Path planning method of breeding net cage cleaning mechanical arm

Abstract

The invention provides an aquaculture net cage cleaning mechanical arm path planning method which comprises the steps that S11, three-dimensional coordinate data of a to-be-cleaned area are obtained and subjected to two-dimensional mapping, and a two-dimensional to-be-cleaned area is obtained; s12, if the number of the two-dimensional to-be-cleaned areas is smaller than or equal to a preset threshold value, path planning is conducted on the two-dimensional to-be-cleaned areas directly through an ant colony optimization algorithm; if the number of the two-dimensional to-be-cleaned areas is greater than a preset threshold value, performing block processing on the two-dimensional to-be-cleaned areas through a clustering algorithm to obtain a plurality of two-dimensional to-be-cleaned sub-areas, and performing path planning on the plurality of two-dimensional to-be-cleaned sub-areas through an ant colony optimization algorithm; s13, integrating the optimal paths of the two-dimensional to-be-cleaned area / the two-dimensional to-be-cleaned sub-area to generate a global path; and S14, carrying out inverse mapping on the global path to obtain a three-dimensional space path planning graph. According to the method, through a three-level architecture of regional blocking, clustering optimization and global integration, the task balanced distribution problem during collaborative operation of multiple mechanical arms is systematically solved.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a path planning method for a farming cage cleaning robot arm. Background Art

[0002] In marine cage aquaculture scenarios, nets are prone to the problem of fouling organisms attaching due to long-term immersion. The large accumulation of these organisms will significantly increase the stress load on the floats and fishing nets. In severe cases, it may cause damage to the cage structure and breakage of the fishing net, leading to the escape of a large number of fish. Fouling organisms have always been a major problem in the field of cage aquaculture. Against this background, underwater cleaning robots for aquaculture cage nets have emerged. The current mainstream underwater cleaning robots use the traditional ant colony optimization algorithm (ACO) for path planning, but the area to be cleaned often presents the characteristics of "sparse distribution and large-scale dense distribution coexisting" (for example, the fouling is sparse at the edge of the cage and dense in the center of the net), resulting in the following core problems of the traditional ACO algorithm: 1. As the number of areas to be cleaned increases, the size of the ant colony needs to increase accordingly, resulting in an exponential increase in computational complexity, a significant increase in the number of iterative convergence times, and a sharp increase in computational time.

[0003] 2. For large-scale areas, the traditional ACO algorithm is prone to falling into local optimal solutions, which reduces the accuracy and efficiency of path planning and makes it difficult to find the optimal path within a reasonable time.

[0004] 3. When the area to be cleaned is unevenly distributed, the algorithm has poor adaptability and cannot specifically handle the path optimization needs of areas with different densities.

[0005] In existing technologies, the time complexity of using the ACO algorithm alone for path planning is positively correlated with the number of iterations, the size of the ant colony, and the number of areas to be cleaned. For large numbers of contaminated areas to be cleaned, blindly increasing the number of ant colonies leads to redundant calculations, with numerous ants repeatedly exploring known paths, significantly increasing computational costs. Furthermore, traditional ACO algorithms lack regional preprocessing capabilities and are unable to optimize block-by-block based on regional distribution characteristics, resulting in inefficient planning in complex scenarios. Summary of the Invention

[0006] The purpose of the present invention is to provide a path planning method for aquaculture cage cleaning robot arm, which systematically solves the problem of balanced task allocation during collaborative operation of multiple robot arms through a three-level architecture of regional blocking-cluster optimization-global integration.

[0007] To achieve the above objectives, the present invention provides a method for path planning of a cage cleaning robot arm, the method comprising: S11, obtaining the three-dimensional coordinate data of the area to be cleaned and performing two-dimensional mapping to obtain a two-dimensional area to be cleaned; S12, if the number of the two-dimensional areas to be cleaned is less than or equal to the preset threshold, directly performing path planning for the two-dimensional areas to be cleaned using the ant colony optimization algorithm; If the number of two-dimensional areas to be cleaned is greater than a preset threshold, the two-dimensional areas to be cleaned are divided into blocks using a clustering algorithm to obtain several two-dimensional sub-areas to be cleaned, and then the ant colony optimization algorithm is used to plan paths for the sub-areas to be cleaned. S13, integrating the optimal paths of the two-dimensional area to be cleaned / the two-dimensional sub-areas to be cleaned to generate a global path; S14. Inversely map the global path to obtain a three-dimensional space path planning map.

[0008] Furthermore, in step S11, the three-dimensional coordinate data of the area to be cleaned is obtained, which specifically includes: Use image recognition algorithm to analyze the net image captured by the camera, output several areas to be cleaned, and extract the coordinates of the bounding box of each area to be cleaned; Connect the bounding box coordinates of each area to be cleaned in sequence to form several connected areas; Perform contour analysis on each connected area and calculate the center coordinates to obtain the three-dimensional coordinate data (x', y', z') corresponding to each area to be cleaned.

[0009] Furthermore, in step S11, the three-dimensional coordinate data of the area to be cleaned is subjected to two-dimensional mapping, specifically including: The spherical coordinate mapping method is used to convert the three-dimensional coordinate data (x', y', z') of the area to be cleaned into two-dimensional plane coordinates ( ), the expression is:

[0010] Wherein, R is the radius of the cage; is the horizontal azimuth, ; is the zenith angle, , the constant 11520 is the size of the grid for discretizing the three-dimensional surface.

[0011] Furthermore, in step S14, the global path is reverse mapped, specifically including: The coordinate points of the two-dimensional path ( ) The spherical coordinate mapping method is used to perform inverse mapping and restore the three-dimensional coordinate data (x', y', z') of the area to be cleaned to obtain a three-dimensional space path planning map.

[0012] Furthermore, in step S12, path planning is performed on the two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned, specifically including: Initialize ant colony parameters, including the number of ants, pheromone factor α, heuristic factor β, pheromone evaporation rate ρ, and pheromone intensity Q; Construct a path planning model for each 2D area to be cleaned / 2D sub-area to be cleaned. Ants start from any area to be cleaned and select the next area to be cleaned based on the state transition probability. The state transition probability formula is:

[0013] in, is the state transition probability of the kth ant from area i to j, is the pheromone concentration from region i to region j in the tth iteration; is the Euclidean distance between two regions The inverse of , i.e., heuristic information; is the set of areas that ant k has not visited when it is in area i, is an area in the set of other areas that the k-th ant has not visited when it is in area i; After all ants complete a traversal, the pheromone concentration is updated, and the expression is:

[0014] in, is the pheromone concentration from region i to region j in the t+1th iteration, is the number of ants on the path in the tth iteration The total amount of pheromones left on is the number of ants on the path in this iteration The amount of pheromone left on the ant cycle model is calculated using the expression:

[0015] Among them, L is the total length of the k-th ant’s path; Repeat the above ant colony optimization process until the path length between each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned converges to a preset value or reaches a preset number of iterations, and obtain the path planning for each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned.

[0016] Furthermore, in step S12, the two-dimensional area to be cleaned is divided into blocks, specifically including: S21, determining the number K of clusters of the two-dimensional area to be cleaned according to the elbow method; S22. Randomly select K initial cluster centers and calculate the Euclidean distance from each coordinate point to each cluster center in the two-dimensional area to be cleaned. The expression is:

[0017] in, is the Euclidean distance; S23, comparing the Euclidean distances of each coordinate point to each cluster center, and assigning the coordinate point to the region to which the cluster center belongs that is less than a preset distance threshold; S24. If the Euclidean distances of the coordinate points to multiple cluster centers are the same, they are randomly assigned to the areas belonging to the cluster centers with the same distances; S25. After the coordinate points are divided, the cluster centers of the sub-regions to be cleaned are recalculated. The expression is:

[0018] in, is the horizontal coordinate of the cluster center of the sub-region to be cleaned, is the number of areas to be cleaned in the Kth area, is the ordinate of the cluster center of the sub-region to be cleaned; S26 , repeating steps S23 and S25 until the change in each cluster center is less than a preset threshold or reaches a preset number of iterations, thereby obtaining a number of sub-regions to be cleaned.

[0019] Furthermore, the elbow method is used to determine the K value, which specifically includes: Calculate the sum of squared errors for each k value, the expression is:

[0020] in, is the preset number of clusters, is the sample set of the i-th cluster, for The samples in is the distance from sample x to the cluster center, expressed as:

[0021] , is the i-th cluster center, the cluster center point The update formula is as follows:

[0022] in, is the coordinate of the jth sample of the i-th cluster, is the jth sample, is the number of samples in the i-th cluster; by The value is the horizontal axis, Draw a curve for the vertical axis, and consider the elbow point where the curve changes from a rapid decline to a slow decline as the optimal number of clusters for the corresponding K value.

[0023] Furthermore, in step S12, when the total number of areas to be cleaned after the block processing is greater than a preset threshold, the total area to be cleaned is divided into more than or equal to 2 total sub-areas to be cleaned, and path planning is performed on each sub-area to be cleaned.

[0024] Furthermore, in step S14, the optimal paths of each area are integrated, specifically including: The optimal paths of each area to be cleaned / sub-area to be cleaned are connected through spatially adjacent connection points in the order of spatial positions to form a global path.

[0025] Furthermore, connections are made through spatially adjacent connection points, specifically including: From each path of the area to be cleaned, the boundary point closest to the center of the adjacent area to be cleaned is extracted, the Euclidean distance between the boundary point and the boundary point of the adjacent area to be cleaned is calculated, and the calculated Euclidean distances are compared pairwise. The point pair with the smallest Euclidean distance between the boundary point and the boundary point of the adjacent area to be cleaned is selected as the connection point.

[0026] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a path planning method for a cage cleaning robot arm. The method divides a large area to be cleaned into multiple sub-areas to be cleaned through K-means clustering. Each sub-area is independently optimized by ant colony optimization, which significantly reduces the problem scale of a single planning and reduces the computational time by about 6%-8%. The method is particularly effective in large-scale scenarios. The number of clusters is dynamically adjusted according to the size of the area to be cleaned. The method is applicable to scenarios with different numbers of areas to be cleaned, ranging from 10 to more than 100, and has good versatility. Regional planning reduces the blindness of ant searches and significantly reduces the number of iterative convergences. For 30 areas to be cleaned, the number of convergences is reduced from about 110 before the improvement to about 95, and the error fluctuation range is reduced from 7.64% to 3.47%. At the same time, regional processing can adjust ant colony parameters for areas to be cleaned of different densities, avoiding the local optimal problem of traditional algorithms under complex distributions. The path length error is significantly reduced, and the block results can be directly assigned to multiple cleaning robots for collaborative work, coordinating work time and further improving cleaning efficiency. The present invention systematically solves the problem of balanced task allocation during collaborative operation of multiple robotic arms through a three-level architecture of regional segmentation, cluster optimization, and global integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work. Figure 1 A schematic flow chart of a method for path planning of a robot arm for cleaning aquaculture cages provided by an embodiment of the present invention; Figure 2 A schematic diagram of an ant colony optimization algorithm for a path planning method of a farming cage cleaning robot arm provided by an embodiment of the present invention; Figure 3 A schematic diagram of the path planning method for a robot arm for cleaning aquaculture cages provided by an embodiment of the present invention; Figure 4 A schematic diagram of determining the K value using the elbow rule in a path planning method for an aquaculture cage cleaning robot arm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0029] Reference Figure 1 This embodiment provides a method for planning a path for a farming cage cleaning robot arm, the method comprising: S11 , obtaining three-dimensional coordinate data of the area to be cleaned and performing two-dimensional mapping to obtain a two-dimensional area to be cleaned.

[0030] S12: If the number of the two-dimensional areas to be cleaned is less than or equal to a preset threshold, the path planning for the two-dimensional areas to be cleaned is performed directly using an ant colony optimization algorithm.

[0031] If the number of two-dimensional areas to be cleaned is greater than a preset threshold, the two-dimensional areas to be cleaned are divided into blocks using a clustering algorithm to obtain several two-dimensional sub-areas to be cleaned, and then the ant colony optimization algorithm is used to plan paths for the several two-dimensional sub-areas to be cleaned.

[0032] S13, integrating the optimal paths of the two-dimensional area to be cleaned / the two-dimensional sub-areas to be cleaned to generate a global path.

[0033] S14. Inversely map the global path to obtain a three-dimensional space path planning map.

[0034] This embodiment addresses the path planning challenge for curved surface cleaning by mapping the underwater 3D net coordinates onto a 2D plane. The number of clusters, K, is dynamically adjusted based on the spatial density of the area to be cleaned (the contaminated area). This replaces the traditional fixed partitioning scheme. Using the elbow algorithm, the number of clusters is dynamically set, assigning smaller clusters to sparse areas and larger clusters to dense areas. This overcomes the traditional fixed partitioning's adaptability to unevenly distributed scenes and significantly improves subsequent path planning efficiency.

[0035] The Ant Colony Optimization (ACO) algorithm runs independently on each two-dimensional area to be cleaned or two-dimensional sub-area to be cleaned, leveraging a local pheromone positive feedback mechanism for rapid convergence. The number of two-dimensional areas to be cleaned or two-dimensional sub-areas to be cleaned is limited to ≤20, reducing the computational complexity of a single ACO run to polynomial level. A state transition probability formula is used to guide ant search, and pheromones are updated in conjunction with an ant cycle model. The global large-scale TSP problem is decomposed into multiple small-scale sub-problems to prevent the algorithm from falling into local optimality. The optimal paths of each area to be cleaned are connected according to the principle of spatial proximity, supporting balanced task allocation across multiple robotic arms. A globally continuous trajectory is generated, with the center coordinates of the optimal paths within the area to be cleaned serving as the connection point. Paths within the area to be cleaned are dynamically assigned to different robotic arms based on their lengths, enabling parallel cleaning. This eliminates redundant movement when joining paths within the area to be cleaned, improving the efficiency of multi-robot collaboration.

[0036] As a preferred embodiment, in step S11, obtaining the three-dimensional coordinate data of the area to be cleaned specifically includes: An image recognition algorithm is used to analyze the net image captured by the camera, output several areas to be cleaned, and extract the coordinates of the boundary box of each area to be cleaned.

[0037] Connect the bounding box coordinates of each area to be cleaned in sequence to form several connected areas.

[0038] Perform contour analysis on each connected area and calculate the center coordinates to obtain the three-dimensional coordinate data (x', y', z') corresponding to each area to be cleaned.

[0039] In this embodiment, the YOLOv5 algorithm can be used to identify the net clothing image captured by the camera, output the coordinates of the boundary box of the stained area, connect the coordinates of the boundary box of the stained area in sequence to form a connected area, perform contour analysis on the connected area, calculate the center coordinates (x', y', z') of each connected area, and output the three-dimensional coordinate data (x', y', z') corresponding to each connected area.

[0040] As a preferred embodiment, in step S11, the three-dimensional coordinate data of the area to be cleaned is subjected to two-dimensional mapping, which specifically includes: The spherical coordinate mapping method is used to convert the three-dimensional coordinate data (x', y', z') of the area to be cleaned into two-dimensional plane coordinates ( ), the expression is:

[0041] Where R is the cage radius (unit: meter); is the horizontal azimuth, ; is the zenith angle, and the extreme singularity can be avoided by limiting the range, The constant 11520 is the size of the grid for discretizing the three-dimensional surface. The three-dimensional surface is discretized into a grid (about 130 million pixels), ensuring that the spacing between cleaning points is greater than the grid resolution. At the same time, the coordinates are rounded to simplify the calculation.

[0042] In this embodiment, the cage surface is a three-dimensional structure. The actual distance between regions should be the geodesic distance along the surface. The cage surface can be approximated as a plane, or the region with a large curvature can be converted into a two-dimensional problem by using a coordinate transformation mapping method. Thus, the three-dimensional coordinates (x', y', z') of the center of the area to be cleaned are obtained and converted into two-dimensional plane coordinates ( ).

[0043] As a preferred embodiment, in step S14, the global path is reverse mapped, specifically including: The coordinate points of the two-dimensional path ( ) The spherical coordinate mapping method is used to perform inverse mapping and restore the three-dimensional coordinate data (x', y', z') of the area to be cleaned to obtain a three-dimensional space path planning diagram.

[0044] In this embodiment, after the optimal traversal path is generated in the two-dimensional plane by ACO (ant colony algorithm) or improved K-means-ACO, it is inversely mapped back to the three-dimensional space, and the controller converts the two-dimensional path points ( ) The inverse transformation is achieved through the spherical coordinate mapping method, and the three-dimensional coordinates are restored to (x', y', z'). These three-dimensional coordinates serve as the target motion trajectory points of the cleaning disk at the end of the robot arm to guide the movement of the robot arm.

[0045] As a preferred embodiment, in step S12, path planning is performed on the two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned, specifically including: Initialize ant colony parameters, including the number of ants, pheromone factor α, heuristic factor β, pheromone evaporation rate ρ, and pheromone intensity Q; Construct a path planning model for each 2D area to be cleaned / 2D sub-area to be cleaned. Ants start from any area to be cleaned and select the next area to be cleaned based on the state transition probability. The state transition probability formula is:

[0046] in, is the state transition probability of the kth ant from area i to j, is the pheromone concentration from region i to region j in the tth iteration; is the Euclidean distance between two regions The inverse of , i.e., heuristic information; is the set of areas that ant k has not visited when it is in area i, is an area in the set of other areas that the k-th ant has not visited when it is in area i; After all ants complete a traversal, the pheromone concentration is updated, and the expression is:

[0047] in, is the pheromone concentration from region i to region j in the t+1th iteration, is the number of ants on the path in the tth iteration The total amount of pheromones left on is the number of ants on the path in this iteration The amount of pheromone left on the ant cycle model is calculated using the expression:

[0048] Among them, L is the total length of the k-th ant’s path; Repeat the above ant colony optimization process until the path length between each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned converges to a preset value or reaches a preset number of iterations, and obtain the path planning for each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned.

[0049] In this embodiment, refer to Figure 2 and Figure 3 , the entire area to be cleaned is divided into multiple sub-areas to be cleaned, and the ant colony optimization algorithm (ACO) is applied to each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned for path planning. The computational complexity of ACO typically increases sharply with the increase in the problem scale, often exponentially or at a high-order polynomial level. By decomposing a large-scale problem into multiple small-scale sub-problems, the number of nodes in each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned is far less than the global number of nodes. When running ACO on a single two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned, the space of path combinations that the ants need to explore and the size of the pheromone matrix that needs to be maintained are significantly reduced, greatly reducing the amount of computation required for a single iteration, accelerating the algorithm's convergence speed, and significantly shortening the overall solution time.

[0050] As a preferred embodiment, in step S12, the two-dimensional area to be cleaned is divided into blocks, specifically including: S21, determining the number K of clusters of the two-dimensional area to be cleaned according to the elbow method; S22. Randomly select K initial cluster centers and calculate the Euclidean distance from each coordinate point to each cluster center in the two-dimensional area to be cleaned. The expression is:

[0051] in, is the Euclidean distance; S23, comparing the Euclidean distances of each coordinate point to each cluster center, and assigning the coordinate point to the region to which the cluster center belongs that is less than a preset distance threshold; S24. If the Euclidean distances of the coordinate points to multiple cluster centers are the same, they are randomly assigned to the areas belonging to the cluster centers with the same distances; S25. After the coordinate points are divided, the cluster centers of the sub-regions to be cleaned are recalculated. The expression is:

[0052] in, is the horizontal coordinate of the cluster center of the sub-region to be cleaned, is the number of areas to be cleaned in the Kth area, is the ordinate of the cluster center of the sub-region to be cleaned; S26 , repeating steps S23 and S25 until the change in each cluster center is less than a preset threshold or reaches a preset number of iterations, thereby obtaining a number of sub-regions to be cleaned.

[0053] In this embodiment, K-means clustering is used to divide the area to be cleaned into several sub-areas, keeping the size of each sub-area within the efficient processing range of the ant colony algorithm (typically ≤ 20 areas / cluster). This reduces the time complexity of a single planning from [k] / [k] (where K is the number of clusters), thereby reducing computational overhead by 40%-60%. Through independent cluster optimization, large-scale, complex path searches are decomposed into several small-scale local optimization problems. Combined with the pheromone positive feedback mechanism of the ant colony algorithm, local optimal solutions are globally integrated to form a better global solution, reducing path length error by 35%-50% compared to traditional algorithms. By dynamically adjusting the number of clusters K to accommodate different density distributions and supporting balanced task allocation when multiple robotic arms collaborate, multi-robot cleaning efficiency is improved by 20%-30%.

[0054] As a preferred embodiment, the elbow method is used to determine the K value, which specifically includes: Calculate the sum of squared errors for each k value, the expression is:

[0055] in, is the preset number of clusters, is the sample set of the i-th cluster, for The samples in is the distance from sample x to the cluster center, expressed as:

[0056] , is the i-th cluster center, the cluster center point The update formula is as follows:

[0057] in, is the coordinate of the jth sample of the i-th cluster, is the jth sample, is the number of samples in the i-th cluster; by The value is the horizontal axis, Draw a curve for the vertical axis, and consider the elbow point where the curve changes from a rapid decline to a slow decline as the optimal number of clusters for the corresponding K value.

[0058] In this embodiment, refer to Figure 4 The elbow method transforms the abstract search for the optimal K value into a visual judgment by plotting the sum of squared errors (SSE) for different numbers of clusters (K values). The SSE represents the sum of the squared distances of all sample points to the center of the region to be cleaned, reflecting the compactness of the region to be cleaned. The inflection point is the recommended K value. Before the inflection point, the SSE decreases sharply as K increases, while after the inflection point, the SSE decreases more gradually. The elbow method simply runs the K-Means algorithm for different K values (usually K = 1 to a reasonable upper limit, such as 10) and calculates the SSE for each value.

[0059] As a preferred embodiment, in step S12, when the total number of areas to be cleaned after the block processing is greater than a preset threshold, the total area to be cleaned is divided into more than or equal to 2 total sub-areas to be cleaned, and then path planning is performed on each sub-area to be cleaned.

[0060] In this embodiment, when the area to be cleaned / sub-area to be cleaned is greater than 20 areas, the area greater than the preset number will be divided into two areas. After the total area is divided into two or more areas, ant colony path planning is performed on the areas to be cleaned within the divided areas respectively, and finally path integration is performed, which balances the clustering effect and model complexity.

[0061] As a preferred embodiment, in step S14, integrating the optimal paths of each area specifically includes: The optimal paths of each area to be cleaned / sub-area to be cleaned are connected through spatially adjacent connection points in the order of spatial positions to form a global path.

[0062] In this embodiment, the optimal paths of each area to be cleaned / sub-area to be cleaned are connected in sequence through spatially adjacent connection points to form a global path. Under the premise of ensuring the continuity and passability of the physical path, the optimality of each area to be cleaned / sub-area to be cleaned is maintained to the maximum extent. At the same time, the redundant movement between areas is eliminated through the direct connection strategy of adjacent points, thereby achieving dual optimization of global cleaning efficiency and energy consumption control. It not only avoids repeated calculation of planned areas, but also ensures the continuity of robot movement through the natural connection of spatial topology, so that the overall cleaning efficiency of the system approaches the optimal level.

[0063] As a preferred embodiment, connection is performed through spatially adjacent connection points, specifically including: From each path of the area to be cleaned, the boundary point closest to the center of the adjacent area to be cleaned is extracted, the Euclidean distance between the boundary point and the boundary point of the adjacent area to be cleaned is calculated, and the calculated Euclidean distances are compared pairwise. The point pair with the smallest Euclidean distance between the boundary point and the boundary point of the adjacent area to be cleaned is selected as the connection point.

[0064] In this embodiment, by selecting the point pairs with the smallest Euclidean distance among the boundary points of adjacent areas as connection points, the cross-area movement path can be shortened to the greatest extent, reducing the robot's idle driving distance and time loss, thereby significantly improving the global cleaning efficiency.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for path planning of a farming cage cleaning robot arm, characterized in that: The method comprises: S11, obtaining the three-dimensional coordinate data of the area to be cleaned and performing two-dimensional mapping to obtain a two-dimensional area to be cleaned; S12, if the number of the two-dimensional areas to be cleaned is less than or equal to the preset threshold, directly performing path planning for the two-dimensional areas to be cleaned using the ant colony optimization algorithm; If the number of two-dimensional areas to be cleaned is greater than a preset threshold, the two-dimensional areas to be cleaned are divided into blocks using a clustering algorithm to obtain several two-dimensional sub-areas to be cleaned, and then the ant colony optimization algorithm is used to plan paths for the sub-areas to be cleaned. S13, integrating the optimal paths of the two-dimensional area to be cleaned / the two-dimensional sub-areas to be cleaned to generate a global path; S14. Inversely map the global path to obtain a three-dimensional space path planning map.

2. The method for path planning of aquaculture cage cleaning robot arm according to claim 1, characterized in that: In step S11, the three-dimensional coordinate data of the area to be cleaned is obtained, which specifically includes: Use image recognition algorithm to analyze the net image captured by the camera, output several areas to be cleaned, and extract the coordinates of the bounding box of each area to be cleaned; Connect the bounding box coordinates of each area to be cleaned in sequence to form several connected areas; Perform contour analysis on each connected area and calculate the center coordinates to obtain the three-dimensional coordinate data (x', y', z') corresponding to each area to be cleaned.

3. The method for path planning of aquaculture cage cleaning robot arm according to claim 2, characterized in that: In step S11, the three-dimensional coordinate data of the area to be cleaned is subjected to two-dimensional mapping, which specifically includes: The spherical coordinate mapping method is used to convert the three-dimensional coordinate data (x', y', z') of the area to be cleaned into two-dimensional plane coordinates ( ), the expression is: Wherein, R is the radius of the cage; is the horizontal azimuth, ; is the zenith angle, , the constant 11520 is the size of the grid for discretizing the three-dimensional surface.

4. The method for path planning of aquaculture cage cleaning robot arm according to claim 3, characterized in that: In step S14, the global path is reverse mapped, specifically including: The coordinate points of the two-dimensional path ( ) The spherical coordinate mapping method is used to perform inverse mapping and restore the three-dimensional coordinate data (x', y', z') of the area to be cleaned to obtain a three-dimensional space path planning map.

5. The method for path planning of aquaculture cage cleaning robot arm according to claim 1, characterized in that: In step S12, path planning is performed on the two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned, specifically including: Initialize ant colony parameters, including the number of ants, pheromone factor α, heuristic factor β, pheromone evaporation rate ρ, and pheromone intensity Q; Construct a path planning model for each 2D area to be cleaned / 2D sub-area to be cleaned. Ants start from any area to be cleaned and select the next area to be cleaned based on the state transition probability. The state transition probability formula is: in, is the state transition probability of the kth ant from area i to j, is the pheromone concentration from region i to region j in the tth iteration; is the Euclidean distance between two regions The inverse of , i.e., heuristic information; is the set of areas that ant k has not visited when it is in area i, is an area in the set of other areas that the k-th ant has not visited when it is in area i; After all ants complete a traversal, the pheromone concentration is updated, and the expression is: in, is the pheromone concentration from region i to region j in the t+1th iteration, is the number of ants on the path in the tth iteration The total amount of pheromones left on is the number of ants on the path in this iteration The amount of pheromone left on the ant cycle model is calculated using the expression: Among them, L is the total length of the k-th ant’s path; Repeat the above ant colony optimization process until the path length between each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned converges to a preset value or reaches a preset number of iterations, and obtain the path planning for each two-dimensional area to be cleaned / two-dimensional sub-area to be cleaned.

6. The method for path planning of aquaculture cage cleaning robot arm according to claim 1, characterized in that: In step S12, the two-dimensional area to be cleaned is divided into blocks, which specifically includes: S21, determining the number K of clusters of the two-dimensional area to be cleaned according to the elbow method; S22. Randomly select K initial cluster centers and calculate the Euclidean distance from each coordinate point to each cluster center in the two-dimensional area to be cleaned. The expression is: in, is the Euclidean distance; S23, comparing the Euclidean distances of each coordinate point to each cluster center, and assigning the coordinate point to the region to which the cluster center belongs that is less than a preset distance threshold; S24. If the Euclidean distances of the coordinate points to multiple cluster centers are the same, they are randomly assigned to the areas belonging to the cluster centers with the same distances; S25. After the coordinate points are divided, the cluster centers of the sub-regions to be cleaned are recalculated. The expression is: in, is the horizontal coordinate of the cluster center of the sub-region to be cleaned, is the number of areas to be cleaned in the Kth area, is the ordinate of the cluster center of the sub-region to be cleaned; S26 , repeating steps S23 and S25 until the change in each cluster center is less than a preset threshold or reaches a preset number of iterations, thereby obtaining a number of sub-regions to be cleaned.

7. The method for path planning of aquaculture cage cleaning robot arm according to claim 6, characterized in that: Use the elbow method to determine the K value, which includes: Calculate the sum of squared errors for each k value, the expression is: in, is the preset number of clusters, is the sample set of the i-th cluster, for The samples in is the distance from sample x to the cluster center, expressed as: , is the i-th cluster center, the cluster center point The update formula is as follows: in, is the coordinate of the jth sample of the i-th cluster, is the jth sample, is the number of samples in the i-th cluster; by The value is the horizontal axis, Draw a curve for the vertical axis, and consider the elbow point where the curve changes from a rapid decline to a slow decline as the optimal number of clusters for the corresponding K value.

8. The method for path planning of aquaculture cage cleaning robot arm according to claim 6, characterized in that: In step S12, when the total number of areas to be cleaned after the block processing is greater than a preset threshold, the total area to be cleaned is divided into more than or equal to 2 total sub-areas to be cleaned, and path planning is performed on each sub-area to be cleaned.

9. The method for path planning of aquaculture cage cleaning robot arm according to claim 1, characterized in that: In step S14, the optimal paths of each area are integrated, specifically including: The optimal paths of each area to be cleaned / sub-area to be cleaned are connected through spatially adjacent connection points in the order of spatial positions to form a global path.

10. The method for path planning of aquaculture cage cleaning robot arm according to claim 1, characterized in that: Connect through spatially adjacent connection points, including: From each path of the area to be cleaned, the boundary point closest to the center of the adjacent area to be cleaned is extracted, the Euclidean distance between the boundary point and the boundary point of the adjacent area to be cleaned is calculated, and the calculated Euclidean distances are compared pairwise. The point pair with the smallest Euclidean distance between the boundary point and the boundary point of the adjacent area to be cleaned is selected as the connection point.

Images