Fish pond precise feeding robot system based on YOLOv8

Abstract

The invention discloses a fishpond precise feeding robot system based on YOLOv8. The fishpond precise feeding robot system comprises a robot control module, a data acquisition and processing subsystem, a freshwater fish feeding data acquisition subsystem, a cloud server and a wireless information transmission subsystem. The system can intelligently judge the feeding opportunity and the feed amount by detecting the gathering condition of fish schools in real time, the feed utilization rate is remarkably improved, healthy growth of the fish schools is promoted, and therefore the overall breeding benefit is improved; due to automatic operation of the system, dependence on manual feeding is reduced, the labor cost is reduced, feed residues are reduced through precise feeding, the water quality is improved, and sustainable development of the breeding environment is maintained; the system has real-time monitoring and data analysis functions, can provide historical data records, supports decision optimization of breeding management, and realizes a modernized and intelligent breeding mode.

Description

Technical Field

[0001] The present invention relates to the field of intelligent breeding technology, including freshwater fish management technology, computer vision technology, intelligent control technology, artificial intelligence technology, network transmission technology and software development technology. Specifically, the invention relates to a fish pond precision feeding robot system based on YOLOv8. Background Art

[0002] With the rapid development of aquaculture, large-scale and intensive management of fish ponds has gradually become the mainstream. However, the traditional feeding method mainly relies on manual operation, and there are problems such as uneven feeding and feed waste, which affects the growth efficiency of fish and the stability of the water environment. In addition, there are significant differences in the feeding behavior and needs of different fish groups, and it is difficult to meet the needs of each fish pond through unified standardized feeding. Therefore, how to use intelligent means to accurately feed according to the actual needs of fish groups has become a key issue that needs to be solved in the field of aquaculture. Precision feeding can not only improve feed utilization, but also effectively reduce water pollution and improve the economic benefits of aquaculture.

[0003] In recent years, with the rapid development of artificial intelligence technology, the application of computer vision in the field of aquaculture has gradually emerged. In particular, the continuous advancement of target detection algorithms, such as the YOLO series of models, has shown excellent performance in processing real-time image data. YOLOv8 has high-precision and high-efficiency target detection capabilities, and can accurately identify and locate fish schools in complex backgrounds. In addition, automated control systems based on unmanned equipment have also been widely used in agriculture and fisheries. Combined with deep learning technology and automated robotic systems, accurate detection and feeding control of fish schools can be achieved. Therefore, the development of a fish pond precision feeding robot system based on YOLOv8, which can intelligently control feeding by detecting the distribution of fish schools, has important application value. Summary of the invention

[0004] In view of the problems in the background technology, the present invention constructs an intelligent feeding robot system based on fish gathering detection, collects fish gathering information, judges the fish feeding needs based on the model, and then gives the robot feeding decision to achieve accurate feeding of fish farming.

[0005] Technical solution:

[0006] A fish pond precision feeding robot system based on YOLOv8 includes a robot control subsystem, a feeding control subsystem, a data acquisition and processing subsystem, a cloud server, and a wireless transmission module, wherein:

[0007] Feeding robot subsystem: controls the robot to move along the track at the edge of the fish pond, and accurately controls the feeding of feed according to the aggregation and activity status of the fish school to avoid feed waste or shortage;

[0008] Image acquisition subsystem: used to collect videos of fish feeding;

[0009] Data acquisition and processing subsystem: Use the YOLOv8 model to process fish school images in real time, obtain the number, location and distribution of fish, and conduct comprehensive analysis in combination with water quality parameters;

[0010] Cloud server: stores and analyzes fish data and water quality data, generates feeding plans, and supports remote monitoring and operation;

[0011] Furthermore, the feeding robot subsystem includes a core control board module, a power supply module, a feeding module, a photoelectric sensor module, a drive module, a wireless information transmission module, an image acquisition module and a track, wherein:

[0012] The core control board module is a single-chip control board with STM32F103ZET6 as the main chip, which includes a power supply interface, a 485 interface, and a control signal sending I / O port. The core control board plays the role of stopping the robot's movement, providing 3.3V or 5V power supply for external devices, and uploading and receiving command data in the feeding robot;

[0013] The power supply module is a 48V output rechargeable lithium battery, which is responsible for supplying power to the core control board module, drive module, photoelectric sensor module and wireless information transmission module of the feeding robot;

[0014] The driving module includes a relay and a motor. After receiving the signal from the core control board, the relay closes, causing the motor to rotate and drive the feeding robot to move.

[0015] The feeding module consists of a silo, a fan, a spiral shaft, spiral blades, a relay and a discharge barrel. The relay closes after receiving a signal from the core control board, driving the spiral shaft to rotate. The feed is sent into the discharge barrel through the spiral shaft and spiral blades and is sent out by the fan to achieve feed spreading.

[0016] The photoelectric sensor module consists of two photoelectric sensors installed at the front and rear ends of the robot, which can sense obstacles at close range and transmit signals to the STM32 core control board to control the robot to stop moving when encountering obstacles.

[0017] The wireless information transmission module includes a serial port server and a WIFI box, which is powered by a power supply module and has an RS485 interface. It can connect to the wireless signal of the WIFI box, receive instructions sent by the cloud server, and convert them into RS485 signals and send them to the STM32 microcontroller.

[0018] Use a step-down device to convert DC 48V into DC 12V to power the wireless information transmission module;

[0019] The track is made of 40mm*40mm standard angle iron parts laid and connected, and is laid on the shore of one side of the fish pond.

[0020] Furthermore, the image acquisition subsystem consists of a Hikvision camera, a WIFI box, and a cloud server. It is installed on the right side of the robot's forward direction, at a height of 70 cm, and shoots at a 70-degree angle to the ground. The image resolution is 1920×1080, which is used to shoot images of fish feeding and fish aggregation. The camera data is transmitted to the cloud server in real time through WIFI signals.

[0021] Preferably, the system also includes a data display module for users to view current feeding control information, image data and historical data saved by the cloud server.

[0022] Specifically, the data acquisition and processing module constructs a standardized freshwater fish precision feeding database through the following steps:

[0023] S1-1 extracts a set of RGB fish school images at regular intervals and selects high-quality images for analysis;

[0024] S1-2 constructs a Gaussian filter to denoise the image and eliminate water impurities and camera noise. The formula is as follows:

[0025]

[0026] Where I(x,y) is the denoised image, I0(x,y) is the original image, and σ is the standard deviation of the filter;

[0027] S1-3 converts the denoised image from RGB space to HSV color space to separate the fish and the water background. The conversion formula is as follows:

[0028] V=max(R,G,B)

[0029]

[0030] Among them, H represents hue, which is defined according to the different pixel colors; S represents saturation, and V represents brightness;

[0031] S1-4 For the obtained HSV color space image, the background subtraction method is used to extract the foreground of the fish based on the dynamic characteristics of the fish and the static background of the water; the background modeling formula is as follows:

[0032] M t (x,y)=αI t (x,y)+(1-α)M t-1 (x,y)

[0033] Among them, M t(x, y) is the background model of the current frame, I t (x, y) is the current image frame, α is the update rate;

[0034] S1-5 performs morphological processing on the binarized fish foreground image, applies dilation and erosion operations, eliminates small noise points and connects the fish body parts; the specific formula is as follows:

[0035]

[0036] Among them, I is the binary image, B is the structural element;

[0037] S1-6 uses the aggregation formula to calculate the aggregation degree of the fish school in the image obtained in step S1-5 to obtain its characteristic parameters;

[0038] S1-7 repeats steps S1-1 to S1-6, extracts the obtained characteristic parameters to construct a freshwater fish precise feeding database, and ends after the data collection is completed.

[0039] Specifically, in step S1-6, clusters of fish are obtained by clustering, and multiple characteristic parameters such as fish density, fish number and average distance are obtained through the clusters, where each cluster represents a group of relatively concentrated fish. The characteristic parameter calculation steps are as follows:

[0040] S1-6-1 Based on the bounding box coordinates output by YOLOv8, the contour area of ​​the fish school in the image is calculated through contour detection; the area of ​​each contour is defined as A i , add up all contour areas to get the total clustering area:

[0041]

[0042] Where N represents the number of detected contours.

[0043] S1-6-2 Analyze the spatial distribution of fish in the fish pond based on their location coordinates;

[0044] S1-6-3 According to the aggregation area and density of fish schools, the aggregation value is obtained:

[0045]

[0046] Among them, A image is the area of ​​the entire image region. c is the density obtained by clustering. This aggregation degree is used for subsequent feeding decisions, that is, when the fish aggregation degree exceeds a certain threshold, the system will determine that the fish in the area are hungry and trigger the feeding action.

[0047] In S1-6-2, the spatial distribution of the fish school in the fish pond is analyzed, namely, the centroid position, the radius and density of the cluster, and the calculation steps are as follows:

[0048] 1) Calculate the centroid C of each cluster c Position, which indicates the average position of the fish cluster in the pond:

[0049]

[0050] Among them, x i and i is the coordinate of the i-th fish, and the distribution center of the fish school is obtained;

[0051] 2) Cluster radius R e It is defined as the average distance from all fish in a cluster to the centroid of the cluster, reflecting the spatial distribution range and compactness of the fish school:

[0052]

[0053] in, and are the coordinates of the cluster centroid; the smaller the radius, the higher the degree of aggregation of the fish school;

[0054] 3) Density of fish school D c is the number of fish N c The ratio between the area of ​​the cluster and the area it covers. For a cluster that is approximately circular, the density is defined as:

[0055]

[0056] Density indicates how compactly a school of fish is clustered.

[0057] Specifically, the data acquisition and processing subsystem constructs a fish school detection model through the following steps:

[0058] S2-1 is a dataset of annotated fish images, which contains the location, size and category information of each fish;

[0059] S2-2 uses different clustering methods to train as a fish detection model;

[0060] S2-3 uses precision, recall and F1 value as evaluation indicators to determine the optimal model as the fish detection model.

[0061] Specifically, in step S2-2, different clustering methods include: K-means clustering, DBSCAN clustering, and Euclidean clustering.

[0062] Specifically, in step S2-3, the calculation method of each evaluation index is as follows:

[0063]

[0064]

[0065] Beneficial effects of the present invention

[0066] (1) By real-time detection of fish aggregation, the system can intelligently determine the timing and amount of feeding, significantly improve feed utilization, promote the healthy growth of fish, and thus improve the overall breeding efficiency;

[0067] (2) The automated operation of the system reduces reliance on manual feeding, reduces labor costs, and reduces feed residues through precise feeding, improves water quality, and maintains the sustainable development of the aquaculture environment;

[0068] (3) The system has real-time monitoring and data analysis functions. The system can provide historical data records, support decision-making optimization of breeding management, and realize a modern and intelligent breeding model.

[0069] (4) The accuracy of the amount of freshwater fish fed is closely related to the living environment and health status of freshwater fish. The method provided by the present invention obtains the number of freshwater fish, aggregation and other information in a non-contact manner, controls the amount of freshwater fish fed, and is beneficial to improving the welfare level of freshwater fish. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The system block diagram of the present invention is

[0071] Figure 2 The system workflow diagram of the present invention is

[0072] Figure 3 Schematic diagram of the feeding robot module structure

[0073] Figure 4 Concept design of the feeding robot

[0074] Figure 5 System block diagram of fish gathering detection system DETAILED DESCRIPTION

[0075] The present invention will be further described below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto:

[0076] Combination Figure 1 , Figure 2 The present invention provides a fish pond precision feeding robot system based on YOLOV8, which includes a robot control module (core control board module, power supply module, feeding module, photoelectric sensor module, drive module, wireless information transmission module, image acquisition module and track), image acquisition subsystem, data acquisition and processing subsystem, and cloud server, among which:

[0077] Feeding robot subsystem: controls the robot to move along the track at the edge of the fish pond, collects fish school images and water quality data in real time, and ensures comprehensive monitoring of the fish pond; accurately controls the feeding of feed according to the aggregation and activity status of the fish school to avoid feed waste or shortage;

[0078] Image acquisition subsystem: used to collect fish school videos, including fish school feeding videos and fish school aggregation videos; fish school videos are captured at a fixed frequency to obtain fish school images;

[0079] Data acquisition and processing subsystem: Build a standardized freshwater fish precision feeding database, use the YOLOv8 model as a standardized freshwater fish precision feeding database to train a fish school detection model; process fish school images in real time, obtain the number, location and distribution of fish based on the fish school detection model, and conduct a comprehensive analysis in combination with water quality parameters;

[0080] Cloud server: stores and analyzes fish data and water quality data, generates feeding plans, and supports remote monitoring and operation.

[0081] Combination Figure 2 and Figure 4 , a system workflow diagram of the present invention is given. The method can solve the difficulty of calculating the number of fish when the fish are feeding, so as to achieve accurate feeding amount of feed, reduce fish pond pollution, and improve the welfare level of freshwater fish.

[0082] Combination Figure 3 This application designs an intelligent feeding robot. The robot is powered by a rechargeable lithium battery, moves on a fixed track, and receives instructions from the host computer through the wireless information transmission module to control the feeding machine to perform feeding tasks. Including:

[0083] The driven wheel 1 is located under the vehicle and is mainly used to support the weight of the vehicle body;

[0084] Bracket 2, used to fix the silo and the feeder;

[0085] Silo 3, for loading feed;

[0086] The feeding machine 4 is operated by sending instructions to the feeding machine through the control module to feed the fish pond;

[0087] A discharge tube 5, through which the bait is ejected from the feeding machine and then passes;

[0088] A support platform 6 is located below the feeding machine, and has four circular openings on the platform for fixing the feeding machine at the four corners of the feeding machine;

[0089] Charging port 7, for charging the battery of the feeding robot;

[0090] The power supply module 8 uses a 48V rechargeable lithium battery and is responsible for supplying power to the core control board module, drive module, feeding module, photoelectric sensor module, and wireless information transmission module of the feeding robot;

[0091] The driving wheel 9 is under the vehicle body and drives the robot forward;

[0092] The motor 10 is connected to the relay, which closes after receiving the control signal from the host computer, causing the motor to rotate and realize the robot moving forward;

[0093] The waterproof box 11 includes a core control board module and a wireless information transmission module. The core control board module uses the STM32F103ZET6 minimum system board as the core, and is equipped with a power supply interface, a 485 interface, a sensor interface, etc. The control board plays a role in controlling the movement and stop of the robot, and receiving and converting the control instructions of the host computer in the feeding robot; the wireless information transmission module is used to connect to the wireless network, and is connected to the control board module through the 485 interface to realize remote communication between the server and the control system;

[0094] Image acquisition module 12 uses Hikvision CS-C2C-32WFR camera, which is installed on the right side of the robot's forward direction. It is shot at a 70-degree vertical angle to the ground at a height of 70 cm. The image resolution is 1920×1080. It is used to capture images of fish feeding and fish aggregation. The camera data is transmitted to the cloud server in real time through WIFI signals.

[0095] The photoelectric sensor modules 13 are installed on the front and rear sides of the robot respectively. After sensing the iron plates or other obstacles at both ends of the track, they send pulse signals. After the control board module captures them, it controls the robot to stop.

[0096] Combination Figure 5 In this embodiment, the fish gathering detection system includes a Hikvision camera (1920*1080), a local host, a switch, an NVR and a cloud database. A Hikvision camera (CS-C2C-32WFR) is installed at a height of 60 cm in the feeding direction of the feeding robot. From 1 minute before each feeding operation to the end of the feeding operation, the fish in the fish pond are photographed vertically at 45 degrees to the ground. The video data is saved in real time to the hard disk and uploaded to the cloud server.

[0097] Combination Figure 2 , the complete steps of model construction are described below with a specific example:

[0098] S1 Data Collection: Based on the fish gathering detection system, the video is obtained, and the image data is captured from the video in real time at a fixed frequency, with a capture frequency of three frames per second.

[0099] S2 dataset construction: The captured images are preprocessed using image processing methods to obtain the training set;

[0100] The specific image processing steps are as follows:

[0101] S2-1 uses the constructed Gaussian filter to perform noise reduction for blur and noise problems;

[0102] S2-2 In order to separate the fish from the background of the water, the image is first converted from the RGB space to the HSV color space, and then the background subtraction method is used to extract the foreground of the fish based on the dynamic characteristics of the fish and the static background of the water.

[0103] S2-3 then applies dilation and erosion operations to the binarized image to eliminate small noise points and connect the fish body parts.

[0104] S2-4 uses the aggregation formula to calculate the density of the fish school in the image and obtains its characteristic parameters.

[0105] S3 network design: balances network speed and detection accuracy, and uses YOLOv8 as the detection network model.

[0106] S4 YOLOv8 model training: Use the dataset images to train the YOLOv8 model as a fish gathering target detector and perform cluster analysis. According to the freshwater fish gathering density, the following evaluation criteria are obtained:

[0107] 1) Low Aggregation: Freshwater fish aggregation values ​​are less than 0.2, where the fish are dispersed over a larger area, indicating a low degree of aggregation, which usually corresponds to lower feed requirements.

[0108] 2) Medium aggregation: The aggregation value of freshwater fish is between 0.2-0.5. At this time, the fish school is aggregated in a certain area, indicating that the feed demand is general.

[0109] 3) High aggregation: The aggregation value of freshwater fish is between 0.5-0.8. At this time, the fish are highly concentrated in a small area, indicating obvious aggregation behavior, which usually corresponds to high feed demand.

[0110] 4) Excessive aggregation: The aggregation value of freshwater fish is greater than or equal to 0.8. At this time, the fish school is highly concentrated in a very small area, indicating extremely strong aggregation behavior, which usually corresponds to extremely high feed requirements.

[0111] S5 Statistical analysis and visualization: The fish aggregation detection results are transmitted to the backend database, searched by SQL statements, published to the front-end visualization page and synchronized to the feeding robot.

[0112] Specifically, the freshwater fish aggregation assessment information is fed back to the feeding robot, and the feeding robot performs accurate feeding according to the preset conditions, which specifically includes the following steps:

[0113] S1. When preparing for feeding, the feeding robot subsystem receives the start command through the WIFI signal, and the single-chip control board controls the output interface, controls the drive transmission module and further controls the movement of the feeding robot.

[0114] S2: Wait for the robot to go through the preset moving time and arrive at the preset feeding point, and the microcontroller will control the robot to stop at a fixed point.

[0115] S3. When feeding begins, the single-chip microcomputer controls the relay on and off to control the feeding port of the feeding machine to open or close, and controls the on and off time of the relay to achieve the expected feeding strategy amount of freshwater fish feed.

[0116] S4. During feeding, the freshwater fish feeding data collection subsystem will be turned on at the same time. The camera will collect the freshwater fish feeding data before and after feeding. The post-feeding image is used to record the gathering position of the fish school, and also record the feeding time and feeding amount.

[0117] Specifically, the feeding strategy in S4 is as follows: the feeding amount is related to the total weight of the freshwater fish in the fish pond. By presetting the relevant value of the feeding amount and the total weight and the average weight of the fish in the pond, different functions are selected according to the aggregation degree of the freshwater fish to calculate the feeding amount. The feeding time is calculated according to the feeding amount per second when the feeding port is opened.

[0118] Specifically, the above freshwater fish feeding amount function is:

[0119]

[0120] The default feeding rate is 2% (i.e., 2g of feed is required for every 100g of fish per day), which can be adjusted in real time in the background according to seasonal weather.

[0121] Specifically, the above feeding duration function is:

[0122]

[0123] Preferably, it also includes a freshwater fish feeding management web platform and an Android terminal, which technically interact with the front and back ends to achieve real-time data updating and visualization.

[0124] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A fish pond precision feeding robot system based on YOLOv8, characterized in that It includes robot control module, image acquisition subsystem, data acquisition and processing subsystem, and cloud server, among which: Feeding robot subsystem: controls the robot to move along the track at the edge of the fish pond, collects fish school images and water quality data in real time, and ensures comprehensive monitoring of the fish pond; accurately controls the feeding of feed according to the aggregation and activity status of the fish school to avoid feed waste or shortage; Image acquisition subsystem: used to collect fish school videos, including fish school feeding videos and fish school aggregation videos; fish school videos are captured at a fixed frequency to obtain fish school images; Data acquisition and processing subsystem: Build a standardized freshwater fish precision feeding database, use the YOLOv8 model as a standardized freshwater fish precision feeding database to train a fish school detection model; process fish school images in real time, obtain the number, location and distribution of fish based on the fish school detection model, and conduct a comprehensive analysis in combination with water quality parameters; Cloud server: stores and analyzes fish data and water quality data, generates feeding plans, and supports remote monitoring and operation.

2. The system according to claim 1, characterized in that The feeding robot subsystem includes a core control board module, a power supply module, a feeding module, a photoelectric sensor module, a drive module, a wireless information transmission module, an image acquisition module and a track, among which: The core control board module is a single-chip control board with STM32F103ZET6 as the main chip, which includes a power supply interface, a 485 interface, and a control signal sending I / O port. The core control board plays the role of stopping the robot's movement, providing 3.3V or 5V power supply for external devices, and uploading and receiving command data in the feeding robot; The power supply module is a 48V output rechargeable lithium battery, which is responsible for supplying power to the core control board module, drive module, photoelectric sensor module and wireless information transmission module of the feeding robot; The driving module includes a relay and a motor. The relay closes after receiving a signal from the core control board, causing the motor to rotate and drive the feeding robot to move. The feeding module consists of a silo, a fan, a spiral shaft, spiral blades, a relay and a discharge barrel. The relay is closed after receiving a signal from the core control board, driving the spiral shaft to rotate. The feed is sent into the discharge barrel through the spiral shaft and spiral blades and sent out by the fan to achieve feed spreading. The photoelectric sensor module consists of two photoelectric sensors installed at the front and rear ends of the robot, which can sense obstacles at close range and transmit signals to the STM32 core control board to control the robot to stop moving when encountering obstacles. The wireless information transmission module includes a serial port server and a WIFI box, which is powered by a power supply module and has an RS485 interface. It can connect to the wireless signal of the WIFI box, receive instructions sent by the cloud server, and convert them into RS485 signals and send them to the STM32 microcontroller. A step-down converter is used to convert the DC 48V into DC 12V to supply power to the wireless information transmission subsystem block; The track is made of 40mm*40mm standard angle iron parts laid and connected, and is laid on the shore of one side of the fish pond.

3. The system according to claim 1, characterized in that The data acquisition and processing subsystem constructs a standardized freshwater fish precision feeding database through the following steps: S1-1 extracts a set of RGB fish school images at regular intervals and selects high-quality images for analysis; S1-2 constructs a Gaussian filter to denoise the image and eliminate water impurities and camera noise. The formula is as follows: Where I(x,y) is the denoised image, I0(x,y) is the original image, and σ is the standard deviation of the filter; S1-3 converts the denoised image from RGB space to HSV color space to separate the fish and the water background. The conversion formula is as follows: V=max(R,G,B) Among them, H represents hue, which is defined according to the different pixel colors; S represents saturation, and V represents brightness; S1-4 For the obtained HSV color space image, the background subtraction method is used to extract the foreground of the fish based on the dynamic characteristics of the fish and the static background of the water; the background modeling formula is as follows: M t (x,y)=αI t (x,y)+(1-α)M t-1 (x,y) Among them, M t (x, y) is the background model of the current frame, I t (x, y) is the current image frame, α is the update rate; S1-5 performs morphological processing on the binarized fish image, applies dilation and corrosion operations, eliminates small noise points and connects the fish body parts; the specific formula is as follows: Dilation: Erosion: Among them, I is the binary image, B is the structural element; S1-6 calculates the aggregation degree of the fish school in the image obtained by S1-5 to obtain its characteristic parameters; S1-7 repeats steps S1-1 to S1-6, extracts the obtained characteristic parameters to construct a freshwater fish precise feeding database, and ends after the data collection is completed.

4. A system according to claim 1 or 3, characterized in that The image acquisition subsystem consists of a Hikvision camera, a WIFI box, and a cloud server. It is installed on the right side of the robot's forward direction, at a height of 70 cm, and shoots at a 70-degree angle to the ground. The image resolution is 1920×1080, which is used to capture images of fish feeding and the aggregation of fish. The camera data is transmitted to the cloud server in real time through WIFI signals. The track is made of standard angle iron parts laid and connected, and is laid on the shore on one side of the fish pond.

5. The system according to claim 3, characterized in that By using clustering, we can get the clusters where fish gather, and get multiple characteristic parameters such as fish density, fish number and average distance through the clusters, where each cluster represents a group of relatively concentrated fish. The characteristic parameter calculation steps are as follows: S1-6-1 Based on the bounding box coordinates output by YOLOv8, the contour area of ​​the fish school in the image is calculated through contour detection; the area of ​​each contour is defined as A i , add up all contour areas to get the total clustering area: Where N represents the number of detected contours; S1-6-2 Analyze the spatial distribution of fish in the fish pond based on their location coordinates; S1-6-3 According to the area and density of the fish gathering area, the aggregation value is obtained: Among them, A image is the area of ​​the entire image region, D c is the density obtained by clustering, and the aggregation degree G is used for subsequent feeding decisions.

6. The system according to claim 5, characterized in that In S1-6-2, the spatial distribution of the fish school in the fish pond is analyzed, namely, the centroid position, the radius and density of the cluster, and the calculation steps are as follows: 1) Calculate the centroid C of each cluster c Position, which indicates the average position of the fish cluster in the pond: Among them, x i and i is the coordinate of the i-th fish, and the distribution center of the fish school is obtained; 2) Cluster radius R e It is defined as the average distance from all fish in a cluster to the centroid of the cluster, reflecting the spatial distribution range and compactness of the fish school: in, and are the coordinates of the cluster centroid; the smaller the radius, the higher the degree of aggregation of the fish school; 3) Density of fish school D c is the number of fish N c The ratio between the area of ​​the cluster and the area it covers. For a cluster that is approximately circular, the density is defined as: Density indicates how compactly a school of fish is clustered.

7. The system according to claim 1, characterized in that The data acquisition and processing subsystem builds a fish detection model through the following steps: S2-1 is a dataset of annotated fish images, which contains the location, size and category information of each fish; S2-2 uses different clustering methods to train as a fish detection model; S2-3 uses precision, recall and F1 value as evaluation indicators to determine the optimal model as the fish detection model.

8. The system according to claim 7, characterized in that In step S2-2, different clustering methods include: K-means clustering, DBSCAN clustering, and Euclidean clustering.

9. The system according to claim 7, characterized in that In step S2-3, the calculation method of each evaluation index is as follows: Among them, TP represents true positive examples, FP represents false positive examples, FN represents false negative examples, and F1 value is the harmonic mean of precision and recall.

10. The system according to claim 1, characterized in that The cloud server generates a feeding plan, and the feeding duration function is: Among them, the feeding amount of freshwater fish is obtained by the following formula: Among them: low aggregation: the aggregation value of freshwater fish is less than 0.2; medium aggregation: the aggregation value of freshwater fish is between 0.2-0.5; high aggregation: the aggregation value of freshwater fish is between 0.5-0.8; excessive aggregation: the aggregation value of freshwater fish is greater than or equal to 0.8.

Images