Multi-unmanned aerial vehicle cooperative detection method for urban approaching operation

Abstract

In order to solve the technical problems of low reconnaissance efficiency, insufficient target detection precision, limited cooperative capability and poor environmental adaptability of an existing multi-unmanned aerial vehicle cooperative detection method, the invention provides a multi-unmanned aerial vehicle cooperative detection method for urban approaching operation. Through the technical means of detection area division, path planning optimization, multi-modal target detection, dynamic task scheduling and the like, the problems of low detection area coverage efficiency, insufficient target detection precision, limited cooperative capability, poor environmental adaptability and the like in the prior art are solved, and efficient, accurate and robust detection tasks in a complex urban environment are realized. And during path planning, an improved A star algorithm and an improved genetic algorithm are adopted, so that the running speed and the planning precision can be improved.

Description

Technical Field

[0001] The present invention relates to the field of measurement and testing technology, and in particular to a multi-UAV collaborative detection method. Background Art

[0002] With the rapid development of drone technology, drones are increasingly being used in public safety, emergency rescue, and other fields. In urban environments, in particular, drone reconnaissance capabilities have become a key technology for solving complex tasks. Urban environments are characterized by high levels of complexity and uncertainty, such as densely populated buildings, narrow streets, and numerous obstacles. Traditional single-drone detection methods struggle to meet the requirements for efficient and comprehensive detection. Therefore, multi-drone collaborative detection methods have emerged.

[0003] The existing multi-UAV collaborative detection methods have the following main problems:

[0004] 1) Low detection efficiency;

[0005] In complex urban environments, it is difficult for a single drone to quickly cover a large area, especially in areas with dense buildings. The drone needs to frequently adjust its flight path, resulting in low detection efficiency.

[0006] 2) Insufficient target detection accuracy;

[0007] There are a large number of camouflaged and hidden targets in urban environments, and traditional detection methods are difficult to achieve accurate detection.

[0008] 3) Limited collaborative capabilities;

[0009] The coordination capabilities between multiple drones are insufficient, and mission areas and detection resources cannot be effectively allocated, resulting in resource waste and mission conflicts.

[0010] 4) Poor environmental adaptability;

[0011] Dynamic changes in urban environments (such as traffic, weather, obstacles, etc.) place higher demands on the autonomous decision-making capabilities of drones, and existing drone detection methods are difficult to meet these requirements. Summary of the Invention

[0012] In order to solve the technical problems of low reconnaissance efficiency, insufficient target detection accuracy, limited coordination capability and poor environmental adaptability of existing multi-UAV collaborative detection methods, the present invention proposes a multi-UAV collaborative detection method for urban approach operations.

[0013] The technical solution adopted by the present invention to solve the above technical problems is:

[0014] A multi-UAV collaborative detection method for urban approach operations is unique in that it includes the following steps:

[0015] Step 1: Determine the detection area and its boundaries based on the urban terrain and detection mission requirements;

[0016] Step 2: According to the detection mission requirements, set the payload type and parameters of the drone so that the drone's detection area can cover the detection area; determine the shape of the detection area. If it is a convex quadrilateral, proceed to step 3A; if it is a concave quadrilateral, proceed to step 3B; if it is a complex polygon, proceed to step 3C;

[0017] Step 3A: Divide the detection area into the photoelectric payload and spectral payload traversal areas, and the ratio of the two areas is equal to the ratio of the detection capabilities of the photoelectric payload and spectral payload UAVs;

[0018] Step 4A: Generate scan lines within the traversal area of the photoelectric payload and the spectral payload respectively, and distribute them as evenly as possible to each photoelectric payload and each spectral payload UAV. The sub-area that each UAV needs to traverse is the sub-area covered by its assigned scan line.

[0019] Step 5A: Use the improved A* algorithm to plan the UAV’s path from the takeoff assembly point to the detection area;

[0020] Step 6A: Use the improved genetic algorithm to plan the UAV’s path within the detection area;

[0021] Step 7A: Use the improved A* algorithm to plan the drone's path from the detection area to the recovery area. The process ends.

[0022] Step 3B: Draw a segmentation line from the concave point of the concave quadrilateral. Scan the segmentation line with the concave point as the axis to decompose the detection area into n convex polygonal sub-areas of equal / similar areas; n is equal to the total number of drones;

[0023] Step 4B: Generate a scan line for the detection area using a reference edge translation scanning method; the reference edge is the convex edge in the detection area closest to the drone takeoff assembly point;

[0024] Step 5B: Use the improved A* algorithm to plan the UAV’s path from the takeoff assembly point to the detection area;

[0025] Step 6B: Use the improved genetic algorithm to plan the UAV’s path within the detection area;

[0026] Step 7B: Use the improved A* algorithm to plan the drone's path from the detection area to the recovery area. The process ends.

[0027] Step 3C: Decompose the detection area into multiple convex polygonal sub-areas using a convex decomposition method;

[0028] Step 4C: Scan line generation;

[0029] Step 4C.1: Generate scan lines in each sub-region using the reference edge translation scanning method, and merge adjacent sub-regions with the same scan line direction;

[0030] Step 4C.2: Calculate the scan lines allocated to the EO / spectral payload UAVs based on the ratio of the number of EO / spectral payload UAVs to the total number of scan lines. From the merged sub-areas, select sub-areas with a number of scan lines close to the number of scan lines allocated to the EO / spectral payload UAVs as the EO / spectral payload mission areas. The remaining sub-areas are designated as spectral / EO payload sub-areas.

[0031] Step 4C.3: If the photoelectric payload task area is a convex quadrilateral, directly use the reference edge translation scanning method to regenerate the scan lines. Otherwise, use the convex decomposition method to decompose it into multiple sub-areas, and then use the reference edge translation scanning method to regenerate the scan lines for each sub-area. If the spectral payload task area is a convex quadrilateral, directly use the reference edge translation scanning method to regenerate the scan lines. Otherwise, use the convex decomposition method to decompose it into multiple sub-areas, and then use the reference edge translation scanning method to regenerate the scan lines.

[0032] The reference edge is the convex edge in each sub-area that is closest to the drone take-off assembly point;

[0033] Step 5C: Use the improved A* algorithm to plan the UAV's path from the takeoff assembly point to the detection area;

[0034] Step 6C: Use the improved genetic algorithm to plan the UAV's path within the detection area;

[0035] Step 7C: Use the improved A* algorithm to plan the drone's path from the detection area to the recovery area. The process ends.

[0036] The heuristic function of the improved A* algorithm is a dynamic weighted heuristic function:

[0037] f(x,y)=g(x,y)+(w+p)h(x,y)

[0038] Where g(n) and h(n) are the actual cost function and the estimated cost function, respectively; p is the offset, p = 0.001; w is the weight coefficient, when h(n) < 18, w = 0.8; when h(n) ≥ 18, w = 3;

[0039] The improved genetic algorithm is based on the existing genetic algorithm, uses a grid potential evaluation method to optimize the initial population, and adds a cubic spline interpolation function to optimize the path generated by the genetic algorithm so that the genetic algorithm finally outputs a smooth path.

[0040] Furthermore, in step 4A, scan lines are assigned to the drone according to the following principles:

[0041] Principle ①: The difference in the number of spectral payload scan lines assigned to any two spectral payload drones should be as small as possible, and the difference in the number of photoelectric payload scan lines assigned to any two photoelectric payload drones should be as small as possible;

[0042] Principle ②: Ensure that as many photoelectric payload drones as possible are assigned the same number of photoelectric payload scan lines with the same parity, and as many spectral payload drones as possible are assigned the same number of spectral payload scan lines with the same parity;

[0043] Principle ③: Make sure that as many photoelectric payload drones as possible are assigned an even number of photoelectric payload scan lines, and as many spectral payload drones as possible are assigned an even number of spectral payload scan lines;

[0044] The priority of principle ①>the priority of principle ②>the priority of principle ③.

[0045] Furthermore, the methods of steps 5A and 5B are the same, specifically:

[0046] Step 5.1: The endpoints of the scan lines on both sides of each sub-area that are closest to the drone take-off assembly point are used as the scanning starting points of each sub-area. The endpoints of each sub-area that are closer to the scanning starting point are selected as the entry points of each sub-area.

[0047] Step 5.2: Use the improved A* algorithm to generate the path from the drone takeoff assembly point to the entry point of each sub-area.

[0048] Furthermore, step 5C is specifically as follows:

[0049] Step 5C.1: From the endpoints of the scan lines within the detection area, select the endpoint closest to the drone takeoff assembly point O as the entry point D s ; When there are multiple nearest endpoints, select one at random;

[0050] Step 5C.2: Plan the path of each UAV from the takeoff assembly point to the entry point of the detection area using the improved A* algorithm.

[0051] Furthermore, step 6A is specifically as follows:

[0052] Step 6A.1: Connect the scan lines generated in step 4A end to end to form an initial path. The endpoints of the scan lines are the waypoints on the initial path.

[0053] Step 6A.2: Calculate the flight strip spacing of the UAV carrying the spectral payload and the UAV carrying the electro-optical payload during scanning respectively;

[0054] Step 6A.3: Select a turning strategy;

[0055] For fixed-wing drones, choose a turning strategy based on the following two situations:

[0056] Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight lanes through circular turns (i.e., U-turn strategy or η-turn strategy).

[0057] Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips;

[0058] For quadrotor drones, when the turn is a right angle + right angle, the minimum turning radius is not considered and the drone can turn directly. When the turn is an acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to a right angle + right angle turn.

[0059] Step 6A.4: Determine the return point of each drone j in its assigned sub-area j

[0060] Step 6A.5: Calculate the operating time t of drone j scanning sub-area j j ;

[0061]

[0062] Where n j The number of flight strips that the j-th UAV covers and scans the j-th sub-area, n j -1 is the number of turns required for the drone to cover the jth area, d 航间1 with d 航间2 are the flight path spacing of the UAV equipped with optoelectronic payload and spectral payload, v j is the level flight speed of the UAV in the jth sub-area, L j is the average length of a single scan line in the jth sub-region;

[0063] Step 6A.6: Optimize the waypoints on the initial path generated in step 6.1 using the improved genetic algorithm;

[0064] The obstacle information, the entry point selected in step 5.1, the scanning direction and initial flight direction, the initial path and waypoints generated in step 6.1, the flight path spacing calculated in step 6.2, the minimum turning radius of the UAV, the turning strategy in step 6.3, the return point determined in step 6.4, and the working time t calculated in step 6.5 are recorded. jThe improved genetic algorithm takes the working time as the target and the obstacle information as the constraint condition, optimizes the waypoints determined in step 6.1, and outputs the working time t j The smallest optimal path.

[0065] Furthermore, step 6B is specifically as follows:

[0066] Step 6B.1: Connect the scan lines generated in step 4B end to end to form an initial path;

[0067] Step 6B.2: Calculate the flight strip spacing between the UAV carrying the spectral payload and the UAV carrying the electro-optical payload during scanning.

[0068] Step 6B.3: Determine the waypoints on the initial path;

[0069] The remaining three edges of the detection area, except the reference edge, are shifted inward by a distance k. The intersection of the indented boundary line and the scan lines generated in step 4 is the waypoint position. L is the smaller value of L1 and L2, L1 and L2 are the projection widths of the photoelectric load and the spectral load on the ground respectively;

[0070] Step 6B.4: Assign the waypoints determined in step 6B.3 to each UAV;

[0071] Starting from the scan start point, calculate the arrival time of each waypoint, and calculate the duration t allocated to each drone based on the arrival time of the last waypoint. Compare the arrival time of each waypoint with the duration t allocated to each drone one by one, and assign each waypoint to the corresponding drone to ensure that the working time of each drone is as similar as possible.

[0072] Step 6B.5: Select a turning strategy;

[0073] For fixed-wing drones, choose a turning strategy based on the following two situations:

[0074] Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight strips through circular turns (i.e., U-turn strategy or η-turn strategy);

[0075] Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips;

[0076] For quadrotor drones, when the turn is a right angle + right angle, the minimum turning radius is not considered and the drone can turn directly. When the turn is an acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to a right angle + right angle turn.

[0077] Step 6B.6: Optimize the waypoints assigned to each UAV using the improved genetic algorithm;

[0078] The obstacle information, the entry point selected in step 5, the scanning direction and the initial flight direction, the initial path generated in step 6B.1, the flight strip spacing calculated in step 6B.2, the minimum turning radius of the UAV, the turning strategy in step 6B.5, the waypoints assigned to each UAV in step 6B.4, and the duration t of each UAV calculated in step 6B.4 are input into the improved genetic algorithm. The improved genetic algorithm optimizes the waypoints assigned to each UAV with duration t as the target and obstacle information as the constraint condition, and outputs the optimal waypoint allocation set with the minimum duration t. These optimal waypoint allocation sets constitute the optimal path.

[0079] Furthermore, step 6C is specifically as follows:

[0080] Step 6C.1: Connect the scan lines generated in step 4C end to end to form an initial path. The endpoints of each scan line are waypoints on the initial path.

[0081] Step 6C.2: Calculate the flight strip spacing between the UAV carrying the spectral payload and the UAV carrying the electro-optical payload during scanning respectively;

[0082] Step 6C.3: Select a turning strategy;

[0083] For fixed-wing drones, choose a turning strategy based on the following two situations:

[0084] Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight strips through circular turns (i.e., U-turn strategy or η-turn strategy);

[0085] Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips;

[0086] For quadrotor drones, when the turn is a right angle + right angle, the minimum turning radius is not considered and the drone can turn directly. When the turn is an acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to a right angle + right angle turn.

[0087] Step 6C.4: Calculate the route set for each UAV and the time it takes for any UAV k to perform the coverage task;

[0088] Create a graph G = (V, E) associated with the area to be covered. The coverage problem can be constructed as a vehicle routing problem. Solve the vehicle routing problem to obtain the route set for each drone; then calculate the time T it takes for drone k to perform the coverage task. k ;

[0089] Step 6C.5: Optimize the route set of each UAV obtained in step 6C.4 using the improved genetic algorithm;

[0090] The obstacle information (obstacle height and / or interference signal, which can be obtained by map or conventional measurement equipment), the entry point determined in step 5C, the initial path and waypoints in step 6C.1, the flight path spacing in step 6C.2, the minimum turning radius of the UAV, the turning strategy in step 6C.3, the route set in step 6C.4, and the time T spent by UAV k to perform the coverage task are recorded. k The improved genetic algorithm is input together with the time T k As the goal, optimize the route set and output time T k The smallest set of optimal paths.

[0091] Furthermore, the methods of steps 7A, 7B, and 7C are the same, and the improved A* algorithm is used to generate the return path of the UAV from the detection area to the recovery area. During the return process, a time collision avoidance strategy is used to assign a return time T to each UAV. j , ensuring that the UAV with a shorter mission path returns first; the return time T j Calculate according to the following formula:

[0092] T j =(1-g i β′)T

[0093] Among them, T j is the time interval from when UAV j takes off from the assembly point to when it is required to return, T is the battery life of UAV j, g i For the return order, the optimal path obtained based on the improved genetic algorithm is used to sort each UAV from small to large according to the mission path length, and the values are assigned to 0, 1, …, n-1, where n is the total number of UAVs; β′ is taken as 0.1-0.5.

[0094] Furthermore, the method for optimizing the initial population using the grid potential evaluation method is:

[0095] Divide the UAV coverage area into grid cells. According to the determined grid size, the entire flight area is divided into uniform grid cells to form a two-dimensional grid matrix.

[0096] Initially, all grid potentials are assigned to 0. Each time a UAV visits one of the grids, the grid potential value of the grid is increased by 1.

[0097] Compare the grid potentials of all grids around the current UAV's grid location and select one or more cells with the smallest grid potential as the candidate grids or the grids for the next waypoint. If there is only one grid with the smallest grid potential around it, the next waypoint is generated. If there are multiple grids with the smallest grid potential, a unique waypoint for the next step can be generated based on different grid patterns.

[0098] The grid mode is selected based on the slope and direction of the candidate grids. First, the grid cells in the candidate grid set are sorted by slope size, and the one with the gentlest slope is selected as the next path point. If there are cases where the slopes are the same, the grid cell with the slope direction closest to the current drone movement direction is selected as the next grid cell to be reached.

[0099] Furthermore, the convex decomposition adopts a trapezoidal decomposition method, which adopts a sweep line algorithm, treating each vertex in the detection area as an event, and the sweep line is a vertical line sweeping across the detection area from left to right; when encountering an event, the sweep line will extend the ray upward and downward until it intersects with the edges of the detection area above and below the event, and form a trapezoidal sub-area at the event according to the event type; after the sweep line completes the rightmost event, the trapezoidal decomposition is completed.

[0100] Beneficial effects of the present invention:

[0101] 1. The present invention solves the problems of low detection area coverage efficiency, insufficient target detection accuracy, limited coordination capability, and poor environmental adaptability in the prior art through technical means such as detection area division, path planning optimization, multimodal target detection (corresponding to step 5 of Examples 1-3), and dynamic task scheduling (corresponding to steps 4 and 6 of Examples 1-3), thereby realizing efficient, accurate, and robust detection tasks in complex urban environments.

[0102] 2. The present invention adopts the method based on trapezoidal decomposition (such as Figure 2 、 3, 6), the optimal coverage path generation based on the improved genetic algorithm, and the multi-UAV collaborative algorithm based on the improved A-star algorithm achieve the beneficial effects of improving detection coordination, detection efficiency and detection accuracy, saving UAV resources and enhancing detection safety. The convex decomposition method adopted by the present invention refers to decomposing the target area into multiple non-intersecting trapezoidal areas, where the triangle can be regarded as a degenerate trapezoid with a base length of 0, and the rectangle can be regarded as a special trapezoid with four right angles. When selecting the UAV flight path and turning strategy, the present invention optimizes the obtuse turning part, thereby improving the safe flight of the UAV during detection.

[0103] 3. The present invention first divides the detection area based on the proportion of drones carrying different payloads and payload parameters, and then plans the detection path of the drone cluster according to the drone parameters and the detection area, thereby maximizing the detection capability of the drone cluster and achieving the most accurate detection effect under a certain scale. Therefore, it can accurately detect in an environment with a large number of camouflaged targets and hidden targets.

[0104] 4. The present invention first determines relevant parameters of the drone cluster, such as the maximum range, maximum flight altitude, and carried payload type, based on the detection mission requirements. Then, it allocates a region for each drone based on the area of the entire detection area and the initial position of the drone. This method has high fault tolerance and can meet the requirements of the autonomous decision-making capabilities of drones in dynamic changes in urban environments.

[0105] 5. By adding a weight coefficient to the heuristic function of the existing A* algorithm, the improved A* algorithm does not need to always pursue the optimal path when it is very close to the end point. Instead, it prioritizes quickly reducing the estimated cost, thereby avoiding traversing a large number of unnecessary nodes and greatly improving the algorithm's running speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 It is a flow chart of the method of the present invention.

[0107] Figure 2 This is a schematic diagram of the area division of the area traversal task payload.

[0108] Figure 3 It is a schematic diagram of the convex decomposition of the concave quadrilateral detection area at the notch.

[0109] Figure 4 It is a flow chart of dividing a concave quadrilateral detection area into multiple convex quadrilateral sub-detection areas.

[0110] Figure 5 It is a schematic diagram of scan line segmentation in concave quadrilateral detection area.

[0111] Figure 6It is a schematic diagram of convex decomposition of complex arbitrary polygon detection area.

[0112] Figure 7 This is a schematic diagram of eight event types when a complex polygonal detection area is decomposed into a trapezoidal shape.

[0113] Figure 8 This is a schematic diagram of establishing the coordinate system of the line sweeping algorithm.

[0114] Figure 9 It is a schematic diagram of sub-area entry point selection.

[0115] Figure 10 is a schematic diagram of the turning strategy.

[0116] Figure 11 This is a schematic diagram of the relationship between UAV payload projection and flight path spacing.

[0117] Figure 12 This is a schematic diagram of the method for determining waypoints when the mission area is a concave quadrilateral.

[0118] Figure 13 It is a schematic diagram of the UAV adopting the η-shaped turning strategy.

[0119] Figure 14 This is a schematic diagram of the grid potential of the grid cells around the drone.

[0120] Figure 15 This is a schematic diagram of the route optimization when the UAV makes an obtuse-angle turn. DETAILED DESCRIPTION

[0121] In order to make the technical solutions and advantages of the present invention clearer and easier to understand, the present invention is further described in detail below with reference to the accompanying drawings.

[0122] Figure 1 The four parts of the drone cluster system, namely flight status information processing, load information processing, state control information processing, and information interaction module, belong to the operation of the drone cluster ground control station. Their connection with the area segmentation and path planning algorithm part focused on by the present invention is that the load information and flight status information of the ground control station provide dependencies and constraints for the method of the present invention. The ground control station generates a drone cluster control signal based on the results of the method of the present invention, controls the drone cluster through the information interaction module, and receives the detection results.

[0123] Example 1:

[0124] This embodiment is a multi-UAV collaborative detection method for a detection area in the shape of a convex quadrilateral.

[0125] like Figure 1As shown, the multi-UAV collaborative detection method for urban approach operations provided in this embodiment specifically includes the following steps:

[0126] Step 1: Determine the detection area;

[0127] Use satellite imagery or drone-generated aerial imagery to model the city's terrain and generate a 3D map. Based on the city's terrain and detection mission requirements, mark the area where close-proximity detection is required on the 3D map and define the boundaries of the detection area.

[0128] Step 2: Set the payload parameters and drone parameters.

[0129] Based on the detection area and its boundaries determined in Step 1, set the parameters for each payload carried by the drone, including the detection range, resolution, and field of view of the electro-optical and spectral payloads. Also, set the drone's level flight speed, endurance, and altitude. The combined reconnaissance area of the electro-optical and spectral payloads should cover the entire detection area.

[0130] Step 3: Detection area division;

[0131] The detection area is divided into the photoelectric load traversal area and the spectral load traversal area, and then the photoelectric load traversal area and the spectral load traversal area are divided into multiple sub-areas; the sub-areas are allocated to each drone according to the number of drones, and only one drone is allocated to each sub-area, and the drone allocated to each sub-area carries a photoelectric payload or a spectral payload.

[0132] Step 3.1: Establish a coordinate system;

[0133] like Figure 2 As shown in the figure, an inertial coordinate system is established with the drone takeoff assembly point O as the origin, the direction from the origin to the target area as the Y direction, the upward direction perpendicular to the bottom surface as the Z direction, and the rightward direction perpendicular to the YOZ plane as the X direction. The coordinates of the four sides and their endpoints of the detection area are determined in the inertial coordinate system.

[0134] Step 3.2: Select the reference edge;

[0135] Among the four edges of the detection area, the edge closest to the gathering point O is selected as the reference edge.

[0136] Step 3.3: Divide the detection area into the photoelectric load traversal area and the spectral load traversal area;

[0137] Assume that among the n drones participating in the detection mission, the number of drones equipped with optoelectronic payloads is n1, and their scan line spacing is d1; the number of drones equipped with spectral payloads is n2, and their scan line spacing is d2. In the same time, the scanning area of drones at the same speed is proportional to their scan line spacing. When dividing the detection area, the ratio of the area traversed by the optoelectronic payload to the area traversed by the spectral payload should be equal to the ratio of the drones' detection capabilities, that is, satisfying:

[0138]

[0139] Where S sg is the area covered by the UAV equipped with the optoelectronic payload, that is, the area traversed by the optoelectronic payload; S gp is the area covered by the UAV carrying the spectral payload, that is, the area traversed by the spectral payload.

[0140] Reference Figure 2 ,The method of dividing the detection area into the photoelectric load traversal area and the spectral load traversal area is as follows:

[0141] First, the endpoint of the two endpoints of the reference edge whose line with the assembly point O forms a larger angle with the X-axis of the inertial coordinate system is recorded as D1, and the other endpoint is recorded as D2. Then, the remaining endpoints of the detection area are recorded as D3 and D4 in a counterclockwise direction.

[0142] Then, select point F1 on the edge of D1D2 and point F2 on the edge of D3D4, and use F1F2 to traverse the detection area, thereby dividing the detection area into the photoelectric load traversal area D1F1F2D4 and the spectral load traversal area F1D2D3F2, and satisfying:

[0143]

[0144] Step 4: Generate and distribute scan lines within the detection area;

[0145] Step 4.1: Generate scanning lines within the photoelectric load traversal area D1F1F2D4;

[0146] Assume that the longest side of the photoelectric load traversal area D1F1F2D4 is D1D4, and let the scanning lines in the photoelectric load traversal area D1F1F2D4 be parallel to D1D4. The distance between the first scanning line and the edge of D1D4 is The distance between two adjacent scanning lines is d1, which is equal to half of the projection width L1 of the photoelectric payload on the ground (the scanning line spacing here is calculated by the parameters of the payload itself and is only used as the basis for generating scanning lines. It is different from the flight strip spacing during the actual scanning of the UAV). The beginning and end of the scanning line are respectively on the two parallel sides (D1F1 and D4F2) of the photoelectric payload traversal area, and the total number of scanning lines in the photoelectric payload traversal area can be obtained. sg for:

[0147]

[0148] in, is the rounding down symbol; according to the distance from the scanning line to D1D4 from small to large, the scanning lines are recorded as photoelectric load scanning line 1, photoelectric load scanning line 2, ..., photoelectric load scanning line X sg ;W sg The maximum width of the photoelectric load traversal area, respectively solve the distance from point F1 and F2 to D1 D4, W sg It is the value with the larger distance from F1 and F2 to D1 and D4.

[0149] Step 4.2: X sg The root scan line is assigned to n1 UAVs equipped with optoelectronic payloads. The traversal area covered by the scan line assigned to each UAV is the sub-area assigned to it;

[0150] According to the number of scan lines X sg The sub-area division and allocation based on the number of drones n1 can be divided into the following two cases:

[0151] 1) When n1=1:

[0152] At this time, there is only one UAV equipped with an optoelectronic payload, so there is no need to allocate scan lines. The optoelectronic payload traversal area D1F1F2D4 is the sub-area that the UAV needs to traverse.

[0153] 2) When n1 ≥ 2:

[0154] At this time, there are multiple UAVs equipped with optoelectronic payloads. When allocating scan lines, the following three scan line allocation principles should be met:

[0155] Principle ①: X should be used as much as possible. sg The scan lines are evenly distributed to n1 drones, that is, the difference in the number of scan lines allocated to any two drones should be as small as possible to ensure that the working time of each drone is roughly the same;

[0156] Principle ②: The parity of the number of scan lines assigned to as many drones as possible should be the same, that is, at most one drone should have a different parity of scan lines from the rest, to ensure that as many drones as possible leave the optoelectronic payload traversal area D1F1F2D4 from the same side after completing the scan;

[0157] Principle ③: As many drones as possible should be assigned an even number of scan lines. In this case, after completing the scan, the drone will leave the optoelectronic payload traversal area D1F1F2D4 on the same side of the entry point, which will not increase the length of the drone's mission route.

[0158] The priority of principle ①>the priority of principle ②>the priority of principle ③.

[0159] Based on the above scan line allocation principle, X sg The scan lines are assigned to n1 UAVs, and the area covered by the scan line assigned to each UAV is the sub-area it needs to scan and detect. For example, when 3 scan lines are distributed to 2 drones, the numbers of scan lines allocated to these 2 drones are 2 and 1 respectively, and the traversed sub-areas assigned to these 2 drones are the sub-areas covered by 2 scan lines and the sub-area covered by 1 scan line respectively; when 15 scan lines are distributed to 3 drones, each drone is allocated 5 scan lines, and the sub-areas assigned to them are all sub-areas covered by 5 scan lines; when 13 scan lines are distributed to 4 drones, the numbers of scan lines allocated to these 4 drones are 3, 3, 3, and 4 respectively. Among these 4 drones, the sub-areas assigned to them are all sub-areas covered by 3 scan lines, and one drone is assigned a sub-area covered by 4 scan lines; when 18 scan lines are distributed to 4 drones, the numbers of scan lines allocated to these 4 drones are 4, 4, 4, and 6 respectively. Among these 4 drones, the sub-areas assigned to them are all sub-areas covered by 4 scan lines, and one drone is assigned a sub-area covered by 6 scan lines.

[0160] Step 4.3: Use the same method as steps 4.1-4.2 to assign the sub-areas that the n2 UAVs carrying spectral payloads need to traverse.

[0161] The detection area of this embodiment is a convex quadrilateral, and since drones equipped with the same payload can be considered isomorphic, the completion of the scan line allocation represents the completion of the detection area allocation, where the sub-area allocated to the photoelectric payload drone is recorded as The sub-area of spectrum payload UAV allocation is denoted as For the task allocation between the same type of UAVs, a random allocation strategy is adopted and a numbering method is adopted. The UAVs assigned to sub-area j are recorded as UAVs. j .

[0162] Step 5: Plan a path from the drone cluster’s takeoff assembly point to the detection area;

[0163] This step first selects appropriate entry points for each subregion based on its geometry, allowing each drone to quickly enter it. Then, a modified A* algorithm is used to plan paths for each drone from its takeoff assembly point to each subregion's entry point. The modified A* algorithm dynamically adjusts the weights between actual and estimated costs, prioritizing paths that quickly reduce estimated costs.

[0164] The specific method is as follows:

[0165] Step 5.1: Select the entry point;

[0166] Step 5.1.1) Determine the scanning starting point of each sub-region j

[0167] The endpoint of the scanning line on both sides of sub-region j, the endpoint closest to the gathering point O, is used as the scanning starting point of sub-region j.

[0168] Step 5.1.2) Select the entry point of sub-region j

[0169] like Figure 9 As shown, the endpoint of the edge where the sub-area j coincides with the reference edge D1D2 of the entire detection area, the endpoint with the larger angle between the line connecting the sub-area j and the X-axis of the inertial coordinate system and the gathering point O is recorded as The other endpoint is recorded as Then, go counterclockwise along subregion j and record the remaining endpoints as and

[0170] The entry point of sub-region j is selected based on the following two situations:

[0171] When the scanning starting point of sub-region j From endpoint When it is closer, the entry point of the jth drone For endpoints The scanning direction of the drone is The initial flight direction of the drone is like Figure 9 The traversal sub-areas Sn and Sn-1 in correspond to this situation.

[0172] When the scanning starting point of sub-region j From endpoint When it is closer, the entry point of the jth drone For endpoints The scanning direction of the drone is The initial flight direction of the drone is like Figure 9 The traversal sub-areas S1 and S2 in correspond to this situation.

[0173] Step 5.2: Plan the entry points of each drone from the takeoff assembly point O to each sub-area j through the improved A* algorithm The path between

[0174] The heuristic function of the A* algorithm is:

[0175] f(n)=g(n)+h(n)

[0176] Where g(n) is the actual cost function, h(n) is the estimated cost function, and the weight ratio of g(n) and h(n) directly affects the search efficiency and path quality of the algorithm.

[0177] The difference between the present invention and the existing A* algorithm is that the distance between the node and the target point is dynamically adjusted by the weight coefficient w in the heuristic function of the existing A* algorithm, and the heuristic function of the A* algorithm is improved into a dynamic weighted heuristic function:

[0178] f(n)=g(n)+w·h(n)

[0179] At the same time, an offset p is introduced into the dynamic weighted heuristic function to ensure that when multiple nodes have the same heuristic function value, the node closer to the end point is preferred. The dynamic weighted heuristic function after introducing the offset p is as follows:

[0180] f(n)=g(n)+w·h(n)+p

[0181] Where p is a small positive number, usually p = 0.001;

[0182] The improved A* algorithm can plan the path of each UAV from the takeoff assembly point O to the detection area. By adjusting the weight of the actual and estimated costs, the improved A* algorithm can balance the actual or estimated costs. The heuristic function of the improved A* algorithm can be optimized as follows:

[0183] f(x,y)=g(x,y)+(w+p)h(x,y)

[0184] Through extensive experimental verification, we added a two-stage dynamic weighting to the h function: when h(n) < 18, the weight coefficient w = 0.8; when h(n) ≥ 18, the weight coefficient w = 3. The offset p is always 0.001.

[0185] Step 6: Plan the path of each drone in the detection area;

[0186] Step 6.1: Connect the scan lines generated in step 4 end to end to form an initial path. The endpoints of the scan lines are the waypoints on the initial path.

[0187] Step 6.2: Calculate the flight strip spacing of the UAV with spectral payload and the UAV with electro-optical payload respectively;

[0188] During the UAV scanning process, it is necessary to determine the flight strip spacing d 航间 For optoelectronic and spectral payloads, since the accuracy distribution of the payload projection ellipse on the short axis is symmetrical, and considering that the projection of the UAV data link antenna in the mission area needs to be perpendicular to the scanning direction during the scanning process, the short axis of the UAV payload projection ellipse is also perpendicular to the scanning direction. Figure 11 To ensure that the detection area can be fully covered and the targets in the area can be effectively identified, the flight strip spacing d 航间 The determination of needs to consider the flight strip overlap rate α.

[0189] According to the minor axis b of the payload projection ellipse and the flight path overlap ratio α, the flight path spacing d of the photoelectric payload UAV and the spectral payload UAV is calculated using the following formula 航间 :

[0190]

[0191] Where b is the minor axis of the projected ellipse of the optoelectronic payload / spectral payload, which is determined when the UAV leaves the factory and is a known quantity.

[0192] Step 6.3: Select a turning strategy;

[0193] When the drone flies from waypoint 1 to waypoint 2 and then reaches waypoint 3, it may make an acute angle + an obtuse angle turn, or a right angle + a right angle turn.

[0194] ①For fixed-wing drones:

[0195] Whether it is an acute angle + obtuse angle turn or a right angle + right angle turn, it is necessary to follow the flight strip spacing d 航间 , minimum turning radius R min (known quantity at the time of leaving the factory, or can be calculated by known conventional methods based on the drone's own parameters) and the distance d between the two end points of the adjacent flight strip AB The turning strategy is selected according to the following two situations:

[0196] Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight paths through circular turns (i.e. U-turn strategy or η-turn strategy). Figure 10As shown in the figure, the red trajectory is the turning trajectory. Figures (a) and (b) show the turning strategy when the angle between AB and L1 is obtuse, and Figures (c) and (d) show the turning strategy when the angle between AB and L1 is perpendicular. Figures (a) and (c) are conventional U-turn strategies, while Figures (b) and (d) are η-turn strategies.

[0197] Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips.

[0198] The η-turn strategy is as follows Figure 13 As shown, the red part in the figure is the turning trajectory, A is the end point of the current flight path, B is the starting point of the next flight path, and the turning process is from A to B. The direction before the turn is L1, and the direction after the turn is L2. (a) Figure is the η-shaped turning strategy when the angle between AB and L1 is an obtuse angle, and (b) Figure is the η-shaped turning strategy when the angle between AB and L1 is an acute angle.

[0199] Since the flight path spacing between the photoelectric payload UAV and the spectral payload UAV is different, the strategies selected by the photoelectric payload UAV and the spectral payload UAV according to the above two situations may be different.

[0200] ②For quadrotor drones:

[0201] When it is a right angle + right angle turn, the minimum turning radius is not considered and the turn can be made directly;

[0202] When the turn is acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to right angle + right angle turn. Figure 15 As shown in the figure, the drone flies from waypoint 1 to waypoint 2 and then to waypoint 3, demonstrating an acute-angle + obtuse-angle turn. For a quadrotor drone, we draw a perpendicular line through waypoint 2, extend it to waypoint 3, and use the intersection of this perpendicular line and the extended line of waypoint 3 as the newly added waypoint. This transforms the drone's flight turns from acute-angle + obtuse-angle to right-angle + right-angle turns. This simplified right-angle + right-angle turn route is simpler and shorter than a circular turn route, saving time.

[0203] Step 6.4: Determine the return point of each drone j in its assigned sub-area j

[0204] We can approximately assume that the projection of the center of mass of the drone in the detection area coincides with the payload projection ellipse. The point where the drone is located after completing its traversal detection mission in the sub-area is called the return point. The return point It is possible to be at the entry point of its sub-area j On the same side of the , it is also possible to be at the entry point of its sub-area j The return point of the drone can be determined based on the number of scan lines in the sub-area j to which the drone j is assigned.

[0205] The return point of the drone is a vector The end point can be calculated by Determine the return point

[0206] When the number of flight strips in sub-region j is odd, vector Calculated by the following formula:

[0207]

[0208] When the number of flight strips in sub-region j is even, vector Calculated by the following formula:

[0209]

[0210] Where, d 航间1 with d 航间2 are the flight strip spacings of UAVs equipped with optoelectronic payloads and spectral payloads, n1 is the number of UAVs equipped with optoelectronic payloads, and n is the total number of UAVs. Indicates the entry point from the jth UAV into the jth sub-area The return point when the jth sub-area scan ends vector, Indicates the entry point of the drone from sub-area j To the first scan line endpoint vector.

[0211] Step 6.5: Calculate the working time t of drone j scanning sub-area j j ;

[0212] The time t that the jth UAV traverses and scans its assigned sub-area j j It is directly related to the number of flight strips and the number of turns. In this invention, the turning trajectory of the UAV is approximately regarded as a semicircle. At this time, the time t for the jth UAV to scan sub-area j is j for:

[0213]

[0214] Where n j The number of flight strips that the j-th UAV covers and scans the j-th sub-area, n j-1 is the number of turns required for the drone to cover the jth area, d 航间1 with d 航间2 are the flight path spacing of the UAV equipped with optoelectronic payload and spectral payload, v j is the level flight speed of the UAV in the jth sub-area, L j is the average length of a single scan line in the jth sub-region.

[0215] Step 6.6: Optimize the waypoints on the initial path generated in step 6.1 using the improved genetic algorithm;

[0216] Obstacle information (obstacle height and / or interference signal, which can be obtained through maps or conventional measurement equipment), the entry point selected in step 5.1, the scanning direction and initial flight direction, the initial path and waypoints generated in step 6.1, the flight strip spacing calculated in step 6.2, the minimum turning radius of the UAV, the turning strategy in step 6.3, the return point determined in step 6.4, and the working time t calculated in step 6.5 are calculated. j The improved genetic algorithm takes the working time as the target and the obstacle information as the constraint condition, optimizes the waypoints determined in step 6.1, and outputs the working time t j The smallest optimal path.

[0217] The improved genetic algorithm in this embodiment will be described in detail at the end of the specification.

[0218] Step 7: Plan the path of the drone cluster from the detection area to the recovery area;

[0219] After the drone completes its detection mission, it uses the same modified A* algorithm as in step 5.2 to plan its return path from the detection area to the recovery area, ensuring its safe return to the assembly area. The modified A* algorithm uses a dynamic weighting method, introducing a dynamic weight coefficient w into the heuristic function. This ensures that when the distance to the destination is far, the algorithm prioritizes paths that quickly reduce the estimated cost (that is, the path with the shortest distance), reducing unnecessary node expansion. The modified A* algorithm demonstrates higher search efficiency in return path planning and significantly reduces computation time.

[0220] During the return process, since there are many waypoints near the assembly point O, this embodiment considers how to avoid collisions between drones. Specifically, a time collision avoidance strategy is adopted, and a return time T is assigned to each drone according to the length of its mission path. j , ensuring that drones with shorter mission paths return first.

[0221] Return time T j Calculate according to the following formula:

[0222] T j =(1-g i β′)T

[0223] Among them, T j is the time interval from when UAV j takes off from the assembly point to when it is required to return, T is the battery life of UAV j (assuming that all UAVs in the UAV cluster are of the same model), g i For the return order, based on the optimal path obtained in step 6, sort each drone according to the mission path length from small to large, and assign the corresponding g to each drone in turn. i The values are assigned as 0, 1, …, n-1, where n is the total number of drones. The value of β′ is 0.1-0.5, which can be selected according to the actual situation. The principle is: drones with short mission paths can be assigned larger β', and drones with high priority (for example, when there is a special requirement that a drone needs to return as soon as possible, the priority of the drone is set to high) can be assigned larger β'.

[0224] Example 2:

[0225] This embodiment is a multi-UAV collaborative detection method for a detection area in the shape of a concave quadrilateral.

[0226] like Figure 1 As shown, the specific method of this embodiment includes the following steps:

[0227] Step 1: Determine the detection area;

[0228] Use satellite imagery or drone-generated aerial imagery to model the city's terrain and generate a 3D map. Based on the city's terrain and detection mission requirements, mark the area where close-proximity detection is required on the 3D map and define the boundaries of the detection area.

[0229] Step 2: Set the payload parameters and drone parameters.

[0230] Based on the detection area and its boundaries determined in Step 1, set the parameters for each payload carried by the drone, including the detection range, resolution, and field of view of the electro-optical and spectral payloads. Also, set the drone's level flight speed, endurance, and altitude. The combined reconnaissance area of the electro-optical and spectral payloads should cover the entire detection area.

[0231] Step 3: Convert the concave quadrilateral detection area into multiple convex quadrilateral sub-regions;

[0232] In this embodiment, the detection area is a concave quadrilateral. If the area allocation method, as in Example 1, only considers scan lines, the detection area of the concave quadrilateral will have a large disparity in the allocated area. Therefore, for this type of detection area, the concave quadrilateral detection area is first converted into multiple convex quadrilateral sub-areas of as equal size as possible. In implementing this conversion method, the case where the concave quadrilateral detection area needs to be covered by two drones is first considered, and then the case where the concave quadrilateral detection area needs to be covered by a cluster of n (n>2) drones is finally extended to the case where the concave quadrilateral detection area needs to be covered by a cluster of n (n>2) drones.

[0233] The specific conversion method is as follows:

[0234] Step 3.1: Calculate the total area of the detection area and the coverage area allocated to each drone;

[0235] Calculate the total area of the entire concave quadrilateral detection area Area(CP), and calculate the area required to be covered by a single drone according to the number of drones participating in the detection mission n: AreaRequired(CP) = Area(CP) / n. The coverage area allocated to the i-th drone is recorded as Area(CP i ).

[0236] Step 3.2: Decompose the concave quadrilateral detection area into multiple convex quadrilateral sub-areas with the same area as much as possible;

[0237] Reference Figure 3 , this is the case where only two drones are required to cover the concave quadrilateral detection area. In this case, the concave quadrilateral detection area needs to be decomposed into two convex quadrilateral sub-areas. The decomposition method is: draw a dividing line from the concave point (notch) of the concave quadrilateral detection area. The area of partition CP1 on the right side of the dividing line is the coverage area allocated to the first drone, and the area of partition CP2 on the left side of the dividing line is the coverage area allocated to the second drone. Extend the edges on both sides of the concave point of the concave quadrilateral into the detection area. The area between the extension lines is the scanning range. Within this scanning range, the dividing line is rotated and scanned around the concave point as the axis to find the common boundary of partitions CP1 and CP2. This common boundary is the common boundary of the two convex quadrilateral sub-areas (the triangle can be regarded as a degenerate convex quadrilateral) obtained after decomposition:

[0238] When the area of the partition CP1 on the right side of the dividing line is smaller than the required coverage area AreaRequired(CP) for a single UAV, that is, Area(CP1)<AreaRequired(CP), the dividing line is scanned clockwise with the concave point of the concave quadrilateral as the axis within the scanning range until the area of the partition CP1 on the right side of the dividing line is equal to the required coverage area AreaRequired(CP) for a single UAV, that is, Area(CP1)=AreaRequired(CP), then the scanning is stopped. The current position of the dividing line is the common boundary of the two convex quadrilateral sub-areas, and the decomposition of the concave quadrilateral detection area is completed.

[0239] Similarly, when the area of the partition CP1 on the right side of the dividing line is larger than the coverage area AreaRequired(CP) required for a single UAV, that is, Area(CP1)>AreaRequired(CP), the dividing line is scanned counterclockwise with the concave point of the concave quadrilateral as the axis within the scanning range until the area of the partition CP1 on the right side of the dividing line is equal to the coverage area AreaRequired(CP) required for a single UAV, that is, Area(CP1)=AreaRequired(CP), then the scanning is stopped. The current position of the dividing line is the common boundary of the two convex quadrilateral sub-areas, and the decomposition of the concave quadrilateral detection area is completed.

[0240] When the concave quadrilateral detection area needs to be covered by n drones, each time a convex quadrilateral sub-area is divided by the above-mentioned dividing line rotation scanning method, the remaining detection area continues to be divided by the above-mentioned dividing line rotation scanning method until n convex quadrilateral sub-areas are divided. The flow chart of the division of the concave quadrilateral detection area is as follows: Figure 4 shown.

[0241] Step 4: Generate scan lines within the detection area;

[0242] like Figure 5 As shown in Figure 1, the reference edge translation scanning method is used to generate scan lines. The core idea is to find a convex edge closest to the drone take-off assembly point O, and use this convex edge as the reference edge to sequentially translate and generate scan lines.

[0243] by Figure 5 Taking the concave quadrilateral detection area shown as an example, we first need to determine which edge is closest to the drone takeoff assembly point O. Assume that the distances from point O to the four vertices of the concave quadrilateral are d1, d2, d3, and d4, and the distances from point O to the four sides of the concave quadrilateral are D1, D2, D3, and D4, respectively. Here, D1, D2, D3, and D4 are not the strict geometric distances from the assembly point O to the four sides of the concave quadrilateral, but are positively correlated with the strict geometric distances and can be used to determine the distance relationship between the assembly point O and the four sides of the concave quadrilateral, where:

[0244] D1=d1+d2

[0245] D2=d2+d3

[0246] D3=d3+d4

[0247] D4=d4+d1

[0248] Compare the sizes of D1, D2, D3, and D4. The edge corresponding to the minimum value among them is the edge closest to the assembly point O.

[0249] Next, we need to determine whether the edge closest to the assembly point O is a convex edge: we extend the closest edge and determine whether the extension line intersects the concave quadrilateral detection area at any point other than the four vertices. If there are additional intersections, the edge is concave; if there are no additional intersections, the edge is convex. If the edge closest to the assembly point O is a convex edge, we use this edge as the reference edge. Otherwise, we use the same method to determine whether the second closest edge to the assembly point O is a convex edge, and so on until we find an edge that meets the conditions and serves as the reference edge.

[0250] Finally, the reference edges are translated in sequence to determine the scan line:

[0251] Assuming that the flight altitude of the optoelectronic payload UAV performing the detection mission is h1, the projection width L1 of the optoelectronic payload on the UAV on the ground is:

[0252]

[0253] Assuming that the flight altitude of the spectral payload UAV performing the detection mission is h2, the projection width L2 of the spectral payload on the UAV on the ground is:

[0254]

[0255] Where δ is the lateral field of view of the photoelectric payload, and ε is the longitudinal field of view of the spectral payload.

[0256] Take L as the smaller value of L1 and L2, then the distance between the first scanning line and the reference edge in the entire detection area is The interval between two adjacent scan lines thereafter is L, until the scan line generated after the reference edge is translated has no intersection with the detection area, and the determination of all scan lines in the entire detection area is completed.

[0257] The detection area of this embodiment is a concave quadrilateral. Since the scan line lengths of the entire concave quadrilateral detection area vary greatly, it is impossible to distribute them to the drones as evenly as possible. Therefore, after the scan line is determined, scan line allocation is not performed here. Instead, waypoints are determined based on the scan line in the subsequent step 6 and the waypoints are allocated to each drone.

[0258] Step 5: Plan a path from the drone cluster’s takeoff assembly point to the detection area;

[0259] This step first selects appropriate entry points for each subregion based on its geometry, allowing each drone to quickly enter it. Then, a modified A* algorithm is used to plan paths for each drone from its takeoff assembly point to each subregion's entry point. The modified A* algorithm dynamically adjusts the weights between actual and estimated costs, prioritizing paths that quickly reduce estimated costs.

[0260] The specific method is as follows:

[0261] Step 5.1: Select the entry point;

[0262] Step 5.1.1) Determine the scanning starting point of each convex quadrilateral sub-region j

[0263] The endpoint of the scan lines on both sides of sub-region j that is closest to the gathering point O is used as the starting point of sub-region j.

[0264] Step 5.1.2) Select the entry point for the drone to enter sub-area j

[0265] like Figure 9 As shown in the figure, the edge closest to the gathering point O in the sub-region j is taken as the sub-region reference edge, and the endpoint of the sub-region reference edge with the larger angle between the line connecting the gathering point O and the X-axis of the inertial coordinate system is recorded as The other endpoint is recorded as Then, go counterclockwise along subregion j and record the remaining endpoints as and

[0266] The entry point of sub-region j is selected based on the following two situations:

[0267] When the scanning starting point of sub-region j From endpoint When closer, the drone's entry point For endpoints The scanning direction of the drone is The initial flight direction of the drone is

[0268] When the scanning starting point of sub-region j From endpoint When closer, the drone's entry point For endpoints The scanning direction of the drone is The initial flight direction of the drone is

[0269] Step 5.2: Use the improved A* algorithm to plan the entry points of each drone from the takeoff assembly point O to each sub-area j in the detection area The improved A* algorithm used here is exactly the same as the improved A* algorithm used in step 5.2 of Example 1, and will not be repeated here.

[0270] Step 6: Plan the path of each drone in the detection area;

[0271] Step 6.1: Connect the scan lines generated in step 4 end to end to form an initial path;

[0272] Step 6.2: Calculate the flight path spacing for the electro-optical and spectral payload drones during scanning. (This step can also be performed after allocating waypoints, as long as it is performed before the turning strategy)

[0273] As in Example 1, to ensure that the detection area can be fully covered and the targets in the area can be effectively identified, the determination of the flight strip spacing d requires the flight strip overlap rate α. The flight strip spacing d is calculated using the following formula for both the photoelectric payload UAV and the spectral payload UAV: 航间 :

[0274]

[0275] Where b is the minor axis of the projected ellipse of the optoelectronic payload / spectral payload, which is determined when the UAV leaves the factory and is a known quantity.

[0276] Step 6.3: Determine the waypoints on the initial path;

[0277] In this embodiment, the entire detection area is a concave quadrilateral. In order to minimize the scanning of areas outside the detection area and improve the scanning efficiency, a waypoint is required. The three sides of the detection area except the reference side need to be shifted inward by a distance k. The intersection of the indented boundary line and the scanning lines generated in step 4 is the waypoint position, as shown in FIG. Figure 12 As shown by the black origin in the middle. L is the smaller value of L1 and L2. L1 and L2 are the projection widths of the photoelectric load and the spectral load on the ground, respectively, which have been calculated in the above step 4.

[0278] Step 6.4: Assign the waypoints determined in step 6.3 to each UAV;

[0279] After determining the coordinates of all waypoints in the entire detection area, it is necessary to calculate the time of arrival of each waypoint based on the time t j ,waypoints are assigned to individual drones, ensuring that each drone has roughly the same working time.

[0280] Assume that the number of waypoints is m, and the arrival time t of each waypoint is calculated from the scanning starting point. j for:

[0281]

[0282] In the formula, uav s is the flight speed of the UAV, i is the number of the waypoint, i=1,2,3,…m; j=1,2…,m; d i-1 is the distance between the i-1th waypoint and the i-th waypoint;

[0283] The time t allocated to each drone is:

[0284]

[0285] In the formula, num uvv is the total number of drones; t m is the arrival time of the last waypoint m starting from the scan start point.

[0286] By comparing the arrival time t of each waypoint one by one j Allocate each waypoint to the corresponding drone according to the time t allocated to each drone, ensuring that the working time of each drone is roughly the same. At this time, it can not only efficiently cover the detection area but also reduce resource waste.

[0287] For ease of understanding, let's use an example. Assume there are five waypoints and two drones. Starting from the scan start point, the arrival times for the five waypoints are: t1 = 0 min, t2 = 1 min, t3 = 3 min, t4 = 4 min, and t5 = 6 min. The time allocated to each drone is t = 3 min. For the first drone, since the time from the start point to the first waypoint is t1 = 0 < 3 min, the time from the start point to the second waypoint is t2 = 1 < 3 min, and the time from the start point to the third waypoint is t3 = 3 min, the first drone should follow the path between waypoints 1-3. For the second drone, the time from the scan start point to the remaining waypoints (waypoints 4 and 5) is t4 = 4 < 2t = 6 min, and t5 = 6 min = 2t. Therefore, the second drone should follow the path between waypoints 4 and 5.

[0288] Step 6.5: Select a turning strategy;

[0289] When the drone flies from waypoint 1 to waypoint 2 and then reaches waypoint 3, it may make an acute angle + obtuse angle turn or a right angle + right angle turn. The method for selecting a turning strategy here is exactly the same as that in step 6.3 of Example 1, and is also divided into two types of fixed-wing drones and quad-rotor drones for explanation. Among them: for fixed-wing drones, the turning strategy needs to be selected based on the relationship between the flight strip spacing, the minimum turning radius and the distance between the two end points of adjacent flight strips; for quad-rotor drones, when encountering an acute angle + obtuse angle turn, a new waypoint needs to be introduced to convert it into a right angle + right angle turn, such as Figure 15 As shown, no further details are given here.

[0290] Step 6.6: Optimize the waypoints assigned to each UAV using the improved genetic algorithm;

[0291] The obstacle information (obstacle height and / or interference signal, which can be obtained through maps or conventional measuring equipment), the entry point selected in step 5.1, the scanning direction and initial flight direction, the initial path generated in step 6.1, the flight strip spacing calculated in step 6.2, the minimum turning radius of the UAV, the turning strategy in step 6.5, the waypoints assigned to each UAV in step 6.4, and the duration t of each UAV calculated in step 6.4 are input into the improved genetic algorithm. The improved genetic algorithm optimizes the waypoints assigned to each UAV with duration t as the target and obstacle information as the constraint condition, and outputs the optimal waypoint allocation set with the minimum duration t. These optimal waypoint allocation sets constitute the optimal path.

[0292] The improved genetic algorithm in this embodiment will be described in detail at the end of the specification.

[0293] Step 7: Plan the path of the drone cluster from the detection area to the recovery area;

[0294] After the drone completes its detection mission, it uses the same modified A* algorithm as in step 5 to plan its return path from the detection area to the recovery area, ensuring its safe return to the assembly area. The modified A* algorithm uses a dynamic weighting method, introducing a dynamic weight coefficient w into the heuristic function. This ensures that when the distance to the destination is far, the algorithm prioritizes paths that quickly reduce the estimated cost (i.e., the path closest to the destination), reducing unnecessary node expansion. This improved A* algorithm demonstrates higher search efficiency in return path planning, significantly reducing computation time.

[0295] During the return process, since the waypoints are relatively dense near the assembly point O, this embodiment, like Example 1, also adopts the same time collision avoidance strategy as Example 1 to avoid collisions between drones, which will not be repeated here.

[0296] Example 3:

[0297] This embodiment is a multi-UAV collaborative detection method for detection areas with complex polygonal shapes. A complex polygon here refers to a non-convex polygon with more than 4 sides or a polygon containing a no-fly zone.

[0298] The specific method of this embodiment includes the following steps:

[0299] Step 1: Determine the detection area;

[0300] Use satellite imagery or drone-generated aerial imagery to model the city's terrain and generate a 3D map. Based on the city's terrain and detection mission requirements, mark the area where close-proximity detection is required on the 3D map and define the boundaries of the detection area.

[0301] Step 2: Set the payload parameters and drone parameters.

[0302] Based on the detection area and its boundaries determined in Step 1, set the parameters for each payload carried by the drone, including the detection range, resolution, and field of view of the electro-optical and spectral payloads. Also, set the drone's level flight speed, endurance, and altitude. The combined reconnaissance area of the electro-optical and spectral payloads must cover the entire detection area.

[0303] Step 3: Convex decomposition of the detection area;

[0304] In area coverage tasks, if the detection area is convex and contains no obstacles, parallel track coverage planning is relatively simple, with the main task being to determine the arrangement direction of the scan lines. When scanning in different directions along the detection area, the scan line lengths are almost the same, but the number of scan lines can vary significantly, increasing the length of the drone's flight path between scan lines. Therefore, the total distance of the drone's area coverage path depends primarily on the number of scan lines, meaning that the total distance will decrease as the number of scan lines decreases. The number of scan lines in a given scanning direction is also proportional to the minimum height of the convex polygon in that direction. Therefore, the area coverage problem can be simplified to solving the minimum height (or width) of the convex polygon and its corresponding direction. In this case, scan lines can be generated along a direction perpendicular to the minimum height (or width).

[0305] However, in actual area coverage detection tasks, the detection area is often complex and includes no-fly zones. Therefore, it is necessary to decompose the complex detection area into multiple convex polygonal sub-areas. Each sub-area can be covered by parallel scan lines in different directions. The minimum number of scan lines in each sub-area can be determined by the sub-area's minimum height. To obtain an optimal solution to the non-convex polygonal area coverage problem, the sum of the minimum heights of the sub-areas must be minimized. Therefore, when processing the detection area, it is necessary to find a convex decomposition of the non-convex domain with the minimum height sum.

[0306] Since the scan line is a set of parallel flight trajectories in the area coverage task, this embodiment adopts a trapezoidal decomposition method for the convex decomposition of the detection area. The trapezoidal decomposition consists of sub-areas in the shape of a trapezoid or a triangle (which can be regarded as a degenerate trapezoid), which is convenient for the arrangement of the scan lines. The trapezoidal decomposition adopts a sweep line algorithm, which regards each vertex in the detection area as an event. In order to form the area decomposition, the sweep line is a vertical line that sweeps across the polygonal detection area from left to right. When encountering an event (vertex), it will extend the ray upward and downward until it intersects with the edges of the detection area above and below the event, and form a trapezoidal sub-area at the event according to the event type. After the scan line completes the rightmost event, the trapezoidal decomposition will be completed. The disadvantage of trapezoidal decomposition is that it will produce redundant convex polygonal sub-areas. Therefore, after completing the trapezoidal decomposition, some adjacent convex polygonal sub-areas can be merged into a larger convex polygonal sub-area to avoid redundancy.

[0307] by Figure 7 Taking the complex polygon detection area shown above as an example, the vertices of the complex polygon are numbered in counterclockwise order, while the vertices of the no-fly zone (obstacle) are numbered in clockwise order. In this case, the interior of the polygon is always located to the left of each edge when following the order. The sweep line is perpendicular to the X-axis and sweeps horizontally from left to right.

[0308] In the trapezoidal decomposition, all vertices are classified into five types of events: OPEN, CLOSE, SPLIT, MERGE, and INFLECTION, where INFLECTION is further divided into FLOOR-CONVEX events, FLOOR-CONCAVE events, CEIL-CONVEX events, and CEIL-CONCAVE events. These types of events are defined as follows:

[0309] ⑴OPEN event: The two adjacent vertices of vertex v are on the right side of the scan line, and the internal angle at v is less than π.

[0310] ⑵SPLIT event: The two adjacent vertices of vertex v are on the right side of the scan line, and the internal angle at v is greater than π.

[0311] ⑶CLOSE event: The two adjacent vertices of vertex v are on the left side of the scan line, and the internal angle at v is less than π.

[0312] ⑷MERGE event: The two adjacent vertices of vertex v are on the left side of the scan line, and the internal angle at v is greater than π.

[0313] ⑸FLOOR-CONVEX event: the previous adjacent vertex v of vertex v prev is on the left side of the scan line, and its next adjacent vertex v next Located to the right of the scan line, and the interior angle at v is less than π.

[0314] ⑹FLOOR-CONCAVE event: the previous adjacent vertex v of vertex v prev is on the left side of the scan line, and its next adjacent vertex v next Located to the right of the scan line, and the interior angle at v is greater than π.

[0315] ⑺CEIL-CONVEX event: the previous adjacent vertex v of vertex v prev is on the right side of the scan line, and its next adjacent vertex v next Located on the left side of the scan line, and the interior angle at v is less than π.

[0316] ⑻CEIL-CONCAVE event: the previous adjacent vertex v of vertex v prev is on the right side of the scan line, and its next adjacent vertex v next Located on the left side of the scan line, and the interior angle at v is greater than π.

[0317] Examples of the eight event types are shown in the figure. Node (1) in the figure is an OPEN event, node (5) is a CLOSE event, node (9) is a SPLIT event, node (10) is a MERGE event, nodes (2, 3, 4, 10) are FLOOR-CONVEX events, node (11) is a FLOOR-CONCAVE event, nodes (6, 7, 8, 14) are CEIL-CONVEX events, and nodes (13, 15) are CEIL-CONCAVE events.

[0318] When applying the sweep line algorithm, the coordinate system needs to be established first. Since the complex task area decomposition is only performed on the plane, it is only necessary to establish a plane coordinate system with the assembly point as the origin and the east direction as the x Axis direction, north is the y-axis direction to establish a plane coordinate system, and any sweep line direction is selected Then, the coordinate axis is rotated so that the x-axis direction of the coordinate system is aligned with Parallel. When applying the sweep line algorithm, the plane coordinate system is as follows Figure 8 shown.

[0319] When applying the sweep line algorithm to decompose the detection area, the x-coordinates of the events are first sorted in ascending order. During the sweep process, the edges of the detection area that intersect the sweep line are stored in a balanced binary tree L. Each subregion in the sweep line algorithm can be represented by two lists: the celing list and the floor list. The sweep line algorithm starts from left to right, visits each event in order, and performs different operations based on the event type:

[0320] ⑴OPEN event: Insert the two associated edges of this event into the balanced binary tree L, create a new sub-region, and add the vertex represented by this event to the floor list of the sub-region.

[0321] ⑵SPLIT event: Search for the detection area edges above and below the event in the balanced binary tree L, then determine the intersection of the sweep line with the upper edge, as well as the intersection of the sweep line with the lower edge. At this time, the sub-region to which the event belongs is found, and the intersection of the sweep line with the upper edge and the intersection of the sweep line with the lower edge are added to the celing list and floor list of the cell respectively. The current sub-region is then considered to have been constructed. Then, two new sub-regions are generated, and the vertices represented by this event are added to the L list of the upper new sub-region and the celing list of the lower new sub-region respectively. The intersection of the sweep line and the upper edge is added to the celing list of the upper new sub-region, and the intersection of the lower sweep line and the lower edge is added to the floor list of the lower new sub-region. The two associated edges of the event are then inserted into L.

[0322] (3) FLOOR-CONVEX event: Find the subregion to which this event belongs. Add the vertex represented by this event to the floor list of the current subregion. Delete the left-hand edge associated with this event from L and insert the right-hand edge associated with this event into L.

[0323] 4. CEIL-CONVEX event: Find the subregion to which this event belongs. Add the vertex represented by this event to the celing list of the current subregion. Delete the left-hand edge associated with this event from L and insert the right-hand edge associated with this event into L.

[0324] ⑸FLOOR-CONCAVE event: Delete the left associated edge of the event from L, search for the edge above the event in L, and determine the intersection of the sweep line and the edge above the event. Then add the right associated edge of the event to L to find the sub-region to which the event belongs. Add the vertices of this event to the floor list of the current sub-region, and add the intersection of the sweep line and the edge above the event to the celing list of the current sub-region. The current sub-region is then constructed. Then create a new sub-region, add the vertices of the event to the floor list of the new sub-region, and add the intersection of the sweep line and the edge above the event to the celing list of the new sub-region.

[0325] ⑹CEIL-CONCAVE event: Delete the left associated edge of the event from L, search for the edge below the event in L, and determine the intersection of the sweep line and the edge below the event. Then add the right associated edge of the event to L and find the sub-region to which the event belongs. Add the vertices of this event to the celing list of the current sub-region, and add the intersection of the sweep line and the edge below the event to the floor list of the current sub-region. The current sub-region is then constructed. Then create a new sub-region, add the vertices of this event to the celing list of the new sub-region, and add the intersection of the sweep line and the edge below the event to the floor list of the new sub-region.

[0326] ⑺MERGE event: Delete the two associated edges of the event from L, search for the edges above and below the event in L, and then determine the intersection of the scan line with the upper edge, and the intersection of the scan line with the lower edge. Find the two sub-regions to which this event belongs. Add the vertices of this event to the floor list of the upper sub-region and the celing list of the lower sub-region respectively. Add the intersection points on the upper edge to the celing list of the upper sub-region, and add the intersection points on the lower edge to the floor list of the lower sub-region. Then both sub-regions have been constructed. Then, create a new sub-region. Add the intersection points on the upper edge to the celing list of the new sub-region, and add the intersection points on the lower edge to the floor list of the new sub-region.

[0327] ⑻CLOSE event: Delete the two edges associated with this event from L. Find the subregion to which this event belongs. Add the vertex of this event to the floor list of the current subregion. The current subregion is now constructed.

[0328] After the scan line traverses all events, the decomposition of the complex polygon detection area can be considered complete. However, when decomposing the area, the best scan line decomposition direction is unknown. When all scan line directions are perpendicular to the scan decomposition direction, the best scan direction must be perpendicular to one of the boundaries or obstacle edges. In this embodiment, only the direction perpendicular to the detection area boundary or the edge of the no-fly zone is selected as the scan line direction. For each scan line direction, the above (1)-(8) are repeated to obtain multiple sub-area decomposition schemes, and the sum of the minimum heights of the sub-areas is determined respectively. The convex decomposition scheme with the smallest sum of the minimum heights is selected as the final decomposition scheme to obtain the final decomposed convex polygon sub-area.

[0329] Step 4: Divide the detection area into photoelectric and spectral payload task areas, and generate scan lines for each;

[0330] Step 4.1: Generate scan lines;

[0331] The same reference edge translation scanning method as step 4 in Example 2 is used to create parallel scan lines in each convex polygon sub-region. At this time, the reference edges are the convex edges closest to the drone takeoff assembly point in each convex polygon sub-region.

[0332] First, solve the projection widths L1 and L2 of the photoelectric payload UAV and the spectral payload UAV on the ground respectively, and calculate the number of scan lines in each convex polygon sub-region j using the following formula and

[0333]

[0334] where s∈(0,1) represents the image overlap ratio, which is usually required to connect images in the area coverage of visual sensors. is the minimum height of the j-th convex polygon sub-region.

[0335] Then use the following formula to solve the scan line spacing in each convex polygon sub-region j:

[0336]

[0337] Among them, N l for and Small value in .

[0338] Finally, based on the solved number and spacing of scan lines in each sub-region, the scan lines in each convex polygon sub-region are generated respectively.

[0339] Step 4.2: Merge adjacent sub-regions with the same scan line direction;

[0340] Since each subregion is covered independently, when parallel trajectories of two adjacent subregions have the same trajectory direction, redundant trajectories covering these two adjacent subregions may be generated. If two convex polygonal subregions have the same trajectory direction and are completely adjacent to each other, they can be merged to avoid redundant trajectories.

[0341] The definitions of adjacency and full adjacency of two polygons are described as follows:

[0342] Consider a polygon P1 with n vertices and a polygon P2 with m vertices. If the edge v of P1 1i v 1(i+1) The edge v between (i∈[1,n]) and P2 2j v 2(j+1) (j∈[1,m]) coincides, then P1 and P2 are adjacent. When P1 and P2 are adjacent, the coincident edges are v 1i v 1(i+1) and v 2jv 2(j+1) In the case of v 1i =v 2j (v 1i =v 2(j+1) or v 1(i+1) =v 2j ), then P1 and P2 are completely adjacent.

[0343] By detecting region decomposition and sub-region adjacency, the merging process can be divided into four steps:

[0344] First, assign each sub-polygon a unique number. Second, iterate over all of these sub-polygons. For each sub-polygon, test all completely adjacent sub-polygons to determine whether the two sub-polygons meet the merge criteria. If so, change the adjacent sub-polygon's number to the current sub-polygon's number. Third, sort the sub-polygons according to their group numbers. Finally, merge sub-polygons with the same group number, completing the merging of redundant regions.

[0345] Step 4.3: Divide the entire detection area into the photoelectric payload task area and the spectral payload task area;

[0346] Step 4.3.1: Based on the number of scan lines in each sub-area obtained in Step 4.1, calculate the total number of scan lines for the entire detection area. Calculate the proportion of EO payload UAVs to the total number of UAVs. Multiply this proportion by the total number of scan lines to determine the number of EO scan lines required for the EO payload UAV. Then, select several sub-areas from all the sub-areas merged in Step 4.2. The sum of the scan line numbers in these selected sub-areas should be closest to the number of EO scan lines. Determine whether the selected sub-area combination is unique. If not, select the sub-area combination with the largest number of adjacent sub-areas and assign it to the EO payload UAV as EO payload mission area P1. Finally, assign the remaining sub-areas to the spectral payload UAV as spectral payload mission area P2.

[0347] Step 4.3.2: Decompose the homogeneous load region and regenerate the scan lines of each sub-region;

[0348] If the photoelectric payload task area P1 is a convex quadrilateral, the scan line is directly regenerated using the method in step 4.1; otherwise, it is necessary to use the same convex decomposition method as step 3 to decompose it into a set of multiple convex polygonal sub-areas, and then use the method in step 4.1 to regenerate the scan line in each current convex polygonal sub-area.

[0349] If the task area P2 of the spectral load is a convex quadrilateral, the scan line is directly regenerated using the method in step 4.1; otherwise, it is necessary to use the same convex decomposition method as step 3 to decompose it into a set of multiple convex polygonal sub-areas, and then use the method in step 4.1 to regenerate the scan line in each current convex polygonal sub-area.

[0350] Step 5: Plan a path from the drone cluster’s takeoff assembly point to the detection area;

[0351] In this step, only one entry point is selected based on the geometry of the entire detection area. Then, the improved A* algorithm, which is the same as that in Example 1, is used to plan the entry path of the UAV from the takeoff assembly point to the detection area.

[0352] The specific method is as follows:

[0353] Step 5.1: Select the entry point;

[0354] From the endpoints of the scan line within the detection area, select the endpoint closest to the drone takeoff assembly point O as the entry point D s When there are multiple nearest endpoints, just select any one.

[0355] Step 5.2: Use the improved A* algorithm to plan the movement of each drone from the takeoff assembly point O to the entry point D of the detection area s The improved A* algorithm used here is exactly the same as the improved A* algorithm used in step 5.2 of Example 1, and will not be repeated here.

[0356] Step 6: Plan the path of each drone in the detection area;

[0357] Step 6.1: Connect the scan lines generated in step 4.3.2 end to end to form an initial path. The endpoints of each scan line are waypoints on the initial path.

[0358] Step 6.2: Calculate the flight strip spacing d during drone scanning;

[0359] As in Example 1, to ensure that the detection area can be fully covered and the targets in the area can be effectively identified, the determination of the flight strip spacing d requires the flight strip overlap rate α. The flight strip spacing d is calculated using the following formula for both the photoelectric payload UAV and the spectral payload UAV: 航间 :

[0360]

[0361] Where b is the minor axis of the projected ellipse of the optoelectronic payload / spectral payload, which is determined when the UAV leaves the factory and is a known quantity.

[0362] Step 6.3: Select a turning strategy;

[0363] When the drone flies from waypoint 1 to waypoint 2 and then reaches waypoint 3, it may make an acute angle + obtuse angle turn or a right angle + right angle turn. The method for selecting a turning strategy here is exactly the same as that in step 6.3 of Example 1, and is also divided into two types of fixed-wing drones and quad-rotor drones for explanation. Among them: for fixed-wing drones, the turning strategy needs to be selected based on the relationship between the flight strip spacing, the minimum turning radius and the distance between the two end points of adjacent flight strips; for quad-rotor drones, when encountering an acute angle + obtuse angle turn, a new waypoint needs to be introduced to convert it into a right angle + right angle turn, such as Figure 15 As shown, no further details are given here.

[0364] Step 6.4: Solve for the set of routes for each UAV and calculate the time it takes for any UAV k to perform the coverage task;

[0365] First, we need to create a graph G = (V, E) associated with the detection area to be covered. The set of scan line endpoints and drone recovery points can be considered the node set V of the graph G = (V, E). Each node in the node set V is assigned a corresponding number. The drone assembly point is numbered as node 1, and the remaining scan line endpoints are numbered sequentially according to the numbering of adjacent nodes on the same scan line. This way, after the numbering is completed, the two endpoints of each scan line are necessarily adjacent odd-even numbers. The edge set E consists of all the straight lines connecting the N nodes, forming a complete graph G.

[0366] The graph G can be represented by an N×N cost matrix C. Since the UAV used in this scheme is a quad-rotor UAV, the flight path length between two points is not constrained by the initial heading, so its elements are C ij It is given by the straight-line distance or the shortest broken-line distance between the spatial coordinates of nodes i and j. The cost matrix C is time-invariant and symmetric, i.e. And the elements in the cost matrix C satisfy the triangle inequality, that is

[0367] Once a graph G is created that represents the area to be covered, the coverage problem can be formulated as a vehicle routing problem (VRP). In this type of problem, a set of customers must be visited by a set of vehicles. To transform the area coverage problem into a VRP, each drone is modeled as a vehicle and each endpoint of a scan line is modeled as a customer. Furthermore, constraints can be used to force vehicles to use certain pre-specified edges of the graph G in their routes, ensuring that each scan line is assigned a drone for coverage.

[0368] Finally, by solving the VRP, the route set of each UAV can be obtained.

[0369] Define the constant C ij Represents the flight cost of the edge (i, j) between nodes i and j, a binary variable Indicates whether the kth UAV flies from vertex i to vertex j. Let the constant is the flight speed of UAV k from vertex i to vertex j, and the constant t s Preparation time for the drone, L k is the battery life of drone k, m is the number of drones required to cover the mission, M is the total number of drones available for this mission, O is the number of landing gears available for the mission, and N is the number of nodes covering the mission. Finally, the variable d k The additional time required to launch UAV k represents the actual time required from the start of the mission to the launch of the UAV.

[0370] The time it takes for UAV k to perform the coverage mission is:

[0371]

[0372] Step 6.5: Use the improved genetic algorithm to optimize the route set of each UAV obtained in step 6.4;

[0373] The obstacle information (obstacle height and / or interference signal, which can be obtained through maps or conventional measuring equipment), the entry point determined in step 5, the initial path and waypoints in step 6.1, the flight strip spacing in step 6.2, the minimum turning radius of the UAV, the turning strategy in step 6.3, the route set in step 6.4 and the time Tk spent by UAV k to perform the coverage task are input into the improved genetic algorithm. The improved genetic algorithm optimizes the route set with time Tk as the target and outputs the optimal path set with the minimum time Tk.

[0374] The improved genetic algorithm in this embodiment will be described in detail at the end of the specification.

[0375] Step 7: Plan the path of the drone cluster from the detection area to the recovery area;

[0376] After the drone completes its detection mission, it uses the same improved A* algorithm as in step 5 to plan its return path from the detection area to the recovery area, ensuring its safe return to the assembly area. The improved A* algorithm uses a dynamic weighting method, introducing a dynamic weighting coefficient w into the heuristic function. This ensures that when the distance to the destination is far, the algorithm prioritizes paths that quickly reduce the estimated cost (that is, the path with the shortest distance), reducing unnecessary node expansion. The improved A* algorithm demonstrates higher search efficiency in return path planning, significantly reducing computation time.

[0377] During the return process, since the waypoints are relatively dense near the assembly point O, this embodiment, like Example 1, also adopts the same time collision avoidance strategy as Example 1 to avoid collisions between drones, which will not be repeated here.

[0378] In step 6 of Examples 1-3 above, the same improved genetic algorithm is used to obtain the optimal path. The only differences between the improved genetic algorithm and the existing genetic algorithm are: 1. The initial population in the existing genetic algorithm is optimized using a grid potential evaluation method; 2. Based on the existing genetic algorithm, a spline interpolation function is added to optimize the generated path. The rest of the process is the same as the existing genetic algorithm.

[0379] ①Optimize the initial population;

[0380] In existing area coverage path planning methods based on genetic algorithms, the initial population is mostly randomly generated, which leads to a strong randomness in the results and has a corresponding impact on the optimality of the results. Therefore, the present invention follows the following strategy for optimization:

[0381] If a drone's initial position isn't within its assigned subarea, the starting point of its coverage path within that subarea is uncertain, requiring a path with the optimal starting point. If a drone's initial position is within its assigned subarea, the coverage efficiency is optimal when its coverage path starts at its initial position. Therefore, this paper designs a grid potential evaluation method specifically for the first case, generating different coverage paths as the initial population by varying the path's starting point.

[0382] The grid potential evaluation method includes two grid evaluation strategies, namely grid potential (GP) and grid mode (GM), to facilitate the UAV to quickly and conveniently select the grid during movement. The grid potential represents the value corresponding to each reachable grid in the target area, and its value range is 0 to infinity; the grid mode represents the different priorities of the adjacent grid cells of the grid where the UAV is located.

[0383] Divide the grid cells in the drone coverage area, and divide the entire flight area into uniform grid cells according to the determined grid size to form a two-dimensional grid matrix. The grid size needs to be determined according to the mission requirements and the complexity of the environment. Generally, it is a few meters to more than ten meters when performing low-altitude flight missions in complex urban environments. In open farmland or mountainous areas, the grid side length can be appropriately increased. Initially, all grid potentials are assigned to 0. Each time the drone visits one of the grids, the grid potential value of the grid is increased by 1. The specific calculation formula is as follows:

[0384] GP(G i )=0,G i Not visited

[0385] GP(G i)=GP(G i )+1,G i Visited

[0386] Compare the grid potentials of all grids surrounding the current drone's location and select one or more cells with the smallest grid potential as candidate grids or the grid for the next waypoint. If there is only one grid with the smallest grid potential, the next waypoint is generated. If there are multiple grids with the smallest grid potential, a unique waypoint for the next step can be generated based on different grid patterns.

[0387] The drone's grid pattern is determined by the slope and direction of the candidate grids. Different grid patterns represent different priorities for candidate grids. Based on the slope and direction of each candidate grid, the flatness and slope direction of the grid can be determined. The flatter the terrain and the grid whose slope direction is closest to the current direction of movement of the drone have the highest priority. This design minimizes the drone's energy consumption and ensures that the drone selects the optimal solution at each step. The specific priority calculation method for the candidate grid is to first sort the grid cells in the candidate grid set by slope size, and select the one with the flattest slope as the next path point. If there are cases where the slope is the same, the grid cell with the slope direction closest to the current direction of movement of the drone is selected as the next grid cell to be reached.

[0388] like Figure 14 As shown in , the UAV selects the grid cell with the smallest grid potential among the grid cells around the current grid as the next path point, as shown in Figure 14 As shown in Figure (a); when there are multiple grid cells with the smallest grid potential around the grid where the drone is currently located, the priority of the grid cells with the same and smallest potential values is determined, and the grid with the highest priority is selected as the next path point, as shown in Figure 1. Figure 14 As shown in Figure (b).

[0389] ②Path optimization:

[0390] The optimal path for each drone calculated by a traditional genetic algorithm is composed of many straight line segments, which does not conform to the drone's motion characteristics. Therefore, this invention uses a cubic spline interpolation function to perform cubic spline interpolation based on the waypoints on the path generated by the genetic algorithm to optimize the drone's path. Ultimately, the path output by the genetic algorithm is a smooth path. This allows the drone controller to better control the drone's navigation using this trajectory information, thereby reducing mission completion time and achieving a more optimal path.

Claims

1. A multi-UAV collaborative detection method for urban approach operations, characterized in that: The following steps are involved: Step 1: Determine the detection area and its boundaries based on the urban terrain and detection mission requirements; Step 2: According to the detection mission requirements, set the payload type and parameters of the drone so that the drone's detection area can cover the detection area; determine the shape of the detection area. If it is a convex quadrilateral, proceed to step 3A. If it is a concave quadrilateral, go to step 3B; If it is a complex polygon, go to step 3C; Step 3A: Divide the detection area into the photoelectric payload and spectral payload traversal areas, and the ratio of the two areas is equal to the ratio of the detection capabilities of the photoelectric payload and spectral payload UAVs; Step 4A: Generate scan lines within the traversal area of the photoelectric payload and the spectral payload respectively, and distribute them as evenly as possible to each photoelectric payload and each spectral payload UAV. The sub-area that each UAV needs to traverse is the sub-area covered by its assigned scan line. Step 5A: Use the improved A* algorithm to plan the UAV’s path from the takeoff assembly point to the detection area; Step 6A: Use the improved genetic algorithm to plan the UAV’s path within the detection area; Step 7A: Use the improved A* algorithm to plan the drone's path from the detection area to the recovery area. The process ends. Step 3B: Draw a segmentation line from the concave point of the concave quadrilateral. Scan the segmentation line with the concave point as the axis to decompose the detection area into n convex polygonal sub-areas of equal / similar areas; n is equal to the total number of drones; Step 4B: Generate a scan line for the detection area using a reference edge translation scanning method; the reference edge is the convex edge in the detection area closest to the drone takeoff assembly point; Step 5B: Use the improved A* algorithm to plan the UAV’s path from the takeoff assembly point to the detection area; Step 6B: Use the improved genetic algorithm to plan the UAV’s path within the detection area; Step 7B: Use the improved A* algorithm to plan the drone's path from the detection area to the recovery area. The process ends. Step 3C: Decompose the detection area into multiple convex polygonal sub-areas using a convex decomposition method; Step 4C: Scan line generation; Step 4C.1: Generate scan lines in each sub-region using the reference edge translation scanning method, and merge adjacent sub-regions with the same scan line direction; Step 4C.2: Calculate the scan lines allocated to the EO / spectral payload UAVs based on the ratio of the number of EO / spectral payload UAVs to the total number of scan lines. From the merged sub-areas, select sub-areas with a number of scan lines close to the number of scan lines allocated to the EO / spectral payload UAVs as the EO / spectral payload mission areas. The remaining sub-areas are designated as spectral / EO payload sub-areas. Step 4C.3: If the photoelectric payload task area is a convex quadrilateral, directly use the reference edge translation scanning method to regenerate the scan lines. Otherwise, use the convex decomposition method to decompose it into multiple sub-areas, and then use the reference edge translation scanning method to regenerate the scan lines of each sub-area. If the spectral load task area is a convex quadrilateral, the reference edge translation scanning method is directly used to regenerate the scan line. Otherwise, the convex decomposition method is used to decompose it into multiple sub-areas, and then the reference edge translation scanning method is used to regenerate the scan line. The reference edge is the convex edge in each sub-area that is closest to the drone take-off assembly point; Step 5C: Use the improved A* algorithm to plan the UAV's path from the takeoff assembly point to the detection area; Step 6C: Use the improved genetic algorithm to plan the UAV's path within the detection area; Step 7C: Use the improved A* algorithm to plan the drone's path from the detection area to the recovery area. The process ends. The heuristic function of the improved A* algorithm is a dynamic weighted heuristic function: f(x,y)=g(x,y)+(w+p)h(x,y) Where g(n) and h(n) are the actual cost function and the estimated cost function, respectively; p is the offset, p = 0.001; w is the weight coefficient, when h(n) < 18, w = 0.8; when h(N) ≥ 18, W = 3; The improved genetic algorithm is based on the existing genetic algorithm, uses a grid potential evaluation method to optimize the initial population, and adds a cubic spline interpolation function to optimize the path generated by the genetic algorithm so that the genetic algorithm finally outputs a smooth path.

2. The multi-UAV collaborative detection method for urban approach operations according to claim 1 is characterized by: In step 4A, scan lines are assigned to the drone according to the following principles: Principle ①: The difference in the number of spectral payload scan lines assigned to any two spectral payload drones should be as small as possible, and the difference in the number of photoelectric payload scan lines assigned to any two photoelectric payload drones should be as small as possible; Principle ②: Ensure that as many photoelectric payload drones as possible are assigned the same number of photoelectric payload scan lines with the same parity, and as many spectral payload drones as possible are assigned the same number of spectral payload scan lines with the same parity; Principle ③: Make sure that as many photoelectric payload drones as possible are assigned an even number of photoelectric payload scan lines, and as many spectral payload drones as possible are assigned an even number of spectral payload scan lines; The priority of principle ①>the priority of principle ②>the priority of principle ③.

3. The multi-UAV collaborative detection method for urban approach operations according to claim 2 is characterized by: Steps 5A and 5B are the same, specifically: Step 5.1: The endpoints of the scan lines on both sides of each sub-area that are closest to the drone take-off assembly point are used as the scanning starting points of each sub-area. The endpoints of each sub-area that are closer to the scanning starting point are selected as the entry points of each sub-area. Step 5.2: Use the improved A* algorithm to generate the path from the drone takeoff assembly point to the entry point of each sub-area.

4. The multi-UAV collaborative detection method for urban approach operations according to claim 3 is characterized by: Step 5C is specifically as follows: Step 5C.1: From the endpoints of the scan lines within the detection area, select the endpoint closest to the drone takeoff assembly point O as the entry point D s ; When there are multiple nearest endpoints, select one at random; Step 5C.2: Plan the path of each UAV from the takeoff assembly point to the entry point of the detection area using the improved A* algorithm.

5. The multi-UAV collaborative detection method for urban approach operations according to claim 4 is characterized in that: Step 6A is specifically as follows: Step 6A.1: Connect the scan lines generated in step 4A end to end to form an initial path. The endpoints of the scan lines are the waypoints on the initial path. Step 6A.2: Calculate the flight strip spacing of the UAV carrying the spectral payload and the UAV carrying the electro-optical payload during scanning respectively; Step 6A.3: Select a turning strategy; For fixed-wing drones, choose a turning strategy based on the following two situations: Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight lanes through circular turns (i.e., U-turn strategy or η-turn strategy). Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips; For quadrotor drones, when the turn is a right angle + right angle, the minimum turning radius is not considered and the drone can turn directly. When the turn is an acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to a right angle + right angle turn. Step 6A.4: Determine the return point of each drone j in its assigned sub-area j Step 6A.5: Calculate the operating time t of drone j scanning sub-area j j ; Where n j The number of flight strips that the j-th UAV covers and scans the j-th sub-area, n j -1 is the number of turns required for the drone to cover the jth area, d 航间1 with d 航间2 are the flight path spacing of the UAV equipped with optoelectronic payload and spectral payload, v j is the level flight speed of the UAV in the jth sub-area, L j is the average length of a single scan line in the jth sub-region; Step 6A.6: Optimize the waypoints on the initial path generated in step 6.1 using the improved genetic algorithm; The obstacle information, the entry point selected in step 5.1, the scanning direction and initial flight direction, the initial path and waypoints generated in step 6.1, the flight path spacing calculated in step 6.2, the minimum turning radius of the UAV, the turning strategy in step 6.3, the return point determined in step 6.4, and the working time t calculated in step 6.5 are recorded. j The improved genetic algorithm takes the working time as the target and the obstacle information as the constraint condition, optimizes the waypoints determined in step 6.1, and outputs the working time t j The smallest optimal path.

6. The multi-UAV collaborative detection method for urban approach operations according to claim 5 is characterized by: Step 6B is specifically as follows: Step 6B.1: Connect the scan lines generated in step 4B end to end to form an initial path; Step 6B.2: Calculate the flight strip spacing between the UAV carrying the spectral payload and the UAV carrying the electro-optical payload during scanning. Step 6B.3: Determine the waypoints on the initial path; The remaining three edges of the detection area, except the reference edge, are shifted inward by a distance k. The intersection of the indented boundary line and the scan lines generated in step 4 is the waypoint position. L is the smaller value of L1 and L2, L1 and L2 are the projection widths of the photoelectric load and the spectral load on the ground respectively; Step 6B.4: Assign the waypoints determined in step 6B.3 to each UAV; Starting from the scan start point, calculate the arrival time of each waypoint, and calculate the duration t allocated to each drone based on the arrival time of the last waypoint. Compare the arrival time of each waypoint with the duration t allocated to each drone one by one, and assign each waypoint to the corresponding drone to ensure that the working time of each drone is as similar as possible. Step 6B.5: Select a turning strategy; For fixed-wing drones, choose a turning strategy based on the following two situations: Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight strips through circular turns (i.e., U-turn strategy or η-turn strategy); Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips; For quadrotor drones, when the turn is a right angle + right angle, the minimum turning radius is not considered and the drone can turn directly. When the turn is an acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to a right angle + right angle turn. Step 6B.6: Optimize the waypoints assigned to each UAV using the improved genetic algorithm; The obstacle information, the entry point selected in step 5, the scanning direction and the initial flight direction, the initial path generated in step 6B.1, the flight strip spacing calculated in step 6B.2, the minimum turning radius of the UAV, the turning strategy in step 6B.5, the waypoints assigned to each UAV in step 6B.4, and the duration t of each UAV calculated in step 6B.4 are input into the improved genetic algorithm. The improved genetic algorithm optimizes the waypoints assigned to each UAV with duration t as the target and obstacle information as the constraint condition, and outputs the optimal waypoint allocation set with the minimum duration t. These optimal waypoint allocation sets constitute the optimal path.

7. The multi-UAV collaborative detection method for urban approach operations according to claim 6 is characterized by: Step 6C is specifically as follows: Step 6C.1: Connect the scan lines generated in step 4C end to end to form an initial path. The endpoints of each scan line are waypoints on the initial path. Step 6C.2: Calculate the flight strip spacing between the UAV carrying the spectral payload and the UAV carrying the electro-optical payload during scanning respectively; Step 6C.3: Select a turning strategy; For fixed-wing drones, choose a turning strategy based on the following two situations: Strip spacing d 航间 Meeting 2Rs min ≤d 航间 , the UAV can achieve smooth transition between flight strips through circular turns (i.e., U-turn strategy or η-turn strategy); Strip spacing d 航间 Satisfy d 航间 <2R min ≤d AB , the UAV needs to use the η-shaped turning strategy to achieve the transition between flight strips; For quadrotor drones, when the turn is a right angle + right angle, the minimum turning radius is not considered and the drone can turn directly. When the turn is an acute angle + obtuse angle, a vertical line is introduced at waypoint 2 to simplify the acute angle + obtuse angle turn to a right angle + right angle turn. Step 6C.4: Calculate the route set for each UAV and the time it takes for any UAV k to perform the coverage task; Create a graph G = (V, E) associated with the area to be covered. The coverage problem can be formulated as a vehicle routing problem. Solve the vehicle routing problem to obtain a set of routes for each drone. Then calculate the time Tk it takes for drone k to perform the coverage task. Step 6C.5: Optimize the route set of each UAV obtained in step 6C.4 using the improved genetic algorithm; The obstacle information (obstacle height and / or interference signal, which can be obtained by map or conventional measurement equipment), the entry point determined in step 5C, the initial path and waypoints in step 6C.1, the flight path spacing in step 6C.2, the minimum turning radius of the UAV, the turning strategy in step 6C.3, the route set in step 6C.4, and the time T spent by UAV k to perform the coverage task are recorded. k The improved genetic algorithm is input together with the time T k As the goal, optimize the route set and output time T k The smallest set of optimal paths.

8. The multi-UAV collaborative detection method for urban approach operations according to claim 7 is characterized by: Steps 7A, 7B, and 7C are the same. They all use the improved A* algorithm to generate the return path of the UAV from the detection area to the recovery area, and use the time collision avoidance strategy during the return process to assign a return time T to each UAV. j , ensuring that the UAV with a shorter mission path returns first; the return time T j Calculate according to the following formula: T j =(1-g i β′)T Among them, T j is the time interval from when UAV j takes off from the assembly point to when it is required to return, T is the battery life of UAV j, g i For the return order, the optimal path obtained based on the improved genetic algorithm is used to sort each UAV from small to large according to the mission path length, and the values are assigned to 0, 1, …, n-1, where n is the total number of UAVs; β′ is taken as 0.1-0.

5.

9. The multi-UAV collaborative detection method for urban approach operations according to any one of claims 1 to 8, characterized in that: The method for optimizing the initial population using the grid potential evaluation method is: Divide the UAV coverage area into grid cells. According to the determined grid size, the entire flight area is divided into uniform grid cells to form a two-dimensional grid matrix. Initially, all grid potentials are assigned to 0. Each time a UAV visits one of the grids, the grid potential value of the grid is increased by 1. Compare the grid potentials of all grids around the current grid position of the drone, and select one or more cells with the smallest grid potential as the candidate grid for the next waypoint or the grid for the next waypoint; When there is only one grid with the smallest grid potential around it, the next path point is generated; when there are multiple grids with the smallest grid potential, the next unique path point can be generated according to different grid patterns; The grid mode is selected based on the slope and direction of the candidate grids. First, the grid cells in the candidate grid set are sorted by slope size, and the one with the gentlest slope is selected as the next path point. If there are cases where the slopes are the same, the grid cell with the slope direction closest to the current drone movement direction is selected as the next grid cell to be reached.

10. The multi-UAV collaborative detection method for urban approach operations according to claim 9, characterized in that: The convex decomposition adopts a trapezoidal decomposition method, which uses a sweep line algorithm to treat each vertex in the detection area as an event. The sweep line is a vertical line that sweeps the detection area from left to right. When encountering an event, the sweep line extends the ray upward and downward until it intersects with the edges of the detection area above and below the event, forming a trapezoidal sub-area at the event according to the event type. After the scan line completes the rightmost event, the trapezoidal decomposition is complete.

Images