A method and system for intelligent feeding control in aquaculture

Abstract

This invention relates to the field of intelligent fisheries technology, and particularly to an intelligent feeding control method and system for aquaculture. The method includes: collecting raw monitoring data and performing standardized preprocessing; extracting features from the preprocessed data to obtain multimodal behavioral features; calculating a fish feeding intention index based on the multimodal behavioral features and the preprocessed data, and determining whether to initiate the feeding process; when feeding is initiated, calculating the feeding amount based on the fish feeding intention index and feeding at appropriate intervals; after feeding, obtaining feeding response data based on the multimodal behavioral features and calculating a comprehensive feedback score; and adaptively updating the feeding amount and feeding interval based on the comprehensive feedback score to achieve closed-loop control. This method solves problems in aquaculture feeding such as inaccurate identification of feeding status, delayed response to water quality changes, lack of closed-loop feedback mechanisms, reliance on human experience, and insufficient data reliability and traceability.

Description

Technical Field

[0001] This invention relates to the field of intelligent fishery technology, and in particular to an intelligent feeding control method and system for aquaculture. Background Technology

[0002] In the aquaculture sector, feeding techniques have evolved from purely manual operation to semi-automated modes such as timed and quantitative feeding, and auxiliary methods such as camera monitoring and water quality testing have been gradually introduced, enabling basic perception of fish activity and the aquatic environment. Some systems, combined with simple control algorithms, can automatically execute feeding tasks based on preset parameters, improving feeding efficiency and ease of operation. In terms of water quality management, existing technologies can acquire key parameters such as dissolved oxygen, temperature, and pH in real time through sensors, and use them as a reference for adjusting feeding. At the same time, some aquaculture management platforms have begun to digitize and platformize feeding records, enabling centralized storage and historical retrieval of feeding data, providing basic data support for scientific aquaculture.

[0003] However, the aforementioned aquaculture feeding systems generally suffer from problems such as limited data, centralized control, opaque algorithms, and insufficient scheduling optimization, making it difficult to meet the needs of intelligent, scalable, and traceable modern aquaculture. Summary of the Invention

[0004] This invention provides an intelligent feeding control method and system for aquaculture to solve technical problems such as inaccurate identification of feeding status, delayed response to water quality changes, lack of closed-loop feedback mechanism, reliance on human experience, and insufficient data reliability and traceability in aquaculture feeding.

[0005] The present invention provides an intelligent feeding control method and system for aquaculture, which specifically includes the following technical solutions:

[0006] A method for intelligent feeding control in aquaculture includes the following steps:

[0007] S1. Collect raw monitoring data and perform standardized preprocessing to obtain preprocessed data; extract features from the preprocessed data to obtain multimodal behavioral features; calculate the fish feeding intention index based on the multimodal behavioral features and the preprocessed data.

[0008] S2. Based on the fish feeding intention index, determine whether to start the feeding process; when feeding is started, calculate the feeding amount based on the fish feeding intention index and feed according to the feeding interval; after feeding, obtain feeding response data based on multimodal behavioral characteristics and calculate the comprehensive feedback score; based on the comprehensive feedback score, adaptively update the feeding amount and feeding interval to achieve closed-loop control.

[0009] Preferably, S1 specifically includes:

[0010] By weighted summation of multimodal behavioral features, combined with dissolved oxygen concentration in preprocessed data, and by introducing a group movement dynamic amplification factor, the feeding intention index of fish groups is calculated.

[0011] Preferably, S1 specifically includes:

[0012] By performing optical flow analysis on the multimodal behavioral characteristics, the rate of change of the kinetic energy of the fish swarm per unit time was calculated, and a dynamic amplification coefficient was introduced to obtain the dynamic amplification factor of the swarm movement.

[0013] Preferably, S2 specifically includes:

[0014] The feeding intention index of the fish population is compared with the preset feeding intention threshold. When the feeding intention index of the fish population is greater than the feeding intention threshold, the feeding process is started. The feeding interval is set by combining the exponential decay function, and a mapping relationship is established between the feeding interval and the feeding intention index of the fish population.

[0015] Preferably, S2 specifically includes:

[0016] Target detection and motion trajectory analysis are performed on the image data in the preprocessed data to obtain the fish gathering center and swimming direction, and a feeding strategy is formulated.

[0017] Preferably, S2 specifically includes:

[0018] Based on the fish feeding intention index, combined with the feeding intention threshold and the preset maximum feeding amount per bite, the feeding amount is calculated, and feeding is carried out in accordance with the feeding strategy.

[0019] Preferably, S2 specifically includes:

[0020] After feeding is performed, based on the feeding response data and the stability of fish aggregation, normalization and weighted fusion calculations are performed to obtain a comprehensive feedback score. Based on the comprehensive feedback score, the feeding amount and feeding interval are adaptively updated, and the feeding intention threshold is optimized to achieve adaptive closed-loop control. The stability of fish aggregation is calculated based on the image data in the preprocessed data.

[0021] An intelligent feeding control system for aquaculture includes the following components:

[0022] Multimodal sensing unit module, feeding intention monitoring module, intelligent feeding execution module, and feedback optimization module;

[0023] The multimodal sensing unit module collects raw monitoring data and performs standardized preprocessing to obtain preprocessed data; it then extracts features from the preprocessed data to obtain multimodal behavioral features, which are stored in the feedback optimization module; the multimodal sensing unit module is connected to the feeding intention monitoring module, the intelligent feeding execution module, and the feedback optimization module.

[0024] The feeding intention monitoring module calculates the fish feeding intention index based on multimodal behavioral characteristics and dissolved oxygen concentration in the preprocessed data, and stores it in the feedback optimization module. When the fish feeding intention index is updated, it automatically notifies the intelligent feeding execution module.

[0025] The intelligent feeding execution module compares the fish feeding intention index with a preset feeding intention threshold. When the feeding intention index is greater than the feeding intention threshold, the feeding process is initiated, and the feeding interval is set using an exponential decay function. The module performs target detection and motion trajectory analysis on the image data in the preprocessed data and formulates a feeding strategy. Based on the fish feeding intention index and the feeding intention threshold, the module calculates the feeding amount and implements the feeding strategy accordingly. The feeding intention threshold, feeding interval, and feeding amount are stored in the feedback optimization module.

[0026] The feedback optimization module obtains feeding response data based on multimodal behavioral characteristics after feeding, and calculates a comprehensive feedback score by combining the stability of fish aggregation. Based on the comprehensive feedback score, it adaptively adjusts the feeding amount and feeding interval, and optimizes the feeding intention threshold to achieve adaptive closed-loop control.

[0027] The beneficial effects of the technical solution of the present invention are:

[0028] 1. Multimodal Precision Monitoring: Through multimodal fusion of underwater cameras, water quality sensors, sound detection modules and auxiliary sensing devices, fish behavior and environmental parameters are comprehensively collected. After feature extraction and normalization, comparable behavioral features are formed, which significantly improves the accuracy and real-time performance of fish feeding intention index calculation.

[0029] 2. Intelligent Triggering and Dynamic Control: When the fish feeding intention index exceeds the feeding intention threshold, the feeding process is automatically triggered. The target detection and trajectory analysis algorithm determines the fish gathering center and swimming direction, thereby formulating a feeding strategy, selecting a single feeding port or multiple points for dispersed feeding, reducing fish competition and feed waste.

[0030] 3. Feedback-driven continuous optimization: After feeding, a comprehensive feedback score is calculated based on the stability of fish aggregation and feeding response data. Based on the comprehensive feedback score, the feeding amount and feeding interval are optimized and updated through machine learning. The feeding intention threshold is iteratively optimized by combining the strategy gradient method with multiple feeding backtracking verifications, thereby achieving a unity of real-time adjustment and long-term optimization.

[0031] 4. Closed-loop intelligent control: This invention forms a closed-loop control mechanism of "monitoring-execution-feedback-re-monitoring", which can automatically adjust feeding parameters when feeding status changes or environmental conditions fluctuate, thereby improving the stability, robustness and adaptability of the intelligent feeding control system.

[0032] 5. Reduce costs and improve efficiency: Precise feeding reduces uneaten feed waste, improves water quality, lowers feed costs, promotes even feeding and healthy growth of fish, and improves aquaculture efficiency. Attached Figure Description

[0033] Figure 1 This is a structural diagram of an intelligent feeding control system for aquaculture as described in this invention;

[0034] Figure 2 This is a flowchart of an intelligent feeding control method for aquaculture as described in this invention. Detailed Implementation

[0035] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0037] The following description, in conjunction with the accompanying drawings, details a smart feeding control method and system for aquaculture provided by this invention.

[0038] See attached document Figure 1 The diagram illustrates a structural diagram of an intelligent feeding control system for aquaculture provided by an embodiment of the present invention. The system includes the following components:

[0039] Multimodal sensing unit module, feeding intention monitoring module, intelligent feeding execution module, and feedback optimization module;

[0040] The multimodal sensing unit module automatically collects raw monitoring data through multimodal sensing devices deployed in the aquaculture water body, including underwater cameras, water quality sensor groups, and sound detection modules, and performs standardized preprocessing to obtain preprocessed data. Feature extraction is performed on the preprocessed data to obtain multimodal behavioral features, which are then stored in the feedback optimization module. The multimodal sensing unit module is connected to the feeding intention monitoring module, the intelligent feeding execution module, and the feedback optimization module.

[0041] The feeding intention monitoring module calculates the fish feeding intention index based on multimodal behavioral characteristics and dissolved oxygen concentration in the preprocessed data, and stores it in the feedback optimization module. When the fish feeding intention index is updated, it automatically notifies the intelligent feeding execution module.

[0042] The intelligent feeding execution module compares the fish feeding intention index with a preset feeding intention threshold. When the feeding intention index is greater than the feeding intention threshold, the feeding process is initiated, and the feeding interval is set using an exponential decay function. The module performs target detection and motion trajectory analysis on the image data in the preprocessed data and formulates a feeding strategy. Based on the fish feeding intention index and the feeding intention threshold, the module calculates the feeding amount and implements the feeding strategy accordingly. The feeding intention threshold, feeding interval, and feeding amount are stored in the feedback optimization module.

[0043] The feedback optimization module acquires multimodal behavioral characteristics after feeding in real time, obtaining feeding response data (such as fish school aggregation degree, feeding duration, and uneaten food ratio). Combined with the stability of fish school aggregation, it calculates a comprehensive feedback score. Based on the comprehensive feedback score, it adaptively updates the feeding amount and feeding interval, and backtracks to verify the optimization of the feeding intention threshold through the strategy gradient method. The updated results are written into the feedback optimization module and synchronized to the intelligent feeding execution module, thereby realizing adaptive closed-loop control of monitoring-execution-feedback-remonitoring.

[0044] See attached document Figure 2 The diagram illustrates a flowchart of an intelligent feeding control method for aquaculture according to an embodiment of the present invention. The method includes the following steps:

[0045] S1. Collect raw monitoring data and perform standardized preprocessing to obtain preprocessed data; extract features from the preprocessed data to obtain multimodal behavioral features; calculate the fish feeding intention index based on the multimodal behavioral features and the preprocessed data.

[0046] Multimodal sensing devices are deployed in the aquaculture water to collect raw monitoring data. These devices include: underwater cameras (responsible for collecting image data, including fish phase, density, and behavior); a water quality sensor array (responsible for collecting environmental physicochemical data, including dissolved oxygen (DO) concentration, temperature, pH, and, if necessary, conductivity and ammonia nitrogen concentration); a sound detection module (used to capture feeding sounds and group activity sounds); and other auxiliary sensors (buoy-type position sensors, current meters, and light sensors). The raw monitoring data undergoes standardized preprocessing to obtain preprocessed data. This standardized preprocessing includes cleaning, noise reduction, missing data completion, time synchronization, and normalization and dimensionless processing to ensure comparability of different modal data at a uniform scale. All methods used are well-known to those skilled in the art and will not be elaborated upon here.

[0047] Feature extraction is performed on the preprocessed data using existing feature engineering techniques to output multimodal behavioral features;

[0048] Furthermore, based on multimodal behavioral characteristics, combined with dissolved oxygen concentration in the preprocessed data, and introducing a group movement dynamic amplification factor, a fish feeding intention index is calculated to comprehensively characterize the feeding needs of the fish in the near future. The specific formula for the fish feeding intention index is defined by combining linear weighting of multimodal behavioral characteristics with environmental sensitivity correction:

[0049] ,

[0050] in, Indicates the current time The fish feeding intention index is a continuous value, limited to 0 to 1, which is used to represent the strength of the fish's feeding intention. The larger the value of the fish feeding intention index, the more likely the fish are to need to feed, thereby guiding the execution delay and edge computing of the feeding strategy. It is the Sigmoid function, an S-shaped non-linear function that smoothly controls the effect of dissolved oxygen concentration on the feeding intention of fish by mapping the input value to the range of 0–1. This is the dissolved oxygen sensitivity coefficient, obtained experimentally, and is a positive value, typically between 0.1 and 10. The higher the value, the more sensitive the fish's feeding intentions are to changes in dissolved oxygen; It is the first Individual behavioral characteristics; It is the first The feature weights of each behavioral feature are set according to expert experience, and the values ​​range from [-1, 1]. It is the total number of behavioral characteristics; It is a weighted sum of multimodal behavioral features; It is the current moment. The dissolved oxygen concentration after pretreatment was obtained from the pretreatment data. This is a dissolved oxygen concentration threshold, set based on expert experience, used to characterize the critical point between active and inhibited feeding in fish populations. Its value ranges from 3 to 7. When the dissolved oxygen concentration after pretreatment... Above the dissolved oxygen concentration threshold When the dissolved oxygen concentration is below the dissolved oxygen concentration threshold after pretreatment, the fish’s feeding intention increases; when the dissolved oxygen concentration is below the threshold, the fish’s feeding intention decreases. It is the current moment. The rate of change of the kinetic energy of the fish school movement per unit time is used to quantify the dynamic changes in group behavior. It is obtained by performing optical flow analysis on the multimodal behavioral characteristics, calculating the regional mean of the square of the optical flow velocity amplitude, and performing time difference. This is a well-known technique in the art and will not be elaborated here. It is the dynamic amplification factor, which can be compared. The regression analysis yielded a result that was fitted to the actual feeding intensity. This result was used to amplify (or attenuate) changes in feeding intention caused by sudden changes in population dynamics. The regression results were then scaled to between 0 and 2 using a Sigmoid scaling function. The desire to eat is amplified when exercise intensifies. If the intensity is relatively low, the impact of movement changes is weaker; the actual feeding intensity is obtained by existing image recognition and acoustic detection methods, used to characterize the feeding activity of the fish during the feeding process, and will not be elaborated here; the regression method is a technique well known to those skilled in the art, and will not be elaborated here; the first item Introducing the dissolved oxygen concentration after pretreatment The difference between the dissolved oxygen concentration and the dissolved oxygen concentration threshold, after pretreatment, represents the dissolved oxygen concentration. The second item suppresses the feeding intention index of fish when it is low; The third term is a weighted sum of multimodal behavioral features, representing the comprehensive feature result extracted from direct monitoring data such as visual, acoustic, and fish density data; As a dynamic amplification factor for group movement, it emphasizes the enhancement or weakening of the feeding hospital index of fish groups during group mutations (such as sudden accelerated aggregation). It is used to make a secondary correction based on the weighted sum of behavioral characteristics, reflecting the real-time effect of group dynamics.

[0051] S2. Based on the fish feeding intention index, determine whether to start the feeding process; when feeding is started, calculate the feeding amount based on the fish feeding intention index and feed according to the feeding interval; after feeding, obtain feeding response data based on multimodal behavioral characteristics and calculate the comprehensive feedback score; based on the comprehensive feedback score, adaptively update the feeding amount and feeding interval to achieve closed-loop control.

[0052] The feeding intention index of the fish population is compared with a feeding intention threshold to determine whether a trigger condition is met: when the feeding intention index is greater than the feeding intention threshold (i.e., the trigger condition is met), the intelligent feeding control system for aquaculture starts the feeding process; when the feeding intention index is less than or equal to the feeding intention threshold, feeding is stopped; the feeding intention threshold is... The values ​​are set based on expert experience and range from 0.3 to 0.7.

[0053] Set the feeding interval when the triggering conditions are met. Combining the exponential decay function, the feeding interval is... Fish feeding intention index Establish a mapping relationship so that when the fish feeding intention index Feeding intervals when the temperature rises The feeding intention index of fish decreases according to an exponential pattern. When the temperature drops, the feeding interval should be adjusted accordingly. It extends according to an exponential pattern, thereby achieving dynamic setting;

[0054] Furthermore, by using target detection and motion trajectory analysis algorithms, the preprocessed image data is processed and analyzed to obtain the fish gathering center and swimming direction, and a feeding strategy is formulated to determine whether to feed one fish at a time or multiple fish at a time: if the fish are highly concentrated near a certain feeding point and swim in the same direction, then one fish at a time is used; if the fish are scattered or have high density in multiple areas (i.e., the fish spatial distribution is multi-centered or covers a large area), then multiple fish at a time is used.

[0055] Furthermore, based on the fish feeding intention index, combined with the feeding intention threshold and the maximum single-batch feeding amount, the feeding amount is calculated. When feeding a single fish, if the feeding amount exceeds the maximum single-batch feeding amount, the excess portion is distributed in batches according to the feeding interval. When feeding multiple fish, the weights are calculated using a product method based on the fish density in the area covered by each feeding point, the distance between each feeding point and the fish gathering center, and the amount of uneaten food. The feeding amount is then distributed to each feeding point according to the weight ratio. The preprocessed image data is taken from the preprocessed data. The target detection and motion trajectory analysis algorithms and the product method are well-known techniques to those skilled in the art and will not be described in detail here. The amount of uneaten food is obtained by identifying the number or proportion of uneaten feed particles from the preprocessed image data using a target detection and segmentation algorithm. It can also be corrected by acoustic detection or changes in water quality parameters, which will not be described in detail here.

[0056] The specific formula for calculating the amount of feed is as follows:

[0057] ,

[0058] in, This is the feeding amount calculated for the current cycle, representing the final feeding output result executed by the intelligent feeding control system for aquaculture. It is the maximum amount of feed per single feeding, which is the upper limit of feeding specified by the equipment or breeding strategy. It represents the maximum amount of feed that cannot be exceeded per single feeding under the most ideal feeding conditions. This represents the sampling time index of the intelligent feeding control system for aquaculture, used to identify the calculation cycle of the fish feeding intention index and the corresponding feeding amount; At any moment Fish feeding intention index; It is the feeding willingness threshold, with a value range of 0.5 to 0.9. It is the minimum willingness requirement to start feeding, set based on expert experience. This refers to the amount of uneaten feed currently detected, which is the amount of feed remaining after the last or previous feedings. This is the maximum allowable amount of uneaten feed, set according to expert experience, and is set to 5% to 20% of the maximum single-bite feeding amount Qmax. If the amount of uneaten feed exceeds the maximum allowable amount of uneaten feed, it is considered overfeeding, and subsequent feeding needs to be reduced or stopped. This indicates the ratio of the current fish feeding intention index to the feeding intention threshold. This is a residual bait correction factor; if the amount of residual bait is close to the maximum allowable amount of residual bait (i.e., ... ≈ If the residual feed correction factor approaches 0, it means that no further feeding is needed. If the amount of residual feed is very small ( If the residual feed correction factor is close to 1, it means that feeding can proceed normally.

[0059] After feeding, based on multimodal behavioral characteristics, feeding response data (including feeding duration, uneaten food ratio, and feeding efficiency) and fish aggregation stability were obtained. These data were then normalized and weighted to calculate a comprehensive feedback score. The value ranges from 0 to 1, and the larger the value of the comprehensive feedback score, the better the feeding effect. The stability of the fish group aggregation is calculated by target detection and trajectory analysis of the preprocessed image data. For example, it is characterized by the variance of the spatial distribution of the fish group, the consistency of the velocity direction, or the fluctuation amplitude of the group's center of gravity. These are technical means well known to those skilled in the art and will not be described in detail here.

[0060] Based on feeding amount Overall feedback score Using existing machine learning optimization methods, the feeding amount and feeding interval are optimized. Adaptive updates are performed, and the feeding intention threshold is optimized by backtracking and verifying the results of multiple feedings using a policy gradient method. The updated results are written to the feedback optimization module and synchronized to the intelligent feeding execution module, thereby realizing adaptive closed-loop control of monitoring-execution-feedback-re-monitoring; the machine learning optimization method and policy gradient method are well known to those skilled in the art and will not be described in detail here.

[0061] In summary, a method and system for intelligent feeding control in aquaculture has been developed.

[0062] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0063] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent feeding control in aquaculture, characterized in that, Includes the following steps: S1. Collect raw monitoring data using multimodal sensing devices, including: an underwater camera responsible for collecting image data, such as fish school phase, fish school density, and behavioral movements; a water quality sensor group responsible for collecting environmental physicochemical data, such as dissolved oxygen concentration, temperature, and pH; a sound detection module responsible for capturing feeding sounds and group activity sounds; and other sensors, including a buoy-type position sensor, a current meter, and a light sensor. Standardize and preprocess the raw monitoring data to obtain preprocessed data. The standardization and preprocessing process includes cleaning, noise reduction, missing data completion, time synchronization, and normalization and dimensionless processing. Extract features from the preprocessed data to obtain multimodal behavioral features. Based on the multimodal behavioral features and the preprocessed data, calculate the fish school feeding intention index using the following formula: ， in, Indicates the current time Fish feeding intention index; It is the Sigmoid function; It is the dissolved oxygen sensitivity coefficient; It is the first Individual behavioral characteristics; It is the first Feature weights for each behavioral characteristic; It is the total number of behavioral characteristics; It is the current moment. Dissolved oxygen concentration after pretreatment; It is the dissolved oxygen concentration threshold; It is the current moment. The rate of change of the kinetic energy of the fish school movement per unit time is obtained by performing optical flow analysis on the multimodal behavioral characteristics, calculating the regional mean of the square of the optical flow velocity amplitude, and then performing time difference analysis. It is the dynamic amplification factor; As a dynamic amplification factor for group movement; S2. Compare the fish feeding intention index with a preset feeding intention threshold. When the fish feeding intention index is greater than the feeding intention threshold, initiate the feeding process. Combine this with an exponential decay function to set the feeding interval, establishing a mapping relationship between the feeding interval and the fish feeding intention index. When the fish feeding intention index increases, the feeding interval shortens exponentially; when the fish feeding intention index decreases, the feeding interval lengthens exponentially. Perform target detection and motion trajectory analysis on the image data in the preprocessed data to obtain the fish gathering center and swimming direction, and formulate a feeding strategy. When feeding is initiated, calculate the feeding amount based on the fish feeding intention index, and feed the fish according to the feeding strategy and feeding interval. The formula for calculating the feeding amount is: ， in, This is the feeding amount calculated for the current cycle; This is the maximum amount that can be fed in a single feeding. This indicates the sampling time index of the intelligent feeding control system for aquaculture; It is the threshold for the willingness to eat; This represents the amount of residual bait currently detected. This is the maximum amount of uneaten bait allowed. After feeding, feeding response data is obtained based on multimodal behavioral characteristics, and a comprehensive feedback score is calculated. Based on the comprehensive feedback score, the feeding amount and feeding interval are adaptively updated to achieve closed-loop control.

2. The intelligent feeding control method for aquaculture according to claim 1, characterized in that, S2 specifically includes: After feeding is performed, based on the feeding response data and the stability of fish aggregation, normalization and weighted fusion calculations are performed to obtain a comprehensive feedback score. Based on the comprehensive feedback score, the feeding amount and feeding interval are adaptively updated, and the feeding intention threshold is optimized to achieve adaptive closed-loop control. The stability of fish aggregation is calculated based on the image data in the preprocessed data.

3. An intelligent feeding control system for aquaculture, applied to the intelligent feeding control method for aquaculture as described in claim 1, characterized in that, Includes the following parts: Multimodal sensing unit module, feeding intention monitoring module, intelligent feeding execution module, and feedback optimization module; The multimodal sensing unit module collects raw monitoring data through multimodal sensing devices, including: an underwater camera responsible for collecting image data, such as fish school phase, fish school density, and behavioral movements; a water quality sensor group responsible for collecting environmental physicochemical data, such as dissolved oxygen concentration, temperature, and pH; a sound detection module responsible for capturing feeding sounds and group activity sounds; and other sensors, including a buoy-type position sensor, a current meter, and a light sensor. The module performs standardized preprocessing on the raw monitoring data to obtain preprocessed data; it then extracts features from the preprocessed data to obtain multimodal behavioral features, which are stored in the feedback optimization module. The multimodal sensing unit module is connected to the feeding intention monitoring module, the intelligent feeding execution module, and the feedback optimization module. The feeding intention monitoring module calculates the fish feeding intention index based on multimodal behavioral characteristics and dissolved oxygen concentration in the preprocessed data, and stores it in the feedback optimization module. When the fish feeding intention index is updated, it automatically notifies the intelligent feeding execution module. The intelligent feeding execution module compares the fish feeding intention index with a preset feeding intention threshold. When the feeding intention index is greater than the feeding intention threshold, the feeding process is initiated, and the feeding interval is set using an exponential decay function. The module performs target detection and motion trajectory analysis on the image data in the preprocessed data and formulates a feeding strategy. Based on the fish feeding intention index and the feeding intention threshold, the module calculates the feeding amount and implements the feeding strategy accordingly. The feeding intention threshold, feeding interval, and feeding amount are stored in the feedback optimization module. The feedback optimization module obtains feeding response data based on multimodal behavioral characteristics after feeding, and calculates a comprehensive feedback score by combining the stability of fish aggregation. Based on the comprehensive feedback score, it adaptively adjusts the feeding amount and feeding interval, and optimizes the feeding intention threshold to achieve adaptive closed-loop control.

Images