Fish identification system and method based on multi-source data analysis

Abstract

The invention discloses a fish identification system and method based on multi-source data analysis, and relates to the technical field of fish identification, and the method comprises the steps: collecting a local flow field velocity vector generated by fish swimming, and obtaining continuous time sequence hydrodynamic disturbance data; filtering processing is carried out, and a disturbance speed is obtained based on the reference flow velocity; reconstructing a local three-dimensional velocity field by adopting hydrodynamic force constraint; detecting a wake event based on a disturbance energy threshold value, and segmenting the continuous disturbance signal into independent fish passing events; fish wake features are extracted, and fish characterization feature vectors are constructed; and carrying out correlation calculation based on a Pearson's correlation coefficient method, and carrying out fish identification according to a calculation result. Compared with the prior art depending on optical imaging or single-point flow measurement, the method can stably work in natural water areas with turbidity, dark light, poor visibility and the like, and has higher environmental adaptability, richer wake feature expression ability and higher recognition accuracy.

Description

Technical Field

[0001] This invention relates to the field of fish identification technology, specifically a fish identification system and method based on multi-source data analysis. Background Technology

[0002] Fish identification and aquatic ecosystem monitoring are key technical aspects of aquaculture, natural water ecosystem protection, and fisheries resource assessment. Traditional fish identification methods primarily rely on optical imaging equipment, such as underwater cameras, sonar imagers, and infrared vision systems, to classify and identify fish by analyzing their external outlines, grayscale textures, or acoustic imaging features. However, in natural water bodies, factors such as turbidity, uneven lighting, severe obstruction, and frequent bubble interference often lead to unstable optical signals, resulting in missing or distorted image information and a significant decrease in identification accuracy. Furthermore, many fish species are difficult to effectively identify using imaging methods during group activities, nocturnal activities, or low-visibility environments.

[0003] In recent years, hydrodynamic signal analysis has gradually become a new method for biological monitoring. Fish generate specific flow field disturbances during swimming through tail fin wagging and body undulations. Their wake vortex structure, energy distribution, and tail wagging frequency, among other hydrodynamic characteristics, exhibit species-specific differences and can serve as potential biological features without direct observation of the fish's appearance. Existing technologies for hydrodynamic identification largely rely on single-point flow measurement devices or low-dimensional velocity data, classifying fish activity through simple time-series or frequency features. However, the limited dimensionality of velocity data collected from single or sparse measurement points makes it difficult to reconstruct the three-dimensional spatial structure of fish wakes and effectively characterize key hydrodynamic features such as vortex distribution and disturbance propagation paths, thus limiting identification accuracy.

[0004] Furthermore, when fish swim through the sensor array area, their hydrodynamic disturbance signals are often interfered with by multiple sources of noise, such as tidal changes, seabed reflection noise, and mechanical noise. Existing methods generally lack multi-level filtering and noise suppression strategies for complex aquatic environments, making it impossible to reliably acquire disturbance information. In terms of feature construction, existing technologies typically rely on a small number of single-dimensional features, lacking a complete description of wake dynamics. In terms of recognition methods, algorithms based on simple distance metrics or traditional classifiers cannot fully explore the fine correlation between fish wakes and standard samples, resulting in insufficient class discrimination capabilities.

[0005] To address the aforementioned issues, it is necessary to provide a fish identification method and system that can stably acquire hydrodynamic disturbance signals in natural waters, reconstruct three-dimensional velocity fields and extract multi-dimensional wake features, and possess high robustness and high identification accuracy. Summary of the Invention

[0006] The purpose of this invention is to provide a fish identification system and method based on multi-source data analysis to solve the problems raised in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a fish identification method based on multi-source data analysis, specifically including the following steps:

[0008] By collecting the local flow field velocity vector generated by fish swimming through a miniature multi-point water flow velocity sensor array deployed in a preset water area, continuous time series hydrodynamic disturbance data is obtained.

[0009] The hydrodynamic disturbance data is filtered, and the disturbance velocity is obtained based on the reference flow velocity.

[0010] Based on the perturbation velocity, the local three-dimensional velocity field is reconstructed using hydrodynamic constraints;

[0011] Based on the perturbation energy threshold detection of wake events, continuous perturbation signals are segmented into independent fish passage events; fish wake features are extracted from the three-dimensional velocity field based on independent fish passage events, and fish characterization feature vectors are constructed.

[0012] The correlation between the fish characterization feature vector and the pre-recorded standard feature samples is calculated based on the Pearson correlation coefficient method, and fish identification is performed based on the correlation calculation results.

[0013] Furthermore, by using a miniature multi-point water flow velocity sensor array deployed in a predetermined water area to collect the local flow field velocity vectors generated by fish swimming, continuous time-series hydrodynamic disturbance data is obtained, specifically:

[0014] A miniature multi-point water flow velocity sensor array is deployed in the target water area according to a preset grid structure, so that a fixed spatial interval is formed between adjacent sensing units; covering the three-dimensional water area that fish may pass through.

[0015] Each sensing unit uses its built-in triaxial microvelocity probe to simultaneously acquire instantaneous flow velocity information of the water in the x, y, and z directions, forming a local flow field velocity vector. , Characterized as: ;in, , and These represent the instantaneous flow velocities of the water body in the x, y, and z directions collected by the sensing unit at time t, where t represents the time marker. As fish swim across the array, a spatially continuous flow field disturbance sequence is formed. Specifically, each sensing unit in the miniature multi-point water flow velocity sensor array outputs the instantaneous flow velocities of the water body in the x, y, and z directions at a high-frequency sampling rate, and time synchronization is performed to construct a continuous time-series hydrodynamic data matrix D, represented as: Where i represents the matrix row index, N represents the number of sensing units, k represents the matrix column index, and T represents the number of sampling times.

[0016] Optional,

[0017] The water flow velocity sensor array is a miniature MEMS triaxial microflow velocity sensor, which uses at least one of fiber optic communication and acoustic modulation communication for data transmission.

[0018] Furthermore, the hydrodynamic disturbance data is filtered, and the disturbance velocity is obtained based on the reference flow velocity, specifically including:

[0019] A combined filtering strategy is used to process hydrodynamic disturbance data, including: removing low-frequency background noise such as tidal noise based on bandpass filtering; removing instantaneous mechanical noise based on median filtering; and suppressing high-frequency noise caused by seabed topographic reflection based on wavelet thresholding.

[0020] The flow velocities of the water body after the filtering process are expressed as follows: , and ; Forming a local flow field filtered velocity vector Characterized as:

[0021] A baseline flow velocity vector for a fishless state was established using environmental baseline modeling methods. ;

[0022] The disturbance velocity vector is characterized as .

[0023] Furthermore, based on the perturbation velocity, a local three-dimensional velocity field is reconstructed using hydrodynamic constraints, specifically:

[0024] Step S1: Establish a vector field model for the disturbance velocity in the local area, and express the disturbance velocity as a continuous function that varies with space and time; at the same time, collect the disturbance velocity observation values ​​measured by the sensing unit at its spatial location.

[0025] Step S2: Establish a mapping operator from the continuous function to the discrete observation point, so that any disturbance velocity can be projected onto the position of the sensing unit, thereby obtaining the correspondence between the vector field model prediction value and the observed disturbance velocity value;

[0026] Step S3: Reconstruct the local three-dimensional velocity field by constructing an objective function that includes a data fitting term, a divergence-free constraint term, and a smoothing constraint term;

[0027] The data fitting term is used to measure the deviation between the vector field model prediction and the observed perturbation velocity; and weights are applied to different measurement points according to the noise level.

[0028] The divergence-free constraint term is used to force or weaken the velocity field to satisfy the incompressibility condition, so that the reconstruction result conforms to the fluid physical properties.

[0029] The smoothing constraint term introduces a spatial smoothing operator to ensure reasonable continuity of the velocity field in space and suppress local oscillations caused by noise.

[0030] Step S4: Discretize the perturbation velocity in the local area into sampling points in the reconstruction area, construct an algebraic solution system, and use the optimal solution as the local three-dimensional velocity field.

[0031] Furthermore, based on the perturbation energy threshold detection of wake events, continuous perturbation signals are segmented into independent fish passage events; based on these independent fish passage events, fish wake features are extracted from the three-dimensional velocity field to construct a fish characterization feature vector, specifically:

[0032] The continuous disturbance signal is segmented into independent fish passage events using a sliding window-based disturbance energy threshold determination method. Specifically, the local energy is obtained based on the flow velocity of the filtered water body in the x, y, and z directions. , characterized as: Local energy Exceeding the threshold When a fish passage event is detected, the perturbation sequence is segmented, and the segmentation result is denoted as... ,in, , and Let G and G represent the events of the passage of the 1st, ..., g, ..., Gth independent fish, respectively, where G is a positive integer;

[0033] For an isolated fish, fish wake features are extracted from a three-dimensional velocity field via event g. A fish characterization feature vector is constructed based on these extracted wake features, where the fish wake features are denoted as: ;in, , and Let J and J represent the tail trace features of independent fish passing through event g, where J is a positive integer; for a certain fish tail trace feature The characteristic vector representing fish is denoted as: ;in, , and These represent the characteristics of fish tail tracks. The 1st, ..., qth, ..., Qth eigenvalues;

[0034] Furthermore, the correlation between the fish characterization feature vector and the pre-recorded standard feature samples is calculated based on the Pearson correlation coefficient method. Based on the correlation calculation results, fish whose correlation calculation results are greater than the predetermined threshold R0 are identified as the same species of fish.

[0035] Fish whose correlation calculation results are not greater than the pre-set threshold R0 are taken as new samples, and the corresponding fish characterization feature vectors are added to the standard feature samples.

[0036] A fish identification system based on multi-source data analysis, the fish identification system includes a water flow velocity acquisition module, a hydrodynamic data preprocessing module, a three-dimensional velocity field reconstruction module, a wake feature extraction module, and a fish association calculation module;

[0037] The water flow velocity acquisition module is used to acquire the local flow field velocity vector generated by fish swimming by a miniature multi-point water flow velocity sensor array deployed in a preset water area, and obtain continuous time series hydrodynamic disturbance data.

[0038] The hydrodynamic data preprocessing module is used to filter the hydrodynamic disturbance data and obtain the disturbance velocity based on the reference flow velocity.

[0039] The three-dimensional velocity field reconstruction module is used to detect wake events based on the disturbance energy threshold and to segment continuous disturbance signals into independent fish passage events;

[0040] The wake feature extraction module is used to extract fish wake features from a three-dimensional velocity field based on events of independent fish, and to construct a fish characterization feature vector.

[0041] The fish association calculation module is used to calculate the correlation between the fish characterization feature vector and the pre-recorded standard feature samples based on the Pearson correlation coefficient method, and to identify fish based on the correlation calculation results.

[0042] Furthermore, it also includes a sample update module; the sample update module is used to add new fish species to the standard feature samples to achieve data updates.

[0043] Compared with existing technologies, the beneficial effects of this invention are as follows: This application acquires high-dimensional hydrodynamic disturbance information generated by fish swimming by deploying a miniature multi-point triaxial flow velocity sensor array in a preset water area. Combined with multi-level filtering strategies such as bandpass filtering, median filtering, and wavelet threshold denoising, it effectively suppresses tidal background noise, mechanical noise, and high-frequency reflection noise, making the disturbance signal more reliable. Based on this, a three-dimensional velocity field reconstruction method with hydrodynamic constraints is introduced. Data fitting, divergence-free constraints, and smoothing constraints are used to complete sparse measurement point information, obtaining a continuous and realistic three-dimensional wake structure, significantly improving the restoration accuracy of wake dynamic characteristics. Independent fish passage events are automatically identified through disturbance energy threshold detection, and multi-dimensional features such as wake frequency, vorticity distribution, tail swing period, tail swing amplitude, energy spectrum, and disturbance propagation speed are extracted from the reconstructed flow field to construct a feature vector comprehensively characterizing fish hydrodynamic behavior. Then, based on Pearson correlation calculation and similarity with standard samples, high-precision identification of fish species is achieved, and the standard sample library can be automatically expanded according to thresholds to achieve self-learning capability. Compared to existing technologies that rely on optical imaging or single-point flow measurement, this application can work stably in natural waters with turbidity, low light, and poor visibility. It has stronger environmental adaptability, richer wake feature expression capabilities, and higher identification accuracy, making it suitable for long-term ecological monitoring and intelligent fisheries management in large-scale waters. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the fish identification method based on multi-source data analysis of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example: Figure 1 As shown, the present invention provides a technical solution, a fish identification method based on multi-source data analysis, specifically including the following steps:

[0047] By collecting the local flow field velocity vector generated by fish swimming through a miniature multi-point water flow velocity sensor array deployed in a preset water area, continuous time series hydrodynamic disturbance data is obtained.

[0048] The hydrodynamic disturbance data is filtered, and the disturbance velocity is obtained based on the reference flow velocity.

[0049] Based on the perturbation velocity, the local three-dimensional velocity field is reconstructed using hydrodynamic constraints;

[0050] Based on the perturbation energy threshold detection of wake events, continuous perturbation signals are segmented into independent fish passage events; fish wake features are extracted from the three-dimensional velocity field based on independent fish passage events, and fish characterization feature vectors are constructed.

[0051] The correlation between the fish characterization feature vector and the pre-recorded standard feature samples is calculated based on the Pearson correlation coefficient method, and fish identification is performed based on the correlation calculation results.

[0052] Furthermore, by using a miniature multi-point water flow velocity sensor array deployed in a predetermined water area to collect the local flow field velocity vectors generated by fish swimming, continuous time-series hydrodynamic disturbance data is obtained, specifically:

[0053] A miniature multi-point water flow velocity sensor array is deployed in the target water area according to a preset grid structure, so that a fixed spatial interval is formed between adjacent sensing units; covering the three-dimensional water area that fish may pass through.

[0054] Each sensing unit uses its built-in triaxial microvelocity probe to simultaneously acquire instantaneous flow velocity information of the water in the x, y, and z directions, forming a local flow field velocity vector. , Characterized as: ;in, , and These represent the instantaneous flow velocities of the water body in the x, y, and z directions collected by the sensing unit at time t, where t represents the time marker. As fish swim across the array, a spatially continuous flow field disturbance sequence is formed. Specifically, each sensing unit in the miniature multi-point water flow velocity sensor array outputs the instantaneous flow velocities of the water body in the x, y, and z directions at a high-frequency sampling rate, and time synchronization is performed to construct a continuous time-series hydrodynamic data matrix D, represented as: Where i represents the matrix row index, N represents the number of sensing units, k represents the matrix column index, and T represents the number of sampling times.

[0055] Optional,

[0056] The water flow velocity sensor array is a miniature MEMS triaxial microflow velocity sensor, which uses at least one of fiber optic communication and acoustic modulation communication for data transmission.

[0057] Furthermore, the hydrodynamic disturbance data is filtered, and the disturbance velocity is obtained based on the reference flow velocity, specifically including:

[0058] A combined filtering strategy is used to process hydrodynamic disturbance data, including: removing low-frequency background noise such as tidal noise based on bandpass filtering; removing instantaneous mechanical noise based on median filtering; and suppressing high-frequency noise caused by seabed topographic reflection based on wavelet thresholding.

[0059] The flow velocities of the water body after the filtering process are expressed as follows: , and ; Forming a local flow field filtered velocity vector Characterized as:

[0060] A baseline flow velocity vector for a fishless state was established using environmental baseline modeling methods. ;

[0061] The disturbance velocity vector is characterized as .

[0062] Furthermore, based on the perturbation velocity, a local three-dimensional velocity field is reconstructed using hydrodynamic constraints, specifically:

[0063] Step S1: Establish a vector field model for the disturbance velocity in the local area, and express the disturbance velocity as a continuous function that varies with space and time; at the same time, collect the disturbance velocity observation values ​​measured by the sensing unit at its spatial location.

[0064] Step S2: Establish a mapping operator from the continuous function to the discrete observation point, so that any disturbance velocity can be projected onto the position of the sensing unit, thereby obtaining the correspondence between the vector field model prediction value and the observed disturbance velocity value;

[0065] Step S3: Reconstruct the local three-dimensional velocity field by constructing an objective function that includes a data fitting term, a divergence-free constraint term, and a smoothing constraint term;

[0066] The data fitting term is used to measure the deviation between the vector field model prediction and the observed perturbation velocity; and weights are applied to different measurement points according to the noise level.

[0067] The divergence-free constraint term is used to force or weaken the velocity field to satisfy the incompressibility condition, so that the reconstruction result conforms to the fluid physical properties.

[0068] The smoothing constraint term introduces a spatial smoothing operator to ensure reasonable continuity of the velocity field in space and suppress local oscillations caused by noise.

[0069] Step S4: Discretize the perturbation velocity in the local area into sampling points in the reconstruction area, construct an algebraic solution system, and use the optimal solution as the local three-dimensional velocity field.

[0070] It should be noted that since the filtered disturbance velocities mostly come from single points or sparse measurement points (sensing units), they cannot cover the complete flow field of a local area. Reconstruction can fill in the missing spatial information and form a continuous three-dimensional flow field distribution. Secondly, the flow field disturbances caused by targets such as fish are three-dimensional. Discrete velocity data alone cannot reflect their spatial propagation laws, intensity distribution, and directional changes. A three-dimensional field can restore the true form of the disturbance. Therefore, it is necessary to reconstruct the local three-dimensional velocity field.

[0071] In this embodiment, the local three-dimensional velocity field is reconstructed by constructing an objective function that includes a data fitting term, a divergence-free constraint term, and a smoothing constraint term. The local three-dimensional velocity field to be reconstructed is denoted as... ; , and Let x represent the flow velocity of the three-dimensional velocity field at time t in the three directions at point x; denote the measured perturbation velocity at the position of the sensing unit. The observed value is ;

[0072] Define measurement operator Mapping the continuous field to the set of measurement points: ;

[0073] The objective function is:

[0074] ;

[0075] in, Indicates the data fitting term; Indicates a term without divergence constraints; Indicates the smoothing constraint term; Represents the residual;

[0076] , and Indicates the regularization weight; Represents the smoothing operator; Indicates the weight of the measurement point; Represents the gradient operator;

[0077] In this application, the smoothing operator is selected as the Laplace operator, or a higher-order derivative can also be selected, which serves to control noise;

[0078] in, A simplified linear version of the incompressible viscous Navier–Stokes:

[0079] U represents a three-dimensional velocity field, the same as... p represents a pressure scalar field; This indicates the water density in the area; Indicates the dynamic viscosity coefficient; Represents a volume force vector; Represents the time partial derivative;

[0080] It should be noted that the fluid is an approximately incompressible flow, and the forced flow is... By adding a momentum residual term to the objective function, the field not only becomes divergent but also satisfies the approximate equilibrium of the fluid motion equations.

[0081] The solution process can be achieved through interpolation and Helmholtz-Hodge projection; where,

[0082] The initial field is obtained by interpolating on the grid using spatial interpolation (RBF / Kriging / kernel regression). For the initial field Perform Helmholtz–Hodge decomposition; in local regions Internal scalar potential Solving the Poisson equation, we can characterize it as follows: Apply boundary conditions ( or Then construct a divergence-free field: The iterative process eventually yields the local three-dimensional velocity field. ;

[0083] Among them, considering momentum consistency, it is possible to Based on this, perform a least squares correction (linearizing the Navier-Stokes minimum residual).

[0084] Furthermore, based on the perturbation energy threshold detection of wake events, continuous perturbation signals are segmented into independent fish passage events; based on these independent fish passage events, fish wake features are extracted from the three-dimensional velocity field to construct a fish characterization feature vector, specifically:

[0085] The continuous disturbance signal is segmented into independent fish passage events using a sliding window-based disturbance energy threshold determination method. Specifically, the local energy is obtained based on the flow velocity of the filtered water body in the x, y, and z directions. , characterized as: Local energy Exceeding the threshold When a fish passage event is detected, the perturbation sequence is segmented, and the segmentation result is denoted as... ,in, , and Let G and G represent the events of the passage of the 1st, ..., g, ..., Gth independent fish, respectively, where G is a positive integer;

[0086] For an isolated fish, fish wake features are extracted from a three-dimensional velocity field via event g. A fish characterization feature vector is constructed based on these extracted wake features, where the fish wake features are denoted as: ;in, , and Let J and J represent the tail trace features of independent fish passing through event g, where J is a positive integer; for a certain fish tail trace feature The characteristic vector representing fish is denoted as: ;in, , and These represent the characteristics of fish tail tracks. The 1st, ..., qth, ..., Qth eigenvalues;

[0087] The wake features include wake frequency, etc.; among them, vorticity distribution features, tail swing period features, tail swing amplitude features, etc. can also be used to form a fish characterization feature vector for correlation calculation.

[0088] Among them, the wake frequency is determined by the perturbation velocity of a certain point of interest or node. Perform short-time Fourier transform to find the main frequency peaks get.

[0089] ;

[0090] in, Indicates a point of interest The perturbation velocity at the location; finding the main frequency peak through short-time Fourier transform is a classic frequency extraction technique based on signal processing, which is widely used in fluid dynamics wake analysis and has clear physical significance; by reasonably selecting STFT parameters and verification strategies, reliable wake frequency estimates can be obtained.

[0091] Eddy power distribution is used to reflect the magnitude of the angular velocity of fluid micro-elements;

[0092] Eddy power is distributed on the sensor array (discrete sensing unit) and estimated using finite difference;

[0093] In this embodiment, the vortex distribution statistics during zebrafish cruising were obtained as follows: maximum positive vortex: 0.41s⁻¹, maximum negative vortex: -0.82s⁻¹, RMS vortex: 0.11s⁻¹, average vortex: -0.008s⁻¹, positive vortex circulation: 8.1236m² / s, negative vortex circulation: -10.8063m² / s;

[0094] Eddy volume analysis of the escape response of medaka fish: maximum positive eddy volume: 4.92 s⁻¹, maximum negative eddy volume: -3.86 s⁻¹, RMS eddy volume: 0.49 s⁻¹, average eddy volume: 0.056 s⁻¹, positive eddy volume circulation: 47.0860 m² / s, negative eddy volume circulation: -22.3497 m² / s.

[0095] The tail wagging periodicity indicates the time required for a fish to complete one full tail wagging motion; it reflects the rhythmic frequency of the tail fin's movement and directly affects propulsion efficiency.

[0096] The tail sway amplitude characteristic indicates the maximum distance the tail fin deviates from the body midline during the swaying process; it reflects the intensity of the tail sway motion.

[0097] It should be noted that if fish characterization feature vectors are composed of vorticity distribution characteristics, tail swing period characteristics, and tail swing amplitude characteristics, at least three different feature combinations must be ensured.

[0098] Furthermore, the correlation between the fish characterization feature vector and the pre-recorded standard feature samples is calculated based on the Pearson correlation coefficient method. Based on the correlation calculation results, fish whose correlation calculation results are greater than the predetermined threshold R0 are identified as the same species of fish.

[0099] Fish whose correlation calculation results are not greater than the pre-set threshold R0 are taken as new samples, and the corresponding fish characterization feature vectors are added to the standard feature samples.

[0100] When calculating the correlation between the fish characterization feature vector and the pre-recorded standard feature sample based on the Pearson correlation coefficient method, it should be noted that the pre-recorded standard feature sample should be a change sequence. The correlation between the change sequences of the same tail feature is calculated to determine whether they belong to the same fish.

[0101] In this application, the main frequency peak is selected for the wake frequency. The correlation between the corresponding change sequences is calculated to make a judgment.

[0102] The Pearson correlation coefficient is a classic statistical method for measuring the degree of linear correlation between two continuous variables. Its core purpose is to quantify the strength and direction of the linear association between variables, with a value range of [-1, 1]. This application uses the Pearson correlation coefficient to calculate the similarity between independent fish passage events and standard feature samples, thereby identifying fish.

[0103] Since the Pearson correlation coefficient method is an existing technology, it will not be explained in detail here.

[0104] The pre-set threshold R0 can be preset based on the average value and standard deviation of the correlation coefficients between several fish characterization feature vectors and pre-recorded standard feature samples in history. The larger the pre-set threshold R0, the higher the accuracy requirement for fish identification, and therefore the higher the monitoring accuracy requirement for the sensing unit. Conversely, the smaller the pre-set threshold R0, the lower the accuracy requirement for fish identification, but the monitoring accuracy requirement for the sensing unit is not so high. The specific setting can be made according to the actual situation, and no restrictions are imposed here.

[0105] A fish identification system based on multi-source data analysis includes a water flow velocity acquisition module, a hydrodynamic data preprocessing module, a three-dimensional velocity field reconstruction module, a wake feature extraction module, and a fish association calculation module.

[0106] The water flow velocity acquisition module is used to acquire the local flow field velocity vector generated by fish swimming by a miniature multi-point water flow velocity sensor array deployed in a preset water area, and obtain continuous time series hydrodynamic disturbance data;

[0107] The hydrodynamic data preprocessing module is used to filter the hydrodynamic disturbance data and obtain the disturbance velocity based on the reference flow velocity;

[0108] The three-dimensional velocity field reconstruction module is used to detect wake events based on the perturbation energy threshold, and to segment continuous perturbation signals into independent fish passage events;

[0109] The wake feature extraction module is used to extract fish wake features from a three-dimensional velocity field based on events of independent fish, and to construct fish characterization feature vectors;

[0110] The fish association calculation module is used to calculate the correlation between the fish characterization feature vector and the pre-recorded standard feature samples based on the Pearson correlation coefficient method, and to identify fish based on the correlation calculation results.

[0111] Furthermore, it also includes a sample update module; the sample update module is used to add new fish species to the standard feature samples to achieve data updates.

[0112] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A fish identification method based on multi-source data analysis, characterized in that: Specifically, the steps include the following: By collecting the local flow field velocity vector generated by fish swimming through a miniature multi-point water flow velocity sensor array deployed in a preset water area, continuous time series hydrodynamic disturbance data is obtained. The hydrodynamic disturbance data is filtered, and the disturbance velocity is obtained based on the reference flow velocity. Based on the perturbation velocity, the local three-dimensional velocity field is reconstructed using hydrodynamic constraints; Based on the detection of wake events using the perturbation energy threshold, continuous perturbation signals are segmented into independent fish passage events; Based on the extraction of fish wake features from a three-dimensional velocity field through events by independent fish, a fish characterization feature vector is constructed. The correlation between the fish characterization feature vector and the pre-recorded standard feature samples is calculated based on the Pearson correlation coefficient method, and fish identification is performed based on the correlation calculation results.

2. The fish identification method based on multi-source data analysis according to claim 1, characterized in that: By collecting local flow field velocity vectors generated by fish swimming through a miniature multi-point water flow velocity sensor array deployed in a predetermined water area, continuous time-series hydrodynamic disturbance data is obtained, specifically: A miniature multi-point water flow velocity sensor array is arranged in the target water area according to a preset grid structure, so that a fixed spatial interval is formed between adjacent sensing units. Each sensing unit uses its built-in triaxial microvelocity probe to simultaneously acquire instantaneous flow velocity information of the water in the x, y, and z directions, forming a local flow field velocity vector. , Characterized as: ;in, , and These represent the instantaneous flow velocities of the water body in the x, y, and z directions collected by the sensing unit at time t, where t represents the time marker. As fish swim across the array, a spatially continuous flow field disturbance sequence is formed. Specifically, each sensing unit in the miniature multi-point water flow velocity sensor array outputs the instantaneous flow velocities of the water body in the x, y, and z directions at a high-frequency sampling rate, and time synchronization is performed to construct a continuous time-series hydrodynamic data matrix D, represented as: Where i represents the matrix row index, N represents the number of sensing units, k represents the matrix column index, and T represents the number of sampling times.

3. The fish identification method based on multi-source data analysis according to claim 2, characterized in that: The hydrodynamic disturbance data is filtered, and the disturbance velocity is obtained based on the reference flow velocity, specifically as follows: A combined filtering strategy is employed to process hydrodynamic disturbance data, including: removing low-frequency background noise using a bandpass filter; removing instantaneous mechanical noise using a median filter; and suppressing high-frequency noise caused by seabed topographic reflections using wavelet thresholding. The flow velocities of the water body after the filtering process are expressed as follows: , and ; Forming a local flow field filtered velocity vector Characterized as: A baseline flow velocity vector for a fishless state was established using environmental baseline modeling methods. ; The disturbance velocity vector is characterized as .

4. The fish identification method based on multi-source data analysis according to claim 3, characterized in that: Based on the perturbation velocity, the local three-dimensional velocity field is reconstructed using hydrodynamic constraints, specifically including the following steps: Step S1: Establish a vector field model for the disturbance velocity in the local area, and express the disturbance velocity as a continuous function that varies with space and time; at the same time, collect the disturbance velocity observation values ​​measured by the sensing unit at its spatial location. Step S2: Establish a mapping operator from the continuous function to the discrete observation point, so that any disturbance velocity can be projected onto the position of the sensing unit, thereby obtaining the correspondence between the vector field model prediction value and the observed disturbance velocity value; Step S3: Reconstruct the local three-dimensional velocity field by constructing an objective function that includes a data fitting term, a divergence-free constraint term, and a smoothing constraint term; The data fitting term is used to measure the deviation between the vector field model prediction and the observed perturbation velocity; the divergence-free constraint term is used to ensure that the reconstruction result conforms to the fluid physics characteristics; the smoothing constraint term suppresses local oscillations caused by noise by introducing a spatial smoothing operator. Step S4: Discretize the perturbation velocity in the local area into sampling points in the reconstruction area, construct an algebraic solution system, and use the optimal solution as the local three-dimensional velocity field.

5. The fish identification method based on multi-source data analysis according to claim 4, characterized in that: Based on the perturbation energy threshold detection of wake events, continuous perturbation signals are segmented into independent fish passage events; fish wake features are extracted from the three-dimensional velocity field based on independent fish passage events, and a fish representation feature vector is constructed, specifically as follows: The continuous disturbance signal is segmented into independent fish passage events using a sliding window-based disturbance energy threshold determination method. Specifically, the local energy is obtained based on the flow velocity of the filtered water body in the x, y, and z directions. , characterized as: Local energy Exceeding the threshold When a fish passage event is detected, the perturbation sequence is segmented, and the segmentation result is denoted as... ,in, , and Let G and G represent the events of the passage of the 1st, ..., g, ..., Gth independent fish, respectively, where G is a positive integer; For an isolated fish, fish wake features are extracted from a three-dimensional velocity field via event g. A fish characterization feature vector is constructed based on these extracted wake features, where the fish wake features are denoted as: ;in, , and Let J and J represent the tail trace features of independent fish passing through event g, where J is a positive integer; for a certain fish tail trace feature The characteristic vector representing fish is denoted as: ;in, , and These represent the characteristics of fish tail tracks. The 1st, ..., qth, ..., Qth eigenvalues.

6. The fish identification method based on multi-source data analysis according to claim 5, characterized in that: The correlation between the fish characterization feature vector and pre-recorded standard feature samples is calculated based on the Pearson correlation coefficient method. Fish identification is then performed based on the correlation calculation results. Specifically: The correlation between the fish characterization feature vector and the pre-recorded standard feature sample is calculated based on the Pearson correlation coefficient method. Based on the correlation calculation results, fish whose correlation calculation results are greater than the predetermined threshold R0 are identified as the same species. Fish whose correlation calculation results are not greater than the pre-set threshold R0 are taken as new samples, and the corresponding fish characterization feature vectors are added to the standard feature samples.

7. The fish identification method based on multi-source data analysis according to claim 2, characterized in that: Also includes: The water flow velocity sensor array is a miniature MEMS triaxial microflow velocity sensor, which uses at least one of fiber optic communication and acoustic modulation communication for data transmission.

8. A fish identification system based on multi-source data analysis, employing the fish identification method based on multi-source data analysis as described in any one of claims 1-7, characterized in that: The fish identification system includes a water flow velocity acquisition module, a hydrodynamic data preprocessing module, a three-dimensional velocity field reconstruction module, a wake feature extraction module, and a fish association calculation module; The water flow velocity acquisition module is used to acquire the local flow field velocity vector generated by fish swimming by a miniature multi-point water flow velocity sensor array deployed in a preset water area, and obtain continuous time series hydrodynamic disturbance data. The hydrodynamic data preprocessing module is used to filter the hydrodynamic disturbance data and obtain the disturbance velocity based on the reference flow velocity. The three-dimensional velocity field reconstruction module is used to detect wake events based on the disturbance energy threshold and to segment continuous disturbance signals into independent fish passage events; The wake feature extraction module is used to extract fish wake features from a three-dimensional velocity field based on events of independent fish, and to construct a fish characterization feature vector. The fish association calculation module is used to calculate the correlation between the fish characterization feature vector and the pre-recorded standard feature samples based on the Pearson correlation coefficient method, and to identify fish based on the correlation calculation results.

9. A fish identification system based on multi-source data analysis according to claim 8, characterized in that: It also includes a sample update module; the sample update module is used to add new fish species to the standard feature samples to achieve data updates.

Images