A zoned fish farming device with electric fence and robot collaboration

Abstract

The present invention relates to a zoned fish farming device in which an electric fence and a robot cooperate, and belongs to the field of aquaculture. The device is composed of a main control module, a pulse fish-driving module, an above-water robot module, and an underwater robot module. The water area is divided into different fish-farming areas by using an electric fence, and the voltage of the electric fence is adjusted to restrict the cross-zone swimming of fish. Above-water and underwater robots are deployed in the buffer zone to work together, and the underwater robot monitors whether there are fish that mistakenly enter the buffer zone through a three-dimensional sensor, a depth sensor, and an infrared camera, and monitors the position of the fish. The above-water robot transmits the monitoring data to the main control module through a WiFi module, and the infrared camera on the shore captures the LED light signal of the above-water robot to determine its two-dimensional position, and tracks the fish in combination with the three-dimensional position information of the underwater robot. The main control module mobilizes the robot to drive the fish to the fishing area, and classifies and manages the fish according to the fish species. The present invention can realize the zoned farming of multiple fish species in the same water area, saving resources and costs.

Description

Technical Field

[0001] The invention belongs to the field of aquaculture, and relates to a zoned fish farming device coordinated by an electric fence and a robot. Background Art

[0002] With the rapid development of aquaculture, aquaculture efficiency and resource utilization have become the focus of the industry. The traditional aquaculture model has the following problems:

[0003] (1) Low resource utilization

[0004] Traditional aquaculture models usually mix different fish species in the same area, resulting in excessive stocking density, deteriorating water quality, slow fish growth, and even disease.

[0005] In order to avoid mixing of fish species, the water area needs to be divided into multiple areas, which causes waste of water resources and increases the difficulty of management.

[0006] (2) Difficulty in cost control

[0007] The traditional breeding model requires a large number of manpower to carry out inspection, feeding, fishing and other tasks, and the labor cost remains high.

[0008] Although the use of high-precision equipment such as sonar for fish monitoring is effective, it is costly and difficult to popularize in small and medium-sized farms.

[0009] (3) Difficulty in management

[0010] Traditional farming models make it difficult to monitor fish activities in real time and are unable to detect abnormal situations in a timely manner, such as fish mistakenly entering other areas or becoming sick.

[0011] Traditional fishing methods are time-consuming, labor-intensive, inefficient, and can easily cause harm to fish.

[0012] (4) Environmental pollution risks:

[0013] Traditional breeding models can easily lead to eutrophication of water bodies and cause environmental pollution.

[0014] Polyculture of fish can easily lead to the spread of diseases, causing economic losses and environmental pollution.

[0015] In response to the above problems, there is an urgent need for an intelligent, low-cost zoning aquaculture system that can achieve efficient management of multiple fish species in the same waters, improve resource utilization, reduce aquaculture costs, reduce manpower requirements and management difficulties, and reduce environmental pollution risks. Summary of the invention

[0016] In view of this, the purpose of the present invention is to provide a zoned fish farming device that cooperates with an electric fence and a robot, combining the physical isolation of the pulse electric fence with the automatic monitoring and fish-driving functions of the intelligent robot to achieve zoned farming of different fish. The cross-regional management of fish is achieved by adjusting the electric fence voltage. When fish mistakenly enter the buffer zone, the robot can automatically identify and guide the fish back to the designated area through path planning. The device not only improves the utilization rate of water resources and reduces farming costs, but also reduces manpower requirements and management difficulties. It is particularly suitable for aquaculture environments that require mixed farming of different fish species and has broad application prospects.

[0017] In order to achieve the above object, the present invention provides the following technical solutions:

[0018] A zoned fish farming device using electric fences and robots in collaboration, comprising:

[0019] The high-voltage pulse device 1 is connected to an electric fence 14, and the electric fence 14 is suspended by a steel wire rope 15, and is used to divide the water area into at least two fish farming areas and at least one buffer zone;

[0020] Water robot module 7, including:

[0021] The water robot WiFi module is connected to the main control module 2 and is used to communicate with the main control module;

[0022] A power module, providing power to the water robot module 7;

[0023] Four thrusters 8, front, back, left, and right, control the movement of the water robot module 7 in a two-dimensional plane;

[0024] LED light 5, used to display the position of the water robot module 7;

[0025] The buoy 6 provides the ability to float on water;

[0026] Underwater robot module 9, including:

[0027] A three-dimensional sensor for collecting data from the underwater robot module 9;

[0028] A depth sensor 13, used to collect underwater depth data;

[0029] A lifting device controls the movement and posture adjustment of the underwater robot module 9 in a direction perpendicular to the two-dimensional plane; an underwater front infrared camera 11 and an underwater rear infrared camera 12 are used to monitor whether fish have mistakenly entered the buffer zone and their location information;

[0030] Main control module 2, including:

[0031] a display screen for displaying information about the location of fish and the robot;

[0032] A main control WiFi module is connected to the water robot module 7 and is used to communicate with the water robot module 7;

[0033] A shore charging device 3, used to charge the robot;

[0034] The shore infrared camera 4 is used to capture the LED light 5 of the water robot module 7 and determine its two-dimensional plane coordinates.

[0035] Furthermore, the electric fence 14 is arranged between the fish farming area and the buffer area.

[0036] Furthermore, the high-voltage pulse device 1 adjusts the pulse voltage of the electric grid 14 according to the reaction intensity of different fish to the voltage, so as to prevent different fish from swimming across zones.

[0037] Furthermore, the above-water robot module 7 and the underwater robot module 9 are connected via a connecting cable 12 .

[0038] Furthermore, the main control module 2 determines the relative position of the fish and the robot according to the two-dimensional plane coordinates of the above-water robot module 7 and the three-dimensional position data transmitted by the underwater robot module 9 .

[0039] Furthermore, the main control module 2 calls a preset path planning algorithm according to the relative positions of the fish and the robot, controls the propeller 8 of the water robot module 7 to adjust its moving direction, and at the same time directs the underwater robot module 9 to adjust its depth through the lifting device, so as to guide the fish that have mistakenly entered the buffer zone to the preset fishing fish box 19.

[0040] Furthermore, the main control module 2 integrates a shore infrared camera 4 for capturing the LED light 5 of the water robot module 7 and determining its position in a two-dimensional plane.

[0041] Furthermore, the main control module 2 is integrated with a shore charging device 3 for charging the robot.

[0042] Furthermore, the underwater robot module 9 is configured with a PP_LiteSeg lightweight semantic segmentation model for real-time recognition and classification of fish.

[0043] The working process of the present invention is completed through the coordination of the main control module, the electric grid system, and the robot module, realizing the automated management of different fish species in zoned farming. The specific working principle is as follows:

[0044] 1. Startup of the main control module and electric fence system: When the main control module is powered on, the pulse fish-driving module is started synchronously. The pulse electric fence forms an electric barrier between the first fish farming area, the second fish farming area and the buffer zone, and controls the cross-zone behavior of fish according to the preset voltage to ensure that each fish stays in its own farming area.

[0045] 2. Underwater monitoring and positioning: The underwater robot is responsible for monitoring whether there are fish that have mistakenly entered the buffer zone within its coverage area. Its infrared camera works continuously, and the deployed PP_LiteSeg lightweight semantic segmentation model is used to identify and classify fish in real time. If a fish crosses the electric fence and mistakenly enters the buffer zone, the underwater robot determines the three-dimensional coordinates and depth position of the fish through a three-dimensional sensor and a depth sensor. The robot transmits the detected fish position data and identification information to the surface robot via a cable.

[0046] 3. Data transmission and position locking: After receiving the data from the underwater robot, the water robot uses its built-in WiFi module to transmit the data to the main control module. At the same time, the infrared camera on the shore captures the LED light signal of the water robot to determine the position of the water robot in the two-dimensional plane of the breeding area. The main control module combines the two-dimensional coordinates of the water robot and the three-dimensional position data transmitted by the underwater robot to determine the precise relative position of the fish and the robot.

[0047] 4. Path planning and fish chasing operation: Based on the relative position of the fish and the robot, the main control module calls the preset path planning algorithm, controls the propeller of the water robot to adjust its moving direction, and commands the underwater robot to adjust its depth through the lifting device. Under the coordination of the main control module, multiple robots work together to gradually guide the fish that have strayed into the buffer zone to the preset fishing box.

[0048] 5. Fish classification and management: When the fish are successfully driven into the fishing box, the staff will classify them according to the characteristics of different fish species and put them back to the corresponding first fish farming area or second fish farming area to avoid mixed breeding of fish and ensure the accuracy of zoning management.

[0049] 6. Robot charging management: When the robot module detects that the power level is lower than the set threshold, the main control module sends a charging command to it, and the robot automatically returns to the charging device on the shore for charging to ensure that it can continue to work.

[0050] The beneficial effects of the present invention are:

[0051] (1) Improve resource utilization

[0052] Electric fences are used to divide the water area into multiple areas to achieve zoning breeding of different fish species, avoid mixing of fish species, increase breeding density and save water resources.

[0053] The electric grid voltage can be flexibly adjusted according to the growth habits and breeding needs of different fish species to achieve more refined zoning management.

[0054] (2) Reduce breeding costs

[0055] The robots automatically complete tasks such as inspection, feeding, and fishing, greatly reducing labor costs.

[0056] Compared with high-precision equipment such as sonar, electric fences are cheaper and easier to maintain.

[0057] (3) Reduce manpower requirements and management difficulty

[0058] The robot automatically monitors fish activities, detects abnormal situations in a timely manner, and handles them, reducing the workload of manual inspections.

[0059] The main control module can monitor the status of fish and robots in real time, and perform intelligent scheduling and control to reduce management difficulty.

[0060] (4) Improve breeding efficiency

[0061] According to the needs of different fish species, precise feeding is carried out to improve feed utilization and promote fish growth.

[0062] Robots work together to quickly drive fish into fishing nets, improving fishing efficiency.

[0063] (5) Reduce environmental pollution risks

[0064] Zoning breeding avoids mixing of fish species and reduces water pollution.

[0065] Avoid contact with fish species to reduce the risk of disease transmission.

[0066] (6) Environmentally friendly

[0067] The electric fence operates with low voltage pulses, which is harmless to fish and the environment and meets environmental protection requirements.

[0068] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0070] Figure 1 This is a layout diagram of the application scenario of the present invention;

[0071] Figure 2 This is a module diagram of the present invention.

[0072] Figure numerals: high-voltage pulse equipment 1; main control module 2; shore charging device 3; shore infrared camera 4; LED light 5; buoy 6; water robot module 7; thruster 8; underwater robot module 9; underwater front infrared camera 10; underwater rear infrared camera 11; connecting cable 12; depth sensor 13; electric grid 14; wire rope 15; buffer zone 16; first fish farming area 17; second fish farming area 18; fish catching box 19. DETAILED DESCRIPTION

[0073] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0074] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0075] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0076] 1. Device Principle

[0077] See also Figure 1This device needs to collect images of the underwater environment and fish in the fish farming area to form a data set, and trains it based on the PP_LiteSeg lightweight semantic segmentation network in the Paddle framework, and deploys the trained model to the underwater front infrared camera 10 and the underwater rear infrared camera 11 of the underwater robot module 9, so that the device can accurately capture the fish and their positions that have mistakenly entered the buffer zone 16; the main control module 2 plans the path to reach the target and drive the target to the fishing box 19 according to the relative position of the robot and the fish, and transmits instructions to the water robot module 7 and the underwater robot module 9 through the WiFi module.

[0078] (1) Fish-driving model and algorithm

[0079] ①System startup

[0080] The main control module turns on the power, the pulse fish-chasing module is automatically activated, and the above-water and underwater robots start up respectively and enter the monitoring state.

[0081] ②Fish detection

[0082] The underwater robot monitors the buffer zone through infrared cameras, 3D sensors and depth sensors. When a fish strays into the buffer zone, the robot immediately locks on the fish's position.

[0083] ③ Position transmission

[0084] The robot transmits the real-time location information of the fish to the water robot through a cable. The water robot then transmits the data back to the main control module through the water robot WiFi module.

[0085] At the same time, the infrared camera on the shore captures the LED light of the water robot, and the main control module determines the two-dimensional plane coordinates of the water robot, and combines the information from the underwater sensor to determine the three-dimensional position of the underwater robot and the position of the fish.

[0086] ④Path planning

[0087] The main control module calculates the attraction (to guide the robot towards the fish) and repulsion (to avoid collisions with obstacles and other robots) based on the relative positions of the fish and the robot.

[0088] By synthesizing the attractive and repulsive forces, a total guiding force is obtained to guide the robot's movement.

[0089] ⑤Robot movement

[0090] According to the resultant force calculated by the main control module, the robot automatically adjusts its movement direction.

[0091] The water robot moves in a plane through its thrusters, while the underwater robot adjusts its depth using a lifting device.

[0092] The entire process ensures that the robot quickly approaches the target fish while avoiding obstacles to ensure the optimal path.

[0093] ⑥Fish chasing operation

[0094] After the robot approaches the target fish, multiple robots work together to drive the fish to the vicinity of the fish box through underwater speakers.

[0095] The main control module continuously monitors the position of the fish and dynamically adjusts the paths of each robot to ensure that the fish are smoothly guided into the fishing net.

[0096] ⑦Fish classification and stocking

[0097] When the fish are driven to the catch box, the staff will classify them according to their type and put them back into Class A or Class B fish farming areas to avoid mixed farming.

[0098] For fish that have strayed into the buffer zone and have been monitored, we need to consider the robot reaching its vicinity and driving the fish into the fishing net through the speaker in the water. In this process, we need to consider the smoothness of the robot's motion path, obstacle avoidance (preventing collisions between robots and between robots and boundaries), and multi-robot collaborative path planning. Next, we will establish a path planning model for robots to collaboratively drive fish based on the potential field method, and adjust the robot's motion by calculating the vector of the resultant force.

[0099] Optimization goals

[0100] In this optimization model, the main goal is to achieve the underwater robot to drive the fish through the potential field method, while introducing path smoothing, obstacle avoidance and multi-robot collaboration technology, so that the robot can:

[0101] Move smoothly to avoid undue disturbance or shock to the fish;

[0102] Dynamically avoid obstacles in water (e.g., buffer zones, robots);

[0103] Overall it is efficient in driving fish to the preset fishing net area.

[0104] Attractive potential field optimization

[0105] The attraction potential field acts to guide the fish and the robot to the target fishing net area. In order to introduce path smoothness, an exponential form of attraction potential field based on the target distance is used, so that the attraction gradually decreases when approaching the target, rather than decreasing linearly, making the path smoother.

[0106] Attractive potential energy function:

[0107]

[0108] Where n is the number of robots. α is the decay rate of the control attraction. f is the position of the fish (x f ,y f , z f ), p n,i is the target position driven by the i-th robot (x n,i ,y n,i , z n,i ). att,i is the strength coefficient of the attractive potential field of the ith robot.

[0109] Attraction (expressed as a vector):

[0110]

[0111] That is: multiple robots jointly guide the attraction of fish, and the total attraction is the combined force of the attractions of each robot.

[0112] Repulsive potential field optimization

[0113] In addition to the attraction of fish to the fishing area, when multiple robots work together, they also need to consider obstacle avoidance between robots and obstacles in the water. We add a repulsive potential field to each robot to avoid collisions between robots and obstacles and boundaries in the water, and between robots.

[0114] Robot and fish

[0115] When the robot approaches the fish, the repulsive force increases sharply, causing the robot to move away from the fish. When the robot exceeds the safe distance, the repulsive force is 0 and no longer affects the robot.

[0116] The repulsive potential energy function between the robot and the fish:

[0117]

[0118] Among them, p r is the robot's position (x r ,y r , z r ). rep is the coefficient of the repulsive potential field. d0 is the threshold distance of the repulsive range.

[0119] The repulsive force between the robot and the fish (expressed as a vector):

[0120]

[0121] Robot and obstacles

[0122] When the robot approaches an obstacle, the repulsive force increases sharply, causing the robot to move away from the obstacle. When the robot exceeds the safe distance, the repulsive force is 0 and no longer affects the robot.

[0123] The repulsive potential energy function between the robot and the obstacle:

[0124]

[0125] Among them, p0 is the position of the obstacle (x0, y0, z0)

[0126] The repulsive force between the robot and the obstacle (expressed as a vector):

[0127]

[0128] Robots and robots

[0129] A certain distance must be maintained between robots.

[0130] The repulsive potential energy function between robots:

[0131]

[0132] in, and are the positions of the jth and ith robots respectively. d0 is the minimum safe distance between robots, ensuring that the robots maintain a reasonable distance to avoid collision.

[0133] The repulsive force between robots (expressed as a vector):

[0134]

[0135] Calculation of combined force

[0136] In summary, the total force of the fish and the robot will be determined by the attractive and repulsive forces.

[0137] The net force on the fish (expressed as a vector):

[0138] F t,f (o f )=F att (p f )+F rep (o f )

[0139] The robot's net force (expressed as a vector):

[0140]

[0141] (2) Path planning model for multi-robot collaborative fish herding based on potential field method

[0142] Decision variables

[0143] In the path planning model of multi-robot collaborative fish herding, the factors that affect path planning are the initial positions of the robots and fish, and the movement speed of the robots. The buffer boundary is fixed, and it is assumed that there are no obstacles in the buffer (except the boundary and the robot). Therefore, the decision variable of this model is the initial position of the robot. The robot's initial position p f 、Robot's movement speed v ri (t).

[0144] Objective Function

[0145] Minimize the combined force of the robots to maximize the efficiency of herding fish.

[0146]

[0147] Constraints

[0148] 1. Position constraints

[0149] The initial and target position constraints of the fish: The current position of the fish p f It should gradually approach the fishing net position p from its initial position n .

[0150] ∥p f -p n ∥≤τ

[0151] Here, τ is the allowable error range to ensure that the fish are accurately driven into the fishing net.

[0152] Robot position constraints: The position of each robot They must operate within the sub-area they are responsible for and cannot exceed the regional boundaries.

[0153]

[0154] Among them, R i is the sub-area that the i-th robot is responsible for.

[0155] 2. Collision constraints

[0156] Minimum safe distance between robots:

[0157]

[0158] Minimum distance between fish and obstacles: When fish are moving, they must avoid obstacles in the water and ensure that the distance between them and the obstacles is not less than the safety threshold.

[0159] ∥p0-p f ∥≤d0

[0160] 3. Potential field constraints

[0161] Attractive potential field constraint: The attraction to the fish is determined by the positions of multiple robots. The fish position p f The attractive potential field should be gradually reduced to ensure that the fish are driven to the target area.

[0162] U att (p f )→0

[0163] Repulsive potential field constraint: The repulsive potential field between robots and between fish and obstacles needs to be gradually increased to ensure that they avoid collision.

[0164]

[0165] U rep,r (p f )→∞

[0166] 3. Path smoothing optimization

[0167] In order to achieve a smooth path, we can introduce Bezier curve path planning when each robot moves to ensure that the movement paths of the fish and the robot are more natural and smooth.

[0168] Bezier curve path planning

[0169] P(t)=(1-t) 2 P0+2(1-t)tP1+t 2 P2,0≤t≤1

[0170] Among them, P0 is the starting position of the robot, P1 is the control point of the path (dynamically adjusted by the potential field method), and P2 is the target position (the position of the fish or the fishing net).

[0171] but:

[0172]

[0173] Among them, K max is the maximum curvature of the robot path.

[0174] 4. Dynamic constraints

[0175] The speed and acceleration limits of the robot: the robot's movement speed v ri (t) and acceleration a ri (t) It needs to be limited within the permitted range to prevent moving too fast or too slow which will affect the effect of driving fish.

[0176] v ri (t)≤v max

[0177] a ri (t)≤a max

[0178] In summary, we have established a path planning model for multi-robot collaborative fish-herding based on the potential field method:

[0179]

[0180]

[0181] Next, we can dynamically adjust the robot's movement by calculating the magnitude and direction of the resultant force in real time based on the potential field method.

[0182] (3) Target Detection Algorithm

[0183] PP-LiteSeg is a lightweight semantic segmentation network based on the PaddlePaddle framework, which is mainly used for efficient object detection and semantic segmentation tasks, especially suitable for embedded devices or resource-constrained environments, such as fish detection in underwater robot systems. The following is the detailed principle of PP-LiteSeg and its specific application in the invention patent.

[0184] 1. Overview of PP-LiteSeg network structure

[0185] The PP-LiteSeg network is a lightweight model optimized for real-time semantic segmentation tasks. It combines some streamlined designs in the classic segmentation network and incorporates some new lightweight modules. It is mainly based on the encoder-decoder network architecture design and is divided into two core parts:

[0186] Encoder part: mainly used for feature extraction, extracting multi-scale feature information of fish and underwater environment.

[0187] Decoder part: responsible for gradually upsampling the features extracted by the encoder and restoring them to the same size as the input image to output the category information of each pixel (different categories such as fish, buffer, background, etc. in the patent of this invention).

[0188] PP-LiteSeg optimizes network performance and model size in the following ways:

[0189] Depthwise Separable Convolutions: This is an extremely efficient convolution method that decomposes the standard convolution into two steps: depthwise convolution and point-by-point convolution, which greatly reduces the amount of calculation and model parameters, thereby ensuring that the calculation speed is accelerated without significantly losing accuracy.

[0190] Lightweight encoder: Use optimized lightweight encoders such as MobileNetV3 or ShuffleNet, which are designed to significantly reduce the computational complexity of the network while maintaining high accuracy, making them suitable for deployment in embedded devices.

[0191] Multi-scale Context Aggregation Module: PP-LiteSeg introduces a multi-scale context information aggregation module in the decoder, which helps the model better capture targets of different scales, such as different postures, distances, and movement speeds of fish.

[0192] Lightweight decoder: Compared with other segmentation networks, the decoder of PP-LiteSeg has been further simplified and adopts a step-by-step upsampling method to gradually restore high-resolution features, reducing the additional computational burden.

[0193] 2. Working principle of PP-LiteSeg

[0194] The workflow of PP-LiteSeg can be summarized as the following steps:

[0195] ① Image input: underwater images collected by the underwater robot’s infrared camera are used as input. These images may contain different types of fish and the surrounding underwater environment.

[0196] ② Feature extraction (encoder part): The input image is first subjected to multi-level feature extraction through the encoder module. The encoder extracts low-level and high-level features of the image through multiple convolutional layers, including important features such as edge information, texture features, and outlines of fish.

[0197] ③ Multi-scale feature aggregation: In the decoder part, PP-LiteSeg can better distinguish different categories such as fish, background, buffer zone, etc. by introducing a multi-scale context aggregation module and combining feature information at different levels. In this device, this module can help the system identify fish that have mistakenly entered the buffer zone and their location.

[0198] ④ Upsampling and fine segmentation (decoder part): The decoder gradually upsamples the feature map to restore it to the same resolution as the input image, and gradually restores the details to output a segmented image of the same size as the original image, in which each pixel is labeled as a certain category. In this application, the output image will clearly label the fish and the surrounding environment, which facilitates the main control module to make decisions and path planning.

[0199] ⑤Semantic segmentation results: The output of the PP-LiteSeg model is a semantic segmentation map, in which pixels of different categories are annotated with different category labels. In actual application scenarios, the segmentation results can accurately distinguish fish from other objects or backgrounds, and identify fish that have mistakenly entered the buffer zone.

[0200] 3. Application of PP-LiteSeg in the present invention

[0201] In this patent, PP-LiteSeg is used to accurately detect and segment fish in underwater environments. The workflow of the entire system is as follows:

[0202] Data collection and preprocessing: A large amount of image data about underwater environment and fish is collected through the infrared camera of the underwater robot. These image data will be used as the dataset for training the PP-LiteSeg model. Image data preprocessing includes operations such as image enhancement, scaling, and normalization.

[0203] Model training: The collected image dataset is input into the Paddle framework and trained based on the PP-LiteSeg lightweight semantic segmentation network. By continuously optimizing the model parameters, the network gradually learns to identify fish, buffer boundaries, and other important background information in the underwater environment. Due to the lightweight design of PP-LiteSeg itself, the model training efficiency is high during the training process and has good deployment performance on embedded devices.

[0204] Model deployment: The trained PP-LiteSeg model is deployed to the infrared camera in the underwater robot module to achieve real-time fish detection. The robot can use this model to analyze the underwater images captured by the camera in real time, accurately detect whether the fish has mistakenly entered the buffer zone, and return the location of the fish.

[0205] Path planning: The main control module receives the semantic segmentation results sent back by the underwater robot, and plans the optimal path for the robot to drive the fish back to the fish farming area or guide the fish to the fishing net according to the position and relative path of the fish. The main control module communicates with the robot via WIFI and sends movement instructions to ensure that the fish can be effectively caught.

[0206] 4. Advantages and applicability of PP-LiteSeg

[0207] Lightweight design: PP-LiteSeg’s lightweight design makes it ideal for deployment in resource-constrained environments, such as underwater robots. Embedded devices such as Jetson Nano can efficiently run the model, achieving real-time object detection and semantic segmentation.

[0208] Efficiency and accuracy: PP-LiteSeg achieves accurate target detection at different scales through a multi-scale feature aggregation module, which helps to improve the segmentation accuracy of fish in the water and the surrounding environment. In the application of the present invention, this ensures that the system can promptly and accurately identify fish that have strayed into the buffer zone.

[0209] Low power consumption and high performance: PP-LiteSeg can run on devices with lower power consumption while maintaining high segmentation accuracy and real-time performance, which is very suitable for application scenarios such as underwater robots that have strict requirements on power consumption and computing power.

[0210] 5. Model Training

[0211] In order to better demonstrate the application of PP-LiteSeg in the fish detection task, a specific neural network training example is given below, including the dataset structure, specific settings during training, and the selection and explanation of evaluation indicators.

[0212] (1) Training data example

[0213] Assume that some fish image datasets are collected underwater in a fish farming area, mainly including RGB images and corresponding annotated images (semantic segmentation labels). The structure of the dataset is as follows:

[0214]

[0215] Each training image contains underwater fish and the surrounding environment, and the label image is a semantic segmentation annotation, where each pixel value in the label image represents a category in the image. For example:

[0216] 0 means background, 1 means fish, 2 means other underwater objects or areas (such as stones, plants, etc. Note: this can be added according to different needs. In this device, the focus is on the detection of fish, so this item is not described in detail here)

[0217] (2) Training process

[0218] Training the PP-LiteSeg model mainly includes the following steps:

[0219] Training environment:

[0220] Use the PaddlePaddle deep learning framework and use GPU devices for accelerated training (such as NVIDIA Jetson devices).

[0221] Hyperparameter settings:

[0222] Batch Size: 16

[0223] Learning Rate: 0.001 (use the learning rate scheduler to gradually reduce the learning rate)

[0224] Epochs: 100

[0225] Optimizer: Adam optimizer or SGD optimizer

[0226] Data Augmentation:

[0227] In order to enhance the generalization ability of the model, the training data is usually enhanced. Common data enhancement methods include random cropping, rotation, scaling, color jittering, etc. of the image.

[0228] Loss function:

[0229] The cross entropy loss function is used as the main loss function for pixel-level classification tasks.

[0230] (3) Evaluation indicators

[0231] After training, a series of indicators are used to evaluate the performance of the model to ensure the accuracy and generalization ability of the model in detecting fish and their locations. The present invention uses the evaluation indicator Intersection over Union (IoU) (Note: because we focus on the detection of fish), which is described as follows:

[0232] The intersection-over-union ratio is used to measure the overlap between the predicted result and the true label. For each category (such as fish), the ratio of the intersection and union of the predicted area and the true area is calculated. The formula is as follows:

[0233]

[0234] The higher the IoU value, the better the model's segmentation effect on the category. Usually, we pay attention to the IoU value of the fish category to ensure that the model can accurately detect the location of the fish.

[0235] (4) Training result example

[0236] After 100 epochs of training, the evaluation indicators of the model are as follows:

[0237] IoU for fish detection: 83.7%

[0238] Global Accuracy (aAcc): 92.3%

[0239] Average Class Accuracy (mAcc): 88.5%

[0240] Mean Intersection over Union (mIoU): 80.4%

[0241] (The following three are other types of evaluation indicators)

[0242] This shows that the model has high segmentation accuracy in the task of fish detection and can effectively identify and segment fish in water. In addition, the model has strong generalization ability and can maintain high accuracy in different underwater environments.

[0243] The trained PP-LiteSeg model will be deployed in the infrared camera of the underwater robot to detect the position of fish and their relative movement trajectory in real time, and perform path planning through the main control module to drive the fish into the fishing net or guide them out of the buffer zone.

[0244] 2. Device Structure

[0245] The device is composed of four core modules: a pulse fish-driving module, a main control module 2, an above-water robot module 7 and an underwater robot module 9. The main control module 2 is also responsible for controlling the operation of the pulse fish-driving module. Figure 2 shown.

[0246] The pulse fish-driving module is composed of a high-voltage pulse device 1 and an electric grid 14. The high-voltage pulse device 1 is connected to the electric grid 14, and the electric grid 14 is suspended by a steel wire rope 15 to divide the water area into at least two fish-raising areas and at least one buffer zone. The high-voltage pulse device 1 adjusts the pulse voltage of the electric grid 14 according to the reaction intensity of different fish to the voltage.

[0247] The main control module 2 integrates a display screen, a WiFi module, a shore charging device 3, a shore infrared camera 4 and a power module. The main control module 2 is connected to the pulse fish-driving module through the power module, and is responsible for supplying the working energy of the electric grid 14. The main control module 2 is responsible for commanding and coordinating the various subsystems in the entire breeding device, monitoring the dynamics of fish and robots, classifying fish and planning paths, and communicating with the display device on the shore to display the location of fish and robots in real time. When the battery level of the robot is lower than a certain threshold, the main control module 2 sends a command to the shore charging device 3 for charging.

[0248] The water robot module 7 is composed of a WiFi module, a charging module, a lithium battery module, an LED light 5, a buoy 6 and four propellers 8 on the front, back, left and right sides. The four propellers 8 on the front, back, left and right sides control the movement of the robot as a whole in the two-dimensional plane of the fish farming device, and realize the communication between the robot and the main control module 2 through the WiFi module. The charging and lithium battery modules are responsible for the power supply to the water robot module 7 and the underwater robot module 9. The LED light 5 displays the position of the water robot module 7 in the two-dimensional plane in real time. The buoy 6 provides the ability to float on the water and maintain the stability of the water robot module 7.

[0249] The underwater robot module 9 is composed of a three-dimensional sensor, a depth sensor 13, a lifting device, a robot body, an underwater front infrared camera 10 and an underwater rear infrared camera 11. The above-water robot module 7 and the underwater robot module 9 are connected by a connecting cable 12. The three-dimensional sensor and the depth sensor 13 are responsible for real-time data collection of the underwater robot module 9, and the WiFi module of the above-water robot module 7 transmits the data to the main control module 2 through the connecting cable 12. The lifting device is responsible for controlling the movement of the underwater robot module 9 in a direction perpendicular to the two-dimensional plane of the fish farming device and the position adjustment of the robot. The underwater front infrared camera 10 and the underwater rear infrared camera 11 are responsible for monitoring whether the fish enter the buffer zone 16 by mistake and its position and other related information.

[0250] The parameters of each component are as follows:

[0251] High-voltage pulse equipment: The input voltage is 220V AC, and the output high-voltage pulse voltage is 300V to 500V, which is used for electric fences to prevent fish from swimming across areas. The pulse frequency is adjustable to adapt to the physiological reactions of different fish. The pulse equipment must support continuous working ability and stability. The applicable equipment is Tester voltage pulse generator, model Test-1200, with a pulse frequency range of 1-20Hz and an output power of 300W.

[0252] Main control module: Adopts Jetson Nano embedded system, integrated with WIFI module (working frequency 2.4GHz, maximum transmission rate 150Mbps), touch screen (7 inches, resolution 1024x600), power module (12V / 3A), shore infrared camera (resolution 1080P, night vision distance 50m) and other equipment. The system communicates with the above-water and underwater robots through WIFI, performs path planning, data processing and command sending. The WIFI module can use ESP8266 or LiChuang SCM-WF01, which has low power consumption and high stability.

[0253] Shore charging device: wireless charging equipment, operating frequency at 50Hz, maximum output power 100W, can quickly charge the water robot to resume work, the charging device adopts IP67 waterproof standard to ensure safe and reliable operation in humid environments. Shenzhen Renaway's RWC-100 wireless charger can be used, with a charging efficiency of up to 85%.

[0254] Shore infrared camera: equipped with infrared night vision function, with a resolution of 1080P and a night vision distance of up to 50 meters, used to monitor the status of the water surface and LED lights in real time and assist the main control module in positioning. Dahua DH-HAC-HFW1239 camera can be used, with IP67 waterproof function, suitable for outdoor environment.

[0255] LED light: installed on the water robot, with a power of 1W, a brightness of up to 200 lumens, and a red color, which is convenient for identifying the position of the water robot under the monitoring of the shore infrared camera at night. You can use Cree Xlamp XP-E2 red LED, which has high luminous efficiency and good stability.

[0256] Float: Made of high-density polyethylene (HDPE), each buoy can bear 50 kg, is corrosion-resistant and wear-resistant, and ensures the stability of the robot and other surface equipment in the water. The size of the buoy is 500mm×500mm×400mm.

[0257] Water robot module: integrated WIFI module (ESP8266), power module (12V lithium battery, capacity 20000mAh), and four electric thrusters.

[0258] Propellers: The four electric propellers of the water robot are installed in different directions. Each propeller has a power of 50W and a maximum thrust of 5 kg, which is suitable for coping with different water flow environments and supports flexible control of the water robot. The recommended device is the BlueRobotics T200 propeller, with an operating voltage of 12V to 16V.

[0259] Underwater robot module: It integrates a 3D sensor (working depth range 0-100 meters, accuracy ±0.1 meters), a lifting device (power 30W, supports depth adjustment from 0 to 50 meters), and two front and rear infrared cameras with a resolution of 720P and night vision function to monitor fish dynamics. The sensor model can be Blue Robotics Bar30, and the lifting device can be Hercules T50.

[0260] Underwater front and rear infrared cameras: used to monitor the fish activities and environmental images in front, with a resolution of 720P, a night vision distance of 10 meters, and support working in low-light environments. The device can choose Reolink RLC-410 infrared camera, with an operating voltage of 12V.

[0261] Connecting cable: used for data and power transmission between above-water and underwater robots. The cable is made of water-resistant and tear-resistant materials, supports simultaneous transmission of power and signals, and is a Jellico IP67 waterproof cable with a length of up to 10 meters and a voltage support of 12V to 24V.

[0262] Depth sensor: Installed on the underwater robot, supports a water depth range of 0 to 100 meters, with an accuracy of ±0.1 meters, real-time monitoring of water depth, and adjusting the depth of the robot in conjunction with the lifting device. Blue Robotics Bar30 sensor can be used.

[0263] Electric fence: Made of corrosion-resistant steel, it is arranged between the first and second fish farming areas. The electric fence is 10 meters long, each with a diameter of 20mm. The electric fence can withstand high-voltage pulses of 300V to 500V to ensure that fish are effectively prevented from crossing areas.

[0264] Steel wire rope: used to fix electric grids and buoys, made of 304 stainless steel, with a diameter of 5 mm. The length is customized according to the actual water environment. It has a tensile strength of 1,000 kg and strong corrosion resistance.

[0265] Buffer zone: It is set between the first fish breeding area and the second fish breeding area, with a water area of ​​about 10 square meters, to prevent fish from crossing into other areas and ensure that fish that stray into other areas can be driven back in time. The isolation area is separated by a pulse electric fence.

[0266] First fish farming area: One of the water area divisions, covering an area of ​​about 1,000 square meters, dedicated to raising specific types of fish.

[0267] The second fish farming area: with the same area as the first fish farming area, is used to breed different fish species. Electric fences ensure the isolation of fish in the two areas.

[0268] Fish box: The size is 2m x 1m x 1m, made of stainless steel, anti-corrosion, specially used for temporary storage of caught fish. The bottom of the fish box is equipped with drainage holes to ensure that the water level in the tank remains stable.

[0269] 3. Device workflow:

[0270] The power of the main control module 2 is turned on, and the pulse fish-driving module is also turned on. The underwater robot module 9 uses the underwater front infrared camera 10 and the underwater rear infrared camera 11 in the area it is responsible for to monitor whether there are fish that have mistakenly entered the buffer zone 16. If a fish is detected to have mistakenly entered the buffer zone 16, the underwater robot module 9 determines its position in the water through the three-dimensional sensor and the depth sensor 13, and at the same time, the underwater front infrared camera 10 and the underwater rear infrared camera 11 are combined with the PP_LiteSeg semantic segmentation model to identify the fish in real time and lock its position, and transmit the relevant data information to the water robot module 7 through the connecting cable 12, and then the WiFi module transmits the information to the main control module 2. The shore infrared camera 4 captures the LED light 5 of the water robot module 7, and the main control module 2 determines the precise position of the underwater robot module 9 and the fish based on its two-dimensional plane coordinates and the transmitted relevant data information. According to the relative position of the fish and the robot, the path planning algorithm is called to control the propulsion system 8 of the water robot module 7 and adjust the lifting device of the underwater robot module 9, so as to approach the target fish. The fish are driven to the preset fishing box 19 by collaboratively controlling the water robot module 7 and the underwater robot module 9. The entire fish-driving process is based on the instructions issued by the main control module 2 to ensure that the fish are guided to the designated area. After the fish are driven to the fishing box 19, the staff will put them back to the first fish farming area 17 or the second fish farming area 18 according to the type of fish to avoid mixing of fish and ensure zoned farming. When the robot power is lower than a certain threshold, the main control module 2 sends a command to the robot to control it to charge at the shore charging device 3.

[0271] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A zoned fish farming device with electric fence and robot collaboration, characterized in that: include: A high voltage pulse device (1) is connected to an electric grid (14), the electric grid (14) is suspended by a steel wire rope (15), and is used to divide a water area into at least two fish farming areas and at least one buffer area; The water robot module (7) comprises: A WiFi module of the water robot is connected to the main control module (2) and is used to communicate with the main control module; A power module, providing power to the aquatic robot module (7); Four thrusters (8) at the front, back, left and right sides control the movement of the water robot module (7) in a two-dimensional plane; LED light (5), used to display the position of the water robot module (7); A buoy (6) provides the ability to float on water; The underwater robot module (9) comprises: A three-dimensional sensor for collecting data from the underwater robot module (9); A depth sensor (13), used for collecting underwater depth data; A lifting device, for controlling the movement and posture adjustment of the underwater robot module (9) in a direction perpendicular to the two-dimensional plane; An underwater front infrared camera (10) and an underwater rear infrared camera (11) are used to monitor whether fish have strayed into the buffer zone and their location information; The main control module (2) comprises: a display screen for displaying information about the location of fish and the robot; A main control WiFi module, connected to the aquatic robot module (7) and used for communicating with the aquatic robot module (7); A shore charging device (3), used for charging the robot; The shore infrared camera (4) is used to capture the LED light (5) of the water robot module (7) to determine its two-dimensional plane coordinates; The electric fence (14) is arranged between the fish farming area and the buffer zone; The high-voltage pulse device (1) adjusts the pulse voltage of the electric grid (14) according to the intensity of the reaction of different fish to the voltage, so as to prevent different fish from swimming across zones; The above-water robot module (7) and the underwater robot module (9) are connected via a connecting cable (12); The main control module (2) determines the relative position of the fish and the robot based on the two-dimensional plane coordinates of the above-water robot module (7) and the three-dimensional position data transmitted by the underwater robot module (9); The main control module (2) calls a preset path planning algorithm according to the relative positions of the fish and the robot, controls the propeller (8) of the water robot module (7) to adjust its moving direction, and at the same time instructs the underwater robot module (9) to adjust its depth through the lifting device, so as to guide the fish that have mistakenly entered the buffer zone to the preset fishing box (19); The main control module (2) is integrated with a shore infrared camera (4) for capturing the LED light (5) of the aquatic robot module (7) to determine its position in a two-dimensional plane.

2. The electric fence and robot coordinated zoned fish farming device according to claim 1 is characterized in that: The main control module (2) is integrated with a shore charging device (3) for charging the robot.

3. The electric fence and robot coordinated zoned fish farming device according to claim 1 is characterized in that: The underwater robot module (9) is configured with a PP_LiteSeg lightweight semantic segmentation model for real-time recognition and classification of fish.

Images