Visual fishery feeding control system and method based on AI drive

Abstract

The invention discloses a visual fishery feeding control system and method based on AI driving, and relates to the technical field of aquaculture automation control, and the system comprises a mobile positioning unit, a feeding execution unit, a visual analysis unit, and a decision control unit. A local and cloud collaborative intelligent management mechanism guarantees the breeding stability, a local PLC receives track positions, sensor data and fish school characteristics in real time, the moving speed of a feeding bin and the opening degree of a valve are adjusted through a PID algorithm, and a cloud platform stores historical data and analyzes the fish school feeding rule; a precise feeding system is realized through cooperation of a self-defined feeding plan, abnormal alarm, an air track and visual analysis, the system realizes cross-pond efficient coverage of a plurality of culture ponds through an annular air track structure, and a cooperative calculation subunit extracts fish school density, swimming speed and other characteristics through a YOLOv8 algorithm, calculates the feeding amount, and realizes the accurate feeding of the culture ponds through the YOLOv8 algorithm. And feed putting is dynamically matched with fish school demands.

Description

Technical Field

[0001] This invention relates to the field of automated control technology for aquaculture, specifically to an AI-driven, visualized fishery feeding control system and method. Background Technology

[0002] Against the backdrop of the industrialization and large-scale development of aquaculture, precise feeding is a key link affecting aquaculture efficiency and resource utilization. Traditional manual feeding methods rely on experience and judgment, which have problems such as high labor intensity, uneven feeding, and difficulty in adapting to efficient management of multiple ponds.

[0003] Currently, existing automated feeding equipment still has significant limitations. For example, ground-mounted rail feeders are limited by the layout of the ponds, have poor mobility, and are susceptible to humid and corrosive environments, resulting in low equipment reliability. Other fixed-point feeding systems rely solely on timed or single water quality sensors to trigger feeding, failing to detect the actual feeding status of the fish in real time. This leads to a mismatch between the amount of feed and the fish's needs, easily causing feed waste or underfeeding.

[0004] In addition, existing systems often lack a data loop for perception, decision-making, and execution. Feeding equipment and monitoring systems are independent of each other, and farmers still need to manually observe videos to adjust feeding strategies. They fail to achieve adaptive regulation based on fish behavior, have limited intelligence, and cannot support long-term optimization of precision aquaculture.

[0005] Therefore, existing technologies cannot meet the intelligent feeding requirements of large-scale farming, which require full coverage, dynamic perception, and precise execution. There is an urgent need for an integrated feeding system that combines advanced mobile positioning, visual analysis, and intelligent decision-making. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an AI-driven, visualized fishery feeding control system and method. This system ensures aquaculture stability through a collaborative intelligent management mechanism between local and cloud platforms. The local PLC controller receives real-time data on track position, sensor data, and fish characteristics, and adjusts the feeding bin's movement speed and valve opening using a PID algorithm. The cloud platform stores historical data and analyzes fish feeding patterns, supporting customized feeding plans and anomaly alarms. Furthermore, the collaboration between the aerial track and visualized analysis achieves a precise feeding system. The system uses a circular aerial track structure to overcome the movement limitations of ground equipment, enabling efficient cross-pond coverage across multiple aquaculture ponds. The collaborative computing subunit extracts fish density, swimming speed, and other characteristics using the YOLOv8 algorithm to calculate the feed amount. The decision control unit drives the electromagnetic pulse valve to execute the feed delivery, forming a complete closed loop of spatial positioning, status perception, dosage calculation, and execution feedback, dynamically matching feed delivery with fish needs.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, an AI-driven visual fishery feeding control system, the system comprising: a mobile positioning unit, a feeding execution unit, a visualization analysis unit, and a decision control unit; The mobile positioning unit is used to construct an aerial track network across the aquaculture ponds. It is erected in the area above the aquaculture pond group and has a built-in double-track guide structure and servo drive motor to drive the feeding execution unit to move flexibly in both directions along the ±X / ±Y axes. The feeding execution unit is suspended below the track and includes an electromagnetic pulse valve, a microwave level sensor and a weight sensor. It controls the valve opening and feeding time by PWM to control the discharge amount, and monitors the remaining amount of feed in the feeding bin and the cumulative feed weight during the feeding process in real time. The visualization analysis unit integrates a camera and a collaborative computing subunit. The camera's optical axis is vertically downward to fully cover the surface of a single aquaculture pond. The collaborative computing subunit is equipped with a fish behavior analysis AI model based on the YOLOv8 algorithm, which outputs data on fish density, swimming speed, and feeding areas in real time. The decision control unit includes a local PLC controller and a cloud platform. The PLC executes the feeding logic control, and the cloud platform supports remote feeding plan customization, real-time data dashboard, and abnormal alarms.

[0008] Furthermore, in the mobile positioning unit, the aerial track is designed in a ring layout. The aerial track has a built-in double-track guide structure and a servo drive motor. The double-track guide structure consists of two parallel track beams and a guide slider embedded in the inner side of the track beams. The guide slider is connected to the feeding execution unit. The servo drive motor is connected to the track beam through a gear and rack transmission mechanism. The motor has a built-in absolute encoder for real-time acquisition and feedback of displacement data, which enables position control and speed adjustment of the feeding execution unit. The position control accuracy is ≤5cm and the speed adjustment range is 0.3-1m / s. The feeding execution unit is suspended below the track and includes an electromagnetic pulse valve, a microwave level sensor, and a weight sensor. The electromagnetic pulse valve controls its opening by receiving PWM pulse signals to adjust the discharge rate. The microwave level sensor is installed on the top of the inner wall of the feeding hopper to detect the remaining feed and trigger a feeding reminder when the remaining feed is insufficient. The weight sensor is integrated into the bottom of the feeding hopper and is used to measure the cumulative weight of the feed in real time.

[0009] Furthermore, the opening time of the electromagnetic pulse valve is determined by... The calculation is performed, where Q is the target feeding amount, preset by the decision control unit based on the parameters of the aquaculture pond, and K is the feed output coefficient. A represents the valve opening percentage; The weight sensor monitors the amount of feed in real time and closes the valve when the value reaches ±2% of the Q value, so that the error of a single feeding is ≤±100g.

[0010] Furthermore, the visualization analysis unit integrates a camera and a collaborative computing subunit, with the camera's optical axis vertically downwards to cover the entire aquaculture pond area; The collaborative computing subunit is connected to the camera via Ethernet and is used to receive image data and perform fish behavior analysis. The process includes: Image preprocessing: Denoising, contrast enhancement, and distortion correction are performed on the original image; Model Inference: Load a YOLOv8-based AI model for fish swarm behavior analysis and output fish swarm density. Average swimming speed and coordinates of the food-grabbing area ; Data output: The behavioral feature data is packaged into JSON format and sent to the decision control unit.

[0011] Furthermore, the construction process of the fish swarm behavior analysis AI model is as follows: Dataset construction: Collect images of fish in aquaculture ponds containing different fish species, densities, and lighting conditions, and label the target bounding box coordinates and fish activity levels. Model training: The SGD optimizer was used, with an initial learning rate α = 0.01, momentum μ = 0.9, weight decay λ = 0.0005, and training epochs. In each iteration, the loss function is used. Optimize model parameters, where, For classifying losses, To regress the loss, These are the regression loss weight coefficients; Feature extraction: Calculate fish density using a trained model. Average swimming speed and the center coordinates of the food-grabbing area Wherein, the fish density In the formula, The number of fish detected by the model. The surface area of ​​the aquaculture pond. The average swimming speed is a correction factor used to compensate for the bias caused by incomplete detection of underwater fish populations. In the formula, , The first The fish in the Frame and the The center coordinates of the frame This refers to the camera's frame rate.

[0012] Furthermore, the decision control unit includes a local PLC controller and a cloud platform; The local PLC controller establishes real-time communication with the mobile positioning unit, the feeding execution unit, and the vision analysis unit for: Receive the real-time position of the feeding execution unit and control its movement and positioning; Receives data from the microwave level sensor and triggers a clearance alarm. Receive data from the weight sensor and control the electromagnetic pulse valve via PWM; It receives fish behavior data output from the visual analysis unit and generates feeding instructions by combining the parameters of the breeding pond.

[0013] Furthermore, the cloud platform is used for: Store historical feeding data and support querying and exporting; Provides a visual interface for customizing feeding plans; Real-time display of fish status and equipment operating parameters; It will issue an alarm and record when there is equipment failure or abnormal fish population.

[0014] On the other hand, an AI-driven, visualized fishery feeding control method is described, with the following specific steps: S1. Load the feeding plan and the list of preset coordinates of the breeding pond; S2. Drive the feeding execution unit to move to the target position according to an S-shaped curve, and switch to low-speed mode when it gets close; S3. Collect video streams and output fish behavior data through AI model analysis; S4. Calculate the amount of feed and control the valve to open, with real-time feedback from the weight sensor; S5. Record feeding data and synchronize it to the cloud to generate a decision report.

[0015] Compared with existing technologies, this AI-driven, visualized fishery feeding control system and method has the following advantages: I. The intelligent management mechanism of this invention, which integrates local and cloud-based systems, ensures the stability of aquaculture. The local PLC controller receives real-time data on track position, sensor data, and fish characteristics, and adjusts the feeding hopper's movement speed and valve opening through a PID algorithm. The cloud platform stores historical data and analyzes fish feeding patterns, supports customized feeding plans and anomaly alarms. Furthermore, the collaboration between the aerial track and visual analysis enables a precise feeding system. The system overcomes the movement limitations of ground equipment through a ring-shaped aerial track structure, achieving efficient cross-pond coverage across multiple aquaculture ponds. The collaborative computing subunit extracts characteristics such as fish density and swimming speed using the YOLOv8 algorithm, calculates the feeding amount, and the decision control unit drives the electromagnetic pulse valve to execute the feeding, forming a complete closed loop of spatial positioning, status perception, dosage calculation, and execution feedback, so that feed delivery dynamically matches the needs of the fish.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is an operation flowchart of an AI-driven, visualized fishery feeding control system. Figure 2 This is a flowchart of the visual analysis and feeding control process of the decision control unit in Example 1; Figure 3 This is a schematic diagram of the overall structure of an AI-driven, visualized fishery feeding control system. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] Example 1: This example details the application process of an AI-driven visual fishery feeding control system. The system enables flexible movement of the feeding bin through a mobile positioning unit, precise control of feed delivery by the feeding execution unit, analysis of fish behavior by the visual analysis unit using AI video recognition technology, and collaborative management between the local and cloud-based decision control units. This forms a complete closed loop from fish status perception to precise feeding, effectively improving the intelligence and precision of aquaculture.

[0021] This system consists of four core parts: a mobile positioning unit, a feeding execution unit, a visualization analysis unit, and a decision control unit, which together complete the intelligent feeding task for fish in aquaculture ponds.

[0022] When the system starts, a preset feeding plan and a list of preset coordinates of the breeding pond are loaded on the cloud platform or local terminal of the decision control unit. These preset coordinates are set by manual on-site measurement based on the specific layout of the breeding pond, and are intended to provide a precise target location reference for the movement of the feeding bin, so as to ensure that the feeding bin can accurately reach each feeding area of ​​the breeding pond.

[0023] The aforementioned mobile positioning unit provides support for the movement of the feeding execution unit. The track is designed as a ring structure, erected above the breeding pond. The guide structure of the track consists of two parallel track beams and guide sliders embedded in the inner side of the track beams. The guide sliders are connected to the feeding execution unit. When the feeding bin needs to move, the guide sliders slide within the track beams, restricting the feeding bin's degree of freedom of movement, ensuring that it can only move along the track direction, thus guaranteeing the stability and accuracy of the feeding bin's movement. The servo drive motor is the power source for the feeding bin's movement, connected to the track beams through a gear and rack transmission mechanism. The absolute encoder built into the motor can collect the displacement data of the feeding bin in real time and feed this data back to the local PLC controller of the decision control unit. The PLC controller calculates the feeding position based on the preset target position and the real-time position fed back by the encoder. The system tracks the movement distance and speed of the feeding bin, then sends control commands to the servo drive motor to achieve bidirectional movement of the feeding bin along the ±X / ±Y axes. During track positioning, the servo motor drives the feeding bin to move according to an S-shaped acceleration and deceleration curve. The reason for using the S-shaped acceleration and deceleration curve is that it allows the feeding bin to accelerate and decelerate smoothly during the start and stop phases, avoiding the feed shaking inside the feeding bin due to sudden speed changes, which would affect the accuracy of the feeding amount. At the same time, it can also reduce the impact on the track and motor, extending the equipment life. When the feeding bin is 50cm away from the preset position of the target breeding pond, the system will automatically switch to low-speed mode to further improve the positioning accuracy and ensure that the feeding bin can accurately reach the target position. Its position control accuracy can reach ≤5cm, and the speed adjustment range is 0.3-1m / s. The staff can adjust it within this range according to the actual feeding needs.

[0024] The feeding execution unit, suspended below the overhead track, is a key component for achieving precise feeding. It includes an electromagnetic pulse valve, a microwave level sensor, and a weight sensor. The electromagnetic pulse valve controls feed dispensing by receiving PWM pulse signals from a local terminal. The valve opening is controlled by the duty cycle of the PWM pulse signal, which determines the valve's opening time and the feed dispensing rate per unit time. Operators can adjust the PWM pulse signal parameters through the decision control unit to precisely control the feeding amount. The opening time is determined by the formula... The calculation involves Q, the target feeding amount preset by the decision control unit based on pond parameters, K, the discharge coefficient, and A, the valve opening percentage. This formula, based on feed discharge patterns, correlates the target feeding amount with the discharge coefficient, pond area, time, and valve opening to accurately calculate the required valve opening time, ensuring precise feeding accuracy. The microwave level sensor is installed on the top of the inner wall of the feeding hopper. Its working principle involves emitting microwave signals and receiving reflected waves to calculate the distance to the feed surface. When the remaining feed level is detected to be below a preset threshold, the sensor generates an electrical signal and sends it to... The decision control unit triggers a feeding reminder, which promptly alerts staff to replenish feed and prevents insufficient feed from affecting the normal operation of feeding. The weight sensor is integrated into the load-bearing structure at the bottom of the feeding hopper. It measures the cumulative weight during the feeding process in real time by sensing changes in the weight of the feed and synchronizes the data to the local terminal. When the weight sensor detects that the amount of feed has reached ±2% of the target amount Q, the system will control the electromagnetic pulse valve to close, so that the error of a single feeding is ≤±100g, further ensuring the accuracy of the feeding amount. Through dual control, the consistency between the actual feeding amount and the target feeding amount is ensured.

[0025] The visualization analysis unit integrates a 4K high-definition waterproof camera and a collaborative computing subunit. The camera's optical axis vertically downwards covers the entire aquaculture pond, and the 4K high-definition resolution ensures clear and detailed images of the fish. The waterproof design allows it to adapt to the humid environment of the aquaculture pond. The collaborative computing subunit is connected to the camera via Ethernet, receives real-time image data, and performs fish behavior analysis. Its specific working steps are as follows: Image preprocessing: The original images captured by the camera are denoised, contrast enhanced, and distortion corrected to output standardized images. Denoising is to eliminate noise interference in the image and improve image quality. Contrast enhancement makes the fish school more distinct from the background, which is convenient for subsequent target recognition. Distortion correction is due to the inherent distortion of the camera lens. Correction can ensure the geometric accuracy of the image and provide a reliable image basis for fish school behavior analysis.

[0026] Model Construction: A fish behavior analysis AI model trained using the YOLOv8 algorithm is loaded. This model analyzes preprocessed images, outputs fish target data, and then calculates fish behavior characteristics, including fish density. Average swimming speed and the center coordinates of the food-grabbing area The construction process of the aforementioned AI model for fish behavior analysis is as follows: Dataset construction: Collect images of fish in aquaculture ponds containing different fish species, densities, and lighting conditions, and label the target bounding box coordinates and fish activity levels. Model training: The SGD optimizer was used, with an initial learning rate α = 0.01, momentum μ = 0.9, weight decay λ = 0.0005, and training epochs. In each iteration, the loss function is used. Optimize model parameters, where, The classification loss measures the accuracy of the model in classifying the fish population. The regression loss is used to measure the accuracy of the model's prediction of the coordinates of the target box for the fish swarm. The regression loss weight coefficient is used to balance the importance of classification loss and regression loss. Through a large number of training iterations, the model can accurately identify fish schools and analyze their behavioral characteristics. Data output link: The calculated fish behavior characteristics are packaged into JSON format data and sent to the decision control unit via Ethernet, with the transmission frequency matching the camera frame rate. This ensures that the decision control unit obtains the latest status of the fish in real time, providing timely data support for dynamic feeding.

[0027] During visual analysis, cameras continuously capture video streams from the aquaculture pond. A collaborative computing subunit processes and analyzes these streams, outputting real-time data on fish density, swimming speed, and feeding areas. This data provides a basis for the decision-making control unit to formulate feeding strategies. Among these, fish density... ,in, The number of fish detected by the model. The surface area of ​​the aquaculture pond. This is a correction factor used to compensate for the bias caused by incomplete detection of underwater fish populations. This deviation can be compensated for, making the calculated fish density more accurate; average swimming speed In the formula, , The first The fish in the Frame and the The center coordinates of the frame The formula calculates the displacement of each fish between two adjacent frames, combines the frame rate to calculate the fish's swimming speed, and finally takes the average to obtain the average swimming speed of the fish school, which can reflect the activity level of the fish school.

[0028] The decision control unit includes a local PLC controller and a cloud platform, responsible for coordinating the work of each module to achieve intelligent feeding. The local PLC controller receives data from the mobile positioning unit on the displacement of the feeding hopper, data from the feeding execution unit on the remaining feed and feed weight, and data from the visualization analysis unit on fish behavior. Based on this data, the PLC controller executes the feeding logic control, such as... Figure 2 As shown: First, images are received from the camera. Based on fish density, average swimming speed, and coordinates of the feeding area, combined with pond parameters, the target feeding amount Q is calculated. Then, based on the value of Q, the discharge coefficient K of the electromagnetic pulse valve, and the valve opening A, the amount of feed is calculated using the formula... The system calculates the valve opening time and controls the opening and closing of the electromagnetic pulse valve to achieve precise feeding. At the same time, the PLC controller also controls the operation of the servo drive motor based on the feeding bin position data fed back by the motion positioning unit, achieving precise positioning and speed adjustment of the feeding bin. The weight sensor monitors the feeding amount in real time and sends the data to the local PLC controller. When the feeding amount reaches ±2% of the target feeding amount Q, the PLC controller controls the electromagnetic pulse valve to close, ensuring that the single feeding error is ≤±100g.

[0029] The cloud platform communicates with the local PLC controller via Ethernet and mainly performs the following functions: Store historical feeding data, including feeding time, pond number, feed amount, fish status parameters, etc., and support data query and export by date and pond number. This historical data provides important basis for staff to analyze the growth pattern and feeding habits of fish, and helps to optimize the feeding plan. Provides a visual operation interface for the terminal: allows users to customize feeding plans, including feeding time, frequency, basic feed amount, etc. After the plan is set, it is sent to the local PLC controller and stored. Users can flexibly adjust the feeding plan according to different breeding stages and fish growth conditions to improve breeding efficiency. The system displays real-time fish population status dashboards and equipment operating parameters for each aquaculture pond, including track position, remaining feed, and motor temperature. Staff can monitor the condition of each pond and the operational status of the equipment through the cloud platform, enabling them to promptly identify and address any issues. Anomaly Alarm: When equipment malfunction or abnormal fish behavior is detected, the cloud platform sends dual alarms via SMS and mobile app, and automatically records the time of occurrence and related data. This feature can promptly alert staff to take action and minimize losses.

[0030] During the data closed-loop process, the system records feeding time, pond location coordinates, actual feed amount, and fish behavior data, and synchronizes this data to the cloud platform. The cloud platform analyzes and processes this data to generate aquaculture decision reports, providing aquaculture suggestions to staff and realizing a complete closed loop from data collection to decision support.

[0031] In summary, this embodiment demonstrates in detail the specific implementation process of an AI-driven visual fishery feeding control system. Through the flexible movement of the mobile positioning unit, the precise feeding control of the feeding execution unit, the fish behavior analysis of the visualization analysis unit, and the local and cloud collaborative management of the decision control unit, the system realizes a complete closed loop from fish status perception to precise feeding.

[0032] Example 2: Figure 1 As shown in Example 1, this example elaborates on the specific steps of a visual fishery feeding control system based on AI in positioning the feeding bin and controlling the feeding. The specific steps are as follows: (1) Develop a feeding plan Based on the actual breeding conditions such as the species, growth stage, and quantity of fish in the breeding ponds, the staff formulates a feeding plan on the decision control unit. The plan includes key information such as the feeding time, frequency, and basic feed amount for each breeding pond, providing overall guidance for subsequent feeding work.

[0033] (2) System loading pool slot list The decision control unit loads the pool position list from a preset storage area (local PLC controller storage or cloud platform synchronization). The pool position list contains the coordinate information of each preset position in the aquaculture pool, which is used to determine the target position of the track robot.

[0034] (3) Initialization of single-pool feeding cycle control The system initiates a single-pond feeding cycle control process, setting the initial value of the cycle count variable n to 1. n is used to sequentially select each preset position in the pond position list to achieve feeding coverage of different aquaculture pond areas.

[0035] (4) Retrieve the preset coordinates of the nth pool : The decision control unit retrieves the coordinates of the preset position of the nth aquaculture pond from the loaded pond position list based on the current cycle count n. .

[0036] (5) Track robot walking The decision control unit sends control commands to the servo drive motor of the mobile positioning unit. Based on the received commands, the servo drive motor drives the track robot to move along the aerial track through the gear and rack transmission mechanism. During the movement, the absolute encoder built into the servo drive motor collects the displacement data of the track robot in real time and feeds it back to the local PLC controller of the decision control unit so as to monitor and adjust the movement status in real time. Meanwhile, the track robot moves according to an S-shaped acceleration and deceleration curve. This curve allows the track robot to accelerate and decelerate smoothly during the start-up and stopping phases, avoiding the shaking of feed in the feeding bin due to sudden speed changes. When the track robot is 50cm away from the target preset position, it automatically switches to low speed mode to further improve positioning accuracy and ensure that it can accurately reach the target position.

[0037] (6) Reach the preset coordinates of the nth pool : Under the precise control of servo drive motors, the track robot moves along the aerial track and eventually reaches the preset coordinates of the nth aquaculture pond. At this point, the decision control unit will receive a feedback signal that the track robot has reached the target position, preparing for the subsequent feeding operation.

[0038] (7) Calculate the amount of feed in real time After the track robot reaches the target position, the visualization analysis unit starts working. The integrated 4K high-definition waterproof camera vertically downwards to collect images of the fish in the breeding pond. The collaborative computing subunit receives real-time image data and performs fish behavior analysis. An AI model for fish behavior analysis trained using the YOLOv8 algorithm is loaded to analyze preprocessed images, output fish target data, and then calculate fish behavior characteristics, including fish density. Average swimming speed and the center coordinates of the food-grabbing area ; Based on these fish behavior characteristics and the parameters of the aquaculture pond, the decision control unit calculates the target feeding amount Q.

[0039] (8) Open the solenoid valve to feed. The decision control unit, based on the calculated target feeding amount Q, the discharge coefficient K of the electromagnetic pulse valve, and the valve opening A, calculates the opening time t of the electromagnetic pulse valve and sends a control signal to the electromagnetic pulse valve of the feeding execution unit to open the valve for feeding. The electromagnetic pulse valve controls its opening degree by receiving PWM pulse signals sent by the local terminal.

[0040] (9) Achieve the target feeding amount (Q±2%) During the feeding process, the weight sensor at the bottom of the feeding execution unit monitors the amount of feed in real time and synchronizes the data to the local PLC controller of the decision control unit. When the weight sensor detects that the amount of feed reaches ±2% of the target amount of feed Q, it will send a signal to the decision control unit.

[0041] (10) Close the valve and generate a feeding record. After receiving a signal that the amount of feed has reached the target range, the decision control unit immediately sends a closing command to the electromagnetic pulse valve to close the valve and stop feeding. At the same time, the system will automatically generate a record of the feeding, including the feeding time, pond location coordinates, actual feed amount, and fish behavior data, and synchronize this data to the cloud platform for subsequent querying, analysis, and generation of aquaculture decision reports.

[0042] (11) Determine whether the feeding plan has been completed. The system will check whether the current loop count n has traversed all the preset positions in the pool position list. If not, the value of n will be incremented by 1, and then return (4) to retrieve the coordinates of the next preset position and repeat the above process from the movement of the track robot to the generation of the feeding record. If all preset positions have been fed, proceed to the next step.

[0043] (12) The orbital robot returns to its starting position. After the entire feeding plan is completed, the decision control unit sends instructions to the mobile positioning unit to control the track robot to return to the starting position along the aerial track, preparing for the next feeding task.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An AI-driven based visualized fishery feeding control system, characterized in that, The system comprises a mobile positioning unit, a feeding execution unit, a visual analysis unit, and a decision control unit. The mobile positioning unit is used to construct an aerial track network across the aquaculture ponds, which is erected above the area of the aquaculture pond group, and is internally provided with a double-track guide structure and a servo drive motor to drive the feeding execution unit to move bidirectionally along the ±X / ±Y axes flexibly. The feeding execution unit is suspended below the track and comprises an electromagnetic pulse valve, a microwave material level sensor, and a weight sensor, which are used to control the valve opening and the discharging time through PWM control, and to monitor the remaining amount of feed in the feeding bin and the cumulative weight of the feed during the feeding process in real time. The visual analysis unit is integrated with a camera and a cooperative computing subunit, the camera optical axis is vertically downward to completely cover the water surface of a single aquaculture pond, and the cooperative computing subunit is loaded with a fish school behavior analysis AI model based on the YOLOv8 algorithm to output the fish school density, swimming speed, and feeding area data in real time. The decision control unit comprises a local PLC controller and a cloud platform, the PLC executes the feeding logic control, and the cloud platform supports remote feeding plan customization, real-time data dashboard, and abnormal alarm.

2. The AI-driven visualized fishery feeding control system according to claim 1, wherein In the mobile positioning unit, the aerial track is designed in a ring layout, the aerial track is internally provided with a double-track guide structure and a servo drive motor, the double-track guide structure is composed of two track beams arranged in parallel and guide sliders embedded in the inner sides of the track beams, and the guide sliders are connected with the feeding execution unit. The servo drive motor is connected with the track beams through a gear and rack transmission mechanism, an absolute value encoder is internally provided in the motor to collect displacement data and feedback in real time, the position control and speed adjustment of the feeding execution unit are realized, the position control accuracy is ≤5 cm, and the speed adjustment range is 0.3-1 m / s. The feeding execution unit is suspended below the track and comprises an electromagnetic pulse valve, a microwave material level sensor, and a weight sensor. The electromagnetic pulse valve controls the valve opening through the reception of a PWM pulse signal to adjust the discharging amount. The microwave material level sensor is installed on the top of the inner wall of the feeding bin to detect the remaining amount of feed and trigger a feeding reminder when the remaining amount is insufficient. The weight sensor is integrated at the bottom of the feeding bin to measure the cumulative weight of the feed in real time.

3. The AI-driven visualized fishery feeding control system according to claim 2, wherein, The electromagnetic pulse valve opening time is calculated by Q / K, wherein Q is a target feeding amount, K is a discharge coefficient, and A is a valve opening percentage; The weight sensor monitors the feed amount in real time, and the valve is closed when the Q value reaches ±2%, so that the single feeding error is ≤±100 g.

4. The AI-driven visualized fishery feeding control system according to claim 1, wherein, In the visual analysis unit, a camera and a cooperative computing subunit are integrated, and the camera optical axis is vertically downward to cover the whole area of the aquaculture pond. The cooperative computing subunit and the camera are connected through Ethernet to receive image data and perform fish school behavior analysis, and the process includes: Image preprocessing: denoising, contrast enhancement, and distortion correction are performed on the original image; Model inference: Load YOLOv8-based fish school behavior analysis AI model, output fish school density , average swimming speed , and food stealing area coordinates ; Data output: behavior feature data is packaged in JSON format and sent to the decision control unit.

5. The AI-driven visualized fishery feeding control system according to claim 4, wherein, The construction process of the fish school behavior analysis AI model is as follows: Data set construction: fish school image of aquaculture ponds containing different fish species, density, and lighting conditions is collected, and fish school target box coordinates and fish school activity labels are labeled; Model training: SGD optimizer, initial learning rate a = 0.01, momentum u = 0.9, weight decay l = 0.0005, training rounds , the loss function in each iteration Optimize model parameters, where, is the classification loss, is the regression loss, is the regression loss weight coefficient; Feature extraction: calculating fish density by the trained model , average swimming speed and foraging area center coordinates , wherein the fish density , wherein, is the number of fish detected by the model, is the water surface area of the culture pond, is a correction factor for compensating for the deviation that the underwater fish group is not completely detected, the average swimming speed , wherein, , are the center coordinates of the i-th fish in the j-th frame and the k-th frame, respectively, is the frame rate of the camera.​​​ 6. The AI-driven visualized fishery feeding control system according to claim 1, wherein, The decision control unit comprises a local PLC controller and a cloud platform. The local PLC controller establishes real-time communication with the mobile positioning unit, the feeding execution unit, and the visual analysis unit, and is used for: Receiving real-time position of the feeding execution unit to control its movement and positioning; Receiving microwave level sensor data to trigger a margin alarm; Receiving weight sensor data to control the electromagnetic pulse valve through PWM; Receiving fish school behavior data output by the visual analysis unit to generate a feeding instruction in combination with the parameters of the breeding pond.

7. The AI-driven visualized fishery feeding control system according to claim 1, wherein, The cloud platform is used for: Storing historical feeding data and supporting query and export; Providing a visual interface to customize a feeding plan; Real-time display of fish school status and device operation parameters; Alarm and record in case of device failure or fish school anomaly.

8. An AI-driven visual fishery feeding control method, applicable to the AI-driven visual fishery feeding control system of any one of claims 1-7, characterized in that, The specific steps of the method are: S1, loading a feeding plan and a preset position coordinate list of the breeding pond; S2, driving the feeding execution unit to move to a target position according to an S-shaped curve, and switching to a low-speed mode when approaching; S3, collecting a video stream, and outputting fish school behavior data through AI model analysis; S4, calculating the amount of feed and controlling the valve to open, and feeding back in real time through the weight sensor; S5, recording the feeding data and synchronizing to the cloud to generate a decision report.

Images