preset threshold of 0.05, the noise suppression effect is considered good. In this case, the signal is marked as an intermediate signal. This method ensures that the signal quality meets the requirements of subsequent processing by quantifying noise features.

[0072] In one possible implementation, when generating the fused signal using a weighted fusion method, weights are assigned based on the signal-to-noise ratio of the signal source.

[0073] For example, among signals from multiple antennas, signals with high signal-to-noise ratios (SNRs) are assigned a weight of 0.6, and those with low SNRs are assigned a weight of 0.4. A weighted average is then used to generate a fused signal. This method can combine the advantages of multiple signal sources and improve the overall signal quality.

[0074] For example, when analyzing the frequency domain characteristics of the fused signal, a Fast Fourier Transform (FFT) is used to convert the signal to the frequency domain to check for any abnormal peaks. Suppose an abnormal peak at 2.4 GHz is found in the spectrum, which could be caused by interference from an external Wi-Fi device. In this case, an adaptive filter is used to suppress this frequency point, generating a final noise-suppressed fused signal. This method can accurately locate and eliminate specific interference.

[0075] In one possible implementation, the Pearson correlation coefficient is used to verify the correlation between the final fused signal and the original coverage data. Assuming a correlation coefficient of 0.95, which is higher than a preset threshold of 0.9, the signal processing flow is considered complete. This verification method ensures that the processed signal maintains a high degree of consistency with the original data and is suitable for signal optimization in long-distance coverage scenarios.

[0076] Furthermore, the process of determining the three-dimensional motion vectors includes:

[0077] The direction vectors of the noise-suppressed fused signal are extracted using signal processing techniques to obtain a set of direction vectors;

[0078] The radial velocity component is calculated from the set of direction vectors using the Doppler effect algorithm to obtain the radial velocity value;

[0079] If the radial velocity value exceeds the preset threshold, the phase difference information is extracted from the fused signal data to obtain the phase difference dataset;

[0080] The phase difference dataset is smoothed using the Kalman filter algorithm to obtain optimized phase difference data;

[0081] Based on the optimized phase difference data and radial velocity values, the three-dimensional motion vectors are calculated to obtain a set of three-dimensional motion vectors.

[0082] The least squares method is used to fit the three-dimensional motion vector set to obtain smooth motion vectors;

[0083] Based on the smoothed motion vector, the motion trajectory of the target object is determined, and three-dimensional motion trajectory data is obtained.

[0084] For example, in one possible implementation, the Doppler effect algorithm is used to calculate the radial velocity component. The set of direction vectors contains the target's motion direction, and the velocity information is obtained by analyzing the frequency shift.

[0085] For example, if a radar operates at a frequency of 10 GHz and detects a target frequency offset of 200 Hz, the estimated radial velocity is approximately 3 m / s. If this velocity exceeds a preset threshold (e.g., 2 m / s), subsequent phase difference extraction is triggered. The phase difference dataset reflects the differences in signal propagation between different sensors.

[0086] For example, two receiving antennas are spaced 0.5 meters apart, with a signal phase difference of π / 4, to pinpoint the precise location of a target.

[0087] Specifically, Kalman filtering is used to smooth phase difference datasets. Phase difference data is often affected by environmental noise, such as thermal noise or multipath effects. By using Kalman filtering, the phase change trend of the target is predicted, and optimized phase difference data is output.

[0088] For example, the initial phase difference fluctuated by ±0.1 radians, and after filtering, the fluctuation was reduced to ±0.02 radians, improving data stability. Based on the optimized phase difference and radial velocity, the three-dimensional motion vector was calculated.

[0089] For example, by combining the velocity of 3 m / s and the phase difference of π / 4, the motion components of the target on the x, y, and z axes can be derived, forming a three-dimensional set of motion vectors.

[0090] In one embodiment, the least squares method is used to fit a three-dimensional set of motion vectors to generate smooth motion vectors. The vector set may contain discrete points due to noise, and a continuous trajectory is generated through fitting.

[0091] For example, the target's motion vector points over 5 seconds are fitted into a smooth curve to reduce jitter. Finally, the target's trajectory is determined based on the smoothed motion vector.

[0092] For example, a trajectory shows a target moving from coordinates (0, 0, 0) to (10, 5, 2), reflecting its spatial motion path. This trajectory data is used for target tracking or behavior prediction.

[0093] For example, the advantage of the above method is that through multi-step processing, from the original signal to the final trajectory, noise interference is gradually reduced, data accuracy is improved, the integrity from signal acquisition to trajectory generation is ensured, and highly reliable motion analysis results can be provided.

[0094] Furthermore, the process of obtaining the positioning coordinates includes:

[0095] Trajectory sequences are extracted based on three-dimensional motion vectors through time series analysis.

[0096] Based on the trajectory sequence, the particle set is initialized using a particle filter algorithm to generate an initial state distribution;

[0097] The particle weights are iteratively updated using a particle filtering algorithm to predict motion complexity.

[0098] If the predicted motion complexity exceeds a preset threshold, the number of particles is adjusted to generate an updated state distribution.

[0099] Based on the updated state distribution, calculate the weighted average position and generate preliminary positioning coordinates;

[0100] After smoothing the initial positioning coordinates using Kalman filtering, the trajectory sequence is optimized using the least squares method to generate the final positioning coordinates.

[0101] Furthermore, the process of obtaining the monitoring dataset includes:

[0102] The original positioning data is obtained by using positioning coordinates, and the original positioning data is preprocessed by using signal filtering technology to obtain a positioning dataset that has been denoised.

[0103] Based on the location dataset, the data is encoded using a hybrid mode transmission protocol to generate an encoded transmission data stream;

[0104] For the encoded transmission data stream, calculate the data transmission integrity index. If the integrity index is lower than the preset threshold, it is marked as abnormal transmission data.

[0105] An error detection mechanism is used to analyze abnormal transmission data, determine the error type and location, and generate an error tag dataset.

[0106] Based on the error-marked dataset, activate the retransmission mechanism to resend the damaged data segment and obtain the retransmitted data stream.

[0107] Data verification technology is used to verify the retransmitted data stream and generate an initial monitoring dataset;

[0108] The initial monitoring dataset is processed by data integration and storage operations to obtain the new monitoring dataset.

[0109] For example, when obtaining raw positioning data through precise positioning coordinates, a high-precision GNSS receiver is used to capture satellite signals. The receiver collects data once per second to generate a raw dataset containing latitude, longitude, altitude, and timestamps.

[0110] It should be noted that the raw data is often biased due to environmental noise such as multipath effect or signal blockage, so preprocessing is required.

[0111] Specifically, the signal filtering technique involved in this embodiment uses median filtering to remove outliers.

[0112] For example, when analyzing 10 consecutive seconds of location data, if the longitude of a point suddenly jumps and significantly deviates from the normal trajectory, median filtering will replace this outlier point with the median value of the surrounding data, generating a smooth, denoised dataset. This method effectively preserves the trajectory trend while reducing noise interference.

[0113] In one embodiment, a hybrid mode transport protocol combines TCP and UDP features to encode the noise-reduced dataset.

[0114] For example, TCP ensures data integrity, while UDP improves transmission speed. In urban monitoring, the encoded data stream can contain location coordinates, timestamps, and checksums, generating a data stream of approximately 1KB / s. If network jitter causes data loss, the integrity index calculation detects the missing data through checksums and comparisons.

[0115] For example, when the integrity index is below 90%, it is marked as abnormal data transmission.

[0116] For example, the error detection mechanism uses cyclic redundancy check (CRC) to analyze abnormal data. Suppose a checksum mismatch is found in a data segment, the error type is identified as a bit flip, and the error location is byte 500 of the data stream. The error tag dataset records this information for subsequent processing. After the retransmission mechanism is activated, the system only retransmits the damaged 500-600 byte data segment, avoiding a full retransmission and improving efficiency.

[0117] Specifically, data verification technology verifies retransmitted data streams through hash verification.

[0118] For example, the MD5 value of the retransmitted data segment is compared with the original value; if they match, reliability is confirmed. The reliable initial monitoring dataset is then integrated into a JSON format, containing time series data and location coordinates, and stored in a local database.

[0119] For example, approximately 3,600 records are generated per hour and stored as structured data containing time, location, and status to ensure the accuracy and traceability of subsequent analyses.

[0120] In one embodiment, data integration and storage employ distributed database optimization.

[0121] For example, using MongoDB to store monitoring data, each record includes longitude, latitude, altitude, and collection time for easy and quick querying. The stored procedures also include timestamp indexes to optimize retrieval efficiency. This approach supports real-time monitoring and historical data analysis, providing a reliable basis for urban management.

[0122] Furthermore, the process of obtaining the final fish biomass value includes:

[0123] Raw data is obtained from the monitoring dataset. Through data extraction and processing, relevant data on fish density and body length estimation are separated to obtain a structured dataset.

[0124] If the structured dataset contains fish density features, then the random forest algorithm is used to train the density model to obtain the predicted fish density.

[0125] Body length-related features are extracted from structured datasets, and a body length estimation model is trained using a linear regression algorithm to obtain the estimated body length value.

[0126] Based on the predicted fish density and estimated body length, a comprehensive parameter set is generated using a parameter fusion method.

[0127] If the comprehensive parameter set meets the preset biomass calculation conditions, the intermediate biomass parameters are obtained by merging the predicted fish density and estimated body length using the weighted average method.

[0128] Based on the intermediate biomass parameters, the biomass value of the fish population is calculated using biomass calculation logic.

[0129] The final fish biomass value was obtained by standardizing the fish biomass value.

[0130] For example, in the field of marine fisheries monitoring, the process of obtaining raw data from monitoring datasets is achieved by acquiring underwater echo signals using sonar equipment. The sonar equipment emits sound waves at a fixed frequency, receives the reflected signals from fish schools, and generates a raw dataset containing information such as depth and intensity. Suppose a single acquisition yields 1000 sets of echo data, each set including signal intensity and depth information. In the data extraction and processing stage, signal segmentation techniques are used to separate features related to fish density and body length.

[0131] For example, the portion of the signal strength above a certain threshold is considered a fish density feature, while the echo duration is associated with the fish length. After processing, a structured dataset is generated, containing 500 sets of density features and 300 sets of body length features.

[0132] In one possible implementation, a random forest algorithm is used to train the density model based on the fish density characteristics. The random forest generates a predicted fish density value by constructing multiple decision trees and combining the predictions of each tree. Assuming the training data contains density data from 100 monitoring points over the past month, the model predicts a fish density of 50 fish per cubic meter in a certain area.

[0133] It should be noted that the advantage of random forests lies in their ability to effectively avoid overfitting when dealing with high-dimensional data, ensuring stable prediction results.

[0134] For example, the body length estimation model is constructed using a linear regression algorithm. Body length-related features, such as echo duration, are extracted from a structured dataset. Assuming a dataset contains 200 sets of body length features, ranging from 10 to 50 centimeters, the linear regression model, trained on historical body length data, predicts the average body length of a fish population to be 30 centimeters. This method is computationally simple and can quickly respond to real-time monitoring needs.

[0135] In one possible implementation, the parameter fusion method integrates the predicted fish density and estimated body length into a comprehensive parameter set. Assuming a predicted density of 50 fish per cubic meter and an estimated body length of 30 centimeters, a comprehensive parameter set is generated through weighted fusion, with weights allocated based on data reliability, such as density accounting for 60% and body length for 40%. If the comprehensive parameter set meets the biomass calculation conditions, such as a density greater than 10 fish per cubic meter and a body length greater than 15 centimeters, then the biomass calculation stage begins.

[0136] For example, the weighted average method is used to fuse density and body length data to generate an intermediate biomass parameter. Assuming a density weight of 0.6 and a body length weight of 0.4, the calculated intermediate biomass parameter is 1200 grams per cubic meter. The biomass calculation logic further incorporates water area and depth; assuming a monitoring area of ​​1000 square meters and an average depth of 10 meters, the calculated fish biomass value is 1200 tons. Standardization then normalizes the biomass value to a standard unit, such as biomass per hectare, facilitating cross-regional comparisons.

[0137] In one possible implementation, standardization is achieved by dividing the biomass value by the area of ​​the monitored region.

[0138] For example, dividing 1200 tons of biomass by 100 hectares yields a standardized biomass value of 12 tons per hectare. This method facilitates fisheries management in assessing resource distribution and optimizing fishing plans.

[0139] Furthermore, the process of determining the optimized biomass distribution map includes:

[0140] Fish biomass data were collected using sonar sensors, and the sonar signals were preliminarily processed to obtain initial biomass distribution data.

[0141] Based on the initial biomass distribution data, a Gaussian mixture model was used to perform cluster analysis on the data to determine the spatial distribution pattern of the fish population.

[0142] Based on the spatial distribution pattern, the mean and variance of biomass in each cluster region are calculated to obtain the statistical characteristics of the distribution pattern;

[0143] If the variance of the statistical features exceeds the preset threshold, a Bayesian regression model is used to correct the bias in the biomass data and generate corrected biomass distribution data.

[0144] Using the corrected biomass distribution data, a smooth biomass distribution map is generated using the nuclear density estimation method, resulting in an optimized distribution pattern.

[0145] Based on the optimized distribution pattern, the spatial resolution and density gradient of the distribution map are extracted to determine the uniformity of biomass distribution.

[0146] Based on the uniformity assessment results, an interpolation algorithm is used to supplement the data in low-density areas, generating the final biomass distribution map.

[0147] For example, when collecting fish biomass data using sonar sensors, multibeam sonar equipment is used to scan a specific body of water to obtain high-resolution echo signals. Preliminary processing of the sonar signals involves signal denoising and feature extraction to generate initial biomass distribution data. Assuming a fish farm where sonar scanning covers 1000 square meters of water, the collected signals show that fish are mainly concentrated in areas with a depth of 10-20 meters. After preliminary processing, distribution data showing a biomass value per square meter ranging from 0.5 to 2.0 kg is obtained. This initial data provides the foundation for subsequent analysis.

[0148] Specifically, when using the Gaussian mixture model to perform cluster analysis on the initial biomass distribution data, the spatial distribution of fish populations is divided into three modes: dense, moderate, and sparse.

[0149] For example, in the aforementioned fishery, the model identified densely populated areas in the center of the water with a mean biomass of approximately 1.8 kg / m², while the peripheral areas exhibited a sparse pattern with a mean biomass of only 0.6 kg / m². The clustering results clearly demonstrate the spatial heterogeneity of the fish population, providing a basis for further analysis.

[0150] In one embodiment, when calculating the mean and variance of biomass for each cluster region, it was found that the variance of dense regions was 0.3, exceeding the preset threshold of 0.2, indicating significant data fluctuation. To address this, a Bayesian regression model was used for bias correction.

[0151] For example, by training the model with historical biomass data and environmental variables (such as water temperature and salinity), the biomass values ​​in densely populated areas can be corrected to more closely approximate the actual distribution. After correction, the mean is adjusted to 1.7 kg / m², and the variance is reduced to 0.15. This correction improves the reliability of the data.

[0152] For example, based on corrected biomass distribution data, the nuclear density estimation method generates a smoothed biomass distribution map. Assuming a bandwidth of 0.5 is used in a fish farm, the generated distribution map has a spatial resolution of 0.1 kg / m², showing a gradual decrease in biomass from the center to the edges. The optimized distribution pattern visually reflects the aggregation characteristics of the fish population, facilitating subsequent analysis.

[0153] Specifically, when extracting the spatial resolution and density gradient of the distribution map, a density gradient of 0.05 kg / m² / m in the central region indicates a relatively uniform biomass distribution. A higher gradient in the peripheral regions, such as 0.1 kg / m² / m, indicates uneven distribution. The uniformity assessment results guide data supplementation strategies.

[0154] For example, in low-density areas (biomass below 0.5 kg / m²), Kriging interpolation is used to supplement data and generate a smooth final biomass distribution map, ensuring the accuracy of fishery management decisions.

[0155] In one embodiment, the final biomass distribution map is used for fishery resource assessment.

[0156] For example, fishery managers can identify high-density areas using distribution maps, optimize fishing routes, reduce resource waste, and simultaneously protect fish populations in low-density areas, promoting ecological balance. The advantage of this method is that it improves the accuracy and spatial resolution of biomass assessment, providing data support for sustainable fisheries management.

[0157] Furthermore, the process of obtaining monitoring response data includes:

[0158] Multidimensional spatiotemporal data are collected through sensor networks to obtain the original distributed dataset;

[0159] Feature analysis methods are used to extract spatiotemporal variation features from the original distributed dataset to obtain a feature vector set;

[0160] If the deviation between the feature vector set and the preset normal distribution model exceeds a threshold, it is determined to be a distribution anomaly, and an anomaly detection result is obtained.

[0161] Based on the anomaly detection results, a real-time alarm signal is generated, and alarm signal data is obtained;

[0162] The alarm signal data triggers the notification system, and the notification distribution record is obtained.

[0163] Response data is extracted from notification distribution records, and its timeliness and coverage are analyzed to obtain monitoring response data;

[0164] Based on the monitoring response data, the parameters of the anomaly detection model are updated to obtain the optimized distribution model.

[0165] For example, when collecting multidimensional spatiotemporal data through sensor networks, a combination of sonar and underwater optical sensors can be used to obtain the dynamic distribution of fish biomass. The sonar sensors collect 1,000 sets of depth and density data per second, while the optical sensors record the movement trajectory and swarming behavior of the fish, forming a raw dataset containing timestamps, location coordinates, and biomass density.

[0166] For example, in monitoring a certain sea area, a spatiotemporal dataset containing 100,000 data points is generated every 10 minutes, covering a monitoring area of ​​100 square kilometers. This multidimensional data acquisition method can capture the dynamic changes of fish schools, providing a high-precision foundation for subsequent analysis.

[0167] In one possible implementation, principal component analysis is used to extract spatiotemporal variation features from the original dataset, such as the rate of change of fish density and the periodic fluctuations in the direction of movement.

[0168] For example, when analyzing data from a fishing ground, it was found that fish density exhibits regular changes within the tidal cycle, with a density peak occurring every 6 hours. The feature vector set records the temporal and spatial weights of these changes. If the feature vector set deviates from the preset normal distribution model by more than 20%, it is judged as an anomaly.

[0169] For example, during a monitoring session, the fish density in a certain area suddenly dropped by 50%, far exceeding the normal fluctuation range, triggering anomaly detection.

[0170] Specifically, once the anomaly detection result is generated, it will notify the monitoring personnel through a real-time alarm signal.

[0171] For example, if the system detects an abnormal decrease in fish density in a certain area, it generates an alarm signal containing the time, location, and severity of the anomaly, and pushes it to the management team via SMS and application. The alarm signal data records the timestamp and reception status of each push.

[0172] For example, an alarm can be pushed to 10 management terminals within 5 seconds, achieving a coverage rate of 95% and ensuring a rapid response.

[0173] In one possible implementation, the analysis of notification distribution records assesses system efficiency by statistically analyzing response time and reception range.

[0174] For example, in one anomaly, notification distribution records showed that 90% of terminals acknowledged receipt within 10 seconds, covering all key monitoring stations. Analysis revealed that areas with high response timeliness could adjust monitoring strategies more quickly, such as increasing sensor density or adjusting patrol routes, thereby improving monitoring effectiveness.

[0175] For example, when updating the parameters of the anomaly detection model, the model threshold is optimized based on the response data.

[0176] For example, one analysis found that the initial threshold setting was too sensitive, causing 30% of normal fluctuations to be misjudged as abnormal. By adjusting the threshold to 25%, the false alarm rate was reduced to 5%, and the optimized distribution model could more accurately reflect the true distribution of the fish population. This optimization method improves the model's adaptability and monitoring accuracy by continuously learning data features.

[0177] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for monitoring fish schools in marine ranches based on acoustic imaging, characterized in that, include: By collecting reflected signals and radio propagation data through an acoustic sensor array, and integrating multi-dimensional algorithms to process the initial fish density and motion state parameters, a preliminary judgment result is obtained. Based on the preliminary judgment results, attenuation compensation features in long-distance coverage are extracted, and adaptive beamforming is used to adjust signal transmission parameters to determine the compensated coverage range data. The noise interference component is obtained from the compensated coverage data, and the signal fusion is corrected by Kalman filtering to obtain a noise-suppressed fused signal. Based on the noise-suppressed fused signal, directional vector analysis is performed, and the radial velocity component is calculated using the Doppler effect. If the radial velocity component exceeds a preset threshold, the phase difference information is fused to determine the three-dimensional motion vector. The trajectory tracking sequence is obtained from the three-dimensional motion vector, and the motion trajectory is predicted by particle filtering to obtain the positioning coordinates; Based on the positioning coordinates, the data is transmitted in a hybrid mode using the integrated communication protocol. If the transmission integrity is lower than a preset threshold, error detection and retransmission are activated to obtain the monitoring dataset. Based on the monitoring dataset, density model and body length estimation parameters are extracted, and biomass calculation logic is integrated to obtain the final fish biomass value. Based on the analysis of the distribution pattern of the final fish biomass value, a statistical model was used to correct the deviation and determine the optimized biomass distribution map. An alarm signal is generated based on the optimized biomass distribution map. If the biomass distribution is abnormal, a real-time notification is triggered, and monitoring response data is obtained.

2. The method according to claim 1, characterized in that, The process of obtaining preliminary judgment results includes: Reflected signal data and radio propagation data are acquired through an acoustic sensor array. Signal preprocessing methods are used to denoise and standardize the data to obtain a processed signal dataset. Fish density parameters and motion state parameters are extracted from the processed signal dataset. Principal component analysis algorithm is used to reduce the dimensionality of the data to obtain the dimensionality-reduced feature parameter set. For the dimensionality-reduced feature parameter set, the k-means clustering algorithm is applied to estimate the fish density distribution, and a fish density distribution model is obtained. Based on the fish density distribution model and the motion state parameters, the Kalman filter algorithm is used to predict the movement trajectory of the fish and obtain the movement trajectory estimation result. If the motion trajectory estimation result exceeds the preset trajectory deviation threshold, then the feature parameter set is analyzed a second time, the clustering parameters are adjusted, and an optimized density distribution model is obtained. By combining the optimized density distribution model and motion trajectory estimation results with multi-dimensional algorithms for comprehensive analysis, the fish behavior determination results are obtained. The fish behavior determination results were smoothed using a weighted average method to initially obtain the fish density and movement status determination results.

3. The method according to claim 1, characterized in that, The process of determining the compensated coverage data includes: The attenuation compensation features are obtained from the preliminary judgment results, and the signal attenuation degree and environmental interference factors are extracted using signal processing techniques to obtain the attenuation compensation feature set. Based on the attenuation compensation feature set, adaptive beamforming technology is used to calculate the beamforming angle and transmit power adjustment parameters, and determine the adjusted signal transmission parameters. By adjusting the signal transmission parameters and combining them with the signal propagation model, the propagation characteristics of the signal over a long coverage area are simulated to obtain data on the change of the coverage boundary. If the coverage boundary change data exceeds the preset coverage range threshold, the beamforming angle and transmit power parameters are adjusted, the propagation characteristics are re-simulated, and new coverage boundary change data are determined. Based on the new coverage boundary change data, the minimum mean square error algorithm is used to optimize the signal transmission parameters, and the optimized signal transmission parameter set is obtained. Based on the optimized signal transmission parameter set, the signal strength distribution within the long-distance coverage area is calculated, and the compensated coverage area data is determined.

4. The method according to claim 1, characterized in that, The process of obtaining a noise-suppressed fused signal includes: The noise interference component is separated from the coverage area data, and a preliminary noise component is obtained by using signal decomposition technology; Based on the initial noise components, Kalman filtering is applied to iteratively correct the coverage data to obtain the corrected signal data. By extracting residual noise features from the corrected signal data, the intermediate signal for noise suppression is determined. If the noise characteristics of the intermediate signal are lower than a preset threshold, a weighted fusion method is used to generate a fused signal. Analyze the frequency domain characteristics of the fused signal to determine whether there are any abnormal interference components; If abnormal interference components exist, the fused signal is processed a second time through adaptive filtering to obtain the final noise-suppressed fused signal.

5. The method according to claim 1, characterized in that, The process of determining three-dimensional motion vectors includes: The direction vectors of the noise-suppressed fused signal are extracted using signal processing techniques to obtain a set of direction vectors; The radial velocity component is calculated from the set of direction vectors using the Doppler effect algorithm to obtain the radial velocity value; If the radial velocity value exceeds a preset threshold, phase difference information is extracted from the fused signal data to obtain a phase difference dataset; The phase difference dataset is smoothed using the Kalman filter algorithm to obtain optimized phase difference data; Based on the optimized phase difference data and radial velocity value, the three-dimensional motion vector is calculated to obtain a set of three-dimensional motion vectors; The least squares method is used to fit the three-dimensional motion vector set to obtain smooth motion vectors; Based on the smoothed motion vector, the motion trajectory of the target object is determined, and three-dimensional motion trajectory data is obtained.

6. The method according to claim 1, characterized in that, The process of obtaining positioning coordinates includes: Based on the three-dimensional motion vector, trajectory sequences are extracted through time series analysis; Based on the trajectory sequence, the particle set is initialized using a particle filtering algorithm to generate an initial state distribution; The particle weights are iteratively updated using a particle filtering algorithm to predict motion complexity. If the predicted motion complexity exceeds a preset threshold, the number of particles is adjusted to generate an updated state distribution. Based on the updated state distribution, calculate the weighted average position and generate preliminary positioning coordinates; After smoothing the initial positioning coordinates using Kalman filtering, the trajectory sequence is optimized using the least squares method to generate the final positioning coordinates.

7. The method according to claim 1, characterized in that, The process of obtaining the monitoring dataset includes: The original positioning data is obtained through the positioning coordinates, and the original positioning data is preprocessed using signal filtering technology to obtain a positioning dataset after noise reduction. Based on the location dataset, the data is encoded using a hybrid mode transmission protocol to generate an encoded transmission data stream; For the encoded transmission data stream, a data transmission integrity index is calculated. If the integrity index is lower than a preset threshold, it is marked as abnormal transmission data. The abnormal transmission data is analyzed through an error detection mechanism to determine the error type and location, and an error tag dataset is generated. Based on the error-marked dataset, the retransmission mechanism is activated to resend the damaged data segment, resulting in a retransmitted data stream. The retransmitted data stream is verified using data verification technology to generate an initial monitoring dataset; The initial monitoring dataset is then processed through data integration and storage operations to obtain the monitoring dataset.

8. The method according to claim 1, characterized in that, The process of obtaining the final fish biomass value includes: Based on the monitoring dataset, raw data is obtained, and through data extraction and processing, relevant data on fish density and body length estimation are separated to obtain a structured dataset. If the structured dataset contains fish density features, then the density model is trained using the random forest algorithm to obtain the predicted fish density. Body length-related features are extracted from the structured dataset, and a body length estimation model is trained using a linear regression algorithm to obtain the estimated body length value. Based on the predicted fish density and estimated body length, a comprehensive parameter set is generated using a parameter fusion method. If the comprehensive parameter set meets the preset biomass calculation conditions, the intermediate biomass parameters are obtained by fusing the predicted fish density and estimated body length using a weighted average method. Based on the intermediate biomass parameters, the biomass value of the fish population is calculated using biomass calculation logic. The final fish biomass value is obtained by standardizing the fish biomass value.

9. The method according to claim 1, characterized in that, The process of determining the optimized biomass distribution map includes: Fish biomass data were collected using sonar sensors, and the sonar signals were preliminarily processed to obtain initial biomass distribution data. Based on the initial biomass distribution data, a Gaussian mixture model was used to perform cluster analysis on the data to determine the spatial distribution pattern of the fish population. Based on the spatial distribution pattern, the mean and variance of biomass in each cluster region are calculated to obtain the statistical characteristics of the distribution pattern; If the variance of the statistical feature exceeds a preset threshold, a Bayesian regression model is used to correct the bias in the biomass data, generating corrected biomass distribution data. Using the corrected biomass distribution data, a smooth biomass distribution map is generated using the nuclear density estimation method, resulting in an optimized distribution pattern. Based on the optimized distribution pattern, the spatial resolution and density gradient of the distribution map are extracted to determine the uniformity of biomass distribution. Based on the uniformity assessment results, an interpolation algorithm is used to supplement the data in low-density areas, generating the final biomass distribution map.

10. The method according to claim 1, characterized in that, The process of obtaining monitoring response data includes: Multidimensional spatiotemporal data are collected through sensor networks to obtain the original distributed dataset; Feature analysis methods are used to extract spatiotemporal variation features from the original distributed dataset to obtain a feature vector set; If the deviation between the feature vector set and the preset normal distribution model exceeds a threshold, it is determined to be a distribution anomaly, and an anomaly detection result is obtained. Based on the anomaly detection results, a real-time alarm signal is generated, and alarm signal data is obtained; The alarm signal data triggers the notification system, resulting in a notification distribution record. Response data is extracted from the notification distribution records, and the timeliness and coverage are analyzed to obtain monitoring response data; Based on the monitoring response data, the anomaly detection model parameters are updated to obtain the optimized distribution model.

Images