Unmanned aerial vehicle landing control method, device, system, apparatus, medium and product

Abstract

This application provides a method, apparatus, system, device, medium, and product for controlling the landing of a drone, relating to the field of drone control technology. The method includes: acquiring image data from a visual marker tag, radio frequency data transmitted by multiple radio frequency tags (RFID tags) affixed to the visual marker tag, and environmental data of the drone; determining target control parameters based on the image data, RFID data, and environmental data; and controlling the drone to land based on the target control parameters. This application uses RFID tags affixed to the visual marker tag. These RFID tags are unaffected by lighting conditions, enabling accurate drone positioning even in low-light conditions or when the visual marker tag is partially obscured. Furthermore, utilizing multimodal data to determine the parameters for controlling the drone improves the accuracy of the control parameters compared to a single sensor or tagging system, thereby enhancing landing accuracy.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a UAV landing control method, device, system, equipment, medium, and product. Background Technology

[0002] Currently, most drones use GPS positioning systems, visual recognition systems, or environmental perception sensor systems to guide their landing. These systems rely on external markers or sensor data to calculate relative position and avoid obstacles. Visual recognition systems are highly dependent on lighting conditions and the camera's angle of view; in insufficient light or when the marker is partially obscured, recognition efficiency is greatly reduced, affecting landing accuracy and safety. Environmental perception sensor systems scan the surrounding environment, automatically detect obstacles in the landing area, and adjust the drone's landing trajectory. While this effectively avoids obstacles, it does not provide a relative positional relationship with the specific landing target, thus failing to improve landing accuracy.

[0003] Therefore, it is necessary to provide a technical solution to improve the landing accuracy of drones. Summary of the Invention

[0004] This application provides a method, apparatus, system, equipment, medium, and product for controlling the landing of unmanned aerial vehicles (UAVs), which can improve the landing accuracy of UAVs.

[0005] This application provides a method for controlling the landing of an unmanned aerial vehicle (UAV), including: Acquire image data of visual marker tags, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tags, and environmental data of the UAV; Based on the image data, the radio frequency data, and the environmental data, the target control parameters are determined. The drone is controlled to land based on the target control parameters.

[0006] As one embodiment, determining the target control parameters based on the image data, the radio frequency data, and the environmental data includes: Based on the image data, the radio frequency data, and the environmental data, a fusion feature vector is determined; The fused feature vector is input into a multimodal machine learning model to obtain the target control parameters output by the multimodal machine learning model.

[0007] As one embodiment, determining the fused feature vector based on the image data, the radio frequency data, and the environmental data includes: Based on the image data, determine the pose vector and illumination intensity of the visual marker label; Based on the radio frequency data, the relative distance between the drone and the radio frequency tag, the angular deviation between the drone and the radio frequency tag, and the redundancy score of all the radio frequency tags are determined. The attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, and the environmental data are fused to obtain a fused feature vector.

[0008] As one embodiment, determining the relative distance between the drone and the RFID tag, the angular deviation between the drone and the RFID tag, and the redundancy score of all the RFID tags based on the radio frequency data includes: Based on the radio frequency data, the arrival time distance between the drone and each of the radio frequency tags is determined; The relative distance between the drone and the RFID tag is determined by weighted fusion of the arrival time distances. Based on the arrival time distances and the positions of the RFID tags, a least squares model is constructed, and the least squares model is solved to determine the position vector of the UAV; based on the position vector of the UAV and the positions of the RFID tags, the angular deviation between the UAV and the RFID tags is determined. Based on the consistency of the arrival time distances, the redundancy score of all the RFID tags is determined.

[0009] As one embodiment, before performing weighted fusion on the arrival time distances, the method further includes: Determine the difference between each arrival time distance and the expected distance, and eliminate arrival time distances with a difference greater than a preset distance threshold.

[0010] As one embodiment, the environmental data includes laser data and sonar data. The process of fusing the attitude vector, the illumination intensity, the relative distance, the angular deviation, the redundancy score, and the environmental data to obtain a fused feature vector includes: Based on the attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, the signal-to-noise ratio and historical performance indicators corresponding to the laser data and the sonar data, weighted fusion is performed to obtain fused data for each mode; The fused data of each modality is input into an adaptive Kalman filter to obtain a fused feature vector; The signal-to-noise ratio and the historical performance metrics are used to determine the weights.

[0011] As an example, the multimodal machine learning model includes a linear speed prediction model, a linear angle prediction model, and a reinforcement learning model. The linear speed prediction model is used to determine the predicted speed based on the fused feature vector. The linear angle prediction model is used to determine the predicted angle based on the fused feature vector. The state space of the reinforcement learning model is used to determine target control parameters based on the predicted fused feature vector constructed from the predicted speed and the predicted angle. The target control parameters include speed adjustment parameters and / or angle adjustment parameters for the UAV.

[0012] As an example, the attention weights of the multimodal machine learning model are determined based on the time-of-arrival distance between the drone and each of the RFID tags.

[0013] As one embodiment, the process of controlling the UAV to land based on the target control parameters further includes: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. If the flight state of the UAV is determined to be deviated based on the predicted flight state vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

[0014] As one embodiment, the process of controlling the UAV to land based on the target control parameters further includes: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. Based on the predicted flight state vector and the observation noise vector, the observation vector of the UAV is determined, wherein the observation noise vector is determined based on the environmental data; If the flight state of the UAV is determined to be deviated based on the observation vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

[0015] As one embodiment, it also includes: If the UAV has completed landing based on the predicted flight state vector, the image data, the radio frequency data, the environmental data, and the target control parameters are stored.

[0016] As one embodiment, it also includes: If the drone has not completed landing, return to the steps of acquiring image data from the visual marker tag, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tag, and environmental data of the drone.

[0017] As one embodiment, the visual marker label has four radio frequency tags arranged in an array.

[0018] This application also provides a drone landing control device, including: The acquisition module is used to acquire image data of the visual marker tag, radio frequency data transmitted by multiple radio frequency tags set on the visual marker tag, and environmental data of the UAV; The determination module is used to determine target control parameters based on the image data, the radio frequency data, and the environmental data; The control module is used to control the UAV to land based on the target control parameters.

[0019] This application also provides a drone landing control system, including a drone and a visual marker tag, wherein the visual marker tag is provided with multiple radio frequency tags; The drone is used to acquire image data from a visual marker tag, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tag, and environmental data of the drone; based on the image data, the radio frequency data, and the environmental data, it determines target control parameters; and based on the target control parameters, it controls the drone to land.

[0020] As one embodiment, the drone integrates an RFID reader, a visual sensor, a laser sensor, and a sonar sensor. The RFID reader is used to receive RFID data transmitted by multiple RFID tags, the visual sensor is used to acquire image data of the visual tags, the laser sensor is used to acquire laser data, and the sonar sensor is used to acquire sonar data.

[0021] As an example, the drone is also used for offline pre-training of a multimodal machine learning model and online fine-tuning of the multimodal machine learning model.

[0022] As one embodiment, the plurality of radio frequency tags are integrated in an array on the visual tag.

[0023] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described drone landing control methods.

[0024] This application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the UAV landing control method as described above.

[0025] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described drone landing control methods.

[0026] This application provides a drone landing control method, apparatus, system, device, medium, and product. The method includes: acquiring image data from a visual marker tag, radio frequency data transmitted by multiple radio frequency tags (RFID tags) affixed to the visual marker tag, and environmental data of the drone; determining target control parameters based on the image data, the radio frequency data, and the environmental data; and controlling the drone to land based on the target control parameters. This application uses RFID tags affixed to the visual marker tag. These RFID tags are unaffected by lighting conditions, enabling accurate drone positioning even in low-light conditions or when the visual marker tag is partially obscured. Furthermore, this application utilizes multimodal data to determine the parameters for controlling the drone, which improves the accuracy of the control parameters compared to a single sensor or tagging system, thereby improving landing accuracy. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating the drone landing control method provided in this application.

[0029] Figure 2 This is a schematic diagram of the payload terminal of the UAV provided in this application.

[0030] Figure 3 This is a schematic diagram of the visual marker label provided in this application.

[0031] Figure 4 This is a schematic diagram of the unmanned aerial vehicle (UAV) landing control device provided in this application.

[0032] Figure 5 This is a schematic diagram of the unmanned aerial vehicle landing control system provided in this application.

[0033] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] Figure 1 This is a flowchart illustrating the drone landing control method provided in this application, as shown below. Figure 1 As shown, the UAV landing control method provided in this application includes steps S110-S130.

[0036] Step S110: Acquire image data of the visual marker tag, radio frequency data transmitted by multiple radio frequency tags set on the visual marker tag, and environmental data of the UAV.

[0037] Optionally, the UAV landing control method provided in this application is applied to a UAV. The UAV's payload terminal integrates an RFID reader, a visual sensor, multiple environmental perception sensors, and a processor. The RFID reader is used to emit amplitude shift keying (ASK) signals and receive response signals from multiple RFID tags. Specifically, the RFID data refers to the response signals sent back by the RFID tags. The environmental perception sensors are used to monitor the environmental conditions of the UAV landing area in real time, such as obstacles and wind speed.

[0038] In another embodiment, such as Figure 2 As shown, the payload terminal of the UAV can integrate an RF reader, a visual sensor, multiple environmental perception sensors, a processing unit, and a control unit. The processing unit includes a high-performance processor, which is used to determine the target control parameters based on image data, RF data, and environmental data, and send the target control parameters to the control unit. The control unit is used to control the UAV based on the target control parameters.

[0039] Optionally, the number of RFID tags can be 3-5, and the multiple RFID tags are integrated in an array on the visual tag. Figure 3 A schematic diagram shows four RFID tags integrated into a visual tag. The four RFID tags are located at the four corners of the visual tag, forming a rectangular array to support multi-point time of arrival (TOA) measurement and redundancy verification. The visual tag with integrated RFID tags not only provides visual identification but also responds to the ASK signal emitted by the RFID reader on the payload terminal, sending an electromagnetic response. This allows the UAV to be verified by both RFID and visual data, greatly improving the accuracy and reliability of landing.

[0040] Optionally, the visual tagging uses Apriltag code as the tag. Apriltag is a black and white square QR code that can be quickly recognized by machines. It is mainly used as a reference point for visual positioning, allowing machines (such as drones and robots) to determine their own position and posture through cameras.

[0041] Step S120: Determine target control parameters based on the image data, the radio frequency data, and the environmental data.

[0042] Optionally, embodiments of this application analyze the flight dynamics, changes in the flight environment, and potential flight risks during the drone landing process based on image data, radio frequency data, and environmental data, and obtain the drone landing strategy, i.e., target control parameters, through machine learning.

[0043] Step S130: Control the UAV to land based on the target control parameters.

[0044] Optionally, radio frequency data can be used to determine whether the distance between the UAV controlled based on target control parameters and the visual marker tag is less than or equal to a set distance. If so, the UAV is determined to have completed landing; otherwise, the UAV is determined not to have completed landing. This embodiment of the application achieves real-time monitoring and adjustment during the UAV landing process by cyclically executing steps S110-S130, adapting to any possible environmental changes and ensuring the safety of the UAV landing.

[0045] Understandably, this application sets an RFID tag on a visual marker tag. The RFID tag is not affected by light. The dual-modal positioning significantly improves the positioning robustness and can accurately locate the UAV even in low light conditions or when the visual marker tag is partially obscured, supporting landing in complex scenarios. In addition, this application also uses multimodal data to determine the parameters for controlling the UAV. Compared with a single sensor or tagging system, it can improve the accuracy of the control parameters, thereby improving the landing accuracy.

[0046] As one embodiment, determining the target control parameters based on the image data, the radio frequency data, and the environmental data includes: Based on the image data, the radio frequency data, and the environmental data, a fusion feature vector is determined; The fused feature vector is input into a multimodal machine learning model to obtain the target control parameters output by the multimodal machine learning model.

[0047] Optionally, image data, radio frequency data, and environmental data can be processed and fused separately to obtain a fused feature vector.

[0048] Optionally, the multimodal machine learning model employs an attention mechanism to determine the target control parameters through supervised learning prediction and reinforcement learning adjustment.

[0049] It is understood that the embodiments of this application fuse multimodal data, enabling UAVs to accurately assess / adapt to complex environments and improve landing safety and success rate. This application also utilizes multimodal machine learning models to predict target control parameters, which can enhance adaptive capabilities, reduce human intervention and errors, and significantly improve flight efficiency and generalization, especially in resource-constrained UAVs.

[0050] As one embodiment, determining the fused feature vector based on the image data, the radio frequency data, and the environmental data includes: Based on the image data, determine the pose vector and illumination intensity of the visual marker label; Based on the radio frequency data, the relative distance between the drone and the radio frequency tag, the angular deviation between the drone and the radio frequency tag, and the redundancy score of all the radio frequency tags are determined. The attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, and the environmental data are fused to obtain a fused feature vector.

[0051] Optionally, in this embodiment, image frames are acquired using a visual camera, analyzed, and processed to determine a pose vector. The pose vector characterizes the rotation or translation angle of the visual tag and can be expressed as follows: The unit is degrees. Specifically, the quadrilateral formed by the four corner points of the visual tag is detected in the image frame. By comparing the deformation relationship between this quadrilateral and a standard square, the rotation angle or translation angle of the visual camera relative to the tag is deduced using a perspective transformation algorithm.

[0052] Optionally, the illumination intensity of an image frame can be calculated by calculating the average or median of the grayscale values ​​of all pixels in the image frame.

[0053] Optionally, this application reduces errors such as multipath interference, noise, or clock drift during TOA measurement by using radio frequency data from multiple RFID tags, and reduces errors by fusing multiple RFID data through redundancy verification and least squares optimization.

[0054] Optionally, environmental data can be weighted and fused to improve data reliability and accuracy.

[0055] It is understandable that this application fuses image data, radio frequency data, and environmental data to construct a fused feature vector with environmental and pose information, which is beneficial to improving positioning accuracy and the reliability of environmental perception.

[0056] As one embodiment, determining the relative distance between the drone and the RFID tag, the angular deviation between the drone and the RFID tag, and the redundancy score of all the RFID tags based on the radio frequency data includes: Based on the radio frequency data, the arrival time distance between the drone and each of the radio frequency tags is determined; The relative distance between the drone and the RFID tag is determined by weighted fusion of the arrival time distances. Based on the arrival time distances and the positions of the RFID tags, a least squares model is constructed, and the least squares model is solved to determine the position vector of the UAV; based on the position vector of the UAV and the positions of the RFID tags, the angular deviation between the UAV and the RFID tags is determined. Based on the consistency of the arrival time distances, the redundancy score of all the RFID tags is determined.

[0057] Optionally, embodiments of this application propose a cross-correlation function. The cross-correlation function is used to estimate the time delay between the ASK signal sent by the RFID reader and the response signal returned by the RFID tag. The expression is as follows: ; Where s(t) is the ASK signal sent by the RFID reader; Indicated based on time shift For response signal The sliding motion is performed; t is a time variable in seconds. By finding the position of the maximum value of R(τ), the time delay can be estimated, and then the TOA distance can be calculated.

[0058] The general formula for calculating TOA distance is as follows: ; d represents the distance in meters; c represents the speed of light constant, approximately 3 × 10^8 m / s. To estimate the delay, in seconds, it is determined by argmax R(τ), where argmax is used to characterize the input value that makes the function reach its maximum value.

[0059] This application embodiment implements array processing of multiple integrated RFID tags, and reduces the impact of multipath interference by performing angle calculation through multi-point TOA differential.

[0060] Optionally, before performing weighted fusion on each of the arrival time distances, the method further includes: Determine the difference between each arrival time distance and the expected distance, and eliminate arrival time distances with a difference greater than a preset distance threshold.

[0061] Optionally, this application embodiment uses four RFID tags as an example. Based on the RFID data, the time-of-arrival distance between the drone and each of the RFID tags can be expressed as the TOA distance. k=1-4. Compare the TOA distance with the geometrically expected distance of the RFID tag array. The difference between them, if | - If the distance threshold is exceeded, the TOA distance will be excluded. The distance threshold is determined based on the historical signal-to-noise ratio (SNR) of the RFID tag.

[0062] For example, suppose the diagonal distance of the rectangle of the RFID tag array is fixed at √2. If cm, then the distance threshold can be 0.2m.

[0063] The formula for calculating the relative distance between the drone and the RFID tag is as follows: ; in, The distance is relative, in meters; k is the index, ranging from 1 to 4, with no unit. The weight corresponding to the TOA distance is dimensionless, ranging from 0 to 1, and is determined based on the signal-to-noise ratio (SNR) of the RFID tag. = / ∑ , ∑ This represents summing the signal-to-noise ratios of all RFID tags, where j is the summation index; The TOA distance for the k-th RFID tag is in meters.

[0064] Optionally, the least squares model can be expressed as follows: ; in, Let [x, y] be the position vector of the UAV, in meters ([x, y]); min is the minimization operation. The distance is in Euclidean form, in meters. The known location of the k-th RFID tag is in meters.

[0065] The least squares model is solved using the gradient descent algorithm to obtain the UAV's position vector. The gradient descent algorithm iterates 10 times with a learning rate of 0.01. For example, in applying the least squares method to solve for the UAV's position vector, a rectangular constraint of the RFID tag matrix is ​​used, such as the tag distance matrix D = [[0,10, ...]. ,10],...].

[0066] It should be noted that if the number of RFID tags changes, the weights and constraint matrices of the TOA distance need to be adjusted, but the least squares model remains unchanged.

[0067] The drone's position vector and the positions of each RFID tag are in the same coordinate system. By inputting the drone's position vector and the RFID tag's position into the angle deviation calculation formula, the angle deviation between the drone and each RFID tag can be obtained.

[0068] Optionally, the redundancy score of all the RFID tags can be obtained based on the average consistency of each TOA distance.

[0069] Understandably, this application achieves multi-label redundant TOA distance fusion through anomaly detection, weighted averaging, and least squares geometric optimization of TOA distance, significantly improving positioning robustness and maintaining high-precision positioning even in visually limited environments such as night or fog and rain, enabling landing in complex scenarios.

[0070] As one embodiment, the environmental data includes laser data detected by a laser sensor and sonar data detected by a sonar sensor. The process of fusing the attitude vector, the illumination intensity, the relative distance, the angular deviation, the redundancy score, and the environmental data to obtain a fused feature vector includes: Based on the attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, the signal-to-noise ratio and historical performance indicators corresponding to the laser data and the sonar data, weighted fusion is performed to obtain fused data for each mode; The fused data of each modality is input into an adaptive Kalman filter to obtain a fused feature vector; The signal-to-noise ratio and the historical performance metrics are used to determine the weights.

[0071] Optionally, environmental data is used to characterize the environmental conditions of the UAV landing area, including obstacles and wind speed, wherein laser data is used to characterize obstacle information in the landing area, and sonar data is used to characterize wind speed in the landing area.

[0072] Optionally, the calculation formula for the fused data of each modality is as follows: ; in, The data represents the fused data from various modalities, with units depending on the data (e.g., meters for location); 'i' is the index, without units (1 to N); 'N' is the total number of sensors, without units (e.g., 4: RFID, vision, LIDAR, sonar). Weight, unitless (between 0 and 1); The measurement data is from the i-th sensor, and the unit depends on the sensor, such as distance for RFID or depth for LIDAR.

[0073] Optionally, the expression for the weights calculated based on the signal-to-noise ratio and the historical performance metrics is as follows: ; Let be the signal-to-noise ratio of the i-th sensor, in decibels (dB), representing the current signal quality; The sum of the SNR of all sensors is used for normalization; α is an adjustment factor, dimensionless (typical value 0.5), used to balance SNR and performance; This is a historical performance indicator, without units, with a value range of 0-1, and can be calculated based on the accuracy of the first 100 fusions.

[0074] Optionally, the adaptive Kalman filter includes a prediction formula and an update formula. The prediction formula is as follows: Update formula , A represents the predicted state, in meters (e.g., position); A is the state transition matrix, without units (describing dynamics). B represents the state at the previous moment; B is the control matrix, which is dimensionless; u is the control input, such as speed adjustment. For updating the state; K is the Kalman gain, dimensionless, calculated based on covariance; z is the observation vector, in meters, used for fusion. As input; H is the observation matrix, which has no unit.

[0075] Understandably, this application performs multimodal fusion of data from multiple types of sensors, combines SNR / historical performance to achieve advanced weighted averaging, and utilizes Kalman filtering to achieve noise filtering and uncertainty adaptation. This is beneficial for UAVs to accurately assess / adapt to complex environments such as dynamic obstacles, thereby improving landing safety and success rate.

[0076] As an example, the multimodal machine learning model includes a linear speed prediction model, a linear angle prediction model, and a reinforcement learning model. The linear speed prediction model is used to determine the predicted speed based on the fused feature vector. The linear angle prediction model is used to determine the predicted angle based on the fused feature vector. The state space of the reinforcement learning model is used to determine target control parameters based on the predicted fused feature vector constructed from the predicted speed and the predicted angle. The target control parameters include speed adjustment parameters and / or angle adjustment parameters for the UAV.

[0077] Optionally, the attention weights of the multimodal machine learning model are determined based on the time-of-arrival distance between the drone and each of the RFID tags.

[0078] Optionally, embodiments of this application utilize supervised learning techniques to train a multimodal machine learning model in order to predict optimal flight and landing parameters under specific environmental conditions.

[0079] Specifically, a historical dataset, constructed using multiple simulations and real landing data, is pre-trained offline on the GPU of the payload terminal and then fine-tuned online.

[0080] The custom loss function for minimizing the multimodal machine learning model is as follows: ; L represents the total loss, which is dimensionless; n represents the number of samples; and y represents the true parameter (such as the actual speed). For prediction; β is the attention regularization coefficient, which can take a value of 0.1; The attention weight norm is used to balance prediction accuracy and feature attention.

[0081] The linear velocity prediction model is as follows: ; v represents the predicted velocity, in m / s; This is the intercept, in m / s; These are regression coefficients, which are unitless. These are attention weights, unitless, ranging from 0 to 1, calculated using Softmax, prioritizing highly relevant features such as... ; For the measurement data of the i-th sensor, for example, = Derived from sonar, =d is derived from RFID; M is the number of features; The error term is expressed in m / s and is assumed to follow a normal distribution.

[0082] The linear angle prediction model is as follows: ; in, Predicted angle, in degrees; Intercept, in degrees; This is a coefficient, without units.

[0083] The formula for calculating the attention layer weights is as follows: ; Q / K / V is a query / key / value matrix, derived from feature vectors; For dimensions.

[0084] The reinforcement learning model employs reinforcement learning techniques to learn and optimize its performance from each execution, gradually improving the flight path decision-making process through continuous trial and error. Specifically, it uses the Deep Q-Network (DQN) algorithm, running in real-time with edge computing acceleration. The terminal GPU processes the Q-network in parallel, reducing latency. The state space includes fused feature vectors, and the action space consists of parameter adjustments, such as +Δvor + Δθ.

[0085] The reward function for a reinforcement learning model is as follows: ; Where R represents the reward, which has no unit; , , All are weights, typically 1, 1, 2; Position error, in meters; Decision delay, in seconds; The safety score is 0-1, and +1 if there is no collision.

[0086] Optionally, embodiments of this application may use machine learning algorithms such as decision trees, random forests, or neural networks to predict and optimize flight control parameters to ensure stability and safety in complex environments, wherein flight control parameters are adjusted.

[0087] For example, a multilayer perceptron (MLP) neural network is used, trained, and then deployed as an edge model. The optimization formula for the MLP neural network is as follows: ; in, To optimize the parameter vector, it can be used as the target control parameter, with mixed units, such as [m / s, degrees]; argmin is the minimization operation; The total loss is unitless; p is the UAV's position vector. The input feature vector; λ is the target (e.g., actual parameter); k is the index; λ is the regularization coefficient, 0.01; Let L2 be the norm of the parameter vector.

[0088] Understandably, this application analyzes flight dynamics during landing and automatically adjusts control parameters to adapt to environmental changes and potential flight risks. Unlike common machine learning parameter prediction in existing technologies, this application integrates TOA distance and multimodal environmental data to achieve closed-loop feedback optimization, rather than single-modal prediction as in existing technologies. This application also proposes multimodal feature fusion based on an attention mechanism, prioritizing key features (such as high SNR RFID data) to improve prediction accuracy. Furthermore, this application proposes real-time training accelerated by edge computing, using the GPU accelerator of the payload terminal for online learning, reducing latency, and is suitable for resource-constrained UAV scenarios.

[0089] As one embodiment, the process of controlling the UAV to land based on the target control parameters further includes: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. If the flight state of the UAV is determined to be deviated based on the predicted flight state vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

[0090] Optionally, the expression for predicting the flight state vector is as follows: ; The flight state vector at time t, in mixed units (e.g., [position in meters, velocity in m / s, angle in degrees]^T); This is the state transition matrix, which is unitless (used to describe system dynamics, such as [[1,Δt],[0,1]]). The state vector of the previous time step, same as ; This is a control matrix, without units (used to map inputs to states). To control the input vector, unit mixing (e.g., [acceleration m / s²]) is used. 2 ,rudder angle]); This is the process noise vector, with the same units. (Follows a Gaussian distribution N(0,Q)); t is the time step.

[0091] Optionally, the difference between the predicted flight state vector and the target flight state vector is used to determine whether there is a deviation in the flight state of the UAV. If the difference between the predicted flight state vector and the target flight state vector is greater than a threshold, it is determined that there is a deviation in the flight state of the UAV and the flight state of the UAV needs to be adjusted.

[0092] Understandably, this application uses historical flight data and current environmental data to predict changes in the flight status of the UAV, enabling closed-loop feedback and ensuring the UAV's adaptability.

[0093] As one embodiment, the process of controlling the UAV to land based on the target control parameters further includes: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. Based on the predicted flight state vector and the observation noise vector, the observation vector of the UAV is determined, wherein the observation noise vector is determined based on the environmental data; If the flight state of the UAV is determined to be deviated based on the observation vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

[0094] Optionally, the expression for the predicted flight state vector is as described in the above embodiment and will not be repeated here. The expression for the observation vector is as follows: ; For observation vectors, unit mixing (e.g., fused distance / angle); This is the observation matrix, which has no units. Same as above; To observe the noise vector, the units are the same. It follows a Gaussian distribution N(0,R).

[0095] Optionally, the difference between the observation vector and the target observation vector is used to determine whether there is a deviation in the flight state of the UAV. If the difference between the observation vector and the target observation vector is greater than a threshold, it is determined that there is a deviation in the flight state of the UAV and the flight state of the UAV needs to be adjusted.

[0096] Understandably, this application uses historical flight data and current environmental data to predict changes in the flight status of the UAV, enabling closed-loop feedback and ensuring the UAV's adaptability.

[0097] As an example, the UAV landing control method provided in this application further includes the following steps: If the UAV has completed landing based on the predicted flight state vector, the image data, the radio frequency data, the environmental data, and the target control parameters are stored.

[0098] Optionally, the image data, the radio frequency data, the environmental data, and the target control parameters are stored in a historical database for use as samples to train or adjust a multimodal machine learning model.

[0099] It is understood that, when the UAV is determined to have completed landing based on the predicted flight state vector, this application stores the image data, the radio frequency data, the environmental data, and the target control parameters to provide data support for subsequent flight or landing.

[0100] As an example, the UAV landing control method provided in this application further includes the following steps: If the drone has not completed landing, return to the steps of acquiring image data from the visual marker tag, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tag, and environmental data of the drone.

[0101] Optionally, the drone may be determined to have completed landing based on the predicted flight state vector. If the drone is determined not to have completed landing, the process may return to step S110.

[0102] Understandably, this application implements closed-loop control to ensure the adaptability of the drone when it determines that the drone has not completed landing.

[0103] As one embodiment, the visual marker label has four radio frequency tags arranged in an array.

[0104] It is understood that the embodiments of this application achieve dual-modal fusion processing by integrating an array of four RFID tags embedded with Apriltag and a multi-tag redundant TOA fusion algorithm, combined with high-resolution visual data, to solve the problems of single-source interference and occlusion, and improve the reliability and accuracy in complex environments.

[0105] The drone landing control device provided in this application is described below. The drone landing control device described below can be referred to in correspondence with the drone landing control method described above.

[0106] Figure 4 This is a schematic diagram of the unmanned aerial vehicle (UAV) landing control device provided in this application, as shown below. Figure 4 As shown, this application also provides a drone landing control device, including: The acquisition module 410 is used to acquire image data of the visual marker tag, radio frequency data transmitted by multiple radio frequency tags set on the visual marker tag, and environmental data of the UAV; The determining module 420 is used to determine target control parameters based on the image data, the radio frequency data, and the environmental data; The control module 430 is used to control the UAV to land based on the target control parameters.

[0107] As one embodiment, the determining module 420 is used for: Based on the image data, the radio frequency data, and the environmental data, a fusion feature vector is determined; The fused feature vector is input into a multimodal machine learning model to obtain the target control parameters output by the multimodal machine learning model.

[0108] As one embodiment, the determining module 420 is used for: Based on the image data, determine the pose vector and illumination intensity of the visual marker label; Based on the radio frequency data, the relative distance between the drone and the radio frequency tag, the angular deviation between the drone and the radio frequency tag, and the redundancy score of all the radio frequency tags are determined. The attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, and the environmental data are fused to obtain a fused feature vector.

[0109] As one embodiment, the determining module 420 is used for: Based on the radio frequency data, the arrival time distance between the drone and each of the radio frequency tags is determined; The relative distance between the drone and the RFID tag is determined by weighted fusion of the arrival time distances. Based on the arrival time distances and the positions of the RFID tags, a least squares model is constructed, and the least squares model is solved to determine the position vector of the UAV; based on the position vector of the UAV and the positions of the RFID tags, the angular deviation between the UAV and the RFID tags is determined. Based on the consistency of the arrival time distances, the redundancy score of all the RFID tags is determined.

[0110] As one embodiment, the determining module 420 is used for: Determine the difference between each arrival time distance and the expected distance, and eliminate arrival time distances with a difference greater than a preset distance threshold.

[0111] As one embodiment, the environmental data includes laser data and sonar data, and the determining module 420 is used for: Based on the attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, the signal-to-noise ratio and historical performance indicators corresponding to the laser data and the sonar data, weighted fusion is performed to obtain fused data for each mode; The fused data of each modality is input into an adaptive Kalman filter to obtain a fused feature vector; The signal-to-noise ratio and the historical performance metrics are used to determine the weights.

[0112] As an example, the multimodal machine learning model includes a linear speed prediction model, a linear angle prediction model, and a reinforcement learning model. The linear speed prediction model is used to determine the predicted speed based on the fused feature vector. The linear angle prediction model is used to determine the predicted angle based on the fused feature vector. The state space of the reinforcement learning model is used to determine target control parameters based on the predicted fused feature vector constructed from the predicted speed and the predicted angle. The target control parameters include speed adjustment parameters and / or angle adjustment parameters for the UAV.

[0113] As an example, the attention weights of the multimodal machine learning model are determined based on the time-of-arrival distance between the drone and each of the RFID tags.

[0114] As one embodiment, the control module 430 is used for: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. If the flight state of the UAV is determined to be deviated based on the predicted flight state vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

[0115] As one embodiment, the control module 430 is used for: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. Based on the predicted flight state vector and the observation noise vector, the observation vector of the UAV is determined, wherein the observation noise vector is determined based on the environmental data; If the flight state of the UAV is determined to be deviated based on the observation vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

[0116] As one embodiment, the control module 430 is used for: If the UAV has completed landing based on the predicted flight state vector, the image data, the radio frequency data, the environmental data, and the target control parameters are stored.

[0117] As one embodiment, the control module 430 is used for: If the drone has not completed landing, return to the steps of acquiring image data from the visual marker tag, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tag, and environmental data of the drone.

[0118] As one embodiment, the visual marker label has four radio frequency tags arranged in an array.

[0119] It should be noted that the UAV landing control device provided in this application has the same technical effects as the UAV landing control method, which will not be elaborated further.

[0120] The UAV landing control system provided in this application is described below. The UAV landing control system described below can be referred to in correspondence with the UAV landing control method described above.

[0121] Figure 5 This is a schematic diagram of the unmanned aerial vehicle (UAV) landing control system provided in this application, as shown below. Figure 5 As shown, this application also provides a drone landing control system, including a drone and a visual marker tag, wherein the visual marker tag is provided with multiple radio frequency tags.

[0122] The drone is used to acquire image data from a visual marker tag, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tag, and environmental data of the drone; based on the image data, the radio frequency data, and the environmental data, it determines target control parameters; and based on the target control parameters, it controls the drone to land.

[0123] As one embodiment, the drone integrates an RFID reader, a visual sensor, a laser sensor, and a sonar sensor. The RFID reader is used to receive RFID data transmitted by multiple RFID tags, the visual sensor is used to acquire image data of the visual tags, the laser sensor is used to acquire laser data, and the sonar sensor is used to acquire sonar data.

[0124] Optionally, the RFID reader is integrated into the drone's payload terminal to emit ASK signals and receive response signals from Apriltag with an RFID tag array.

[0125] Optionally, laser and sonar sensors are also integrated into the payload terminal to monitor environmental conditions in the landing area in real time, specifically for detecting obstacles and wind speed, respectively.

[0126] Optionally, the vision sensor uses a high-resolution camera to scan Apriltag codes with RFID tags to assist in position and orientation determination.

[0127] As an example, the drone is also used for offline pre-training of a multimodal machine learning model and online fine-tuning of the multimodal machine learning model.

[0128] Optionally, the drone's payload terminal also integrates a processing unit consisting of a high-performance processor for offline pre-training of a multimodal machine learning model and online fine-tuning of the multimodal machine learning model to optimize flight and landing strategies.

[0129] Optionally, the drone's payload terminal also integrates a control unit, which dynamically adjusts the drone's flight parameters, such as speed, angle, and attitude, based on the data analysis results from the processing unit to ensure a precise landing.

[0130] Optionally, the UAV includes a data collection module, a data processing and fusion module, a machine learning module, and a flight control execution module at the software level. The data collection module is mounted on the aforementioned multimodal sensors, the data processing and fusion module and the machine learning module are mounted on the processing unit, and the flight control execution module is mounted on the control unit. The data collection module is used to collect data from RFID readers, environmental perception sensors, and visual sensors in real time. The data processing and fusion module is used to analyze and fuse the collected data to provide accurate environmental and location information. The machine learning module is used to dynamically adjust and optimize the landing strategy based on historical data and real-time input. The flight control execution module is used to use historical flight data and current environmental data to generate future flight states through iterative prediction and output them to the machine learning module for real-time adjustment.

[0131] As one embodiment, the plurality of radio frequency tags are integrated in an array on the visual tag.

[0132] Optionally, the method for controlling drone landing based on the drone landing control system includes the following steps.

[0133] Step 1: Pre-landing data collection.

[0134] When the drone approaches the designated landing point, the payload terminal automatically activates the RFID reader, emits an ASK signal, activates the RFID tag on the Apriltag, and enables it to start transmitting signals back, providing raw TOA data.

[0135] By integrating sensors such as lasers and sonar into the payload terminal, environmental parameters such as obstacles and wind speed around the landing area are monitored in real time to ensure that the landing area is unobstructed and the wind speed is within a safe range, thus providing a safe landing environment for the drone.

[0136] The visual sensor scans the Apriltag with the RFID tag, captures image data, and accurately calculates the relative position and attitude of the drone and the Apriltag.

[0137] Step 2: Data analysis and processing.

[0138] The processing unit performs extraction at a sampling cycle of 20 times per second, activating when the UAV enters the landing preparation phase (distance <30m). Through PCA (Principal Component Analysis) dimensionality reduction and real-time filtering (e.g., median filtering to remove noise), a feature vector is formed. Example: [d, , , , , , , , (9-dimensional vector). After extraction, it is directly input into a multimodal machine learning model to achieve multimodal fusion.

[0139] Specifically, the processing unit analyzes the signals returned by the RFID tags, uses the TOA algorithm to calculate the distance and angle with each RFID tag, obtains the accurate position and orientation information of the drone and the Apriltag tags, and fuses and analyzes the environmental perception sensor data and visual sensor data with the RFID data to create a comprehensive environmental and location information model, thereby improving the positioning accuracy and the reliability of environmental perception.

[0140] The processing unit applies a pre-trained multimodal machine learning model to dynamically adjust flight control parameters, optimize flight paths and landing strategies, and adapt to different landing environments and conditions, such as adjusting the angle by +5° in high wind scenarios, based on current environment and location data and through linear regression / reinforcement learning.

[0141] Step 3: Flight control and execution.

[0142] The control unit adjusts the speed, angle, and attitude of the drone based on the output of the processing unit. For example, if the model predicts a position deviation of >0.2m, the speed is adjusted by -10% to ensure that the drone lands safely along the calculated optimal path.

[0143] During landing, environmental changes and flight status are continuously monitored, and data is fed back to the machine learning module for fine-tuning when necessary. For example, when wind speed suddenly increases, model noise is used. The compensation path adapts to any immediate changes that may occur, such as sudden increases in wind speed or moving obstacles, ensuring the safety of drone landing.

[0144] After the drone successfully lands, the system performs a self-check to confirm the status of all systems, verify the final match with the target, record data and feed it back to the historical database for future machine learning training, verify the success of the landing process, and provide data support for subsequent flights.

[0145] In summary, this application achieves dual-modal fusion processing by integrating multiple RFID tags with Apriltags in an integrated array and using a multi-tag redundant TOA fusion algorithm, combined with high-resolution visual data. This solves the problems of single-source interference and occlusion, improving reliability and accuracy in complex environments (such as fog, rain, or indoor environments). This application also applies an attention-based machine learning model to analyze the fused data, dynamically adjusting parameters to adapt to changes such as sudden increases in wind speed, improving response speed and adaptability, reducing human error, and supporting non-shipborne, variable scenarios, demonstrating the innovative advantages of machine learning-driven approaches. Furthermore, this application integrates real-time LIDAR / sonar monitoring with RFID / visual data fusion to accurately assess dynamic environments such as obstacle movement, reducing risks and ensuring safe landing in complex terrain. Finally, this application uses a high-performance processor on the payload terminal to process multi-source data, supporting real-time machine learning execution and closed-loop feedback, ensuring efficient edge computing, improving operational fluency and accuracy, and is particularly suitable for resource-constrained UAVs.

[0146] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a UAV landing control method, which includes: Acquire image data of visual marker tags, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tags, and environmental data of the UAV; Based on the image data, the radio frequency data, and the environmental data, the target control parameters are determined. The drone is controlled to land based on the target control parameters.

[0147] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the UAV landing control method provided by the above methods, the method including: Acquire image data of visual marker tags, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tags, and environmental data of the UAV; Based on the image data, the radio frequency data, and the environmental data, the target control parameters are determined. The drone is controlled to land based on the target control parameters.

[0149] Furthermore, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the UAV landing control method provided by the methods described above, the method comprising: Acquire image data of visual marker tags, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tags, and environmental data of the UAV; Based on the image data, the radio frequency data, and the environmental data, the target control parameters are determined. The drone is controlled to land based on the target control parameters.

[0150] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0151] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for controlling the landing of an unmanned aerial vehicle (UAV), characterized in that, include: Acquire image data of visual marker tags, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tags, and environmental data of the UAV; Based on the image data, the radio frequency data, and the environmental data, the target control parameters are determined. The drone is controlled to land based on the target control parameters.

2. The UAV landing control method according to claim 1, characterized in that, The determination of target control parameters based on the image data, the radio frequency data, and the environmental data includes: Based on the image data, the radio frequency data, and the environmental data, a fusion feature vector is determined; The fused feature vector is input into a multimodal machine learning model to obtain the target control parameters output by the multimodal machine learning model.

3. The UAV landing control method according to claim 2, characterized in that, The step of determining the fused feature vector based on the image data, the radio frequency data, and the environmental data includes: Based on the image data, determine the pose vector and illumination intensity of the visual marker label; Based on the radio frequency data, the relative distance between the drone and the radio frequency tag, the angular deviation between the drone and the radio frequency tag, and the redundancy score of all the radio frequency tags are determined. The attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, and the environmental data are fused to obtain a fused feature vector.

4. The UAV landing control method according to claim 3, characterized in that, The step of determining the relative distance between the drone and the RFID tag, the angular deviation between the drone and the RFID tag, and the redundancy score of all the RFID tags based on the radio frequency data includes: Based on the radio frequency data, the arrival time distance between the drone and each of the radio frequency tags is determined; The relative distance between the drone and the RFID tag is determined by weighted fusion of the arrival time distances. Based on the arrival time distances and the positions of the RFID tags, a least squares model is constructed, and the least squares model is solved to determine the position vector of the UAV; based on the position vector of the UAV and the positions of the RFID tags, the angular deviation between the UAV and the RFID tags is determined. Based on the consistency of the arrival time distances, the redundancy score of all the RFID tags is determined.

5. The UAV landing control method according to claim 4, characterized in that, Before performing the weighted fusion of the arrival time distances, the method further includes: Determine the difference between each arrival time distance and the expected distance, and eliminate arrival time distances with a difference greater than a preset distance threshold.

6. The UAV landing control method according to claim 3, characterized in that, The environmental data includes laser data and sonar data. The fusion of the attitude vector, illumination intensity, relative distance, angular deviation, redundancy score, and environmental data to obtain a fused feature vector includes: Based on the attitude vector, the illumination intensity, the relative distance, the angle deviation, the redundancy score, the signal-to-noise ratio and historical performance indicators corresponding to the laser data and the sonar data, weighted fusion is performed to obtain fused data for each mode; The fused data of each modality is input into an adaptive Kalman filter to obtain a fused feature vector; The signal-to-noise ratio and the historical performance metrics are used to determine the weights.

7. The UAV landing control method according to claim 2, characterized in that, The multimodal machine learning model includes a linear speed prediction model, a linear angle prediction model, and a reinforcement learning model. The linear speed prediction model is used to determine the predicted speed based on the fused feature vector. The linear angle prediction model is used to determine the predicted angle based on the fused feature vector. The state space of the reinforcement learning model is used to determine target control parameters based on the predicted fused feature vector constructed from the predicted speed and the predicted angle. The target control parameters include speed adjustment parameters and / or angle adjustment parameters for the UAV.

8. The UAV landing control method according to claim 2, characterized in that, The attention weights of the multimodal machine learning model are determined based on the time-of-arrival distance between the drone and each of the RFID tags.

9. The UAV landing control method according to claim 1, characterized in that, The process of controlling the UAV to land based on the target control parameters also includes: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. If the flight state of the UAV is determined to be deviated based on the predicted flight state vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

10. The UAV landing control method according to claim 1, characterized in that, The process of controlling the UAV to land based on the target control parameters also includes: The historical flight state vector of the UAV is input into the state prediction model to obtain the predicted flight state vector output by the state prediction model. Based on the predicted flight state vector and the observation noise vector, the observation vector of the UAV is determined, wherein the observation noise vector is determined based on the environmental data; If the flight state of the UAV is determined to be deviated based on the observation vector, the process returns to the step of determining the target control parameters based on the image data, the radio frequency data, and the environmental data.

11. The UAV landing control method according to claim 9 or 10, characterized in that, Also includes: If the UAV has completed landing based on the predicted flight state vector, the image data, the radio frequency data, the environmental data, and the target control parameters are stored.

12. The UAV landing control method according to claim 1, characterized in that, Also includes: If the drone has not completed landing, return to the steps of acquiring image data from the visual marker tag, radio frequency data transmitted by multiple radio frequency tags attached to the visual marker tag, and environmental data of the drone.

13. The UAV landing control method according to claim 1, characterized in that, The visual marker label has four radio frequency tags arranged in an array.

14. A drone landing control device, characterized in that, include: The acquisition module is used to acquire image data of the visual marker tag, radio frequency data transmitted by multiple radio frequency tags set on the visual marker tag, and environmental data of the UAV; The determination module is used to determine target control parameters based on the image data, the radio frequency data, and the environmental data; The control module is used to control the UAV to land based on the target control parameters.

15. A landing control system for unmanned aerial vehicles (UAVs), characterized in that, This includes drones and visual marker tags, wherein the visual marker tags are equipped with multiple radio frequency tags; The drone is used to acquire image data of visual marker tags, radio frequency data transmitted by multiple radio frequency tags set on the visual marker tags, and environmental data of the drone; Based on the image data, the radio frequency data, and the environmental data, the target control parameters are determined. The drone is controlled to land based on the target control parameters.

16. The unmanned aerial vehicle landing control system according to claim 15, characterized in that, The drone integrates an RFID reader, a visual sensor, a laser sensor, and a sonar sensor. The RFID reader is used to receive RFID data sent by multiple RFID tags, the visual sensor is used to acquire image data of the visual tags, the laser sensor is used to acquire laser data, and the sonar sensor is used to acquire sonar data.

17. The UAV landing control system according to claim 15 or 16, characterized in that, The drone is also used for offline pre-training of multimodal machine learning models and online fine-tuning of the multimodal machine learning models.

18. The unmanned aerial vehicle landing control system according to claim 15, characterized in that, The multiple radio frequency tags are integrated in an array on the visual tag.

19. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the UAV landing control method as described in any one of claims 1 to 13.

20. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the UAV landing control method as described in any one of claims 1 to 13.

21. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the UAV landing control method as described in any one of claims 1 to 13.

Images