Three-dimensional multi-dimensional ocean swimming animal monitoring method based on artificial intelligence

Abstract

The invention provides a three-dimensional multi-dimensional oceanic nekton monitoring method based on artificial intelligence, and relates to the technical field of oceanic nekton monitoring, and the method comprises the steps: deploying a temperature-salinity-depth section plotter and an acoustic Doppler flow velocity section plotter, collecting data, configuring an environment parameter data set, obtaining a three-dimensional environment parameter data set, and obtaining a three-dimensional environment parameter data set; and a multi-dimensional environment parameter database is generated. Various ocean fishery resource investigation means are fused, an advanced artificial intelligence method is applied, three-dimensional multi-dimensional dynamic monitoring of the ocean swimming animals is achieved, on one hand, the number composition and space-time distribution of the ocean swimming animals, especially the vertical distribution characteristics can be monitored, the population dynamics of the ocean swimming animals and the relation between the population dynamics and environmental factors can be revealed, and on the other hand, the population dynamics of the ocean swimming animals can be monitored. By monitoring the swimming animals, basic data can be provided for population resource evaluation, and the method serves formulation of fishing quota of ocean economic swimming animals, delimitation of ocean protection areas and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ocean swimming animal monitoring. Specifically, it relates to a three-dimensional multi-dimensional ocean swimming animal monitoring method based on artificial intelligence. BACKGROUND

[0002] As an important part of global marine ecosystems, the ocean fishery ecosystem has unique fishery biota characteristics due to its wide spatial scale and strong environmental heterogeneity. Swimming animals, as consumers in the ocean food chain, play an important role in maintaining the material cycle and energy flow of the ecosystem. They are also an important part of human exploitation of ocean fishery resources. The distribution and behavior of ocean swimming animals are influenced by ocean currents, water temperature, and prey environment, and their dynamic changes are important indicators of ocean ecosystem monitoring.

[0003] Currently, ocean swimming animal monitoring has multiple limitations in terms of technology and ecological factors, which restricts the accurate understanding of its dynamic changes. Traditional monitoring methods such as trawl sampling have certain sampling bias. The distribution of swimming animals is significantly influenced by ocean currents and climate change, and traditional fixed-point monitoring cannot reflect the rapid changes in their spatial and temporal dynamics. In addition, the distribution of swimming animals is highly correlated with factors such as plankton, topography, and water temperature, but existing monitoring focuses on a single species and lacks systematic observation of the "environment-prey-biological" linkage mechanism. SUMMARY

[0004] The present application aims to address the shortcomings of the prior art by providing a three-dimensional multi-dimensional ocean swimming animal monitoring method based on artificial intelligence. This method integrates various ocean fishery resource survey methods and uses advanced artificial intelligence methods to achieve three-dimensional multi-dimensional dynamic monitoring of ocean swimming animals. On the one hand, it can monitor the number composition and spatial and temporal distribution of ocean swimming animals, especially the vertical distribution characteristics, revealing the population dynamics and their relationship with environmental factors. On the other hand, through monitoring of swimming animals, it can provide basic data for population resource assessment, serving the formulation of ocean economic swimming animal catch quotas and the delineation of marine protected areas.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A three-dimensional multi-dimensional ocean swimming animal monitoring method based on artificial intelligence, comprising the following steps: Step S100, deploy a temperature-salinity-depth profiler and an acoustic Doppler current profiler, collect data, and configure an environmental parameter data set to obtain a three-dimensional environmental parameter data set and generate a multi-dimensional environmental parameter database.

[0006] Step S200, deploy the multi-dimensional swimming animal collection device, and collect data to obtain the upper swimming animal species and quantity distribution data set and the multi-dimensional acoustic inversion data set, and configure the multi-dimensional swimming animal composition and distribution data structure.

[0007] Step S300, based on the three-dimensional environmental parameter data set and the multi-dimensional swimming animal composition and distribution data structure, data fusion is performed to obtain a fusion input data set, and a random forest regression model and a deep neural network model are configured and trained, an environmental factor and swimming animal distribution multivariate mapping relationship structure is configured, a random forest regression mapping structure and a deep neural network mapping structure are obtained, and the most suitable habitat environment parameter combination and model performance index are output.

[0008] Step S400, based on the random forest regression mapping structure and the deep neural network mapping structure, generate the swimming animal community structure spatiotemporal distribution map, calculate the Shannon diversity index and the Simpson diversity index, and perform spatial statistical analysis.

[0009] Step S500, deploy a remote sensing satellite platform and an unmanned aerial vehicle patrol platform, collect macro-scale environmental parameter data and pre-process, configure an extended area prediction input data set, obtain the extended area prediction input data set, predict and calculate the extended area swimming animal spatial distribution, and identify the high probability distribution area and spatiotemporal dynamic analysis.

[0010] As a preferred scheme of the present application, the step S100 specifically comprises: Step S100.1, deploy a temperature-salinity-depth profiler and an acoustic Doppler current profiler, and collect data.

[0011] In the target ocean area, preset the investigation station, and fix and install the temperature-salinity-depth profiler CTD and the acoustic Doppler current profiler ADCP on the deck hoisting system and the ship body bottom mounting rack of the oceanographic research ship respectively.

[0012] After the oceanographic research ship sails to the investigation station, the deck hoisting system of the oceanographic research ship is started, the temperature-salinity-depth profiler is controlled to be vertically lowered to the set detection depth at a stable speed, and in the lowering process, the temperature parameter, the salinity parameter and the depth parameter of the water body are continuously collected through the integrated temperature sensor, the salinity sensor and the pressure sensor, and the detection depth is set according to the sea area investigation task requirement.

[0013] The collected environmental parameters are transmitted in real time to the data collection and processing module on the oceanographic research ship through the communication cable, and are recorded and stored according to the corresponding depth coordinates and time labels, and a continuous vertical temperature-salinity-depth profile data set is constructed.

[0014] The acoustic Doppler current profiler is continuously working during the voyage of the oceanographic survey ship, adopts the preset frequency multi-beam acoustic wave signal to emit downward, the frequency shift is received through the echo, the movement speed of the suspended particles in the water body is measured according to the Doppler effect, and the flow velocity and the flow direction angle in different depth sections are obtained.

[0015] The acoustic Doppler current profiler divides the whole water column into multiple equal-interval depth layers, and outputs the corresponding velocity vector of each layer, including: velocity size, direction angle and vertical depth coordinate.

[0016] Step S100.2, based on the data collected by the temperature-salinity-depth profiler and the acoustic Doppler current profiler, the configuration environment parameter data set is obtained, the three-dimensional environment parameter data set is obtained, and the multi-dimensional environment parameter database is generated.

[0017] The temperature parameter, salinity parameter and depth parameter collected by the temperature-salinity-depth profiler are converted into a unified space reference system and time synchronization processing with the flow velocity, flow direction angle and depth parameter collected by the acoustic Doppler current profiler, and the three-dimensional environment parameter data set covering temperature, salinity, depth, flow velocity and flow direction five types of physical factors is constructed by structured coding in the data processing module according to the unified format.

[0018] As a preferred scheme of the application, the step S200 is specifically: Step S200.1, deploying a swimming animal multi-dimensional acquisition device, the swimming animal multi-dimensional acquisition device includes: a surface environment DNA sampling device, a middle-upper layer trawl net system and a Simrad EK80 type wideband fishery acoustic detection system.

[0019] Based on the surface environment DNA sampling device, the middle-upper layer trawl net system and the Simrad EK80 type wideband fishery acoustic detection system, the middle-upper layer swimming animal species and quantity distribution data set and the multi-dimensional acoustic inversion data set are obtained.

[0020] In the target ocean area, the swimming animal multi-dimensional acquisition device is deployed according to the investigation station, the biological sampling range of the surface layer, the middle-upper layer and the multi-depth layer is set at each sampling point, and a biological sampling grid covering the vertical and horizontal dimensions is formed.

[0021] The surface environment DNA sampling device collects 3 liters to 5 liters of seawater samples at each surface sampling point, and immediately filters the samples on site through a filter membrane with a pore size of 0.45 microns to obtain a filter membrane sample rich in swimming animal genetic material.

[0022] The filter membrane sample is transported to the laboratory through the cold chain, DNA is extracted using a commercial environmental DNA extraction kit, and the extracted DNA fragments are amplified and sequenced using the Illumina NovaSeq high-throughput sequencing platform.

[0023] The sequencing results are compared to the reference database NCBINT database or FISH-BOL database to obtain the species classification unit and corresponding sequence depth of the swimming animals, and a mid-upper swimming animal species composition data set containing the species name, classification level, relative abundance, and sampling point identifier is generated.

[0024] The mid-layer trawl is used for horizontal trawling within the specified water depth range. The four-piece mid-layer trawl has a main dimension of 916 meshes x 400 mm, a net opening circumference of 366.4 m, a net body length of 120 m, and a single capsule structure. The net opening part uses large mesh, and the net body part uses machine-knitted mesh. The double-blade net plate uses a single hand cable connection method, and the trawling speed is set to 3.5-4.5 nmile / h, with a trawling time of 1 hour.

[0025] The swimming animal samples caught by the trawl device are classified, counted, and weighed, and the corresponding tail number and mass data of each species are recorded, along with the trawl depth, sampling latitude and longitude coordinates, and sampling time, forming a mid-upper swimming animal species and quantity distribution data set with three-dimensional labels of survey point, water depth layer, and species.

[0026] The Simrad EK80 type wideband fisheries acoustic detection system continuously emits 38 kHz, 70 kHz, 120 kHz, and 200 kHz frequency band wideband acoustic signals during the voyage of the oceanographic research ship, and records the vertical echo scattering intensity.

[0027] The Simrad EK80 type wideband fisheries acoustic detection system performs data acquisition at a pulse emission frequency of 5 times per second, divides the entire water column into multiple depth sections with a depth resolution of 0.5 meters, and performs background noise elimination and target intensity inversion processing on the original acoustic echo data. Combined with system calibration data and the acoustic scattering characteristic parameters of swimming animal target species, the number density value of swimming animals per unit volume in each depth layer is obtained by inversion, forming a multi-dimensional acoustic inversion data set including depth coordinates, latitude and longitude coordinates, number density, and acoustic frequency band identifier.

[0028] Step S200.2, configure multi-dimensional swimming animal composition and distribution data structure.

[0029] Based on the obtained mid-upper layer swimming animal species and quantity distribution data set and multi-dimensional acoustic inversion data set, format uniform processing, spatial coordinate, depth label standardization, time synchronization and structured coding processing are performed, and a multi-dimensional swimming animal composition and distribution data structure facing the survey station is constructed, which contains species name, quantity density, classification level, spatial position and sampling method label.

[0030] As a preferred scheme of the present application, the step S300 specifically comprises: Step S300.1, based on the three-dimensional environmental parameter data set and the multi-dimensional swimming animal composition and distribution data structure, data fusion is performed to obtain a fusion input data set, and a random forest regression model and a deep neural network model are configured.

[0031] Step S300.2, based on the fusion input data set, the random forest regression model and the deep neural network model are trained.

[0032] The fusion input data set is used to train the random forest regression model and the deep neural network model respectively, and the five-fold cross-validation method is used to verify the prediction performance of the random forest regression model and the deep neural network model.

[0033] The fusion input data set is randomly divided into five subsets, four of which are used for training the random forest regression model and the deep neural network model, and one of which is used for verifying the random forest regression model and the deep neural network model, the training and verification are repeated five times, and the model prediction error of each verification is recorded, when the trained random forest regression model and the deep neural network model show stable prediction performance on unseen data, the training is stopped.

[0034] Step S300.3, based on the trained random forest regression model and the deep neural network model, a multivariate mapping relationship structure of environmental factors and swimming animal distribution is configured, a random forest regression mapping structure and a deep neural network mapping structure are obtained, and the most suitable habitat environmental parameter combination and model performance index are output.

[0035] The trained random forest regression model and the deep neural network model are respectively parameterized to obtain the random forest regression mapping structure and the deep neural network mapping structure.

[0036] The random forest regression mapping structure includes: input environmental factor dimension, decision tree number, maximum tree depth, feature weight parameter and prediction error boundary.

[0037] The deep neural network mapping structure includes: input feature dimension, hidden layer node number, weight matrix, bias vector, activation function type and output species existence probability threshold.

[0038] The optimal habitat environment parameter combination and model performance index are recorded as output files of feature selection and prediction performance evaluation.

[0039] As a preferred scheme of the present application, the step S400 specifically comprises: Step S400.1, generating a swimming animal community structure spatiotemporal distribution map based on a random forest regression mapping structure and a deep neural network mapping structure.

[0040] The environment parameter combination in the spatial analysis unit is respectively input into the random forest regression mapping structure and the deep neural network mapping structure, and the resource density value of the swimming animal and the species existence probability value of the swimming animal in each spatial analysis unit are predicted and obtained.

[0041] The resource density value and the species existence probability value predicted and obtained in each spatial analysis unit are subjected to spatial interpolation and smoothing processing, and a swimming animal population number density spatial distribution map and a species existence probability spatial distribution map in the target ocean sea area are output.

[0042] Step S400.2, calculating a Shannon diversity index and a Simpson diversity index based on the swimming animal community structure spatiotemporal distribution map, obtaining a Shannon diversity index spatial distribution map and a Simpson diversity index spatial distribution map, and performing spatial statistical analysis.

[0043] The spatiotemporal distribution characteristics and the change trend of the high diversity region and the low diversity region are subjected to spatial statistical analysis, the spatial diffusion characteristics of the high diversity region and the low diversity region changing with time are analyzed, and the main environmental factors causing the change of the high diversity region and the low diversity region are analyzed based on temperature parameters, salinity parameters, depth parameters, flow velocity parameters and flow direction parameters.

[0044] An analysis result file including a Shannon diversity index and a Simpson diversity index spatiotemporal distribution map, a spatial statistical analysis data table and a trend change graph is generated, and the spatial distribution characteristics, the spatiotemporal dynamic change trend, the community stability and the diversity condition of the ecosystem of the swimming animal community structure in the target ocean sea area are comprehensively characterized.

[0045] As a preferred scheme of the present application, the step S500 specifically comprises: Step S500.1, deploying a remote sensing satellite platform and an unmanned aerial vehicle patrol platform, and collecting macro-scale environment parameter data.

[0046] In the extended observation area outside the target ocean sea area, a remote sensing satellite platform and an unmanned aerial vehicle patrol platform are deployed, and the remote sensing satellite platform comprises a MODIS remote sensing satellite, a Sentinel-3 remote sensing satellite and a VIIRS remote sensing satellite.

[0047] The unmanned aerial vehicle patrol platform comprises a multi-rotor unmanned aerial vehicle and a fixed-wing unmanned aerial vehicle.

[0048] The remote sensing satellite platform performs remote sensing observation on the sea surface temperature parameter, the sea surface chlorophyll concentration parameter and the sea surface wind speed parameter in the extended observation area with a spatial resolution of 1 kilometer x 1 kilometer and a sampling period of 12 hours, to obtain macro-scale environmental parameter data.

[0049] The unmanned aerial vehicle patrol platform patrols the sea area along the grid patrol route planned in advance at a patrol flight height of 300 meters in the extended observation area, to obtain a sea surface visible light image, an infrared temperature spectrum, a local wind speed parameter and a local wind direction parameter with a spatial resolution of not less than 10 meters, and to supplement the local environmental parameter data with high spatial precision.

[0050] Step S500.2, data preprocessing is performed based on the macro-scale environmental parameter data, and an extended area prediction input data set is configured, to obtain an extended area prediction input data set, The macro-scale environmental parameter data obtained by the remote sensing satellite platform and the local environmental parameter data obtained by the unmanned aerial vehicle patrol platform are subjected to spatial coordinate re-projection, time synchronization resampling and data format unification processing.

[0051] The sea surface temperature parameter, the sea surface chlorophyll concentration parameter, the sea surface wind speed parameter, the local wind speed parameter and the local wind direction parameter subjected to the spatial coordinate re-projection processing are subjected to spatial interpolation and edge splicing processing with the three-dimensional environmental parameter data set in the original investigation area, and an extended area prediction input data set is obtained.

[0052] Step S500.3, prediction calculation is performed based on the extended area prediction input data set, to generate an extended area swimming animal spatial distribution map and identify a high probability distribution area.

[0053] The extended area prediction input data set is input into a random forest regression mapping structure and a deep neural network mapping structure, to perform swimming animal resource density value prediction calculation and swimming animal species existence probability value prediction calculation in the extended area.

[0054] The random forest regression mapping structure outputs the swimming animal resource density value in each spatial analysis unit in the extended area, and the deep neural network mapping structure outputs the swimming animal species existence probability value in each spatial analysis unit in the extended area.

[0055] The swimming animal resource density value and the swimming animal species existence probability value predicted in each spatial analysis unit in the extended area are stored in a grid data structure, and corresponding prediction time stamps, latitude and longitude spatial coordinates and depth level labels are marked.

[0056] The spatial interpolation and smoothing processing are performed on the predicted results of the swimming animal resource density value and the swimming animal species existence probability value in the extended area to generate a swimming animal population quantity density spatial distribution map and a swimming animal species existence probability spatial distribution map of the extended area.

[0057] According to the predicted calculation results, the threshold value of the swimming animal resource density value is defined as 10 tails / cubic meter, the threshold value of the swimming animal species existence probability value is defined as 0.6, and the spatial analysis unit with the resource density value greater than 10 tails / cubic meter and the species existence probability value greater than 0.6 is marked as a high-probability distribution area.

[0058] The swimming animal spatial distribution map of the extended area, the high-probability distribution area mask map, the swimming animal resource density value and the swimming animal species existence probability value data table are output.

[0059] Step S500.4, performing spatio-temporal dynamic analysis based on the swimming animal spatial distribution map of the extended area, and outputting a large-range prediction result file.

[0060] Based on the swimming animal spatial distribution map of the extended area and the high-probability distribution area mask map, and by using a spatial statistical analysis tool, the spatial migration characteristics of the high-probability distribution area at different prediction time stamps are analyzed.

[0061] The swimming animal population quantity density spatial distribution map, the swimming animal species existence probability spatial distribution map, the high-probability distribution area mask map, the swimming animal resource density value and the species existence probability value data table in the extended area are summarized to generate an extended area prediction result file containing spatio-temporal dynamic analysis map and trend change analysis.

[0062] Compared with the prior art, the present application has the following beneficial effects: 1. By deploying a temperature-salinity-depth profiler and an acoustic Doppler current profiler, and combining the spatial movement characteristics of an oceanographic research vessel, a three-dimensional environmental parameter data set covering the vertical and horizontal dimensions is constructed, which not only covers key physical factors such as temperature, salinity, depth, flow rate and flow direction, but also ensures the consistency of the data in the spatial reference system and the time axis, effectively improving the applicability and accuracy of the environmental factor data in the swimming animal distribution modeling.

[0063] 2. By integrating surface environmental DNA sampling, trawl sampling and wideband fisheries acoustic detection, multi-dimensional composition and distribution information of surface and upper-middle layer swimming animals are systematically obtained, and a unified structure data structure is constructed, which fully integrates macro and micro scale biological information, significantly enhances the representation ability of species diversity, spatial distribution and quantity density, and solves the problem of single data dimension and missing distribution information in the traditional monitoring method.

[0064] 3. By constructing a random forest regression model and a deep neural network model that fuse three-dimensional environmental parameters with multi-dimensional distribution data of swimming animals, and performing multiple rounds of cross-validation training, a stable and efficient prediction mapping structure is established. This mapping structure not only improves the accuracy of predicting the spatial distribution of swimming animals in the target area, but also outputs the optimal habitat environmental factor combination based on feature importance scores, providing data support for environmental regulation and resource assessment. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 A flowchart of an artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0067] Please refer to Figure 1 , Figure 1 A flowchart of an artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method is provided for the embodiments of the present application.

[0068] In the present embodiment, an artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method can include steps S100, S200, S300, S400 and S500.

[0069] Step S100, deploy a temperature-salinity-depth profiler and an acoustic Doppler current profiler, and collect and configure environmental parameter data sets to obtain a three-dimensional environmental parameter data set and generate a multi-dimensional environmental parameter database.

[0070] Step S200, deploy a multi-dimensional swimming animal collection device and collect data to obtain a mid-upper layer swimming animal species and quantity distribution data set and a multi-dimensional acoustic inversion data set, and configure a multi-dimensional swimming animal composition and distribution data structure.

[0071] Step S300, based on the three-dimensional environmental parameter data set and the multi-dimensional swimming animal composition and distribution data structure, perform data fusion to obtain a fusion input data set, configure a random forest regression model and a deep neural network model, and train the model. Configure the multivariate mapping relationship structure of environmental factors and swimming animal distribution, obtain the random forest regression mapping structure and the deep neural network mapping structure, and output the optimal habitat environmental parameter combination and the model performance index.

[0072] Step S400, based on the random forest regression mapping structure and the deep neural network mapping structure, generate the spatiotemporal distribution map of the swimming animal community structure, calculate the Shannon diversity index and the Simpson diversity index, and perform spatial statistical analysis.

[0073] Step S500, deploy remote sensing satellite platforms and unmanned aerial vehicle patrol platforms, collect macro-scale environmental parameter data and pre-process, configure an extended area prediction input data set, derive an extended area prediction input data set, predict and calculate the spatial distribution of swimming animals in the extended area, and identify high-probability distribution areas and perform spatiotemporal dynamic analysis.

[0074] In some embodiments, the step S100, in particular: Step S100.1, deploy a temperature-salinity-depth profiler and an acoustic Doppler current profiler, and collect data.

[0075] In the target ocean area, preset the investigation station, and fix and install the temperature-salinity-depth profiler CTD and the acoustic Doppler current profiler ADCP on the deck hoisting system and the bottom mounting bracket of the oceanographic research ship, respectively.

[0076] After the oceanographic research ship sails to the investigation station, start the deck hoisting system of the oceanographic research ship, control the temperature-salinity-depth profiler to be vertically lowered at a stable rate to a set detection depth, and continuously collect the temperature, salinity, and depth parameters of the water body through the integrated temperature sensor, salinity sensor, and pressure sensor during the lowering process. The detection depth is set according to the requirements of the sea area investigation task.

[0077] The collected environmental parameters are transmitted in real time to the data acquisition and processing module on the oceanographic research ship through a communication cable, and are recorded and stored according to the corresponding depth coordinates and time labels, and a continuous vertical temperature-salinity-depth profile data set is constructed.

[0078] The acoustic Doppler current profiler continuously works during the sailing of the oceanographic research ship, emits downward a multi-beam acoustic wave signal at a preset frequency, measures the movement speed of suspended particles in the water body according to the Doppler effect through the frequency shift of the received echo, and inversely obtains the flow velocity and flow direction angle in different depth sections.

[0079] The acoustic Doppler current profiler divides the entire water column into multiple equally spaced depth layers and outputs the corresponding velocity vector of each layer, including: velocity magnitude, direction angle, and vertical depth coordinate.

[0080] Step S100.2, configure the environmental parameter data set based on the data collected by the temperature-salinity-depth profiler and the acoustic Doppler current profiler, obtain a three-dimensional environmental parameter data set, and generate a multi-dimensional environmental parameter database.

[0081] The temperature parameters, salinity parameters and depth parameters collected by the temperature-salinity-depth profiler are converted into a unified spatial reference system and time-synchronized with the flow velocity, flow direction angle and depth parameters collected by the acoustic Doppler current profiler. The data are structured and coded in a unified format in the data processing module, and a three-dimensional environmental parameter data set covering temperature, salinity, depth, flow velocity and flow direction is constructed.

[0082] It should be noted that the unified spatial reference system conversion, time synchronization processing and unified format structured coding are well-known techniques in the art, and will not be described in detail here.

[0083] The three-dimensional environmental parameter data set is archived in units of survey stations, and is labeled with multi-dimensional tags according to time sequence, depth level and geographic coordinates to generate a multi-dimensional environmental parameter database for the target sea area.

[0084] In some embodiments, the step S200 specifically includes: Step S200.1, deploying a swimming animal multi-dimensional collection device, which includes a surface environmental DNA sampling device, a mid-upper layer trawl net system and a Simrad EK80 type wideband fisheries acoustic detection system.

[0085] Based on the surface environmental DNA sampling device, the mid-upper layer trawl net system and the Simrad EK80 type wideband fisheries acoustic detection system, the mid-upper layer swimming animal species and quantity distribution data set and the multi-dimensional acoustic inversion data set are obtained.

[0086] In the target ocean area, the swimming animal multi-dimensional collection device is deployed according to the survey station, and the biological sampling range of the surface, mid-upper layer and multi-depth layer is set at each sampling point to form a biological sampling grid covering the vertical and horizontal dimensions.

[0087] The surface environmental DNA sampling device collects 3-5 liters of seawater samples at each surface sampling point, and immediately filters the samples on site through a filter membrane with a pore size of 0.45 microns to obtain a filter membrane sample rich in swimming animal genetic material.

[0088] The filter membrane sample is transported to the laboratory through cold chain preservation, DNA extraction is performed using a commercial environmental DNA extraction kit, and the extracted DNA fragments are amplified and sequenced using an Illumina NovaSeq high-throughput sequencing platform.

[0089] The sequencing results are compared to the reference database NCBINT database or FISH-BOL database to obtain the species classification unit and corresponding sequence depth of the swimming animals, and a mid-upper swimming animal species composition data set containing the species name, classification level, relative abundance and sampling point identification is generated.

[0090] The middle layer trawl is used for horizontal trawling operation in a set water depth range. The four-piece middle layer trawl has a main dimension of 916 meshes by 400 mm, a mesh opening circumference of 366.4 m, a net body length of 120 m, and a single capsule structure. The mesh opening part uses large mesh, and the net body part uses machine-knitted mesh. The double-blade net plate uses a single hand cable connection mode. The trawling speed is set to 3.5-4.5 nmile / h, and the trawling time is 1 hour.

[0091] The swimming animal samples caught by the trawl device are classified, counted and weighed. The corresponding tail number and mass data of each species are recorded, and the trawling depth, sampling longitude and latitude coordinates and sampling time are marked to form a mid-upper swimming animal species and quantity distribution data set with three-dimensional labels of survey point, water depth layer and species.

[0092] The Simrad EK80 type wideband fisheries acoustic detection system is controlled to continuously emit wideband acoustic signals at 38 kHz, 70 kHz, 120 kHz and 200 kHz frequency bands during the voyage of the oceanographic research ship, and the vertical echo scattering intensity is recorded.

[0093] The Simrad EK80 type wideband fisheries acoustic detection system performs data acquisition at a pulse emission frequency of 5 times per second, divides the entire water column into multiple depth sections with a depth resolution of 0.5 meters, performs background noise elimination and target strength inversion processing on the original acoustic echo data, and combines system calibration data and acoustic scattering characteristic parameters of swimming animal target species to obtain the number density value of swimming animals per unit volume in each depth layer, forming a multi-dimensional acoustic inversion data set including depth coordinates, longitude and latitude coordinates, number density and acoustic frequency band identification.

[0094] Step S200.2, configure multi-dimensional swimming animal composition and distribution data structure.

[0095] Based on the obtained mid-upper swimming animal species and quantity distribution data set and multi-dimensional acoustic inversion data set, format uniform processing, spatial coordinate, depth label standardization, time synchronization and structured coding processing are performed, and a multi-dimensional swimming animal composition and distribution data structure oriented to survey stations is constructed, containing species name, number density, classification level, spatial position and sampling method label.

[0096] It should be noted that, due to the format uniform processing, spatial coordinates, depth label standardization, time synchronization and structured coding processing are well known to those skilled in the art, and will not be described in detail here.

[0097] In some embodiments, the step S300, in particular: Step S300.1, based on the three-dimensional environmental parameter data set and the multi-dimensional swimming animal composition and distribution data structure, the fusion input data set is obtained, and the random forest regression model and the deep neural network model are configured.

[0098] The temperature, salinity, depth, flow rate and flow direction in the three-dimensional environmental parameter data set are fused with the species name, resource density, classification level, spatial position and sampling method label in the multi-dimensional swimming animal composition and distribution data structure. In the fusion process, the longitude and latitude coordinates of the survey station, the depth level and the sampling time are used as the only index to match the environmental parameter data with the corresponding swimming animal distribution data, and the fusion input data set is obtained.

[0099] Based on the obtained fusion input data set, a random forest regression model and a deep neural network model are constructed respectively, and the input feature vector X of the random forest regression model is: , T is the temperature parameter, S is the salinity parameter, D is the depth parameter, V is the flow rate parameter, is the flow direction parameter, and the output result is the swimming animal resource density at the corresponding position , the model adopts the error optimization objective function: ; In the formula: MSE is the mean square error loss function of the random forest regression model, is the observed true value of the swimming animal resource density, is the predicted value of the swimming animal resource density by the random forest regression model, and n is the total number of training samples of the random forest regression model.

[0100] The input feature vector of the deep neural network model is consistent with that of the random forest regression model, and the network structure is set as follows: the number of nodes in the input layer is 5, the first hidden layer and the second hidden layer are 128 nodes respectively and use ReLU activation function, and the output layer uses Sigmoid function to output the probability value of species existence , the deep neural network model training process adopts binary cross entropy loss function for optimization, specifically: ; In the formula: is the binary cross entropy loss function of the deep neural network model, is the observed true label value of the species existence, which is 0 or 1, is the species presence probability value predicted by the deep neural network model, which takes value between 0 and 1, and m is the total number of training samples of the deep neural network model.

[0101] Step S300.2, training the random forest regression model and the deep neural network model based on the fused input data set.

[0102] The random forest regression model and the deep neural network model are trained respectively using the fused input data set, and the five-fold cross-validation method is used to verify the prediction performance of the random forest regression model and the deep neural network model.

[0103] The fused input data set is randomly divided into five subsets, of which four subsets are used for training the random forest regression model and the deep neural network model, and one subset is used for verifying the random forest regression model and the deep neural network model. The training and verification are repeated five times, and the model prediction error of each verification is recorded. When the trained random forest regression model and the deep neural network model show stable prediction performance on unseen data, the training is stopped.

[0104] Step S300.3, configuring the multivariate mapping relationship structure of environmental factors and swimming animal distribution based on the trained random forest regression model and the deep neural network model, obtaining the random forest regression mapping structure and the deep neural network mapping structure, and outputting the optimal habitat environmental parameter combination and the model performance index.

[0105] The trained random forest regression model and the deep neural network model are respectively parameterized to obtain the random forest regression mapping structure and the deep neural network mapping structure.

[0106] The random forest regression mapping structure includes: input environmental factor dimension, decision tree number, maximum tree depth, feature weight parameter and prediction error boundary.

[0107] The deep neural network mapping structure includes: input feature dimension, hidden layer node number, weight matrix, bias vector, activation function type and output species presence probability threshold.

[0108] The random forest regression mapping structure and the deep neural network mapping structure jointly represent the spatial response characteristics of the number density and species presence probability of swimming animals under different environmental parameter combinations.

[0109] The feature importance scores of temperature parameters, salinity parameters, depth parameters, flow rate parameters and flow direction parameters are extracted from the random forest regression mapping structure, and the optimal habitat environmental parameter combination is formed based on the descending order of the score values. The model accuracy, recall rate and F1 score are output from the deep neural network mapping structure, and the performance index is calculated as: ; ; ; In the formula: Accuracy is the prediction accuracy in the performance index of the deep neural network model, TP is the number of samples in the model prediction result in which the true class is positive and the model prediction is positive, TN is the number of samples in the model prediction result in which the true class is negative and the model prediction is negative, FP is the number of samples in the model prediction result in which the true class is negative but the model prediction is positive, FN is the number of samples in the model prediction result in which the true class is positive but the model prediction is negative, Recall is the recall rate in the performance index of the deep neural network model, F1-score is the comprehensive evaluation score in the performance index of the deep neural network model, and Precision is the precision in the model prediction result, i.e. the proportion of actual positives in the samples predicted as positive by the model.

[0110] The optimal habitat environment parameter combination and model performance index are recorded as the output file of feature selection and prediction performance evaluation.

[0111] In some embodiments, the step S400, in particular: Step S400.1, generating a spatiotemporal distribution map of the swimming animal community structure based on the random forest regression mapping structure and the deep neural network mapping structure.

[0112] Based on the random forest regression mapping structure and the deep neural network mapping structure, the target ocean area is divided into horizontal grids with a latitude and longitude grid scale of 1°x1°, and a plurality of vertical water layers are divided along the vertical direction with an interval of 500 meters, to form spatial analysis units intersected by the horizontal grids and the vertical water layers. The environmental parameter combination composed of the temperature parameter, the salinity parameter, the depth parameter, the flow velocity parameter and the flow direction parameter in each spatial analysis unit.

[0113] The environmental parameter combination in the spatial analysis unit is respectively input into the random forest regression mapping structure and the deep neural network mapping structure to predict the resource density value of the swimming animal and the species existence probability value of the swimming animal in each spatial analysis unit.

[0114] The resource density value and the species existence probability value predicted in each spatial analysis unit are subjected to spatial interpolation and smoothing processing, and the swimming animal population number density spatial distribution map and the species existence probability spatial distribution map in the target ocean area are output.

[0115] Step S400.2, based on the spatiotemporal distribution map of the swimming animal community structure, calculate the Shannon diversity index and the Simpson diversity index, obtain the Shannon diversity index spatial distribution map and the Simpson diversity index spatial distribution map, and perform spatial statistical analysis.

[0116] According to the generally accepted standard in the current field of ecology and environmental biological monitoring, Shannon diversity index value > 3 and Simpson diversity index value > 0.7 are defined as high diversity areas, and Shannon diversity index value < 1 and Simpson diversity index value < 0.3 are defined as low diversity areas.

[0117] The spatial statistical analysis is performed on the spatiotemporal distribution characteristics and variation trend of the high diversity area and the low diversity area, the spatial diffusion characteristics of the high diversity area and the low diversity area are analyzed with time variation, and the main environmental factors causing the variation of the high diversity area and the low diversity area are analyzed based on temperature parameters, salinity parameters, depth parameters, flow rate parameters and flow direction parameters.

[0118] The analysis result file including the Shannon diversity index and the Simpson diversity index spatiotemporal distribution map, the spatial statistical analysis data table and the trend change graph is generated, and the spatial distribution characteristics, the spatiotemporal dynamic variation trend, the community stability and the diversity condition of the ecological system of the swimming animal community structure in the target ocean area are fully characterized.

[0119] In some embodiments, the step S500, in particular: Step S500.1, deploying a remote sensing satellite platform and an unmanned aerial vehicle patrol platform, and collecting macro-scale environmental parameter data.

[0120] In the extended observation area outside the target ocean area, a remote sensing satellite platform and an unmanned aerial vehicle patrol platform are deployed, and the remote sensing satellite platform includes a MODIS remote sensing satellite, a Sentinel-3 remote sensing satellite and a VIIRS remote sensing satellite.

[0121] The unmanned aerial vehicle patrol platform includes a multi-rotor unmanned aerial vehicle and a fixed-wing unmanned aerial vehicle.

[0122] The remote sensing satellite platform performs remote sensing observation on the sea surface temperature parameters, the sea surface chlorophyll concentration parameters and the sea surface wind speed parameters in the extended observation area with a spatial resolution of 1 kilometer x 1 kilometer and a sampling period of 12 hours, and obtains the macro-scale environmental parameter data.

[0123] The unmanned aerial vehicle patrol platform patrols the sea area along the previously planned grid patrol route at a patrol flight height of 300 meters to obtain a sea surface visible light image with a spatial resolution of not less than 10 meters, an infrared temperature spectrum, a local wind speed parameter and a local wind direction parameter, and supplement local environmental parameter data with high spatial accuracy.

[0124] Step S500.2, data preprocessing based on macro-scale environmental parameter data, and configuring an extended area prediction input data set, obtaining an extended area prediction input data set, The macro-scale environmental parameter data obtained by the remote sensing satellite platform and the local environmental parameter data obtained by the unmanned aerial vehicle patrol platform are subjected to spatial coordinate re-projection, time synchronization resampling and data format unification processing.

[0125] The sea surface temperature parameter, sea surface chlorophyll concentration parameter, sea surface wind speed parameter, local wind speed parameter and local wind direction parameter subjected to spatial coordinate re-projection processing are subjected to spatial interpolation and edge splicing processing with the three-dimensional environmental parameter data set in the original investigation area, and an extended area prediction input data set is obtained.

[0126] Step S500.3, prediction calculation based on the extended area prediction input data set, generating an extended area swimming animal spatial distribution map, and identifying a high probability distribution area.

[0127] The extended area prediction input data set is input into a random forest regression mapping structure and a deep neural network mapping structure to perform swimming animal resource density value prediction calculation and swimming animal species existence probability value prediction calculation in the extended area.

[0128] The random forest regression mapping structure outputs the swimming animal resource density value in each spatial analysis unit of the extended area, and the deep neural network mapping structure outputs the swimming animal species existence probability value in each spatial analysis unit of the extended area.

[0129] The swimming animal resource density value and the swimming animal species existence probability value predicted in each spatial analysis unit of the extended area are stored in a grid data structure, and the corresponding prediction time stamp, latitude and longitude spatial coordinates and depth level label are marked.

[0130] The swimming animal resource density value prediction result and the swimming animal species existence probability value prediction result in the extended area are subjected to spatial interpolation and smoothing processing to generate a swimming animal population number density spatial distribution map and a swimming animal species existence probability spatial distribution map of the extended area.

[0131] According to the prediction calculation result, the threshold value of the swimming animal resource density value is defined as 10 individuals per cubic meter, the threshold value of the swimming animal species existence probability value is defined as 0.6, and the spatial analysis unit with the resource density value greater than 10 individuals per cubic meter and the species existence probability value greater than 0.6 is marked as a high-probability distribution area.

[0132] The swimming animal spatial distribution map of the extended area, the high-probability distribution area mask map, the swimming animal resource density value, and the swimming animal species existence probability value data table are output.

[0133] Step S500.4, performing spatio-temporal dynamic analysis based on the swimming animal spatial distribution map of the extended area, and outputting a large-range prediction result file.

[0134] Based on the swimming animal spatial distribution map of the extended area and the high-probability distribution area mask map, and by using a spatial statistical analysis tool, the spatial migration characteristics of the high-probability distribution area at different prediction time stamps are analyzed.

[0135] The swimming animal population number density spatial distribution map, the swimming animal species existence probability spatial distribution map, the high-probability distribution area mask map, the swimming animal resource density value, and the species existence probability value data table in the extended area are summarized to generate an extended area prediction result file containing spatio-temporal dynamic analysis map and trend change analysis.

[0136] In the above content, in actual application, first, a plurality of survey stations are set in a target ocean area, a CTD is fixedly installed on a deck hoisting system of a research vessel, and an ADCP is fixedly installed on a ship bottom mounting bracket. After the research vessel reaches the preset survey station, the deck hoisting system is started to control the CTD to be vertically lowered to a detection depth at a stable speed, and temperature, salinity, and depth parameters of the water body are collected in real time by temperature, salinity, and pressure sensors during the lowering process. The ADCP continuously emits acoustic signals and measures the movement speed of suspended particles in the water body during the sailing of the research vessel to obtain flow velocity and flow direction angle data of different depth sections.

[0137] Subsequently, the temperature, salinity, and depth data obtained by the CTD and the flow velocity, flow direction, and depth data obtained by the ADCP are subjected to unified spatial reference system conversion and time synchronization processing, a three-dimensional stereoscopic environmental parameter data set covering temperature parameters, salinity parameters, depth parameters, flow velocity parameters, and flow direction parameters is constructed, and is archived in units of survey stations to form a multi-dimensional environmental parameter database.

[0138] After that, the multi-dimensional collection device of swimming animals including the surface environmental DNA sampling device, the mid-upper layer trawl system and the Simrad EK80 type wideband fisheries acoustic detection system is deployed inside each investigation station; the surface environmental DNA sampling device collects seawater and obtains filter membrane samples after on-site filtration, and the species composition data of mid-upper layer swimming animals are obtained through DNA extraction and sequencing; the mid-upper layer trawl system performs horizontal trawl operation and catches swimming animals, and the captured swimming animals are classified, counted and weighed to form the species and quantity distribution data of mid-upper layer swimming animals; the Simrad EK80 type wideband fisheries acoustic detection system emits different frequency band acoustic signals, and the number density of swimming animals in each depth layer is obtained through target intensity inversion to form a multi-dimensional acoustic inversion data set.

[0139] Then, based on the three-dimensional environmental parameter data set and the multi-dimensional swimming animal composition and distribution data structure, the fusion input data set is obtained; the fusion input data set is used to construct a random forest regression model and a deep neural network model respectively, and the models are trained, and the five-fold cross-validation method is used to verify the prediction performance of the models; the trained random forest regression mapping structure and deep neural network mapping structure are obtained, the optimal habitat environmental parameter combination is output, and the performance indicators of the model are calculated, including the prediction accuracy, recall rate and F1 score of the model, to form the feature selection and prediction performance evaluation output file.

[0140] Then, based on the random forest regression mapping structure and the deep neural network mapping structure, the target sea area is horizontally and vertically gridded to obtain the swimming animal resource density and species existence probability data in each spatial analysis unit; and spatial interpolation and smoothing processing are performed to generate the swimming animal population number density spatial distribution map and the species existence probability spatial distribution map; the Shannon diversity index and the Simpson diversity index of each spatial analysis unit are calculated, and the corresponding spatial distribution map is output, the high diversity area and the low diversity area are defined and identified, the spatial statistical analysis is performed, and the analysis result file including the trend change graph is generated.

[0141] Finally, the MODIS remote sensing satellite, the Sentinel-3 remote sensing satellite, the VIIRS remote sensing satellite, and the multi-rotor unmanned aerial vehicle and fixed-wing unmanned aerial vehicle patrol platform are deployed in the extended observation area of the target ocean area to collect macro-scale environmental parameter data and perform spatial coordinate re-projection and time synchronization processing, so as to obtain an extended area prediction input data set; the extended area prediction input data set is input into a random forest regression mapping structure and a deep neural network mapping structure for prediction calculation, so as to generate an extended area swimming animal spatial distribution map and identify a high probability distribution area; the swimming animal spatial distribution in the extended area is analyzed in time and space in combination with the remote sensing satellite and unmanned aerial vehicle patrol data, and finally an extended area prediction result file containing a dynamic analysis map and trend change analysis is generated.

[0142] While embodiments of the present application have been shown and described, it is to be understood that the embodiments described are merely divergences, modifications, replacements and variations of the embodiments, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based stereoscopic multi-dimensional ocean swimming animal monitoring method, characterized by, The steps include: S100: deploy a temperature-salinity-depth profiler and an acoustic Doppler current profiler, collect data, configure an environmental parameter data set, obtain a three-dimensional environmental parameter data set, and generate a multi-dimensional environmental parameter database; S200, deploying a multi-dimensional swimming animal collection device and performing data collection to obtain a mid- and upper-layer swimming animal species and quantity distribution data set and a multi-dimensional acoustic inversion data set, and configuring a multi-dimensional swimming animal composition and distribution data structure; S300, performing data fusion based on a three-dimensional environmental parameter data set and a multi-dimensional swimming animal composition and distribution data structure, obtaining a fused input data set, configuring a random forest regression model and a deep neural network model and training, configuring a multivariate mapping relationship structure between environmental factors and swimming animal distribution, obtaining a random forest regression mapping structure and a deep neural network mapping structure, and outputting an optimal habitat environmental parameter combination and model performance indicators; S400, based on the random forest regression mapping structure and deep neural network mapping structure, generates spatiotemporal distribution maps of swimming animal community structure, calculates Shannon diversity index and Simpson diversity index, and performs spatial statistical analysis; S500. Deploy remote sensing satellite platforms and drone inspection platforms, collect macro-scale environmental parameter data and pre-process them, configure the extended area prediction input data set, obtain the extended area prediction input data set, predict and calculate the spatial distribution of swimming animals in the extended area, identify high-probability distribution areas and conduct spatiotemporal dynamic analysis.

2. The artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method according to claim 1, wherein, The S100 is specifically: S100.

1. Deploy a temperature-salinity-depth profiler and an acoustic Doppler current profiler and collect data; In the target ocean area, a survey station is preset, and the temperature-salinity-depth profiler (CTD) and the acoustic Doppler current profiler (ADCP) are fixedly installed on the deck hoisting system and the bottom mounting frame of the ocean-going survey vessel respectively; After the ocean-going survey vessel sails to the survey station, the deck hoisting system of the ocean-going survey vessel is activated to control the temperature-salinity-depth profiler to be vertically lowered to the set detection depth at a stable rate. During the lowering process, the temperature parameters, salinity parameters and depth parameters of the water body are continuously collected through the integrated temperature sensor, salinity sensor and pressure sensor. The detection depth is set according to the requirements of the marine survey mission; The collected environmental parameters are transmitted in real time via communication cables to the data acquisition and processing module on the ocean survey vessel, and are recorded and stored according to the corresponding depth coordinates and time tags, and a continuous vertical temperature-salinity-depth profile data set is constructed; The acoustic Doppler current profiler works continuously while the ocean-going survey vessel is sailing. It transmits a multi-beam acoustic signal at a preset frequency downward, receives the echo frequency shift, and uses the Doppler effect to measure the velocity of suspended particles in the water. It then inverts the velocity and direction of the flow in different depth sections. The acoustic Doppler current profiler divides the entire water column into multiple equally spaced depth layers and outputs the velocity vector corresponding to each layer, including: velocity magnitude, direction angle and vertical depth coordinate; S100.

2. Configure an environmental parameter data set based on data collected by the temperature-salinity-depth profiler and the acoustic Doppler current profiler, obtain a three-dimensional environmental parameter data set, and generate a multi-dimensional environmental parameter database; The temperature, salinity, and depth parameters collected by the temperature-salinity-depth profiler are converted into a unified spatial reference system and time-synchronized with the flow velocity, flow direction angle, and depth parameters collected by the acoustic Doppler current profiler. They are then structured and encoded in a unified format in the data processing module to construct a three-dimensional environmental parameter data set covering five physical factors: temperature, salinity, depth, flow velocity, and flow direction.

3. The artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method according to claim 1, wherein, The S200 is specifically: S200.

1. Deploy a multi-dimensional swimming animal collection device, comprising: a surface environmental DNA sampling device, a pelagic trawl system, and a Simrad EK80 broadband fishery acoustic detection system; Using surface environmental DNA sampling devices, pelagic trawl systems, and the Simrad EK80 broadband fishery acoustic detection system, we collected data on the distribution of pelagic swimming animals and their abundance, as well as multidimensional acoustic inversion data. In the target ocean area, multi-dimensional swimming animal collection equipment is deployed according to the survey station. The biological sampling range of the surface layer, the middle layer and multiple depth layers is set at each sampling point, and a biological sampling grid covering the vertical and horizontal dimensions is formed; The surface environmental DNA sampling device collects 3 to 5 liters of seawater samples at each surface sampling point and immediately filters the samples on-site through a filter membrane with a pore size of 0.45 microns to obtain a filter membrane sample rich in the genetic material of swimming animals; The filter membrane samples were transported to the laboratory through cold chain storage, DNA was extracted using a commercial environmental DNA extraction kit, and the extracted DNA fragments were amplified and sequenced using the Illumina NovaSeq high-throughput sequencing platform; The sequencing results were aligned to the reference database NCBINT or FISH-BOL to obtain the species taxonomic units and corresponding sequence depths of swimming animals, and a data set of mid- and upper-level swimming animal species composition was generated, including species name, taxonomic level, relative abundance, and sampling site identification; A mid-water trawl was used for horizontal trawling within the set water depth range. The four-piece mid-water trawl had a main size of 916 mesh × 400 mm, a net opening circumference of 366.4 m, and a net body length of 120 m. The net adopted a single-blank structure, with a large mesh opening and a machine-woven mesh body. Double-leaf mesh board, using a single-handed line connection method, set the towing speed to 3.5-4.5nmile / h, and the towing time is 1 hour; The swimming animal samples caught by the trawl device are classified, counted and weighed, and the number and mass data of each species are recorded. The trawl depth, sampling latitude and longitude coordinates and sampling time are annotated to form a data set of the species and number distribution of mid- and pelagic swimming animals with three-dimensional labels of survey point, water depth and species. The Simrad EK80 type wideband fishery acoustic detection system continuously emits 38 kHz, 70 kHz, 120 kHz and 200 kHz frequency band wideband acoustic signals during the voyage of the oceanographic survey ship, and records the vertical echo scattering intensity; The Simrad EK80 type wideband fishery acoustic detection system collects data at a pulse emission frequency of 5 times per second, divides the entire water column into multiple depth sections with a depth resolution of 0.5 meters, performs background noise elimination and target intensity inversion processing on the original acoustic echo data, and combines system calibration data and acoustic scattering characteristic parameters of swimming animal target species to obtain the number density value of swimming animals per unit volume in each depth layer, forming a multi-dimensional acoustic inversion data set including depth coordinates, latitude and longitude coordinates, number density and acoustic frequency band identifier; S200.2, configure a multi-dimensional swimming animal composition and distribution data structure; Based on the obtained upper layer swimming animal species and number distribution data set and the multi-dimensional acoustic inversion data set, perform format uniform processing, spatial coordinate, depth label standardization, time synchronization and structured coding processing, and construct a multi-dimensional swimming animal composition and distribution data structure oriented to the survey station, containing species name, number density, classification level, spatial position and sampling method label.

4. The artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method of claim 1, wherein, The S300, specifically: S300.1, based on the three-dimensional environmental parameter data set and the multi-dimensional swimming animal composition and distribution data structure, data fusion is performed to obtain a fusion input data set, and a random forest regression model and a deep neural network model are configured; The temperature, salinity, depth, flow rate and flow direction in the three-dimensional environmental parameter data set are fused with the species name, resource density, classification level, spatial position and sampling method label in the multi-dimensional swimming animal composition and distribution data structure, and in the fusion process, the latitude and longitude coordinates of the survey station, the depth level and the sampling time are used as the only index to match the environmental parameter data with the corresponding swimming animal distribution data, and a fusion input data set is obtained; Based on the obtained fusion input data set, a random forest regression model and a deep neural network model are constructed respectively, the input feature vector X of the random forest regression model is: , T is a temperature parameter, S is a salt parameter, D is a depth parameter, V is a flow rate parameter, is a flow direction parameter, and the output result is the swimming animal resource density at the corresponding position ; The input feature vector of the deep neural network model is consistent with the random forest regression model, and the network structure is set as follows: the number of nodes of the input layer is 5, the first hidden layer and the second hidden layer are 128 nodes respectively and adopt the ReLU activation function, and the output layer adopts the Sigmoid function to output the probability value of the existence of the species , and the binary cross entropy loss function is used for optimization in the training process of the deep neural network model. S300.2, based on the fusion input data set, train the random forest regression model and the deep neural network model; The fusion input data set is used to train the random forest regression model and the deep neural network model respectively, and the five-fold cross-validation method is used to verify the prediction performance of the random forest regression model and the deep neural network model; The fusion input data set is randomly divided into five subsets, four of which are used for training the random forest regression model and the deep neural network model, and one is used for verifying the random forest regression model and the deep neural network model, the training and verification are repeated five times, and the model prediction error of each verification is recorded, when the trained random forest regression model and the deep neural network model show stable prediction performance on unseen data, the training is stopped; S300.3, based on the trained random forest regression model and the deep neural network model, configure the multivariate mapping relationship structure of environmental factors and swimming animal distribution, obtain the random forest regression mapping structure and the deep neural network mapping structure, and output the optimal habitat environmental parameter combination and the model performance index; The trained random forest regression model and the deep neural network model are respectively parameterized to obtain the random forest regression mapping structure and the deep neural network mapping structure; The random forest regression mapping structure includes: input environmental factor dimension, decision tree number, maximum tree depth, feature weight parameter and prediction error boundary; The deep neural network mapping structure includes: input feature dimension, hidden layer node number, weight matrix, bias vector, activation function type and output species existence probability threshold; The feature importance scores of the temperature parameter, the salinity parameter, the depth parameter, the flow rate parameter and the flow direction parameter are extracted from the random forest regression mapping structure, and the optimal habitat environmental parameter combination is formed in descending order based on the score values. The model accuracy, recall rate and F1 score are output from the deep neural network mapping structure, and the performance index is calculated as follows: ; ; ; In the formula: Accuracy is the prediction accuracy in the performance index of the deep neural network model, TP is the number of samples in the model prediction result that are actually positive and predicted positive, TN is the number of samples in the model prediction result that are actually negative and predicted negative, FP is the number of samples in the model prediction result that are actually negative but predicted positive, FN is the number of samples in the model prediction result that are actually positive but predicted negative, Recall is the recall rate in the performance index of the deep neural network model, F1-score is the comprehensive evaluation score in the performance index of the deep neural network model, and Precision is the accuracy rate in the model prediction result, that is, the proportion of actual positive samples in the samples predicted positive by the model; The optimal habitat environmental parameter combination and the model performance index are recorded as the output file of feature selection and prediction performance evaluation.

5. The artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method according to claim 1, wherein, The S400, in particular: S400.1, based on the random forest regression mapping structure and the deep neural network mapping structure, generate the spatiotemporal distribution map of swimming animal community structure; Based on the random forest regression mapping structure and the deep neural network mapping structure, the target ocean area is divided into horizontal grids with a grid scale of 1°×1°, and a plurality of vertical water layers are divided along the vertical direction with an interval of 500 meters, to form spatial analysis units intersected by the horizontal grids and the vertical water layers. The environmental parameter combination composed of the temperature parameter, the salinity parameter, the depth parameter, the flow rate parameter and the flow direction parameter in each spatial analysis unit; The environmental parameter combination in the spatial analysis unit is input into the random forest regression mapping structure and the deep neural network mapping structure respectively, and the resource density value of swimming animals and the species existence probability value of swimming animals in each spatial analysis unit are predicted. The resource density values and the species existence probability values obtained by the intra-space analysis unit are spatially interpolated and smoothed, and a swimming animal population quantity density spatial distribution map and a species existence probability spatial distribution map in the target ocean area are output; S400.

2. Based on the swimming animal community structure spatio-temporal distribution map, the Shannon diversity index and the Simpson diversity index are calculated, the Shannon diversity index spatial distribution map and the Simpson diversity index spatial distribution map are obtained, and spatial statistical analysis is performed; According to the generally accepted standard in the current field of ecology and environmental biology monitoring, Shannon diversity index value > 3 and Simpson diversity index value > 0.7 are defined as high diversity areas, Shannon diversity index value < 1 and Simpson diversity index value < 0.3 are defined as low diversity areas; The spatio-temporal distribution characteristics and variation trends of the high diversity areas and the low diversity areas are subjected to spatial statistical analysis, the spatial diffusion characteristics of the high diversity areas and the low diversity areas with time are analyzed, and the main environmental factors causing the changes of the high diversity areas and the low diversity areas are analyzed based on temperature parameters, salinity parameters, depth parameters, flow rate parameters and flow direction parameters; The analysis result file including the Shannon diversity index and the Simpson diversity index spatio-temporal distribution map, the spatial statistical analysis data table and the trend change graph is generated, and the spatial distribution characteristics, the spatio-temporal dynamic variation trend, the community stability and the diversity condition of the ecosystem of the swimming animal community structure in the target ocean area are comprehensively characterized.

6. The artificial intelligence-based three-dimensional multi-dimensional ocean swimming animal monitoring method of claim 1, wherein, The S500 specifically comprises: S500.

1. Deploying a remote sensing satellite platform and an unmanned aerial vehicle patrol platform, and collecting macro-scale environmental parameter data; Outside the target ocean area, a remote sensing satellite platform and an unmanned aerial vehicle patrol platform are deployed in the extended observation area, the remote sensing satellite platform comprises a MODIS remote sensing satellite, a Sentinel-3 remote sensing satellite and a VIIRS remote sensing satellite; The unmanned aerial vehicle patrol platform comprises a multi-rotor unmanned aerial vehicle and a fixed-wing unmanned aerial vehicle; The remote sensing satellite platform performs remote sensing observation on the sea surface temperature parameter, the sea surface chlorophyll concentration parameter and the sea surface wind speed parameter in the extended observation area with a spatial resolution of 1 kilometer x 1 kilometer and a sampling period of 12 hours, and obtains the macro-scale environmental parameter data; The unmanned aerial vehicle patrol platform patrols the sea area along the pre-planned grid patrol route at a patrol flight height of 300 meters in the extended observation area, and obtains a sea surface visible light image, an infrared temperature map, a local wind speed parameter and a local wind direction parameter with a spatial resolution of not less than 10 meters, thereby supplementing the local environmental parameter data with high spatial precision; S500.

2. Based on the macro-scale environmental parameter data, data preprocessing is performed, and an extended area prediction input data set is configured, and the extended area prediction input data set is obtained, The macro-scale environmental parameter data obtained by the remote sensing satellite platform and the local environmental parameter data obtained by the unmanned aerial vehicle patrol platform are subjected to spatial coordinate re-projection, time synchronization resampling and data format unification processing; The sea surface temperature parameter, sea surface chlorophyll concentration parameter, sea surface wind speed parameter, local wind speed parameter and local wind direction parameter processed by spatial coordinate re-projection are spatially interpolated and edge-spliced with the original survey area three-dimensional environmental parameter data set to obtain an extended area prediction input data set; S500.3, based on the extended area prediction input data set, performing prediction calculation to generate an extended area swimming animal spatial distribution map, and identifying a high probability distribution area; The extended area prediction input data set is input into a random forest regression mapping structure and a deep neural network mapping structure to perform swimming animal resource density value prediction calculation and swimming animal species existence probability value prediction calculation in the extended area; The random forest regression mapping structure outputs the swimming animal resource density value in each spatial analysis unit of the extended area, and the deep neural network mapping structure outputs the swimming animal species existence probability value in each spatial analysis unit of the extended area; The swimming animal resource density value prediction result and the swimming animal species existence probability value prediction result in the extended area are spatially interpolated and smoothed to generate a swimming animal population number density spatial distribution map and a swimming animal species existence probability spatial distribution map of the extended area; According to the prediction calculation result, the threshold value of the swimming animal resource density value is defined as 10 tails / cubic meter, and the threshold value of the swimming animal species existence probability value is defined as 0.6, and the spatial analysis unit with a resource density value greater than 10 tails / cubic meter and a species existence probability value greater than 0.6 is marked as a high probability distribution area; S500.4, based on the extended area swimming animal spatial distribution map, performing spatio-temporal dynamic analysis and outputting a large-scale prediction result file; Based on the swimming animal spatial distribution map of the extended area and the high probability distribution area mask map, and using a spatial statistical analysis tool, the spatial migration characteristics of the high probability distribution area under different prediction timestamps are analyzed; The swimming animal population number density spatial distribution map, the swimming animal species existence probability spatial distribution map, the high probability distribution area mask map, the swimming animal resource density value and the species existence probability value data table in the extended area are summarized to generate an extended area prediction result file containing spatio-temporal dynamic analysis map and trend change analysis.

Images