Method and system for controlling cleaning robot for hull surface under waterline

Abstract

The invention provides a control method and system for a cleaning robot on the surface of a ship under a waterline, and the method comprises the steps: collecting an image of the ship under the waterline, recognizing attachments on the ship based on the image of the ship, judging the positions of the attachments at all positions, and carrying out the path planning of the cleaning robot based on the positions of the attachments at all positions. And controlling the cleaning robot to arrive at the appointed attachment along the path to carry out cleaning treatment. According to the invention, the cleaning robot can be automatically controlled to clean attachments on the ship body, so that the cleanness of the ship body is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of ship hull cleaning robot control, and in particular to a control method and system of a ship hull surface cleaning robot below waterline. Background Art

[0002] The bottom of the ship is soaked in seawater for a long time. Over time, rust will appear on the surface of the hull, and some algae will gradually grow. As these algae continue to grow, they can slowly provide food for other marine animals, so they will also attract other marine organisms (such as barnacles). The longer the ship drifts in the sea, the more seafood will be absorbed by the bottom of the ship, and the surface structure of the bottom of the ship will gradually change. The impact is that the ship's sailing speed will slow down, and it will also cause an increase in fuel consumption. The first reason is that these seafood on the bottom of the ship increase resistance, which will also increase fuel consumption; the second reason is that these seafood on the bottom of the ship will also increase the weight of the ship, thereby increasing fuel consumption, and the bottom of the ship carries more marine organisms, which may also cause biological invasions to the port where it is docked.

[0003] In the prior art, cleaning is often done manually or by a cleaning robot. The manual cleaning method is inefficient, while the cleaning method of a cleaning robot is often manually controlled, which is time-consuming and labor-intensive, and cannot be automatically controlled or automatically cleaned. Therefore, it is very necessary to design a control method and system for a cleaning robot on the surface of a ship under the waterline. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, an object of the present invention is to provide a control method and system for a robot for cleaning pollution on a hull surface below the waterline.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A first aspect of the present invention provides a control method for a robot for cleaning pollution on a hull surface below a waterline, comprising:

[0007] Acquire images of the hull below the waterline;

[0008] Identify the attachments on the hull based on the image of the hull and determine the location of each attachment;

[0009] Path planning for the pollution-cleaning robot based on the locations of various attachments;

[0010] Control the cleaning robot to reach the designated attachment along the path and carry out cleaning.

[0011] Preferably, the attachments on the hull are identified based on the image of the hull, and the positions of the attachments are determined, specifically:

[0012] Build an attachment recognition model based on the improved EfficientNet network;

[0013] Obtain various images of ship hull attachments, perform data enhancement on them, and construct a ship hull attachment dataset;

[0014] The attachment recognition model is trained based on the hull attachment data set to obtain a trained attachment recognition model;

[0015] The collected images of the hull below the waterline are input into the trained attachment recognition model to obtain the locations of various attachments.

[0016] Preferably, the improvement process of the EfficientNet network is as follows: the MBConv module in the shallow layer of the EfficientNet-B0 model in the EfficientNet network is replaced with a Fused-MBConv module, that is, the 1×1 convolution operation and the 3×3 depth-separable convolution in the main branch of the original MBConv module are replaced with an ordinary 3×3 convolution operation, the last stage with a step size of 1 in the original EfficientNet network structure is removed, and the 5×5 convolution kernel in the original structure is replaced with a 3×3 convolution kernel.

[0017] Preferably, data enhancement is performed on the images of various types of hull attachments, specifically as follows:

[0018] The images of various types of hull attachments are subjected to set transformation, brightness adjustment, saturation adjustment, contrast adjustment, random noise addition, as well as mirroring, scaling and rotation operations to achieve data enhancement.

[0019] Preferably, the path planning of the cleaning robot is performed based on the locations of the attachments at various locations, specifically:

[0020] Get the location of each attachment;

[0021] Get the location of the pollution-cleaning robot;

[0022] Based on the position of the cleaning robot, the position of the attachment closest to the cleaning robot is obtained;

[0023] Path planning is performed based on the improved ant colony algorithm according to the position of the cleaning robot and the position of the attachment closest to the cleaning robot;

[0024] Control the cleaning robot to reach the designated attachment along the path and perform cleaning;

[0025] Repeat the above steps until all the attachments are removed.

[0026] Preferably, path planning is performed based on an improved ant colony algorithm, specifically:

[0027] Taking the position of the cleaning robot as the starting point and the position of the attachment closest to the cleaning robot as the end point, the starting point and the end point area are divided into different grid areas based on the grid method, and the coordinates of the starting point and the end point are determined;

[0028] Initialize the parameters of the ant colony algorithm, set the maximum number of iterations K, the ant colony size M, the pheromone importance factor a, the heuristic factor β, and the pheromone volatility coefficient ρ;

[0029] When searching for a path, the ants select the next node based on the state transition probability formula. When searching for a path, the ant colony determines the next moving direction by judging the pheromone concentration and heuristic information. The ants move along the path until they reach the end or complete this iteration.

[0030] Record the length of each path and the number of ants on each path, and store the information in the cell structure CELL. Subsequent pheromones are updated based on this information.

[0031] According to the recorded information, find the best path and the worst path in this iteration, and adjust the pheromone concentration on the path according to the pheromone update rule;

[0032] If the maximum number of iterations K is reached, the optimal path is output, otherwise, the taboo table TABUK is cleared and a new iteration is performed;

[0033] The obtained optimal path is optimized, and unnecessary turning points in the path are removed based on the triangle pruning method to obtain the final path.

[0034] The present invention also provides a control system for a robot for cleaning pollution on a hull surface below the waterline, comprising:

[0035] The cleaning robot body is used to clean the attached objects;

[0036] An image acquisition module is provided on the body of the pollution-cleaning robot or the hull and is used to acquire images of the hull below the waterline;

[0037] The attachment recognition module is used to identify the attachments on the hull based on the image of the hull and determine the location of each attachment;

[0038] The path planning module is used to plan the path of the pollution cleaning robot based on the locations of various attachments.

[0039] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] The present invention provides a method and system for controlling a cleaning robot on a hull surface below the waterline, wherein the method comprises collecting an image of the hull below the waterline, identifying attachments on the hull based on the image of the hull, determining the positions of the attachments at various locations, planning a path for the cleaning robot based on the positions of the attachments at various locations, and controlling the cleaning robot to reach the designated attachments along the path to perform cleaning. The present invention can realize automatic control of the cleaning robot to clean the hull attachments, thereby ensuring the cleanliness of the hull. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the process of identifying hull attachments;

[0044] Figure 3 This is a schematic diagram of the path planning process. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The purpose of the present invention is to provide a control method and system for a robot for cleaning the surface of a hull below the waterline, which can automatically clean the attachments on the hull and is easy to use.

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a control method for a robot for cleaning pollution on a hull surface below the waterline, comprising:

[0049] Step 100: Collecting an image of the hull below the waterline;

[0050] Step 200: Identify the attachments on the hull based on the image of the hull and determine the location of each attachment;

[0051] Step 300: planning a path for the cleaning robot based on the locations of the attachments;

[0052] Step 400: Control the cleaning robot to reach the designated attachment along the path and perform cleaning.

[0053] In step 100, an image of the hull below the waterline is collected, specifically:

[0054] An image acquisition module is provided on the hull or the pollution-cleaning robot to acquire images of the hull below the waterline, wherein the image acquisition module may adopt a waterproof camera;

[0055] like Figure 2 As shown, in step 200, the attachments on the hull are identified based on the image of the hull, and the positions of the attachments are determined, specifically:

[0056] Step 201: constructing an attachment recognition model based on the improved EfficientNet network;

[0057] Step 202: Acquire various types of ship attachment images, perform data enhancement on them, and construct a ship attachment dataset;

[0058] Step 203: training the attachment recognition model based on the hull attachment data set to obtain a trained attachment recognition model;

[0059] Step 204: input the collected image of the hull below the waterline into the trained attachment recognition model to obtain the location of each attachment.

[0060] EfficientNet is a new type of network that is improved and optimized based on the traditional convolutional neural network. It is one of the current classic and efficient classification networks. Traditional convolutional neural networks generally expand the network by adjusting the resolution of the input image, the network depth, and the number of convolution channels. The model that scales any one dimension alone will gradually tend to saturation. In the scaling process of the entire neural network, it is key to comprehensively balance all dimensions. Therefore, EfficientNet uses a model composite scaling method to comprehensively expand the three. The specific method is to specify a composite coefficient φ to simultaneously constrain image resolution, network width, and network depth. The specific expression is:

[0061]

[0062] Where d is the network depth, w is the network channel width, r is the image resolution, γ is the resolution multiplier factor found by the network, α is the depth multiplier factor found by the network, β is the width multiplier factor found by the network, φ is the mixing factor, and st is the constraint condition;

[0063] In step 201, the improvement process of the EfficientNet network is as follows: the MBConv module in the shallow layer of the EfficientNet-B0 model in the EfficientNet network is replaced with a Fused-MBConv module, that is, the 1×1 convolution operation and the 3×3 depth-separable convolution in the main branch of the original MBConv module are replaced with an ordinary 3×3 convolution operation, and the last stage with a step size of 1 in the original EfficientNet network structure is removed to reduce the number of network parameters, and the 5×5 convolution kernel in the original structure is replaced with a 3×3 convolution kernel, and the stacked layer structure is used to compensate for the influence of the small convolution kernel on the receptive field.

[0064] The present invention provides an embodiment to compare the attachment recognition model with various models, wherein the weights of the improved EfficientNet network are initialized and trained, the data set selects the hull attachment data set, and the ratio of the training data set to the test data set is 9:1. Since the results of the convolutional neural network are affected by multiple hyperparameters, the learning rate represents the step size of each convergence, which determines the speed of convergence, and the batch size represents the number of training samples selected each time, which determines the direction of convergence. For this reason, the present invention selects Epoch as 200 iterations, the initial learning rate is set to 0.01, the batch size is set to 8, the multiplier factor of the network width is set to 1.4, the multiplier factor of the network depth is set to 1.8, and the remaining parameters are set to the default values. The recognition results of the improved EfficinetNet network, GoogleNet, VGG16, MobileNet and the standard EfficientNet network on the constructed hull attachment image data set are shown in Table 1;

[0065] Table 1 Comparison of recognition accuracy and training time of different models

[0066] Model Recognition accuracy Training time Improved EfficinetNet network 95.6％ 68min Standard EfficinetNet network 90.1％ 85min GoogleNet 86.2％ 105min VGG16 87.3％ 115min MobileNet 91.6％ 92min

[0067] As can be seen from Table 1, the improved EfficientNet network model has improved the recognition accuracy of hull attachments compared with GoogleNet, VGG16, MobileNet and the standard EfficientNet model, which are increased by 9.4%, 8.3%, 4% and 5.5% respectively. Under the same number of iterations, the improved training time is also reduced compared with the standard EfficientNet network model. In general, the recognition performance of the improved EfficientNet network model in the hull attachment image dataset is improved compared with other models, which verifies its effectiveness.

[0068] In step 202, data enhancement is performed on the images of various types of hull attachments, specifically:

[0069] Perform set transformation, brightness adjustment, saturation adjustment, contrast adjustment, random noise addition, mirroring, scaling and rotation operations on various ship attachment images to achieve data enhancement;

[0070] The present invention provides an embodiment, which performs data enhancement on various images of ship hull attachments based on a DCGAN data enhancement method;

[0071] Introduce the DCGAN data enhancement method: DCGAN is a derivative of the generative adversarial neural network (GAN). On the basis of GAN, it introduces a convolutional neural network to replace the generator and discriminator structures. It also uses some techniques to improve the convergence speed of the network and alleviate the problem of gradient disappearance.

[0072] The objective function of the DCGAN adversarial network is consistent with that of the GAN, and its expression is:

[0073]

[0074] In the formula, x is the real data, P data (x) is the real data distribution, z is the noise input to the G network, P z (z) is the random noise distribution, G(z) is the data generated by the G network, D(x) is the probability that the D network judges whether the real data is real, D[G(z)] is the probability that the D network judges whether the data generated by G is real, and the value function min G max D V(D,G) is the minimax game between D and G, E x～Pdata(x) It is expected that x will be data (x), is the mathematical expectation of the discriminator to identify the real data, E z～Pz(z) It is expected that z will be z (z) is obtained as the mathematical expectation of the random noise distribution;

[0075] The training of DCGAN is an alternating process, which is essentially a maximum and minimum optimization problem;

[0076] 1) When training the discriminator D, the generation network G is fixed, the parameters of the judgment network D (discriminator D) are optimized, and the gradient ascent algorithm is used to increase the value of the objective function. The expression is:

[0077]

[0078] 2) When training the generator G, fix the discriminant network D, optimize the parameters of the generator network G, and use the gradient descent algorithm to reduce the value of the objective function. The expression is:

[0079]

[0080] The present invention can also use other models to replace the improved EfficientNet network model. For example, the improved GoogLeNet network is used. The GoogLeNet network has a deeper network depth and a wider network width in structure, which means that GoogLeNet has better image feature extraction and representation capabilities. Due to its special Inception module and output module in the GoogLeNet network structure, although the scale of the GoogLeNet network is large, the number of parameters of the network is the least. The number of parameters of the GoogLeNet network is 7M, which is about one twentieth of the number of parameters of the VGG network. A smaller network parameter number means that the network has lower computing performance requirements for the computer and a faster detection speed of the network.

[0081] Improve the network based on the GoogLeNet network, further optimize the network model, simplify the network scale, and reduce the number of network parameters, so as to meet the requirements of fast, accurate, and lightweight detection of the model;

[0082] The improved design for GoogLeNet still continues the improvement and simplification ideas in the two directions of network depth and network width. First, after the image preprocessing, the image size collected by the present invention is balanced and the size change is not large. The first three layers in GoogLeNet are used to extract three layers of convolution layers of different sizes, which can simplify and retain a layer of 3*3 convolution layers. Secondly, in order to avoid the loss of feature information in the post-extraction stage of the network due to insufficient features and too deep network layers, the improved network reduces the number of Inception modules in the original network, reducing the original 9 Inception modules to 6 Inception modules, and adding a pooling layer in every two Inception modules to adjust the image size. The Inception module still retains the original 4-channel two-level structure design. The number of Inception convolution kernels draws on the convolution kernel setting of the VGG network, which is increased by multiples from 64 to 512, which makes the entire network structure more regular and modular, which is convenient for the setting and modification of the network, and is conducive to the writing of program code, reducing the amount of code, and facilitating computer hardware acceleration. Finally, the network sets up a three-classification output layer and adds softmax to standardize the result output to obtain the probability distribution of all categories.

[0083] In addition, other algorithms may be selected for implementation according to specific requirements. The present invention only provides these two algorithms.

[0084] like Figure 3 As shown, in step 300, the path planning of the cleaning robot is performed based on the location of the attachments at various locations, specifically:

[0085] Step 301: Obtain the location of each attachment;

[0086] Step 302: Obtain the location of the cleaning robot;

[0087] Step 303: acquiring the position of the attachment closest to the cleaning robot based on the position of the cleaning robot;

[0088] Step 304: performing path planning based on an improved ant colony algorithm according to the position of the cleaning robot and the position of the attachment closest to the cleaning robot;

[0089] Step 305: Control the cleaning robot to reach the designated attachment along the path and perform cleaning;

[0090] Step 306: Repeat the above steps until all attachments are removed.

[0091] The ant colony algorithm improvement recorded in the present invention is mainly based on the traditional ant colony algorithm. The traditional ant colony algorithm is not introduced here. The improvements include:

[0092] Improvement of the heuristic information function: The traditional ant colony algorithm has shown significant application potential in many optimization problems due to its unique bionic principles and good distributed computing characteristics. However, in the process of extensive application and in-depth research, the traditional ant colony algorithm has some limitations and shortcomings in practical applications. The main disadvantages include: blindness in the initial search stage, slow convergence speed, and easy to fall into local optimal solutions. These problems mainly stem from the fact that the pheromone concentration on each path is the same in the initial stage of the algorithm, and there is a lack of effective heuristic information to guide the search direction, resulting in limited search efficiency and path optimization capabilities. In view of these shortcomings of the traditional ant colony algorithm, the present invention adds the concept of artificial potential field gravity to the heuristic information function. The core idea of ​​this strategy is to introduce the relative position and distance information between the target point and the heuristic information to enhance the ant's directionality to the target during the search process. The improved heuristic information function takes into account the relative distance from the ant's current position to the target point, and guides the ant to explore the target area more effectively by adjusting the distance factor and the influence of the artificial potential field gravity;

[0093] The improved heuristic information function is expressed as:

[0094]

[0095] Wherein, μ is a constant used to adjust the degree of influence of the artificial potential field, and its value range is between (0, 1). λ is a proportional coefficient used to adjust the influence of the distance of the target point on the inspiration. ij is the distance from node i to node j, d(j,E) is the distance from node j to the target point E. The closer the node is to the target point, the greater the potential energy it receives, which can increase the probability of the path being selected.

[0096] In order to further enhance the search performance and adaptability of the algorithm, the following innovative formula is proposed:

[0097]

[0098] In the formula, θ is a newly introduced exponential parameter used to adjust d ij The influence of the heuristic information increases the sensitivity of the model to distance changes. σ is another adjustment factor used to balance the direct distance d ij The innovative heuristic information function not only maintains the strong directionality to the target point, but also improves the ability of the algorithm to avoid local optimal solutions and accelerate convergence in complex environments.

[0099] Improve the pheromone update rule: Although the traditional ant colony algorithm has shown good search capabilities in a variety of optimization problems, as the number of iterations increases, the algorithm is prone to the accumulation of pheromone concentrations, causing the search process to fall into a local optimal solution, thereby affecting the ability to find the optimal path. This is because during the iteration of the algorithm, the accumulated pheromones on the path will cause subsequent ants to tend to choose paths that have been frequently visited by the previous ants, thereby reducing the diversity of the search space and the possibility of exploring new paths. In order to overcome this problem, the present invention proposes an adaptive pheromone adjustment mechanism, the purpose of which is to enhance the global search capability of the algorithm and avoid premature convergence to the local optimal solution by dynamically adjusting the pheromone concentration. The core strategy is to introduce an adaptive adjustment method based on the difference between the pheromones of the previous generation of optimal path and the worst path on the basis of the original pheromone update rule. It not only considers the pheromone increment left by the ants on the current path, but also considers the difference in pheromone concentration between the optimal path and the worst path, so as to adjust the update amount of pheromones, thereby prompting the algorithm to explore new paths while effectively avoiding the algorithm from falling into the local optimal solution;

[0100] The adaptive pheromone update formula is:

[0101]

[0102] Among them, τ ij (t+1) represents the path at the next moment<i,j> The pheromone concentration on the surface, (1-ρ)τ ij (t) represents the naturally volatile part of the pheromone, ρ is the pheromone volatility, Δτ ij is the kth ant on the path<i,j> The increase of pheromones left on the It is the additional pheromone adjustment amount based on the difference between the best path and the worst path pheromone, and the calculation formula is:

[0103]

[0104] Among them, v and w represent the number of ants on the optimal path and the worst path respectively, Q is the pheromone intensity coefficient, L min and L max Represent the optimal path length and the worst path length respectively;

[0105] In this way, when updating pheromones, not only the rewards for good paths are increased, but also certain penalties are imposed on poor paths, thereby achieving adaptive adjustment of the search direction during the algorithm's iteration process;

[0106] In order to further improve the flexibility and adaptability of the algorithm, a dynamic adjustment factor α(t) is introduced to adjust the influence of the additional pheromone adjustment amount so that it changes dynamically with the number of iterations. In this way, the algorithm can increase the tendency to explore new paths in the early stage and focus more on optimizing the discovered good paths in the later stage. The introduction of the dynamic adjustment factor is:

[0107]

[0108] Among them, α(t) is a dynamic factor that changes with the number of iterations t, which aims to balance the relationship between exploration and utilization to adapt to the search needs at different stages.

[0109] In step 304, path planning is performed based on the improved ant colony algorithm, specifically:

[0110] Taking the position of the cleaning robot as the starting point and the position of the attachment closest to the cleaning robot as the end point, the starting point and the end point area are divided into different grid areas based on the grid method, and the coordinates of the starting point and the end point are determined;

[0111] Initialize the parameters of the ant colony algorithm, set the maximum number of iterations K, the ant colony size M, the pheromone importance factor α, the heuristic factor β, and the pheromone volatility coefficient ρ;

[0112] When searching for a path, the ants select the next node based on the state transition probability formula. When searching for a path, the ant colony determines the next moving direction by judging the pheromone concentration and heuristic information. The ants move along the path until they reach the end or complete this iteration.

[0113] Record the length of each path and the number of ants on each path, and store the information in the cell structure CELL. Subsequent pheromones are updated based on this information.

[0114] According to the recorded information, find the best path and the worst path in this iteration, and adjust the pheromone concentration on the path according to the pheromone update rule;

[0115] If the maximum number of iterations K is reached, the optimal path is output, otherwise, the taboo table TABUK is cleared and a new iteration is performed;

[0116] The obtained optimal path is optimized, and unnecessary turning points in the path are removed based on the triangle pruning method to obtain the final path;

[0117] Introducing the triangle pruning method, which is an efficient path optimization method. The core is to identify and connect adjacent turning points on the path and check whether the newly formed straight path crosses obstacles. If not, these turning points can be safely removed to make the path straighter and shorter.

[0118] Let P = {p1, p2, ..., p n} is the set of all points on the path, which is also the original path of the ant colony. Let V = {v1,v2,...,v n} is the set of turning points on the path that need to be optimized, where Let O be the set of obstacles. For any two adjacent turning points v i and v i+1 If the straight line path Do not cross any obstacles, that is:

[0119]

[0120] Then we can remove all the intermediate points between these two turning points and directly connect v i and v i+1 , the updated path is expressed as:

[0121]

[0122] Among them, P′ represents the optimized path, P\{v i+1 ,v i+2 ,...,v i+k-1} means to remove v from the path i and v i+1 All the intermediate points between Indicates that a new straight path is added to the path. Since unnecessary turning points and intermediate points are removed, the optimized path P′ is straighter and shorter, improving the performance of the robot in practical applications.

[0123] The present invention provides an embodiment to verify the advantages of the improved ant colony algorithm proposed in the present invention in path planning. The present invention compares the improved ant colony algorithm proposed in the present invention with the traditional ant colony algorithm by setting four key performance indicators (optimal path length, running time, number of first convergence iterations, and turning points). The experimental environment of this embodiment is: Processor Intel(R) Core(TM) i7-9750HCPU@2.60GHz2.59Ghz, MATLAB R2018b, the simulation parameters are shown in Table 2;

[0124] Table 2 Simulation experiment parameters

[0125]

[0126]

[0127] The simulation results are shown in Table 3;

[0128] Table 3 Simulation results

[0129]

[0130] Compared with the traditional ant colony algorithm, the algorithm of the present invention reduces the shortest path length by 31.5%, and achieves a significant reduction of 44.4% and 59% in the number of iterations and turning points respectively. This result not only demonstrates the high efficiency of the algorithm in path optimization, but also reflects its significant improvement in iteration speed and path optimization, fully meeting the dual requirements of efficiency and practicality of robot path planning.

[0131] The present invention also provides a control system for a robot for cleaning pollution on a hull surface below the waterline, comprising:

[0132] The cleaning robot body is used to clean the attached objects;

[0133] The present invention provides an embodiment of a cleaning robot body, which is composed of a fuselage, a chain-driven electromagnetic adsorption crawler mechanism and a cleaning mechanism. The density of the robot is controlled to be slightly lower than the density of seawater, and wireless control is adopted at the same time, so that the robot can float to the surface of the water by itself in the event of an accident while reducing the control influence of buoyancy on electromagnetic force, and be dragged back by a control line. The chain-driven crawler mechanism of the robot is magnetic and can move in all directions on the bulkhead and hull. When the cleaning operation starts, the electromagnet on the belly of the fuselage is turned on, and the magnetic force is controlled by controlling the current, so that the robot can withstand the reaction force generated by high-pressure jet cleaning and the water flow resistance during the navigation of the ship and firmly adsorb on the hull and cabin, so that the tank cleaning robot can smoothly perform cleaning work on the hull and bulkhead, and the crew can realize real-time observation of the cleaning work through the front and rear camera images of the robot, and remote control of the robot makes tank cleaning more convenient and intelligent. When washing the bulkhead, the crew controls the left and right swing motor on the jet mechanism via wire to swing the nozzle in the desired direction of washing. At the same time, they can control the size of the nozzle opening to change the amount of water sprayed, and turn on the nylon brush mechanism to scrub stubborn stains.

[0134] The chain drive crawler mechanism is mainly composed of a permanent magnetic crawler, a signal receiver and two independent drive motors (distributed on the crawlers on the left and right sides). Permanent magnets and electromagnetic adsorption units are evenly distributed on the chain, which can be adsorbed on the cargo hold wall and the hull and roll back and forth with the chain to achieve forward and turning movements. When operating at high altitudes and on the bottom of the ship, the tank cleaning robot needs to overcome its own gravity, the shaking force when encountering waves hitting the hull, the reaction force of the sprayed impact water, and the water flow resistance. In order to enable the robot to perform cleaning operations stably during the navigation of the ship, in addition to the permanent magnetic crawler, the robot is also equipped with an electromagnetic adsorption unit on the crawler. The electromagnet consists of a permanent magnetic adsorption unit and a magnetic isolation plate. The magnetic isolation plate can effectively reduce the repulsive force between the permanent magnetic adsorption units and increase the attraction of the magnet to the hull and the bulkhead. When the robot starts cleaning, the high-pressure jet nozzle adjusts the amount of water sprayed according to the stubbornness of the stains on the bulkhead. The reaction force of different sizes will make the robot unstable. The resistance of the water flow when cleaning the bottom of the ship during navigation will also make the body unstable. By using the existing electronic control technology, the magnetic force of the electromagnet can be controlled, so that the robot can be firmly attached to the hull and bulkhead to perform stable work.

[0135] The jet cleaning mechanism is divided into two high-pressure nozzles in front and behind the fuselage, which can be sprayed and then cleaned. The high-pressure nozzle is composed of a control system, a nozzle, a valve, a pipeline system, etc. Before the robot works, the crew connects the high-pressure water pump on the ship to the water connector of the robot to obtain a continuous source of high-pressure water. The nozzle uses the change of the inner hole area to gather high-pressure water and convert it into kinetic energy to spray in the form of a high-speed jet, so as to achieve the desired cleaning effect. The control system is composed of integrated circuits, which are mainly used to control the opening and closing of the jet valve and the size of the valve opening to control the amount of water sprayed. The pipeline is mainly used to transport high-pressure water. The nozzle is the core of the tank cleaning robot. The balanced cylindrical jet nozzle can reduce the resistance of the water flow sprayed by the nozzle. At the same time, the front and rear nozzles are sprayed at the same time, almost offsetting the reaction force generated by the spray. In the face of different dirt, in order to achieve a good cleaning effect, the tank cleaning crew must first understand the working conditions on site, which are generally determined by the characteristics of the dirt, the effect and efficiency of cleaning, the cleaning medium, the shape of the cleaning object and the on-site environment.

[0136] The nylon brush roller mechanism consists of multiple nylon brush rollers at the bottom of the machine, a drive motor, a control system, etc. During the cleaning operation, the nylon brush drive motor is turned on, and the motor drives the nylon brush roller to rotate, brushing the bulkhead and hull to achieve the desired cleaning effect. The control system is composed of integrated circuits, which are mainly used to control the opening and closing of the drive motor, control the rotation speed, and select brush strips with different lengths, thicknesses and densities of bristles, thereby controlling the cleaning intensity. The nylon brush roller is the core of the cleaning mechanism. When the bristles in individual areas are severely worn and broken, the brush strip can be removed and replaced separately without replacing the entire strip, saving material costs and improving efficiency.

[0137] Other detection equipment may also be additionally provided on the pollution cleaning robot, for example, an infrared thermal imager, a temperature measuring device, a sensor, an ultrasonic sensor group, a three-axis gyroscope, a GPS positioning module, and the like.

[0138] An image acquisition module is provided on the body of the pollution-cleaning robot or the hull and is used to acquire images of the hull below the waterline;

[0139] The attachment recognition module is used to identify the attachments on the hull based on the image of the hull and determine the location of each attachment;

[0140] The path planning module is used to plan the path of the pollution cleaning robot based on the locations of various attachments.

[0141] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0142] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A control method for a robot for cleaning pollution on the surface of a ship below the waterline, It is characterized in that include: Acquire images of the hull below the waterline; Based on the improved EfficientNet network and the image of the hull, the attachments on the hull are identified and the positions of the attachments are determined. The improvement process of the EfficientNet network is as follows: the MBConv module in the shallow layer of the EfficientNet-B0 model in the EfficientNet network is replaced with a Fused-MBConv module, that is, the 1×1 convolution operation and the 3×3 depth-separable convolution in the main branch of the original MBConv module are replaced with a common 3×3 convolution operation, the last stage with a step size of 1 in the original EfficientNet network structure is removed, and the 5×5 convolution kernel in the original structure is replaced with a 3×3 convolution kernel; Path planning for the pollution-cleaning robot based on the locations of various attachments; Control the cleaning robot to reach the designated attachment along the path and carry out cleaning.

2. The method according to claim 1, It is characterized in that Identify the attachments on the hull and determine the location of each attachment, specifically: Build an attachment recognition model based on the improved EfficientNet network; Obtain various images of ship hull attachments, perform data enhancement on them, and construct a ship hull attachment dataset; The attachment recognition model is trained based on the hull attachment data set to obtain a trained attachment recognition model; The collected images of the hull below the waterline are input into the trained attachment recognition model to obtain the locations of various attachments.

3. The method according to claim 2, It is characterized in that Data enhancement is performed on various images of hull attachments, specifically: The images of various types of hull attachments are subjected to set transformation, brightness adjustment, saturation adjustment, contrast adjustment, random noise addition, as well as mirroring, scaling and rotation operations to achieve data enhancement.

4. The method according to claim 1 or 2, It is characterized in that The path planning of the cleaning robot is carried out based on the location of the attachments, specifically: Get the location of each attachment; Get the location of the pollution-cleaning robot; Based on the position of the cleaning robot, the position of the attachment closest to the cleaning robot is obtained; Path planning is performed based on the improved ant colony algorithm according to the position of the cleaning robot and the position of the attachment closest to the cleaning robot; Control the cleaning robot to reach the designated attachment along the path and perform cleaning; Repeat the above steps until all the attachments are removed.

5. The method according to claim 4, It is characterized in that Path planning is performed based on the improved ant colony algorithm, specifically: Taking the position of the cleaning robot as the starting point and the position of the attachment closest to the cleaning robot as the end point, the starting point and the end point area are divided into different grid areas based on the grid method, and the coordinates of the starting point and the end point are determined; Initialize the parameters of the ant colony algorithm, set the maximum number of iterations K, the ant colony size M, the pheromone importance factor a, the heuristic factor β, and the pheromone volatility coefficient ρ; When searching for a path, the ants select the next node based on the state transition probability formula. When searching for a path, the ant colony determines the next moving direction by judging the pheromone concentration and heuristic information. The ants move along the path until they reach the end or complete this iteration. Record the length of each path and the number of ants on each path, and store the information in the cell structure CELL. Subsequent pheromones are updated based on this information. According to the recorded information, find the best path and the worst path in this iteration, and adjust the pheromone concentration on the path according to the pheromone update rule; If the maximum number of iterations K is reached, the optimal path is output, otherwise, the taboo table TABUK is cleared and a new iteration is performed; The obtained optimal path is optimized, and unnecessary turning points in the path are removed based on the triangle pruning method to obtain the final path.

6. A robot control system for cleaning pollution on the surface of a ship below the waterline. It is characterized in that include: The cleaning robot body is used to clean the attached objects; An image acquisition module is provided on the body of the pollution-cleaning robot or the hull and is used to acquire images of the hull below the waterline; The attachment recognition module is used to identify the attachments on the hull based on the image of the hull and determine the location of each attachment; The path planning module is used to plan the path of the pollution cleaning robot based on the locations of various attachments.

Images