A method for planning a flight path of a cluster of unmanned aerial vehicles (uavs) for forest fire edge detection

Abstract

The application provides a kind of unmanned aerial vehicle cluster path planning method for forest fire edge, first obtain the quasi-forest fire edge graph after edge expansion, and calculate to obtain scanning range convex polygon, the distance of discrete point in fire area to each side of scanning range convex polygon is solved, the shortest distance is regarded as the importance index of the discrete point, and according to the index, draw the equal importance line graph, and the different regions in the graph are weighted, the spacing of adjacent path of unmanned aerial vehicle scanning in the region is determined according to the weight of region, to improve the lateral overlap rate of edge region unmanned aerial vehicle scanning, so as to improve the aerial image precision of unmanned aerial vehicle;Then construct multi-unmanned aerial vehicle scanning range distribution function and unmanned aerial vehicle three-dimensional path cost function, the optimal position of range segmentation ray and the best three-dimensional path planning are solved by using particle swarm optimization algorithm and genetic simulated annealing algorithm.The application is reliable and practical, and has good adaptability for unmanned aerial vehicle cluster path planning in forest fire scene.

Description

Technical Field

[0001] This invention relates to the fields of forest fire monitoring and UAV trajectory planning, specifically to a method for planning the trajectory of UAV swarms that emphasizes the edge of forest fires. Background Technology

[0002] When a fire occurs, obtaining information about the fire situation and accurately assessing the intensity and scale of the fire are crucial for providing decision-making support to command and decision-making bodies and for timely and effective control and extinguishing of the fire.

[0003] Currently, drones have effectively met these needs. Equipped with cameras, high-resolution cameras, forward-looking infrared (FLIR) sensors, and image transmission equipment, drones can scan and cover target areas, transmitting video images or storing high-definition photos in real time for processing upon return to the ground. Compared to aircraft, drones are cost-effective and highly suitable for high-risk tasks such as monitoring forest fires. Drone aerial monitoring has become a primary means of obtaining fire information during forest fires. However, the current level of intelligence in drone operation for forest fire monitoring is not high. Due to significant influence from the flight environment, crashes frequently occur during operation, necessitating an intelligent method to replace manual control. To achieve autonomous flight of drones, it is essential to rely on drone trajectory planning to guide them in avoiding obstacles, maximizing time and energy savings while efficiently completing tasks.

[0004] Most existing trajectory planning algorithms only complete the trajectory planning for a single UAV, while forest fire monitoring tasks typically require the collaboration of multiple UAVs. However, existing multi-UAV trajectory planning algorithms do not consider the varying importance of information in different areas during fire information acquisition and the continuous spread of forest fires. In command and decision-making, acquiring information on the movement and changes at the fire edge is crucial, directly impacting the lives of firefighters and trapped personnel. Furthermore, temperature, humidity, and gas-related information in the fire area near the fire edge can be used to predict fire spread; this information is more important than information from the inner fire area. Additionally, due to the continuous spread of fires, the scanning range is often set too small, failing to acquire fire edge information, or too large, reducing efficiency. Therefore, using existing UAV swarm trajectory planning algorithms for fire monitoring results in low efficiency, insufficient edge information acquisition, and low accuracy in fire prediction and prediction.

[0005] Therefore, there is an urgent need for a drone swarm trajectory planning method that takes into account both the importance of the edge of forest fires and the ability to comprehensively and efficiently acquire information on the fire situation. Summary of the Invention

[0006] This invention proposes a method for planning the flight paths of UAV swarms that emphasizes the edge of forest fires, aiming to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for planning the flight paths of unmanned aerial vehicle (UAV) swarms that emphasizes the edge of forest fires, comprising the following steps:

[0008] S1. Acquire aerial images of the forest fire area, use the aerial images to obtain an approximate forest fire edge map, and expand the edge by a certain distance to obtain an expanded simulated forest fire edge map.

[0009] S2, select K1 specific points on the simulated forest fire edge map, use the Melkman algorithm to find convex polygons for these K1 points, and use a range simplification algorithm based on the outward expansion distance to reduce the range of the convex polygons;

[0010] S3, rasterize the simulated forest fire edge map, calculate the distance from each discrete point to each edge line of the convex polygon, and take the shortest distance as the importance index of the discrete point.

[0011] S4. Find the point with the highest importance index, divide its corresponding importance index into K2 equal parts, draw an equal importance line graph, and divide the fire area into K2 closed graphs with different importance weights.

[0012] S5: Several segmentation rays radiate from the point with the highest importance index to divide the scanning range. The minimum circumscribed triangle of each segmented scanning range is obtained by using the rotating caliper method, and importance weights are assigned to them.

[0013] The specific implementation of step S5 includes:

[0014] S501, find the point M with the highest importance index, radiate a rays from point M, divide the convex polygon in step S2 into a parts, and form a convex hull with two adjacent rays a1, a2 and the edge line.

[0015] S502, the minimum circumscribed triangle of the convex hull is obtained by using the rotating caliper method. This triangle is composed of rays a1, a2 and a base edge, which serves as the scanning range of a single UAV.

[0016] S503, translate the bottom edge to other lines of equal importance to form a closed region divided into multiple levels, while using the weights from step S4. The closed region consists of two dividing rays, the bottom edge, and the translated bottom edge.

[0017] S6, set the spacing between adjacent UAV tracks corresponding to each closed shape according to the importance weight of each closed shape;

[0018] S7. Construct a multi-UAV scanning range allocation function, use particle swarm optimization algorithm to calculate the global optimal position of the segmented ray, and obtain the scanning area of ​​each UAV.

[0019] S8 constructs the cost function of the UAV's three-dimensional trajectory and uses the genetic simulated annealing algorithm to obtain the optimal three-dimensional trajectory plan.

[0020] Furthermore, the specific implementation of step S1 includes:

[0021] S101, Obtain existing environmental information in the fire area, including digital surface model and GPS positioning;

[0022] S102: Use drones or helicopters to fly high above the fire and take images of the overall fire area.

[0023] S103, an image edge smoothing algorithm is used to obtain an approximate forest fire edge map from the overall fire range image;

[0024] S104, GPS positioning was used to obtain the location information of the edge point of the forest fire;

[0025] S105, expand the forest fire edge map by a distance E to obtain the expanded simulated forest fire edge map.

[0026] Furthermore, the mathematical model for determining the outward expansion distance E in step S105 is as follows:

[0027]

[0028] Where maxD is the maximum distance between two points on the forest fire edge map, i.e., the distance between the two points furthest apart on the forest fire edge map, minD is the minimum distance between two points on the forest fire edge map, and r is a given value.

[0029] Furthermore, the specific implementation of step S2 includes:

[0030] S201, K1 takes the value of 8. Find these 8 specific points, which are the x-coordinate extreme points a and b, y-coordinate extreme points c and d on the simulated forest fire edge map, and the x-coordinate extreme points e and f and y-coordinate extreme points j and k after rotating the image clockwise by 45 degrees.

[0031] S202, use the Melkman algorithm to find the convex polygon formed by these 8 azimuth points;

[0032] S203 uses a range simplification algorithm based on the outward expansion distance to generate a convex polygon with a reduced range.

[0033] Furthermore, the specific implementation of step S3 includes:

[0034] S301, rasterize the simulated forest fire edge map to obtain the coordinates of each discrete point (x). i y i );

[0035] S302, assuming the number of edge points of the convex polygon is n, calculate the distance P[i](x) from each discrete point to each endpoint of the edge line of the convex polygon. pi ,y pi The distance M i ;

[0036] S303, calculate the distance N from each discrete point to each edge line of the convex polygon. i ;

[0037] S304, N i The shortest distance in the interval is used as the importance index I of the discrete point.

[0038] Furthermore, the specific implementation of step S4 includes:

[0039] S401, find the point M with the highest importance index, that is, the discrete point corresponding to the maximum importance index value. The importance index of this discrete point is Imax.

[0040] S402, K2 is an integer greater than five, which is set to 10 here. Therefore, ten lines of equal importance are formed by connecting discrete points with the same importance I=0, Imax / 10, Imax / 5, 3Imax / 10, 2Imax / 5, Imax / 2, 3Imax / 5, 7Imax / 10, 4Imax / 5, and 9Imax / 10, which divide the fire area into ten closed parts.

[0041] S403 assigns weights to the nth region counting inwards from the edge using an importance weighting function. 0 <n<11,

[0042]

[0043] Where e is the natural logarithm.

[0044] Furthermore, in step S6, the mathematical model for calculating the distance between adjacent UAV tracks in the nth closed region is as follows:

[0045]

[0046] Where N is the maximum distance between adjacent UAV tracks as originally defined, and Distant(n) is the distance between adjacent UAV tracks in the nth closed region, counting inwards from the edge. This represents the importance weight of the nth closed region, calculated from the edge inwards.

[0047] Furthermore, the multi-UAV scanning range allocation function in step S7 is:

[0048]

[0049] Where j1 and j2 are weighting coefficients, representing the emphasis between the sum of all UAV flight distances and the balance of flight distances of individual UAVs, j1 + j2 = 1, Assign represents the inferiority of scan range allocation, S 总 The total distance traveled by all drones is defined as:

[0050]

[0051] in, a S represents the number of available drones. i Let be the flight path of the i-th drone;

[0052] MAD is the average absolute deviation of the flight path of each UAV, defined as:

[0053]

[0054] in, for a The average flight distance of each drone is represented by MAD, which reflects the similarity in flight distances among the drones.

[0055] Furthermore, the cost function for the UAV's three-dimensional trajectory in step S8 is:

[0056]

[0057] Where N represents the total number of tracks, f i h represents the length of the i-th segment of the trajectory. i Let w1 and w2 represent the altitude of the i-th segment of the flight path, and w1 and w2 are weighting coefficients, representing the emphasis on flight distance and flight altitude, respectively, and w1 + w2 = 1.

[0058] This invention provides a drone swarm trajectory planning method that emphasizes the edge of forest fires. It utilizes various data sources, including digital surface models, GPS positioning, and high-altitude imagery of the fire area, to construct an environmental model of the drone scanning region. Considering the continuous outward expansion of forest fires, the Melkman algorithm is used to calculate the convex polygon of the scanning range. A range simplification algorithm based on the outward expansion distance is employed to make the scanning range closer to the actual extent of the forest fire. Finally, the rotating caliper method is used to find the minimum circumscribed triangle of the scanning range for a single drone, expanding the scanning range to a certain extent and avoiding the problem of missing fire edge information due to fire expansion. Simultaneously, considering the importance of fire edge region information, the distances from discrete points in the fire area to each side of the convex polygon of the scanning range are calculated. The shortest distance is used as the importance index for that discrete point, and an equal importance line graph is drawn based on this index. Different regions in the graph are weighted, and the spacing between adjacent drone scanning tracks in that region is determined by the region weight to improve the lateral overlap rate of drone scanning in the edge region, thereby improving the accuracy of drone aerial imagery. Furthermore, extending the information collection time of drone sensors near the fire edge region further improves the accuracy of fire spread prediction. Then, a multi-UAV scanning range allocation function and a UAV 3D trajectory cost function are constructed. Particle swarm optimization and genetic simulated annealing algorithms are used to solve for the optimal position of the range segmentation ray and the best 3D trajectory planning. This invention fully considers the characteristics of continuous outward expansion of forest fires and the greater importance of information in the fire edge area. In practical applications, this method can efficiently generate UAV swarm trajectory planning schemes in a short time, comprehensively and accurately covering the forest fire area while collecting high-precision information on the fire edge area, further improving the accuracy of fire spread prediction and providing strong support for forest fire emergency command and dispatch.

[0059] In summary, this invention is reliable and practical, and has good adaptability to drone swarm trajectory planning in forest fire scenarios. It has taken relevant measures to reasonably set the scanning range and improve the accuracy of information collection in the fire edge area, and has achieved excellent results. It has good practicality and feasibility. Attached Figure Description

[0060] Figure 1 This is a schematic diagram showing the outward expansion of the edge of a forest fire.

[0061] Figure 2 The flowchart shows how to calculate the convex polygon of the scan range using the Melkman algorithm and then reduce the range of the convex polygon using a range simplification algorithm based on the outward expansion distance. Figure 2 (a)-(h) illustrate the process of calculating the convex polygon of the scan range using the Melkman algorithm. Figure 2 (i)-(j) demonstrate the process of reducing the extent of a convex polygon using an extent simplification algorithm based on the outward extension distance.

[0062] Figure 3 This is a line graph showing the degree of equal importance.

[0063] Figure 4 This is a schematic diagram of the scan range segmentation.

[0064] Figure 5 This is a schematic diagram of an S-shaped scan of the target area, with the track spacing decreasing from the inside out.

[0065] Figure 6 This is a flowchart of an embodiment of the present invention. Specific Implementation

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the present invention will be further described below with reference to the accompanying drawings of the embodiments.

[0068] The embodiment provides a method for planning the flight paths of UAV swarms that emphasizes the edge of forest fires, specifically employing the following technical solutions:

[0069] S1: Acquire aerial images of the forest fire area, use the aerial images to obtain an approximate forest fire edge map, and expand the edge by a certain distance to obtain an expanded simulated forest fire edge map.

[0070] S2: Select K1 specific points on the simulated forest fire edge map, use the Melkman algorithm to find convex polygons for these K1 points, and use a range simplification algorithm based on the outward expansion distance to reduce the range of the convex polygons;

[0071] S3: Rasterize the simulated forest fire edge map, calculate the distance from each discrete point to each edge line of the convex polygon, and take the shortest distance as the importance index of the discrete point.

[0072] S4: Find the point with the highest importance index, divide its corresponding importance index into K2 equal parts, draw an equal importance line graph, and divide the fire area into K2 closed graphs with different importance weights;

[0073] S5: The scanning range is divided into several segments radiating from the point with the highest importance index. The minimum circumscribed triangle of each segmented scanning range is obtained by using the rotating caliper method, and importance weights are assigned to them.

[0074] Step S5 includes:

[0075] S501: A rays radiate from point M, dividing the convex polygon in step S2 into a parts. Adjacent rays a1 and a2 and the edge line form a convex hull.

[0076] S502: The minimum circumscribed triangle of the convex hull is obtained by using the rotating caliper method. This triangle is composed of rays a1, a2 and a base edge, which serves as the scanning range of a single UAV.

[0077] Using the smallest circumscribed triangle as the final scanning range allows the range to be expanded to a certain extent, which is consistent with the continuous expansion of forest fires and avoids the problem of setting the scanning range too small. At the same time, the UAV can perform an S-shaped scan parallel to the base of the smallest circumscribed triangle, thereby enhancing the acquisition of edge information by increasing the lateral overlap rate of the UAV track in the edge area.

[0078] S503: Translate the bottom edge to the other 9 lines of equal importance, forming a closed region divided into ten levels, while inheriting the previously calculated weights. The closed region consists of two dividing rays, the bottom edge, and the translated bottom edge. Figure 4 It is by Figure 3 (Originated from an iso-importance line plot) Figure 4 The outermost edge is the base of each smallest circumscribed triangle. These bases are then translated inwards to the lines of equal importance inside, thus forming... Figure 4 . Figure 4 It consists of five large triangles, one of which is divided into ten parts, one of which is a closed area of ​​a hierarchy.

[0079] S6: Set the spacing between adjacent UAV tracks for each closed region according to the importance weight of each closed region;

[0080] S7: Construct a multi-UAV scanning range allocation function, use particle swarm optimization algorithm to calculate the global optimal position of the segmented ray, and obtain the scanning area of ​​each UAV;

[0081] S8: Construct the cost function of the UAV's 3D trajectory and use the genetic simulated annealing algorithm to obtain the optimal 3D trajectory plan.

[0082] In one implementation, step S1 includes:

[0083] S101: Obtain existing environmental information in the fire area, including digital surface models and GPS positioning;

[0084] S102: Use drones or helicopters to fly high above the fire and take images of the overall fire area;

[0085] Due to low visibility over the fire site, aerial photography was poor, resulting in only a rough image of the overall fire area.

[0086] S103: Use an image edge smoothing algorithm to obtain an approximate forest fire edge map from the overall fire range image;

[0087] S104: Use GPS positioning to obtain the location information of the edge point of the forest fire;

[0088] S105: Expand the forest fire edge map by a distance E to obtain the expanded simulated forest fire edge map.

[0089] Furthermore, the mathematical model for determining the outward expansion distance E in step S105 is as follows:

[0090]

[0091] Where maxD is the maximum distance between two points on the edge map of a forest fire, that is, the distance between the two points furthest apart on the edge map of a forest fire, minD is the minimum distance between two points on the edge map of a forest fire, and r is a given value determined by the intensity of the fire. The stronger the fire, the smaller r is. The default value is 10.

[0092] Figure 1 It shows the process of the forest fire's edge spreading outwards.

[0093] In one implementation, step S2 includes:

[0094] S201: K1 takes the value of 8. Find these 8 specific points, which are the x-coordinate extreme points a and b, the y-coordinate extreme points c and d on the simulated forest fire edge map, the x-coordinate extreme points e and f, and the y-coordinate extreme points j and k after rotating the image 45 degrees clockwise.

[0095] S202: Use the Melkman algorithm to find the convex polygon formed by these 8 azimuth points;

[0096] To make the scanning range closer to the actual range of forest fires, while reducing the outer expansion range, speeding up the calculation, and improving the efficiency of the method, especially to improve the speed of calculating the importance index of discrete points in step S3, it is necessary to simplify the range of the convex polygon.

[0097] S203: Generate a convex polygon with a reduced range using a range simplification algorithm based on the outward expansion distance.

[0098] Furthermore, the specific steps of the Melkman algorithm in step S202 are as follows:

[0099] S2021: Input a simple polyline P[i] with 8 vertices. Add any three initial points from the 8 points to the queue D[], such that P[2] is at the top and bottom of D[], and in D[], P0, P1, and P2 form a counterclockwise triangle;

[0100] S2022: Process each point after i=2 in sequence. For P[i], check whether P[i] is inside D. If P[i] is to the left of the directed line segment D[bot]D[bot+1] and the directed line segment D[top-1]D[top], then it is determined that P[i] is inside D. Skip P[i] and continue to process the next point.

[0101] S2023: When P[i] is to the right of the directed line segment D[bot]D[bot+1], delete D[bot] and make P[i] the bottom of the double-ended queue D[]. When P[i] is to the right of the directed line segment D[top-1]D[top], delete D[top] and make P[i] the top of the double-ended queue D[].

[0102] S2024: When all points have been processed, D[] is the final convex polygon result.

[0103] Figure 2 (a)-(h) in the diagram illustrate the process of calculating the convex polygon of the scan range using the Melkman algorithm.

[0104] To make the final scanning range closer to the actual range of the forest fire, while reducing the outer expansion range, speeding up the calculation, and improving the efficiency of the method, especially to improve the speed of calculating the importance index of discrete points in step S3, it is necessary to simplify the range of the convex polygon.

[0105] Furthermore, the specific steps of the range simplification algorithm based on the outward expansion distance E in step S203 are as follows:

[0106] S2031: Find the distance d[i] from point P[i] in the dequeue D[] to line segment P[i+1]P[i-1];

[0107] S2032: If d[i] <= E, then determine that the triangle is located in the outer expansion range, and delete P[i] from D[].

[0108] S2033: Traverse all points in D[] to obtain the simplified convex polygon range;

[0109] In one implementation, the mathematical model for calculating the distance d[i] in step S2031 is as follows:

[0110]

[0111] Where p is the half perimeter of the triangle, a is the length of line segment P[i+1]P[i-1], and b and c are the lengths of the other two sides of the triangle.

[0112] Figure 2(i)-(j) in the diagram illustrate the process of reducing the range of a convex polygon using a range simplification algorithm based on the outward expansion distance.

[0113] In one embodiment, step S3 includes:

[0114] S301: Rasterize the image to obtain the coordinates of each discrete point (x). i y i );

[0115] S302: Assuming the number of edge points of a convex polygon is n, calculate the distance P[i](x) from each discrete point to each endpoint of the edge line of the convex polygon. pi ,y pi The distance M i ;

[0116] S303: Calculate the distance N from each discrete point to each edge line of the convex polygon. i ;

[0117] S304: N i The shortest distance in the interval is used as the importance index I of the discrete point.

[0118] Furthermore, in S302, the formula for calculating the distance from each discrete point to the endpoints of each edge line of the convex polygon is:

[0119]

[0120] Furthermore, the formula for calculating the distance from a discrete point to the edge line in S303 is as follows:

[0121]

[0122] Where a is the distance between line segments P[i] and P[i+1], and b is M. i c is M i+1 Let p be the semi-perimeter of the triangle formed by points P[i], P[i+1] and the discrete point.

[0123] Furthermore, in S304, the importance index of discrete points is I = min{N1, N2…N…} n}

[0124] In one implementation, step S4 includes:

[0125] S401: Find the point M with the highest importance index, that is, the discrete point corresponding to the maximum importance index value. The importance index of this discrete point is Imax.

[0126] S402: K2 is an integer greater than five, which is set to 10 here. Therefore, ten lines of equal importance are formed by connecting discrete points with the same importance I=0, Imax / 10, Imax / 5, 3Imax / 10, 2Imax / 5, Imax / 2, 3Imax / 5, 7Imax / 10, 4Imax / 5, and 9Imax / 10, thus dividing the fire area into ten closed parts.

[0127] Among them, the equal importance line diagram is as follows Figure 3 As shown;

[0128] S403: Assign weight P to the nth region counting inwards from the edge using an importance weighting function. n (0) <n<11)。

[0129] Furthermore, the importance weighting function in step S402 is:

[0130]

[0131] Where e is the natural logarithm.

[0132] In one implementation, the mathematical model for calculating the distance between adjacent UAV tracks in step S6 is as follows:

[0133]

[0134] Where N is the maximum distance between adjacent UAV tracks as originally defined, and Distant(n) is the distance between adjacent UAV tracks in the nth closed region, counting inwards from the edge. This represents the importance weight of the nth closed region, calculated from the edge inwards.

[0135] By reducing the distance between adjacent flight paths of drones at the edge of a fire, the lateral overlap rate can be increased, thereby improving the accuracy of drone aerial imagery. At the same time, the information collection time of drone sensors near the edge of a fire can be extended, further improving the accuracy of fire spread prediction.

[0136] like Figure 5 As shown, an S-shaped scan is used on the target area, with the track spacing gradually decreasing from the inside to the outside.

[0137] Specifically, the number of rays 'a' in step S6 is the number of available drones, and the drones are equipped with gas sensors, temperature sensors, and humidity sensors.

[0138] In one implementation, the multi-UAV scanning range allocation function in step S7 is:

[0139]

[0140] Where j1 and j2 are weighting coefficients, representing the emphasis between the sum of all UAV flight distances and the balance of flight distances among individual UAVs, j1 + j2 = 1. Assign represents the poorness of scan range allocation; the smaller the Assign, the better the scan range allocation. 总 The total distance traveled by all drones is defined as:

[0141]

[0142] in, a S represents the number of available drones. i Let be the flight path of the i-th drone;

[0143] MAD is the average absolute deviation of the flight path of each UAV, defined as:

[0144]

[0145] in, for a The MAD (mean squared distance) is the average distance traveled by each drone. It is used to reflect how similar the distances traveled by each drone are. The smaller the MAD, the closer the distances traveled by each drone are and the closer the completion times are.

[0146] In one implementation, the cost function of the UAV's three-dimensional trajectory in step S8 is:

[0147]

[0148] Where N represents the total number of tracks, f i h represents the length of the i-th segment of the trajectory. i Let w1 and w2 represent the altitude of the i-th segment of the flight path, and w1 and w2 are weighting coefficients, representing the emphasis on flight distance and flight altitude, respectively, and w1 + w2 = 1.

[0149] The specific embodiments mentioned in this invention are merely illustrative examples of the method steps of this invention. Those skilled in the art can make corresponding modifications or additions to the specific implementation steps mentioned, but any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A method for planning the flight paths of unmanned aerial vehicle (UAV) swarms, emphasizing the edge of forest fires, characterized in that... Includes the following steps: S1. Acquire aerial images of the forest fire area, use the aerial images to obtain an approximate forest fire edge map, and expand the edge by a certain distance to obtain an expanded simulated forest fire edge map. S2, select K1 specific points on the simulated forest fire edge map, use the Melkman algorithm to find convex polygons for these K1 points, and use a range simplification algorithm based on the outward expansion distance to reduce the range of the convex polygons; S3, rasterize the simulated forest fire edge map, calculate the distance from each discrete point to each edge line of the convex polygon, and take the shortest distance as the importance index of the discrete point. S4. Find the point with the highest importance index, divide its corresponding importance index into K2 equal parts, draw an equal importance line graph, and divide the fire area into K2 closed graphs with different importance weights. S5: Several segmentation rays radiate from the point with the highest importance index to divide the scanning range. The minimum circumscribed triangle of each segmented scanning range is obtained by using the rotating caliper method, and importance weights are assigned to them. The specific implementation of step S5 includes: S501, find the point M with the highest importance index, radiate a rays from point M, divide the convex polygon in step S2 into a parts, and form a convex hull with two adjacent rays a1, a2 and the edge line. S502, the minimum circumscribed triangle of the convex hull is obtained by using the rotating caliper method. This triangle is composed of rays a1, a2 and a base edge, which serves as the scanning range of a single UAV. S503, translate the bottom edge to other lines of equal importance to form a closed region divided into multiple levels, while using the weights from step S4. The closed region consists of two dividing rays, the bottom edge, and the translated bottom edge. S6, set the spacing between adjacent UAV tracks corresponding to each closed shape according to the importance weight of each closed shape; S7. Construct a multi-UAV scanning range allocation function, use particle swarm optimization algorithm to calculate the global optimal position of the segmented ray, and obtain the scanning area of ​​each UAV. S8 constructs the cost function of the UAV's three-dimensional trajectory and uses the genetic simulated annealing algorithm to obtain the optimal three-dimensional trajectory plan.

2. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: The specific implementation of step S1 includes: S101, Obtain existing environmental information in the fire area, including digital surface model and GPS positioning; S102: Use drones or helicopters to fly high above the fire and take images of the overall fire area. S103, an image edge smoothing algorithm is used to obtain an approximate forest fire edge map from the overall fire range image; S104, GPS positioning was used to obtain the location information of the edge point of the forest fire; S105, expand the forest fire edge map by a distance E to obtain the expanded simulated forest fire edge map.

3. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 2, characterized in that: The mathematical model for determining the outward expansion distance E in step S105 is as follows: Where maxD is the maximum distance between two points on the forest fire edge map, i.e., the distance between the two points furthest apart on the forest fire edge map, minD is the minimum distance between two points on the forest fire edge map, and r is a given value.

4. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: The specific implementation of step S2 includes: S201, K1 takes the value of 8. Find these 8 specific points, which are the x-coordinate extreme points a and b, y-coordinate extreme points c and d on the simulated forest fire edge map, and the x-coordinate extreme points e and f and y-coordinate extreme points j and k after rotating the image clockwise by 45 degrees. S202, use the Melkman algorithm to find the convex polygon formed by these 8 azimuth points; S203 uses a range simplification algorithm based on the outward expansion distance to generate a convex polygon with a reduced range.

5. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: The specific implementation of step S3 includes: S301, rasterize the simulated forest fire edge map to obtain the coordinates of each discrete point (x). i y i ); S302, assuming the number of edge points of the convex polygon is n, calculate the distance P[i](x) from each discrete point to each endpoint of the edge line of the convex polygon. pi ,y pi The distance M i ; S303, calculate the distance N from each discrete point to each edge line of the convex polygon. i ; S304, N i The shortest distance in the interval is used as the importance index I of the discrete point.

6. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: The specific implementation of step S4 includes: S401, find the point M with the highest importance index, that is, the discrete point corresponding to the maximum importance index value. The importance index of this discrete point is Imax. S402, K2 is an integer greater than five, which is set to 10 here. Therefore, ten lines of equal importance are formed by connecting discrete points with the same importance I=0, Imax / 10, Imax / 5, 3Imax / 10, 2Imax / 5, Imax / 2, 3Imax / 5, 7Imax / 10, 4Imax / 5, and 9Imax / 10, which divide the fire area into ten closed parts. S403 assigns weights to the nth region counting inwards from the edge using an importance weighting function. 0 <n<11, Where e is the natural logarithm.

7. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: In step S6, the mathematical model for calculating the distance between adjacent UAV tracks in the nth closed region is as follows: Where N is the maximum distance between adjacent UAV tracks as originally defined, and Distant(n) is the distance between adjacent UAV tracks in the nth closed region, counting inwards from the edge. This represents the importance weight of the nth closed region, calculated from the edge inwards.

8. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: The multi-UAV scanning range allocation function in step S7 is: Where j1 and j2 are weighting coefficients, representing the emphasis between the sum of all UAV flight distances and the balance of flight distances of individual UAVs, j1 + j2 = 1, Assign represents the inferiority of scan range allocation, S 总 The total distance traveled by all drones is defined as: in, a S represents the number of available drones. i Let be the flight path of the i-th drone; MAD is the average absolute deviation of the flight path of each UAV, defined as: in, for a The average flight distance of each drone is represented by MAD, which reflects the similarity in flight distances among the drones.

9. The method for planning drone swarm paths that emphasizes the edge of forest fires as described in claim 1, characterized in that: The cost function for the UAV's 3D trajectory in step S8 is: Where N represents the total number of tracks, f i h represents the length of the i-th segment of the trajectory. i Let w1 and w2 represent the altitude of the i-th segment of the flight path, and w1 and w2 are weighting coefficients, representing the emphasis on flight distance and flight altitude, respectively, and w1 + w2 = 1.

Images