Automatic feeding system for training and breeding juvenile plectropomus leopardus

Abstract

The invention provides an automatic bait casting system for training and breeding juvenile plectropomus leopardus, which comprises a basic data module connected with an automatic bait training device and used for acquiring original data of a breeding pond and extracting feeding basic data; the dynamic association module is used for carrying out multi-level cross association modeling, causal inference association, sub-mode mining and classification adaptation to obtain an ingestion-turbidity dynamic association model and outputting an ingestion-turbidity relationship; the hunger index module is used for constructing a hunger index model and outputting the hunger index of each culture pond; the feeding strategy module is used for calculating the next feeding interval, the strip-shaped bait-live bait proportion, the actual time point of next feeding and the bait amount per hour and judging whether extra meal or increase of the next feeding interval is needed or not to obtain a self-adaptive feeding strategy; and the feeding driving module is used for controlling the running action of the automatic bait training device. Scientific feeding can be achieved, the feeding survival rate, the growth rate and the bait training effect are remarkably improved, feed waste is reduced, and intelligent and automatic management of the whole process is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of juvenile fish training and feeding, in particular to an automatic feeding system for training juvenile leopard gill bass. Background Art

[0002] The spiny grouper, also known as the eastern grouper, holds a prominent position in marine fisheries and culinary traditions due to its unique appearance and high economic value. Found in the tropical waters of the Indian and western Pacific Oceans, it is named for the tiny, star-like white spots covering its body. In Chaozhou cuisine, cooking the grouper raw is a classic, showcasing its firm, chewy texture. Nutritionally, grouper is rich in over twenty essential trace elements and vitamins, boasting high protein, low fat, and low cholesterol. It nourishes the liver and kidneys, while the "horsebene" in its skin is a beauty ingredient. The "brain gold" in its bones is also crucial for brain development and health. Its ease of digestion and high absorption make it a high-quality ingredient that combines deliciousness with health.

[0003] Among the many species of spiny gill perch, the leopard spiny gill perch is a highly representative species. It primarily inhabits coral-rich lagoons, seaward reefs, and outer reef slopes. These areas are rich in food resources, providing the leopard spiny gill perch with an ample supply of prey. The leopard spiny gill perch is a ferocious and voracious eater, primarily consuming fish and occasionally crustaceans. This unique diet is key to its survival and reproduction in nature, but it also presents significant challenges in captivity.

[0004] Currently, the artificial domestication of leopard gill bass is still in the exploratory stage, especially the bait domestication technology for juveniles is still blank. Traditional artificial exploratory domestication mainly relies on the experience and subjective judgment of the breeder, and has many drawbacks. Breeders need to frequently observe the condition of the juveniles and manually adjust the amount of bait and feeding time. This is not only time-consuming and labor-intensive, but also makes it difficult to ensure that every juvenile fish receives the appropriate food, resulting in uneven domestication results and extremely low efficiency. In this context, the development and application of automatic bait training devices have become a key direction to solve the problem, but existing devices have obvious defects in automation and intelligence.

[0005] At the automation level, the mechanical structure and control system of existing automatic bait training devices are insufficient. The bait delivery system of many devices is relatively simple and cannot flexibly adjust the shape, size and delivery frequency of bait according to the feeding needs of leopard gill perch juveniles at different growth stages. For example, juveniles may initially require small, soft baits, and as they grow, the demand for larger granular baits increases, but it is difficult for the device to achieve this dynamic adjustment. In addition, the device has a weak ability to respond to environmental parameters. Changes in water temperature and water quality will affect the feeding activity of juveniles, but most automatic bait training devices cannot perceive these environmental factors in real time and adjust the feeding strategy accordingly, resulting in waste of bait or insufficient feeding of juveniles.

[0006] In terms of intelligence, existing devices lack the ability to accurately identify and analyze the feeding behavior of juvenile leopard gill bass. Juvenile feeding patterns are complex and variable. For example, if they are in poor health, their food intake will decrease, and if they are frightened, they may stop eating. However, the device cannot accurately capture this behavioral information through vision, sensors, and other technologies. Furthermore, the device's learning and adaptive capabilities are limited. It is unable to establish an effective feeding model based on historical feeding data and juvenile growth, making it difficult to implement personalized, intelligent feeding training programs. For example, it is impossible to develop differentiated feeding training strategies for different batches of juveniles with different growth rates, resulting in suboptimal feeding results. Summary of the Invention

[0007] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide an automatic feeding system for the training and rearing of leopard gill perch juveniles, which can feed scientifically, significantly improve the survival rate, growth rate and training effect of feeding, reduce feed waste, and realize intelligent and automated management of the entire process.

[0008] To achieve the above-mentioned object, the present invention provides the following solution: an automatic feeding system for training juvenile leopard gill perch, comprising:

[0009] A basic data module is used to obtain raw data from each aquaculture pond using an automatic bait training device, a water turbidity sensor, and a camera, and to extract basic feeding data based on the raw data;

[0010] A dynamic association module is used to perform multi-level cross-correlation modeling, causal inference association, sub-pattern mining and classification adaptation based on the basic feeding data to obtain a feeding-turbidity dynamic association model and output the feeding-turbidity relationship;

[0011] A hunger index module is used to construct a hunger index model based on the feeding-turbidity relationship, and output the hunger index of each aquaculture pond using the hunger index model;

[0012] A feeding strategy module is used to calculate the next feeding interval, the ratio of strip bait to live bait, the actual time of the next feeding, and the amount of bait per hour based on the hunger index, and to determine whether an additional meal or an increase in the next feeding interval is needed based on the hunger index and a preset threshold, thereby obtaining an adaptive feeding strategy;

[0013] A feeding drive module, configured to control the operation of the automatic bait training device according to the adaptive feeding strategy;

[0014] Wherein, the basic data module, the dynamic association module, the hunger index module, the feeding strategy module and the feeding drive module are interconnected.

[0015] Optionally, the automatic bait training device is used to automatically feed live bait at a predetermined time according to a preset bait amount. The automatic bait training device includes a power supply and a microcontroller connected to the power supply, a bait storage unit, and a water pump.

[0016] Optionally, the bait storage unit includes a strip bait storage subunit, a first solenoid valve connected to the strip bait storage subunit, a live bait storage subunit, a second solenoid valve connected to the live bait storage subunit, a bait mixing subunit connected to the first solenoid valve and the second solenoid valve, a main solenoid valve connected to the bait mixing subunit, and a feeding channel connected to the main solenoid valve.

[0017] Optionally, the basic data module includes:

[0018] A raw data unit is configured to deploy a water turbidity sensor and a camera at the edge of the aquaculture pond, automatically collect monitoring data from the automatic bait training device, the water turbidity sensor, and the camera using an Internet of Things data platform, obtain raw data, and store the raw data in a cloud database; the raw data includes a count sequence of juvenile fish feeding actions, a sequence of water turbidity decreases, a timing of bait delivery, and a timing of feeding actions;

[0019] a data processing unit for filtering out initial transient values, performing downward trend statistics, and identifying downward inflection points for the water turbidity decline sequence to obtain turbidity data, and performing image segmentation and target detection on image data in the raw data to obtain feeding times;

[0020] The relationship enhancement unit is used to construct a functional relationship between the number of feeding times and the turbidity data using statistical regression or machine learning methods to obtain basic feeding data.

[0021] Optionally, the dynamic association module includes:

[0022] A multi-level cross-correlation unit is used to set the spatial area of ​​the aquaculture pond, the juvenile fish group, and the collection terminal as nodes based on the GNN model, introduce time series features into the nodes, and define edges to obtain a multi-level cross-correlation model. The multi-level cross-correlation model is used to output the dynamic correlation coefficient of feeding-turbidity and the abnormal propagation path of each area of ​​the aquaculture pond to obtain the feeding-turbidity field pattern;

[0023] The causal inference unit is used to set the feeding event variables, turbidity change variables, and other interference factor variables, and build a structural causal model based on the set variables. Then, the do-calculus is used to perform intervention estimation on the probability distribution of turbidity change to isolate external variable factors and quantify the direct contribution of feeding to turbidity change.

[0024] The sub-pattern recognition unit is used to perform multi-dimensional time series data clustering on feeding-turbidity-water quality, separate the peak feeding pattern, the low-peak feeding pattern and the equipment intervention pattern, and extract the feeding-turbidity joint change sub-pattern through cluster analysis. The local correlation model is then used to perform personalized fitting on the feeding-turbidity joint change sub-pattern to obtain a feeding-turbidity dynamic correlation model. The feeding-turbidity dynamic correlation model is used to output the feeding-turbidity relationship. The fitting function expression of the personalized fitting is:

[0025]

[0026] in, is the turbidity change under the kth sub-mode, A t is the frequency or intensity of food intake during the same period, C t For environment variables, E t is the device status indicator, is the regression weight of each influencing factor under the kth mode, ε t is the noise term.

[0027] Optionally, the hunger index module includes:

[0028] An inference framework building unit is used to train a classification model using the feeding-turbidity relationship, introduce temporal dynamic features into the classification model to obtain a feeding posture model, and use the feeding posture model to infer the number of feeding events and the spatiotemporal distribution of behaviors to obtain inferred feeding data; the temporal dynamic features include action duration, action interval, and group synchronization behavior;

[0029] A multi-modal data unit is used to align the action curve of the inferred feeding data with the downward trend of the turbidity data to construct a combined feature and calculate a combined feature vector; the combined feature includes the number of predations per unit time, the average turbidity decline rate during the same period, the duration of the predation behavior, and the time of the turbidity abnormal inflection point;

[0030] The regression modeling unit is used to design a regression model with the target variable being the actual feeding evaluation score, deploy the regression model into the predation posture model to obtain a hunger index model, and input the combined feature vector into the hunger index model to obtain the hunger index of each breeding pond.

[0031] Optionally, the multimode data unit includes:

[0032] A mode definition subunit is used to define the characteristic mode of predation behavior, the characteristic mode of water turbidity and environment, and the characteristic mode of spatial distribution based on the inferred feeding data and the turbidity data, and to perform asynchronous alignment and missing completion on each modal data;

[0033] The vector calculation subunit is used to automatically calculate the contribution weight of each modality in the reasoning task using the multi-head attention mechanism, and then construct the combined features and calculate the combined feature vector based on the contribution weight. The calculation expression of the combined feature vector is:

[0034]

[0035] Among them, f t is the combined feature vector after the final multi-mode fusion, is the original eigenvector of the mth mode at the tth moment, W m is the linear transformation weight of the m-th modal feature space, is the contribution weight of the mth mode at the tth moment.

[0036] Optionally, the feeding strategy module includes:

[0037] The feeding interval adjustment unit is used to preset a basic fixed interval time, and adjust the basic fixed interval time according to the change rate of the hunger index to obtain the next feeding interval; the calculation expression of the next feeding interval is:

[0038] t feed(i+1) =t base ·(1-α·H t )+β·ΔH

[0039] Among them, t feed(i+1) is the recommended interval time for the next feeding, t base is the basic fixed interval time, H t is the real-time hunger index at the current moment, α is the hunger adjustment weight coefficient, ΔH is the rate of change of the hunger index within 5-15 minutes, and β is the dynamic weighted adjustment coefficient;

[0040] The bait ratio adjustment unit is used to automatically calculate the ratio of strip bait to live bait according to the hunger index and the training stage; the calculation expression of the strip bait to live bait ratio is:

[0041] r=g(S stage ,H t )=γ·S stage +(1-γ)·H t

[0042] Among them, r is the current ratio of strip bait to live bait, S stage is the classification coefficient of the training bait stage, H t is the real-time hunger index at the current moment, γ is the training phase-straw weight distribution coefficient;

[0043] The feeding calculation unit is used to calculate the actual time of the next feeding and the amount of bait per hour based on the next feeding interval and the ratio of the strip bait to the live bait, and to determine whether the hunger index is continuously higher than the hunger threshold. If so, an additional meal is provided; and to determine whether the hunger index is continuously lower than the saturation threshold. If so, the next feeding interval is automatically extended.

[0044] Optionally, the feeding drive module includes:

[0045] a valve control unit, configured to calculate and control the operating time of the first solenoid valve, the second solenoid valve, the main solenoid valve, and the water pump using the microcontroller according to the adaptive feeding strategy;

[0046] The calculation expression of the operating time of the first solenoid valve is:

[0047]

[0048] Among them, x is the opening time of the first electromagnetic valve, α is the amount of strip bait in the current round, Q dry is the actual average flow rate of strip bait;

[0049] The calculation expression of the operating time of the second solenoid valve is:

[0050]

[0051] Among them, y is the opening time of the second electromagnetic valve, β is the amount of live bait in the current round, Q live is the actual average flow of live bait;

[0052] The calculation expression of the water pump running time is:

[0053] τ=θ1·(x+y)+θ2

[0054] Among them, τ is the water pump operation time, θ1 is the unit feeding mixing time coefficient, and θ2 is the minimum guaranteed mixing benchmark time;

[0055] The calculation expression of the total solenoid valve operating time is:

[0056] δ=λ1·(x+y)+λ2

[0057] Among them, δ is the total solenoid valve full opening time, λ1 is the time coefficient required for the release of unit mixed bait, and λ2 is the mechanical hysteresis compensation time;

[0058] The log recording unit is used to record the time point of each round of feeding, the execution seconds of each action and the equipment status, obtain the action sequence log, and organize the action sequence log into a log table for feeding tracing.

[0059] The present invention provides an automatic feeding system for training juvenile leopard gill perch, and discloses the following technical effects:

[0060] 1. Accurate and reliable data acquisition: 1) Accurate data capture: By integrating the IoT water turbidity sensor and the camera, the original data of juvenile fish predation, turbidity and feeding are automatically collected in real time to realize the quantitative extraction of basic feeding data. And through statistical modeling or machine learning, the feeding behavior is functionally associated with the turbidity change, which can accurately and scientifically capture the feeding behavior of juvenile fish and their feedback to the environment, and more reasonably reflect the actual growth environment; 2) Strong data correlation: Based on the graph neural network, spatial areas, terminals, and individual juvenile fish are modeled as nodes, and the time and space characteristics are fully embedded, which deeply reveals the complex, multi-level, and highly time-varying laws between the behavior of juvenile fish and the environment, locates abnormal propagation and mode switching, and significantly improves the explanatory power of actual aquaculture dynamics, realizes the accurate characterization of the feeding-turbidity relationship, and subsequent personalized adaptive management to improve aquaculture efficiency.

[0061] 2. Real feedback on the status of juvenile fish: Based on behavioral recognition and environmental multimodality, combined with temporal and spatial distribution characteristics, a predation posture model is constructed. A multi-head attention mechanism is introduced to dynamically calculate the weights of various behavioral and environmental characteristics, output a comprehensive feature vector, and build a regression model with feeding assessment as the goal. The "hunger index" is quantitatively output, breaking away from the limitations of manual experience. The real hunger and health status of the aquaculture group is quantitatively reflected in real time, which can better judge feeding strategies, respond to group changes in a personalized manner, and improve survival and conversion rates.

[0062] 3. Intelligent feeding strategy: It can dynamically adjust the feeding interval, the ratio of strip bait to live bait, the total feeding amount and time points according to the hunger index and staged feeding goals, determine whether additional meals are needed, change the interval, etc., and realize "on-demand supply, staged, and staggered" intelligent feeding, no longer relying on fixed time periods, reducing feed waste, slowing down water quality deterioration, and increasing the success rate of strip bait conversion.

[0063] 4. Automated feeding: Through automated control of valves, water pumps, and feeding and mixing units, automatic feeding is achieved with precision in each round. The feeding process is fully logged and traceable, achieving closed-loop automation without human intervention, reducing operational errors, and greatly improving aquaculture stability and management efficiency.

[0064] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 A schematic diagram of the system architecture provided by an embodiment of the present invention;

[0067] Figure 2 A schematic diagram of the architecture of an automatic bait training device provided by an embodiment of the present invention;

[0068] Figure 3 A schematic diagram of the intelligent feeding process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0070] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0071] like Figure 1 As shown, the present invention provides an automatic feeding system for training juvenile leopard gill perch, including a basic data module connected to an automatic feeding device, a dynamic association module, a hunger index module, a feeding strategy module and a feeding drive module.

[0072] 1. Automatic bait training device

[0073] like Figure 2As shown, the automatic bait training device is used to automatically feed live bait at a fixed time according to a preset bait amount. The automatic bait training device includes a power supply and a microcontroller connected to the power supply, a bait storage unit and a water pump.

[0074] The bait storage unit includes a strip bait storage subunit, a first solenoid valve connected to the strip bait storage subunit, a live bait storage subunit, a second solenoid valve connected to the live bait storage subunit, a bait mixing subunit connected to the first solenoid valve and the second solenoid valve, a main solenoid valve connected to the bait mixing subunit, and a feeding channel connected to the main solenoid valve.

[0075] 2. Basic data module

[0076] like Figure 3 As shown, it is used to obtain the raw data of each breeding pond using an automatic bait training device, a water turbidity sensor and a camera, and extract the basic feeding data based on the raw data; the basic data module includes:

[0077] 2.1 Original Data Unit

[0078] It is used to deploy water turbidity sensors and cameras at the edge of the breeding pond, and use the Internet of Things data platform to automatically collect monitoring data of the automatic bait training device, the water turbidity sensor and the camera to obtain raw data, and store the raw data in a cloud database; the raw data includes a counting sequence of juvenile fish predation actions, a sequence of water turbidity decrease, a bait-feeding time point and a feeding action time point.

[0079] Water turbidity sensor: Use a high-sensitivity water turbidity sensor and fix it on the edge of the aquaculture pond to ensure that the measurement position is consistent each time and the data is highly comparable.

[0080] Camera and AI recognition system: Deploy high-definition cameras and connect them to AI image recognition modules to monitor the feeding movements and number of juvenile fish in real time.

[0081] Data collection and management: Develop an IoT data platform to automatically collect the above hardware outputs and record them in categories for subsequent analysis.

[0082] 2.2 Data processing unit

[0083] It is used to filter out initial transient values, perform downward trend statistics and identify downward inflection points for the water turbidity decline sequence to obtain turbidity data, and perform image segmentation and target detection on the image data in the original data to obtain the number of feeding times.

[0084] 2.3 Relationship Enhancement Unit

[0085] It is used to construct a functional relationship between the feeding frequency and the turbidity data using statistical regression or machine learning methods to obtain basic feeding data. The basic feeding data not only reflects a single feeding action, but also quantifies the feeding efficiency, innovatively realizing the mutual verification of two indicators.

[0086] 3. Dynamic association module

[0087] It is used to perform multi-level cross-correlation modeling, causal inference association, sub-pattern mining and classification adaptation based on the basic feeding data to obtain a feeding-turbidity dynamic association model and output the feeding-turbidity relationship; the dynamic association module includes:

[0088] 3.1 Multi-level cross-correlation units

[0089] like Figure 3 As shown, based on the GNN model, the spatial area of ​​the breeding pond, the juvenile fish group and the collection terminal are set as nodes, and time series features are introduced into the nodes. The time series features include the number of feeding actions, local turbidity changes, and local water environment parameters.

[0090] Edges are defined, and the edges between nodes represent their physical proximity, fluid dynamics influence, or potential causal relationship between feeding and environmental changes. A multi-level cross-correlation model is obtained by combining nodes and edges. The multi-level cross-correlation model is used to output the feeding-turbidity dynamic correlation coefficient and abnormal propagation path of each area of ​​the aquaculture pond to obtain the feeding-turbidity field pattern.

[0091] 3.2 Causal Inference Unit

[0092] It is used to set feeding event variables, turbidity change variables and other interference factor variables, and build a structural causal model based on the set variables. Then, do-calculus is used to perform intervention estimation on the probability distribution of turbidity change to isolate external variable factors and quantify the direct contribution of feeding to turbidity change.

[0093] 3.3 Sub-pattern recognition unit

[0094] It is used to cluster multidimensional time series data of feeding, turbidity and water quality, separate peak feeding patterns, low-peak feeding patterns and equipment intervention patterns, and extract feeding-turbidity joint change sub-patterns through cluster analysis. Then, the local correlation model is used to perform personalized fitting on the feeding-turbidity joint change sub-patterns to obtain a feeding-turbidity dynamic correlation model. The feeding-turbidity dynamic correlation model is used to output the feeding-turbidity relationship. The fitting function expression of the personalized fitting is:

[0095]

[0096] in, is the turbidity change under the kth sub-mode, At is the frequency or intensity of food intake during the same period, C t For environment variables, E t is the device status indicator, is the regression weight of each influencing factor under the kth mode, ε t is the noise term.

[0097] 4. Hunger Index Module

[0098] like Figure 3 As shown, it is used to construct a hunger index model based on the feeding-turbidity relationship, and use the hunger index model to output the hunger index of each aquaculture pond; the hunger index module includes:

[0099] 4.1 Reasoning Framework Building Unit

[0100] It is used to train a classification model using the feeding-turbidity relationship, and introduce temporal dynamic features into the classification model to obtain a predation posture model, and use the predation posture model to infer the number of predation events and the spatiotemporal distribution of behavior to obtain inferred feeding data; the temporal dynamic features include action duration, action interval and group synchronization behavior, so as to enhance the model's sensitivity to feeding peaks.

[0101] 4.2 Multimode Data Unit

[0102] Used to align the action curve of the inferred feeding data and the downward trend of the turbidity data to construct a combined feature and calculate a combined feature vector; the combined feature includes the number of predations per unit time, the average turbidity decline rate during the same period, the duration of the predation behavior, and the time of the turbidity abnormal inflection point; the multimodal data unit includes:

[0103] 4.2.1 Modal Definition Subunit

[0104] It is used to define the characteristic modes of predation behavior, the characteristic modes of water turbidity and environment, and the characteristic modes of spatial distribution based on the inferred feeding data and the turbidity data, and to perform asynchronous alignment and missing completion on each modality data.

[0105] Behavioral characteristic modalities: such as behavior counts per unit time, feeding action recognition scores, behavioral timing patterns, etc.

[0106] Water turbidity and environmental characteristic modes: such as turbidity change rate, water temperature, dissolved oxygen, pH, etc.

[0107] Spatial distribution characteristic modes: such as intelligent positioning of fish density distribution, spatial center of gravity changes, etc.

[0108] 4.2.2 Vector Computation Subunit

[0109] It is used to automatically calculate the contribution weight of each modality in the reasoning task using the multi-head attention mechanism, and then construct the combined features and calculate the combined feature vector based on the contribution weight; the calculation expression of the combined feature vector is:

[0110]

[0111] Among them, f t is the combined feature vector after the final multi-mode fusion, is the original eigenvector of the mth mode at the tth moment, W m is the linear transformation weight of the m-th modal feature space, is the contribution weight of the mth mode at the tth moment.

[0112] 4.3 Regression Modeling Unit

[0113] A regression model is designed with the target variable being the actual feeding evaluation score, the regression model is deployed in the predation posture model to obtain a hunger index model, and the combined feature vector is input into the hunger index model to obtain the hunger index of each breeding pond.

[0114] Target variable: such as bait intake per unit time.

[0115] 5. Feeding strategy module

[0116] like Figure 3 As shown, it is used to calculate the next feeding interval, the ratio of strip bait to live bait, the actual time of the next feeding and the amount of bait per hour according to the hunger index, and determine whether to add a meal or increase the next feeding interval according to the hunger index and a preset threshold, so as to obtain an adaptive feeding strategy; the feeding strategy module includes:

[0117] 5.1 Feeding interval adjustment unit

[0118] It is used to preset a basic fixed interval time, and adjust the basic fixed interval time according to the change rate of the hunger index to obtain the next feeding interval; the calculation expression of the next feeding interval is:

[0119] t feed(i+1) =t base ·(1-α·H t )+β·ΔH

[0120] Among them, t feed(i+1) is the recommended interval time for the next feeding, t base is the basic fixed interval time, H tis the real-time hunger index at the current moment, ranging from 0 to 1. The larger the value, the hungrier the person is. α is the hunger adjustment weight coefficient, ranging from 0 to 1. ΔH is the rate of change of the hunger index within 5-15 minutes, which is used to observe peak feeding or emergency needs. β is the dynamic weighted adjustment coefficient.

[0121] 5.2 Feeding ratio adjustment unit

[0122] It is used to automatically calculate the ratio of strip bait to live bait according to the hunger index and the bait training stage; the calculation expression of the ratio of strip bait to live bait is:

[0123] r=g(S stage ,H t )=γ·S stage +(1-γ)·H t

[0124] Among them, r is the current ratio of strip bait to live bait, the value is 0-1, 0 is all live bait, 1 is all strip bait, S stage is the classification coefficient of the training bait stage, H t is the real-time hunger index at the current moment, γ is the training phase-straw weight distribution coefficient;

[0125] 5.3 Feeding calculation unit

[0126] It is used to calculate the actual time of the next feeding and the amount of bait per hour according to the next feeding interval and the ratio of the strip bait to the live bait, and to determine whether the hunger index is continuously higher than the hunger threshold. If so, a meal is added. It is also used to determine whether the hunger index is continuously lower than the saturation threshold. If so, the next feeding interval is automatically extended, as shown in Table 1 below:

[0127] Table 1 Feeding schedule

[0128]

[0129]

[0130] 6. Feeding drive module

[0131] like Figure 3 As shown, it is used to control the operation of the automatic bait training device according to the adaptive feeding strategy; the feeding drive module includes:

[0132] 6.1 Valve control unit

[0133] Used to calculate and control the operating time of the first solenoid valve, the second solenoid valve, the main solenoid valve and the water pump using the microcontroller according to the adaptive feeding strategy.

[0134] The calculation expression of the operating time of the first solenoid valve is:

[0135]

[0136] Among them, x is the opening time of the first electromagnetic valve, α is the amount of strip bait in the current round, Q dry is the actual average flow rate of strip bait;

[0137] The calculation expression of the operating time of the second solenoid valve is:

[0138]

[0139] Among them, y is the opening time of the second electromagnetic valve, β is the amount of live bait in the current round, Q live is the actual average flow of live bait;

[0140] The calculation expression of the water pump running time is:

[0141] τ=θ1·(x+y)+θ2

[0142] Among them, τ is the water pump operation time, θ1 is the unit feeding mixing time coefficient, and θ2 is the minimum guaranteed mixing benchmark time;

[0143] The calculation expression of the total solenoid valve operating time is:

[0144] δ=λ1·(x+y)+λ2

[0145] Among them, δ is the total solenoid valve full opening time, λ1 is the time coefficient required for the release of unit mixed bait, and λ2 is the mechanical hysteresis compensation time;

[0146] 6.2 Logging Unit

[0147] It is used to record the time point of each round of feeding, the execution seconds of each action and the equipment status, obtain the action sequence log, and organize the action sequence log into a log table for feeding tracing. As shown in Table 2 below:

[0148] Table 2 Delivery action sequence log table

[0149]

[0150] Therefore, the present invention provides an automatic feeding system for the training and rearing of leopard gill perch juveniles, which can feed scientifically, significantly improve the survival rate, growth rate and training effect, reduce feed waste, and realize intelligent and automated management of the entire process.

[0151] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0152] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An automatic feeding system for training juvenile leopard gill perch, characterized in that: Includes connection to automatic bait training device: A basic data module is used to obtain raw data from each aquaculture pond using an automatic bait training device, a water turbidity sensor, and a camera, and to extract basic feeding data based on the raw data; A dynamic association module is used to perform multi-level cross-correlation modeling, causal inference association, sub-pattern mining and classification adaptation based on the basic feeding data to obtain a feeding-turbidity dynamic association model and output the feeding-turbidity relationship; A hunger index module is used to construct a hunger index model based on the feeding-turbidity relationship, and output the hunger index of each aquaculture pond using the hunger index model; A feeding strategy module is used to calculate the next feeding interval, the ratio of strip bait to live bait, the actual time of the next feeding, and the amount of bait per hour based on the hunger index, and to determine whether an additional meal or an increase in the next feeding interval is needed based on the hunger index and a preset threshold, thereby obtaining an adaptive feeding strategy; A feeding drive module, configured to control the operation of the automatic bait training device according to the adaptive feeding strategy; Wherein, the basic data module, the dynamic association module, the hunger index module, the feeding strategy module and the feeding drive module are interconnected.

2. The automatic feeding system for training juvenile leopard gill perch according to claim 1, characterized in that: The automatic bait training device is used for automatically feeding live bait at regular intervals according to a preset bait feeding amount. The automatic bait training device comprises a power supply, a microcontroller connected to the power supply, a bait storage unit and a water pump.

3. The automatic feeding system for training juvenile leopard gill perch according to claim 2, characterized in that: The bait storage unit includes a strip bait storage subunit, a first solenoid valve connected to the strip bait storage subunit, a live bait storage subunit, a second solenoid valve connected to the live bait storage subunit, a bait mixing subunit connected to the first solenoid valve and the second solenoid valve, a main solenoid valve connected to the bait mixing subunit, and a feeding channel connected to the main solenoid valve.

4. The automatic feeding system for training juvenile leopard gill perch according to claim 3, characterized in that: The basic data module includes: A raw data unit is configured to deploy a water turbidity sensor and a camera at the edge of the aquaculture pond, automatically collect monitoring data from the automatic bait training device, the water turbidity sensor, and the camera using an Internet of Things data platform, obtain raw data, and store the raw data in a cloud database; the raw data includes a count sequence of juvenile fish feeding actions, a sequence of water turbidity decreases, a timing of bait delivery, and a timing of feeding actions; a data processing unit for filtering out initial transient values, performing downward trend statistics, and identifying downward inflection points for the water turbidity decline sequence to obtain turbidity data, and performing image segmentation and target detection on image data in the raw data to obtain feeding times; The relationship enhancement unit is used to construct a functional relationship between the number of feeding times and the turbidity data using statistical regression or machine learning methods to obtain basic feeding data.

5. The automatic feeding system for training juvenile leopard gill perch according to claim 4, characterized in that: The dynamic association module includes: A multi-level cross-correlation unit is used to set the spatial area of ​​the aquaculture pond, the juvenile fish group, and the collection terminal as nodes based on the GNN model, introduce time series features into the nodes, and define edges to obtain a multi-level cross-correlation model. The multi-level cross-correlation model is used to output the dynamic correlation coefficient of feeding-turbidity and the abnormal propagation path of each area of ​​the aquaculture pond to obtain the feeding-turbidity field pattern; The causal inference unit is used to set the feeding event variables, turbidity change variables, and other interference factor variables, and build a structural causal model based on the set variables. Then, the do-calculus is used to perform intervention estimation on the probability distribution of turbidity change to isolate external variable factors and quantify the direct contribution of feeding to turbidity change. The sub-pattern recognition unit is used to perform multi-dimensional time series data clustering on feeding-turbidity-water quality, separate the peak feeding pattern, the low-peak feeding pattern and the equipment intervention pattern, and extract the feeding-turbidity joint change sub-pattern through cluster analysis. The local correlation model is then used to perform personalized fitting on the feeding-turbidity joint change sub-pattern to obtain a feeding-turbidity dynamic correlation model. The feeding-turbidity dynamic correlation model is used to output the feeding-turbidity relationship. The fitting function expression of the personalized fitting is: in, is the turbidity change under the kth sub-mode, A t is the frequency or intensity of food intake during the same period, C t For environment variables, E t is the device status indicator, is the regression weight of each influencing factor under the kth mode, ε t is the noise term.

6. The automatic feeding system for training juvenile leopard gill perch according to claim 5, characterized in that: The hunger index module includes: An inference framework building unit is used to train a classification model using the feeding-turbidity relationship, introduce temporal dynamic features into the classification model to obtain a feeding posture model, and use the feeding posture model to infer the number of feeding events and the spatiotemporal distribution of behaviors to obtain inferred feeding data; the temporal dynamic features include action duration, action interval, and group synchronization behavior; A multi-modal data unit is used to align the action curve of the inferred feeding data with the downward trend of the turbidity data to construct a combined feature and calculate a combined feature vector; the combined feature includes the number of predations per unit time, the average turbidity decline rate during the same period, the duration of the predation behavior, and the time of the turbidity abnormal inflection point; The regression modeling unit is used to design a regression model with the target variable being the actual feeding evaluation score, deploy the regression model into the predation posture model to obtain a hunger index model, and input the combined feature vector into the hunger index model to obtain the hunger index of each breeding pond.

7. The automatic feeding system for training juvenile leopard gill perch according to claim 6, characterized in that: The multimode data unit includes: A mode definition subunit is used to define the characteristic mode of predation behavior, the characteristic mode of water turbidity and environment, and the characteristic mode of spatial distribution based on the inferred feeding data and the turbidity data, and to perform asynchronous alignment and missing completion on each modal data; The vector calculation subunit is used to automatically calculate the contribution weight of each modality in the reasoning task using the multi-head attention mechanism, and then construct the combined features and calculate the combined feature vector based on the contribution weight. The calculation expression of the combined feature vector is: Among them, f t is the combined feature vector after the final multi-mode fusion, is the original eigenvector of the mth mode at the tth moment, W m is the linear transformation weight of the m-th modal feature space, is the contribution weight of the mth mode at the tth moment.

8. The automatic feeding system for training juvenile leopard gill perch according to claim 7, characterized in that: The feeding strategy module includes: The feeding interval adjustment unit is used to preset a basic fixed interval time, and adjust the basic fixed interval time according to the change rate of the hunger index to obtain the next feeding interval; the calculation expression of the next feeding interval is: t feed(i+1) =t base ·(1-α·H t )+β·ΔH Among them, t feed(i+1) is the recommended interval time for the next feeding, t base is the basic fixed interval time, H t is the real-time hunger index at the current moment, α is the hunger adjustment weight coefficient, ΔH is the rate of change of the hunger index within 5-15 minutes, and β is the dynamic weighted adjustment coefficient; The bait ratio adjustment unit is used to automatically calculate the ratio of strip bait to live bait according to the hunger index and the training stage; the calculation expression of the strip bait to live bait ratio is: r=g(S stage ,H t )=γ·S stage +(1-γ)·H t Among them, r is the current ratio of strip bait to live bait, S stage is the classification coefficient of the training bait stage, H t is the real-time hunger index at the current moment, γ is the training phase-straw weight distribution coefficient; The feeding calculation unit is used to calculate the actual time of the next feeding and the amount of bait per hour based on the next feeding interval and the ratio of the strip bait to the live bait, and to determine whether the hunger index is continuously higher than the hunger threshold. If so, an additional meal is provided; and to determine whether the hunger index is continuously lower than the saturation threshold. If so, the next feeding interval is automatically extended.

9. The automatic feeding system for training juvenile leopard gill perch according to claim 8, characterized in that: The feeding drive module includes: a valve control unit, configured to calculate and control the operating time of the first solenoid valve, the second solenoid valve, the main solenoid valve, and the water pump using the microcontroller according to the adaptive feeding strategy; The calculation expression of the operating time of the first solenoid valve is: Among them, x is the opening time of the first electromagnetic valve, α is the amount of strip bait in the current round, Q dry is the actual average flow rate of strip bait; The calculation expression of the operating time of the second solenoid valve is: Among them, y is the opening time of the second electromagnetic valve, β is the amount of live bait in the current round, Q live is the actual average flow of live bait; The calculation expression of the water pump running time is: τ=θ1·(x+y)+θ2 Among them, τ is the water pump operation time, θ1 is the unit feeding mixing time coefficient, and θ2 is the minimum guaranteed mixing benchmark time; The calculation expression of the total solenoid valve operating time is: δ=λ1·(x+y)+λ2 Among them, δ is the total solenoid valve full opening time, λ1 is the time coefficient required for the release of unit mixed bait, and λ2 is the mechanical hysteresis compensation time; The log recording unit is used to record the time point of each round of feeding, the execution seconds of each action and the equipment status, obtain the action sequence log, and organize the action sequence log into a log table for feeding tracing.

Images