An underwater area of interest coverage path planning method and device for an unmanned vehicle

Abstract

The application discloses a kind of underwater area of interest coverage path planning method and device of unmanned vehicle, solve the problems such as long operation path, multiple steering times, low coverage algorithm operation efficiency that are faced in large-scale ocean area coverage detection. Extract multiple interest areas in the user concerned depth range from the known information of topographic region;Multiple polygon boundary ranges are generated by clustering interest areas;Improved Bayazit algorithm is proposed to segment multiple polygons based on concave points to obtain multiple sub-convex polygons;Each two sub-convex polygons with common edge and consistent internal parallel line direction are merged to obtain multiple target interest areas;A discrete grouping teaching algorithm is proposed to plan the initial path of the target interest area;A multi-thread fast optimization method is constructed to optimize the initial path and achieve optimal coverage path planning. The application realizes autonomous and efficient detection and coverage planning of underwater topographic interest areas under the premise of limited endurance and limited computing power of unmanned vehicle detection force.

Description

Technical Field

[0001] This invention belongs to the field of underwater topographic detection technology, and more specifically, relates to a method and apparatus for underwater region of interest coverage path planning for unmanned aerial vehicles. Background Technology

[0002] Shallow waters such as shoals, reefs, and nearshore areas are characterized by well-developed marine ecosystems, complex and varied topography, and turbulent currents. For a long time, large marine geological survey vessels have been hampered by draft and complex wind, wave, and current conditions, making it difficult to approach shallow waters for mapping. Traditional manual bathymetry, with its scattered measurement methods, has also proven insufficient for obtaining detailed nautical chart information over a large area and periodically. However, with the development of unmanned aerial vehicles (UAVs) and sonar bathymetry technology, shallow water topographic mapping has undergone a fundamental transformation in recent years. Using UAVs equipped with single-beam or multi-beam bathymetry sonar to detect topographic and geological changes in shallow waters such as islands, reefs, lakes, and dams has become a highly valuable operational method.

[0003] Human exploration activities in vast waters often focus on local regions of interest to complete the mission within limited search resources and time. Based on known, coarse-grained nautical chart information or historical charts, detailed coverage exploration of these regions of interest has become an important operational requirement. Autonomous, full-coverage path planning for these regions of interest often faces challenges such as difficulty in extracting region boundaries, decomposing boundary convexity features, and traversing randomly distributed regions.

[0004] To address the need for fine-grained coverage detection of regions of interest, unmanned aerial vehicle (UAV) mission planning still requires further improvement in areas such as region boundary extraction, region segmentation, and inter-region path planning. This will solve problems such as long coverage paths, numerous turning times, and low computational efficiency of coverage algorithms in existing technologies for large-scale region coverage, so as to achieve autonomous and efficient mission planning and execution using the limited endurance and computing power of UAVs. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and apparatus for underwater region of interest coverage path planning for unmanned aerial vehicles (UAVs). The purpose is to enable UAVs to autonomously and efficiently detect and plan coverage for underwater terrain regions of interest, thereby solving the technical problem of low efficiency in underwater path planning in existing technologies.

[0006] To achieve the above objectives, according to one aspect of the present invention, an underwater region of interest coverage path planning method for an unmanned aerial vehicle is provided, comprising:

[0007] S1: Divide the terrain area information map with known prior information to extract multiple regions of interest within the user's attention depth range; cluster the multiple regions of interest to obtain clustering results, and generate polygonal boundaries of the regions of interest based on the clustering results;

[0008] S2: The improved Bayazit algorithm is used to segment the polygon boundary of the region of interest based on concave points to obtain multiple sub-convex polygons; every two sub-convex polygons with a common edge and internal parallel lines in the same direction are merged to obtain the target region of interest.

[0009] S3: The initial path of the target region of interest is planned using a discrete grouping teaching algorithm; the initial path is optimized using a multi-threaded fast optimization method to obtain the target path.

[0010] In one embodiment, S1 includes:

[0011] S11: Rasterize the terrain area information map according to the user's actual required resolution to obtain multiple area rasteres;

[0012] S12: Extract multiple regions of interest from the multiple region grids according to the user's depth range of interest; use the DBSCAN algorithm to cluster the multiple regions of interest to obtain multiple clustering results;

[0013] S13: Use the α-shapes algorithm to generate the bounding polygons of the regions corresponding to each clustering result, and use them as the boundaries of the polygons of the regions of interest.

[0014] In one embodiment, S2 includes:

[0015] S21: Based on the Bayazit concave polygon segmentation algorithm, the method of directly and greedily connecting the nearest vertex of the reflection arc is improved. When a concave point exists, the nearest vertex within the reflection arc range that is a concave point is connected. If no concave point exists, the nearest vertex is connected, thereby realizing the segmentation of the polygon boundaries corresponding to each region of interest and obtaining multiple sub-convex polygons.

[0016] S22: Merge sub-convex polygons with common edges and parallel lines in the same direction to obtain the target region of interest, thereby reducing the coverage path length and the number of turns.

[0017] In one embodiment, S21 includes:

[0018] Randomly select a vertex of the concave polygon formed by the boundary of the polygon to be decomposed for concavity point detection, and then check each vertex in a counterclockwise direction for concavity point detection; when the first concavity point is found... P i Afterwards, alongP i-1 P i and P i P i+1 Extend them in opposite directions and intersect at the polygon boundary;

[0019] If there are multiple vertices within the range of the intersection of the reverse extension line and the polygon boundary, then select the nearest point within the range that has the concave attribute and is closest to it;

[0020] If there are no vertices within the intersection range of the reflected ray and the polygon, then connect the center points of the two intersection points with auxiliary points to remove the concave points; finally, multiple sub-convex polygons are obtained.

[0021] In one embodiment, S22 includes:

[0022] If adjacent sub-convex polygons share a common edge and the parallel lines planned within the region are in the same direction, then the two sub-convex polygons are merged, and the direction of the parallel lines in the target interest region obtained after the merger is consistent with that before the merger; wherein, the direction of the parallel lines is perpendicular to the minimum span direction of the sub-convex polygons.

[0023] In one embodiment, S3 includes:

[0024] S31: Design a dynamic adaptive neighborhood radius model for planning the first path of the target interest region in the discrete grouping teaching algorithm; use greedy crossover, intermediate sequence, neighborhood mutation, neighborhood inversion and neighborhood shift operators to generate the initial path;

[0025] The neighborhood radius model is as follows: ; Let N be the radius of the neighborhood of the current region center t, and N be the total number of regions. This represents the maximum distance from other regions to the center t of the current region. This represents the minimum distance from other regions to the center t of the current region. Let be the average distance from other regions to the center t of the current region. The minimum distance between the centers of all regions. The average distance between the centers of all regions is given by , it is the current iteration number, and Maxit is the total number of iterations.

[0026] S32: Optimize the initial path using a multi-threaded fast optimization method to merge the initial path after neighborhood improvement, thereby obtaining the target path.

[0027] In one embodiment, S31 includes:

[0028] S311: Set the initialization parameters: total number of iterations and number of students in the group teaching optimization algorithm; generate the first path based on the dynamic adaptive neighborhood radius model; determine the individual corresponding to the shortest total path based on the first path. ;

[0029] S312: Judging individuals If the termination condition is met, the corresponding path planning sequence is output as the initial path; if not, the top 50% and bottom 50% of individuals are divided into two groups based on the total path length of each individual, forming an excellent group and a general group, and S313 is executed.

[0030] S313: For excellent groups, in the main thread, first generate the average level sequence within the group based on the individuals, then cross the average individuals with the individuals within the group based on the greedy crossover operator, and finally perform neighborhood mutation, neighborhood reversal and neighborhood shift processing on the individuals in sequence.

[0031] S314: For a general group, in the sub-thread, the greedy crossover operator is first used to cross the shortest path individual with the group individual, then the individual is processed according to the neighborhood mutation, neighborhood reversal and neighborhood shift operators, and finally the group individual is processed using the neighborhood 3-opt operator.

[0032] S315: After completing S313 and S314, merge the new individual sequences and obtain the individual corresponding to the shortest total path based on the total path of the new individual sequence after merging. ;

[0033] S316: For new individuals Execute S312-S315 until the termination condition is met.

[0034] In one embodiment, S32 includes:

[0035] The excellent team is run on the main thread and the general team is run on a sub-thread to optimize the initial path and finally obtain the target path.

[0036] According to another aspect of the present invention, an underwater region of interest coverage path planning device for an unmanned aerial vehicle is provided, comprising:

[0037] The boundary extraction module is used to divide the terrain area information map with known prior information to extract multiple regions of interest within the user's attention depth range; to cluster the multiple regions of interest to obtain clustering results; and to generate polygonal boundaries of the regions of interest based on the clustering results.

[0038] The region segmentation module is used to segment the polygon boundary of the region of interest based on concave points using an improved Bayazit algorithm to obtain multiple sub-convex polygons; and to merge every two sub-convex polygons that have a common edge and whose internal parallel lines are in the same direction to obtain the target region of interest.

[0039] The region connection module is used to plan the initial path of the target region of interest using a discrete grouping teaching algorithm; and to optimize the initial path using a multi-threaded fast optimization method to obtain the target path.

[0040] According to another aspect of the present invention, an unmanned vehicle is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0041] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0042] (1) Existing concave polygon region segmentation fails to merge adjacent convex polygons based on the consistency of their parallel line directions, resulting in long planned coverage paths and numerous turning times. To address this, this invention proposes an improved Bayazit algorithm. This algorithm prioritizes concave points within the reflection arc range and combines vertex connection selection based on distance from the segmentation point to reduce the number of sub-regions. Furthermore, it merges sub-convex polygons with common edges and consistent parallel line directions to further reduce the number of sub-regions, thereby optimizing the coverage path length and reducing the number of turning times.

[0043] (2) To address the existing inter-regional path planning problem, this invention introduces a dynamic adaptive neighborhood radius model to optimize paths in the group teaching optimization algorithm. The neighborhood radius of this adaptive model gradually decreases with the increase of the iteration number, thus taking into account both global and local search capabilities. At the same time, the hyperbolic tangent function is used to balance the nonlinear relationship of distance, making the distribution of neighborhood radius with the distance of sub-regions smoother. This achieves faster convergence of the connection path between sub-regions and is closer to the true optimal value, providing a fast and accurate path scheme for underwater topographic area information map coverage.

[0044] (3) In view of the high real-time requirements of the embedded system of unmanned vehicle, the present invention divides the iteration process of the general group and the excellent group into two threads to run in parallel, thereby accelerating the algorithm running efficiency and providing an efficient and real-time path planning method for underwater terrain area information map coverage. Attached Figure Description

[0045] Figure 1 This is a flowchart of a method for planning the underwater terrain of interest area information map coverage path for an unmanned vehicle in one embodiment of the present invention;

[0046] Figure 2a , Figure 2b , Figure 2c and Figure 2d In one embodiment of the present invention, DBSCAN clustering and the α-shapes algorithm are used to generate boundary maps of regions of interest, wherein, Figure 2a This is a contour map. Figure 2b To create a rasterized contour map, Figure 2c For DBSCAN clustering graph, Figure 2d Generate maps for the boundaries of α-shapes;

[0047] Figure 3a This is a schematic diagram illustrating the concave polygon segmentation principle of the Bayazit algorithm in one embodiment of the present invention.

[0048] Figure 3b This is a schematic diagram illustrating the principle of concave polygon segmentation using the improved Bayazit algorithm in one embodiment of the present invention.

[0049] Figure 4a This is a schematic diagram of the segmentation result of the original Bayazit algorithm in one embodiment of the present invention;

[0050] Figure 4b This is a schematic diagram of the segmentation result of the improved Bayazit algorithm in one embodiment of the present invention;

[0051] Figure 4c This is a diagram showing the parallel line path planning result of the original Bayazit algorithm in one embodiment of the present invention;

[0052] Figure 4d This is a diagram showing the parallel line path planning result of the improved Bayazit algorithm in one embodiment of the present invention;

[0053] Figure 5a This is a schematic diagram of the region of interest segmentation result before the improvement of the Bayazit algorithm in one embodiment of the present invention;

[0054] Figure 5b This is a schematic diagram of the improved Bayazit algorithm segmentation result of the region of interest in one embodiment of the present invention;

[0055] Figure 6a for Figure 5b A schematic diagram of the results of parallel line planning;

[0056] Figure 6b for Figure 6a A schematic diagram of the parallel line planning results after merging;

[0057] Figure 7a This is a convergence graph of the dynamic adaptive neighborhood radius model of the discrete grouping teaching optimization algorithm of the present invention on a test case with random initialization;

[0058] Figure 7bThis is a convergence graph of the dynamic adaptive neighborhood radius model random mutation of the discrete grouping teaching optimization algorithm of the present invention on test cases;

[0059] Figure 8a , Figure 8b and Figure 8c The figures show the test results using the EIL51, EIL76, and EIL101 test sets from the TSPLIB library, respectively.

[0060] Figure 9a and Figure 9b This is a schematic diagram of the path planning result for a region of interest according to the present invention;

[0061] Figure 10 This is a schematic diagram of a scenario according to an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0063] like Figure 1 As shown, an underwater region of interest (ROI) coverage path planning method for an unmanned aerial vehicle (UAV) is provided, including: S1: dividing the terrain area information map with known prior information to extract multiple ROIs within the user's depth range of interest; clustering the multiple ROIs to obtain clustering results, and generating ROI polygon boundaries based on the clustering results; S2: using an improved Bayazit algorithm to segment the ROI polygon boundaries based on concave points to obtain multiple sub-convex polygons; merging every two sub-convex polygons with a common edge and consistent internal parallel line directions to obtain the target ROI; S3: using a discrete grouping teaching algorithm to plan the initial path of the target ROI; and using a multi-threaded fast optimization method to optimize the initial path to obtain the target path.

[0064] Specifically, S1: Based on the user's required resolution, the prior information terrain map is divided using a raster method; regions of interest (ROIs) within the user's depth of focus are extracted, and clustering can be performed using the DBSCAN algorithm (though this is not the only possible clustering algorithm); the α-shapes algorithm is used to generate polygonal boundaries for the ROIs. The raster method requires determining the length and width of each small cell, i.e., the resolution. This cell resolution is determined by the user's requirements; if meter-level accuracy is required, the resolution must be less than or equal to meter-level accuracy. The extracted ROIs are as follows: Figure 2c As shown, it belongs to a point set, while the "boundary of the region of interest polygon" is as follows: Figure 2d As shown, an outer boundary is generated. S2: The improved Bayazit algorithm is used to segment concave points in the region of interest, and sub-convex polygons with common edges and parallel lines in the same direction are merged. S3: The discrete group teaching algorithm is used to plan the paths between regions of interest. A discrete group teaching algorithm based on neighborhood initialization and neighborhood mutation crossover is proposed to accelerate the convergence speed of the algorithm. An application can have multiple threads. In order to make full use of the advantages of CPU multi-core, the excellent group and the general group are placed in different threads when running the group teaching optimization algorithm, which can run simultaneously and shorten the algorithm running time.

[0065] In one embodiment, S1 includes: S11: rasterizing the terrain area information map according to the resolution required by the user to obtain multiple area rasteres; S12: extracting multiple regions of interest from the multiple area rasteres according to the user's depth of interest range; using the DBSCAN algorithm to cluster the multiple regions of interest to obtain multiple clustering results; S13: using the α-shapes algorithm to generate the bounding polygons of the regions corresponding to each clustering result, and using them as the boundaries of the polygons of the regions of interest.

[0066] Specifically, 1) Topographic contour maps with prior information, such as... Figure 2a As shown, the area is 10km in both length and width. The terrain information map is rasterized according to the resolution required by the user, as follows: Figure 2b The raster resolution shown is 100m*100m. 2) Extract the region of interest (ROI) raster based on the user's depth of interest, and then use the DBSCAN algorithm to cluster the ROI raster region. Specifically, the radius is set to 40m and the number of points is set to 8. The clustering result after removing noise is shown below. Figure 2c 3) The α-shapes algorithm was used to generate the bounding polygon boundaries of each region of interest after classification. Specifically, the simulated rolling circle radius was set to 20m. The results are as follows: Figure 2d .

[0067] In one embodiment, S2 includes: S21: Based on the Bayazit concave polygon segmentation algorithm, improve the method of directly and greedily connecting the nearest vertex of the reflection arc. When a concave point exists, connect the nearest vertex within the reflection arc range that is a concave point. If no concave point exists, connect the nearest vertex, thereby achieving the segmentation of the polygon boundaries corresponding to each region of interest and obtaining multiple sub-convex polygons; S22: Merge the sub-convex polygons with common edges and parallel lines in the same direction to obtain the target region of interest, thereby reducing the coverage path length and the number of turns.

[0068] In S1, the improved Bayazit algorithm refers to: improving the method of directly and greedily connecting the nearest vertex of the reflection arc, selecting the closest vertex within the range of the segmented point and the concave point of the reflection arc, and if no concave point exists, connecting the nearest vertex, effectively reducing the number of concave points while maintaining running efficiency. S22: merging sub-convex polygons with common edges and parallel lines in the same direction, reducing the length of the covered path and the number of turns.

[0069] In one embodiment, S21 includes: randomly selecting a vertex of a concave polygon formed by the boundary of the polygon to be decomposed for concave point determination, and then determining concave points vertex by vertex in a counterclockwise direction; when the first concave point P is found... i Then, along P i-1 P i and P i P i+1 Extend the lines in the opposite direction and intersect them at the polygon boundary. If there are multiple vertices within the intersection range of the reverse extension line and the polygon boundary, select the nearest point with the concave attribute within the range and connect it. If there are no vertices within the intersection range of the reflection line and the polygon, select the auxiliary point connected by the center point of the two intersection points to remove the concave point. Finally, multiple sub-convex polygons are obtained.

[0070] Specifically, randomly select a vertex of the concave polygon to be decomposed, and then check whether the point is concave in a counterclockwise direction. The formula is as follows:

[0071]

[0072] Where i represents the vertex number, N represents the total number of points in the polygon, mod represents the modulo operation, and P i Let i represent the i-th vertex. After finding the first concave vertex i, proceed along point P from i. i-1 P i and P i P i+1 Extend them in opposite directions and intersect at the polygon boundary.

[0073] The concave points need to be segmented as follows Draw its reflection line. If there are multiple vertices within the range where the reflection line intersects with the polygon, select the nearest point within that range that has the concave attribute (using...). Figure 3a For example, the concave points need to be segmented as follows: The range is Select the concave point that is closest to the point. );

[0074] If the intersection of the reflected ray and the polygon does not contain any vertices, then connect the center points of the two intersection points with auxiliary points to remove the concave points. Figure 3b For example, the concave points need to be segmented as follows: If there are no vertices within the intersection range of the reflected ray and the polygon, then select the midpoint of the two intersection points, i.e., the solid circle in the figure); after completing the segmentation of the concave point, the original concave polygon is divided into two parts, and then the above steps of S22 are repeated for the two parts respectively until the segmented sub-polygon has no concave point. This process is a recursive call.

[0075] Specifically, to further highlight the effectiveness of the algorithm in handling cases where there are multiple vertices within the intersection range of two reflection lines, the improved Bayazit segmentation algorithm proposed in this invention only requires one segmentation (e.g., Figure 4b As shown, the concave points that need to be segmented are... The range is Select the concave point that is closest to the point. The original Bayazit segmentation algorithm uses direct distance greedily to select connected vertices. With reflection point (i.e., concave points) cause concave points to remain after segmentation. It needs to be divided twice, such as Figure 4a As shown.

[0076] In one embodiment, S22 includes: if adjacent sub-convex polygons have a common edge and the planned parallel lines within the region have the same direction, then merge the two sub-convex polygons and make the parallel line direction in the target interest region obtained after merging consistent with that before merging; wherein the parallel line direction is perpendicular to the minimum span direction of the sub-convex polygon.

[0077] Specifically, based on the segmented sub-convex polygons, if adjacent sub-convex polygons share a common edge and the planned parallel lines within their regions are aligned (where the parallel lines are perpendicular to the minimum span direction of the sub-convex polygon), then the two sub-convex polygons are merged, and the parallel line directions remain consistent with those before the merge. For example... Figure 4c and 4d The figures show the parallel line path planning results before and after merging, respectively. It is a concave point. For example... Figure 4c As shown, if they are not merged, regional planning needs to be carried out separately for ① and ②, while after merging, they can be planned as a whole as follows. Figure 4d As shown, the planning results can reduce some paths and reduce two roundabout processes.

[0078] To further demonstrate the effects of this embodiment, regarding... Figure 2d The shown nautical chart's region of interest scene was processed using both the improved and unimproved Bayazit segmentation algorithms. Figure 5a and 5b Taking the dashed box shown as an example, the improved Bayazit algorithm reduces the number of polygons segmented from four convex polygons after the original Bayazit algorithm to three, and the shapes are more reasonable, which is more conducive to the planning of parallel line paths in the subsequent region.

[0079] It should be noted that the design within the region uses parallel lines with equal search line spacing, and the search line direction is perpendicular to the minimum span direction of the convex polygon. The relevant theory has been proven and derived in detail in "An Algorithm for UAV Coverage Track Planning in Convex Polygon Regions," and will not be elaborated upon here. Figure 5b The results of parallel line planning are as follows Figure 6a As shown, the search lines are spaced 600 meters apart, and the starting line is 300 meters from the boundary. The planning results indicate that... Figure 6a The area shown in the dashed box contains two adjacent areas ( Figure 6a Regions ① and ② within the dashed box share a common edge and their parallel lines are in the same direction. Therefore, they are merged. The result of the parallel line planning after merging is as follows. Figure 6b As shown.

[0080] In one embodiment, S3 includes: S31: Designing a dynamic adaptive neighborhood radius model for planning the first path of the target interest region in a discrete grouping teaching algorithm; using greedy crossover, intermediate sequence, neighborhood mutation, neighborhood inversion, and neighborhood shift operators to generate the initial path; the neighborhood radius model is: ; Let N be the radius of the neighborhood of the current region center t, and N be the total number of regions. This represents the maximum distance from other regions to the center t of the current region. This represents the minimum distance from other regions to the center t of the current region. Let be the average distance from other regions to the center t of the current region. The minimum distance between the centers of all regions. S32: Optimize the initial path using a multi-threaded fast optimization method, merge the improved initial path from the neighborhood, and thus obtain the target path.

[0081] To further explain greedy crossover, intermediate sequence, neighborhood mutation, neighborhood inversion, and neighborhood shift operators, combined with Figure 7a This will be elaborated further.

[0082] Furthermore, for the greedy crossover operator, firstly, two random numbers are randomly selected based on the length of the individual sequence, such as... Figure 7a(Greedy crossover) (e.g., positions 3 and 5). Then, compare the distances between the two individual sequences at positions 3 and 5. Select the sequence with the smallest distance (e.g., element 3->7->1) and add it to the individual sequence between positions 3 and 5. Furthermore, elements that are repeated between positions 3 and 5 in the individual sequence are removed (e.g., element 1). Finally, the unused portion is added to the sequence in a greedy manner, forming the sequence 2->5->6->3->7->1->4.

[0083] Furthermore, for the intermediate sequence, the intermediate sequence is obtained based on the principle of having the most common elements under the same position number. For n cities and m individuals, the following model can be established:

[0084]

[0085] in, It is the i-th individual. This refers to the nth position of the i-th individual. For example... Figure 7a The (intermediate sequence) shows the positions of the intermediate elements. The resulting sequence is processed sequentially column-wise. First, all used elements at each position are deleted. Then, based on the frequency statistics of the remaining elements, the element with the highest frequency is selected to fill the corresponding position in the output sequence, such as... Figure 7a As shown in the intermediate sequence (e.g., positions 2, 3, 4, 5, 7), however, if more than one element has the same highest frequency, one is randomly selected for output (e.g., element 2 is selected for position 1). If all elements at this position are removed (e.g., elements 1 and 5 are removed from position 6), then the remaining unused elements are randomly selected (e.g., element 6) to fill the sequence, forming the sequence 2->5->3->7->1->6->4.

[0086] Furthermore, for the neighborhood mutation operator, a different random number is randomly selected based on the length of the individual sequence, such as... Figure 7a (Neighborhood mutation) (e.g., element 1 corresponding to position 6), select an element from the neighborhood (e.g., element 5 corresponding to position 2). Finally, swap the element at the position corresponding to the random number with the element at the corresponding position in the neighborhood of that element (e.g., swap elements 1 and 5 corresponding to positions 2 and 6), forming the sequence 2->1->6->3->7->5->4.

[0087] Furthermore, for the neighborhood inversion operator, a different random number is randomly selected based on the length of the individual sequence, such as... Figure 7a(Neighborhood inversion) (e.g., element 1 corresponding to position 2), select an element from the neighborhood range (e.g., element 5 corresponding to position 6). Finally, invert the elements at positions 3 to 6 to form the sequence 2->1->5->7->3->6->4.

[0088] Furthermore, for the neighborhood mutation operator, a different random number is randomly selected based on the length of the individual sequence, such as... Figure 7a (Neighborhood mutation) (e.g., element 1 corresponding to position 2), select an element from the neighborhood (e.g., element 3 corresponding to position 5). Finally, move element 5 after position 2, forming the sequence 2->1->3->5->7->6->4.

[0089] In one embodiment, S31 includes: S311: setting the total number of iterations and the number of students in the group teaching optimization algorithm as initialization parameters; generating a first path based on the dynamic adaptive neighborhood radius model; and determining the individual corresponding to the shortest total path based on the first path. This group-based teaching optimization algorithm primarily addresses the problem of planning connected paths between multiple regions of interest. Therefore, "individual" refers to... Figure 7a The order in which the multiple regions of interest are traversed (or queued) determines how the unmanned aerial vehicle traverses each region of interest in sequence.

[0090] S312: Judging individuals If the termination condition is met, output the corresponding path planning sequence as the initial path; otherwise, based on the total path length of each individual, divide the top 50% and bottom 50% into two groups to form an excellent group and a general group, and execute S313.

[0091] S313: For excellent groups, in the main thread, first generate the average level sequence within the group based on the individuals, then cross the average individuals with the individuals within the group based on the greedy crossover operator, and finally perform neighborhood mutation, neighborhood reversal and neighborhood shift processing on the individuals in sequence.

[0092] S314: For a general group, in the sub-thread, the greedy crossover operator is first used to cross the shortest path individual with the group individual, then the individual is processed according to the neighborhood mutation, neighborhood reversal and neighborhood shift operators, and finally the group individual is processed using the neighborhood 3-opt operator.

[0093] S315: After completing S313 and S314, merge the new individual sequences and obtain the individual corresponding to the shortest total path based on the total path of the new individual sequence after merging. ;

[0094] S316: For new individuals Execute S312-S315 until the termination condition is met.

[0095] The entire dynamic adaptive neighborhood radius model is used in the discrete grouping teaching algorithm process, as follows: Figure 7b As shown.

[0096] 1) First, set the initialization parameters: total number of iterations and number of individuals. Generate an initial path sequence based on the dynamic adaptive neighborhood radius model, and calculate the individual corresponding to the shortest total path based on the sequence. ;

[0097] 2) Determine if the termination condition is met. If not, divide the top 50% and bottom 50% of each individual into two groups to form an excellent group and a general group based on the total path of each individual. If the condition is met, output the optimal sequence and the corresponding total path value.

[0098] 3) For outstanding teams, the main thread first determines... Figure 7a Intermediate individuals generate a sequence of average group levels, and then based on... Figure 7a The greedy crossover operator crosses the average individual with the individuals in the group, and finally performs neighborhood mutation, neighborhood reversal and neighborhood shift processing on the individuals in sequence.

[0099] 4) For general groups, in the sub-thread, the greedy crossover operator is first used to cross the shortest path individual with the individuals within the group, and then according to... Figure 7a Individuals are processed using neighborhood mutation, neighborhood reversal, and neighborhood shift operators, and finally, individuals within the group are processed using the neighborhood 3-opt operator.

[0100] 5) After completion, merge the generated new individual sequences, and obtain the individual corresponding to the shortest total path based on the total path of each merged individual sequence. Then return to part 2) to execute.

[0101] Specifically, to fully compare the dynamic adaptive neighborhood radius model used in discrete grouping teaching algorithms with conventional random initialization, random mutation, random inversion, and random shift operators, the EIL51, EIL76, and EIL101 test sets from the TSPLIB library were used for testing. The test results are as follows: Figure 8a , Figure 8b and Figure 8c As shown, the total number of iterations of the algorithm is set to 1000 generations, and the total number of individuals is set to 100.

[0102] The formula for calculating Relative Error is as follows: In the formula, O represents the optimal solution on the test set, corresponding to 428.87, 545.38, and 642.31 on the EIL51, EIL76, and EIL101 test sets, respectively (the above cases correspond to 51, 76, and 101 cities, respectively), and R is the sum of city distances obtained by the grouping optimization algorithm. From Figure 8a , 8b As can be seen from 8c, the dynamic adaptive neighborhood radius model proposed in this invention has a relatively large advantage in initial values ​​for discrete group teaching algorithms, and as the number of cities increases, the final optimization result of this algorithm has a greater advantage than that of random initialization.

[0103] Specifically, in order to further apply this algorithm to inter-regional path planning, therefore, for example... Figure 6b The segmented region of interest shown in the figure is obtained as follows: Figure 9a The planned path is shown, where the * marks the center of the region. As can be seen from the path planning results (mainly within the black dashed box), the algorithm of this invention is a global optimization, without a direct distance-greedy design. Furthermore, the design automatically returns to the starting point after the unmanned surface vessel completes its coverage task; the starting point can be any region. Traversal within the region is performed according to the parallel line planning results. The region's entrance and exit are selected using a single convex polygon parallel line planning start and end point (e.g., ...). Figure 9b As shown, the entrance is marked with an asterisk (*) and the exit is marked with a solid circle. The starting and ending lines of the parallel lines within a single region each intersect the region at two points. The starting point of the region is selected based on the distance-greedy method of the previous polygon endpoint. The order of the polygon regions is output by the discrete grouping teaching algorithm. Once the starting point is determined, the endpoint is also determined.

[0104] In one embodiment, S32 includes: running the excellent group on the main thread and the general group on a child thread to optimize the initial path and finally obtain the target path.

[0105] The multi-threaded fast optimization method provided in this invention includes: running the excellent group and the general group in different threads, performing their respective optimizations simultaneously, and then merging them back together for subsequent grouping to continue the subsequent iteration process.

[0106] Specifically, to further highlight the invention's proposed approach of placing outstanding teams and general teams on the main thread and child thread respectively, and the direct single-threaded mode, the following steps are taken: Figure 8a , Figure 8b and Figure 8cThe examples were run on dual-threaded and single-threaded systems, respectively. The implementation environment was MATLAB 2020b with an Intel(R) Core(TM) i9-10850K CPU @ 3.60GHz. The overall running time to obtain the optimal inter-region path planning result was 0.5620s and 0.7550s, respectively. As can be seen from the running time, the proposed mode of placing the excellent group and the general group on the main thread and the child thread, respectively, can shorten the solution time compared with the single-threaded mode.

[0107] According to another aspect of the present invention, an underwater region of interest coverage path planning device for an unmanned aerial vehicle is provided, such as... Figure 10 As shown, it includes:

[0108] The boundary extraction module is used to divide the terrain area information map with known prior information to extract multiple regions of interest within the user's depth range of interest; the DBSCAN algorithm is used to cluster the multiple regions of interest to obtain clustering results, and polygonal boundaries of the regions of interest are generated based on the clustering results;

[0109] The region segmentation module is used to segment the polygon boundary of the region of interest based on concave points using an improved Bayazit algorithm to obtain multiple sub-convex polygons; every two sub-convex polygons with a common edge and internal parallel lines in the same direction are merged to obtain the target region of interest.

[0110] The region connection module is used to plan the initial path of the target region of interest using a discrete grouping teaching algorithm; and to optimize the initial path using a multi-threaded fast optimization method to obtain the target path.

[0111] According to another aspect of the present invention, an unmanned vehicle is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0112] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for underwater region of interest coverage path planning for an unmanned aerial vehicle, characterized in that, include: S1: Divide the terrain area information map with known prior information to extract multiple regions of interest within the user's attention depth range; Clustering is performed on the multiple regions of interest to obtain clustering results, and polygonal boundaries of the regions of interest are generated based on the clustering results; S2: The improved Bayazit algorithm is used to segment the polygon boundary of the region of interest based on concave points to obtain multiple sub-convex polygons; every two sub-convex polygons with a common edge and internal parallel lines in the same direction are merged to obtain the target region of interest. S3: Use the discrete grouping teaching algorithm to plan the initial path of the target region of interest; The initial path is optimized using a multi-threaded fast optimization method to obtain the target path; S3 includes: S31: Design a dynamic adaptive neighborhood radius model for planning the first path of the target region of interest in a discrete grouping teaching algorithm; The initial path is generated using greedy crossover, intermediate sequence, neighborhood mutation, neighborhood inversion, and neighborhood shift operators; the neighborhood radius model is as follows: ; Let N be the radius of the neighborhood of the current region center t, and N be the total number of regions. This represents the maximum distance from other regions to the center t of the current region. This represents the minimum distance from other regions to the center t of the current region. Let be the average distance from other regions to the center t of the current region. The minimum distance between the centers of all regions. The average distance between the centers of all regions is given by , it is the current iteration number, and Maxit is the total number of iterations. S32: Optimize the initial path using a multi-threaded fast optimization method to merge the initial path after neighborhood improvement, thereby obtaining the target path.

2. The underwater region of interest coverage path planning method for unmanned aerial vehicles as described in claim 1, characterized in that, S1 includes: S11: Rasterize the terrain area information map according to the user's actual required resolution to obtain multiple area rasteres; S12: Extract multiple regions of interest from the multiple region grids according to the user's depth range of interest; use the DBSCAN algorithm to cluster the multiple regions of interest to obtain multiple clustering results; S13: Use the α-shapes algorithm to generate the bounding polygons of the regions corresponding to each clustering result, and use them as the boundaries of the polygons of the regions of interest.

3. The underwater region of interest coverage path planning method for unmanned aerial vehicles as described in claim 1, characterized in that, S2 includes: S21: Based on the Bayazit concave polygon segmentation algorithm, improve the method of directly and greedily connecting the nearest vertex of the reflection arc. When a concave point exists, connect the nearest vertex within the reflection arc range that is a concave point. If no concave point exists, connect the nearest vertex, thereby achieving the segmentation of the polygon boundaries corresponding to each region of interest and obtaining multiple sub-convex polygons; S22: Merge the sub-convex polygons with common edges and parallel lines in the same direction to obtain the target region of interest, thereby reducing the coverage path length and the number of turns.

4. The underwater region of interest coverage path planning method for unmanned aerial vehicles as described in claim 3, characterized in that, S21 includes: Randomly select a vertex of the concave polygon formed by the boundary of the polygon to be decomposed for concavity point detection, and then check each vertex in a counterclockwise direction for concavity point detection; when the first concavity point is found... P i Afterwards, along P i-1 P i and P i P i+1 Extend them in opposite directions and intersect at the polygon boundary; If there are multiple vertices within the range of the intersection of the reverse extension line and the polygon boundary, then select the nearest point within the range that has the concave attribute and is closest to it; If there are no vertices within the intersection range of the reflected ray and the polygon, then connect the center points of the two intersection points with auxiliary points to remove the concave points; finally, multiple sub-convex polygons are obtained.

5. The underwater region of interest coverage path planning method for unmanned aerial vehicles as described in claim 4, characterized in that, S22 includes: If adjacent sub-convex polygons share a common edge and the parallel lines planned within the region are in the same direction, then the two sub-convex polygons are merged, and the direction of the parallel lines in the target interest region obtained after the merger is consistent with that before the merger; wherein, the direction of the parallel lines is perpendicular to the minimum span direction of the sub-convex polygons.

6. The underwater region of interest coverage path planning method for an unmanned aerial vehicle as described in claim 1, characterized in that, S31 includes: S311: Set the initialization parameters: total number of iterations and number of students in the group teaching optimization algorithm; generate the first path based on the dynamic adaptive neighborhood radius model; determine the individual corresponding to the shortest total path based on the first path. ; S312: Judging individuals If the termination condition is met, the corresponding path planning sequence is output as the initial path; if not, the top 50% and bottom 50% of individuals are divided into two groups based on the total path length of each individual, forming an excellent group and a general group, and S313 is executed. S313: For excellent groups, in the main thread, first generate the average level sequence within the group based on the individuals, then cross the average individuals with the individuals within the group based on the greedy crossover operator, and finally perform neighborhood mutation, neighborhood reversal and neighborhood shift processing on the individuals in sequence. S314: For a general group, in the sub-thread, the greedy crossover operator is first used to cross the shortest path individual with the group individual, then the individual is processed according to the neighborhood mutation, neighborhood reversal and neighborhood shift operators, and finally the group individual is processed using the neighborhood 3-opt operator. S315: After completing S313 and S314, merge the new individual sequences and obtain the individual corresponding to the shortest total path based on the total path of the new individual sequence after merging. ; S316: For new individuals Execute S312-S315 until the termination condition is met.

7. The underwater region of interest coverage path planning method for an unmanned aerial vehicle as described in claim 6, characterized in that, S32 includes: The excellent team is run on the main thread and the general team is run on a sub-thread to optimize the initial path and finally obtain the target path.

8. An underwater region of interest coverage path planning device for an unmanned aerial vehicle, characterized in that, The underwater region of interest coverage path planning method for executing any one of claims 1-7 of the unmanned aerial vehicle includes: The boundary extraction module is used to divide the terrain area information map with known prior information to extract multiple regions of interest within the user's attention depth range; to cluster the multiple regions of interest to obtain clustering results; and to generate polygonal boundaries of the regions of interest based on the clustering results. The region segmentation module is used to segment the polygon boundary of the region of interest based on concave points using an improved Bayazit algorithm to obtain multiple sub-convex polygons; and to merge every two sub-convex polygons that have a common edge and whose internal parallel lines are in the same direction to obtain the target region of interest. The region connection module is used to plan the initial path of the target region of interest using a discrete grouping teaching algorithm; and to optimize the initial path using a multi-threaded fast optimization method to obtain the target path.

9. An unmanned aerial vehicle, comprising a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method according to any one of claims 1 to 7.

Images