Control method and system applied to precise landing of unmanned aerial vehicle and medium

Abstract

The invention relates to a control method and system applied to precise landing of an unmanned aerial vehicle and a medium, and the method comprises the steps: Q1, in the process that the unmanned aerial vehicle executes an operation task, obtaining the positioning data information of the unmanned aerial vehicle in real time based on an airborne GNSS, obtaining the image data information of a landing target point in real time based on an airborne down-looking camera, and obtaining the image data information of the landing target point; acquiring data information of the position of the landing target point in real time based on a UWB positioning device; and Q2, based on the data information of the position of the landing target point, the image data information of the landing target point and the positioning data information of the unmanned aerial vehicle, adopting a CGO-Transform-GRU multi-modal fusion prediction algorithm to predict the pose of the unmanned aerial vehicle in the landing process, and obtaining the predicted data information of the pose of the unmanned aerial vehicle in the landing process. According to the invention, the continuity and real-time performance of the positioning data of the unmanned aerial vehicle are ensured, the positioning drift or air crash risk caused by the network terminal is avoided, and the landing accuracy of the unmanned aerial vehicle is improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a control method, system, and medium for precise landing of UAVs. Background Technology

[0002] With the widespread application of drones in surveying, inspection, logistics, and emergency rescue, higher demands are being placed on the safety and accuracy of autonomous drone landing. Traditional GNSS positioning methods suffer from problems such as signal multipath, obstruction, and error accumulation, making it difficult to guarantee positioning accuracy.

[0003] In the prior art, Chinese patent application (publication number: CN119255191A) discloses a drone delivery method, a drone, and a UWB positioning device. The method includes: the drone acquiring the target operating space of a target balcony; the drone flying into the target operating space based on real-time positioning information; the drone hovering within the target operating space and acquiring hovering position information; the drone broadcasting a first positioning signal to activate the UWB positioning device and send a response signal, the first positioning signal including hovering position information; the drone parsing the response signal to acquire the position information and identification information of the UWB positioning device; the drone constructing a take-off and landing space on the target balcony based on the position information and identification information; and the drone landing on the target balcony based on the take-off and landing space. However, this solution relies solely on UWB as the positioning source. In densely built-up areas, UWB signals are prone to attenuation or blockage. When the UWB signal is abnormal, the drone lacks other positioning sources for compensation, which can easily lead to landing deviation or inability to land.

[0004] In the prior art, Chinese patent application (publication number: CN119270912A) discloses a visual-guided precision landing method for unmanned aerial vehicles (UAVs). This method includes: capturing surrounding environmental information through the UAV's camera to obtain operational environment data; constructing a flight control model using the UAV's own structural parameters; setting a desired landing point; extracting a target image from the corresponding operational environment data based on the desired landing point; establishing a visual mapping between the UAV and the desired landing point based on the target image; calculating the relative position data and the distance to be flown for the UAV based on the visual mapping; obtaining a predicted flight trajectory based on the relative position data and the distance to be flown for the UAV through the flight control model; automatically capturing and tracking the desired landing point during landing; calculating the relative pose information between the UAV and the desired landing point in real time; and adjusting the UAV's flight parameters in real time based on the relative pose information and the predicted flight trajectory. However, this solution mainly relies on the visual system to identify and track the target image at the landing point. While it has good landing accuracy under ideal conditions, when the UAV's camera experiences video stream stuttering or frame drops, the position error accumulates continuously, causing the UAV to continuously deviate from a certain position during landing, resulting in significant deviations in precision landing and a risk of crashing.

[0005] In the prior art, Chinese patent application (publication number: CN109782226A) discloses an autonomous tracking and positioning system and method for unmanned aerial vehicles (UAVs) based on network RTK (Real-Time Kinematics). The system includes: an RTK base station, used to acquire network positioning information from a network of RTK base stations via a 4G network, combine this information with the RTK satellite positioning information of the RTK base stations to calculate the UAV's positioning calibration data, and transmit this data to the UAV's onboard processing equipment; and the UAV's onboard processing equipment, used to combine the UAV's RTK satellite positioning information with the positioning calibration data from the RTK base stations to calculate the UAV's precise position, thus achieving accurate positioning of the UAV. However, this solution only uses a single RTK positioning source and does not incorporate other available local positioning methods (self-built RTK base stations, UWB positioning, visual or inertial navigation compensation). When the RTK signal is lost, the number of satellites is insufficient, or the signal-to-noise ratio decreases, the system cannot automatically switch to other available high-precision positioning sources for compensation, resulting in poor system reliability in complex environments. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the present invention provides a control method, system and medium for precise landing of UAVs, which not only ensures the continuity and real-time nature of UAV positioning data and avoids the risk of positioning drift or crash caused by network terminals, but also improves the accuracy of UAV landing.

[0007] To achieve the above and other related objectives, the present invention provides the following technical solution: A control method for precise landing of unmanned aerial vehicles, the method comprising: Q1. During the operation of the UAV, the UAV's positioning data is acquired in real time based on the airborne GNSS, the image data of the landing target point is acquired in real time based on the airborne downward-looking camera, and the location data of the landing target point is acquired in real time based on the UWB positioning device. Q2. Based on the location data of the landing target point, the image data of the landing target point, and the positioning data of the UAV, the CGO-Transformer-GRU multimodal fusion prediction algorithm is used to predict the pose of the UAV during the landing process, and the predicted pose data of the UAV during the landing process is obtained. Q3. Based on the predicted pose data during the UAV landing process, and combined with the historical pose data of the UAV landing at the target point, an improved charged particle optimization algorithm is used to optimize the pose during the UAV landing process, resulting in optimized pose data during the UAV landing process. Q4. Based on the optimized drone pose data during the landing process, a preset drone pose range threshold is set. If the optimized drone pose during the landing process is within the preset drone pose range threshold, the drone pose is output to complete the landing.

[0008] Furthermore, the method also includes returning to step Q2 if the optimized drone pose during landing exceeds a preset drone pose range threshold, until the optimized drone pose during landing is within the preset drone pose range threshold.

[0009] Furthermore, in step Q2, the prediction of the pose during the UAV landing process using the CGO-Transformer-GRU multimodal fusion prediction algorithm includes: Q21. Based on the location data of the landing target point and the positioning data of the UAV, construct the covariance matrix of the landing target point and the covariance matrix of the UAV's positioning respectively, and obtain the data information of the covariance matrix of the landing target point and the data information of the covariance matrix of the UAV's positioning. Q22. Based on the image data information of the landing target point, perform feature extraction and construct the feature matrix of the landing target point image to obtain the data information of the feature matrix of the landing target point image; Q23. Based on the feature matrix data of the landing target point image, the covariance matrix data of the landing target point position, and the covariance matrix data of the UAV positioning, establish the final fusion feature function F. fused , , , , , Among them, F img For the feature matrix data information of the landing target point image, F ts1 For the data information of the covariance matrix of the landing target point, F ts2 For the data information of the covariance matrix of the drone's positioning, g img G is the gating weight function for the landing target point. ts1 G is the gating weight function for the location of the landing target point. ts2 The gating weight function is used for the positioning of the UAV, and the multimodal feature data of the UAV during the landing process is fused to obtain the multimodal fused feature data information of the UAV. Q24. Input the multimodal fusion feature data of the UAV into the trained gated recurrent neural network to predict the pose of the UAV during the landing process, and obtain the predicted pose data of the UAV during the landing process.

[0010] Furthermore, in step Q24, the trained gated recurrent neural network includes: Q241. Based on the data information of the multimodal fusion features of the UAV, construct a dataset of the multimodal fusion features of the UAV; Q242. Input the dataset of the multimodal fusion features of the UAV into the gated recurrent neural network for training and learning, optimize the network parameters, and obtain the trained gated recurrent neural network.

[0011] Furthermore, in step Q3, the optimization of the UAV's pose during landing using the improved charged particle optimization algorithm includes: Q31. Based on the predicted pose data during the UAV landing process and the historical pose data of the UAV landing at the target point, the charged particle population is initialized, the population parameters and the maximum number of iterations L are determined, and the data information of the initialized charged particle population is obtained. Q32. Based on the data information of the initialized charged particle population, construct an evaluation function P, calculate the fitness value of each individual in the charged particle population, and obtain the data information of the fitness value of each individual in the charged particle population; Q33. Based on the fitness values ​​of the individual charged particle population, and combined with the particle's current position, velocity, and charge, as well as the positions of neighboring particles, update the particle's velocity and position; Q34. Update the global electromagnetic field based on the current positions of all particles; Q35. Repeat steps Q32-Q34 until the maximum number of iterations L is reached, and the optimal solution for the charged particle population is obtained, which is the optimal pose of the UAV during the landing process.

[0012] Furthermore, in step Q32, the evaluation function P is, , Where, x i Let f be the charge value of the i-th individual in the initialized charged particle population, and let f be the charge value with respect to x. i The parabolic function, g is a function of x. i It is a direct proportional function, where α and β are weighting coefficients.

[0013] Furthermore, the positioning data of the UAV is processed using standardized data processing, the image data of the landing target point is processed using standardized data processing, and the location data of the landing target point is processed using standardized data processing.

[0014] To achieve the above and other related objectives, the present invention provides a control system for precise landing of unmanned aerial vehicles (UAVs), used to implement the aforementioned control method for precise landing of UAVs, the system comprising: The data acquisition module is used to acquire the UAV's positioning data in real time based on the airborne GNSS, acquire the image data of the landing target point in real time based on the airborne downward-looking camera, and acquire the location data of the landing target point in real time based on the UWB positioning device. The pose prediction module during the drone landing process is connected to the data acquisition module and is used to predict the pose of the drone during the landing process using the CGO-Transformer-GRU multimodal fusion prediction algorithm to obtain the predicted pose data information of the drone during the landing process. The pose optimization module during the drone landing process is connected to the pose prediction module during the drone landing process. It is used to optimize the pose of the drone during the landing process using an improved charged particle optimization algorithm to obtain the optimized pose data information of the drone during the landing process. The threshold module during the drone landing process is connected to the pose optimization module during the drone landing process. It is used to set a preset pose range threshold for the drone. If the optimized pose of the drone during the landing process is within the preset pose range threshold, the drone pose is output to complete the landing.

[0015] Furthermore, the system also includes a voice broadcast module connected to the threshold module during the drone's landing process, used to broadcast the drone's pose data information in real time during the landing process.

[0016] To achieve the above and other related objectives, the present invention also provides a computer-readable storage medium storing a computer program programmed or configured to perform the control method for precise landing of an unmanned aerial vehicle.

[0017] The present invention has the following positive effects: 1. This invention uses the CGO-Transformer-GRU multimodal fusion prediction algorithm to predict the pose of a UAV during landing, and combines it with an improved charged particle optimization algorithm to optimize the pose of the UAV during landing. This not only enables accurate prediction of the UAV landing process by combining multi-source data, ensuring the stability of UAV operations, but also optimizes the UAV landing process by combining the historical pose of the UAV when landing at the target point, further improving the accuracy and stability of the UAV landing process.

[0018] 2. This invention sets a preset pose range threshold for the UAV. If the pose of the optimized UAV during the landing process is within the preset pose range threshold, the UAV pose is output to complete the landing. This not only ensures the continuity and real-time nature of the UAV positioning data and avoids the risk of positioning drift or crash caused by network terminals, but also improves the accuracy of UAV landing. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a flowchart illustrating the CGO-Transformer-GRU multimodal fusion prediction algorithm of the present invention. Figure 3 This is a flowchart illustrating the improved charged particle optimization algorithm of the present invention. Figure 4 This is a schematic diagram of the system framework of the present invention; Figure 5 This is a schematic diagram of the system layout of the present invention. Detailed Implementation

[0020] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0021] Example 1: As Figure 1 As shown, a control method for precise landing of a drone is provided, the method comprising: Q1. During the operation of the UAV, the UAV's positioning data is acquired in real time based on the airborne GNSS, the image data of the landing target point is acquired in real time based on the airborne downward-looking camera, and the location data of the landing target point is acquired in real time based on the UWB positioning device. Q2. Based on the location data of the landing target point, the image data of the landing target point, and the positioning data of the UAV, the CGO-Transformer-GRU multimodal fusion prediction algorithm is used to predict the pose of the UAV during the landing process, and the predicted pose data of the UAV during the landing process is obtained. Q3. Based on the predicted pose data during the UAV landing process, and combined with the historical pose data of the UAV landing at the target point, an improved charged particle optimization algorithm is used to optimize the pose during the UAV landing process, resulting in optimized pose data during the UAV landing process. Q4. Based on the optimized drone pose data during the landing process, a preset drone pose range threshold is set. If the optimized drone pose during the landing process is within the preset drone pose range threshold, the drone pose is output to complete the landing.

[0022] In this embodiment, the method further includes returning to step Q2 if the optimized drone pose during the landing process exceeds a preset drone pose range threshold, until the optimized drone pose during the landing process is within the preset drone pose range threshold.

[0023] In this embodiment, adaptive attitude tolerance determination and landing execution are performed as follows: Preset attitude range thresholds are set: horizontal deviation ≤ ±2.0 cm, altitude deviation ≤ ±1.5 cm, and attitude angle deviation ≤ ±1.0°. When the optimized attitude meets all thresholds, the flight control system outputs a hovering command and initiates the motor descent program (descent rate 0.3 m / s), while continuously monitoring the UWB altitude value. When Δz_uwb < 0.15 m, the mechanical landing gear is locked, completing the landing. If any dimension exceeds the tolerance, the system returns to Q2 for re-prediction, with a maximum of 3 retries.

[0024] In this embodiment, as Figure 2 As shown, in step Q2, the adoption of CGO-Transfor The mer-GRU multimodal fusion prediction algorithm predicts the pose of a UAV during landing, including: Q21. Based on the location data of the landing target point and the positioning data of the UAV, construct the covariance matrix of the landing target point and the covariance matrix of the UAV's positioning respectively, and obtain the data information of the covariance matrix of the landing target point and the data information of the covariance matrix of the UAV's positioning. Q22. Based on the image data information of the landing target point, perform feature extraction and construct the feature matrix of the landing target point image to obtain the data information of the feature matrix of the landing target point image; Q23. Based on the feature matrix data of the landing target point image, the covariance matrix data of the landing target point position, and the covariance matrix data of the UAV positioning, establish the final fusion feature function F. fused , , , , , Among them, F img For the feature matrix data information of the landing target point image, F ts1 For the data information of the covariance matrix of the landing target point, F ts2 For the data information of the covariance matrix of the drone's positioning, g img G is the gating weight function for the landing target point. ts1 G is the gating weight function for the location of the landing target point. ts2 The gating weight function is used for the positioning of the UAV, and the multimodal feature data of the UAV during the landing process is fused to obtain the multimodal fused feature data information of the UAV. Q24. Input the multimodal fusion feature data of the UAV into the trained gated recurrent neural network to predict the pose of the UAV during the landing process, and obtain the predicted pose data of the UAV during the landing process.

[0025] In this embodiment, a certain type of hexacopter logistics drone (empty weight 2.8kg, payload 1.5kg) was used as an example, and the actual measurement was carried out on a standard concrete platform (3m×3m). Three types of interference scenarios were set up for the experiment: (1) static no wind; (2) level 3 crosswind (wind speed 3.4-5.4m / s); (3) low illumination (illuminance 15 lux, simulating dusk). Each scenario was repeated 50 times for landing.

[0026] Under the CGO-Transformer-GRU multimodal fusion prediction algorithm of this application, the average horizontal error is 1.27cm, the average altitude error is 0.89cm, the average attitude angle error is 1.32°, the success rate is 99.8%, and the average time is 9.2 seconds. From the above data, we can conclude that the UAV landing control method of this application has an interquartile range (IQR) of only 0.91cm when the wind speed reaches 5.2m / s, which is significantly narrower than the 1.73cm of the UWB+Kalman scheme, indicating that it has stronger anti-disturbance robustness.

[0027] In this embodiment, in step Q24, the trained gated recurrent neural network includes: Q241. Based on the data information of the multimodal fusion features of the UAV, construct a dataset of the multimodal fusion features of the UAV; Q242. Input the dataset of the multimodal fusion features of the UAV into the gated recurrent neural network for training and learning, optimize the network parameters, and obtain the trained gated recurrent neural network.

[0028] In this embodiment, as Figure 3 As shown, in step Q3, the optimization of the UAV's pose during landing using the improved charged particle optimization algorithm includes: Q31. Based on the predicted pose data during the UAV landing process and the historical pose data of the UAV landing at the target point, the charged particle population is initialized, the population parameters and the maximum number of iterations L are determined, and the data information of the initialized charged particle population is obtained. Q32. Based on the data information of the initialized charged particle population, construct an evaluation function P, calculate the fitness value of each individual in the charged particle population, and obtain the data information of the fitness value of each individual in the charged particle population; Q33. Based on the fitness values ​​of the individual charged particle population, and combined with the particle's current position, velocity, and charge, as well as the positions of neighboring particles, update the particle's velocity and position; Q34. Update the global electromagnetic field based on the current positions of all particles; Q35. Repeat steps Q32-Q34 until the maximum number of iterations L is reached, and the optimal solution for the charged particle population is obtained, which is the optimal pose of the UAV during the landing process.

[0029] In this embodiment, in step Q32, the evaluation function P is, , Where, x i Let f be the charge value of the i-th individual in the initialized charged particle population, and let f be the charge value with respect to x. i The parabolic function, g is a function of x. i It is a direct proportional function, where α and β are weighting coefficients.

[0030] In this embodiment, the positioning data of the UAV is processed using standardized data processing, the image data of the landing target point is processed using standardized data processing, and the location data of the landing target point is processed using standardized data processing.

[0031] In this embodiment, the present invention has been integrated into a provincial power grid inspection drone system. In a 220kV transmission tower maintenance scenario, the drone needs to automatically drop toolkits onto a metal platform (1.2m × 1.2m) atop the tower. The platform has no GNSS signal, relying solely on four UWB anchor points and H-shaped markers painted on the tower. In 127 consecutive landings, 126 successfully landed within ±5cm of the platform center. The only failure occurred due to strong electromagnetic interference causing a momentary loss of UWB lock; the system automatically switched to vision-driven mode, with a final error of 6.8cm, still meeting safety requirements. Compared to the original pure vision solution (73.1% success rate, average error 14.2cm), this invention saves an average of 2.3 minutes per inspection task and reduces annual maintenance costs by approximately 187,000 yuan.

[0032] First, a pioneering CGO-Transformer-GRU hybrid architecture is proposed, embedding a channel gating alignment mechanism into the front end of the Transformer encoder. This solves the feature mismatch problem caused by large scale differences and inconsistent update frequencies among the GNSS / UWB / visual three modalities. Compared with simple stitching and fusion, this solution reduces the horizontal prediction error by 52.4% on the same test set, verifying the effectiveness of the CGO module in multi-source geometric consistency modeling.

[0033] Second, we propose the IEPO optimization algorithm, which for the first time uses the average of historical successful landing trajectories as the physical prior to embed into the fitness function, and designs a dynamic charge decay and elite migration mechanism, which improves the optimization convergence speed by 2.1 times and the pose calibration accuracy by 41.3% compared with the standard PSO.

[0034] Third, a fully closed-loop attitude tolerance determination mechanism was established, with UWB real-time altitude as the final criterion, which avoids the risk of "false hovering" caused by GNSS altitude drift, reduces the standard deviation of landing impact force to 0.42 N·s (1.87 N·s in the traditional scheme), and significantly extends the landing gear life.

[0035] Example 2: Based on the control method for precise landing of UAVs in Example 1, the present invention will be further explained and described below.

[0036] like Figure 1 As shown, a control method for precise landing of a drone is provided, the method comprising: Q1. During the operation of the UAV, the UAV's positioning data is acquired in real time based on the airborne GNSS, the image data of the landing target point is acquired in real time based on the airborne downward-looking camera, and the location data of the landing target point is acquired in real time based on the UWB positioning device. Q2. Based on the location data of the landing target point, the image data of the landing target point, and the positioning data of the UAV, CGO-Transformer is used. The r-GRU multimodal fusion prediction algorithm predicts the pose of the UAV during the landing process, and obtains the predicted pose data information of the UAV during the landing process. Q3. Based on the predicted pose data during the UAV landing process, and combined with the historical pose data of the UAV landing at the target point, an improved charged particle optimization algorithm is used to optimize the pose during the UAV landing process, resulting in optimized pose data during the UAV landing process. Q4. Based on the optimized drone pose data during the landing process, a preset drone pose range threshold is set. If the optimized drone pose during the landing process is within the preset drone pose range threshold, the drone pose is output to complete the landing.

[0037] In this embodiment, as Figure 4 As shown, the present invention provides a control system for precise landing of unmanned aerial vehicles (UAVs), used to implement the aforementioned control method for precise landing of UAVs. The system includes: The data acquisition module is used to acquire the UAV's positioning data in real time based on the airborne GNSS, acquire the image data of the landing target point in real time based on the airborne downward-looking camera, and acquire the location data of the landing target point in real time based on the UWB positioning device. The pose prediction module during the drone landing process is connected to the data acquisition module and is used to predict the pose of the drone during the landing process using the CGO-Transformer-GRU multimodal fusion prediction algorithm to obtain the predicted pose data information of the drone during the landing process. The pose optimization module during the drone landing process is connected to the pose prediction module during the drone landing process. It is used to optimize the pose of the drone during the landing process using an improved charged particle optimization algorithm to obtain the optimized pose data information of the drone during the landing process. The threshold module during the drone landing process is connected to the pose optimization module during the drone landing process. It is used to set a preset pose range threshold for the drone. If the optimized pose of the drone during the landing process is within the preset pose range threshold, the drone pose is output to complete the landing.

[0038] In this embodiment, the system further includes a voice broadcast module connected to the threshold module during the drone's landing process, used to broadcast the drone's pose data information during the landing process in real time.

[0039] In this embodiment, the present invention provides a computer-readable storage medium storing a computer program programmed or configured to perform the control method for precise landing of a drone.

[0040] In this embodiment, as Figure 5 As shown, 1. This system consists of three parts: airborne terminal, flight control system, and peripheral equipment.

[0041] Airborne Terminal: Installed on the UAV, the airborne terminal is the front-end node for positioning data acquisition and fusion, primarily responsible for acquiring and preprocessing multi-source positioning information. Its core tasks include: 1) Deploy a network RTK client to achieve centimeter-level GNSS differential positioning; 2) Equipped with a UWB Tag module, enabling close-range high-frequency relative positioning; 3) Integrate a downward-looking camera to perform visual recognition of the landing point and PnP pose calculation; 4) The above data is fused and processed through an embedded computing platform, and the results are transmitted to the flight control system in real time via serial port and MAVLink protocol.

[0042] Flight Control System: The flight control system is the core control and decision-making unit of this system, responsible for UAV flight attitude control, navigation calculation, and safety protection. Its main functions are as follows: 1) Execute positioning fusion algorithm: Perform multi-source fusion of GNSS / RTK, UWB, visual and inertial navigation (IMU) data from the airborne terminal, and use algorithms such as extended Kalman filter (EKF) to dynamically evaluate positioning accuracy; 2) Adaptive positioning switching: Automatically switches to the optimal positioning source based on environmental characteristics (such as network latency, light intensity, signal obstruction); 3) Flight control: Generate position, speed, and attitude control commands based on fused positioning data to ensure smooth control of the UAV during landing; 4) Safety protection: When visual recognition is lost, UWB signal is abnormal, or RTK delay exceeds the threshold, the backup positioning source and inertial navigation calculation are automatically activated to achieve fault-tolerant control and avoid fall or deviation caused by error accumulation.

[0043] Peripheral equipment: Peripheral equipment includes self-built RTK base stations, UWB anchor arrays, etc.

[0044] 1) Self-built RTK base station: used to provide local RTCM differential data in the absence of network, and send it to the airborne terminal through the data transmission link to achieve high-precision positioning; 2) UWB anchor array: Deployed in the landing area to assist UAVs in high-frequency positioning correction during the low-altitude phase; 2. Detailed composition of the airborne terminal.

[0045] The airborne components include a network RTK client, a UWB tag module, and a downward-facing camera.

[0046] Network RTK client: Subscribes to NTRIP service to receive differential data in real time and calculates centimeter-level position information. When network delay or signal loss occurs during the landing phase, it automatically switches to self-built base station RTK or UWB / INS fusion mode. UWB Tag Module: Performs time-of-flight (TOF) ranging with 3-4 UWB anchor points deployed in the landing area, calculates the three-dimensional position of the UAV relative to the anchor point coordinate system, and corrects UWB drift and signal jitter through IMU compensation algorithm to enhance stability in occluded environments.

[0047] Downward-looking camera: Activated when the drone enters the landing area, it recognizes the preset QR code, performs image coordinate calculation, and combines the camera's intrinsic parameters and calibration parameters to calculate the relative pose between the drone and the landing point using the PnP algorithm. It outputs the relative position error to the flight control system for closed-loop control compensation to achieve precise landing.

[0048] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0049] In summary, this invention not only ensures the continuity and real-time nature of UAV positioning data, avoiding the risk of positioning drift or crash caused by network terminals, but also improves the accuracy of UAV landing.

[0050] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A control method for precise landing of unmanned aerial vehicles (UAVs), characterized in that, The method includes: Q1. During the operation of the UAV, the UAV's positioning data is acquired in real time based on the airborne GNSS, the image data of the landing target point is acquired in real time based on the airborne downward-looking camera, and the location data of the landing target point is acquired in real time based on the UWB positioning device. Q2. Based on the location data of the landing target point, the image data of the landing target point, and the positioning data of the UAV, the CGO-Transformer-GRU multimodal fusion prediction algorithm is used to predict the pose of the UAV during the landing process, and the predicted pose data of the UAV during the landing process is obtained. Q3. Based on the predicted pose data during the UAV landing process, and combined with the historical pose data of the UAV landing at the target point, an improved charged particle optimization algorithm is used to optimize the pose during the UAV landing process, resulting in optimized pose data during the UAV landing process. Q4. Based on the optimized drone pose data during the landing process, a preset drone pose range threshold is set. If the optimized drone pose during the landing process is within the preset drone pose range threshold, the drone pose is output to complete the landing.

2. The control method for precise landing of unmanned aerial vehicles according to claim 1, characterized in that, The method further includes returning to step Q2 if the optimized drone pose during landing exceeds a preset drone pose range threshold, until the optimized drone pose during landing is within the preset drone pose range threshold.

3. The control method for precise landing of unmanned aerial vehicles according to claim 1, characterized in that, In step Q2, the prediction of the pose during the UAV landing process using the CGO-Transformer-GRU multimodal fusion prediction algorithm includes: Q21. Based on the location data of the landing target point and the positioning data of the UAV, construct the covariance matrix of the landing target point and the covariance matrix of the UAV's positioning respectively, and obtain the data information of the covariance matrix of the landing target point and the data information of the covariance matrix of the UAV's positioning. Q22. Based on the image data information of the landing target point, perform feature extraction and construct the feature matrix of the landing target point image to obtain the data information of the feature matrix of the landing target point image; Q23. Based on the feature matrix data of the landing target point image, the covariance matrix data of the landing target point position, and the covariance matrix data of the UAV positioning, establish the final fusion feature function F. fused , , ， ， ， Among them, F img For the feature matrix data information of the landing target point image, F ts1 For the data information of the covariance matrix of the landing target point, F ts2 For the data information of the covariance matrix of the drone's positioning, g img G is the gating weight function for the landing target point. ts1 G is the gating weight function for the location of the landing target point. ts2 The gating weight function is used for the positioning of the UAV, and the multimodal feature data of the UAV during the landing process is fused to obtain the multimodal fused feature data information of the UAV. Q24. Input the multimodal fusion feature data of the UAV into the trained gated recurrent neural network to predict the pose of the UAV during the landing process, and obtain the predicted pose data of the UAV during the landing process.

4. The control method for precise landing of unmanned aerial vehicles according to claim 3, characterized in that, In step Q24, the trained gated recurrent neural network includes: Q241. Based on the data information of the multimodal fusion features of the UAV, construct a dataset of the multimodal fusion features of the UAV; Q242. Input the dataset of the multimodal fusion features of the UAV into the gated recurrent neural network for training and learning, optimize the network parameters, and obtain the trained gated recurrent neural network.

5. The control method for precise landing of unmanned aerial vehicles according to claim 1, characterized in that, In step Q3, the optimization of the UAV's pose during landing using the improved charged particle optimization algorithm includes: Q31. Based on the predicted pose data during the UAV landing process and the historical pose data of the UAV landing at the target point, the charged particle population is initialized, the population parameters and the maximum number of iterations L are determined, and the data information of the initialized charged particle population is obtained. Q32. Based on the data information of the initialized charged particle population, construct an evaluation function P, calculate the fitness value of each individual in the charged particle population, and obtain the data information of the fitness value of each individual in the charged particle population; Q33. Based on the fitness values ​​of the individual charged particle population, and combined with the particle's current position, velocity, and charge, as well as the positions of neighboring particles, update the particle's velocity and position; Q34. Update the global electromagnetic field based on the current positions of all particles; Q35. Repeat steps Q32-Q34 until the maximum number of iterations L is reached, and the optimal solution for the charged particle population is obtained, which is the optimal pose of the UAV during the landing process.

6. The control method for precise landing of unmanned aerial vehicles according to claim 5, characterized in that, In step Q32, the evaluation function P is, ， Where, x i Let f be the charge value of the i-th individual in the initialized charged particle population, and let f be the charge value with respect to x. i The parabolic function, g is a function of x. i It is a direct proportional function, where α and β are weighting coefficients.

7. The control method for precise landing of unmanned aerial vehicles according to claim 1, characterized in that: The drone's positioning data has undergone standardized data processing, the image data of the landing target point has undergone standardized data processing, and the location data of the landing target point has undergone standardized data processing.

8. A control system for precise landing of unmanned aerial vehicles (UAVs), characterized in that, The system is used to implement the control method for precise landing of a drone as described in any one of claims 1-7, the system comprising: The data acquisition module is used to acquire the UAV's positioning data in real time based on the airborne GNSS, acquire the image data of the landing target point in real time based on the airborne downward-looking camera, and acquire the location data of the landing target point in real time based on the UWB positioning device. The pose prediction module during the drone landing process is connected to the data acquisition module and is used to predict the pose of the drone during the landing process using the CGO-Transformer-GRU multimodal fusion prediction algorithm to obtain the predicted pose data information of the drone during the landing process. The pose optimization module during the drone landing process is connected to the pose prediction module during the drone landing process. It is used to optimize the pose of the drone during the landing process using an improved charged particle optimization algorithm to obtain the optimized pose data information of the drone during the landing process. The threshold module during the drone landing process is connected to the pose optimization module during the drone landing process. It is used to set a preset pose range threshold for the drone. If the optimized pose of the drone during the landing process is within the preset pose range threshold, the drone pose is output to complete the landing.

9. The control system for precise landing of unmanned aerial vehicles according to claim 8, characterized in that, The system also includes a voice broadcast module, which is connected to the threshold module during the drone's landing process, and is used to broadcast the drone's pose data information in real time during the landing process.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is programmed or configured to perform the control method for precise landing of an unmanned aerial vehicle as described in any one of claims 1 to 7.

Images