Intelligent fishing method for effectively improving fishing efficiency

Abstract

The invention relates to the technical field of intelligent fishing, in particular to an intelligent fishing method for effectively improving fishing efficiency, which comprises the following steps: S1, constructing a fish school calculation sub-model: recording historical fish school data, environment data and fishing data, and constructing a fishing calculation model according to the data; s2, ecological protection and optimization: setting a dynamic threshold value and a trapping forbidding rule, monitoring environmental data in real time, and embedding the data and a fishing policy rule into a calculation model to realize strategy self-adaption of different water area scenes; on the basis of the fish school calculation sub-model and the environment prediction sub-model, satellite remote sensing and sonar detection are combined, the positioning error of the fish school gathering area is greatly reduced, the navigation path optimization time of the fishing boat can be shortened, the fish catch per unit time is greatly improved, and through the S-shaped navigation path and the electronic pulse driving technology, the fish school gathering efficiency is improved, and the fishing efficiency is improved. Edge computing equipment is matched to coordinate multiple ships to form a surrounding net, and the cluster fishing success rate is greatly increased.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fishing, and particularly relates to an intelligent fishing method that effectively improves fishing efficiency. Background Art

[0002] Marine fishing refers to the fishing activities of marine fish and other aquatic economic animals and plants in the ocean using various fishing gears, fishing boats and equipment, mainly including various natural seawater animals and plants such as fish, shrimps, crabs, shellfish, pearls, and algae. The marine fishing industry is a traditional marine industry and an important part of the marine aquaculture industry. From the aspect of fishing ground utilization, it is generally divided into inshore fishing, offshore fishing, outer sea fishing and deep sea fishing. With the development of technology, in order to improve fishing efficiency, intelligent fishing has emerged. Intelligent fishing is a way to achieve efficient fishing by using artificial intelligence technology to locate, identify and track fish. By using devices such as sensors and cameras, the intelligent fishing system can monitor the activities of fish in real time and use machine learning algorithms for positioning and identification. The system will automatically adjust the fishing intensity according to the activities of fish to improve fishing efficiency.

[0003] For example, a marine fishing method with the application number CN201110343424.4 and the authorization announcement date of November 27, 2013. It uses at least one fishing boat to cooperate with two or more LED fish attracting floating platforms. The LED fish attracting floating platforms are independently set, and the fishing boat can move between the above LED fish attracting floating platforms for fishing at each point; the LED fish attracting floating platform is remotely controlled by the fishing boat to light up and gather fish. The present invention can achieve continuous fishing operations back and forth, achieving low energy consumption and high efficiency, replacing traditional high energy consumption and low efficiency fishing modes such as trawling, promoting fishing operations, and being environmentally friendly. The LED fish attracting floating platform can independently float on the water, and the underwater LED fish gathering lamp and the water surface LED fish gathering lamp are used to gather fish and attract fish. The fishing operation is more flexible, and since it is powered by a storage battery, the sound of the storage battery power supply is small and there is no noise. It is remotely controlled by the fishing boat to light up and gather fish, improving fishing efficiency, saving energy and protecting the environment.

[0004] Although the above and in the prior art have improved fishing efficiency by setting up intelligent fishing devices, there are still a large number of disadvantages, which are specifically as follows:

[0005] (1) Low efficiency: relying on manual experience and lacking accurate prediction of fish school dynamics;

[0006] (2) Ecological damage: overfishing and accidental capture of non-target species leading to resource depletion;

[0007] (3) Technical limitations: low utilization rate of single sensor data and insufficient intelligence of fishing equipment;

[0008] In view of this, an intelligent fishing method is designed to effectively improve fishing efficiency and solve the above problems. Summary of the Invention

[0009] The object of the present invention is to provide an intelligent fishing method that can effectively improve fishing efficiency to solve the above deficiencies in the prior art.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] An intelligent fishing method that can effectively improve fishing efficiency, including:

[0012] Step S1. Construct a fish school operator model: collect historical fish school data, environmental data and fishing data, and construct a fishing calculus model based on the data;

[0013] It should be noted that the historical fish school data includes fish school species data and biological behavior interaction data, the environmental data includes water quality and temperature data, water flow and terrain data, and the fishing data includes fishing pressure data and ecological protection measure data.

[0014] The fishing calculus model is divided into a fish school operator model, an environmental prediction sub-model, a fishing operator model and a decision-making sub-model, and the construction steps of the fishing calculus model are as follows:

[0015] Step S1-1. Data acquisition and processing: Obtain historical fish school data, environmental data and fishing data through the Internet, artificial communication and government announcements, convert the data into a unified format, and then perform feature extraction on the data to obtain key information in the data. The key information is fish school species information, trajectory information, fish age information, quantity information, location information, etc.;

[0016] Step S1-2. Model construction and training: Select a deep learning network architecture, input the key information to construct a model, then set parameters and construct a loss function, and optimize the model according to the loss function. The class task combines FocalLoss to alleviate the class imbalance problem (such as rare fish species identification);

[0017] It should be noted that the CIoULoss (Complete Intersection over Union) is used for the loss function in the object detection task

[0018] to replace the traditional MSE and improve the regression accuracy of the bounding box.

[0019] Step S1-3. Verification and deployment: Divide the test set to verify the generalization ability, focus on complex scenarios, use TensorRT to quantize the model, compress the model volume to 1 / 4 of the original size, improve the operation efficiency of edge devices, and then optimize the computational graph structure for embedded devices, remove redundant layers and merge operators.

[0020] The fish school operator model is used to calculate and predict fish school data, and the specific steps are as follows:

[0021] Step S2-1. Model initialization and parameter setting:

[0022] (2) Initialization of fish school state: Set the fish school scale and initial distribution range, the position vector and fitness value carried by each individual in the fish school, and then define the environmental parameter space, including water temperature gradient, dissolved oxygen concentration and food density distribution field;

[0023] (2) Configuration of behavior rules: Foraging behavior: Individuals move towards the neighborhood with higher fitness, simulating the gradient ascent process, and the step size is dynamically adjusted; Aggregation behavior: Calculate the local density threshold, and when the number of individuals within the sensing radius exceeds the threshold, trigger group aggregation to reduce the dispersion;

[0024] Step S2-2. Dynamic behavior simulation and interaction:

[0025] (1) Cooperative operation of multiple behaviors: Following behavior: Individuals track the neighboring individual with the highest fitness, and if the target area is not oversaturated, update the position, otherwise perform random walk;

[0026] Random behavior: Introduce position perturbation with a probability of 0.1 to avoid the algorithm falling into a local optimal solution;

[0027] (2) Environmental response mechanism: Encode the marine environmental parameters as the weights of the fitness function to drive the fish school to migrate to the high-resource area;

[0028] Step S2-3. Prediction model optimization and iteration:

[0029] (1) Hybrid model fusion: Combine the fish school algorithm with the BP neural network: The fish school optimizes the initial values of the neural network weights to improve the convergence speed, introduce a convolutional layer to process spatially distributed data, and extract the spatio-temporal features of the fish school movement trajectory;

[0030] (2) Parameter adaptive adjustment: Dynamically adjust the behavior rule parameters based on historical prediction errors, and update the network weights through backpropagation;

[0031] Step S2-4. Prediction result generation:

[0032] (2) Data output form: Generate a heat map of fish school density and a probability distribution map of migration paths, and the time granularity can be refined to the hourly level;

[0033] It should be noted that the cross-validation method is used for verification and calibration. 80% of the historical observation data is divided for training, and 20% is used for testing. Calculate the root mean square error RMSE < 15%, compare with satellite remote sensing monitoring data, and correct the response deviation of the model to sudden environmental changes (such as sudden ocean current changes).

[0034] The environmental prediction sub-model is used to predict the future changes in the fishing environment, and the specific steps are as follows:

[0035] Step S3-1. Model initialization and parameter setting:

[0036] (1) Definition of environmental parameters: Set the core environmental variables, including key indicators such as water temperature (daily variation range ±2°C), dissolved oxygen concentration (threshold range 4-8 mg / L), and chlorophyll concentration (correlated with plankton abundance); and construct an initial parameter distribution field in combination with historical monitoring data (such as satellite remote sensing, sensor network), with a spatial resolution of not less than 1 km × 1 km;

[0037] (2) Configuration of dynamic driving mechanism: Encode meteorological data (such as wind speed, air pressure) and oceanographic parameters (such as ocean current speed, salinity gradient) as driving factors for environmental evolution, and define the coupling relationship between parameters, such as the non-linear response formula for the decrease in dissolved oxygen solubility caused by the increase in water temperature;

[0038] Step S3-2. Simulation of dynamic environmental parameters:

[0039] (1) Numerical model solution: Discretize partial differential equations using the finite difference method or the finite volume method to simulate the spatio-temporal evolution of water quality parameters (such as the diffusion process of dissolved oxygen), and then introduce a random perturbation term (such as Gaussian noise) to characterize environmental uncertainty and enhance the model's prediction ability for emergencies;

[0040] (2) Multi-time scale prediction:

[0041] Short-term prediction (24-72 hours): Update the initial field of the model based on real-time data assimilation technology, and the prediction accuracy error < 10%;

[0042] Long-term trend analysis (monthly / yearly): Combine climate model outputs (such as ENSO index) to evaluate the periodic change law of environmental parameters;

[0043] Step S3-3. Multi-source data fusion and model optimization:

[0044] (1) Heterogeneous data integration: Integrate satellite remote sensing data (such as MODIS chlorophyll products), real-time monitoring data from buoy sensors, and historical fishing records of fishing vessels, and use the Kalman filter or particle filter algorithm to correct the prediction deviation of the model and improve data consistency;

[0045] (2) Parameter sensitivity analysis: Identify the dominant environmental variables through the Morris screening method or the Sobol index method (such as the contribution degree of water temperature to fish migration > 60%), and dynamically adjust the model weights to preferentially optimize the prediction accuracy of highly sensitive parameters;

[0046] Step S3-4. Prediction Result Generation and Risk Assessment:

[0047] (1) Output Form: Generate the environmental parameter change curve for the next 72 hours (such as the predicted daily decline of dissolved oxygen concentration) and the spatial thermal map, mark the high-risk areas (such as the sea area with dissolved oxygen < 4mg / L), and trigger the fishing operation warning;

[0048] (2) Ecological Impact Assessment: Input the prediction results into the fish population operator model, evaluate the impact of environmental changes on fish population distribution (such as a 20% decrease in aggregation) and resource quantity (such as a ±15% fluctuation in biomass), and combine with the sustainable fishing threshold (such as the maximum allowable fishing intensity coefficient k = 0.3) to formulate a dynamic fishing strategy;

[0049] It should be noted that the subsequent model verification and iterative update are specifically as follows:

[0050] (1) Cross-validation mechanism: Divide the training set (80% historical data) and the test set (20%), calculate the root mean square error (RMSE < 8%) and the Nash efficiency coefficient (NSE > 0.75), then compare with the actual observed data (such as Argo float profile data), correct the systematic bias of the model, deploy the incremental learning framework, update the model parameters every 24 hours to adapt to the gradual and sudden changes of environmental parameters, and combine with the reinforcement learning strategy to dynamically adjust the model complexity and computational resource allocation (such as grid encryption area focusing on high-gradient change areas).

[0051] The fishing operator model dynamically simulates the fish behavior based on the fish population trajectory operator model and the environmental prediction sub-model, and the specific steps are as follows:

[0052] Step S4-1. Initialization and Parameter Setting:

[0053] (3) Divide the fish population age groups (1 to 4 years old) according to the calculation results of the fish population operator model, define the initial quantity distribution and biological parameters (natural mortality rate of 0.8 / year, spawning quantity gradient, etc.) of each age group, and set the fishing strategy constraint conditions: for example, only allow fishing of 3-year-old and 4-year-old fish, and the fishing intensity coefficient is related to the mesh size (such as 13mm mesh);

[0054] (4) Based on the dynamic parameters (water temperature, dissolved oxygen, chlorophyll concentration) output by the environmental prediction sub-model, construct a 1km×1km grid environmental field, and map the environmental parameters into fish behavior driving factors (such as high chlorophyll areas are marked as priority foraging areas);

[0055] Step S4-2. Dynamic Simulation and Interactive Computation:

[0056] (4) Operation of the fish population trajectory operator model: Simulate the change of the age group quantity based on the Leslie matrix, and the formula is:

[0057] N t+1 = A·N t - C(k)·N t

[0058] where A is a transition matrix containing the reproduction rate and survival rate, C(k) is a fishing intensity coefficient matrix, and N t represents the age structure vector of the fish population at time t, usually a column vector, and N t+1 represents the age structure vector of the fish population at time t + 1, which evolves from the population state and dynamic process at the current time. Subsequently, the individual migration paths can be adjusted by combining fish behavior rules (foraging, schooling, following). The migration step size is positively correlated with the environmental gradient (such as food density);

[0059] (5) Driven by the environmental prediction sub - model: The finite volume method is used to predict environmental parameters (such as changes in dissolved oxygen concentration) in the next 72 hours. The model input is updated in real - time through data assimilation technology. For the triggering of environmental mutation warnings (such as areas with dissolved oxygen < 4mg / L), the fish population migration path is dynamically adjusted to avoid risk areas;

[0060] (6) Dynamic adjustment of fishing intensity: According to the age - based fish population density distribution, the fishing intensity coefficient k is optimized to meet the constraint condition:

[0061] ∑(K n ·N n ) ≤ sustainable threshold

[0062] where K n represents the fishing intensity of fish at age n, N n represents the number of individuals in the fish population at age n. n is the minimum fish age allowed to be fished. The sustainable threshold is the upper limit of the allowable maximum catch. Exceeding this value will lead to an unsustainable population size, and the value is determined by the environmental carrying capacity and the intrinsic growth rate of the population;

[0063] Step S4 - 3. Multi - model coupling and data fusion:

[0064] (3) Cross - model interaction mechanism: The fish population trajectory operator sub - model outputs the probability distribution of the migration path, and the environmental prediction sub - model provides the dynamic environmental field. The two are corrected through Kalman filtering, and Monte Carlo sampling is introduced to simulate the impact of random events (such as ocean current mutations) on the fish population distribution, generating multi - scenario prediction results;

[0065] (4) Calculate the fish population biomass per unit area (such as the average mass of 4 - year - old fish is 22.99g / fish), combine with the fishing intensity to generate a catch prediction curve, and optimize the fishing strategy through dynamic programming algorithm to ensure the stability of the fish population. The objective function is:

[0066] max (economic benefits) S.T N t+1 ≥Nt

[0067] Among them, max(economic benefits) maximizes the economic benefits of fishing activities under sustainable constraints, N t+1 , N t respectively represent the fish stock quantity or age structure vector at times t and t + 1), and are calculated through the evolution of the Leslie matrix of the population dynamics model;

[0068] Step S4-4. Prediction result generation and verification:

[0069] (2) Output form: Generate a 72-hour fish stock density heat map (resolution 1km 2 ), mark the high fishing revenue areas and no-fishing areas (such as the spawning and hatching period protection areas), and output the recommended value of the fishing intensity coefficient and the corresponding range of catch fluctuations (such as ±15%);

[0070] It should be noted that for verification and calibration, cross-validation is used to compare historical catch data (such as Argo buoy monitoring data), calculate RMSE (required < 15%) and NSE (required > 0.7), and combine satellite remote sensing data (such as MODIS chlorophyll products) to correct the response deviation of the model to sudden environmental events.

[0071] The decision sub-model intelligently generates a fishing plan based on the calculation results of the fishing operator sub-model, and is specifically as follows:

[0072] Step S5-1. Data fusion and preprocessing: Fusion of the fish stock density distribution (1km×1km grid data) output by the fishing operator sub-model and parameters such as dissolved oxygen and chlorophyll concentration provided by the environmental prediction sub-model to construct a dynamic fishing potential evaluation matrix. Through data cleaning and standardization processing, outliers (such as areas with dissolved oxygen < 3mg / L) are removed, and valid fishing ground data is retained;

[0073] Step S5-2. Data correlation analysis: Using spatio-temporal interpolation technology, convert discrete fish stock trajectory point data into a continuous density field, and combine the tidal cycle to predict the hot spots of fish stock aggregation in the next 24 hours;

[0074] Step S5-3. Dynamic fishing strategy optimization: According to the sustainable threshold formula, calculate the maximum allowable fishing intensity of each grid in real time, and preferentially reduce the fishing intensity in high-density areas through the gradient descent method to avoid the risk of overfishing;

[0075] Step S5-4. Identification and recommendation: Based on the fish stock age structure (such as the proportion of 4-year-old fish ≥ 30%) real-time identified by the YOLOv9 target detection system, dynamically recommend the mesh size (130mm / 160mm) to reduce the juvenile fish bycatch rate (decrease ≥ 27%);

[0076] Step S5-5. Construct a dual objective function: max (economic benefit) and min (ecological disturbance index), generate a Pareto optimal solution set through a non-dominated sorting genetic algorithm (NSGA-II), and combine the environmental prediction sub-model (72-hour prediction accuracy>85%) to avoid high-risk operation periods;

[0077] Step S5-6. Generate a dynamic fishing heat map: mark the priority operation areas and fishing ban areas, and output the recommended fishing intensity value and the expected catch fluctuation range (confidence level 95%).

[0078] Step S2. Ecological protection and optimization: Set dynamic thresholds and fishing ban rules, monitor environmental data in real time, and embed data and fishing policy rules into the calculation model to achieve strategy adaptation in different water scenarios;

[0079] It should be noted that the dynamic threshold and fishing ban rules in step S2 are as follows: setting a catch threshold based on the age structure of the fish school, real-time monitoring of fish specifications through AI recognition technology, triggering the escape channel for young fish, and combining the large fishery model to predict the resource recovery cycle and dynamically delineate the fishing ban area.

[0080] Step S3. Launching and equipment arrangement: The fishing boat is launched and sails, the fishing calculation model is started for the first time, it sails to the designated area according to the calculation result, and the detection system is released to detect the position of the fish school;

[0081] It should be noted that in step S3, the detection system uses an intelligent fish finding boat, and a multimodal perception system is arranged on the intelligent fish finding boat. The multimodal perception system includes an integrated sonar (to detect the position and size of fish schools), an underwater camera (to identify fish species and behavior patterns), and a water quality sensor (to monitor dissolved oxygen, temperature, salinity, etc.).

[0082] Step S4. Path planning and area division: The detection data is input into the fishing calculation model, and the model predicts the future trajectory data of the fish school. The fishing area and driving path are divided according to the trajectory data:

[0083] It should be noted that, in step S4, the fishing area is preferably in shallow water, and the driving path is preferably the shortest and safest route.

[0084] Step S5. Driving and trapping fish schools: The fishing vessel releases the trapping device, which lures the fish schools to move, and combines the electronic pulse driving technology to concentrate the fish schools to the fishing area for fishing through an S-shaped navigation path.

[0085] It should be noted that during trapping in step S5, multiple fishing boats are coordinated based on edge computing devices to form an encirclement net to improve cluster fishing efficiency.

[0086] In the above technical solution, an intelligent fishing method provided by the present invention for effectively improving fishing efficiency has the following beneficial effects:

[0087] (1) Based on the fish swarm operator model and the environmental prediction sub-model, and combining satellite remote sensing and sonar detection, the positioning error of the fish swarm aggregation area is greatly reduced, which can shorten the optimization time of the fishing boat's navigation path, significantly increase the catch per unit time, and also improve the fish swarm concentration efficiency through the S-shaped navigation path and the electronic pulse driving technology. Cooperating with the edge computing device to coordinate multiple boats to form an encirclement net, the success rate of cluster fishing is greatly improved.

[0088] (2) The present invention introduces a fish swarm age structure model and a sustainable fishing formula to ensure a decrease in the bycatch rate of juvenile fish, and real-time monitors parameters such as dissolved oxygen and water temperature to trigger a fishing ban warning in high-risk areas, reducing the ecological disturbance index to the international standard threshold.

[0089] (3) The present invention integrates multi-source data, generates a Pareto optimal solution set through the NSGA-II algorithm, realizes the dual-objective optimization of economic benefits and ecological costs, and is based on Monte Carlo simulation and Kalman filter correction. The model is incrementally updated every 24 hours to adapt to sudden environmental events such as ocean current mutations.

[0090] (4) Through an intelligent and data-driven technical system, the present invention realizes the dual goals of "precision fishing" and "ecological protection", and the economic benefits, sustainability, and technical universality all reach the leading level in the industry, providing a systematic solution for the transformation and upgrading of modern fisheries. Description of the Drawings

[0091] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0092] Figure 1 It is a flowchart of the method provided by an embodiment of an intelligent fishing method for effectively improving fishing efficiency of the present invention.

[0093] Figure 2 It is a schematic diagram of the steps for constructing a fishing calculus model provided by an embodiment of an intelligent fishing method for effectively improving fishing efficiency of the present invention.

[0094] Figure 3 It is a schematic diagram of the calculation of the bottom fish swarm operator model provided by an embodiment of an intelligent fishing method for effectively improving fishing efficiency of the present invention.

[0095] Figure 4Schematic diagram of the environmental prediction sub-model calculation for an embodiment of an intelligent fishing method for effectively improving fishing efficiency according to the present invention.

[0096] Figure 5 Schematic diagram of the fishing operator sub-model calculation for an embodiment of an intelligent fishing method for effectively improving fishing efficiency according to the present invention.

[0097] Figure 6 Schematic diagram of the decision-making steps of the decision sub-model for an embodiment of an intelligent fishing method for effectively improving fishing efficiency according to the present invention. Detailed implementation manner

[0098] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0099] As Figure 1-6 shown, an intelligent fishing method for effectively improving fishing efficiency provided by an embodiment of the present invention includes:

[0100] Step S1. Construct a fish school operator sub-model: collect historical fish school data, environmental data and fishing data, and construct a fishing calculation model based on the data;

[0101] It should be noted that the historical fish school data includes fish school species data and biological behavior interaction data (group survival strategies: (1) Schooling behavior: maintaining the group structure through separation, alignment and cohesion rules, reducing the risk of being preyed upon (for example, when the density of sardine schools reaches 100 fish / m 3 , the probability of being preyed upon decreases by 60%); (2) Foraging and migration: fish schools migrate periodically according to the food distribution (for example, the daily vertical migration depth of tuna can reach 500 meters). Predator-prey relationships: (1) The active areas of top predators (such as sharks) will force fish schools to disperse or change their movement routes; (2) Fish schools interfere with the attack accuracy of predators through the "confusion effect" (such as synchronous turning)). The environmental data includes water quality and temperature data (dissolved oxygen concentration and water temperature directly affect the metabolism and activity of fish schools, low oxygen or extreme temperatures will cause fish schools to migrate to suitable areas, and chlorophyll concentration reflects the abundance of plankton, and high-concentration areas attract fish schools to gather and forage), water flow and terrain data (strong ocean currents will disperse the fish school structure, while slow-flow areas (such as the leeward side of underwater mountains) often form fish school shelters, and underwater terrain (such as reefs, trenches) changes the water flow direction, guiding fish schools to form specific migration paths), and the fishing data includes fishing pressure data (industrial trawling operations lead to the adaptive evolution of fish school behavior, increasing the probability of dispersed movement to avoid fishing nets, sonar detection (frequency 20-200 kHz) and strong light irradiation will cause the stress escape reaction of fish schools), and ecological protection measure data (the closed fishing season policy forces fish schools to concentrate in protected areas during the spawning period, and intelligent sorting devices reduce fishing interference and maintain the natural structure of fish schools).

[0102] The fishing calculus model is divided into a fish school operator sub-model, an environmental prediction sub-model, a fishing operator sub-model, and a decision-making sub-model. The construction steps of the fishing calculus model are as follows:

[0103] Step S1-1. Data acquisition and processing: Obtain historical fish school data, environmental data, and fishing data through the network, artificial communication, and government announcements, convert the data into a unified format, and then perform feature extraction on the data to obtain key information in the data. The key information includes fish species information, trajectory information, fish age information, quantity information, location information, etc.

[0104] Step S1-2. Model construction and training: Select a deep learning network architecture, input the key information to construct a model, then set parameters and construct a loss function, and optimize the model according to the loss function. The FocalLoss is combined for class tasks to alleviate the class imbalance problem (such as rare fish species identification).

[0105] It should be noted that the CIoULoss (Complete Intersection over Union) is used for the loss function in the object detection task to replace the traditional MSE and improve the bounding box regression accuracy.

[0106] (CompleteIntersectionoverUnion) replaces the traditional MSE to improve the bounding box regression accuracy.

[0107] Step S1-3. Verification and deployment: Divide the test set to verify the generalization ability, focus on complex scenarios, use TensorRT to quantize the model, compress the model volume to 1 / 4 of the original size to improve the running efficiency of edge devices, and then optimize the computational graph structure for embedded devices, remove redundant layers, and merge operators.

[0108] The fish school operator sub-model is used to calculate and predict fish school data. The specific steps are as follows:

[0109] Step S2-1. Model initialization and parameter setting:

[0110] (3) Initialization of fish school state: Set the fish school scale and initial distribution range, the position vector and fitness value carried by each individual in the fish school, and then define the environmental parameter space, including water temperature gradient, dissolved oxygen concentration, and food density distribution field.

[0111] (2) Configuration of behavior rules: Foraging behavior: Individuals move to neighborhoods with higher fitness, simulating the gradient ascent process, and the step size is dynamically adjusted; Aggregation behavior: Calculate the local density threshold, and when the number of individuals within the sensing radius exceeds the threshold, group aggregation is triggered to reduce the dispersion.

[0112] Step S2-2. Dynamic behavior simulation and interaction:

[0113] (1) Multi-behavior collaborative operation: Tail-chasing behavior: An individual tracks the neighboring individual with the highest fitness. If the target area is not oversaturated, it updates its position, otherwise it performs a random walk;

[0114] Random behavior: introduce position disturbance with probability 0.1 to prevent the algorithm from falling into the local optimal solution;

[0115] (2) Environmental response mechanism: Encoding marine environmental parameters as fitness function weights to drive fish migration to high-resource areas;

[0116] Step S2-3. Prediction model optimization and iteration:

[0117] (1) Hybrid model fusion: The fish school algorithm is combined with the BP neural network: the fish school optimizes the initial weight value of the neural network to improve the convergence speed, introduces the convolution layer to process the spatial distribution data, and extracts the spatiotemporal characteristics of the fish school's movement trajectory;

[0118] (2) Parameter adaptive adjustment: Dynamically adjust the behavior rule parameters based on historical prediction errors and update the network weights through back propagation;

[0119] Step S2-4. Generate prediction results:

[0120] (3) Data output form: Generate fish density heat map and migration path probability distribution map, with time granularity down to the hour level;

[0121] It should be noted that the cross-validation method was used for verification and calibration, with 80% of the historical observation data used for training and 20% used for testing. The root mean square error (RMSE) was calculated to be <15%. The satellite remote sensing monitoring data was compared to correct the model's response deviation to sudden environmental changes (such as ocean current mutations).

[0122] The environmental prediction sub-model is used to predict future changes in the fishing environment, and the specific steps are as follows:

[0123] Step S3-1. Model initialization and parameter setting:

[0124] (1) Definition of environmental parameters: core environmental variables were set, including water temperature (daily variation ±2°C), dissolved oxygen concentration (threshold range 4-8 mg / L) and chlorophyll concentration (associated with plankton abundance). The initial parameter distribution field was constructed in combination with historical monitoring data (such as satellite remote sensing and sensor networks), with a spatial resolution of no less than 1 km × 1 km.

[0125] (2) Dynamic driving mechanism configuration: Encode meteorological data (such as wind speed and air pressure) and ocean hydrological parameters (such as ocean current speed and salinity gradient) as driving factors of environmental evolution, and define the coupling relationship between parameters, such as the nonlinear response formula for the decrease in dissolved oxygen solubility due to rising water temperature;

[0126] Step S3-2. Dynamic environmental parameter simulation:

[0127] (1) Numerical model solution: Discretize partial differential equations using the finite difference method or the finite volume method to simulate the spatio-temporal evolution of water quality parameters (such as the dissolved oxygen diffusion process), and then introduce a random perturbation term (such as Gaussian noise) to characterize environmental uncertainty and enhance the model's prediction ability for emergencies;

[0128] (2) Multi-time scale prediction:

[0129] Short-term prediction (24 - 72 hours): Update the initial field of the model based on real-time data assimilation technology, and the prediction accuracy error < 10%;

[0130] Long-term trend analysis (monthly / yearly): Combine climate model outputs (such as the ENSO index) to evaluate the periodic variation law of environmental parameters;

[0131] Step S3-3. Multi-source data fusion and model optimization:

[0132] (1) Heterogeneous data integration: Integrate satellite remote sensing data (such as MODIS chlorophyll products), real-time monitoring data from buoy sensors, and historical fishing records of fishing vessels, and use the Kalman filter or particle filter algorithm to correct the prediction deviation of the model to improve data consistency;

[0133] (2) Parameter sensitivity analysis: Identify the dominant environmental variables through the Morris screening method or the Sobol index method (such as the contribution of water temperature to fish migration > 60%), and dynamically adjust the model weights to preferentially optimize the prediction accuracy of highly sensitive parameters;

[0134] Step S3-4. Prediction result generation and risk assessment:

[0135] (1) Output form: Generate a curve of environmental parameter changes in the next 72 hours (such as the predicted daily decline of dissolved oxygen concentration) and a spatial heat map, mark high-risk areas (such as waters with dissolved oxygen < 4 mg / L), and trigger a fishing operation warning;

[0136] (2) Ecological impact assessment: Input the prediction results into the fish population sub-model to evaluate the impact of environmental changes on fish population distribution (such as a 20% decrease in aggregation) and resource quantity (such as a ±15% fluctuation in biomass), and combine the sustainable fishing threshold (such as the maximum allowable fishing intensity coefficient k = 0.3) to formulate a dynamic fishing strategy;

[0137] It should be noted that the subsequent model verification and iterative update are specifically as follows:

[0138] (1) Cross-validation mechanism: Divide the training set (80% historical data) and the test set (20%), calculate the root mean square error (RMSE < 8%) and the Nash efficiency coefficient (NSE > 0.75), then compare with the actual observed data (such as Argo float profile data), correct the systematic bias of the model, deploy an incremental learning framework, update the model parameters every 24 hours, adapt to the gradual and sudden changes of environmental parameters, and combine with reinforcement learning strategies to dynamically adjust the model complexity and computational resource allocation (such as grid encryption area focusing on high-gradient change areas).

[0139] The fishing operator sub-model dynamically simulates the actions of fish schools based on the fish school trajectory operator sub-model and the environmental prediction sub-model, and the specific steps are as follows:

[0140] Step S4-1. Initialize and set parameters:

[0141] (5) Divide the fish school age groups (1 to 4 years old) according to the calculation results of the fish school operator sub-model, define the initial quantity distribution and biological parameters of each age group (natural mortality rate 0.8 / year, spawning quantity gradient, etc.), and set the fishing strategy constraint conditions: for example, only allow fishing of 3-year-old and 4-year-old fish, and the fishing intensity coefficient is related to the mesh size (such as 13mm mesh);

[0142] (6) Based on the dynamic parameters (water temperature, dissolved oxygen, chlorophyll concentration) output by the environmental prediction sub-model, construct a 1km×1km grid environmental field, and map the environmental parameters into fish school behavior driving factors (such as high chlorophyll areas are marked as priority foraging areas);

[0143] Step S4-2. Dynamic simulation and interactive operation:

[0144] (7) Run the fish school trajectory operator sub-model: Simulate the change of the age group quantity based on the Leslie matrix, and the formula is:

[0145] N t+1 =A·N t -C(k)·N t

[0146] where A is the transfer matrix containing the reproduction rate and survival rate, C(k) is the fishing intensity coefficient matrix, and N t represents the age structure vector of the fish school at time t, usually a column vector, and N t+1 represents the age structure vector of the fish school at time t + 1, which evolves from the population state and dynamic process at the current moment. Then, the individual migration path can be adjusted by combining fish behavior rules (foraging, schooling, following), and the migration step size is positively correlated with the environmental gradient (such as food density);

[0147] (8) Driven by the environmental prediction sub-model: The finite volume method is used to predict environmental parameters (such as changes in dissolved oxygen concentration) in the next 72 hours. The model input is updated in real time through data assimilation technology. For the triggering of environmental mutation warnings (such as areas with dissolved oxygen < 4mg / L), the migration path of the fish population is dynamically adjusted to avoid risk areas;

[0148] (9) Dynamically adjusting fishing intensity: According to the density distribution of fish populations of different ages, optimize the fishing intensity coefficient k to meet the constraint conditions:

[0149] ∑(K n ·N n ) ≤ sustainable threshold

[0150] Among them, K n represents the fishing intensity of fish at age n, N n represents the number of individuals in the fish population at age n. n is the minimum age at which fishing is only allowed. The sustainable threshold is the upper limit of the allowable maximum catch. Exceeding this value will lead to an unsustainable population size, and the value is determined by the environmental carrying capacity and the intrinsic growth rate of the population;

[0151] Step S4-3. Multi-model coupling and data fusion:

[0152] (5) Cross-model interaction mechanism: The fish trajectory evolution sub-model outputs the probability distribution of the migration path, and the environmental prediction sub-model provides the dynamic environmental field. The two are corrected through Kalman filtering, and Monte Carlo sampling is introduced to simulate the impact of random events (such as sudden changes in ocean currents) on the fish population distribution, generating multi-scenario prediction results;

[0153] (6) Calculate the biomass of the fish population per unit area (such as the average mass of 4-year-old fish is 22.99g / fish), generate a catch prediction curve in combination with the fishing intensity, and optimize the fishing strategy through the dynamic programming algorithm to ensure the stability of the fish population. The objective function is:

[0154] max (economic benefits) S.T N t+1 ≥N t

[0155] Among them, max (economic benefits) maximizes the economic benefits of fishing activities under sustainable constraints. N t+1 , N t represent the fish population quantity or age structure vector at times t and t + 1 respectively), and are calculated through the evolution of the Leslie matrix of the population dynamics model;

[0156] Step S4-4. Generation and verification of prediction results:

[0157] (3) Output form: Generate a 72-hour fish population density heat map (resolution 1km 2) Mark the areas with high fishing yields and no-fishing areas (such as spawning and hatching period protection areas), and output the recommended values of fishing intensity coefficients and the corresponding ranges of catch fluctuations (such as ±15%);

[0158] It should be noted that for verification and calibration, cross-validation is used to compare historical catch data (such as Argo buoy monitoring data), calculate RMSE (requirement <15%) and NSE (requirement >0.7), and combine satellite remote sensing data (such as MODIS chlorophyll products) to correct the response bias of the model to sudden environmental events.

[0159] The decision-making sub-model intelligently generates a fishing plan based on the calculation results of the fishing operator sub-model, and is specifically as follows:

[0160] Step S5-1. Data fusion and preprocessing: Fusion of the fish density distribution (1km×1km grid data) output by the fishing operator sub-model and parameters such as dissolved oxygen and chlorophyll concentration provided by the environmental prediction sub-model to construct a dynamic fishing potential evaluation matrix. Through data cleaning and standardization processing, outliers (such as areas with dissolved oxygen <3mg / L) are removed, and valid fishing ground data are retained;

[0161] Step S5-2. Data correlation analysis: Using spatio-temporal interpolation technology, convert discrete fish trajectory point data into a continuous density field, and combine the tidal cycle to predict the hot spots of fish aggregation in the next 24 hours;

[0162] Step S5-3. Optimization of dynamic fishing strategies: According to the sustainable threshold formula, calculate the maximum allowable fishing intensity of each grid in real time, and preferentially reduce the fishing intensity in high-density areas through the gradient descent method to avoid the risk of overfishing;

[0163] Step S5-4. Identification and recommendation: Based on the fish age structure (such as the proportion of 4-year-old fish ≥30%) real-time identified by the YOLOv9 target detection system, dynamically recommend the mesh size (130mm / 160mm) to reduce the juvenile fish bycatch rate (decrease ≥27%);

[0164] Step S5-5. Construct a bi-objective function: max (economic benefit) and min (ecological disturbance index), generate a Pareto optimal solution set through the non-dominated sorting genetic algorithm (NSGA-II), and combine the environmental prediction sub-model (72-hour prediction accuracy >85%) to avoid high-risk operation periods;

[0165] Step S5-6. Generate a dynamic fishing heat map: Mark the priority operation areas and no-fishing areas, and output the recommended fishing intensity values and the expected catch fluctuation range (confidence level 95%).

[0166] Step S2. Ecological protection and optimization: Set dynamic thresholds and fishing ban rules, monitor environmental data in real time, and embed data and fishing policy rules into the calculation model to achieve strategy adaptation in different water scenarios;

[0167] It should be noted that the dynamic thresholds and fishing ban rules in step S2 are: setting a catch threshold based on the age structure of the fish population, real-time monitoring of fish specifications through AI recognition technology, triggering escape channels for young fish, and combining the large fishery model to predict resource recovery cycles and dynamically delineate fishing ban areas.

[0168] Step S3. Launching and equipment arrangement: The fishing boat is launched and sails, the fishing calculation model is started for the first time, it sails to the designated area according to the calculation result, and the detection system is released to detect the position of the fish school;

[0169] It should be noted that in step S3, the detection system uses an intelligent fish finding boat, and a multimodal perception system is arranged on the intelligent fish finding boat. The multimodal perception system includes an integrated sonar (to detect the position and size of fish schools), an underwater camera (to identify fish species and behavior patterns), and a water quality sensor (to monitor dissolved oxygen, temperature, salinity, etc.).

[0170] Step S4. Path planning and area division: The detection data is input into the fishing calculation model, and the model predicts the future trajectory data of the fish school. The fishing area and driving path are divided according to the trajectory data:

[0171] It should be noted that in step S4, the fishing area is preferably in shallow water, and the driving path is preferably the shortest and safest route.

[0172] Step S5. Driving and trapping fish schools: The fishing vessel releases the trapping device, which lures the fish schools to move, and combines the electronic pulse driving technology to concentrate the fish schools to the fishing area for fishing through an S-shaped navigation path.

[0173] It should be noted that during trapping in step S5, multiple fishing boats are coordinated based on edge computing devices to form an encirclement net to improve cluster fishing efficiency.

[0174] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An intelligent fishing method for effectively improving fishing efficiency, characterized in that, Including: Step S1. Construct a fish swarm operator model: Collect historical fish swarm data, environmental data, and fishing data, and construct a fishing calculus model based on the data. Step S2. Ecological protection and optimization: Set dynamic thresholds and fishing ban rules, and continuously monitor environmental data. Embed the data and fishing policy rules into the calculus model to achieve strategy adaptation for different water area scenarios. Step S3. Launch the vessel and equipment deployment: Launch the fishing vessel and sail. Initiate the first calculation of the fishing calculus model. Navigate to the designated area according to the calculation results, and release the detection system to detect the position of the fish swarm. Step S4. Route planning and area division: Input the detection data into the fishing calculus model. The model predicts the future trajectory data of the fish swarm. Divide the fishing area and the driving route based on the trajectory data. Step S5. Fish swarm driving and trapping: The fishing vessel releases trapping equipment. The trapping equipment lures the fish swarm to move, and combines with the electronic pulse driving technology to concentrate the fish swarm to the fishing area for fishing through an S-shaped sailing route.

2. The intelligent fishing method for effectively improving fishing efficiency according to claim 1, characterized in that The historical fish swarm data includes fish swarm species data and biological behavior interaction data. The environmental data includes water quality and temperature data, water flow and terrain data. The fishing data includes fishing pressure data and ecological protection measure data.

3. An intelligent fishing method for effectively improving fishing efficiency according to claim 2, characterized in that, In step S1, the fishing calculus model is divided into a fish swarm operator model, an environmental prediction sub-model, a fishing operator model, and a decision-making sub-model. The construction steps of the fishing calculus model are as follows: Step S1-1. Data acquisition and processing: Obtain historical fish swarm data, environmental data, and fishing data through the network, manual communication, and government announcements. Convert the data into a unified format, and then perform feature extraction on the data to obtain the key information in the data. Step S1-2. Model construction and training: Select a deep learning network architecture, input the key information to construct a model, then set parameters and construct a loss function, and optimize the model according to the loss function. Step S1-3. Verification and deployment: Divide the test set to verify the generalization ability, focusing on complex scenarios. Use TensorRT to quantize the model, compress the model volume to 1 / 4 of the original size, improve the operation efficiency of edge devices, and then optimize the computational graph structure for embedded devices, remove redundant layers and merge operators.

4. An intelligent fishing method for effectively improving fishing efficiency according to claim 3, characterized in that, The fish swarm operator model is used to calculate and predict fish swarm data, and the specific steps are as follows: Step S2-1. Model initialization and parameter setting: (1) Initialization of fish swarm state: Set the fish swarm size and initial distribution range, the position vector and fitness value carried by each individual in the fish swarm, and then define the environmental parameter space, including water temperature gradient, dissolved oxygen concentration, and food density distribution field. (2) Configuration of behavior rules: Foraging behavior: Individuals move to neighborhoods with higher fitness, simulating the gradient ascent process, and the step size is dynamically adjusted. Swarming behavior: Calculate the local density threshold, and trigger group aggregation when the number of individuals within the sensing radius exceeds the threshold to reduce the dispersion. Step S2-2. Dynamic behavior simulation and interaction: (1) Cooperative operation of multiple behaviors: Following behavior: Individuals track the neighboring individual with the highest fitness. If the target area is not oversaturated, update the position, otherwise perform a random walk. Random behavior: Introduce position perturbations with a probability of 0.1 to prevent the algorithm from falling into local optimal solutions; (2) Environmental response mechanism: Encode ocean environmental parameters as weights of the fitness function to drive the fish school to migrate towards high-resource areas; Step S2-3. Optimization and iteration of the prediction model: (1) Hybrid model fusion: Combine the fish school algorithm with the BP neural network: The fish school algorithm optimizes the initial values of the neural network weights to improve the convergence speed. Introduce a convolutional layer to process spatially distributed data and extract spatio-temporal features of the fish school movement trajectory; (2) Parameter adaptive adjustment: Dynamically adjust the behavior rule parameters based on historical prediction errors and update the network weights through backpropagation; Step S2-4. Generation of prediction results: (1) Data output form: Generate a heat map of fish school density and a probability distribution map of migration paths, and the time granularity can be refined to the hourly level.

5. The intelligent fishing method for effectively improving fishing efficiency according to claim 4, wherein, The environmental prediction sub-model is used to predict future changes in the fishing environment, and the specific steps are as follows: Step S3-1. Model initialization and parameter setting: (1) Definition of environmental parameters: Set core environmental variables, including key indicators such as water temperature, dissolved oxygen concentration, and chlorophyll concentration; and construct an initial parameter distribution field in combination with historical monitoring data, with a spatial resolution of not less than 1 km × 1 km; (2) Configuration of the dynamic driving mechanism: Encode meteorological data and ocean hydrological parameters as driving factors for environmental evolution, and define the coupling relationship between parameters, such as a non-linear response formula for the decrease in dissolved oxygen solubility caused by an increase in water temperature; Step S3-2. Simulation of dynamic environmental parameters: (1) Numerical model solution: Discretize partial differential equations using the finite difference method or the finite volume method to simulate the spatio-temporal evolution of water quality parameters. Then introduce a random perturbation term to represent environmental uncertainty and enhance the model's prediction ability for emergencies; (2) Multi-time scale prediction: Short-term prediction: Update the model initial field based on real-time data assimilation technology, and the prediction accuracy error < 10%; Long-term trend analysis: Combine climate model outputs to evaluate the periodic change patterns of environmental parameters; Step S3-3. Multi-source data fusion and model optimization: (1) Heterogeneous data integration: Integrate satellite remote sensing data, real-time monitoring data from buoy sensors, and historical fishing records of fishing vessels, and use the Kalman filter or particle filter algorithm to correct the model prediction deviation and improve data consistency; (2) Parameter sensitivity analysis: Identify dominant environmental variables through the Morris screening method or the Sobol index method, and dynamically adjust the model weights to preferentially optimize the prediction accuracy of highly sensitive parameters; Step S3-4. Generation of prediction results and risk assessment: (1) Output form: Generate curves of environmental parameter changes and spatial heat maps for the next 72 hours, mark high-risk areas, and trigger fishing operation warnings; (2) Ecological impact assessment: Input the prediction results into the fish school evolution sub-model to evaluate the impact of environmental changes on fish school distribution and resource quantity, and combine with the sustainable fishing threshold to formulate dynamic fishing strategies.

6. The intelligent fishing method for effectively improving fishing efficiency according to claim 5, characterized in that The fishing operation sub-model dynamically simulates the actions of the fish school based on the fish school trajectory sub-model and the environmental prediction sub-model, and the specific steps are as follows: Step S4-1. Initialization and parameter setting: (1) Divide the fish age groups according to the calculation results of the fish swarm operator model, define the initial quantity distribution and biological parameters of each age group, and set the fishing strategy constraint conditions; (2) Based on the dynamic parameters output by the environmental prediction sub-model, construct a 1km×1km grid environmental field, and map the environmental parameters to the fish behavior driving factors; Step S4-2. Dynamic simulation and interactive operation: (1) Operation of the fish swarm trajectory operator model: Simulate the change of the quantity of age groups based on the Leslie matrix, and the formula is: N t+1 = A·N t - C(k)·N t Among them, A is a transition matrix containing the reproduction rate and survival rate, C(k) is a fishing intensity coefficient matrix, and N t represents the age structure vector of the fish population at time t, usually a column vector, N t+1 represents the age structure vector of the fish population at time t + 1, which evolves from the population state and dynamic process at the current moment. Subsequently, the individual migration path can be adjusted in combination with the fish behavior rules, and the migration step size is positively correlated with the environmental gradient; (2) Driving by the environmental prediction sub-model: Use the finite volume method to predict the environmental parameters in the next 72 hours, update the model input in real time through data assimilation technology, trigger an early warning for environmental mutations, and dynamically adjust the fish migration path to avoid risk areas; (3) Dynamically adjust the fishing intensity: According to the density distribution of the age fish swarm, optimize the fishing intensity coefficient k to meet the constraint conditions: ∑(K n ·N n ) ≤ Sustainable threshold Among them, K n represents the fishing intensity at the age of n fish, N n represents the number of individuals in the fish population at the age of n fish. n is the minimum fish age that is only allowed to be fished. The sustainable threshold is the upper limit of the allowable maximum fishing volume. Exceeding this value will lead to the unsustainability of the population quantity, and the value is determined by the environmental carrying capacity and the intrinsic growth rate of the population; Step S4-3. Multi-model coupling and data fusion: (1) Cross-model interaction mechanism: The fish swarm trajectory operator model outputs the probability distribution of the migration path, and the environmental prediction sub-model provides the dynamic environmental field. The two perform data correction through the Kalman filter, introduce Monte Carlo sampling to simulate the influence of random events on the fish swarm distribution, and generate multi-scenario prediction results; (2) Calculate the fish biomass per unit area, generate a catch prediction curve in combination with the fishing intensity, and optimize the fishing strategy through the dynamic programming algorithm to ensure the stability of the fish population. The objective function is: max(Economic benefits) S.T N t+1 ≥N t Among them, max(economic benefits) maximizes the economic benefits of fishing activities under sustainable constraints, N t+1 , N t respectively represent the fish population quantity or age structure vector at time t and t + 1), and are calculated through the evolution of the Leslie matrix of the population dynamics model; Step S4-4. Generation of prediction results: (1) Output form: Generate a 72-hour fish density heat map, mark the high fishing revenue areas and no-fishing areas, and output the recommended value of the fishing intensity coefficient and the corresponding catch fluctuation range.

7. An intelligent fishing method for effectively improving fishing efficiency according to claim 6, characterized in that, The decision-making sub-model intelligently generates a fishing plan based on the calculation results of the fishing operator model, and specifically as follows: Step S5-1. Data fusion and preprocessing: Fusion of the fish density distribution output by the fishing operator model and parameters such as dissolved oxygen and chlorophyll concentration provided by the environmental prediction sub-model, construct a dynamic fishing potential evaluation matrix, and through data cleaning and standardization processing, eliminate outliers and retain valid fishing ground data; Step S5-2. Data correlation analysis: Use spatio-temporal interpolation technology to convert discrete fish swarm trajectory point data into a continuous density field, and combine the tidal cycle to predict the hot spots of fish swarm aggregation in the next 24 hours; Step S5-3. Optimization of the dynamic fishing strategy: According to the sustainable threshold formula, calculate the maximum allowable fishing intensity of each grid in real time, and preferentially reduce the fishing intensity in the high-density area through the gradient descent method to avoid the risk of overfishing; Step S5-4. Identification and recommendation: Based on the fish age structure real-time identified by the YOLOv9 object detection system, dynamically recommend the mesh size to reduce the juvenile fish bycatch rate; Step S5-5. Construct a double-objective function: max (economic benefit) and min (ecological disturbance index), generate a Pareto optimal solution set through the NSGA-II non-dominated sorting genetic algorithm, and combine the environmental prediction sub-model to avoid high-risk operation periods; Step S5-6. Generate a dynamic fishing heat map: Mark the priority operation areas and no-fishing areas, and output the recommended value of the fishing intensity and the expected catch fluctuation range.

8. An intelligent fishing method for effectively improving fishing efficiency according to claim 1, characterized in that, The dynamic threshold and fishing ban rules in step S2: Set the fishing volume threshold based on the age structure of the fish stock, use AI recognition technology to monitor the fish body size in real time, trigger the escape channel for juvenile fish, and combine the fishery large model to predict the resource recovery period, and dynamically delimit the fishing ban area. In step S3, the detection system selects an intelligent fish finder boat, and a multi-modal perception system is arranged on the intelligent fish finder boat. The multi-modal perception system includes an integrated sonar, an underwater camera, and a water quality sensor.

9. An intelligent fishing method for effectively improving fishing efficiency according to claim 1, characterized in that, In step S4, the fishing area is preferably in shallow water, and the driving path is preferably the shortest and safest route.

10. An intelligent fishing method for effectively improving fishing efficiency according to claim 1, characterized in that, In step S5, during trapping, multiple fishing boats are coordinated based on the edge computing device to form an encirclement net to improve the efficiency of cluster fishing.

Images