A method for UAV aeromagnetic geofollowing flight trajectory planning based on sparse constraints of feature information
By constructing a planning method for an aeromagnetic importance field and a sparse waypoint constraint set, the problems of low efficiency, wasted endurance, and insufficient information utilization in UAV aeromagnetic measurements are solved, achieving an optimal balance between safety, efficiency, and effectiveness.
Patent Information
- Application Number
- CN202611167496.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-03
- Publication Date
- 2026-11-10
AI Technical Summary
Existing UAV aeromagnetic measurement methods suffer from low flight efficiency, wasted endurance, and insufficient utilization of prior aeromagnetic information in complex terrain, making it difficult to balance safety, efficiency, and effectiveness.
A trajectory planning method based on sparse constraints of feature information is adopted. By constructing a deep operator network guided by physical information to generate an aeromagnetic importance field, and combining differentiable Top-K algorithm and multi-objective optimization, a sparse waypoint constraint set is selected to generate the optimal three-dimensional terrain-following flight trajectory.
It significantly improves flight efficiency and endurance utilization, accurately utilizes prior aeromagnetic information, and increases the capture rate of key aeromagnetic anomalies, achieving a balance between safety, efficiency, and effectiveness.
Smart Images

Figure CN123065462A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geophysical aeromagnetic survey trajectory planning technology, and more specifically to a method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information. Background Technology
[0002] Unmanned aerial vehicle (UAV) aeromagnetic surveying has become an important technical means for geophysical airborne exploration. In recent years, with the rapid development of UAV platform technology, how to scientifically balance the effectiveness of aeromagnetic surveying with operational efficiency while ensuring flight safety has become a core challenge in terrain-following flight path planning. Currently, existing UAV terrain-following flight path planning methods are mainly geared towards scenarios such as mapping, inspection, and rescue using multi-rotor platforms. When directly applied to aeromagnetic surveying tasks that also utilize multi-rotor UAVs, the following significant shortcomings are revealed: 1. The inherent contradiction between waypoint setting and flight efficiency To ensure flight safety, existing terrain-following algorithms typically require generating extremely detailed and dense waypoint sequences in complex, undulating terrain. This approach is acceptable for multi-rotor drones primarily tasked with hovering and photography; however, for aeromagnetic operations where a smooth measurement trajectory is essential, this "point-by-point dense following" strategy leads to frequent deceleration, stationary maneuvers, and re-acceleration. Mainstream multi-rotor models, such as the DJI M350 and M400, typically have an effective flight time of only about 30 minutes per sortie. Given the already limited battery life, these non-measurement-essential maneuvers consume precious energy exponentially, severely compressing the effective aeromagnetic measurement time and significantly reducing data acquisition efficiency and the area covered per sortie.
[0003] 2. Planning defects due to insufficient utilization of prior aeromagnetic information
[0004] Existing flight path planning methods only use topographic elevation information as a single input constraint, and their objective function is simply to fit the surface undulations or avoid obstacles. They completely ignore a crucial piece of prior information that can be utilized—the subsurface magnetic structure characteristics revealed by existing small-scale aeromagnetic data in the same area. This directly leads to planned flight paths that often uniformly sweep across the entire area, failing to concentrate extremely limited flight time and measurement resources on areas with significant geological aeromagnetic anomalies, resulting in a double waste of data output and energy consumption.
[0005] 3. Safety, efficiency, and effectiveness are difficult to balance in terrain simulation operations.
[0006] In summary, current methods face a "trilemma" in practical engineering applications: if the goal is to maximize measurement effectiveness, requiring minimal altitude and comprehensive, detailed scanning, efficiency and flight time must be sacrificed, and safety risks increase in complex terrain; conversely, if safety and efficiency are prioritized, relaxing terrain constraints, widening line spacing, and increasing flight altitude, crucial details of aeromagnetic anomalies will inevitably be missed, reducing exploration effectiveness. This inherent theoretical flaw and contradiction stems primarily from the fact that existing algorithms rely solely on topographical geometry and lack feature extraction of geophysical targets and geological bodies.
[0007] Therefore, proposing a UAV aeromagnetic geodetic flight trajectory planning method based on sparse constraints of feature information to solve the difficulties of existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0008] In view of this, the present invention provides a UAV aeromagnetic geodetic flight trajectory planning method based on sparse constraint of feature information, which solves the problems of low flight efficiency and wasted endurance caused by dense waypoint settings in the prior art, the inability to effectively utilize prior aeromagnetic feature information to guide trajectory planning, and the inability to simultaneously achieve flight safety, measurement efficiency and exploration results.
[0009] To achieve the above objectives, the present invention provides the following technical solution: A method for planning UAV aeromagnetic geofollowing flight paths based on sparse constraints of feature information includes the following steps: S1. Acquire multi-source data for the survey area, including digital elevation model and prior small-scale aeromagnetic data. S2. Generate an initial terrain-following flight reference route and candidate waypoint set based on terrain elevation; S3. Construct a deep operator network guided by physical information to learn the mapping from spatial coordinates to aeromagnetic feature importance scores and generate a continuous aeromagnetic importance field. S4. Determine the maximum number of flyable waypoints K based on endurance, and use the differentiable Top-K algorithm to select the K points with the highest importance scores from the candidate waypoint set to form a sparse waypoint constraint set; S5. Using flight safety, shortest total distance, and sparse waypoint constraints as joint optimization objectives, solve for the optimal three-dimensional terrain-following flight path. S6. Output the final waypoint file that conforms to the UAV airborne control system specifications.
[0010] Optionally, the specific content of the multi-source data for the survey area acquired in S1, including the digital elevation model and prior small-scale aeromagnetic data, is as follows: Acquire the digital elevation model of the survey area, as well as existing small-scale aeromagnetic anomaly grid data or scatter data in the same area; the digital elevation model is used to ensure flight safety, and the prior small-scale aeromagnetic data provides prior geological information; The aeromagnetic data is preprocessed, including coordinate alignment, interpolation to a uniform grid, denoising, and polarization, and the output is the label field and total field outliers for training.
[0011] Optionally, the specific content of generating the initial terrain-following flight reference route and candidate waypoint set based on terrain elevation in S2 is as follows: Following the terrain-following flight safety route generation mechanism, a series of survey lines covering the survey area are generated based on the user-defined survey area boundaries, main survey line direction, and line spacing. For each survey line, a dense sequence of initial 3D waypoints is generated according to the set ground clearance and sampling interval, in conjunction with the digital elevation model. The waypoints meet the terrain obstacle avoidance requirements and constitute a candidate waypoint set. .
[0012] Optionally, in S3, a physical information-guided deep operator network is constructed to learn the mapping from spatial coordinates to aeromagnetic feature importance scores, generating a continuous aeromagnetic importance field. The specific details are as follows: A physically enhanced deep operator network is employed to learn a generalized mapping, extracting a continuous and physically consistent aeromagnetic feature importance field from small-scale aeromagnetic data. This is based on a deep operator network structure, including branch networks and a backbone network. Branch network: The input is a function representation of the prior aeromagnetic anomaly field, that is, the set of values of the prior aeromagnetic anomaly field at a fixed sensor position, which is a low-dimensional latent variable after feature encoding, and the output is a set of operator coefficients; Backbone network: The input is any three-dimensional coordinate point in space, and the output is the importance score of the aeromagnetic feature corresponding to the coordinate point; The output of the entire network is represented as:
[0013] in, This represents the input aeromagnetic anomaly field function. Represents spatial coordinates, This represents the output of the branch network. This represents the output of the backbone network. This represents the importance score of the generated aeromagnetic features; After training, deep operator networks can perform training on any given aeromagnetic distribution. It can infer the information importance of any location in the region for aeromagnetic measurements in seconds and output the aeromagnetic importance field.
[0014] Optionally, physical information constraints can be introduced into the training of deep operator networks, and the loss function can be composed of a data fitting term and a physical regularization term:
[0015] Among them, the data fitting term It measures the reconstruction error between the model output and the original aeromagnetic anomaly amplitude, ensuring that the network captures known magnetic feature distributions; the physical regularization term... Based on potential field theory, it requires that the importance field output by the model satisfies the Laplace equation in the source-free region, or satisfies the corresponding Poisson equation constraint at the boundary of the magnetic body; it adopts the following form:
[0016] in, For the Laplace operator, This is the balance coefficient.
[0017] Optionally, in S4, the maximum number of flyable waypoints K is determined based on endurance. The K points with the highest importance scores are selected from the candidate waypoint set using a differentiable Top-K algorithm, forming the sparse waypoint constraint set. The specific content of this constraint set is as follows: Determine the maximum total number of flyable waypoints globally based on the maximum flight range or waypoint limit of the drone model. ; In the candidate waypoint set Query the aeromagnetic importance score of each point. Then, the waypoint constraint set is obtained through the following sparse selection mechanism. :
[0018] The differentiable Top-K algorithm is used to select the K points with the highest importance scores from the candidate waypoint set, forming a sparse waypoint constraint set.
[0019] Optionally, in S5, flight safety, minimum total distance, and passing through a sparse waypoint constraint set are used as joint optimization objectives. The specific content of solving for the optimal three-dimensional terrain-following flight trajectory is as follows: Define the decision variable as the sequence of final three-dimensional waypoints. ,in The objective function is optimized as follows:
[0020] in: Total distance traveled, representing efficiency; For flight safety penalties, defined as This means that a penalty is imposed when the flight altitude is lower than the sum of the terrain elevation and the altitude above the ground. To define sparse waypoints as a sparse constraint penalty term. To the track The shortest distance, if the shortest distance is greater than a small threshold. If so, then punishment will be imposed; , , These are the weighting coefficients; An improved swarm intelligence algorithm is used to solve the problem. The initial solution is set as the initial terrain-following flight reference route in S2. The optimal three-dimensional terrain-following flight trajectory is obtained through iterative optimization.
[0021] Optionally, the specific content of the final waypoint file output by S6 that conforms to the UAV onboard control system specifications is as follows: The optimized 3D waypoint sequence is written with coordinate, altitude, and turning strategy information according to the specific UAV platform's route file specifications, generating the final waypoint file that can be directly loaded into the flight control system. At the same time, navigation control point files and yaw data files are exported as needed.
[0022] As can be seen from the above technical solution, compared with the prior art, the present invention provides a method for planning UAV aeromagnetic terrain-following flight trajectory based on sparse constraints of feature information, which has the following beneficial effects: (1) Significantly improve flight efficiency and endurance utilization: In response to the problem that the existing terrain-following flight method causes frequent acceleration and deceleration of multi-rotor UAVs and serious waste of endurance due to the dense waypoint setting, the present invention reduces the number of control points by constructing an aeromagnetic feature importance field and a differentiable Top-K sparse selection mechanism. For example, a single flight can have at most dozens of control points, and the waypoints with the most aeromagnetic measurement value can replace the hundreds or thousands of dense waypoints that passively follow the terrain in the traditional method. This technical means makes the flight trajectory smoother, greatly reduces the maneuvering actions that are not required for measurement, and concentrates the precious battery power for effective magnetometer data acquisition. It is expected to increase the proportion of effective measurement time per flight to more than 30%, and significantly expand the workable area per flight. (2) Accurately utilize prior aeromagnetic information to significantly improve measurement results: In view of the shortcomings of existing methods that rely solely on terrain elevation and completely ignore existing prior aeromagnetic information, this invention introduces a depth operator network guided by physical information for the first time to generate a continuous and physically consistent aeromagnetic feature importance field from small-scale aeromagnetic data; the sparse point selection strategy guided by this field can automatically tilt limited measurement resources to locations with important geological significance such as aeromagnetic anomaly boundaries and peak areas, avoiding the invalid repeated measurement of the background field by traditional uniform sweeping; through simulation comparison, it can be seen that under the same endurance constraint, the capture rate of key aeromagnetic anomaly features of this invention can be increased by more than 40%, and the structural resolution capability of the obtained data will be significantly enhanced; (3) Achieving the optimal balance of safety, efficiency and effectiveness for the first time: In response to the "trilemma" of safety, efficiency and effectiveness faced by existing technologies, this invention unifies the terrain obstacle avoidance safety constraints, total flight efficiency objectives and sparse waypoint hard constraints in one optimization model through the multi-objective joint optimization framework constructed in step five; and adopts an improved swarm intelligence algorithm for global optimization, so that the generated terrain-following flight trajectory can not only strictly guarantee the safety of the altitude above the ground, but also pass through all key aeromagnetic feature anchor points in sequence with a relatively short global flight path, fundamentally breaking through the theoretical defects of traditional methods that cannot take into account the balance of the three. (4) Strong universality and adaptability to multiple platforms and scenarios: From the route planning process to the final output format, this invention takes into account the endurance characteristics and control specifications of mainstream multi-rotor UAVs (such as DJI M350 / M400, etc.); at the same time, by replacing the digital elevation model and prior aeromagnetic data, this method can be seamlessly transferred to geophysical exploration tasks of different survey areas and different scales, and has good versatility and promotion value. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0024] Figure 1 A schematic diagram comparing the waypoint planning principles of this invention with those of existing technologies is provided for this invention. Figure 2 A flowchart of a method for planning UAV aeromagnetic geofelting flight trajectory based on sparse constraints of feature information provided by the present invention; Figure 3 This is a flowchart of the process of obtaining multi-source data of the survey area in step one of the embodiments of the present invention; Figure 4 This is a flowchart illustrating the process of generating an initial terrain-following flight reference route based on terrain elevation in step two of this embodiment of the invention. Figure 5 This is a flowchart of the physical information-guided deep operator network and aeromagnetic importance field generation process in step three of this embodiment of the invention. Figure 6 The flowchart for generating sparse measurement point selection based on endurance hard constraints in step four of the present invention is provided in the embodiment of the present invention. Figure 7 This is a flowchart of the three-dimensional trajectory planning process for multi-objective optimization in step five of the embodiments of the present invention; Figure 8 The flowchart for outputting waypoint files in step six is provided in the embodiment of the present invention. Detailed Implementation
[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0026] See Figure 2 As shown, this invention discloses a method for planning UAV aeromagnetic terrain-following flight paths based on sparse constraints of feature information, comprising the following steps: S1. Acquire multi-source data for the survey area, including digital elevation model and prior small-scale aeromagnetic data. S2. Generate an initial terrain-following flight reference route and candidate waypoint set based on terrain elevation; S3. Construct a deep operator network guided by physical information to learn the mapping from spatial coordinates to aeromagnetic feature importance scores and generate a continuous aeromagnetic importance field. S4. Determine the maximum number of flyable waypoints K based on endurance, and use the differentiable Top-K algorithm to select the K points with the highest importance scores from the candidate waypoint set to form a sparse waypoint constraint set; S5. Using flight safety, shortest total distance, and sparse waypoint constraints as joint optimization objectives, solve for the optimal three-dimensional terrain-following flight path. S6. Output the final waypoint file that conforms to the UAV airborne control system specifications.
[0027] Furthermore, the specific contents of the multi-source data for the survey area acquired in S1, including the digital elevation model and prior small-scale aeromagnetic data, are as follows: Acquire the digital elevation model of the survey area, as well as existing small-scale aeromagnetic anomaly grid data or scatter data in the same area; the digital elevation model is used to ensure flight safety, and the prior small-scale aeromagnetic data provides prior geological information; The aeromagnetic data is preprocessed, including coordinate alignment, interpolation to a uniform grid, denoising, and polarization, and the output is the label field and total field outliers for training.
[0028] Furthermore, the specific details of generating the initial terrain-following flight reference route and candidate waypoint set based on terrain elevation in S2 are as follows: Following the terrain-following flight safety route generation mechanism, a series of survey lines covering the survey area are generated based on the user-defined survey area boundaries, main survey line direction, and line spacing. For each survey line, a dense sequence of initial 3D waypoints is generated according to the set ground clearance and sampling interval, in conjunction with the digital elevation model. The waypoints meet the terrain obstacle avoidance requirements and constitute a candidate waypoint set. .
[0029] Furthermore, in S3, a physical information-guided deep operator network is constructed to learn the mapping from spatial coordinates to aeromagnetic feature importance scores, generating a continuous aeromagnetic importance field. The specific details are as follows: A physically enhanced deep operator network is employed to learn a generalized mapping, extracting a continuous and physically consistent aeromagnetic feature importance field from small-scale aeromagnetic data. This is based on a deep operator network structure, including branch networks and a backbone network. Branch network: The input is a function representation of the prior aeromagnetic anomaly field, that is, the set of values of the prior aeromagnetic anomaly field at a fixed sensor position, which is a low-dimensional latent variable after feature encoding, and the output is a set of operator coefficients; Backbone network: The input is any three-dimensional coordinate point in space, and the output is the importance score of the aeromagnetic feature corresponding to the coordinate point; The output of the entire network is represented as:
[0030] in, This represents the input aeromagnetic anomaly field function. Represents spatial coordinates, This represents the output of the branch network. This represents the output of the backbone network. This represents the importance score of the generated aeromagnetic features; After training, deep operator networks can perform training on any given aeromagnetic distribution. It can infer the information importance of any location in the region for aeromagnetic measurements in seconds and output the aeromagnetic importance field.
[0031] Furthermore, physical information constraints are introduced into the training of deep operator networks, and the loss function is composed of a data fitting term and a physical regularization term:
[0032] Among them, the data fitting term It measures the reconstruction error between the model output and the original aeromagnetic anomaly amplitude, ensuring that the network captures known magnetic feature distributions; the physical regularization term... Based on potential field theory, it requires that the importance field output by the model satisfies the Laplace equation in the source-free region, or satisfies the corresponding Poisson equation constraint at the boundary of the magnetic body; it adopts the following form:
[0033] in, For the Laplace operator, This is the balance coefficient.
[0034] Furthermore, in S4, the maximum number of flyable waypoints K is determined based on endurance. The K points with the highest importance scores are selected from the candidate waypoint set using a differentiable Top-K algorithm, forming the sparse waypoint constraint set. The specific content of this constraint set is as follows: Determine the maximum total number of flyable waypoints globally based on the maximum flight range or waypoint limit of the drone model. ; In the candidate waypoint set Query the aeromagnetic importance score of each point. Then, the waypoint constraint set is obtained through the following sparse selection mechanism. :
[0035] The differentiable Top-K algorithm is used to select the K points with the highest importance scores from the candidate waypoint set, forming a sparse waypoint constraint set.
[0036] Furthermore, in S5, flight safety, minimum total distance, and passing through sparse waypoints are used as joint optimization objectives. The specific content of solving the optimal three-dimensional terrain-following flight trajectory is as follows: Define the decision variable as the sequence of final three-dimensional waypoints. ,in The objective function is optimized as follows:
[0037] in: Total distance traveled, representing efficiency; For flight safety penalties, defined as This means that a penalty is imposed when the flight altitude is lower than the sum of the terrain elevation and the altitude above the ground. To define sparse waypoints as a sparse constraint penalty term. To the track The shortest distance, if the shortest distance is greater than a small threshold. If so, then punishment will be imposed; , , These are the weighting coefficients; An improved swarm intelligence algorithm is used to solve the problem. The initial solution is set as the initial terrain-following flight reference route in S2. The optimal three-dimensional terrain-following flight trajectory is obtained through iterative optimization.
[0038] Furthermore, the specific content of the final waypoint file output by S6 that conforms to the UAV airborne control system specifications is as follows: The optimized 3D waypoint sequence is written with coordinate, altitude, and turning strategy information according to the specific UAV platform's route file specifications, generating the final waypoint file that can be directly loaded into the flight control system. At the same time, navigation control point files and yaw data files are exported as needed.
[0039] In a specific embodiment, such as Figure 1 The diagram shown is a comparison of the waypoint planning principles of this invention and existing technologies. Based on this, this invention is proposed. This embodiment takes a resurvey task in a known small-scale aeromagnetic anomaly area as an example, and includes the following: This embodiment assumes that there are already 1:50,000 scale aeromagnetic survey results (prior data) in the survey area. The task is to use a DJI M350 multi-rotor drone equipped with a high-precision aeromagnetic survey system to conduct a resurvey of the area with a denser survey network and higher accuracy. The flight time per sortie is approximately 30 minutes, and the requirement is to plan a flight path that ensures safe terrain-following flight while maximizing the survey results.
[0040] Step 1: As Figure 3 As shown, multi-source data acquisition and preprocessing
[0041] (1) Data acquisition
[0042] Digital Elevation Model: Download the GeoTIFF file of the survey area and surrounding regions from a public geographic information database. The resolution is approximately 30m. This data will be used for subsequent terrain obstacle avoidance and height lookup.
[0043] Prior aeromagnetic data: Load the existing 1:50,000 scale aeromagnetic anomaly grid file for the survey area, in a general gridded data format. This grid contains the longitude, latitude, and total field anomaly value ΔT for each grid point.
[0044] (2) Preprocessing of aeromagnetic data
[0045] Perform the following operations using Python's scipy and pyproj libraries: Coordinate alignment: Unify the coordinate system of the aeromagnetic grid data with the digital elevation model to the WGS84 geographic coordinate system (EPSG:4326).
[0046] Interpolation to a unified grid: Bicubic spline interpolation is used to resample the prior aeromagnetic data to a spatial resolution consistent with the digital elevation model, ensuring a one-to-one correspondence between the pixels.
[0047] Noise reduction: Use a 5×5 window median filter to remove high-frequency random noise.
[0048] Pole reduction processing: If the survey area is located in the mid-to-low latitudes, the aeromagnetic ΔT data is processed to reduce the poles to eliminate the influence of tilt magnetization and better correspond the anomaly morphology to the location of the magnetic body. The data after pole reduction is recorded as RTP (Reduced To Pole, pole-reduced magnetic anomaly).
[0049] After preprocessing, a smooth aeromagnetic tag field that is perfectly aligned with the digital elevation model is obtained and stored as a two-dimensional array Mag_map[row, column], where each element is the polarization magnetic anomaly value (unit: nT) of the corresponding grid point.
[0050] Step Two: As Figure 4 As shown, the initial candidate waypoints are generated based on the terrain.
[0051] (1) Generation of survey lines
[0052] The SurveyLineGenerator class is used to generate a series of survey lines covering the survey area, specifically: Input parameters: boundary polygon of the survey area, direction angle of the main survey line (e.g., 30° east of north), and spacing between survey lines (e.g., 100m).
[0053] Projection transformation: Use pyproj to project boundary points from WGS84 coordinates to the UTM plane coordinate system for plane geometric operations.
[0054] Parallel line cutting: In the UTM coordinate system, with the survey line direction as the horizontal axis, a series of parallel straight lines are generated along the vertical direction with the line spacing as the step size. The intersection point of each straight line with the boundary of the survey area is obtained to obtain the effective survey line segment.
[0055] Serpentine sorting: All survey lines are arranged from west to east, with adjacent survey lines connected end to end, forming a serpentine flight sequence to reduce round-trip distance.
[0056] Inverse projection: The endpoints of the survey line segments in UTM coordinates are inverted back to WGS84 geographic coordinates.
[0057] (2) Dense candidate waypoint generation
[0058] Along each survey line, perform the following sub-steps: Dense sampling: A sequence of sampling points is generated at equal intervals from the start point to the end point along the survey line with a sampling interval of 5m.
[0059] Elevation query: Call the TerrainLoader.get_elevations_batch() method to sample the terrain elevation from the digital elevation model for the latitude and longitude of each sampling point.
[0060] Unified Elevation Datum: A Key Step - Using pyproj.Transformer, the WGS84 ellipsoidal heights provided by the digital elevation model are batch-converted to EGM96 elevation heights, i.e., positive altitudes. This step fundamentally eliminates the risk of collisions with mountains caused by inconsistent elevation datums.
[0061] Candidate waypoint generation: For each sampling point, the altitude + altitude above the ground is used as the flight altitude of that point to obtain a dense sequence of three-dimensional candidate waypoints. The format of each waypoint is: longitude, latitude, and flight altitude.
[0062] The candidate waypoints for all survey lines are summarized to form the complete set of candidate waypoints, denoted as: ; Where M is the total number of candidate waypoints in the entire set, which can typically range from several thousand to tens of thousands.
[0063] Step 3: Generation of Physics-Information-Guided Deep Operator Networks and Aeromagnetic Importance Fields
[0064] (1) Network structure construction
[0065] Construct a physically enhanced deep operator network, with the network structure as follows: Figure 5 As shown, it consists of a branch network and a backbone network: Branch network: A fully connected neural network is used. The input layer receives a functional representation of the prior aeromagnetic anomaly field—that is, uniformly sampling a fixed number (e.g., 256) of grid point values at fixed locations from the aeromagnetic tag field Mag_map obtained in step one, forming the input vector u. There are 4 hidden layers, each with 128 neurons, and the activation function is GELU. The output layer produces a p-dimensional (e.g., p=64) operator coefficient vector.
[0066] The backbone network also employs a fully connected neural network. The input layer accepts arbitrary three-dimensional coordinates (x, y, z) in space, which are then normalized. There are four hidden layers, each with 128 neurons, using GELU as the activation function. The output layer produces a p-dimensional basis function vector.
[0067] (2) Forward propagation and importance score output
[0068] The dot product of the outputs of the branch network and the backbone network is used to obtain the aeromagnetic feature importance score for this spatial coordinate: ; In the formula: The mapping function represents the entire deep operator network, where θ represents all trainable parameters of the network; The input is represented by the prior aeromagnetic anomaly field function. Represents the spatial coordinates of the query point.
[0069] (3) Loss function design and training
[0070] To overcome the limited accuracy of small-scale data, the loss function integrates data-driven terms and physical constraint terms: ; The data fitting term is the mean square error: ; This means that the importance score output by the network should approximate the normalized amplitude of the prior magnetic anomaly at that location.
[0071] The physical regularization term is based on potential field theory. In a source-free space, the magnetic potential field satisfies the Laplace equation. Therefore, it is required that the sum of the second derivatives of the important output field approaches zero in the region outside the magnetic source: ; In the formula: This represents the number of physical constraint points, obtained through random sampling within space. This represents the physical regularization weight coefficient, which is set to 0.1 in this embodiment. The second-order partial derivatives are calculated using an automatic differentiation mechanism.
[0072] Training was performed using the AdamW optimizer with an initial learning rate of 1×10⁻⁶. -3 The network was trained for 2000 epochs, with the learning rate adjusted using a cosine annealing strategy. After training, the network possessed the ability to infer the aeromagnetic importance of arbitrary spatial coordinates within seconds.
[0073] (4) Generation of aeromagnetic importance field
[0074] The survey area is divided into a uniform three-dimensional grid, with a horizontal resolution consistent with the digital elevation model (DEM). Vertically, it is divided into several layers. Each grid point is input into a trained network to obtain a gridded aeromagnetic feature importance score. Then, through three-dimensional spatial interpolation, a continuous and differentiable aeromagnetic significance map is generated. This map visually represents the aeromagnetic measurement value of each point in space.
[0075] Step 4: Sparse Measurement Point Selection - Generation of Hard Constraints Based on Endurance
[0076] This step quantifies the limited battery life into strict mathematical constraints, and the implementation process is as follows: Figure 6 As shown.
[0077] (1) Determine the maximum number of flyable waypoints K
[0078] The DJI M350 drone supports a maximum of approximately 99 waypoints per flight. Taking into account range and turning losses, this embodiment sets the global maximum number of waypoints K=80.
[0079] (2) Candidate waypoint scoring
[0080] For each point in the complete set of candidate waypoints generated in step two Query the aeromagnetic importance field generated in step three and obtain its score. .
[0081] (3) Differentiable Top-K selection
[0082] The Perturbed Top-K algorithm is used to select the K points with the highest scores from the entire candidate set. For the rating vector Apply random Gaussian perturbations to generate multiple sets of perturbation scores; For each group of disturbance scores, take the indices of the top K maximum values; The expected value of multiple sets of results is used as a continuous relaxation approximation of the final point selection index.
[0083] This process is differentiable, thus allowing gradient propagation in subsequent end-to-end optimization. The selected K waypoints constitute a sparse waypoint constraint set: ; These waypoints are the most valuable "anchor points" for aeromagnetic measurements in the current survey area, and must be crossed by the flight path in the final flight track.
[0084] Step 5: Multi-objective optimization of 3D trajectory planning
[0085] This step unifies safety, efficiency, and effectiveness within a single optimization framework, such as... Figure 7 As shown.
[0086] (1) Definition of decision variables
[0087] Define the decision variable as the final three-dimensional waypoint sequence. ,in, Each waypoint It includes longitude, latitude, and flight altitude.
[0088] (2) Construction of the objective function ; The specific objectives are as follows: Total voyage (i.e., efficiency goals): ; in This represents the geodesic distance between two points.
[0089] Flight safety penalties : ; in for The elevation of the terrain at that location This is the set safe altitude above the ground. This ensures that the flight altitude is not lower than the sum of the terrain altitude and the safe altitude above the ground.
[0090] Sparse constraint penalty : ; That is, sparse waypoint constraint set This is the sum of the shortest distances from each anchor point to the track. Ideally, this should approach zero, indicating that the track precisely passes through each key waypoint.
[0091] (3) Weighting
[0092] In this embodiment, a weighting coefficient is set. (i.e., range weighting) (i.e., high weighting of security penalties) (i.e., the highest weight of sparse constraints) to ensure that safety and measurement results are given priority.
[0093] (4) Optimization solution
[0094] Initial solution generation: The candidate waypoint sequence from step two is used as the initial solution to ensure that the initial track meets the terrain safety constraints.
[0095] Optimization Algorithm: An improved genetic algorithm is used to solve the problem. Real-number encoding is used, with each chromosome corresponding to a set of waypoint sequences; simulated binary crossover is used for crossover, and polynomial mutation is used for mutation. The sparse waypoint constraint set from step four is then used... Points in the dataset are directly encoded into the chromosome, naturally satisfying the hard constraints. The fitness function is taken from the objective function. The reciprocal of.
[0096] Iteration termination condition: The optimal fitness remains unchanged for 50 consecutive generations (relative change less than 1×10). -4 ), or reach the maximum number of iterations (500 generations).
[0097] Output result: The waypoint sequence corresponding to the best individual in the final population is the optimal three-dimensional terrain-following flight track.
[0098] (5) Alternative solution scheme
[0099] If computational resources are limited, a hierarchical optimization strategy can also be adopted: first, divide the survey line into K-1 segments using sparse waypoints, then run the path planning algorithm for each segment to search for the shortest safe path connecting adjacent anchor points in the three-dimensional grid space, and finally stitch them together to form a complete track.
[0100] Step Six: Output Waypoint Files
[0101] The optimal 3D waypoint sequence obtained in step five is output as a flight path file that can be directly executed by the UAV, such as... Figure 8 As shown.
[0102] (1) Generate KMZ file
[0103] Encode the waypoint sequence according to the WPML specification: template.kml: Writes the mission-level configuration, including parameters such as safe takeoff altitude, flight speed, turning mode, post-failure maneuvers, and payload type.
[0104] waylines.wpml: This file contains waypoint-level parameters. Each waypoint includes longitude, latitude, EGM96 altitude, speed, heading mode, and turning damping distance. The turning damping distance is dynamically calculated—it is taken as 1 / 3 of the minimum distance between the current waypoint and the waypoints before and after it—to ensure smooth flight.
[0105] Package the two files above into a ZIP file and change the extension to .kmz. This will create a route file that can be directly imported.
[0106] (2) Exporting auxiliary files
[0107] General KML file: Used to preview routes and waypoints in Google Earth.
[0108] Yaw data file: contains Gauss-Kruger projection coordinates.
[0109] 3D yaw data file: records the three-dimensional coordinates of all control waypoints.
[0110] Flight route log file: Records altitude, terrain elevation, and altitude above ground point by point, and marks the highest altitude point in the entire area, as well as warning messages generated due to point limit.
[0111] Route parameter file: Completely records all input parameters generated this time, making it easy to trace.
[0112] Example Effect Verification: Through the complete implementation of the above six steps, this example achieves the following technical effects: Efficiency improvement: The traditional method reduces hundreds of densely packed waypoints that follow the terrain to 80 sparse but high-value waypoints, significantly reducing non-measurement maneuvers and increasing the effective measurement time by about 30%.
[0113] Enhanced performance: Most of the 80 waypoints accurately cover the main aeromagnetic anomaly peaks and boundary regions revealed by prior data, avoiding invalid measurements of the background field, and improving the capture rate of key anomaly features by about 40%.
[0114] Safety Guarantee: All waypoints strictly meet the altitude constraints and adopt the EGM96 unified altitude benchmark, eliminating the risk of collision with mountains caused by the difference between the ellipsoidal height and the altitude.
[0115] Engineering feasibility: The final output KMZ file can be directly loaded into the DJI M350 flight control system for execution without secondary conversion.
[0116] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0117] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for planning the aeromagnetic geofollowing flight trajectory of an unmanned aerial vehicle (UAV) based on sparse constraints of feature information, characterized in that, Includes the following steps: S1. Acquire multi-source data for the survey area, including digital elevation model and prior small-scale aeromagnetic data. S2. Generate an initial terrain-following flight reference route and candidate waypoint set based on terrain elevation; S3. Construct a deep operator network guided by physical information to learn the mapping from spatial coordinates to aeromagnetic feature importance scores and generate a continuous aeromagnetic importance field. S4. Determine the maximum number of flyable waypoints K based on endurance, and use the differentiable Top-K algorithm to select the K points with the highest importance scores from the candidate waypoint set to form a sparse waypoint constraint set; S5. Using flight safety, shortest total distance, and sparse waypoint constraints as joint optimization objectives, solve for the optimal three-dimensional terrain-following flight path. S6. Output the final waypoint file that conforms to the UAV airborne control system specifications.
2. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 1, characterized in that, The specific contents of the multi-source data for the survey area obtained in S1, including the digital elevation model and prior small-scale aeromagnetic data, are as follows: Acquire the digital elevation model of the survey area, as well as existing small-scale aeromagnetic anomaly grid data or scatter data in the same area; the digital elevation model is used to ensure flight safety, and the prior small-scale aeromagnetic data provides prior geological information; The aeromagnetic data is preprocessed, including coordinate alignment, interpolation to a uniform grid, denoising, and polarization, and the output is the label field and total field outliers for training.
3. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 1, characterized in that, The specific content of generating the initial terrain-following flight reference route and candidate waypoint set based on terrain elevation in S2 is as follows: Following the terrain-following flight safety route generation mechanism, a series of survey lines covering the survey area are generated based on the user-defined survey area boundaries, main survey line direction, and line spacing. For each survey line, a dense sequence of initial 3D waypoints is generated according to the set ground clearance and sampling interval, in conjunction with the digital elevation model. The waypoints meet the terrain obstacle avoidance requirements and constitute a candidate waypoint set. .
4. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 1, characterized in that, In S3, a deep operator network guided by physical information is constructed to learn the mapping from spatial coordinates to aeromagnetic feature importance scores, generating a continuous aeromagnetic importance field. The specific details are as follows: A physically enhanced deep operator network is employed to learn a generalized mapping, extracting a continuous and physically consistent aeromagnetic feature importance field from small-scale aeromagnetic data. This is based on a deep operator network structure, including branch networks and a backbone network. Branch network: The input is a function representation of the prior aeromagnetic anomaly field, that is, the set of values of the prior aeromagnetic anomaly field at a fixed sensor position, which is a low-dimensional latent variable after feature encoding, and the output is a set of operator coefficients; Backbone network: The input is any three-dimensional coordinate point in space, and the output is the importance score of the aeromagnetic feature corresponding to the coordinate point; The output of the entire network is represented as: in, This represents the input aeromagnetic anomaly field function. Represents spatial coordinates, This represents the output of the branch network. This represents the output of the backbone network. This represents the importance score of the generated aeromagnetic features; After training, deep operator networks can perform training on any given aeromagnetic distribution. It can infer the information importance of any location in the region for aeromagnetic measurements in seconds and output the aeromagnetic importance field.
5. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 4, characterized in that, In training deep operator networks, physical information constraints are introduced, and the loss function is composed of a data fitting term and a physical regularization term: Among them, the data fitting term It measures the reconstruction error between the model output and the original aeromagnetic anomaly amplitude, ensuring that the network captures known magnetic feature distributions; the physical regularization term... Based on potential field theory, it requires that the importance field output by the model satisfies the Laplace equation in the source-free region, or satisfies the corresponding Poisson equation constraint at the boundary of the magnetic body; it adopts the following form: in, For the Laplace operator, This is the balance coefficient.
6. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 1, characterized in that, In S4, the maximum number of flyable waypoints K is determined based on endurance. The differentiable Top-K algorithm is then used to select the K points with the highest importance scores from the candidate waypoint set, forming the sparse waypoint constraint set. The specific content of this constraint set is as follows: Determine the maximum total number of flyable waypoints globally based on the maximum flight range or waypoint limit of the drone model. ; In the candidate waypoint set Query the aeromagnetic importance score of each point. Then, the waypoint constraint set is obtained through the following sparse selection mechanism. : The differentiable Top-K algorithm is used to select the K points with the highest importance scores from the candidate waypoint set, forming a sparse waypoint constraint set.
7. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 1, characterized in that, In S5, flight safety, minimum total distance, and passing through a sparse waypoint constraint set are used as joint optimization objectives. The specific content of solving the optimal three-dimensional terrain-following flight trajectory is as follows: Define the decision variable as the sequence of final three-dimensional waypoints. ,in The objective function is optimized as follows: in: Total distance traveled, representing efficiency; For flight safety penalties, defined as This means that a penalty is imposed when the flight altitude is lower than the sum of the terrain elevation and the altitude above the ground. To define sparse waypoints as a sparse constraint penalty term. To the track The shortest distance, if the shortest distance is greater than a small threshold. If so, then punishment will be imposed; , , These are the weighting coefficients; An improved swarm intelligence algorithm is used to solve the problem. The initial solution is set as the initial terrain-following flight reference route in S2. The optimal three-dimensional terrain-following flight trajectory is obtained through iterative optimization.
8. The method for planning UAV aeromagnetic geofollowing flight trajectory based on sparse constraints of feature information according to claim 1, characterized in that, The specific content of the final waypoint file output by S6 that conforms to the UAV airborne control system specifications is as follows: The optimized 3D waypoint sequence is written with coordinate, altitude, and turning strategy information according to the specific UAV platform's route file specifications, generating the final waypoint file that can be directly loaded into the flight control system. At the same time, navigation control point files and yaw data files are exported as needed.
