Shrimp fry hatching method and system based on multi-modal data fusion

Abstract

This invention relates to a method and system for hatching shrimp larvae based on multimodal data fusion, belonging to the field of tiger prawn hatching technology. The invention constructs a reinforcement learning framework based on digital twins and twin model mechanisms, ultimately acquiring image data information of the current target aquaculture area within a preset time period. The reinforcement learning framework is then used to analyze this image data information to find the action sequence that maximizes cumulative rewards. This invention systematically solves the technical challenges of extensive perception, lagging regulation, and experience-dependent decision-making during shrimp larvae hatching through the deep integration of multimodal perception, digital twin modeling, and reinforcement learning decision-making. It significantly improves the survival rate and growth uniformity of larvae during the hatching stage, while reducing the risk of water quality deterioration and energy and material consumption, providing reliable technical support for intelligent aquaculture management.

Description

Technical Field

[0001] This invention relates to the field of technology, and in particular to a method and system for hatching shrimp larvae based on multimodal data fusion. Background Technology

[0002] Shrimp larvae hatching is a crucial initial stage in the aquaculture industry chain, and its hatching quality and survival rate directly affect the economic benefits of subsequent adult shrimp farming. As a high-value aquaculture species, the tiger prawn (Penaeus monodon) is extremely sensitive to environmental factors during its larval hatching process, and traditional hatching management methods have long faced numerous technical bottlenecks.

[0003] Currently, shrimp larvae hatching is mainly managed through manual experience. Technicians observe shrimp larvae activity and water color changes with the naked eye, and manually adjust operational parameters such as feeding amount, water exchange rate, and aeration intensity based on regularly sampled water quality indicators. This model has significant limitations: First, manual observation is subjective and time-sensitive, making it difficult to continuously and quantitatively monitor subtle shrimp larvae behaviors (such as antennal wagging frequency and feeding activity) and early health status (such as gill abnormalities); second, the coupling mechanism between water quality parameters and shrimp larvae growth status is complex, and traditional threshold-based control methods cannot achieve dynamic synergistic optimization of multiple variables; third, the water flow field and shrimp larvae density distribution in the hatching tank exhibit significant spatial heterogeneity, making it difficult for traditional point sensors to comprehensively characterize the system state, leading to control decisions often deviating from the optimal solution.

[0004] In summary, there is an urgent need in this field for a method that can integrate multimodal sensing data, construct a high-fidelity digital twin, and achieve intelligent decision-making in the hatching process based on reinforcement learning, in order to solve the technical problems of insufficient sensing capabilities, lagging regulation, and suboptimal decision-making in traditional shrimp larvae hatching management models. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method and system for hatching shrimp larvae based on multimodal data fusion.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for hatching shrimp larvae based on multimodal data fusion, characterized by comprising the following steps: Collect image data information in the current target aquaculture area, identify the image data information in the current target aquaculture area, and obtain multimodal data information; Obtain design drawing data of the target aquaculture area, and construct a digital twin based on the multimodal data and the design drawing data of the target aquaculture area; Establish the twin model mechanism, and construct a reinforcement learning framework based on the digital twin and the twin model mechanism; The image data information of the current target breeding area within a preset time is obtained, and the image data information of the current target breeding area within the preset time is analyzed using the reinforcement learning framework to find the action sequence that maximizes the cumulative reward.

[0007] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, image data information of the current target aquaculture area is collected, and the image data information of the current target aquaculture area is identified to obtain multimodal data information, specifically as follows: A polarization imaging unit is deployed in the target aquaculture area, using an active LED light source and integrating a linear polarizer. The polarization image is acquired by rotating the polarizer, and the polarization image is processed using a differential polarization processing algorithm to extract the outline of the tiger prawn, the movement of its antennae, and the details of its gills. By performing data augmentation on the polarization image and extracting the enhanced high edge corresponding features, and extracting the cephalothorax and abdominal segments of the tiger prawn, the individual segmented in each frame image is regarded as a node in the image. By introducing temporal information, the displacement of individuals in adjacent frames is tracked using optical flow. When two nodes are found to be stuck together and then separated in consecutive frames, the affiliation of the individuals before sticking together is inferred by the change in edge weights of the graph network. Kernel density estimation is used to map individual locations to heatmaps, outputting density distribution cloud maps of the aquaculture area. At the same time, each node is tracked and identified to complete the multimodal data identification of the developmental stage, health status, feeding behavior, density distribution, and water quality of Penaeus monodon.

[0008] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, the design drawing data information of the target aquaculture area is obtained, and a digital twin is constructed based on the multimodal data information and the design drawing data information of the target aquaculture area, specifically including: Obtain the design drawing data of the target aquaculture area, and construct an initial twin of the aquaculture area based on the design drawing data of the target aquaculture area; A digital twin model, synchronized in real time with the physical incubation pool, is established in the cloud based on the initial twin. Multimodal data on the developmental stage, health status, feeding behavior, density distribution, and water quality of Penaeus monodon were used as input parameters. The digital twin model is dynamically demonstrated using the input parameters to form a digital twin.

[0009] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, a twin model mechanism is established, and a reinforcement learning framework is constructed based on the digital twin and the twin model mechanism, specifically as follows: Computational fluid dynamics was used to simulate the effect of water flow field on the distribution of Penaeus monodon larvae. At the same time, population dynamics model was used to simulate the larval growth curves under different feeding strategies, forming a twin model mechanism. A reinforcement learning framework is constructed based on a deep Q-network, which is used as the decision core to construct a state space, an action space, and a reward function. In the state space, the twin synchronizes the current environmental parameters, developmental stage, health status, and feeding behavior status. In the action space, actions such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic delivery type are configured, and positive rewards such as improved metamorphosis rate, improved survival rate, good feeding activity, and stable water quality are set in the reward function. The deep Q-network sets juvenile mortality, Vibrio outbreaks, excessive ammonia nitrogen, and energy consumption as negative rewards. Before each actual execution, the deep Q-network performs a preset number of simulations in the digital twin to find the action sequence that maximizes the cumulative reward.

[0010] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, image data information of the current target aquaculture area within a preset time is acquired. The reinforcement learning framework is then used to analyze this image data information to find the action sequence that maximizes the cumulative reward. Specifically: Acquire image data information of the current target aquaculture area within a preset time period, and input the image data information of the current target aquaculture area within the preset time period into the reinforcement learning framework for a preset number of simulations; By performing simulations a preset number of times, the system finds the action sequence that maximizes the cumulative reward, outputs action data such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic delivery type, and displays them in a preset manner.

[0011] Furthermore, the shrimp larvae hatching method based on multimodal data fusion also includes: The actual hatching data of tiger prawns within a preset time period are statistically analyzed, and the actual hatching data of tiger prawns within the preset time period is compared with the twin projection data to obtain the deviation rate, and a deviation rate threshold is set. Determine whether the deviation rate is greater than the deviation rate threshold. If the deviation rate is greater than the deviation rate threshold, update the parameters of the deep Q network using the actual data of tiger prawn hatching within the preset time. When the deviation rate is not greater than the deviation rate threshold, the parameters of the deep Q-network remain unchanged.

[0012] A second aspect of the present invention provides a shrimp larvae hatching system based on multimodal data fusion, including a memory and a processor. The memory includes a shrimp larvae hatching method program based on multimodal data fusion. When the shrimp larvae hatching method program based on multimodal data fusion is executed by the processor, it implements the steps of the shrimp larvae hatching method based on multimodal data fusion as described in any one of the present invention.

[0013] A third aspect of the present invention provides a computer-readable storage medium including a shrimp larvae hatching method program based on multimodal data fusion, wherein when the shrimp larvae hatching method program based on multimodal data fusion is executed by a processor, it implements the steps of the shrimp larvae hatching method based on multimodal data fusion as described in any one of the present invention.

[0014] This invention addresses the shortcomings of the prior art and has the following beneficial effects: This invention collects image data from a target aquaculture area, identifies this data to obtain multimodal data, and then acquires design drawings of the target aquaculture area. Based on the multimodal data and the design drawings, a digital twin is constructed, establishing a twin model mechanism. A reinforcement learning framework is then built based on the digital twin and the twin model mechanism. Finally, image data from the target aquaculture area within a preset time period is acquired, and the reinforcement learning framework is used to analyze this data to find the action sequence that maximizes cumulative reward. This invention, through the deep integration of multimodal perception, digital twin modeling, and reinforcement learning decision-making, systematically solves the technical challenges of extensive perception, lagging regulation, and experience-dependent decision-making during shrimp larvae hatching. It significantly improves the survival rate and growth uniformity of larvae during the hatching stage, while reducing the risk of water quality deterioration and energy and material consumption, providing reliable technical support for intelligent aquaculture management. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the overall process of shrimp larvae hatching method based on multimodal data fusion is shown. Figure 2 A system block diagram of a shrimp larvae hatching system based on multimodal data fusion is shown. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] like Figure 1 As shown, the first aspect of the present invention provides a method for hatching shrimp larvae based on multimodal data fusion, comprising the following steps: Collect image data information of the current target breeding area, identify the image data information of the current target breeding area, and obtain multimodal data information; Obtain design drawings and data of the target aquaculture area, and construct a digital twin based on multimodal data and design drawings and data of the target aquaculture area; Establish the mechanism of the twin model and construct a reinforcement learning framework based on the digital twin and the twin model mechanism; Acquire image data of the current target breeding area within a preset time period, and use a reinforcement learning framework to analyze the image data of the current target breeding area within the preset time period to find the action sequence that maximizes the cumulative reward.

[0020] It should be noted that this invention systematically solves the technical problems of extensive perception, lagging regulation, and experience-dependent decision-making during the hatching process of shrimp larvae by deeply integrating multimodal perception, digital twin modeling, and reinforcement learning decision-making. It significantly improves the survival rate and growth uniformity of larvae during the hatching stage, while reducing the risk of water quality deterioration and energy and material consumption, providing reliable technical support for intelligent management of aquaculture.

[0021] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, image data information of the current target aquaculture area is collected, and the image data information of the current target aquaculture area is identified to obtain multimodal data information, specifically: A polarization imaging unit is deployed in the target aquaculture area. It uses an active LED light source and integrates a linear polarizer. The polarization image is acquired by rotating the polarizer. The polarization image is processed using a differential polarization processing algorithm to extract the outline of the tiger prawn, the movement of its antennae, and the details of its gills. By performing data augmentation on polarized images and extracting the enhanced high-edge corresponding features, and extracting the cephalothorax and abdominal segments of the tiger prawn, the individuals segmented in each frame of the image are regarded as nodes in the graph. By introducing temporal information, the displacement of individuals in adjacent frames is tracked using optical flow. When two nodes are found to be stuck together and then separated in consecutive frames, the affiliation of the individuals before sticking together is inferred by the change in edge weights of the graph network. Kernel density estimation is used to map individual locations to heatmaps, outputting density distribution cloud maps of the aquaculture area. At the same time, each node is tracked and identified to complete the multimodal data identification of the developmental stage, health status, feeding behavior, density distribution, and water quality of Penaeus monodon.

[0022] It should be noted that by using kernel density estimation to map individual locations into heat maps and output density distribution cloud maps of the aquaculture area, local over-dense or under-dense areas caused by aerator locations or uneven feeding can be accurately identified. Each node can be tracked and identified through AI recognition technology, thereby identifying the developmental stage, health status, feeding behavior, density distribution, and water quality of the tiger prawn.

[0023] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, design drawing data of the target aquaculture area is obtained, and a digital twin is constructed based on the multimodal data and the design drawing data of the target aquaculture area, specifically including: Obtain the design drawings and data of the target aquaculture area, and construct an initial twin of the aquaculture area based on the design drawings and data of the target aquaculture area; A digital twin model is built in the cloud based on the initial twin and synchronized in real time with the physical incubation pool; Multimodal data on the developmental stage, health status, feeding behavior, density distribution, and water quality of Penaeus monodon were used as input parameters. The digital twin model is dynamically demonstrated using input parameters to form a digital twin.

[0024] It should be noted that the design drawings of the target aquaculture area include data such as length, width, height, and shape features. This invention integrates the design drawings of the target aquaculture area with real-time perceived multimodal data (developmental stage, health status, feeding behavior, density distribution, and water quality parameters) to construct a digital twin model in the cloud that is synchronized in real-time with the physical hatching pond. This twin not only replicates the geometric structure of the aquaculture pond but also, by using multimodal data as dynamic input parameters, achieves real-time mapping and dynamic demonstration of shrimp larvae distribution and growth status, providing a high-precision virtual experimental platform for subsequent simulation and decision optimization.

[0025] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, a twin model mechanism is established, and a reinforcement learning framework is constructed based on the digital twin and the twin model mechanism, specifically as follows: Computational fluid dynamics was used to simulate the effect of water flow field on the distribution of Penaeus monodon larvae. At the same time, population dynamics model was used to simulate the larval growth curves under different feeding strategies, forming a twin model mechanism. A reinforcement learning framework is constructed based on deep Q-networks, with deep Q-networks as the decision core. A state space, action space, and reward function are constructed. In the state space, the twin synchronizes the current environmental parameters, developmental stage, health status, and feeding behavior status. Configure actions such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic delivery type in the action space, and set positive rewards such as improved metamorphosis rate, improved survival rate, good feeding activity, and stable water quality in the reward function. The deep Q-network sets juvenile mortality, Vibrio outbreaks, excessive ammonia nitrogen, and energy consumption as negative rewards. Before each actual execution, the deep Q-network performs a preset number of simulations in the digital twin to find the action sequence that maximizes the cumulative reward.

[0026] It should be noted that this invention establishes a twin model mechanism by coupling computational fluid dynamics simulation and population dynamics model, enabling the quantification of the impact of water flow field distribution on shrimp larvae and growth curves under different feeding strategies. Based on this, a reinforcement learning framework based on a deep Q-network is constructed, using the real-time state of the digital twin as the state space and key operations such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic type as the action space. A multi-objective reward function is designed, encompassing metamorphosis rate, survival rate, feeding activity, water quality stability, and risk events (larval mortality, Vibrio outbreaks, and ammonia nitrogen exceedances). By performing multiple simulations within the digital twin before actual execution, the optimal action sequence that maximizes cumulative rewards is found, achieving a leap from experience-driven, post-event control to data-driven, pre-event optimization, effectively improving the scientific rigor and precision of hatching management.

[0027] Furthermore, in the shrimp larvae hatching method based on multimodal data fusion, image data information of the current target aquaculture area within a preset time is acquired. A reinforcement learning framework is then used to analyze this image data information to find the action sequence that maximizes the cumulative reward. Specifically: Acquire image data information of the current target aquaculture area within a preset time period, and input the image data information of the current target aquaculture area within the preset time period into the reinforcement learning framework for a preset number of simulations; By performing simulations a preset number of times, the system finds the action sequence that maximizes the cumulative reward, outputs action data such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic delivery type, and displays them in a preset manner.

[0028] Furthermore, the shrimp larvae hatching method based on multimodal data fusion also includes: The actual hatching data of tiger prawns within a preset time period are statistically analyzed, and the actual hatching data of tiger prawns within the preset time period is compared with the twin projection data to obtain the deviation rate, and a deviation rate threshold is set. Determine if the deviation rate is greater than the deviation rate threshold. If the deviation rate is greater than the deviation rate threshold, update the parameters of the deep Q network using the actual data of tiger prawn hatching within a preset time. When the deviation rate is not greater than the deviation rate threshold, the parameters of the deep Q-network remain unchanged.

[0029] It should be noted that this invention further statistically analyzes the actual hatching data of tiger prawns within a preset time period and compares it with the twin projection data to calculate the deviation rate. When the deviation rate exceeds a set threshold, the parameters of the deep Q-network are dynamically updated using the actual data. This closed-loop feedback mechanism enables the decision-making model to continuously adapt to uncertainties such as changes in the aquaculture environment and batch differences in shrimp larvae, avoiding model degradation and ensuring decision-making accuracy and system stability under long-term operation.

[0030] In addition, the following steps may be included when identifying the developmental stage of the tiger prawn: A miniature waterproof high-speed microscopic imaging system is deployed in the hatching tank / pond, equipped with a variable focus liquid lens and ring LED supplemental lighting. Multiple sampling points are set in the larval swimming layer, and characteristic images of tiger prawns during the hatching process are collected using the sampling points. To address the characteristics of shrimp larvae being translucent and moving quickly, the Retinex algorithm was used to remove underwater lighting unevenness, and the inter-frame difference method was combined to extract moving targets, reducing interference from background algae or food. An improved YOLOv8-Trans model was constructed, and the Swin Transformer module was introduced into the Backbone network to globally capture the larval appendage features, distinguish the subtle morphological differences between flea larvae stage 2 and flea larvae stage 3, and improve the recognition accuracy of flea larvae stage 2 and flea larvae stage 3. The model outputs the number of individuals and confidence level at different developmental stages in each frame, sets a time window, and counts the population proportion of each stage. When the proportion of a specific reproductive stage exceeds a preset threshold, an abnormality warning is triggered. At the same time, based on the results of the stage judgment, the expert knowledge base is invoked to generate relevant feeding strategies.

[0031] It should be noted that an improved YOLOv8-Trans model was constructed, introducing the SwinTransformer module into the Backbone network to enhance the global correlation capture ability of subtle local features such as larval appendages (e.g., caudal spines, frontal horns), distinguishing the subtle morphological differences between zoea larvae stage 2 and zoea larvae stage 3. Furthermore, the model outputs the number of individuals and confidence scores for different developmental stages (Z1, Z2, Z3, M1, P1, etc.) in each frame. A time window is set to statistically analyze the population proportion of each stage. When the proportion of a specific stage (e.g., Z2 to Z3) exceeds 75%, a metamorphosis warning is triggered. This method can further optimize the accuracy of automatic larval developmental stage identification. Specifically, if the current stage is zoea larvae stage 2 and the density is above a threshold, the system automatically generates an instruction to increase the feeding amount of Artemia nauplii by 20% and supplement with Chlorella for enhanced nutrition. If the current stage is mysid shrimp larvae, the system automatically switches to feeding a mixture of formulated feed and Artemia nauplii in a specific ratio.

[0032] In addition, this method also includes: The metabolic characteristic data of all genotypes of Penaeus monodon that have appeared under different climatic conditions are obtained, and a knowledge graph is constructed based on the metabolic characteristic data of all genotypes of Penaeus monodon that have appeared under different climatic conditions. Obtain historical climate characteristic data of the current target aquaculture area, and perform retrieval using the knowledge graph based on the historical climate characteristic data of the current target aquaculture area; By searching, the metabolic characteristic data of all genotypes that have appeared under the historical climate characteristic data information of the current target breeding area are obtained, and the metabolic characteristic data of all genotypes that have appeared under the historical climate characteristic data information of the current target breeding area are sorted, and the average value of the metabolic characteristic data corresponding to each genotype and the historical climate characteristic data information is calculated. By sorting, the metabolic characteristic data with the highest average value is obtained, and the genotype corresponding to the metabolic characteristic data with the highest average value is used as the priority breeding type for tiger prawns in the current aquaculture area.

[0033] It should be noted that a higher average value indicates a dominant aquaculture advantage in that region for most of the time. This method fully considers climatic factors, enabling the selection of the optimal genotype for cultivation in the current region during the initial selection of tiger prawns, thus optimizing tiger prawn farming. Metabolic characteristic data includes the type and concentration characteristics of metabolites; vigorous metabolism indicates excellent growth of tiger prawns.

[0034] like Figure 2As shown, the second aspect of the present invention provides a shrimp larvae hatching system based on multimodal data fusion, including a memory and a processor. The memory includes a shrimp larvae hatching method program based on multimodal data fusion. When the shrimp larvae hatching method program based on multimodal data fusion is executed by the processor, it implements the steps of any of the shrimp larvae hatching methods based on multimodal data fusion.

[0035] A third aspect of the present invention provides a computer-readable storage medium including a shrimp larvae hatching method program based on multimodal data fusion, wherein when the shrimp larvae hatching method program based on multimodal data fusion is executed by a processor, it implements the steps of any one of the shrimp larvae hatching methods based on multimodal data fusion.

[0036] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0037] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0038] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0039] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0040] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0041] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for hatching shrimp larvae based on multimodal data fusion, characterized in that, Includes the following steps: Collect image data information in the current target aquaculture area, identify the image data information in the current target aquaculture area, and obtain multimodal data information; Obtain design drawing data of the target aquaculture area, and construct a digital twin based on the multimodal data and the design drawing data of the target aquaculture area; Establish the twin model mechanism, and construct a reinforcement learning framework based on the digital twin and the twin model mechanism; The image data information of the current target breeding area within a preset time is obtained, and the image data information of the current target breeding area within the preset time is analyzed using the reinforcement learning framework to find the action sequence that maximizes the cumulative reward.

2. The shrimp larvae hatching method based on multimodal data fusion according to claim 1, characterized in that, Collect image data information from the current target aquaculture area, identify the image data information from the current target aquaculture area, and obtain multimodal data information, specifically: A polarization imaging unit is deployed in the target aquaculture area, using an active LED light source and integrating a linear polarizer. The polarization image is acquired by rotating the polarizer, and the polarization image is processed using a differential polarization processing algorithm to extract the outline of the tiger prawn, the movement of its antennae, and the details of its gills. By performing data augmentation on the polarization image and extracting the enhanced high edge corresponding features, and extracting the cephalothorax and abdominal segments of the tiger prawn, the individual segmented in each frame image is regarded as a node in the image. By introducing temporal information, the displacement of individuals in adjacent frames is tracked using optical flow. When two nodes are found to be stuck together and then separated in consecutive frames, the affiliation of the individuals before sticking together is inferred by the change in edge weights of the graph network. Kernel density estimation is used to map individual locations to heatmaps, outputting density distribution cloud maps of the aquaculture area. At the same time, each node is tracked and identified to complete the multimodal data identification of the developmental stage, health status, feeding behavior, density distribution, and water quality of Penaeus monodon.

3. The shrimp larvae hatching method based on multimodal data fusion according to claim 2, characterized in that, Obtain design drawing data of the target aquaculture area, and construct a digital twin based on the multimodal data and the design drawing data of the target aquaculture area, specifically including: Obtain the design drawing data of the target aquaculture area, and construct an initial twin of the aquaculture area based on the design drawing data of the target aquaculture area; A digital twin model, synchronized in real time with the physical incubation pool, is established in the cloud based on the initial twin. Multimodal data on the developmental stage, health status, feeding behavior, density distribution, and water quality of Penaeus monodon were used as input parameters. The digital twin model is dynamically demonstrated using the input parameters to form a digital twin.

4. The shrimp larvae hatching method based on multimodal data fusion according to claim 1, characterized in that, A twin model mechanism is established, and a reinforcement learning framework is constructed based on the digital twin and the twin model mechanism, specifically as follows: Computational fluid dynamics was used to simulate the effect of water flow field on the distribution of Penaeus monodon larvae. At the same time, population dynamics model was used to simulate the larval growth curves under different feeding strategies, forming a twin model mechanism. A reinforcement learning framework is constructed based on a deep Q-network, which is used as the decision core to construct a state space, an action space, and a reward function. In the state space, the twin synchronizes the current environmental parameters, developmental stage, health status, and feeding behavior status. In the action space, actions such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic delivery type are configured, and positive rewards such as improved metamorphosis rate, improved survival rate, good feeding activity, and stable water quality are set in the reward function. The deep Q-network sets juvenile mortality, Vibrio outbreaks, excessive ammonia nitrogen, and energy consumption as negative rewards. Before each actual execution, the deep Q-network performs a preset number of simulations in the digital twin to find the action sequence that maximizes the cumulative reward.

5. The shrimp larvae hatching method based on multimodal data fusion according to claim 1, characterized in that, The image data of the current target aquaculture area within a preset time period is acquired. The reinforcement learning framework is then used to analyze this image data to find the action sequence that maximizes the cumulative reward. Specifically: Acquire image data information of the current target aquaculture area within a preset time period, and input the image data information of the current target aquaculture area within the preset time period into the reinforcement learning framework for a preset number of simulations; By performing a preset number of simulations, the system finds the action sequence that maximizes the cumulative reward, outputs action data such as feeding amount, water exchange rate, heating rate, oxygenation intensity, and probiotic delivery type, and displays them in a preset manner.

6. The shrimp larvae hatching method based on multimodal data fusion according to claim 5, characterized in that, Also includes: The actual hatching data of tiger prawns within a preset time period are statistically analyzed, and the actual hatching data of tiger prawns within the preset time period is compared with the twin projection data to obtain the deviation rate, and a deviation rate threshold is set. Determine whether the deviation rate is greater than the deviation rate threshold. If the deviation rate is greater than the deviation rate threshold, update the parameters of the deep Q network using the actual data of tiger prawn hatching within the preset time. When the deviation rate is not greater than the deviation rate threshold, the parameters of the deep Q-network remain unchanged.

7. A shrimp larvae hatching system based on multimodal data fusion, characterized in that, The device includes a memory and a processor. The memory includes a program for a shrimp larvae hatching method based on multimodal data fusion. When the processor executes the program for the shrimp larvae hatching method based on multimodal data fusion, it implements the steps of the shrimp larvae hatching method based on multimodal data fusion as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The method includes a shrimp larvae hatching method program based on multimodal data fusion, wherein when the shrimp larvae hatching method program based on multimodal data fusion is executed by a processor, it implements the steps of the shrimp larvae hatching method based on multimodal data fusion as described in any one of claims 1-6.

Images