A multi-frequency acoustic wave and AI algorithm large-scale fish resource detection method

Abstract

The application discloses a large-scale fish resource detection method combining multi-frequency sound waves and AI algorithms, which comprises the following steps: correcting original echo signals to signals under standard environmental characteristics in combination with environmental characteristics; extracting and normalizing spectral characteristics and time sequence distribution information of the echo signals; acquiring frequency response characteristic data of fish species and associating spatial position information thereof; acquiring fish school population proportion through three-dimensional point cloud clustering analysis; identifying fish school abnormal behavior types, regional ranges, and formulating fish school abnormal intervention strategies; predicting time points, regional ranges, abnormal types and reasons of fish school abnormal behaviors in future time periods; evaluating effects of the fish school abnormal intervention strategies in combination with Mahalanobis distance and fish school behavior prediction, and dynamically adjusting the fish school abnormal intervention strategies. The application can improve the precision and efficiency of fish resource detection, meet the needs of dynamic monitoring and accurate management of fish schools in large water areas, and further promote the health and sustainable development of water ecological systems.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aquaculture monitoring, and in particular to a large-scale fish resource detection method based on multi-frequency sound waves and AI algorithms. BACKGROUND

[0002] With the changes in global aquatic ecological environment and the increasing scarcity of fishery resources, scientifically and efficiently monitoring fish resource dynamics has become an important task for ecological protection and fishery management. However, the current fish resource detection and monitoring technology still faces many problems in practical application. The detection capability of traditional monitoring technology for large areas of water is limited, which cannot meet the demand of modern fishery management for real-time monitoring of large-scale fish populations. Secondly, as a non-contact fish resource detection method, multi-frequency sonar technology can obtain fish echo signals in water through sound waves, but there are still many problems in practical application. The propagation of sound waves is significantly affected by the complex characteristics of the water environment, such as temperature, salinity, flow rate, turbidity, etc., resulting in noise, distortion and even loss of part of the echo signal. The interference of these environmental variables makes the quality of the original data collected by multi-frequency sonar unstable, directly affecting the accuracy of subsequent fish identification. In addition, the behavior of fish in natural water has high dynamicity and complexity, including group aggregation, dispersion, change of movement direction, etc. These behavioral changes are usually driven by multiple factors, such as predator threat, environmental disturbance or human activities. However, traditional fish monitoring methods have limited ability to capture these dynamic behavior characteristics, especially in real-time and detailed analysis, making it difficult to achieve timely monitoring and accurate identification of fish behavior. In addition, the spatiotemporal complexity of fish behavior data makes it more difficult to predict future dynamic changes, and there is currently a lack of precise prediction methods for future behavior trends of fish groups, which leads to a lag and passive state of fishery management in response to abnormal behavior. In addition, there are also great technical challenges in identifying and intervening abnormal behavior. At present, the monitoring of fish abnormal behavior mainly relies on artificial experience or single indicator judgment, and lacks a scientific behavior baseline model as a reference, which is easy to cause false negatives or false positives. At the same time, in the monitoring and intervention range adjustment of abnormal behavior expansion trend, the current means also have great deficiencies, especially for the case that abnormal behavior may spread to non-intervention areas, the existing technology is difficult to respond in time and effectively. Therefore, the above problems seriously restrict the improvement of fish resource management efficiency and the scientific protection of aquatic ecosystems, and urgent need for more accurate and efficient technical means to solve. SUMMARY

[0003] The present application provides a large-scale fish resource detection method based on multi-frequency sound waves and AI algorithms to solve the above problems of the prior art, mainly including:

[0004] Adjusting sound wave emission parameters through a multi-frequency sonar device, and correcting the original echo signal to a signal under standard environmental characteristics in combination with environmental characteristics;

[0005] According to the corrected echo signal, through time window segmentation, signal arrival time calculation and denoising processing, the spectral features and timing distribution information of the echo signal are extracted and normalized;

[0006] Through time series analysis, frequency response gradient calculation and envelope curve extraction of multi-frequency signals, frequency response feature data of fish species are generated and associated with spatial location information;

[0007] According to the frequency response feature data of fish under multi-frequency sound waves in the fish sound wave monitoring database, the spatial distribution, population proportion and mixed fish group marking information of the fish school are obtained through signal classification and three-dimensional point cloud clustering analysis;

[0008] According to the fish school echo signal and dynamic data obtained by the multi-frequency sonar receiver in real time, the parameter range of the fish school normal behavior baseline is determined, the Mahalanobis distance between the fish school dynamic data and the fish school normal behavior baseline mean value is calculated, the fish school abnormal behavior type, area range are identified, and the fish school abnormal intervention strategy is formulated;

[0009] According to the historical dynamic data of the fish school behavior mode database, in combination with the fish school normal behavior baseline mean value, the time point, area range, abnormal type and reason of the fish school abnormal behavior in the future time period are predicted, and the fish school abnormal intervention strategy is implemented in advance;

[0010] According to the fish school dynamic data obtained by the multi-frequency sonar receiver, in combination with the Mahalanobis distance and the fish school behavior prediction, the effect of the fish school abnormal intervention strategy is evaluated, and the fish school abnormal intervention strategy and intervention range are dynamically adjusted.

[0011] Further, the adjusting sound wave emission parameters through a multi-frequency sonar device, and correcting the original echo signal to a signal under standard environmental characteristics in combination with environmental characteristics, comprises:

[0012] According to the preset detection water area and the possible distribution range of the target fish species, a multi-frequency sonar device is used to configure multiple frequency ranges, and the order of low-frequency, medium-frequency and high-frequency transmission signals is set; through the sound wave emission system of the multi-frequency sonar device, the sound wave emission power, the sound beam width and the emission angle are adjusted, and the sound wave signals are emitted to the specified water area one by one, the original echo signals of the multi-frequency sound waves are obtained through the multi-frequency sonar receiver, the spatial position information of the sound wave emission is marked, and stored to the fish sound wave monitoring database; the echo signals under the standard environmental characteristics and the echo signals under the non-standard environmental characteristics of the same distance, azimuth and depth are obtained through the fish sound wave monitoring database, the model is trained using a recurrent neural network, and a echo signal correction model is constructed, and the environmental characteristics include the temperature, salinity, flow rate and turbidity of the water body; according to the original echo signals obtained in real time and the environmental characteristics, the echo signal correction model is used to correct the original echo signals to the echo signals under the standard environmental characteristics.

[0013] Further, according to the corrected echo signals, the frequency spectrum characteristics and the time sequence distribution information of the echo signals are extracted and normalized through time window segmentation, signal arrival time calculation and denoising processing, including:

[0014] According to the corrected echo signals, the echo signals are separated based on time window segmentation and signal arrival time calculation, the sound wave echo intensity, scattering mode and arrival time of different frequencies are recorded; the sound wave propagation path in the water area is calculated according to the time delay of the received signals, and the distance, azimuth, depth and position coordinates corresponding to the echo signals are recorded; the acquired echo signals are denoised by a band-pass filter to remove environmental noise and device interference in the non-target frequency range; the time sequence signal is converted into a time-frequency domain signal by a short-time Fourier transform algorithm, and the frequency spectrum characteristic data of the echo is extracted; the amplitude of the echo signal intensity data is normalized by a normalization method, the power amplitudes of different frequency signals are unified, and the frequency distribution and time variation mode are recorded, to obtain the normalized frequency spectrum characteristic data and the time sequence distribution information thereof.

[0015] Further, the frequency response characteristic data of the fish species and the spatial position information thereof are generated by time sequence analysis of the multi-frequency signals, frequency response gradient calculation and envelope curve extraction, including:

[0016] According to the normalized frequency spectrum characteristic data, including the time sequence and the frequency spectrum characteristic of the multi-frequency signal, the amplitude and phase information of the signals of different frequency bands are extracted by using a fast Fourier transform, and the time sequence data under each frequency is recorded; according to the frequency time sequence signals, the formula I = 20·log 10(|S|) converts the time series signal into intensity values, and calculates the mean and variance of the echo signal intensity in each frequency range to obtain the scattering intensity characteristics corresponding to different frequencies, wherein S is the amplitude of the normalized signal; according to the scattering intensity characteristics data corresponding to different frequencies, fitting analysis is performed on the scattering intensity data of different frequencies, a curve of scattering intensity changing with frequency is established, and the gradient value of the curve is calculated extracts the intensity change rate between different frequencies to judge the sensitivity of fish to frequency; according to the range of the frequency change gradient, the characteristics are grouped to form the characteristic template of the frequency response of fish, and the frequency sensitivity characteristics of fish species are obtained, including the frequency response change gradient and the characteristic template; the frequency response data of various fish at each frequency is extracted, the characteristic separation index of different fish at each frequency band is calculated, and the frequency band scaling of the signal characteristic overlapping region is performed; according to the time series signal of each frequency, the envelope curve of the signal is extracted by using Hilbert transform to obtain the shape characteristics of the reflection signal; according to the falling part of the echo signal envelope, an exponential fitting model S(t) = S0·e -R·t calculates the attenuation rate R of the signal to obtain the shape characteristics and attenuation rate characteristics of the fish echo signal, wherein S(t) is the envelope amplitude, and S0 is the starting value of the envelope curve; the scattering intensity characteristics of the sound wave of each frequency, the frequency sensitivity characteristics of fish species, the shape characteristics and the attenuation rate characteristics of the fish echo signal are combined according to the frequency range to generate the frequency response characteristic data table of fish species, and the frequency response characteristic data of fish species is associated with the distance, azimuth, depth and position coordinates of the water area where the fish is located.

[0017] Further comprising, extracting the frequency response data of various fish at each frequency, calculating the characteristic separation index of different fish at each frequency band, and performing frequency band scaling on the signal characteristic overlapping region, specifically comprising:

[0018] extracting various fish at each frequency from the echo signal, according to the formula calculating the frequency response gradient G(f), recording the G(f) value of each fish in the full frequency band and establishing a frequency response gradient database, wherein S(f) represents the normalized signal intensity at frequency f, and Δf represents the frequency increment; according to the frequency response gradient of each fish, using the characteristic separation index formula Calculate the characteristic separation index I(f) of each fish in each frequency band, where G1(f) and G2(f) are the frequency response gradients of the two fish at frequency f, and σ1(f) and σ2(f) are the standard deviations of the signal intensity of the two fish at frequency f; based on a preset separation index threshold, mark the frequency region with a characteristic separation index I(f) less than the preset separation index threshold as a signal characteristic overlap region; perform frequency band scaling on the frequency range of the signal characteristic overlap region through a frequency scaling formula f' = ξ·(f-f0)+f0, where f' is the scaled frequency, ξ is a scaling factor for amplifying or compressing the frequency interval, obtained by fitting historical data, and f0 is the center point of the frequency, maintaining the consistency of frequency band translation; calculate the frequency response gradient G'(f) and the separation index I'(f) again for the scaled signal, if I'(f) of all frequency bands is greater than or equal to the preset separation index threshold, there is no signal characteristic overlap problem, otherwise adjust the scaling factor until the separation index of all frequency bands reaches or exceeds the threshold

[0019] Further, the frequency response characteristic data of fish under multi-frequency sound waves in the fish sound wave monitoring database is used to obtain the spatial distribution, population proportion and mixed fish group marking information of the fish school through signal classification and three-dimensional point cloud clustering analysis, including:

[0020] Through the fish sound wave monitoring database, the frequency response characteristic data of fish species under multi-frequency sound waves of different fish is obtained, a recurrent neural network is used for model training, and a fish recognition model is constructed; according to the real-time acquired frequency response characteristic data of fish species, the fish recognition model is used to determine the fish label and classification confidence of each echo signal; if the classification confidence is lower than the preset confidence threshold, it is marked as an invalid signal and eliminated; the classification results of all echo signals are counted to obtain the overall distribution of fish species in the water area, the proportion and number of different fish populations are determined, and the signal points are mapped to a three-dimensional coordinate system based on the distance, azimuth, depth and position coordinates corresponding to the echo signals, forming point cloud data; according to the point cloud data, a K-means clustering algorithm or a DBSCAN clustering algorithm is used for model training, the signal points belonging to the same fish are input into the clustering algorithm, and the spatial boundary of each fish school is obtained, the size of the fish school in each clustering area, the center position and boundary range of the fish school, wherein the size of the fish school is the number of signal points in the area; the clustering results of different fish in space are detected for overlap, to determine whether two or more species of fish overlap in the same space, if an overlap region is detected, it is marked as a mixed fish school; if there is a mixed fish school, the region is marked separately and the fish species and quantity statistics are output.

[0021] Further, the fish school echo signal and dynamic data obtained in real time by the multi-frequency sonar receiver are used to determine the parameter range of the normal behavior baseline of the fish school, the Mahalanobis distance between the fish school dynamic data and the mean value of the normal behavior baseline of the fish school is calculated to identify the type and range of abnormal behavior of the fish school, and an abnormal intervention strategy for the fish school is developed, including:

[0022] The fish school echo signal is received in real time by the multi-frequency sonar receiver, the timestamp, distance, azimuth, depth and position coordinates of each signal point are obtained, the dynamic data of the fish school are determined, including the position coordinates, group density, movement direction and speed of the fish school; the behavior characteristics of the fish school are determined according to the dynamic data of the fish school, including the density change rate, spatial aggregation characteristics, movement direction consistency and speed change rate of the fish school, the spatial aggregation characteristics are the aggregation degree of the fish school in a specific area, and the movement direction consistency is the similarity of the movement direction of individuals in the group; the K-means clustering algorithm is used for clustering analysis according to the behavior characteristics of the fish school, a normal behavior baseline model of the fish school is established, and the parameter range of the normal behavior baseline of the fish school is determined, including the normal group density change range, aggregation radius and spatial distribution characteristics, average movement direction and speed range; the Mahalanobis distance D between the dynamic data of the fish school and the mean value of the normal behavior baseline of the fish school is calculated by the formula M , to determine the deviation degree of the real-time behavior of the fish school from the normal behavior baseline, wherein ∑ is the covariance matrix; if D M exceeds the preset confidence interval of the normal behavior of the fish school, it is determined that the current behavior of the fish school is abnormal, the abnormal degree of the fish school is judged based on D M and the preset abnormal degree evaluation threshold, and the type and degree of abnormal behavior of the fish school are marked, the type of abnormal behavior includes but is not limited to aggregation, dispersion or high-speed movement, and the degree of abnormality includes mild, moderate and severe; the position coordinates and range of the abnormal behavior of the fish school are determined according to the position coordinates of the fish school, and the group characteristics of the area where the fish school behaves abnormally are counted, including the number and density of the fish school; the behavior characteristics of the fish school in the area where the fish school behaves abnormally are stored in the fish school behavior pattern database; the behavior characteristics of the fish school in the area where the fish school behaves abnormally are obtained from the fish school behavior pattern database, and the abnormal reasons are marked, a random forest algorithm is used for model training to build an abnormal reason identification model of the fish school, and the abnormal reasons include but are not limited to predator threat, environmental disturbance or artificial intervention; the abnormal reason of the fish school is determined using the fish school abnormal reason identification model according to the real-time behavior characteristics of the fish school; based on the time, range, reason and degree of abnormal behavior of the fish school, an abnormal intervention strategy for the fish school is developed and implemented, including but not limited to adjusting water quality parameters, reducing pollution load, increasing dissolved oxygen concentration or optimizing water flow velocity.

[0023] ​Further, the historical dynamic data of the fish school behavior mode database is combined with the normal behavior baseline mean value to predict the time point, area range, abnormal type and cause of the fish school abnormal behavior in a future time period, and the fish school abnormal intervention strategy is implemented in advance, including:

[0024] The historical fish school dynamic data is obtained through the fish school behavior mode database, a long short-term memory network is used for model training, a fish school behavior prediction model is constructed, and fish school dynamic data of the fish school in a preset time period is predicted; based on the predicted fish school dynamic data of the fish school in the preset time period, the Mahalanobis distance of the fish school dynamic data and the normal behavior baseline mean value of the fish school is calculated, and the time point, area range, abnormal type and abnormal cause of the fish school abnormal behavior in the future preset time period are determined; according to the predicted time point of the fish school abnormal behavior, the fish school abnormal intervention strategy is implemented in advance for intervention.

[0025] Further, the fish school dynamic data obtained by the multi-frequency sonar receiver is combined with the Mahalanobis distance and the fish school behavior prediction to evaluate the effect of the fish school abnormal intervention strategy, and the fish school abnormal intervention strategy and the intervention range are dynamically adjusted, including:

[0026] The fish school dynamic data after the fish school abnormal intervention strategy is implemented is obtained through the multi-frequency sonar receiver, and the Mahalanobis distance of the fish school dynamic data and the normal behavior baseline of the fish school is calculated; if the Mahalanobis distance is lower than the preset fish school normal behavior confidence interval, it is judged that the fish school abnormal intervention strategy is effective, and the abnormal behavior area of the fish school returns to normal; if the Mahalanobis distance still exceeds the preset fish school normal behavior confidence interval, it is judged that the fish school abnormal intervention strategy is ineffective, and whether the abnormal area expands to the non-intervention area is judged based on the fish school dynamic data, the fish school abnormal intervention strategy and the intervention range are adjusted, and the fish school abnormal behavior area returns to normal; the fish school dynamic data of the area where the fish school abnormal intervention strategy is implemented in advance is obtained, the fish school dynamic data of the fish school in a preset time period is predicted again based on the fish school behavior prediction model, and the Mahalanobis distance of the fish school dynamic data of the fish school in the preset time period and the normal behavior baseline mean value is calculated according to the predicted fish school dynamic data of the fish school in the preset time period; based on the Mahalanobis distance, it is judged whether the fish school abnormal intervention strategy implemented in advance eliminates the abnormal risk of the abnormal behavior in the future; if the Mahalanobis distance still exceeds the preset fish school normal behavior confidence interval, it is judged that the fish school abnormal intervention strategy implemented in advance is ineffective, and whether the abnormal area expands to the non-intervention area is judged based on the predicted fish school dynamic data of the fish school in the preset time period, and the fish school abnormal intervention strategy and the intervention range are adjusted.

[0027] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0028] The application provides a multi-frequency sound wave and AI algorithm large-scale fish resource detection method. The application realizes high-precision correction of fish echo signals by combining multi-frequency sonar equipment with environmental feature correction technology, effectively eliminates the influence of environmental variables on sound wave propagation, ensures the quality and stability of echo signals, and comprehensively obtains frequency response characteristic data of fish species through normalized spectral feature extraction, multi-frequency signal time series analysis and frequency response characteristic generation, and associates the data with spatial position information, thereby accurately identifying fish species and distribution characteristics. The application accurately identifies the spatial distribution, population proportion and mixed fish group characteristics of fish groups through three-dimensional point cloud clustering analysis, providing high-precision data support for dynamic behavior research and ecological monitoring of fish groups. At the same time, the application can quickly identify abnormal behavior of fish groups and their regional range through real-time monitoring and dynamic data analysis combined with normal behavior baseline models of fish groups, providing a scientific basis for timely intervention of abnormal behavior. In addition, the application uses historical dynamic data combined with prediction models to accurately predict abnormal behavior of fish groups and its causes in future time periods, realizing early intervention and dynamic management of abnormal behavior of fish groups. Through dynamic adjustment of intervention strategies and regional range, abnormal behavior can be effectively eliminated and prevented from spreading. The application not only improves the accuracy and efficiency of fish resource detection, meets the needs of dynamic monitoring and accurate management of fish groups in large-scale water areas, but also quickly identifies the source of problems and effectively intervenes in the case of environmental disturbance or ecological abnormalities, providing a reliable technical foundation and scientific support for fishery resource protection, water ecosystem restoration and environmental management, further promoting the health and sustainable development of water ecological systems. BRIEF DESCRIPTION OF DRAWINGS

[0029] Fig. 1 A flowchart of a multi-frequency sound wave and AI algorithm large-scale fish resource detection method of the application;

[0030] Fig. 2 A schematic diagram of a multi-frequency sound wave and AI algorithm large-scale fish resource detection method of the application;

[0031] Fig. 3 Another schematic diagram of a multi-frequency sound wave and AI algorithm large-scale fish resource detection method of the application. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be described in detail below in combination with the drawings and specific embodiments.

[0033] As Figs. 1-3 , the multi-frequency sound wave and AI algorithm large-scale fish resource detection method of the embodiment can specifically include:

[0034] Step S101, adjust the sound wave emission parameters through the multi-frequency sonar device, and correct the original echo signal to the signal under the standard environmental characteristics combined with the environmental characteristics.

[0035] According to the preset detection water area and the possible species distribution range of the target fish, a multi-frequency sonar device is used to configure multiple frequency ranges, and the order of low-frequency, medium-frequency and high-frequency emission signals is set. Through the sound wave emission system of the multi-frequency sonar device, the sound wave emission power, the sound beam width and the emission angle are adjusted, and the sound wave signals are emitted to the specified water area one by one. The original echo signal of the multi-frequency sound wave is obtained through the multi-frequency sonar receiver, the spatial position information of the sound wave emission is marked, and is stored to the fish sound wave monitoring database. Through the fish sound wave monitoring database, the echo signal under the standard environmental characteristics and the echo signal under the non-standard environmental characteristics of the same distance, azimuth and depth are obtained, a recurrent neural network is used for model training, and an echo signal correction model is constructed. The environmental characteristics include the temperature, salinity, flow rate and turbidity of the water body. According to the real-time obtained original echo signal and the environmental characteristics, the echo signal correction model is used to correct the original echo signal to the echo signal under the standard environmental characteristics.

[0036] Exemplarily, in an aquaculture farm with an area of 1000 square meters and a water depth of 10 meters, multiple types of fish including carp and sea bass are bred, a multi-frequency sonar device is used to configure low-frequency signals of 30-50 kHz, medium-frequency signals of 50-100 kHz and high-frequency signals of 100-300 kHz, and signals are emitted in order from low frequency to high frequency. The sound wave emission system adjusts the sound wave emission power to 150 W, the sound beam width to 30°, and the emission angle to 45° below the horizontal plane, and sequentially emits sound wave signals to different areas of the designated water area. The multi-frequency sonar receiver receives the echo signals and records the spatial position information of the sound wave emission, the spatial coordinates of the emission point are (0, 0, 0), and the spatial position of the target fish is (5m, 3m, -3m), wherein x is the horizontal distance, y is the longitudinal distance, and z is the depth. The received echo signal strength is -40 dB, the echo delay time is 0.02 s, and these data are stored in the fish sound wave monitoring database. From the database, the echo signal of the same target distance (5m, 3m, -3m) under standard environmental characteristics (temperature 20℃, salinity 30‰, flow rate 0.1m / s, turbidity 1NTU) is intensity -35dB, delay time 0.018s, and under non-standard environmental characteristics, temperature 25℃, salinity 35‰, flow rate 0.3m / s, turbidity 5NTU, the echo signal is intensity -50dB, delay time 0.025s, and a recurrent neural network model is trained to construct an echo signal correction model. According to the original echo signal with intensity -50dB and delay time 0.025s obtained in real time and the current environmental characteristics, temperature 25℃, salinity 35‰, flow rate 0.3m / s, and turbidity 5NTU, the echo signal correction model is input, and after model correction, the original echo signal is adjusted to the corrected signal under standard environmental characteristics, intensity -40dB, and delay time 0.02s.

[0037] In step S102, according to the corrected echo signal, the frequency spectrum characteristics and time sequence distribution information of the echo signal are extracted and normalized by time window segmentation, signal arrival time calculation and denoising processing.

[0038] According to the corrected echo signal, the echo signal is separated based on time window segmentation and signal arrival time calculation, and the echo intensity, scattering pattern and arrival time of sound waves of different frequencies are recorded. According to the time delay of the received signal, the sound wave propagation path in the water area is calculated, and the distance, azimuth, depth and position coordinates corresponding to the echo signal are recorded. The obtained echo signal is denoised by a band-pass filter to remove environmental noise and equipment interference outside the target frequency range. The time series signal is converted to a time-frequency domain signal by a short-time Fourier transform algorithm, and the spectral feature data of the echo is extracted. The echo signal intensity data is amplitude normalized using a normalization method, the power amplitudes of signals of different frequencies are unified, and the frequency distribution and time variation pattern are recorded to obtain normalized spectral feature data and time sequence distribution information.

[0039] For example, if the corrected echo signal has a low frequency of 30 kHz, a medium frequency of 80 kHz, and a high frequency of 200 kHz, the intensity is -40 dB, -45 dB, and -50 dB, respectively, and the echo delay time is 0.02 s, 0.018 s, and 0.015 s, respectively. The echo signal is separated based on time window segmentation and signal arrival time calculation, and the echo intensity, scattering pattern and arrival time of sound waves of different frequencies are recorded. According to the time delay of the echo signal, the time delays of the low frequency echo, the medium frequency echo and the high frequency echo are 0.02 s, 0.018 s and 0.015 s, respectively. The propagation path corresponding to the low frequency echo is 30 m, the propagation path corresponding to the medium frequency echo is 27 m, and the propagation path corresponding to the high frequency echo is 22.5 m. Combined with the emission angle and the sound beam direction, the spatial coordinates corresponding to the echo signal are recorded as (25 m, 10 m, -3 m), (20 m, 12 m, -4 m) and (18 m, 15 m, -5 m). The echo signals of 30 kHz, 80 kHz and 200 kHz are denoised by a band-pass filter, respectively, to remove environmental noise such as water flow sound, ship noise and equipment interference outside the target frequency range. The time series signal is converted to a time-frequency domain signal by a short-time Fourier transform algorithm, and the spectral feature data of each frequency is extracted to obtain the spectral peak of the 30 kHz signal near 31 kHz, the spectral peak of the 80 kHz signal near 79 kHz, and the spectral peak of the 200 kHz signal near 201 kHz. The echo signal intensity is amplitude normalized using a normalization algorithm, the intensities of the low frequency, medium frequency and high frequency signals are normalized to the amplitude range of 0.8, 0.7 and 0.6, respectively, the power amplitudes of signals of different frequencies are unified, and the frequency distribution time variation pattern is recorded to obtain the intensity of 30-50 kHz concentrated above 0.75, and the signal intensity reaches a peak within 0.015-0.025 s.

[0040] Step S103, through time series analysis of multi-frequency signal, frequency response gradient calculation and envelope curve extraction, the frequency response characteristic data of fish species is generated and the spatial position information is associated.

[0041] According to the normalized spectral feature data, including the time series of multi-frequency signal and its spectral feature, the amplitude and phase information of different frequency band signals are extracted by using fast Fourier transform, and the time series data at each frequency is recorded. According to the time series signal at each frequency, the formula I = 20·log 10 (|S|) is used to convert the time series signal into intensity value, and the mean and variance of echo signal intensity are calculated in each frequency range to obtain the scattering intensity characteristics corresponding to different frequencies, wherein S is the amplitude of normalized signal. According to the scattering intensity characteristic data corresponding to different frequencies, fitting analysis is performed on the scattering intensity data of different frequencies, a curve of scattering intensity changing with frequency is established, and the gradient value of the curve is calculated The intensity change rate between different frequencies is extracted to judge the sensitivity of fish to frequency. According to the range of frequency change gradient, the characteristics are grouped to form the characteristic template of fish frequency response, and the frequency sensitivity characteristics of fish species are obtained, including the frequency response change gradient and the characteristic template. The frequency response data of various fish at each frequency is extracted, the characteristic separation index of different fish at each frequency band is calculated, and the frequency band scaling of signal characteristic overlapping area is performed. According to the time series signal at each frequency, the envelope curve of the signal is extracted by using Hilbert transform to obtain the shape feature of the reflected signal. According to the falling part of the echo signal envelope, an exponential fitting model S(t) = S0·e -R·t The decay rate R of the signal is calculated to obtain the shape feature and decay rate characteristics of fish echo signal, wherein S(t) is the envelope amplitude, and S0 is the starting value of the envelope curve. The scattering intensity characteristics of each frequency sound wave, the frequency sensitivity characteristics of fish species, the shape feature and decay rate characteristics of fish echo signal are combined according to the frequency range to generate the frequency response characteristic data table of fish species, and the frequency response characteristic data of fish species is associated with the distance, azimuth, depth and position coordinates of the water area where the fish is located.

[0042] For example, according to the normalized spectral feature data, including the time series of multi-frequency signal and its spectral feature, in the three frequency bands of 30 kHz, 80 kHz and 200 kHz, the amplitude of the signal is 0.8, 0.6 and 0.4 respectively, the signal amplitude and phase information of the three frequency bands are extracted by using fast Fourier transform, and the time series signal data is recorded, such as the signal amplitude in the 30 kHz frequency band changes with time, the starting value is 0.8, after 0.01 s, it drops to 0.6, and at 0.02 s, it drops to 0.4. According to these time series signals, the formula I = 20·log 10(|S|) converts the amplitude to the intensity value, and the intensity values of the 30 kHz signal are -1.94 dB, 0.8 amplitude, -4.44 dB, 0.6 amplitude, -7.96 dB, 0.4 amplitude, and the average intensity of the frequency band is -4.78 dB, and the variance is 2.55 dB2, wherein S is the amplitude of the normalized signal. By fitting and analyzing the scattering intensity characteristic data of the three frequency bands of 30 kHz, 80 kHz and 200 kHz, the curve of the scattering intensity changing with the frequency is established, and the average intensity corresponding to 30 kHz, 80 kHz and 200 kHz is -4.78 dB, -6.02 dB and -8.12 dB respectively. The fitting curve shows that the intensity decreases linearly with the increase of the frequency, and the gradient value of the curve is The gradient value is used to judge the sensitivity of fish to frequency, and it is found that fish respond more strongly to the low frequency band of 30-80 kHz, and the response gradually weakens in the high frequency band of 80-200 kHz. According to the range of the frequency gradient, the frequency gradient is divided into low frequency gradient of -1.68 to -1.00, medium frequency gradient of -1.00 to -0.50 and high frequency gradient less than -0.50, forming a characteristic template of fish frequency response, and generating a fish template containing gradient range and frequency response characteristics. According to the time series signal of each frequency band, the envelope curve of the signal is extracted by using Hilbert transform. In the 30 kHz frequency band, the initial amplitude S0 of the envelope curve is 0.8, and then the signal decays with time. The exponential fitting model S(t) = S0·e -R·t The decay rate R of the signal is calculated, and the decay rate R = 50 s -1 , wherein S(t) is the envelope amplitude, and S0 is the initial value of the envelope curve. The scattering intensity characteristics of the sound waves of each frequency, the frequency sensitivity characteristics of the fish, such as strong response to 30 kHz and weak response to 200 kHz, the shape characteristics of the fish echo signal, such as envelope curve and decay rate, are combined to generate a frequency response characteristic data table of fish species, such as marking the target fish as species A, and recording the frequency response data table of 30 kHz as -4.78 dB, the frequency change gradient as -1.68 dB / kHz, and the envelope decay rate as 50 s -1 . The frequency response characteristics of the fish species are associated with the space information of the water area where the fish is located, such as distance 10 m, azimuth angle 45°, depth 5 m, and position coordinates (7.1 m, 7.1 m, -5 m).

[0043] wherein the frequency response data of various fish at each frequency is extracted, the characteristic separation index of different fish at each frequency band is calculated, and the frequency band signal intensity of the signal characteristic overlap area is scaled.

[0044] The frequency response gradient G(f) of each fish is calculated according to the formula The G(f) values of each fish in the full frequency range are recorded and a frequency response gradient database is established, where S(f) represents the normalized signal intensity at frequency f, and Δf represents the frequency increment. According to the frequency response gradient of each fish, the characteristic separation index I(f) of each fish in each frequency band is calculated using the characteristic separation index formula where G1(f) and G2(f) are the frequency response gradients of two fish species at frequency f, and σ1(f) and σ2(f) are the standard deviations of the signal intensity of these two fish species at frequency f. Based on a pre-set separation index threshold, the frequency region with a characteristic separation index I(f) less than the pre-set separation index threshold is marked as a signal characteristic overlap region. The frequency range of the signal characteristic overlap region is scaled by the frequency scaling formula f' = ξ · (f - f0) + f0, where f' is the scaled frequency, ξ is the scaling factor used to amplify or compress the frequency interval, which is fitted from historical data, and f0 is the center point of the frequency, which maintains the consistency of frequency band translation. For the scaled signal, the frequency response gradient G'(f) and the separation index I'(f) are calculated again. If I'(f) of all frequency bands is greater than or equal to the pre-set separation index threshold, there is no signal characteristic overlap problem, otherwise the scaling factor is adjusted until the separation index of all frequency bands reaches or exceeds the threshold.

[0045] For example, the frequency response characteristic data of two fish species are extracted from the echo signal, denoted as fish A and fish B, respectively. In the frequency range f = 10 kHz to f = 50 kHz, the normalized signal intensity S(f) is recorded with a frequency increment Δf of every 2 kHz, such as at f = 10 kHz and f = 12 kHz, the normalized signal intensity of fish A is S A (10) = 0.8 and S A (12) = 0.6, respectively. According to the formula the frequency response gradient of fish A at 10 kHz is -0.1 kHz -1 . The signal intensity of fish B in the same frequency range is 0.75 and 0.7, respectively, so the frequency response gradient of fish B is -0.025 kHz -1 . After recording the G(f) values of each fish in the full frequency range, a frequency response gradient database is established. The characteristic separation index I(f) of each fish in each frequency band is calculated using the characteristic separation index formula For example, at 10 kHz, the frequency response gradients of fish A and B are G A (10) = -0.1 and G B (10) = -0.025, respectively, and the standard deviations of the signal intensity are σ A (10) = 0.05 and σB (10) = 0.03, the feature separation index is 0.9375. The preset feature separation index threshold is 1, so I(10) = 0.9375 < 1.0, indicating that there is a signal feature overlap problem in the 10 kHz frequency range. For the frequency range marked as signal feature overlap, the frequency range is scaled by the frequency scaling formula f' = ξ · (f-f0) + f0, where the scaling factor ξ is obtained by fitting historical data, used to enlarge the frequency interval, and the frequency center point f0 is 11 kHz, used to maintain the consistency of the translation of the frequency range. After calculation, the scaled frequency of 10 kHz is 9.8 kHz, and the scaled result of 12 kHz is 12.2 kHz. After scaling, the frequency range is adjusted to 9.8 kHz to 12.2 kHz, which enlarges the frequency interval and enhances the discrimination ability of different fish in this frequency range. The frequency response gradient and feature separation index are recalculated for the scaled signal. For example, the gradient of fish A after scaling is GA(9.8) = -0.12, the gradient of fish B after scaling is GB(9.8) = -0.02, the signal intensity standard deviation remains unchanged, the feature separation index is calculated as 1.25, which is greater than the preset feature separation index threshold 1, and meets the separation index threshold requirement, so it can be determined that the signal feature overlap problem has been solved. If the separation index still does not meet the threshold requirement, the scaling factor ξ or the preprocessing parameter can be adjusted for further optimization until the separation index of all frequency ranges reaches or exceeds the threshold. B (9.8) = -0.02, the signal intensity standard deviation remains unchanged, the feature separation index is calculated as 1.25, which is greater than the preset feature separation index threshold 1, and meets the separation index threshold requirement, so it can be determined that the signal feature overlap problem has been solved. If the separation index still does not meet the threshold requirement, the scaling factor ξ or the preprocessing parameter can be adjusted for further optimization until the separation index of all frequency ranges reaches or exceeds the threshold.

[0046] Step S104, according to the frequency response characteristic data of the fish sound wave under the multi-frequency sound wave in the fish sound wave monitoring database, the spatial distribution, population proportion and mixed fish group marking information of the fish school are obtained by signal classification and three-dimensional point cloud clustering analysis.

[0047] Through the fish acoustic monitoring database, the frequency response characteristic data of different fish species under multi-frequency acoustic waves is obtained, a recurrent neural network is used for model training, and a fish recognition model is constructed. According to the real-time acquired frequency response characteristic data of the fish species, the fish recognition model is used to determine the fish label and the classification confidence of each echo signal. If the classification confidence is lower than the pre-set confidence threshold, it is marked as an invalid signal and eliminated. The classification results of all echo signals are counted to obtain the overall distribution of fish species in the water area, determine the proportion and quantity of different fish populations, and map the signal points to a three-dimensional space coordinate system based on the distance, azimuth, depth and position coordinates corresponding to the echo signals to form point cloud data. According to the point cloud data, a K-means clustering algorithm or a DBSCAN clustering algorithm is used for model training, the signal points belonging to the same fish species are input into the clustering algorithm, the spatial boundary of each fish school is obtained, the size of the fish school, the center position and the boundary range of the fish school in each clustering area, wherein the size of the fish school is the number of signal points in the area. The clustering results of different fish species in the space are detected for overlap, to determine whether two or more species of fish overlap in the same space, if an overlapping area is detected, it is marked as a mixed fish school. If there is a mixed fish school, the area is marked separately and the fish species and quantity statistics are output.

[0048] For example, the frequency response characteristic data of carp, sea bass and mandarin fish at multiple frequencies 30 kHz, 80 kHz and 200 kHz are obtained from the fish acoustic monitoring database. The average scattering intensity of carp at 30 kHz is 0.8, and the variance is 0.1. The frequency sensitivity shows that the low-frequency response intensity is high. The average scattering intensity of sea bass at 80 kHz is 0.75, and the variance is 0.08. The middle-frequency response is strong. The average scattering intensity of mandarin fish at 200 kHz is 0.85, and the variance is 0.05. The high-frequency response is most obvious. The fish recognition model is constructed by training the frequency response characteristic data through the recurrent neural network. When the echo signals of the multi-frequency acoustic waves are received in real time, the fish recognition model is used to classify each echo signal. The classification results show that the fish labels corresponding to the echo signals are carp, sea bass and mandarin fish, and the classification confidence is 90%, 85% and 60% respectively. Among them, the classification confidence of mandarin fish is lower than the preset threshold 70%, so it is marked as invalid signal and eliminated. Finally, the classified effective echo signals are carp and sea bass. According to the classification results of all echo signals, it is determined that the population proportion of carp in the water area is 60%, and the number is 120. The population proportion of sea bass is 40%, and the number is 80. Combined with the spatial position information of the echo signals, the coordinates corresponding to the carp signals are (10m, 5m, -2m), and the coordinates corresponding to the sea bass signals are (20m, 8m, -3m). Map these coordinate points to the three-dimensional coordinate system to form point cloud data. The K-means clustering algorithm is used to analyze the point cloud data, and the clustering center position of carp is (12m, 6m, -2.5m), and the boundary range of the clustering area is (8-16m, 4-8m, -3--2m). The clustering center position of sea bass is (22m, 10m, -3.5m), and the boundary range of the clustering area is (18-26m, 7-13m, -4--3m). The clustering results of different fish are detected, and it is found that the clustering boundaries of carp and sea bass overlap in some areas. The number of signal points in this area is 30, which is marked as a mixed fish group area. The number of carp in the mixed area is 20, and the number of sea bass is 10. Finally, the mixed fish group area is marked as a special area.

[0049] In step S105, the parameter range of the fish school normal behavior baseline is determined according to the fish school echo signals and dynamic data obtained by the multi-frequency sonar receiver in real time. The Mahalanobis distance between the fish school dynamic data and the mean value of the fish school normal behavior baseline is calculated to identify the type and range of fish school abnormal behavior, and to develop fish school abnormal intervention strategies.

[0050] The fish school echo signal is received in real time by the multi-frequency sonar receiver, the timestamp, distance, azimuth, depth and position coordinates of each signal point are obtained, and the fish school dynamic data are determined, including the position coordinates, group density, movement direction and movement rate of the fish school. According to the fish school dynamic data, the fish school behavior characteristics are determined, including the fish school density change rate, spatial aggregation characteristics, movement direction consistency and speed change rate. The spatial aggregation characteristics are the aggregation degree of the fish school in a specific area, and the movement direction consistency is the similarity of the movement direction of individuals in the group. According to the fish school behavior characteristics, clustering analysis is performed using the K-means clustering algorithm, a fish school normal behavior baseline model is established, and the parameter range of the fish school normal behavior baseline is determined, including the normal group density change range, aggregation radius and spatial distribution characteristics, average movement direction and speed range. The Mahalanobis distance calculation formula The fish school dynamic data and the fish school normal behavior baseline mean Mahalanobis distance D are calculated M , to determine the deviation degree of the fish school real-time behavior from the normal behavior baseline, wherein ∑ is the covariance matrix. If D M exceeds the preset fish school normal behavior confidence interval, it is determined that the current fish school behavior is abnormal, the fish school abnormal degree is judged based on D M and the preset abnormal degree evaluation threshold, and the fish school abnormal behavior type and abnormal degree are marked. The abnormal behavior type includes but is not limited to aggregation, dispersion or high-speed movement, and the abnormal degree includes mild, moderate and severe. According to the position coordinates of the fish school, the position coordinates and area range of the fish school abnormal behavior occurrence are determined, and the group characteristics of the fish school abnormal behavior occurrence area are counted, including the number and density of the fish school. The fish school behavior characteristics of the fish school abnormal behavior occurrence area are stored in the fish school behavior pattern database. The fish school behavior characteristics of the historical fish school abnormal behavior occurrence area are obtained from the fish school behavior pattern database, and the abnormal reasons are marked. A random forest algorithm is used for model training to construct a fish school abnormal reason identification model. The abnormal reasons include but are not limited to predator threat, environmental disturbance or artificial intervention. According to the real-time obtained fish school behavior characteristics, the fish school abnormal reason identification model is used to determine the fish school abnormal reason. Based on the fish school abnormal behavior occurrence time, area range, fish school abnormal reason and abnormal degree, fish school abnormal intervention strategies are developed and implemented, including but not limited to adjusting water quality parameters, reducing pollution load, increasing dissolved oxygen concentration or optimizing water flow velocity.

[0051] For example, the fish school echo signal is received in real time by the multi-frequency sonar receiver, the timestamp, distance, azimuth, depth and position coordinates of each signal point are obtained, and the echo signal of a certain fish school shows that its position coordinates are (10m, 15m, -5m) and the timestamp is 10s. According to the multi-point echo data calculation, the group density of the fish school is 10m 3The fish school contains 20 fish, and the fish school moves at a direction of 45° relative to the north direction at a speed of 0.5 m / s. According to the dynamic data of the fish school, the behavior characteristics of the fish school are determined, including a fish school density change rate of 2 fish per second, a spatial aggregation characteristic showing that the fish school is concentrated in a region with a center point (10 m, 15 m, -5 m) and a radius of 5 m, a motion direction consistency of 85%, indicating that most individuals in the group have consistent motion directions, and a speed change rate of 0.1 m / s per second. According to the behavior characteristics of the fish school, a K-means clustering algorithm is used for clustering analysis of the fish school to establish a normal behavior baseline model of the fish school and determine the normal behavior parameter range of the fish school, such as a normal group density change range of ± 3 fish per second, an aggregation radius range of 3-6 m, a motion direction consistency range of 80%-95%, and a speed range of 0.3-0.7 m / s. The deviation degree of the current behavior of the fish school from the baseline mean is calculated by the Mahalanobis distance calculation formula. If the Mahalanobis distance D M = 4.5 exceeds the preset normal behavior confidence interval of 95%, and the confidence interval corresponds to D M ≤ 3, it is determined that the behavior of the fish school is abnormal. Based on the deviation degree D M = 4.5 and the preset abnormality evaluation threshold, it is determined that the abnormality is a serious abnormality, and it is labeled as a high-speed moving behavior type. According to the position coordinates of the fish school, the range of the region where the abnormal behavior occurs is determined to be a space with a center point (10 m, 15 m, -5 m) and a radius of 5 m. The fish school characteristics of the region are counted, including a fish school quantity of 100 fish and a density of 25 fish per 10 m 3 . The fish school behavior characteristics of the abnormal behavior region are stored in the fish school behavior pattern database, and the abnormal behavior records of similar regions are extracted from the historical data in the database, and the abnormal reasons are labeled, such as the similar situations in the historical data are often related to predator threats. After training the model by a random forest algorithm, a fish school abnormal reason recognition model is constructed, real-time data is input into the fish school abnormal reason recognition model, it is determined that the current abnormal reason is a predator threat, and the fish school abnormal behavior occurs at a time of 10 s, a center point (10 m, 15 m, -5 m), a region range of a radius of 5 m, an abnormal reason of a predator threat, and an abnormal degree of a serious abnormality. A fish school abnormal intervention strategy is developed. The intervention measures include deploying high-frequency sound wave equipment to drive away predators, increasing the dissolved oxygen concentration in the water to reduce the stress of the fish school, and continuously monitoring the environmental parameters of the region, including water temperature, dissolved oxygen, and pollutant concentration, to confirm that the abnormality is resolved.

[0052] In step S106, according to the historical dynamic data of the fish school behavior pattern database, the normal behavior baseline mean is combined to predict the time point, region range, abnormal type and reason of the fish school abnormal behavior in the future time period, and the fish school abnormal intervention strategy is implemented in advance.

[0053] The fish school behavior pattern database is used to obtain historical fish school dynamic data, a long short-term memory network is used for model training, a fish school behavior prediction model is constructed, and fish school dynamic data of the fish school in a preset time period is predicted. Based on the fish school dynamic data of the fish school in the preset time period predicted, the Mahalanobis distance of the fish school dynamic data and the normal behavior baseline mean value of the fish school is calculated, and the time point, the area range, the abnormal type and the abnormal reason of the abnormal behavior of the fish school in the future preset time period are determined. According to the predicted time point of the abnormal behavior of the fish school, the fish school abnormal intervention strategy is implemented in advance for intervention.

[0054] For example, through the fish school behavior pattern database, the fish school dynamic data of the past 1 week is obtained, including the spatial position, the motion direction, the density change, the speed change and other characteristics of the fish school. The historical data record shows that under the sound waves of low frequency 30 kHz, medium frequency 80 kHz and high frequency 200 kHz, the echo signals of different fish schools show that the average motion speed is 0.4 m / s, the density change rate is in the range of ±5 fish per second, and the motion direction consistency is 80%-95%. Using these data, a long short-term memory network is used for model training, and a fish school behavior prediction model is constructed. Through model prediction of the fish school dynamic data in the next 24 hours, it is predicted that in the future 10th hour, the aggregation radius of the fish school at the position (15m, 30m, -4m) will be reduced to below 5m, and the group density will be significantly increased to 100 fish per 10m 3 The motion speed of the 30 fish will increase to 1.2 m / s, and the motion direction consistency will decrease to 60%. The predicted dynamic data and the normal behavior baseline mean value of the fish school are calculated by Mahalanobis distance, and the Mahalanobis distance D M =4.8 is obtained, which exceeds the 95% confidence interval of the normal behavior baseline, corresponding to D M =3, it is determined that the fish school will have abnormal behavior in the 10th hour in this area. Based on the classification output of the model, the abnormal behavior type is predicted to be high-speed movement accompanied by aggregation, and the abnormal reason is predicted to be a predator threat. According to the prediction result, the intervention strategy is formulated and implemented in advance, and high-frequency sound wave equipment is deployed in the area at the position (15m, 30m, -4m) to drive away possible predators before the predicted time arrives, and oxygen increasing equipment is added around to increase the dissolved oxygen concentration to alleviate the stress reaction of the fish school.

[0055] In step S107, the fish school abnormal intervention strategy effect is evaluated according to the fish school dynamic data obtained by the multi-frequency sonar receiver, combined with Mahalanobis distance and fish school behavior prediction, and the fish school abnormal intervention strategy and intervention range are dynamically adjusted.

[0056] The multi-frequency sonar receiver is used to obtain fish dynamic data after the fish abnormal intervention strategy is implemented, and Mahalanobis distance of the fish dynamic data and the fish normal behavior baseline is calculated. If the Mahalanobis distance is lower than the preset fish normal behavior confidence interval, it is judged that the fish abnormal intervention strategy is effective, and the abnormal behavior region of the fish is restored to normal. If the Mahalanobis distance is still beyond the preset fish normal behavior confidence interval, it is judged that the fish abnormal intervention strategy is ineffective, and whether the abnormal region expands to the non-intervention region is judged based on the fish dynamic data, and the fish abnormal intervention strategy and the range of the strategy implementation are adjusted until the fish abnormal behavior region is restored to normal. The fish dynamic data of the region where the fish abnormal intervention strategy is implemented in advance is obtained, the fish dynamic data of the fish in the preset time period is predicted again based on the fish behavior prediction model, and the Mahalanobis distance of the fish dynamic data of the fish in the preset time period and the fish normal behavior baseline mean value is obtained. Based on the Mahalanobis distance, it is judged whether the fish abnormal intervention strategy implemented in advance eliminates the abnormal risk of the future abnormal behavior. If the Mahalanobis distance is still beyond the preset fish normal behavior confidence interval, it is judged that the fish abnormal intervention strategy implemented in advance is ineffective, and whether the abnormal region expands to the non-intervention region is judged based on the fish dynamic data of the fish in the preset time period predicted, and the fish abnormal intervention strategy and the range of the strategy implementation are adjusted.

[0057] For example, the center point of a certain abnormal region is located at (20m, 30m, -3m), and the fish dynamic data after the fish abnormal intervention strategy is implemented is obtained in real time by the multi-frequency sonar receiver. For example, the density of the fish is 20 fish per 10m 3 , the movement speed is 0.8m / s, the direction consistency is 70%, and the aggregation radius is 6m. By comparing these dynamic data with the fish normal behavior baseline, the Mahalanobis distance D M =2.5 is calculated by the Mahalanobis distance formula, which is lower than the threshold value of the preset normal behavior confidence interval of 95% confidence interval, corresponding to D M ≤3, so it is judged that the current fish abnormal intervention strategy is effective, and the abnormal behavior of the region has been restored to normal. Subsequently, the dynamic data of the region intervened in advance is continuously monitored, and the latest fish dynamic data is recorded by the sonar receiver. In the region intervened in advance, the density is maintained at 20 fish per 10m 3 , the movement speed is 0.6m / s, the direction consistency is 85%, and the aggregation radius is 8m. Based on these data, the fish dynamic data in the next 12 hours is predicted again by using the fish behavior prediction model. The prediction result shows that the fish dynamic data in the region tends to be normal in the next 6 hours, and the Mahalanobis distance D M= 2.8, still within the preset normal behavior confidence interval, further confirming that the early intervention strategy is effective in eliminating the future risk of abnormal behavior. Through monitoring of other non-intervention areas, it is found that the fish density in a new area (25m, 35m, -4m) outside the abnormal area rises to 30 fish per 10m 3 M = 4.2, beyond the normal behavior confidence interval, indicating that abnormal behavior begins to expand to non-intervention areas. Therefore, the intervention strategy is adjusted to expand the coverage of the intervention device to the new abnormal area, such as deploying a high-frequency sound device in the (25m, 35m, -4m) area to drive away predators, while increasing the dissolved oxygen concentration around the area to reduce the stress response of the fish school. After expanding the intervention measures, real-time monitoring shows that the fish density in the new area drops to 22 fish per 10m 3 M = 2.9, lower than the preset normal behavior confidence interval, judging that the expanded fish school abnormal intervention strategy successfully controls the abnormal behavior and effectively prevents further expansion of the abnormal area.

[0058] The above description is only the preferred embodiment of the present application and the explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features can be replaced with the technical features disclosed in the present application (but not limited to) with similar functions to form technical solutions.​​

Claims

1. A method for detecting large-scale fish resources by multi-frequency sound waves and AI algorithms, characterized in that, The method comprises: Adjust the sound wave emission parameters through the multi-frequency sonar device, correct the original echo signal to the signal under the standard environmental characteristics combined with the environmental characteristics; According to the corrected echo signal, through time window segmentation, signal arrival time calculation and denoising processing, the spectral characteristics and time sequence distribution information of the echo signal are extracted and normalized; Through time sequence analysis, frequency response gradient calculation and envelope curve extraction of multi-frequency signals, frequency response characteristic data of fish species are generated and their spatial position information is associated, including: According to the normalized spectral feature data, including the time series of the multi-frequency signal and the spectral features thereof, the amplitude and phase information of the signals of different frequency bands are extracted by using fast Fourier transform, and the time series data at each frequency is recorded; according to the time series signals at each frequency, the formula is used to convert the time series signals into intensity values, and the mean and variance of the echo signal intensity are calculated in each frequency range to obtain the scattering intensity characteristics corresponding to different frequencies, wherein S is the amplitude of the normalized signal; according to the scattering intensity characteristic data corresponding to different frequencies, the scattering intensity data of different frequencies are analyzed by fitting, a curve of the scattering intensity changing with the frequency is established, and the gradient value of the curve is calculated to extract the intensity change rate between different frequencies and judge the sensitivity of fish to the frequency; according to the range of the frequency change gradient, the features are grouped to form the feature templates of the frequency response of fish, and the frequency sensitivity features of fish species are obtained, including the frequency response change gradient and the feature templates; the frequency response data of various fish at each frequency is extracted, the characteristic separation index of different fish at each frequency band is calculated, and the frequency band scaling of the signal feature overlapping area is performed; according to the time series signals at each frequency, the envelope curve of the signal is extracted by using Hilbert transform to obtain the shape feature of the reflected signal; according to the falling part of the envelope of the echo signal, an exponential fitting model is used to calculate the decay rate R of the signal to obtain the shape feature and decay rate characteristics of the echo signal of fish, wherein is the envelope amplitude, is the starting value of the envelope curve; the scattering intensity characteristics of the sound waves at each frequency, the frequency sensitivity features of fish species, the shape features and decay rate characteristics of the echo signal of fish are combined according to the frequency range to generate the frequency response feature data table of fish species, and the frequency response feature data of fish species is associated with the distance, azimuth, depth and position coordinates of the water area where the fish is located; According to the frequency response characteristic data of fish under multi-frequency sound waves in the fish sound wave monitoring database, the spatial distribution, population proportion and mixed fish group marking information of fish schools are obtained through signal classification and three-dimensional point cloud clustering analysis; According to the fish school echo signal and dynamic data obtained by the multi-frequency sonar receiver in real time, the parameter range of the normal behavior baseline of the fish school is determined, the Mahalanobis distance between the fish school dynamic data and the mean value of the normal behavior baseline of the fish school is calculated, the fish school abnormal behavior type, area range are identified, and the fish school abnormal intervention strategy is formulated, including: Real-time fish echo signals were received using a multi-frequency sonar receiver to obtain the timestamp, distance, azimuth, depth, and location coordinates of each signal point, thus determining the fish school's dynamic data, including its location coordinates, population density, direction of movement, and speed. Based on this dynamic data, the fish school's behavioral characteristics were determined, including the rate of change in fish density, spatial clustering characteristics, consistency of movement direction, and rate of change in speed. Spatial clustering characteristics refer to the degree of fish aggregation in a specific area, while consistency of movement direction refers to the similarity of movement directions among individuals within the group. Based on these behavioral characteristics, K-means clustering was used to perform cluster analysis, establishing a baseline model for normal fish school behavior. The parameter ranges for this baseline were determined, including the normal range of population density variation, aggregation radius and spatial distribution characteristics, and the average direction of movement and speed range. The Mahalanobis distance formula was then used to calculate the baseline. Calculate the mean Mahalanobis distance between fish school dynamics data and the baseline of normal fish behavior. To determine the degree of deviation between the real-time behavior of the fish school and the baseline of normal behavior, Let be the covariance matrix; if If the behavior exceeds the preset confidence interval for normal fish behavior, the current fish behavior is determined to be abnormal. The system assesses the degree of fish swarm abnormality using a preset anomaly assessment threshold, and labels the type and severity of abnormal behavior. Abnormal behavior types include, but are not limited to, gathering, dispersing, or high-speed movement; severity levels include mild, moderate, and severe. Based on the fish swarm's location coordinates, the system determines the location coordinates and area of ​​the abnormal behavior, and statistically analyzes the population characteristics of the area, including fish number and density. The system stores the fish behavior characteristics of the area where abnormal behavior occurred in a fish behavior pattern database. It then retrieves historical fish behavior characteristics from the database, labels the causes of abnormality, and uses a random forest algorithm to train a model to identify the causes of fish swarm abnormalities. Causes include, but are not limited to, predator threats, environmental disturbances, or human intervention. Based on real-time acquired fish behavior characteristics, the system uses the fish swarm abnormality identification model to determine the cause of the abnormality. Finally, based on the time, area, cause, and severity of the abnormal behavior, the system formulates and implements intervention strategies, including but not limited to adjusting water quality parameters, reducing pollution load, increasing dissolved oxygen concentration, or optimizing water flow velocity. According to the historical dynamic data of the fish school behavior mode database, combined with the mean value of the normal behavior baseline of the fish school, the time point, area range, abnormal type and reason of the fish school abnormal behavior in the future time period are predicted, and the fish school abnormal intervention strategy is implemented in advance; According to the fish school dynamic data obtained by the multi-frequency sonar receiver, combined with the Mahalanobis distance and the fish school behavior prediction, the effect of the fish school abnormal intervention strategy is evaluated, and the fish school abnormal intervention strategy and intervention range are dynamically adjusted.

2. The method of claim 1, wherein, The method comprises: According to the preset detection water area and the possible species distribution range of the target fish, a multi-frequency sonar device is used to configure multiple frequency ranges, and the order of low frequency, medium frequency and high frequency emission signals is set; through the sound wave emission system of the multi-frequency sonar device, the sound wave emission power, beam width and emission angle are adjusted, and the sound wave signals are emitted to the specified water area one by one, the original echo signal of the multi-frequency sound wave is obtained by the multi-frequency sonar receiver, the spatial position information of the sound wave emission is marked, and stored to the fish sound wave monitoring database; through the fish sound wave monitoring database, the echo signal under the standard environmental characteristics and the echo signal under the non-standard environmental characteristics of the same distance, azimuth and depth are obtained, a recurrent neural network is used for model training, and an echo signal correction model is constructed, the environmental characteristics include water temperature, salinity, flow rate and turbidity; according to the real-time obtained original echo signal and environmental characteristics, the echo signal correction model is used to correct the original echo signal to the echo signal under the standard environmental characteristics.

3. The method of claim 1, wherein, The method comprises: According to the corrected echo signal, the echo signal is separated based on time window segmentation and signal arrival time calculation, the echo intensity, scattering mode and arrival time of sound waves of different frequencies are recorded; according to the time delay of the received signal, the sound wave propagation path in the water area is calculated, and the distance, azimuth, depth and position coordinates corresponding to the echo signal are recorded; the obtained echo signal is denoised by a band-pass filter to remove environmental noise and equipment interference in the non-target frequency range; the time sequence signal is converted into a time-frequency domain signal by a short-time Fourier transform algorithm, and the spectral feature data of the echo is extracted; the echo signal intensity data is amplitude normalized by a normalization method, the power amplitude of different frequency signals is unified, and the frequency distribution and time variation mode are recorded to obtain normalized spectral feature data and time sequence distribution information.

4. The method of claim 1, wherein, The frequency response data of various fish species at each frequency is extracted, the characteristic separation index of different fish species at each frequency band is calculated, and the frequency band scaling of the signal feature overlapping area is performed, including: The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula The frequency response gradient of each fish species is calculated from the echo signals according to the formula 5. The method of claim 1, wherein, According to the frequency response feature data of fish in the fish sound wave monitoring database under multi-frequency sound waves, the spatial distribution, population proportion and mixed fish group marking information of the fish school are obtained through signal classification and three-dimensional point cloud clustering analysis, including: Through the fish sound wave monitoring database, the frequency response feature data of different fish species under multi-frequency sound waves is obtained, a recurrent neural network is used for model training, and a fish recognition model is constructed; according to the real-time acquired frequency response feature data of the fish species, the fish recognition model is used to determine the fish label and classification confidence of each echo signal; if the classification confidence is lower than the pre-set confidence threshold, it is marked as invalid signal and removed; the classification results of all echo signals are counted to obtain the overall distribution of fish species in the water area, the proportion and number of different fish populations are determined, and the signal points are mapped into a three-dimensional coordinate system based on the distance, azimuth, depth and position coordinates corresponding to the echo signal to form point cloud data; according to the point cloud data, a K-means clustering algorithm or a DBSCAN clustering algorithm is used for model training, the signal points belonging to the same fish species are input into the clustering algorithm, and the spatial boundary of each fish school is obtained, the size of the fish school in each clustering area, the center position and boundary range of the fish school, wherein the size of the fish school is the number of signal points in the area; the clustering results of different fish species in the space are detected for overlap, to determine whether two or more species of fish overlap in the same space, if an overlapping area is detected, it is marked as a mixed fish school; if there is a mixed fish school, the area is marked separately and the fish species and quantity statistics are output.

6. The method of claim 1, wherein, According to the historical dynamic data of the fish school behavior mode database, combined with the normal behavior baseline mean value, the time point, area range, abnormal type and reason of the fish school abnormal behavior in the future time period are predicted, and the fish school abnormal intervention strategy is implemented in advance, including: The fish school behavior prediction model is constructed by using the long short-term memory network to train the model based on the historical fish school dynamic data obtained from the fish school behavior pattern database, and the fish school dynamic data of the fish school in a preset time period is predicted.

7. The method of claim 1, wherein, The fish school dynamic data obtained from the multi-frequency sonar receiver is combined with the Mahalanobis distance and fish school behavior prediction to evaluate the effect of the fish school abnormal intervention strategy, and the fish school abnormal intervention strategy and intervention range are dynamically adjusted, including: The fish school dynamic data after the fish school abnormal intervention strategy is implemented is obtained by the multi-frequency sonar receiver, and the Mahalanobis distance between the fish school dynamic data and the fish school normal behavior baseline is calculated. If the Mahalanobis distance is lower than the preset fish school normal behavior confidence interval, it is determined that the fish school abnormal intervention strategy is effective, and the fish school abnormal behavior area returns to normal. If the Mahalanobis distance still exceeds the preset fish school normal behavior confidence interval, it is determined that the fish school abnormal intervention strategy is ineffective, and whether the abnormal area expands to the non-intervention area is determined based on the fish school dynamic data, the fish school abnormal intervention strategy and the intervention range are adjusted, and the fish school abnormal behavior area returns to normal. The fish school dynamic data of the area where the fish school abnormal intervention strategy is implemented in advance is obtained, the fish school dynamic data of the fish school in a preset time period is predicted again based on the fish school behavior prediction model, and the Mahalanobis distance between the fish school dynamic data of the fish school in the preset time period and the fish school normal behavior baseline mean is calculated. Whether the fish school abnormal intervention strategy implemented in advance eliminates the future abnormal risk of the abnormal behavior is determined based on the Mahalanobis distance. If the Mahalanobis distance still exceeds the preset fish school normal behavior confidence interval, it is determined that the fish school abnormal intervention strategy implemented in advance is ineffective, and whether the abnormal area expands to the non-intervention area is determined based on the fish school dynamic data of the fish school in the preset time period, and the fish school abnormal intervention strategy and the intervention range are adjusted.

Images