Parawing system homing drop point planning method based on terrain and ground features

Abstract

The invention discloses a parafoil system homing drop point planning method based on terrain and ground features. The method comprises the following steps: screening a nominal landing point based on a terrain analysis model; screening out a plurality of two-dimensional feature points from all nominal landing points; sorting all the two-dimensional feature points to obtain an optimal search sequence; obtaining the height of the piloted parafoil passing through each two-dimensional feature point based on time estimation, and obtaining a three-dimensional space path point; carrying out segmented trajectory planning on an air-drop trajectory between every two adjacent three-dimensional space path points by taking the lowest energy consumption as an optimization target, and integrating trajectory planning results of the piloted parafoil; acquiring a ground object image returned by the piloted parafoil; and carrying out ground feature classification on the ground feature image based on the ground feature classification model, and after all nominal landing points which are not suitable for landing are eliminated, outputting the remaining nominal landing points. The method is applied to the field of trajectory planning, the energy consumption of the piloted parafoil can be effectively reduced, and qualitative and quantitative analysis and evaluation can be carried out on a large-range landing preselected area.

Description

Technical Field

[0001] The invention relates to the technical field of trajectory planning, and in particular to a method for planning a homing point for a parafoil system based on terrain and objects. Background Art

[0002] In an airdrop mission, due to ground communication obstruction, changes in or loss of terrain information in the target area, the dropped materials may be lost, personnel may be injured, and the mission may fail. Therefore, the selection of a suitable landing point is the primary task of an airdrop project, and terrain safety is one of the key factors to be considered. To this end, it is necessary to obtain terrain information of the target area and screen the landing point in advance.

[0003] The existing landing point selection methods are mainly divided into two categories. One is the online selection method during the landing process of the aircraft. During the landing process, the aircraft uses sensors such as lidar and cameras to obtain data such as elevation and landing point images, and then calculates the optimal landing point in real time through the design algorithm. There are shortcomings in this type of method: the pure online selection method has high requirements for the amount of image data and large amount of calculation, and it is impossible to get the landing point in time when the landing area is too large; the offline selection before flight selects the macro area by statistical and distribution analysis of the terrain, but lacks quantitative analysis of the terrain and cannot respond to sudden and drastic changes in the terrain in time. Therefore, the application of image recognition technology to obtain terrain information in real time is a necessary means for airdrop and landing missions. The landing point is pre-screened through the existing satellite map, and then the online selection method is used to verify the feasibility of the selected landing point, so as to take into account both real-time and high efficiency.

[0004] The "pilot" parafoil is a parafoil that is deployed first for terrain exploration, wind field measurement and other advance tasks before subsequent large-scale airdrops on high-value targets. It can obtain the environmental characteristics of the target area in real time, and provide reference situation information and homing routes for cluster deployment. For the homing trajectory planning of the pilot parafoil system, the current mainstream methods include the simple homing method, the optimal control homing method based on controlling the minimum energy, and the segmented homing method. The optimal control homing method has low energy consumption, but the control law changes continuously, and the implementation is difficult. Although the conventional segmented homing method is relatively simple to implement, it has relatively high energy consumption. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a homing landing point planning method for a parafoil system based on terrain and landforms, which can not only effectively reduce the energy consumption of the pilot parafoil, but also perform qualitative and quantitative analysis and evaluation on a large range of pre-selected landing areas.

[0006] To achieve the above object, the present invention provides a method for planning a homing point of a parafoil system based on terrain and features, comprising the following steps:

[0007] Step 1: Analyze the digital elevation map of the pre-landing area based on the terrain analysis model to screen out all nominal landing points with terrain undulations suitable for parafoil landing;

[0008] Step 2, select a number of two-dimensional feature points from all nominal landing points, and the selection rule is: when the pilot parafoil passes through all the two-dimensional feature points, all the nominal landing points can be scanned;

[0009] Step 3, sort all the two-dimensional feature points to obtain the optimal search order;

[0010] Step 4, based on the optimal search sequence, obtain the height of the pilot parafoil when it passes through each two-dimensional feature point based on time estimation, and obtain a three-dimensional space path point corresponding to each two-dimensional feature point;

[0011] Step 5, segmenting the airdrop trajectory between each two adjacent three-dimensional space path points, and taking the minimum energy consumption as the optimization goal, planning each segmented trajectory under the premise of satisfying the constraint conditions, and integrating the trajectory planning results of the pilot parafoil;

[0012] Step 6, controlling the pilot parafoil to scan the surface objects according to the trajectory planning result, and obtaining the image of the objects transmitted back by the pilot parafoil;

[0013] Step 7: classify the object image based on the object classification model, eliminate all nominal landing points that are not suitable for landing, and output the remaining nominal landing points.

[0014] In one embodiment, in step 1, the terrain analysis model uses an inverse pyramid landing point selection algorithm to screen all nominal landing points;

[0015] Among them, the inverse pyramid landing point selection algorithm searches the pre-landing area in a progressive order of flat point search, safe landing point search, reliable obstacle avoidance point search, safe landing area and nominal landing point area selection.

[0016] In one embodiment, in step 2, the step of selecting a plurality of two-dimensional feature points from all nominal landing points is as follows:

[0017] Step 201, adding all nominal landing points to the point set;

[0018] Step 202, randomly selecting a point in the point set as a two-dimensional feature point, and drawing a circle with the selected two-dimensional feature point as the center and r as the radius;

[0019] Step 203, after deleting all points in the circle made in step 202 from the point set, step 202 to step 203 are performed again until the point set is an empty set.

[0020] In one embodiment, in step 3, the optimal search order is obtained based on the simulated annealing algorithm with the goal of minimizing the sum of the distances between adjacent two-dimensional feature points in the search order, and the process is as follows:

[0021] Step 301, obtaining an initial temperature T0, an annealing temperature ΔT and a randomly set initial search order, and setting the current temperature T=T0;

[0022] Step 302, let T = T - ΔT, and calculate the sum L0 of the distances between the adjacent two-dimensional feature points in the current search order;

[0023] Step 303, randomly select two nodes in the current search order and exchange their positions to obtain a new search order, and calculate the sum L of their connection distances;

[0024] Step 304, determine whether L0>L+exp(α·(T0-T) / T0) holds, where α is a control parameter:

[0025] If yes, after taking the new search order as the current search order, proceed to step 305;

[0026] Otherwise, proceed to step 305;

[0027] Step 305, determine whether T<0 holds:

[0028] If so, take the current search order as the optimal search order and output it;

[0029] Otherwise, return to step 302.

[0030] In one embodiment, in step 3, the height of the pilot parafoil when it passes through each two-dimensional feature point is estimated based on time, specifically:

[0031]

[0032] Among them, h i 、h i-1 is the height of the pilot parafoil when it passes through the i-th and i-1-th two-dimensional feature points, τ i-1,i is the estimated time for the pilot parafoil to maneuver from the i-1th two-dimensional feature point to the i-th two-dimensional feature point, ∑τ i-1,i is the sum of all estimated times, h0 is the initial position height of the pilot parafoil, and h min ≤h0≤h max ,h min is the minimum scanning height of the pilot parafoil, h max It is the highest scanning altitude of the pilot parafoil.

[0033] In one embodiment, the estimated time for the pilot parafoil to maneuver from the i-1th two-dimensional feature point to the i-th two-dimensional feature point is specifically:

[0034]

[0035] Among them, l i-1,i is the plane distance between the i-1th two-dimensional feature point and the i-th two-dimensional feature point, V is the velocity modulus of the pilot parafoil, γ is the track angle of the pilot parafoil, Δξ i-2,i-1,i is the maneuvering angle of the line between the i-2nd and i-1st two-dimensional feature points relative to the line between the i-1st and i-th two-dimensional feature points, u max It is the maximum control amount of the pilot parafoil.

[0036] In one embodiment, the minimum scanning height and the maximum scanning height of the pilot parafoil are:

[0037] h min =v / tan(θ / 2)

[0038]

[0039] Where v is the minimum length scale, θ is the maximum field of view of the camera on the pilot parafoil, px scanner Camera resolution carried by the pilot parafoil, px model The image resolution for image recognition.

[0040] In one embodiment, in step 5, when planning the segmented trajectory, the segmented trajectory is divided into N segments in the height direction, and a control amount u is applied between each segment. ij Perform a maneuver;

[0041] The goal of planning the segmented trajectory is to obtain a set of optimal control quantities U i =[u i2 ,…u ij ,…,u iN ], so that the energy consumption is minimized, wherein the particle swarm algorithm is first used to obtain the approximate range of the optimal control quantity, and then the interior point method-obstacle penalty function method is used to obtain the optimal control quantity U within the range, which specifically includes the following steps:

[0042] Step 501, set the number of segments M, the number of segments counter p=1, the critical value of the allowable function J0, the iteration counter T=1, the upper limit of the number of particle swarm iterations T0, the global optimal weight C1 and the local optimal weight C2;

[0043] Step 502: Based on the parafoil coordinate position and direction angle, the optimization function J of the pth segment trajectory is given. p ;

[0044] Step 503: Randomly initialize the pth trajectory input Particle evolution speed Among them, k is the particle number, p is initially set to 1, and u p2 ,…,u pN The value range is [-u max ,u max ],v p2 ,…,v pN The value range is [-u max ,u max ],u max It is the maximum control amount of the pilot parafoil;

[0045] Step 504, Update renew 1, where U p0 is the optimal solution for the pth trajectory, is the local optimal solution of the kth particle in the pth trajectory;

[0046] Step 505, determine |u pj |>u max Is it true, where j = 2, 3, 4, ..., N:

[0047] If so, let u pj in [-u max ,u max ] randomly pick a value;

[0048] Otherwise, let u pj The value remains unchanged;

[0049] Step 506, determine |v pj |>u max Is it true, where j = 2, 3, 4, ..., N:

[0050] If so, let v pj in [-u max ,u max ] randomly pick a value;

[0051] Otherwise, let v pj The value remains unchanged;

[0052] Step 507, determine Is it established:

[0053] If so, then

[0054] Otherwise, make The value remains unchanged;

[0055] Step 508, determine Whether it holds:

[0056] If so, then let

[0057] Otherwise, keep the value of U p0 unchanged;

[0058] Step 509, determine whether J p (U p0 ) < J0 or T > T0 holds;

[0059] If so, then let U p = U p0 and then proceed to Step 5010;

[0060] Otherwise, return to Step 504;

[0061] Step 5010, use the Newton method to minimize J p (U p ) + μ·c(U p ), where μ is the penalty factor and c(U p ) is the out-of-bounds penalty function;

[0062] Step 5011, update p = p + 1 and determine whether p > M is satisfied:

[0063] If so, then output the control matrix U = [U1; U2; …; U M

[0064] Otherwise, return to Step 502.

[0065] In one embodiment, in Step 5, the constraint conditions include boundary constraints, headwind landing constraints, control constraints, and field-of-view angle constraints.

[0066] In one embodiment, in Step 7, the ground object classification model is composed of a trained ResNet34 network.

[0067] Compared with the prior art, the present invention has the following beneficial technical effects:

[0068] 1. The present invention addresses the problem of landing point selection during the homing process of the parafoil system, combines the impacts of terrain and ground objects on landing, and makes a dual selection of the landing point from two perspectives: terrain undulation and ground object screening, effectively avoiding the two problems that insufficient computing power leads to the inability to obtain results in a timely manner and insufficient real-time image results in drastic changes in the landing point terrain and the inability to land.

[0069] 2. The present invention can meet the requirements of real-time task processing and has high accuracy, providing a useful reference for the landing point selection of the parafoil system and the path planning of the pilot-guided parafoil based on the landing point selection. ​BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0071] Figure 1 It is a flow chart of a method for planning a homing point for a parafoil system based on terrain and features in an embodiment of the present invention;

[0072] Figure 2 Schematic diagram of the estimated maneuvering time between two-dimensional feature points in an embodiment of the present invention;

[0073] Figure 3 Schematic diagram of the viewing angle in an embodiment of the present invention;

[0074] Figure 4 It is a contour map of the pre-landing area in an embodiment of the present invention;

[0075] Figure 5 A schematic diagram of the simulation results of the terrain analysis model in an embodiment of the present invention;

[0076] Figure 6 Schematic diagram of the distribution of nominal landing points and two-dimensional feature points in an embodiment of the present invention;

[0077] Figure 7 Schematic diagram of optimization result of two-dimensional feature point search sequence in an embodiment of the present invention;

[0078] Figure 8 This is a pilot parafoil trajectory diagram in an embodiment of the present invention;

[0079] Fig. 9 It is a plane projection diagram of the trajectory of the pilot parafoil in the embodiment of the present invention;

[0080] Fig.10 This is a graph showing changes in the control amount of a pilot parafoil in an embodiment of the present invention;

[0081] Fig.11 Schematic diagram of a simulation prediction atlas in an embodiment of the present invention.

[0082] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0085] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0086] like Figure 1 The present embodiment discloses a method for planning a homing point for a parafoil system based on terrain and features, which mainly includes the following steps:

[0087] Step 1: Analyze the digital elevation map of the pre-landing area based on the terrain analysis model to screen out all nominal landing points with terrain undulations suitable for parafoil landing;

[0088] Step 2, select a number of two-dimensional feature points from all nominal landing points, and the selection rule is: when the pilot parafoil passes through all the two-dimensional feature points, all the nominal landing points can be scanned;

[0089] Step 3, sort all the two-dimensional feature points to obtain the optimal search order;

[0090] Step 4, based on the optimal search sequence, the height of the pilot parafoil when it passes through each two-dimensional feature point is obtained based on time estimation, and a three-dimensional space path point corresponding to each two-dimensional feature point is obtained;

[0091] Step 5, segment the airdrop trajectory between each two adjacent three-dimensional space path points, and plan each segmented trajectory with the minimum energy consumption as the optimization goal, while satisfying the constraint conditions, and integrate the trajectory planning results of the pilot parafoil;

[0092] Step 6, controlling the pilot parafoil to scan the surface objects according to the trajectory planning result, and obtaining the image of the objects transmitted back by the pilot parafoil;

[0093] Step 7: classify the object image based on the object classification model, eliminate all nominal landing points that are not suitable for landing, and output the remaining nominal landing points.

[0094] The soft landing capability of the parafoil cluster system mainly includes three aspects: landing point accuracy, autonomous obstacle avoidance capability, and terrain slope adaptability. Therefore, the terrain analysis model in this embodiment uses the inverse pyramid landing point selection algorithm to screen all nominal landing points. Among them, the inverse pyramid landing point selection algorithm selects the pre-landing area in the order of flat point search, safe landing point search, reliable obstacle avoidance point search, safe landing area and nominal landing area point selection. That is, first search for the flat point area in the pre-landing area, then search for the safe landing point area in the flat point area, then search for the reliable obstacle avoidance point area in the safe landing point area, and finally search for the safe landing area and the nominal landing area in the reliable obstacle avoidance point area, gradually converge from weak to strong, and gradually narrow the selection area of ​​the landing point.

[0095] Considering that the pilot parafoil has a field of view constraint when scanning downward from a high altitude, it cannot scan all nominal landing points. At the same time, if the parafoil is required to pass over each nominal landing point, it will consume too much energy. Therefore, this embodiment selects some two-dimensional feature points from all nominal landing points so that the pilot parafoil can scan the ground features of all nominal landing points in a specific area when passing through the two-dimensional feature points. The specific implementation process is as follows:

[0096] Step 201, adding all nominal landing points to the point set;

[0097] Step 202, randomly selecting a point in the point set as a two-dimensional feature point, and drawing a circle with the selected two-dimensional feature point as the center and r as the radius;

[0098] Step 203, after deleting all points in the circle made in step 202 from the point set, perform steps 202 to 203 again until the point set is an empty set.

[0099] Since there are often multiple two-dimensional feature points, the order in which the pilot parafoil searches for the two-dimensional feature points during the homing process will affect the minimum height of the parafoil and the energy required for the parafoil. In addition, since the total energy of the parafoil is limited, it is necessary to optimize the search order of all the two-dimensional feature points. In this embodiment, the optimization sorting target is to minimize the sum of the connecting line distances between the two-dimensional feature points, that is:

[0100]

[0101] In the formula, (x i ,y i ) is the plane position coordinate of the i-th two-dimensional feature point, and n is the number of two-dimensional feature points;

[0102] In the specific implementation process, this embodiment adopts the simulated annealing algorithm to obtain the optimal search sequence, and the process includes the following steps:

[0103] Step 301, obtaining an initial temperature T0, an annealing temperature ΔT and a randomly set initial search order, and setting the current temperature T=T0;

[0104] Step 302, let T = T - ΔT, and calculate the sum L0 of the distances between the adjacent two-dimensional feature points in the current search order;

[0105] Step 303, randomly select two nodes in the current search order and exchange their positions to obtain a new search order, and calculate the sum L of their connection distances;

[0106] Step 304, determine whether L0>L+exp(α·(T0-T) / T0) holds, where α is a control parameter. The specific value of the control parameter α is adjustable, for example, the value may be α=7:

[0107] If yes, after taking the new search order as the current search order, proceed to step 305;

[0108] Otherwise, proceed to step 305;

[0109] Step 305, determine whether T<0 holds:

[0110] If so, take the current search order as the optimal search order and output it;

[0111] Otherwise, return to step 302.

[0112] This embodiment uses the total route distance as the search target of the optimal search order, randomly selects a route in advance, and optimizes the annealing algorithm to obtain a path with the minimum total distance among all routes. Since the function values ​​of the two end points of the route in the simulated annealing algorithm are equal, and any point can be used as the initial point to minimize the path, it is only necessary to randomly specify one point between the two end points as the starting point.

[0113] After obtaining the optimal search order, the trajectory planning of the pilot parafoil can be carried out to minimize energy consumption and ensure that the parafoil passes through all two-dimensional feature points and scans all nominal landing points under the premise of satisfying the constraint conditions. In order to meet the path constraint conditions, this embodiment designs a parafoil trajectory adaptive segmentation strategy based on time estimation, and the strategy process is as follows:

[0114] First, the pilot parafoil is launched from one end of the optimal search sequence, the first two-dimensional feature point is scanned, and then the time required to fly to the next two-dimensional feature point is estimated. The initial direction is set to point from the starting point to the next two-dimensional feature point. The estimated time for the pilot parafoil to maneuver from the i-1th two-dimensional feature point to the i-th two-dimensional feature point is:

[0115]

[0116] Among them, τ i-1,i is the estimated time for the pilot parafoil to maneuver from the i-1th two-dimensional feature point to the i-th two-dimensional feature point, l i-1,i is the plane distance between the i-1th two-dimensional feature point and the i-th two-dimensional feature point, V is the resultant velocity of the pilot parafoil, γ is the track angle of the pilot parafoil, Δξ i-2,i-1,i is the maneuvering angle of the line between the i-2nd 2D feature point and the i-1th 2D feature point relative to the line between the i-1th 2D feature point and the i-th 2D feature point, i max It is the maximum control amount of the pilot parafoil;

[0117] Assume that the horizontal distance between two-dimensional feature point 2 and two-dimensional feature point 3 is l 2,3 , the line connecting the two-dimensional feature point 1 and the two-dimensional feature point 2, the maneuvering angle relative to the two-dimensional feature point 2 and the two-dimensional feature point 3 is Δξ, and the maneuvering angle is defined as Figure 2 As shown, the theoretical minimum time required for it to fly from two-dimensional feature point 2 to two-dimensional feature point 3 can be estimated as:

[0118]

[0119] Secondly, the minimum scanning height constraint: According to the field of view constraint of the pilot parafoil, the minimum scanning height is estimated to ensure that the scanned image resolution can meet the image recognition requirements. Therefore, the minimum scanning height h of the pilot parafoil is min It can be estimated as:

[0120] h min =v / tan(θ / 2)

[0121] Where v is the minimum length scale in the terrain analysis model, and θ is the maximum field of view of the camera on the pilot parafoil;

[0122] Then, the maximum scanning altitude constraint: Considering that if the parafoil flies at a high altitude, it is impossible to accurately identify the ground objects in the ground area, it is also necessary to specify its maximum flight altitude. According to the field of view angle constraint, the maximum scanning altitude of the pilot parafoil h max for:

[0123]

[0124] Among them, px scanner Camera resolution carried by the pilot parafoil, px model The image resolution for image recognition;

[0125] Finally, fix the position of the two-dimensional feature points and segment the trajectory: According to the estimated maneuvering time and altitude constraints, the position of each two-dimensional feature point in the altitude direction can be fixed as:

[0126]

[0127] Among them, h i 、h i-1 is the height of the pilot parafoil when it passes through the i-th and i-1-th two-dimensional feature points, ∑τ i-1,i is the sum of all estimated times, h0 is the initial position height of the pilot parafoil release;

[0128] By combining the plane coordinates of each two-dimensional feature point and the calculated height of the pilot parafoil when it passes through the two-dimensional feature point, the three-dimensional coordinates of a series of three-dimensional space path points during the pilot parafoil's return process can be determined. The three-dimensional space path points correspond to the two-dimensional feature points one by one. The pilot parafoil is controlled to pass through each three-dimensional space path point in sequence according to the optimal search order, and the track segmentation based on estimated time can be achieved according to each three-dimensional space path point.

[0129] In this embodiment, the four-degree-of-freedom motion equation of the pilot parafoil in the inertial coordinate system is:

[0130]

[0131] Among them, ξ is the velocity direction angle, u is the control input parameter, V s is the horizontal projection speed, V z is the vertical projection velocity, w x is the wind speed component along the x-axis, w y is the wind speed component along the y-axis, is the time derivative of the three-axis components of displacement in the inertial reference system Oxyz;

[0132] When the pilot parafoil performs a given mission, in addition to meeting the boundary constraints, maneuverability constraints, and headwind landing constraints when the parafoil is returning, it also needs to meet the field of view constraints of the carried camera to meet the need for scanning the ground;

[0133] The boundary constraints are: when the pilot parafoil performs the scanning task, it is assumed that the state of each parafoil at the initial moment is known, that is, the initial conditions are known, where (x0, y0, z0) is the initial position of the pilot parafoil, ξ0 is the initial yaw angle of the pilot parafoil, and the plane coordinate state of the parafoil at the two-dimensional feature point is also known, where (xi ,y i ,z i ) is the coordinate of the ith two-dimensional feature point passed, and the terminal boundary condition is that the parafoil is at t f Arriving at the target landing point (x f ,y f ,z f );

[0134] The headwind landing constraint is: before the parafoil reaches the target landing point, it should try to ensure that it is aligned against the wind, that is, the landing direction of the parafoil is opposite to the wind direction, so as to meet the conditions for the bird landing. Assume that the wind direction angle near the target landing point is ξ w , then in order to meet the headwind landing constraint, the direction angle of the parafoil during landing should satisfy ξ(t f )=ξ w ±(2k+1)π,

[0135] The control constraint is: the parafoil system is mainly realized by pulling down the parachute rope during the maneuvering process. However, the pulling amount of the left and right parachute ropes is limited. When the parachute rope is pulled down to the maximum stable unilateral pulling amount, the corresponding turning radius is the minimum value. At this time, the turning angular velocity control amount is the largest. Therefore, the yaw angular velocity control amount is constrained to |u|≤u max , where the maximum control amount u allowed is max Corresponds to the minimum turning radius of the parafoil;

[0136] The field of view angle constraint is: when the pilot parafoil performs scanning tasks, it is necessary to use the camera carried to shoot and scan the ground, and the scanning area of ​​the camera is constrained by its field of view angle and flight altitude. The field of view angle refers to the angle formed by the two edges of the maximum range of the object image of the measured target that can pass through the lens in the optical instrument, with the lens of the optical instrument as the vertex. Figure 3 , let the field of view angle be θ, and the height of the parafoil from the ground be h, then the area of ​​the observable range is S = π(htanθ) 2 .

[0137] In the process of planning the segmented trajectory, considering the maneuverability and calculation difficulty of the parafoil system, the descent trajectory between two three-dimensional space path points of the pilot parafoil is divided into N segments. For example, the segmented trajectory between the i-1th three-dimensional space path point and the i-th three-dimensional space path point is composed of N+1 segment nodes, where the 1st and N+1th segment nodes correspond to the i-1th and i-th three-dimensional space path points respectively, and a control amount u is applied between each segment. ij (j=2,…,N) to perform a maneuver. The goal in planning the segmented trajectory is to obtain a set of optimal control quantities U for each segment. i =[u i2 ,…uij ,…,u iN ], so that the energy consumption is minimized, that is, the objective function is:

[0138]

[0139] Among them, J i-1,i is the objective function of segmented trajectory planning between the i-1th 3D space path point and the i-th 3D space path point, t j-1 ,t j is the time when it passes through the j-1th segment node and the jth segment node;

[0140] In this embodiment, a particle swarm algorithm is combined with an interior point method-obstacle penalty function method to perform segmented trajectory planning, thereby obtaining the optimal control amount of each segment to avoid falling into a local optimal solution too early. The specific implementation process is as follows:

[0141] Step 501, set the number of segments M, the number of segments counter p=1, the critical value of the allowable function J0, the iteration counter T=1, the upper limit of the number of particle swarm iterations T0, the global optimal weight C1 and the local optimal weight C2;

[0142] Step 502: Based on the parafoil coordinate position and direction angle, the optimization function J of the pth segment trajectory is given. p ;

[0143] Step 503: Randomly initialize the pth trajectory input Particle evolution speed Among them, k is the particle number, p is initially set to 1, and u p2 ,…,u pN The value range is [-u max ,u max ],v p2 ,…,v pN The value range is [-u max ,u max ];

[0144] Step 504, Update renew Among them, U p0 is the optimal solution for the pth trajectory, is the local optimal solution of the kth particle in the pth trajectory;

[0145] Step 505, determine |u pj |>u max Is it true, where j = 2, 3, 4, ..., N;

[0146] If so, let u pj in [-u max, u max is randomly selected from;

[0147] Otherwise, keep u pj unchanged;

[0148] Step 506, determine whether |v pj | > u max holds, where j = 2, 3, 4, …, N;

[0149] If so, let v pj be randomly selected from [-u max , u max ;

[0150] Otherwise, keep v pj unchanged;

[0151] Step 507, determine whether holds, where represents the optimized function value when the input of the p-th trajectory is , and represents the optimized function value when the input of the p-th trajectory is ;

[0152] If so, let

[0153] Otherwise, keep unchanged;

[0154] Step 508, determine whether holds, where J p (U p0 ) represents the optimized function value when the input of the p-th trajectory is U p0 ;

[0155] If so, let

[0156] Otherwise, keep U p0 unchanged;

[0157] Step 509, determine whether J p (U p0 ) < J0 or T > T0 holds;

[0158] If so, let U p = U p0 and then proceed to Step 5010;

[0159] Otherwise, return to Step 504;

[0160] Step 5010, use the Newton method to minimize J p (U p ) + μ·c(Up ), where μ is the penalty factor, c(U p ) is the out-of-bounds penalty function;

[0161] Step 5011, update p=p+1, and determine whether p>M is satisfied:

[0162] If so, then the output control matrix U = [U1; U2; ...; U M ]

[0163] Otherwise, return to step 502.

[0164] During the landing process of the parafoil system, not only the impact of the undulation of the terrain on the landing should be considered, but also the threat posed by the existing features to the safe landing should be considered. During the landing mission, if the features are found to be buildings, waters and woodlands, the area is not suitable for the landing mission, and the landing point should be changed, and the area where the existing features are roads or cultivated land should be selected as the landing point. Therefore, after the pilot parafoil is segmented and planned, the pilot parafoil scans the surface features according to the trajectory planning results, and obtains the feature images sent back by the pilot parafoil, and then inputs the feature images into the feature classification model for feature classification, and eliminates all nominal landing points that are not suitable for landing, such as eliminating the nominal landing points in the areas of buildings, waters and woodlands, and retaining the remaining nominal landing points and outputting them, so that more accurate landing point selection can be performed, thereby improving the reliability of the homing mission.

[0165] In this embodiment, in order to realize the recognition of the types of features in the pre-landing area, it is first necessary to obtain pictures of each typical feature. A public channel obtains several satellite images of size 256×256 with a resolution of 30m. The images are divided into five basic features: woodland, cultivated land, water, building and road by manual labeling and classification, thereby constructing the data set required for training. The amount of data for each feature is roughly equal. The data set is divided into a training set and a validation set in a ratio of 7:1.

[0166] In order to select a network model suitable for object classification, the constructed data set was trained using ResNet34, ResNet 101, VGG19, and GoogLeNet, which are commonly used in image classification. The code platform used for training is Python. The training process uses a third-party deep learning platform Pytorch2.1.2, the GPU is NVIDIA Tesla P100, Cuda version 12.1, and 400 rounds of iterative training. The training time of the four network models is shown in Table 1, and the training results are shown in Table 2.

[0167] Table 1

[0168]

[0169] Table 2

[0170]

[0171] As shown in Table 1, the training time of the VGG and ResNet 101 network models is relatively long. The VGG network uses more parameters, which results in a longer model training time; the ResNet network is deeper and uses more residual blocks, which results in its fitting training time being an order of magnitude higher than the other three. As can be seen from Table 2, the accuracy, recall rate, and F1 value of the ResNet34 network are all larger values ​​among the four network models. It has better predictability, lower loss function, better model fit, and smaller error value. Comparing the prediction time, the ResNet 34 network has a shorter prediction time, less than 1s, and can perform fast and accurate predictions, which can meet the basic task requirements of land object recognition. Based on the above evaluation and analysis, this embodiment selects the ResNet 34 network for land object classification.

[0172] The following further describes the method for planning the homing point of the parafoil system based on terrain and features in this embodiment in conjunction with specific numerical simulation analysis.

[0173] In order to verify the effectiveness of the method of this embodiment, it is necessary to select an area with undulating terrain for simulation verification. In this example, a square topographic map with a resolution of 30m and a size of 72px×72px, that is, a side length of 216m×216m, is selected. The corresponding longitude and latitude range is (28.5303°N, 112.3103°E)~(28.5500°N, 112.3300°E) DEM for simulation calculation. The data comes from the Shuttle Radar Topography Mission (SRTM) of the Resources and Environmental Science Data Center of the Chinese Academy of Sciences. The contour map of the area is shown in the figure below. Figure 4 shown.

[0174] The terrain analysis model is used to obtain the flat points, safe landing points, reliable obstacle avoidance points and nominal landing points in the area. Figure 5 The parameter settings are shown in Table 3. Figure 5 It can be seen that the range of landing points in the area is gradually reduced, and finally reduced to the point where the black point is located. The black point in the figure is the working point where the terrain is suitable for performing the parafoil landing mission, that is, the nominal landing point.

[0175] Table 3

[0176]

[0177] The two-dimensional feature points are selected for the nominal landing points selected in the terrain analysis model. The selection results are as follows: Figure 6 As shown, the selected coverage radius r=200m. Figure 6 In the figure, the black area is the nominal landing point selected by terrain analysis, the triangulated point is the two-dimensional feature point, and the circle represents the area covered by the two-dimensional feature point. Figure 6 It can be seen that the selection of two-dimensional feature points simplifies the task and ensures that the nominal landing points are scanned by the carried camera to meet the task requirements.

[0178] Then, the annealing algorithm is used to optimize the search order of each two-dimensional feature point, and the initial temperature is set to T = 1×10 7 , the optimization simulation results are as follows Figure 7 As shown, the endpoint in the upper right corner is selected as the initial point. The search order basically meets the goal of minimizing the connection distance.

[0179] Then, the particle swarm algorithm is combined with the interior point method to solve the optimal trajectory planning problem. Figure 8 As shown, the plane projection of the trajectory is as follows Fig. 9 As shown in the figure. The solid line is the parafoil track, and the dotted line and arrows represent the search order. The control quantity u of each track changes with time as shown in Fig.10 The planned horizontal projection of the flight path passes through each two-dimensional feature point, which meets the mission requirements of the pilot parafoil and controls the discontinuous change of the input quantity, making the actual operation relatively simple.

[0180] Finally, in order to verify the effectiveness of the ground feature analysis model, the ResNet34 network was used to test the ground features and identify the landforms of the mission landing area. A satellite map with a resolution of 30m was used, and the GPU used for prediction was NVIDIA TeslaP100. The prediction atlas was selected as follows: Fig.11 , the recognition results are shown in Table 4. Fig.11 Compared with the results in Table 4, the accuracy of ground object recognition is high, and the accuracy of the 20 recognized graphics reaches 100%. Fig.11 In the above analysis, the average prediction time of each image is within 1 second, which can meet the real-time processing requirements of the task. Furthermore, based on the image recognition results, the nominal landing points that are not suitable for the deployment of landing missions can be eliminated to provide a landing area plan for large-scale airdrops. As shown in Table 4, the landing areas numbered 1-8 are retained, and the landing areas numbered 9-12 are accurately eliminated. In this way, the landing points that are not suitable for the landing mission are eliminated, and more accurate landing point selection can be performed, which improves the reliability of the homing mission.

[0181] Table 4

[0182]

[0183] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for planning the homing point of a parafoil system based on terrain and features, characterized in that: The steps include: Step 1: Analyze the digital elevation map of the pre-landing area based on the terrain analysis model to screen out all nominal landing points with terrain undulations suitable for parafoil landing; Step 2, select a number of two-dimensional feature points from all nominal landing points, and the selection rule is: when the pilot parafoil passes through all the two-dimensional feature points, all the nominal landing points can be scanned; Step 3, sort all the two-dimensional feature points to obtain the optimal search order; Step 4, based on the optimal search sequence, obtain the height of the pilot parafoil when it passes through each two-dimensional feature point based on time estimation, and obtain a three-dimensional space path point corresponding to each two-dimensional feature point one by one; Step 5, segmenting the airdrop trajectory between each two adjacent three-dimensional space path points, and taking the minimum energy consumption as the optimization goal, planning each segmented trajectory under the premise of satisfying the constraint conditions, and integrating the trajectory planning results of the pilot parafoil; Step 6, controlling the pilot parafoil to scan the surface objects according to the trajectory planning result, and obtaining the image of the objects transmitted back by the pilot parafoil; Step 7: classify the object image based on the object classification model, eliminate all nominal landing points that are not suitable for landing, and output the remaining nominal landing points.

2. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1 is characterized in that: In step 1, the terrain analysis model uses an inverse pyramid landing point selection algorithm to screen all nominal landing points; Among them, the inverse pyramid landing point selection algorithm searches the pre-landing area in a progressive order of flat point search, safe landing point search, reliable obstacle avoidance point search, safe landing area and nominal landing point area selection.

3. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1, characterized in that: In step 2, a number of two-dimensional feature points are selected from all nominal landing points, specifically: Step 201, adding all nominal landing points to the point set; Step 202, randomly selecting a point in the point set as a two-dimensional feature point, and drawing a circle with the selected two-dimensional feature point as the center and r as the radius; Step 203, after deleting all points in the circle made in step 202 from the point set, step 202 to step 203 are performed again until the point set is an empty set.

4. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1, characterized in that: In step 3, the goal is to minimize the sum of the distances between adjacent two-dimensional feature points in the search order, and the optimal search order is obtained based on the simulated annealing algorithm. The process is as follows: Step 301, obtaining an initial temperature T0, an annealing temperature ΔT and a randomly set initial search order, and setting the current temperature T=T0; Step 302, let T = T - ΔT, and calculate the sum L0 of the distances between the adjacent two-dimensional feature points in the current search order; Step 303, randomly select two nodes in the current search order and exchange their positions to obtain a new search order, and calculate the sum L of their connection distances; Step 304, determine whether L0>L+exp(α·(T0-T) / T0) holds, where α is a control parameter: If yes, after taking the new search order as the current search order, proceed to step 305; Otherwise, proceed to step 305; Step 305, determine whether T<0 holds: If so, take the current search order as the optimal search order and output it; Otherwise, return to step 302.

5. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1, 2, 3 or 4, characterized in that: In step 3, the height of the pilot parafoil when it passes through each two-dimensional feature point is estimated based on time, specifically: Among them, h i 、h i-1 is the height of the pilot parafoil when it passes through the i-th and i-1-th two-dimensional feature points, τ i-1,i is the estimated time for the pilot parafoil to maneuver from the i-1th two-dimensional feature point to the i-th two-dimensional feature point, ∑τ i-1,i is the sum of all estimated times, h0 is the initial position height of the pilot parafoil, and h min ≤h0≤h max ,h min is the minimum scanning height of the pilot parafoil, h max It is the highest scanning altitude of the pilot parafoil.

6. The method for planning the homing point of a parafoil system based on terrain and features according to claim 5 is characterized in that: The estimated time for the pilot parafoil to maneuver from the i-1th two-dimensional feature point to the i-th two-dimensional feature point is specifically: Among them, l i-1,i is the plane distance between the i-1th two-dimensional feature point and the i-th two-dimensional feature point, V is the velocity modulus of the pilot parafoil, γ is the track angle of the pilot parafoil, Δξ i-2,i-1,i is the maneuvering angle of the line between the i-2nd and i-1st two-dimensional feature points relative to the line between the i-1st and i-th two-dimensional feature points, u max It is the maximum control amount of the pilot parafoil.

7. The method for planning the homing point of a parafoil system based on terrain and features according to claim 5, characterized in that: The minimum scanning altitude and the maximum scanning altitude of the pilot parafoil are respectively: h min =v / time(θ / 2) Where v is the minimum length scale, θ is the maximum field of view of the camera on the pilot parafoil, px scanner Camera resolution carried by the pilot parafoil, px model The image resolution for image recognition.

8. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1, 2, 3 or 4, characterized in that: In step 5, when planning the segmented trajectory, the segmented trajectory is divided into N segments in the height direction, and a control amount u is applied between each segment. ij Perform a maneuver; The goal of planning the segmented trajectory is to obtain a set of optimal control quantities U i =[u i2 ,…u ij ,…,u iN ], so that the energy consumption is minimized, wherein the particle swarm algorithm is first used to obtain the approximate range of the optimal control quantity, and then the interior point method-obstacle penalty function method is used to obtain the optimal control quantity U within the range, which specifically includes the following steps: Step 501, set the number of segments M, the number of segments counter p=1, the critical value of the allowable function J0, the iteration counter T=1, the upper limit of the number of particle swarm iterations T0, the global optimal weight C1 and the local optimal weight C2; Step 502: Based on the parafoil coordinate position and direction angle, the optimization function J of the pth segment trajectory is given. p ; Step 503: Randomly initialize the pth trajectory input Particle evolution speed Among them, k is the particle number, p is initially set to 1, and u p2 ,…,u pN The value range is [-u max ,u max ],v p2 ,…,v pN The value range is [-u max ,u max ],u max It is the maximum control amount of the pilot parafoil; Step 504, Update renew Among them, U p0 is the optimal solution for the pth trajectory, is the local optimal solution of the kth particle in the pth trajectory; Step 505, determine |u pj |>u max Is it true, where j = 2, 3, 4, ..., N: If so, let u pj in [-u max ,u max ] randomly pick a value; Otherwise, let u pj The value remains unchanged; Step 506, determine |v pj |>u max Is it true, where j = 2, 3, 4, ..., N: If so, let v pj in [-u max ,u max ] randomly pick a value; Otherwise, let v pj The value remains unchanged; Step 507, determine Is it established: If so, then Otherwise, make The value remains unchanged; Step 508, determine Is it established: If so, then Otherwise, make U p0 The value remains unchanged; Step 509, determine J p (U p0 ) < J0 or T > T0 holds? If so, let U p =U p0 Then proceed to step 5010; Otherwise, return to step 504; Step 5010, using Newton's method to minimize J p (U p )+μ·c(U p ), where μ is the penalty factor, c(U p ) is the out-of-bounds penalty function; Step 5011, update p=p+1, and determine whether p>M is satisfied: If so, then the output control matrix U = [U1; U2; ...; U M ] Otherwise, return to step 502.

9. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1, 2, 3 or 4, characterized in that: In step 5, the constraints include boundary constraints, headwind landing constraints, control constraints and field of view constraints.

10. The method for planning the homing point of a parafoil system based on terrain and features according to claim 1, 2, 3 or 4, characterized in that: In step 7, the object classification model is composed of a trained ResNet34 network.

Images