An application method and system for an underwater intelligent net cleaning robot

Abstract

The present invention relates to the technical field of robots, and particularly to an application method and system for an underwater intelligent net cleaning robot, including: S1. Using digital twin technology to construct a net cage model on a computer; S2. Importing the model of the robot on the computer and automatically adjusting the relative position between the robot and the net cage; S3. The robot intelligently calibrates the cleaning area; S4. Real-time updating the physical and motion states of the digital twin model; S5. According to the calibration level of the cleaning area and the shape of the attachments, simulating the cleaning process in the digital twin model through a path planning method, optimizing the cleaning path and conducting simulation tests, and converting the test results into actual cleaning instructions; S6. The underwater robot updates its own state, and at the same time feeds back the position, attitude, linear velocity, and angular velocity state data information to the digital twin model. The present invention solves the technical problem of "improving the automation and intelligence level of the net cleaning robot", thereby improving the cleaning efficiency and quality and reducing the labor cost.

Description

Technical Field

[0001] The present invention relates to the technical field of robots, and in particular to an application method and system for an underwater intelligent net cleaning robot. Background Art

[0002] At present, the cleaning of underwater cages mainly relies on manual work, which is not only inefficient but also poses a threat to the safety and health of operators. In addition, existing automated cleaning equipment often has problems such as large mass, high energy consumption, and incomplete cleaning. In view of these deficiencies, the present invention proposes a new application method and system for an underwater intelligent net cleaning robot. Summary of the Invention

[0003] Based on the above problems, the purpose of the present invention is to provide an application method and system for an underwater intelligent net cleaning robot, which solves the problems of low efficiency, high cost, and harsh working environment in the existing underwater cage cleaning operations. Through automation and intelligent technologies, the cleaning efficiency and quality are improved, and the labor cost is reduced at the same time.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] An application method for an underwater intelligent net cleaning robot includes the following steps:

[0006] S1. Use digital twin technology to construct a cage model on a computer as the virtual environment for the robot's cleaning work;

[0007] S2. Place the underwater intelligent net cleaning robot in the water, import the robot's model on the computer, and perform positioning through the sensors on the robot to automatically adjust the relative position between the robot and the cage;

[0008] S3. The robot performs an underwater inspection task, captures the stain image data of the object to be cleaned using an underwater camera, and uses a convolutional neural network to accurately identify and segment the stain area, and combines the geometric shape and size information of the object to intelligently calibrate the cleaning area;

[0009] S4. Collect on-site data through a SCADAS data acquisition device, including video stream, seawater flow rate, underwater pressure, and robot attitude, etc., prepare the collected data into a data set for use in training the digital twin model, train the digital twin model using integral reinforcement learning, and construct a value function and the control strategy and disturbance input at the initial moment, and use an IoT edge gateway to analyze and process to update the physical and motion states of the digital twin model in real time;

[0010] S5. According to the calibration level and shape of the attachments in the cleaning area, the cleaning process is simulated in the digital twin model through the path planning method, the cleaning path is optimized and simulation tests are performed, and the test results are converted into actual cleaning instructions;

[0011] S6. When the water flow and pressure change, the underwater robot takes action based on the received control strategy and disturbance input, updates its own status, and feeds back the position, attitude, linear velocity, and angular velocity status data information to the digital twin model. Further real-time comparison is performed based on the fed-back status data to correct the digital twin, and the thruster power output is adjusted by selecting the optimal system design solution to achieve adaptive posture adjustment of the robot, ensuring the continuity and efficiency of the cleaning process.

[0012] As a preferred solution of the underwater intelligent net cleaning robot application method of the present invention, during the cleaning process, the propeller power is dynamically adjusted through the net force sensor in combination with the type and density distribution of the net cage attachments and the sea condition data, to ensure that the cleaning disc maintains a constant contact force with the surface to be cleaned, thereby achieving efficient cleaning.

[0013] As a preferred solution of the application method of the underwater intelligent net cleaning robot of the present invention, the SCADAS data acquisition device collects field data, including video stream, seawater flow rate, underwater pressure and robot posture, etc., and prepares the collected data into a data set, which includes:

[0014] Geometric data: The geometric data of the net and underwater environment are obtained through laser scanning, sonar or other 3D imaging technologies, such as the size characteristics of the net, the size characteristics and structural characteristics of the underwater robot, the depth of the water, the volume of the water area, etc.

[0015] Dynamic data: Use force sensors and inertial measurement units (IMUs) to obtain dynamic data of the robot and water flow and other environments, such as hydrodynamics, system dynamics, buoyancy, propeller propulsion performance, structural force analysis, etc.

[0016] Operational data: Obtain the operational data of the robot when performing cleaning tasks, such as trajectory, speed, torque, robot input and output characteristics, etc.

[0017] As a preferred solution of the underwater intelligent net cleaning robot application method of the present invention, a convolutional neural network is used to accurately classify image data, and the best candidate area is screened through a non-maximum suppression algorithm to eliminate redundant information and ensure the accuracy of the cleaning process.

[0018] An underwater intelligent net cleaning robot system, using the above-mentioned underwater intelligent net cleaning robot application method, comprises:

[0019] an underwater camera, mounted on the side of the robot, for capturing stain image data;

[0020] Scrubbing disc, installed at the bottom of the robot for cleaning operations;

[0021] Illuminating lamp, used to provide auxiliary lighting for the underwater camera;

[0022] Attitude detection module, including gyroscope, accelerometer, compass, depth sensor, temperature sensor, pressure sensor, etc., used to detect the attitude of the robot;

[0023] Drive module, composed of thrusters driven by electronic speed controllers, used to control the movement of the robot;

[0024] Umbilical cable, used for communication between the underwater robot and the above-water console;

[0025] Above-water console, used to build a digital twin model, conduct simulation tests, and control the robot to move along the cleaning path according to the stain information collected by the camera to perform cleaning operations;

[0026] The beneficial effects of the present invention are:

[0027] The application method and system of the underwater intelligent net cleaning robot provided by the present invention are as follows. First, the digital twin technology constructs a net cage model on a computer as the virtual environment for the robot's cleaning work. An underwater robot is placed in the actual water area, and the model of the robot is also imported into the virtual environment to achieve digital twinning. Second, the underwater cleaning robot conducts inspections. The camera will collect images to obtain the image data of the characteristic parts and stain areas of the object to be cleaned, and perform image preprocessing. Then, a convolutional neural network is used to identify and segment the image data of the stain area of the object to be cleaned. At the same time, on-site data is collected through SCADAS data acquisition equipment, and the data collected in the physical environment is real-time mapped into the digital twin model, including physical, motion, environmental and other attributes. Combining the prior information of the shape and size of the object to be cleaned, the system can automatically plan the optimal cleaning path and conduct a simulation demonstration on the three-dimensional model to verify the feasibility of the path. The application method and system of the underwater intelligent net cleaning robot provided by the present invention, on the one hand, enable manual labor not to work in a harsh cleaning environment. On the other hand, the existing cleaning equipment is bulky, inflexible and energy-consuming. The present invention improves the cleaning efficiency and the degree of intelligence; the cleaning robot based on visual guidance can utilize image information to clean the key parts and stain areas targeted, and can more effectively remove stains; the present invention uses the scraping brush disc at the bottom to achieve contact cleaning; the present invention controls the magnitude of the cleaning force based on the force feedback method to ensure the contact cleaning pressure between the brush disc and the surface of the object to be cleaned, which can not only achieve a high-quality cleaning effect, but also avoid the over-cleaning of the already cleaned area by the brush disc, resulting in damage to the net surface and energy waste; according to the inclination angle of the net cage net, the thruster makes the robot fit the net to maintain or change to the best cleaning posture, and during the cleaning process, the posture of the underwater robot is continuously adjusted to make the brush disc have the largest cleaning coverage area with the surface of the net cage attachment, improving the cleaning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments of the present invention. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the content of the embodiments of the present invention and these drawings.

[0029] Figure 1 It is a schematic flow chart of the application method of the underwater intelligent net cleaning robot provided by the specific embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] To make the technical problems solved, the technical solutions adopted, and the technical effects achieved by the present invention clearer, the technical solutions of the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present invention.

[0031] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It 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 operated in a specific orientation, and thus cannot be construed as a limitation to the present invention. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance. Among them, the terms "first position" and "second position" are two different positions.

[0032] In the description of the present invention, it should be noted that unless otherwise clearly defined and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection or a detachable connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0033] As Figure 1 shown, this embodiment provides an application method for an underwater intelligent net cleaning robot, which is used to solve the problems of harsh underwater net cleaning environment, large workload, low efficiency, and low cleaning quality and intelligence degree, and includes the following steps:

[0034] S1. Use digital twin technology to build a net cage model on a computer as the virtual environment for the robot's cleaning work;

[0035] S2. Put the underwater intelligent net cleaning robot into the water, import the model of the robot on the computer, and perform positioning through the sensors on the robot to automatically adjust the relative position between the robot and the net cage;

[0036] S3. The robot performs an underwater inspection task, captures the stain image data of the object to be cleaned by using an underwater camera, and uses a convolutional neural network to accurately identify and segment the stain area, and combines the geometric shape and size information of the object to intelligently calibrate the cleaning area;

[0037] S4. Collect field data through SCADAS data acquisition equipment, including video stream, seawater flow rate, underwater pressure and robot posture, etc., prepare the collected data into a data set for digital twin model training, use integral reinforcement learning to train the digital twin model, and construct the value function as well as the control strategy and disturbance input at the initial moment. Use the IoT edge gateway for analysis and processing to update the physical and motion states of the digital twin model in real time;

[0038] S5. According to the calibration level and shape of the attachments in the cleaning area, the cleaning process is simulated in the digital twin model through the path planning method, the cleaning path is optimized and simulation tests are performed, and the test results are converted into actual cleaning instructions;

[0039] S6. When the water flow and pressure change, the underwater robot takes action based on the received control strategy and disturbance input, updates its own status, and feeds back the position, attitude, linear velocity, and angular velocity status data information to the digital twin model. Further real-time comparison is performed based on the fed-back status data to correct the digital twin, and the thruster power output is adjusted by selecting the optimal system design solution to achieve adaptive posture adjustment of the robot, ensuring the continuity and efficiency of the cleaning process.

[0040] This paper uses advanced digital twin technology, combined with the professional Simcenter software platform, to build an accurate three-dimensional digital twin model based on the detailed parameters and performance indicators of the underwater net cleaning robot. The model not only includes the physical structure of the robot, but also covers the characteristics of its operating environment - the aquaculture cage. Through rigid-flexible multi-body dynamics simulation, the model can accurately simulate the physical interaction between the robot and the complex underwater environment.

[0041] In order to ensure the real-time and accuracy of the model, the present invention uses SCADAS data acquisition equipment to collect key field data, including video streams, seawater flow rate, underwater pressure, and the robot's real-time posture, etc. These data are quickly analyzed and processed by the IoT edge gateway, and then integrated into the digital twin model in real time, dynamically updating the model's physical and motion properties to reflect actual working conditions.

[0042] Digital twin model establishment process:

[0043] Use CAD software or 3D modeling tools to build a geometric model of the net and underwater environment. The following methods can be used: Point cloud processing: Use the ICP (Iterative Closest Point) algorithm to align the point cloud data with the known model to generate an accurate environmental model.

[0044] Meshing: Convert the geometric model into the mesh model required for finite element analysis for subsequent simulation and analysis.

[0045] 1) Design of the digital twin model:

[0046] (1)

[0047] Wherein, represents the system design scheme, represents the design scheme of the th subsystem, and , are the characteristic parameters of the size characteristics, buoyancy, resultant external force, water flow velocity, water temperature, water area volume, liquid density, underwater robot position, attitude, linear velocity, and angular velocity of the underwater robot respectively;

[0048] 2) Expression of the net buoyancy of the underwater robot:

[0049] (2)

[0050] Wherein, B and G represent the buoyancy and gravity of the underwater robot respectively, ， ;

[0051] 3) 、 、 are respectively x axis, y axis, z axis water flow velocities;

[0052] 4) The expression of the resultant external force on the underwater robot is:

[0053] (3)

[0054] Wherein, and are respectively the repulsive force and the ocean current field force on the underwater robot. At the same time, through computational fluid dynamics, is roughly considered as , is the ocean current force;

[0055] 5) In the input-output characteristics of the underwater robot, the approximate relationship between the input x and the output speed y is:

[0056] (4)

[0057] Wherein, the output speed and the above cubic function relationship is obtained by measuring the operation data of the underwater robot, and Respectively representing the angular velocity of the underwater robot, the front and rear linear velocities, and the up and down linear velocities, all can be approximately expressed by a cubic function relationship between the input and output.

[0058] Establish the kinematic and dynamic models of the underwater robot, considering the influence of underwater hydrodynamics.

[0059] First, establish a coordinate system to describe the movement of the robot, and then derive and establish the kinematic model from the rotation matrix of the coordinate system. Then analyze the force and moment relationships between the coordinate systems of the underwater robot. Finally, select appropriate parameter variables to establish a more accurate dynamic model to more realistically simulate the actual movement of the robot underwater.

[0060] Under the geographical coordinate system of the underwater robot, the displacements of the three axes are described as longitudinal displacement x , lateral displacement y and vertical displacement z , and the attitudes of rotation around the three axes are described as , , , also called roll angle, pitch angle and heading angle. Under the body coordinate system of the underwater robot, the moving speeds of the three axes are described as longitudinal speed , lateral speed and vertical speed , and the angular velocities of rotation around the three axes are described as , , . Under the body coordinate system, the axial forces on the three axes of the robot are described as , , , and the torques around the three axes are described as , , .

[0061] The kinematic model of the robot is a description of the underwater robot in two coordinate systems. Under the geographical coordinate system of the underwater robot, the position is described as , the attitude angle is expressed as , and the position and attitude angle vector is:

[0062] (5)

[0063] The linear velocity of the robot under the body coordinate is expressed as , the angular velocity is expressed as , and the linear velocity and angular velocity vector is expressed as:

[0064] (6)

[0065] For the linear velocity of the robot , the transformation matrix from the vehicle coordinate system to the geographic coordinate system is:

[0066] (7)

[0067] Since is an orthogonal matrix, we have: (8)

[0068] For the angular velocity , the transformation matrix from the vehicle coordinate system to the geographic coordinate system is:

[0069] (9)

[0070] The kinematic model from the vehicle coordinate system of the robot to the geographic coordinate system can be obtained as:

[0071] (10)

[0072] The kinematic model from the geographic coordinate system to the vehicle coordinate system is:

[0073] (11)

[0074] As a rigid body, ignoring the influence of the buoyancy rope, the six-degree-of-freedom dynamic model of the underwater robot in the vehicle coordinate system can be described as: (12)

[0075] In the formula: M is the inertia matrix of the underwater robot, , is the rigid body inertia matrix of the robot, is the added mass matrix; is the Coriolis force and centripetal force matrix, , and are the Coriolis force and centripetal force matrices generated by the robot and the added mass respectively; is the hydrodynamic damping matrix; is the restoring force and moment matrix composed of gravity and buoyancy; is the thruster thrust.

[0076] In terms of data fusion and model calibration, the digital twin model is calibrated by fusing the sensor data in actual operation with the virtual model. Kalman filtering or "Extended Kalman Filter (EKF)" is used to fuse the data of different sensors and correct the model errors.

[0077] (13)

[0078] In the formula: represents the estimated value of the state at time k; the predicted estimated value of the state at time k; represents the Kalman gain at time k; the measurement value at time k; represents the measurement matrix.

[0079] In the face of the deformation or inclination of the net caused by the changes in water flow velocity and pressure, the present invention can realize the adaptive attitude adjustment of the robot by adjusting the power output of the thruster, ensuring the continuity and efficiency of the cleaning process. To improve the reliability of the system, the present invention also optimizes the fault detection and prevention mechanism to ensure that the robot can respond quickly in case of abnormal situations. Through the combination of the digital twin model and the HEEDS rapid optimization and iteration algorithm, the system can continuously analyze and compare the collected operation data, automatically adjust and optimize the model, and enhance the adaptability of the robot to complex environmental changes.

[0080] In terms of motion cooperative control, the present invention uses the virtual-real mapping technology to reflect the changes in the net attitude caused by environmental changes in the digital model in real time. By using the digital twin HEEDS algorithm and machine learning algorithms, the system can analyze and predict the optimal dynamic adjustment strategy to optimize the attitude adjustment of the robot. By precisely controlling the thruster, the robot can adapt to the changes in water flow and pressure, maintain close contact with the net, and ensure the cleaning effect. Use simulation platforms (such as Gazebo, ROS, etc.) to run synchronously with the digital twin model to simulate every step of the underwater robot performing the cleaning task. Make real-time adjustments according to the simulation results and sensor data to ensure the accuracy of the robot in actual operation, using PID control or model predictive control (MPC)

[0081] (14)

[0082] Wherein, is the error between the desired trajectory and the actual trajectory; the control input, usually representing any form of control signal acting on the system; is the proportional gain, used to adjust the reaction intensity of the controller to the current error; is the integral gain used to adjust the reaction of the controller to the error accumulation.

[0083] The underwater intelligent net cleaning robot system of the present invention can use high-definition image information to clean key parts and stain areas specifically by adopting a vision-guided cleaning robot. The design of the bottom scrubbing disk realizes contact cleaning, and the cleaning force control technology based on force feedback ensures an appropriate contact pressure between the scrubbing disk and the surface of the object to be cleaned, which not only guarantees the cleaning quality but also avoids over-cleaning and potential damage to the already cleaned area.

[0084] The thrusters of the underwater robot are not only used to adjust the robot's attitude but also responsible for power distribution to achieve the best cleaning effect. According to the inclination angle of the net cage net and the characteristics of the attachments, the system can dynamically adjust the power output of the thrusters, enabling the robot to maintain the best cleaning attitude during the cleaning process, maximizing the cleaning coverage area, and thus improving the overall cleaning efficiency.

[0085] The design of the cleaning module takes into account the intelligent power distribution of the scrubbing disk. Combining the type characteristics and density distribution of the attachments on the net cage and the real-time sea condition information, the system can study and implement an adaptive cleaning strategy. Through the data feedback of the net-attaching force sensor, the system can dynamically adjust the power output of the thrusters to ensure that the cleaning disk maintains a constant scrubbing force, achieving an efficient and thorough cleaning effect.

[0086] The contact force with the net during the cleaning process is obtained through a force sensor, and the cleaning intensity is adjusted to avoid damaging the net. The admittance control method is used to achieve:

[0087] (15)

[0088] Among them, is the external force; M is the virtual mass; B is the virtual damping; K is the virtual spring stiffness.

[0089] According to the cleaning effect, the robot's motion trajectory is adjusted to ensure the maximization of the cleaning coverage rate and the cleaning effect. The genetic algorithm or particle swarm optimization algorithm is used for global optimization:

[0090] (16)

[0091] Among them, refers to the maximum cleaning efficiency; then refers to the minimization of energy consumption.

[0092] The underwater intelligent net cleaning robot control method and system provided by the present invention, on the one hand, solve the problems existing in the existing underwater cage cleaning operations, such as low efficiency of manual operation, high cost, and harsh working environment. Through automation and intelligent technologies, the cleaning efficiency and quality are improved, and the labor cost is reduced at the same time. On the other hand, the existing cleaning equipment is bulky, inflexible, and has high energy consumption. The present invention improves the cleaning efficiency and degree of intelligence; by using a vision-guided cleaning robot, it can utilize image information to clean the key parts and stain areas targeted, and can more effectively remove stains; the present invention uses a scraping brush disk at the bottom to achieve contact cleaning; the present invention controls the magnitude of the cleaning force based on force feedback to ensure the contact cleaning pressure between the brush disk and the surface of the object to be cleaned, which can not only achieve a high-quality cleaning effect, but also avoid the over-cleaning of the already cleaned area by the brush disk, causing damage to the net surface and energy waste; according to the inclination angle of the cage net, the thrusters make the robot fit the net and maintain or change to the best cleaning posture, and during the cleaning process, continuously adjust the posture of the underwater robot so that its brush disk has the largest cleaning coverage area with the surface of the cage attachments, improving the cleaning efficiency.

[0093] Use vision to identify and locate the stains and key cleaning areas on the surface of the object to be cleaned, and then combine the object to be cleaned identified by vision. According to the prior information of its shape and size, select the corresponding cleaning path for cleaning, so as to achieve accurate and high-quality cleaning operations.

[0094] Preprocess the image of the cage attachments obtained by the camera, and use the retinex algorithm to enhance the image to obtain a feature map, and obtain the target sample candidate box. After passing through the region of interest pooling layer and the trained fully connected classification network, perform target recognition to obtain the attachment classification of the target image to be recognized.

[0095] Recognition process for areas with more attachments: First, use the camera to collect visual images, perform noise reduction on the images using median filtering and Gaussian filtering, and then use the single-scale retinex algorithm (SSR) to eliminate the brightness difference to obtain the preprocessed image; in order to improve the accuracy of target recognition and segmentation, use a method based on convolutional neural network for target recognition and segmentation.

[0096] Preprocess the image of the stain area obtained by the camera, and use a deep convolutional neural network to obtain a feature map, and obtain the target sample candidate box. After passing through the region of interest pooling layer and the trained fully connected classification network, perform target detection to obtain the position information of the stain area of the component object.

[0097] Perform median filtering on the target image. The median filtering method is a non-linear smoothing technique that sets the gray value of each pixel point to the median of the gray values of all pixel points within a certain neighborhood window of that point. Median filtering is a non-linear signal processing technique based on sorting statistics theory that can effectively suppress noise. The basic principle of median filtering is to replace the value of a point in a digital image or digital sequence with the median of the values of all points in a neighborhood of that point, making the surrounding pixel values closer to the true value and thus eliminating isolated noise points.

[0098] The specific method is to use a two-dimensional sliding template of a certain structure, sort the pixels within the template according to the pixel values to generate a two-dimensional data sequence that monotonically increases (or decreases). The output of two-dimensional median filtering is g(x,y) = med{f(x-k,y-l),(k,l∈W)}, where f(x,y) and g(x,y) are the original image and the processed image respectively. W is a two-dimensional template, usually a 3*3 or 5*5 area, and can also be of different shapes, such as linear, circular, cross-shaped, circular ring-shaped, etc. Median filtering has a good effect on filtering impulse noise. Especially when filtering noise, it can protect the edges of the signal and prevent them from being blurred. The median filtering method is very effective in eliminating salt-and-pepper noise.

[0099] Gaussian filtering is a linear smoothing filter that is suitable for eliminating Gaussian noise and is widely used in the denoising process of image processing. Generally speaking, Gaussian filtering is a process of weighted averaging of the entire image. The value of each pixel point is obtained by weighted averaging of itself and other pixel values in its neighborhood. The specific operation of Gaussian filtering is as follows: scan each pixel in the image with a template (or called convolution, mask), and replace the value of the pixel at the center of the template with the weighted average gray value of the pixels in the neighborhood determined by the template. Most of the noise in images belongs to Gaussian noise, so Gaussian filters are also widely used. Gaussian filtering is a linear smoothing filter suitable for eliminating Gaussian noise. The specific operation of Gaussian filtering is as follows: scan each pixel in the image with a user-specified template (or called convolution, mask), and replace the value of the pixel at the center of the template with the weighted average gray value of the pixels in the neighborhood determined by the template.

[0100] One-dimensional Gaussian distribution:

[0101]

[0102] Two-dimensional Gaussian distribution:

[0103]

[0104] The Single Scale Retinex (SSR) algorithm is an image enhancement technique. The Retinex theory is based on the working principle of the human visual system, especially the processing mechanism of the retina for illumination and reflectance. The main purpose of this algorithm is to improve the visual effect of images, especially under low-light conditions. The core idea of the Single Scale Retinex algorithm is to separate the brightness and reflectance of an image. In natural scenes, the pixel values of an image contain not only the reflectance information of objects but also illumination information. The Retinex algorithm attempts to recover the true reflectance of objects from the image, thereby enhancing the visual effect of the image. It is particularly suitable for enhancing images taken under low-light conditions and can significantly improve the brightness and contrast of images.

[0105] Calculate the logarithmic reflectance map:

[0106] R(x,y) = log(I(x,y))

[0107] Background subtraction:

[0108]

[0109] Reflectance recovery:

[0110]

[0111] Since the training of a convolutional neural network relies on high-quality images, and due to the complex and variable cleaning scene environment, the images directly obtained by the camera have poor quality. Therefore, image preprocessing is required before inputting them into the neural network for training to obtain better recognition results. Image preprocessing includes: using median filtering and Gaussian filtering to denoise the image, and then using the Single Scale Retinex algorithm to enhance the brightness and saturation of the image, thus obtaining the preprocessed image. The process of object recognition based on a convolutional neural network mainly includes: first generating a series of target candidate regions, then using the convolutional neural network to extract the convolutional features in the target image, and finally classifying the target image data. For the adjusted sample candidate boxes, non-maximum suppression is used to remove the sample candidate boxes with too large overlapping degrees, and the sample target candidate boxes are obtained.

[0112] Convolutional neural networks have characteristics such as local perception, shared weights, and multi-core convolution. The basic structure of a convolutional neural network consists of an input layer, hidden layers, a fully connected layer, and an objective function. Among them, the hidden layers include several convolutional layers, activation function layers, and pooling layers.

[0113] In computer vision, the input layer provides the original image data for the network. The convolutional layer in the hidden layer extracts the feature information in the image by performing convolution operations on the provided original image data. Among them, the weights of the elements in the convolutional kernel matrix are determined after the network training.

[0114] The activation function layer in the hidden layer takes the result after the convolution operation as the input. The purpose is to introduce non-linear factors to make the network representation ability stronger. Common activation functions include the sigmoid function, Tanh function, and Relu function. The pooling layer, also known as the downsampling layer, is used to reduce the dimension of the feature map, reduce the number of network calculation parameters, and at the same time enable the network to extract more important features. In the pooling layer, the commonly used pooling operation methods are max pooling and average pooling.

[0115] The fully connected layer fuses the local feature information extracted by the hidden layer and feeds the final result to the classifier. The objective function, also known as the loss function, is used to calculate the error between the predicted value of the network output and the provided true value. In regression tasks and classification tasks, the most commonly used objective functions are the quadratic loss function and the cross-entropy loss function respectively. Among them, Equation (a) represents the quadratic loss function, and Equation (b) represents the cross-entropy loss function.

[0116] (a)

[0117] Among them, W represents the weight vector of the network, b is the bias vector of the network, M represents the number of input sample data, x represents the input sample data, y is the true value of the sample data, L is the maximum number of network layers, and a L represents the predicted value of the network.

[0118] (b)

[0119] Among them, represents the predicted value of the network output, and y(i) represents the sample class label 0 or 1 corresponding to the i-th group.

[0120] First, the convolutional neural network layer performs convolutional feature extraction on the original image, then inputs the feature map into the RPN network for candidate box extraction, and finally classifies the candidate boxes and performs bounding box regression to achieve object detection.

[0121] The cleaning path design includes but is not limited to linear and concentric circle types to adapt to different environments and cleaning requirements and achieve more stable and efficient robot control.

[0122] In the virtual environment, use algorithm or Dijkstra algorithm to generate the best cleaning path for the robot:

[0123] Among them, is the actual cost from the starting point to node n, is the estimated cost from node n to the target point.

[0124] In actual operation, the "Dynamic Window Approach (DWA)" is used to dynamically adjust the path to avoid obstacles:

[0125] wherein refers to the velocity component of the robot along the x-axis; refers to the velocity component of the robot along the y-axis.

[0126] Note that the above is only the preferred embodiment of the present invention and the technical principles applied. Those skilled in the art will understand that the present invention is not limited to the specific embodiments here, and various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments only. Without departing from the concept of the present invention, it can also include more other equivalent embodiments, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. An application method of an underwater intelligent net cleaning robot, characterized in that, It includes the following steps: S1. Use digital twin technology to build a cage model on a computer as the virtual environment for the robot's cleaning work; S2. Put the underwater intelligent net cleaning robot into the water, import the robot's model on the computer, and position it through the sensors on the robot to automatically adjust the relative position between the robot and the cage; S3. The robot performs underwater inspection tasks, uses an underwater camera to capture the stain image data of the object to be cleaned, and uses a convolutional neural network to accurately identify and segment the stain area. Combining the geometric shape and size information of the object, it intelligently calibrates the cleaning area; S4. Collect on-site data through SCADAS data acquisition equipment, including video stream, seawater flow rate, underwater pressure, and robot attitude. Prepare the collected data into a dataset for digital twin model training. Use integral reinforcement learning to train the digital twin model, and construct a value function, control strategy, and disturbance input at the initial moment. Use the IoT edge gateway to analyze and process to update the physical and motion states of the digital twin model in real time; S5. According to the calibration level of the cleaning area and the shape of the attachments, simulate the cleaning process in the digital twin model through path planning methods, optimize the cleaning path and conduct simulation tests, and convert the test results into actual cleaning instructions; S6. When the water flow and pressure change, act based on the received control strategy and disturbance input. The underwater robot updates its own state, and at the same time feeds back the position, attitude, linear velocity, and angular velocity state data information to the digital twin model. Further, based on the feedback state data, make a real-time comparison to correct the digital twin. By selecting the optimal system design scheme, adjust the propeller power output to achieve the adaptive attitude adjustment of the robot and ensure the continuity and efficiency of the cleaning process.

2. The application method of the underwater intelligent net cleaning robot according to claim 1, characterized in that Use digital twin technology for parametric modeling, build three-dimensional digital models of the net cleaning robot and the aquaculture cage, accurately simulate the physical interaction between the robot and the cage, use an underwater high-definition camera to capture specific signs for auxiliary positioning, intelligently plan the cleaning path, and verify the feasibility of the path through simulation.

3. The application method of the underwater intelligent net cleaning robot according to claim 1, characterized in that, During the cleaning process, combine the types, density distribution of the cage attachments and sea condition data, and dynamically adjust the propeller power through the net adhesion force sensor to ensure that the cleaning disc maintains a constant contact force with the surface to be cleaned and achieve efficient cleaning.

4. The application method of the underwater intelligent net cleaning robot according to claim 1, characterized in that The said dataset includes: Geometric data: Obtain the size characteristics, structural characteristics of the net, the size characteristics, structural characteristics of the underwater robot, the water depth, and the water area volume through laser scanning, sonar, or other 3D imaging technologies; Dynamic data: Use force sensors and inertial measurement units to obtain data on hydrodynamics, system dynamics, buoyancy, propeller propulsion performance, and structural force analysis; Operation data: Obtain the operation data of the robot when performing the cleaning task, including trajectory, speed, torque, and robot input-output characteristics.

5. The application method of the underwater intelligent net cleaning robot according to claim 1, wherein Use a convolutional neural network to accurately classify the image data, screen the best candidate areas through the non-maximum suppression algorithm, exclude redundant information, and ensure the accuracy of the cleaning process.

6. An intelligent cleaning robot system, characterized in that, Adopt the intelligent net clothing cleaning robot application method described in any one of claims 1-5, including: An underwater camera, arranged on the side of the robot, for capturing stain image data; A scrubbing disc, arranged at the bottom of the robot, for cleaning operations; A lighting lamp, for providing auxiliary lighting for the underwater camera; An attitude detection module, including a gyroscope, an accelerometer, a compass, a depth sensor, a temperature sensor, and a pressure sensor, for detecting the attitude of the robot; A driving module, composed of thrusters driven by electronic speed controllers, for controlling the movement of the robot; An umbilical cable, for communication between the underwater robot and the above-water console; An above-water console, for constructing a digital twin model, conducting simulation tests, and controlling the robot to move along the cleaning path according to the stain information collected by the camera to perform cleaning operations.

Images