Low-altitude unmanned aerial vehicle traffic operation risk early warning method and system

Abstract

The invention provides a low-altitude unmanned aerial vehicle traffic operation risk early warning method and system, and the method comprises the steps: collecting unmanned aerial vehicle state data and external environment data, carrying out the data preprocessing, and generating fusion environment perception data; inputting the fused environmental perception data into a lightweight multi-channel convolutional network, and outputting a local risk score tensor; generating a risk map sequence and a risk drift point sequence in a future time period through a double-input gating recurrent neural network; intercepting an ROI region extending by taking the current position of the aircraft as a starting point and taking the current speed direction as a vector from the risk map sequence, extracting an average risk value of each frame in the ROI region to calculate a future region risk level, and generating a decision suggestion mark; and generating an executable flight control instruction according to the decision suggestion sign to drive the unmanned aerial vehicle to execute. According to the invention, the real-time response capability and safety of the low-altitude unmanned aerial vehicle in a complex flight environment are significantly improved.

Description

Technical Field

[0001] The present invention belongs to the field of low-altitude UAV traffic, and particularly relates to a method and system for early warning of low-altitude UAV traffic operation risks. Background Art

[0002] With the rapid development of UAV technology, low-altitude UAVs have become important application tools in multiple fields such as transportation, logistics, agriculture, and environmental monitoring. The wide application of low-altitude UAVs has brought about a high degree of competition for airspace resources. Especially in densely populated flight areas such as cities and industrial areas, the risks of low-altitude flight are increasing continuously. In order to ensure the flight safety of UAVs in complex environments, various low-altitude UAV risk early warning and anti-collision systems have been proposed at home and abroad. These systems usually rely on the real-time perception data of UAVs and combine traditional flight path planning and control methods to achieve risk avoidance and flight safety management.

[0003] Existing low-altitude UAV traffic management technologies mainly rely on two methods: one is environment perception and local path planning based on sensors. For example, visual sensors, lidar, infrared sensors, etc. are used to monitor the surrounding environment in real time, and the flight trajectory is adjusted through local obstacle avoidance algorithms. The other method is to conduct aircraft scheduling and risk early warning through airspace monitoring and traffic flow management, usually based on static flight paths and fixed rules for management. However, existing technologies still have obvious limitations in dealing with the highly dynamic environment faced by low-altitude UAVs. First of all, the sensor-based risk early warning system lacks flexible adaptability to the rapidly changing flight environment. For example, most existing obstacle avoidance algorithms rely on simple static rules and cannot cope with emergencies or complex airspace environments (such as complex airflows, temporary obstacles, etc.). The airspace management system, on the other hand, relies on fixed flight paths and regular scheduling, fails to adjust and optimize the path in real time, and cannot respond in a timely manner in the face of emergencies. These technologies cannot fully handle the dynamic risks encountered by UAVs during actual flight, such as high flight density, meteorological changes, the appearance of unexpected obstacles, and other factors.

[0004] In addition, most existing low-altitude UAV risk early warning technologies ignore the specific requirements of different flight missions and airspace environments, and fail to comprehensively consider the real-time changes in the airspace and the mission objectives of UAVs for dynamic optimization, resulting in response delays in the system under complex flight scenarios, and even unable to provide timely and effective early warnings in high-risk situations. Generally speaking, existing technologies cannot provide comprehensive, flexible, and real-time risk early warning and behavior adjustment in the complex and changeable low-altitude environment, and there is still much room for improvement in terms of adaptability, efficiency, and accuracy. Summary of the Invention

[0005] The object of the present invention is to propose a method and system for early warning of low-altitude UAV traffic operation risks. Based on real-time environmental perception data and an efficient dynamic risk prediction model, it can flexibly respond to sudden changes in the low-altitude flight environment, adjust the flight path in real time, and provide timely warnings.

[0006] To achieve the above object, in the first aspect of the present invention, a method for early warning of low-altitude UAV traffic operation risks is provided. The method includes: S1. Collect UAV state data and external environment data, perform data preprocessing, and generate fused environmental perception data. Among them, the UAV state data includes position, speed, and flight altitude; the external environment data is sensor data, and the sensors include image data, lidar, and infrared sensors; S2. Input the fused environmental perception data into a lightweight multi-channel convolutional network, and combine it with a spatial attention module to strengthen regional significant risk features, and output a local risk score tensor, indicating the risk degree score at the moment corresponding to the spatial position, and the dimension is consistent with the environmental ground Figure 1 That is, the local risk score tensor is a risk score map, and identify the coordinates of high-risk areas in the risk score map to construct a set of high-risk area coordinates; S3. Obtain a historical risk map sequence, a historical flight state sequence, and a historical sensor dynamic weight sequence according to the risk score map and UAV state data. Combine the current risk score map and the set of high-risk area coordinates, and generate a risk map sequence and a risk drift point sequence for the future time period through a dual-input gated recurrent neural network; S4. Intercept an ROI area starting from the current position of the aircraft and extending in the direction of the current speed vector in the risk map sequence. Extract the average risk value of each frame in the ROI area to calculate the future area risk level. At the same time, calculate the minimum distance between all high-risk area coordinates and the current heading trajectory of the aircraft during this period, and count the number of high-risk points entering the ROI area to construct a high-risk point threat coefficient. Combine the future area risk level, the current speed of the aircraft, and the task urgency to generate a decision recommendation flag; S5. Generate an executable flight control instruction according to the decision recommendation flag to drive the UAV to execute.

[0007] Optionally, the structure of the lightweight multi-channel convolutional network is a three-layer convolutional structure, where the size of the convolutional kernel of each layer is , and the number of channels is 16, 32, and 64 respectively. Among them, the first layer processes image data, the second layer processes the point cloud map of lidar, and the third layer combines the thermal imaging channel of the infrared sensor to realize multi-modal feature joint coding; Among them, a spatial attention module is added after each convolutional layer to strengthen regional significant risk features; and the risk score map is smoothed by spatial gradient regularization constraints to suppress noise interference.

[0008] Optionally, identifying the coordinates of high-risk areas in the risk score map to construct a set of high-risk area coordinates specifically includes: Regarding the risk score map as a high-risk candidate area, and combining the aircraft speed direction and the predicted flight path to determine whether there are high-risk intersection points within the path; if so, record its spatial coordinates and the corresponding risk value as the current high-risk area coordinates.

[0009] Optionally, the dual-input gated recurrent neural network is composed of two parallel GRU modules, specifically: The first GRU module is used to process the risk map sequence, and the second GRU module processes the multi-variable vector sequence. After outputting the intermediate features, they are concatenated and fused, and then the predicted risk map sequence for several future time steps is output through the fusion layer; where the multi-variable vector sequence is the combined flight state, sensor response, and high-risk point density sequence; The first GRU module and the second GRU module are two groups of three-layer 128-unit GRU sub-networks.

[0010] Optionally, the step of generating the risk drift point sequence includes: Obtain the current high-risk point set and the current horizontal speed component of the aircraft, and perform a drift mapping on the current high-risk point set based on the current horizontal speed component of the aircraft to generate a risk drift point sequence.

[0011] Optionally, the future area risk level is calculated as follows: where T is the time window, is the set of grid positions included in the ROI area in the th frame, is the risk prediction value of this point; (x, y) is the current horizontal speed component of the aircraft; The threat coefficient of the high-risk point is constructed as follows: where, is an indicator function used to count the number of high-risk points falling into the ROI area, has the highest value, indicating that the future flight path has the highest intersection degree with the high-risk area; is the current time window; is the current risk drift point sequence.

[0012] Optionally, generating a decision-making recommendation flag by combining the future area risk level, the current speed of the aircraft, and the mission urgency specifically includes: Dividing by using a rule tree method, where is the decision-making recommendation flag, is the mission urgency: If and high , then , indicating continue to fly; If and , then , then it means to maintain the current state and continue to monitor; If or , then: If : , it is recommended to hover in place or wait; If : , it is recommended to plan a detour path.

[0013] Optionally, generating an executable flight control instruction according to the decision-making recommendation flag to drive the UAV to execute specifically includes: If , then the UAV maintains the current speed vector direction, sets the target acceleration to zero, and only makes speed maintenance and attitude stability adjustments; If , then it is the same as , but the high-frequency status monitoring flag is enabled to enter the high-frequency sampling mode, and the system re-checks the risk level every 0.2 s; If , then a deceleration and altitude locking instruction is issued; by applying a reverse acceleration in the current movement direction, the speed vector is attenuated to zero, and at the same time, the altitude locking module is activated to keep the current altitude unchanged; If , then it deviates from the current position of the aircraft by a certain angle in the direction away from the high-risk point, constructs an avoidance course vector, and gives an acceleration direction adjustment.

[0014] Optionally, each generated control instruction is encapsulated into a standard MAVLink control frame and sent to the flight control system for execution via the communication module; the control instruction includes acceleration and an avoidance course vector.

[0015] In another aspect of the present invention, a low-altitude UAV traffic operation risk warning system is provided, and the system includes: Data acquisition module: It is used to collect the UAV status data and external environment data, perform data preprocessing, and generate fused environment perception data. Among them, the UAV status data includes position, speed, and flight altitude; the external environment data is sensor data, and the sensors include image data, lidar, and infrared sensors. Risk identification module: It is used to input the fused environment perception data into a lightweight multi-channel convolutional network, and combine it with a spatial attention module to strengthen the regional significant risk features, and output a local risk score tensor, indicating the risk degree score at the corresponding spatial position at time , with the dimension consistent with the environment Figure 1 , that is, the local risk score tensor is a risk score map, and identify the high-risk area coordinates in the risk score map to construct a high-risk area coordinate set. Risk prediction module: It is used to obtain a historical risk map sequence, a historical flight state sequence, and a historical sensor dynamic weight sequence according to the risk score map and the UAV status data. Combine the current risk score map and the high-risk area coordinate set, and generate a risk map sequence and a risk drift point sequence for the future time period through a dual-input gated recurrent neural network. Decision support module: It is used to intercept an ROI area starting from the current position of the aircraft and extending in the direction of the current speed vector in the risk map sequence, extract the average risk value of each frame in the ROI area to calculate the future area risk level, and at the same time calculate the minimum distance between all high-risk area coordinates and the current flight path of the aircraft during this period, and count the number of high-risk points entering the ROI area to construct a high-risk point threat coefficient, and combine the future area risk level, the current speed of the aircraft, and the task urgency to generate a decision recommendation flag. Flight control module: It is used to generate executable flight control instructions according to the decision recommendation flag to drive the UAV to execute.

[0016] The beneficial technical effects of the present invention are at least as follows: To address the above problems in the existing low-altitude UAV risk warning systems, the present invention proposes a novel low-altitude UAV traffic operation risk warning method and system. Based on real-time environmental perception data and an efficient dynamic risk prediction model, this system can flexibly respond to sudden changes in the low-altitude flight environment, adjust the flight path in real time, and provide timely warnings. The main innovation of the present invention lies in the combination of real-time environmental perception and dynamic airspace risk prediction, breaking the limitations of traditional static path planning and local obstacle avoidance. First, by obtaining detailed surrounding environmental information through multi-source sensor data (such as vision, lidar, infrared sensors, etc.), the system can accurately identify real-time environmental factors such as aircraft, obstacles, and weather changes. At the same time, the system incorporates a dynamic airspace risk prediction module, which continuously optimizes the flight path and decision-making strategies according to real-time changes in flight density, airspace status, and meteorological conditions, warns of potential risks in advance, and avoids collisions of aircraft in high-risk environments.

[0017] Different from the prior art, the present invention can predict changes in traffic density, aircraft behavior, weather impacts, etc. that may occur in the next few minutes to hours through in-depth fusion analysis of real-time sensor data and historical flight data, and combines these prediction results with real-time perception information to provide comprehensive risk assessment and real-time decision-making support. This innovation enables UAVs not only to make quick decisions in the current environment but also to prepare for risk avoidance in advance and optimize flight efficiency. Based on this dynamic adaptation and prediction optimization, the flight system can automatically adjust the flight path according to environmental changes, improving the safety and mission execution efficiency of UAVs in complex airspaces. Through this method, the present invention solves the limitations brought by the static and single warning mechanism in the prior art, significantly enhancing the real-time response ability and safety of low-altitude UAVs in complex flight environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on the following drawings without creative efforts.

[0019] Figure 1 It is a flowchart of a method for warning of low-altitude UAV traffic operation risks according to an embodiment of the present invention.

[0020] Figure 2 It is a framework diagram of a system for warning of low-altitude UAV traffic operation risks according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0022] As Figure 1 shown, a method for early warning of low-altitude UAV traffic operation risks provided by an embodiment of the present invention includes: S1. Collect UAV state data and external environment data, perform data preprocessing, and generate fused environment perception data; wherein, the UAV state data includes position, speed, and flight altitude; the external environment data is sensor data, and the sensors include image data, lidar, and infrared sensors.

[0023] Among them, in a low-altitude flight environment, accurately perceiving the surrounding environment is the primary condition for ensuring the safe flight of UAVs. Environment perception and data collection obtain real-time data around the aircraft by combining multiple sensors (such as vision, lidar, and infrared sensors), and these data will provide essential input information for subsequent risk identification, prediction, and flight path optimization. Especially during flight, the environment may change very quickly, so the real-time and accuracy of data collection and preprocessing are crucial.

[0024] The input data for this step includes the state data of the aircraft and the perception data of the external environment. Specifically: State data of the aircraft: Position ( ): The real-time coordinates of the aircraft in the airspace, usually obtained through the GPS system.

[0025] Speed ( ): The speed of the aircraft in each axis direction, which can be obtained through an inertial navigation system (INS) or a speed sensor.

[0026] Flight altitude ( ): The height of the aircraft relative to the ground or sea level, usually provided by a barometer or the GPS system.

[0027] Perception data of the external environment: Image data from a vision camera: Used to identify obstacles, aircraft, and ground targets. Especially in low-light conditions, this data is crucial.

[0028] Depth map from lidar (LiDAR): Used to calculate the distance between the aircraft and surrounding objects, as well as the precise positioning of obstacles.

[0029] Data from infrared sensors: Provide additional information in low-light or nighttime environments to help sense temperature differences and identify potential obstacles or aircraft.

[0030] The collection of this data will be transmitted to the data processing module in real time and, after preprocessing, fused into unified environmental perception data.

[0031] Specifically, the step details include: First, the aircraft performs environmental perception through multiple sensors. During low-altitude flight, the types and positions of obstacles can be very complex, and the aircraft needs to make decisions based on the surrounding environment at any time. Therefore, the diversity and real-time nature of sensor data are crucial.

[0032] Among them, the multi-sensor deployment in the low-altitude flight environment is designed to obtain real-time perception of the surrounding environment and aircraft status data. For this purpose, the drone is equipped with the following types of sensors: Visual camera: Provide image information of the environment. In the daytime or under good lighting conditions, the visual sensor can efficiently identify surrounding obstacles, aircraft, and ground targets. However, in nighttime or low-light environments, the effectiveness of the camera may be greatly reduced, and at this time, the infrared sensor can provide a strong supplement.

[0033] Light Detection and Ranging (LiDAR): LiDAR is a very important sensor in low-altitude flight. It can accurately measure the distance between the aircraft and surrounding objects and generate a high-precision three-dimensional environmental map. Especially in an environment with dense obstacles, the LiDAR data can provide real-time support for flight path planning.

[0034] Infrared sensor: When the aircraft is in special environments such as low light, nighttime, or haze, the infrared sensor can sense temperature differences and identify possible obstacles.

[0035] Furthermore, after the sensor data is collected, data preprocessing is required, which includes operations such as denoising, time synchronization, and data normalization. Denoising is to eliminate random errors and external interference in the sensor output to ensure data quality. Time synchronization is to ensure the time alignment of different sensor data so that the data of multiple sensors can be compared and fused in the same time dimension. For example, the sampling frequencies of LiDAR and the visual camera may be different, so it is necessary to align the data times of them. Data normalization is to unify the scales of sensor data so that data from different types of sensors can be processed within the same computational framework.

[0036] In specific operations, denoising techniques such as Kalman filtering and particle filtering will be applied. Kalman filtering can be used to optimize the estimation of sensor data, especially when affected by noise and dynamic changes, and can provide a more stable data output. Particle filtering can be used to estimate the uncertainty in complex dynamic systems, especially suitable for the prediction and tracking of aircraft states.

[0037] For example, the formula of Kalman filtering is as follows: Where, is the state estimate at the current time, is the Kalman gain, is the current observation value, is the observation matrix. is the state estimate of the previous step. is the Kalman gain, which determines the weight between the measurement data and the predicted data. is the current observed data. is the observation matrix, which is used to convert the state prediction value into the observation space.

[0038] Furthermore, the data fusion technology combines the data from different sensors to form a unified and accurate environmental perception data set. The innovation point of this step is to adopt an adaptive weighted fusion method, that is, to dynamically adjust the data weights of various sensors according to the flight conditions detected by the environmental perception module (such as light intensity, visibility, meteorological category, etc.). For example, in the night or low-light environment, the performance of the vision sensor decreases, and the system will correspondingly reduce the weight of the vision data and increase the weight of the infrared sensor; while in the sunny day, the visual image is clear, and the system will mainly rely on the visual data. In complex meteorological conditions with low visibility such as thick fog or strong wind, due to its strong penetration ability, the weight of lidar will be significantly increased to enhance the positioning accuracy of obstacles.

[0039] To implement this weight adjustment mechanism, the system will continuously analyze the data from the environmental auxiliary sensors of the aircraft (such as brightness sensors, barometers, hygrometers, etc.) and classify the current environmental conditions. According to different classification results (such as sunny / day / night / thick fog / strong wind, etc.), the configured weight strategy table will be automatically called to dynamically set the weighting coefficient.

[0040] For example: When detected as day / sunny condition: , , ; When detected as night / no street light condition: , , ; When detected as thick fog condition: , , ; Among them, represents the weighting coefficient of the vision sensor; represents the weighting coefficient of the lidar sensor; represents the weighting coefficient of the infrared sensor; specifically, the sensor weighted fusion calculation formula is as follows: Among them, is the fused environmental data, is the data of the rd sensor (vision, lidar, infrared in sequence), is the weighting coefficient dynamically selected by the system according to the environmental type at time (where i is or or ).

[0041] Through this mechanism, this method not only enhances the utilization efficiency of different sensors under their optimal working conditions, but also improves the system's adaptability to sudden environmental changes, ensuring that even in complex or rapidly changing low-altitude flight environments, high-precision environmental perception capabilities can be maintained.

[0042] Furthermore, the fused environmental perception data ( ), that is, the environmental data around the aircraft after fusion, including the real-time state and environmental characteristics of the aircraft, will be passed as input data to the subsequent step (step 2) for feature extraction and risk identification.

[0043] S2. Input the fused environmental perception data into a lightweight multi-channel convolutional network, combined with a spatial attention module, to enhance regional significant risk features, and output a local risk score tensor, representing the risk degree score at the corresponding spatial position at time , with the dimension consistent with the environmental Figure 1 , that is, the local risk score tensor is a risk score map, and identify the coordinates of high-risk areas in the risk score map to construct a set of high-risk area coordinates.

[0044] The goal of this step is to, during the real-time flight process, from the fused environmental perception data output in step one Extract environmental features related to the flight risk altitude and identify potential risk areas based on this. The environment where low-altitude drones fly is complex and changeable, including static obstacles (such as buildings, trees), dynamic targets (such as other drones, birds), and unstable factors (such as sudden air current changes, meteorological changes). Therefore, this step not only needs to extract geometric and motion features from the data, but also needs to estimate the risk level by combining spatio-temporal context. To achieve high robustness and real-time performance, the system uses a method that combines a lightweight convolutional network and temporal window sliding analysis to perform local feature extraction and immediate risk judgment on the environment.

[0045] Specifically, this step directly inherits the output of step one , that is, the fused multi-modal environmental perception data. First, the system will send it into a lightweight multi-channel convolutional feature extraction network. The network structure is a three-layer convolutional structure, where the size of each convolutional kernel is , and the number of channels is 16, 32, and 64 respectively. The first layer processes the original visual image , the second layer processes the point cloud map, and the third layer combines the thermal imaging channel to achieve joint encoding of multi-modal features. The key to this structure is that a spatial attention module is added after each convolutional layer to strengthen regional significant risk features (such as areas with high-density obstacle clusters or hot spots of objects approaching quickly).

[0046] Furthermore, in order to improve the accuracy of risk identification and the forward-looking nature of the system, this step introduces a sliding time analysis mechanism based on a local dynamic window. At each time step , not only the current is considered, but also the data of the past frames is cited for short-term behavior change trend analysis. This mechanism is completed through a temporal feature fusion layer, and outputs a local risk score tensor , indicating the risk degree score at the corresponding spatial position at time . The core formula for calculating the risk score is as follows: Among them, represents the output of the multi-channel convolutional feature extraction module, is the temporal weighting factor, which is adjusted exponentially according to the time distance from the current frame (such as ), is the normalization activation function, which is used to normalize the risk value to between .

[0047] Among them, is the fused perception data at time ; is a three - layer convolutional neural network feature extractor; is a temporal weighting factor, controlling that the weight of recent frames is high and the weight of distant frames is low; is the risk score tensor at the current moment, with dimensions corresponding to the environmental ground Figure 1 to indicate the risk level at each spatial position.

[0048] Subsequently, the system will extract the local maximum region in the risk score map as the high - risk candidate region, and combine the aircraft speed direction and the predicted flight path to determine whether there are high - risk intersection points within the path. If so, record its spatial coordinates and the corresponding risk value as the current real - time risk recognition output.

[0049] Finally, to avoid the system from over - responding under noise interference, the system introduces a regularization constraint based on the spatial domain gradient to smooth the high - frequency fluctuations caused by noise in the risk score map. The regularization term is defined as follows: where, and respectively represent the gradients of the risk map in the and directions, approximated by the Sobel operator. The regularization term is used to suppress the sudden change of local risk values, thereby enhancing the stability and interpretability of the recognition results. is the risk value at the position in the risk score map; is the smoothing regularization term, which is finally used as part of the confidence score in the system for risk result correction.

[0050] Furthermore, the output risk score map : used to calibrate the risk levels of different spatial positions within the current flight area; the set of high - risk area coordinates : the list of risk points output by the system after threshold segmentation and spatial screening, for use by subsequent prediction and path planning modules.

[0051] This step constructs a risk feature extraction mechanism centered on local temporal information, making full use of the multi - modal fusion data output in Step 1. Through joint modeling of spatial attention and time window, it realizes the real - time recognition of high - risk points on the flight path. Compared with traditional methods, this solution shows higher adaptability and robustness in identifying dynamic risks, in - path collision trends, etc., and is especially suitable for scenarios where targets appear quickly, are frequently occluded, and have unstable lighting in the low - altitude environment.

[0052] S3. Obtain a historical risk map sequence, a historical flight state sequence, and a historical sensor dynamic weight sequence based on the risk score map and the UAV status data. Combine the current risk score map and the set of high-risk area coordinates, and use a dual-input gated recurrent neural network to generate a risk map sequence and a risk drift point sequence for a future time period.

[0053] To enhance the forward-looking ability of the system, this step predicts the potential risk change trend during future flights through a multivariate time series analysis method. Different from the traditional time series modeling based only on the risk score map , this solution integrates multiple heterogeneous time dimension data (including the aircraft's own state sequence, environmental sensor response sequence, local risk dynamic changes, etc.) to construct a higher-dimensional and more interactive time series analysis model. On this basis, a dual-input gated recurrent neural network is designed to model the joint evolution of multiple variables, thereby improving the accuracy and robustness of future risk prediction. This mechanism is of great significance for dealing with multi-target interference, sudden meteorological changes, etc. in low-altitude complex dynamic scenarios.

[0054] Specifically, this step takes the output of step two as the main input and synchronously accesses the flight state and sensor behavior variables to form the following multivariate time series input: Current risk score map ; Set of high-risk points ; Historical risk map sequence , each frame being a gridded and normalized two-dimensional risk distribution; Historical flight state sequence , where ; Historical sensor dynamic weight sequence , where ; Sequence of the number of high-risk points .

[0055] Furthermore, these multivariate sequences jointly form a prediction input tensor group and are aligned in time steps through a synchronous time window processing mechanism.

[0056] To fuse the above multi-modal, multi-source, and multi-dimensional time series features, this step designs a dual-input gated recurrent neural network structure (Dual-GRU). This structure consists of two parallel GRU modules: one processes the risk map sequence , and the other processes the multi-variable vector sequence . After outputting the intermediate features, they are concatenated and fused, and then the fused layer outputs a predicted risk map sequence for several future time steps.

[0057] The core expression for risk map prediction of this structure is: Among them, and are two groups of GRU sub-networks with three layers and 128 units. represents the fusion layer, which consists of one layer of fully connected layer plus an activation function.

[0058] Among them, is the two-dimensional risk score map predicted at the future time moment; is the GRU model for processing the risk map sequence; is the GRU model for processing the multi-variable state vector sequence; is the function for fusing and transforming after concatenating the outputs of the two sequences; is the combined flight state, sensor response, and high-risk point density sequence.

[0059] The risk prediction map of the fusion output can identify the high-risk areas that may appear on the future flight trajectory. However, considering that low-altitude aircraft have continuous motion characteristics, only predicting the spatial distribution map is not sufficient to locate the precise conflict points. Therefore, the system synchronously adopts the forward risk point drift mechanism to perform drift mapping based on the flight speed for the current high-risk point set : This mechanism assumes that the high-risk area has a derivable spatial offset path under the inertial action of the aircraft, thereby further improving the prediction accuracy and response time.

[0060] Among them, is the position of the high-risk point after drift at the prediction time ; is the high-risk point identified at the current moment; is the current horizontal speed component of the aircraft; is the prediction time step offset value.

[0061] Output: Risk map sequence in the future time period ; Risk drift point sequence : Used to accurately predict the potential high-collision areas at future moments.

[0062] It is understandable that in this step, through the joint modeling of multivariate time series, not only the evolution trend of the spatial risk map is integrated, but also auxiliary information such as the aircraft's own behavior, sensor status, and local risk density is incorporated to form a highly coupled dynamic prediction model. The parallel design of the dual GRU structure reflects the flexibility of temporal information processing. Combining the drift inference strategy makes the risk prediction more practical and interpretable. This solution effectively solves the problem that single-variable modeling cannot handle complex interaction factors and is a core part of realizing dynamic early warning and active avoidance strategies.

[0063] S4. Intercept a Region of Interest (ROI) in the risk map sequence that starts from the current position of the aircraft and extends along the vector of the current speed direction. Extract the average risk value of each frame in the ROI to calculate the future regional risk level. At the same time, calculate the minimum distance between the coordinates of all high-risk regions and the current flight path of the aircraft during this period, and count the number of high-risk points entering the ROI to construct a threat coefficient for high-risk points. Combine the future regional risk level, the current speed of the aircraft, and the urgency of the task to generate a decision recommendation flag.

[0064] This step aims to aggregate and model the future risk information predicted in Step 3 to form a quantitative risk level and strategy recommendation for the current flight state. During the low-altitude UAV flight, the system not only needs to know whether there is a risk at a certain place, but also needs to form a comprehensive judgment on whether to take evasive actions at this moment. Therefore, this step designs a dynamic risk cumulative grading mechanism based on a multi-step prediction score map and the density of high-risk points, and introduces the current aircraft state parameters and task parameters for joint decision-making, outputting two indicators: the regional risk level ( ), and the action recommendation flag bit ( ).

[0065] Specifically, the core idea of this step is to aggregate the risk scores in the future risk map along the area of the aircraft's predicted flight path to form a regional risk integral, and then combine information such as the density trend of high-risk points and the current flight speed to output two quantifiable indicators.

[0066] First, the system intercepts a dynamic Region of Interest (ROI) in that starts from the current position of the aircraft and extends along the vector of the current speed direction. This area simulates the natural forward flight segment of the UAV without evasion. Then, extract the average risk value of each frame in this ROI to construct the future regional risk level: where, is the The set of grid positions included in the ROI region in the frame, is the risk prediction value of this point. This metric reflects the average risk level faced by the aircraft when moving forward in the current flight direction in the next several frames.

[0067] Then, the system calculates all high-risk points during the prediction period and the minimum distance between them and the current heading trajectory of the aircraft, and counts the number of high-risk points entering the ROI region to construct a high-risk point threat coefficient: Among them, is an indicator function used to count the number of high-risk points falling into the ROI region. The higher the value of, the higher the intersection degree between the future flight path and the high-risk area.

[0068] Finally, the system combines , , the current speed magnitude and the mission urgency , and enters a hierarchical strategy model. This model is divided in the following way using a rule tree: If and , then (continue flying); If and , then (maintain the current state and continue monitoring); If or , then: If : (recommend hovering in place or waiting); If : (recommend detour path planning).

[0069] Furthermore, output: future area risk level : used to quantify the risk aggregation value of the future flight path area; decision recommendation flag : an action flag with values from 0 to 3, for subsequent path planning modules to call.

[0070] This step is based on future multi-frame risk maps and high-risk point prediction results, combines the current flight direction and mission requirements, constructs a comprehensive risk level through spatio-temporal ROI risk integration and risk point density assessment, and outputs action suggestions based on regularization criteria. Different from the traditional rule + threshold judgment method, this solution can take into account both risk maps and drift points, comprehensively evaluate the persistent risk trend in the prediction space, and realize an accurate and flexible decision support module.

[0071] S5. Generate executable flight control instructions according to the decision suggestion flag to drive the UAV to execute.

[0072] Specifically, based on the completion of all risk perception, analysis, and prediction, this step is responsible for transforming the behavior suggestions output in Step 4 and the regional risk level

[0073] into executable flight control instructions to achieve actions such as path avoidance, speed adjustment, or hovering in place. This step no longer performs modeling or evaluation, but directly calls the action control module to implement the flight strategy, thus completing the execution link of the entire low-altitude risk warning closed-loop. The system enters different control strategy templates according to the value and generates corresponding control instruction sets

[0074] under each strategy. The control instructions are expressed in the form of acceleration vectors and angular velocity instructions and are compatible with conventional UAV flight control interfaces. If (continue flying): The system maintains the current speed vector direction, sets the target acceleration to zero, and only makes speed maintenance and attitude stability adjustments: where is the attitude angular velocity control amount, and

[0075] (continue monitoring): The same as

[0076] If (recommended to hover): The system issues deceleration and altitude lock instructions. Apply reverse acceleration in the current movement direction to decay the speed vector to zero, and at the same time activate the altitude lock module to keep the current altitude unchanged: where is the deceleration coefficient, generally taking a value of 0.8 to 1.2, is the current velocity vector, represents the rate of change of altitude, and setting it to zero means altitude locking.

[0077] If (recommended to avoid): The system offsets a certain angle from the current position of the aircraft in the direction away from the high-risk point to construct an avoidance heading vector , and gives an adjustment of the acceleration direction: Among them, is the maximum allowable horizontal acceleration of the system (defined by flight control constraints), is the avoidance direction vector constructed in the reverse direction of the reference drift point direction. For example, if the high-risk point is in the front left of the current flight path, the system selects the right rear direction to correct the heading.

[0078] Each generated control command will be encapsulated into a standard MAVLink control frame and sent to the flight control system for execution via the communication module.

[0079] Such as Figure 2 shown, in another embodiment of the present invention, a low-altitude UAV traffic operation risk warning system is provided, and the system includes: Data acquisition module 301: used to acquire UAV state data and external environment data, perform data preprocessing, and generate fused environmental perception data; among them, the UAV state data includes position, speed, and flight altitude; the external environment data is sensor data, and the sensors include image data, lidar, and infrared sensors; Risk identification module 302: used to input the fused environmental perception data into a lightweight multi-channel convolutional network, combined with a spatial attention module to strengthen regional significant risk features, and output a local risk score tensor, representing the risk degree score at the corresponding spatial position at time , and the dimension is consistent with the environment Figure 1 That is, the local risk score tensor is a risk score map, and the coordinates of the high-risk area are identified in the risk score map to construct a high-risk area coordinate set; Risk prediction module 303: used to obtain a historical risk map sequence, a historical flight state sequence, and a historical sensor dynamic weight sequence according to the risk score map and UAV state data, combined with the current risk score map and the high-risk area coordinate set, and generate a risk map sequence and a risk drift point sequence for the future time period through a dual-input gated recurrent neural network; Decision Support Module 304: It is used to intercept an ROI area in the risk map sequence that starts from the current position of the aircraft and extends with the current speed direction as the vector, extract the average risk value of each frame in the ROI area to calculate the future area risk level, and at the same time calculate the minimum distance between all high-risk area coordinates and the current heading trajectory of the aircraft during the prediction period, and count the number of high-risk points entering the ROI area to construct a high-risk point threat coefficient, and combine the future area risk level, the current speed of the aircraft, and the urgency of the task to generate a decision recommendation flag; Flight Control Module 305: It is used to generate executable flight control instructions according to the decision recommendation flag to drive the UAV to execute.

[0080] It should be noted that the above-described work process is only illustrative and does not limit the protection scope of the present invention. In actual applications, those skilled in the art can select some or all of them according to actual needs to achieve the purpose of the solution of this embodiment, and no limitation is imposed here.

[0081] In addition, for the technical details not described in detail in this embodiment, reference can be made to the parameter operation method provided in any embodiment of the present invention, and details will not be repeated here.

[0082] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0083] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0084] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, 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 is stored in a storage medium (such as a read-only memory / random access memory, magnetic disk, optical disc), and includes several instructions for causing a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in various embodiments of the present invention.

[0085] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A method for early warning of low-altitude unmanned aerial vehicle traffic operation risks, characterized in that, The method includes: S1. Collect the state data of the UAV and the external environment data, perform data preprocessing, and generate fused environment perception data. Among them, the UAV state data includes position, speed, and flight altitude; the external environment data is sensor data, and the sensors include image data, lidar, and infrared sensors. S2. Input the fused environment perception data into a lightweight multi-channel convolutional network, combined with a spatial attention module to enhance regional significant risk features, and output a local risk score tensor, indicating the risk degree score at the spatial position corresponding to the moment The dimension is consistent with the environment map, that is, the local risk score tensor is a risk score map, and identify the coordinates of high-risk areas in the risk score map to construct a set of high-risk area coordinates; S3. Obtain a historical risk map sequence, a historical flight state sequence, and a historical sensor dynamic weight sequence according to the risk score map and the UAV state data. Combine the current risk score map and the high-risk area coordinate set, and generate a risk map sequence and a risk drift point sequence for a future time period through a dual-input gated recurrent neural network. S4. Intercept an ROI area starting from the current position of the aircraft and extending in the direction of the current speed vector in the risk map sequence. Extract the average risk value of each frame in the ROI area to calculate the future area risk level. At the same time, calculate the minimum distance between all high-risk area coordinates and the current flight path of the aircraft during this period, and count the number of high-risk points entering the ROI area to construct a high-risk point threat coefficient. Combine the future area risk level, the current speed of the aircraft, and the task urgency to generate a decision recommendation flag. S5. Generate an executable flight control instruction according to the decision recommendation flag to drive the UAV to execute.

2. The low-altitude UAV traffic operation risk warning method according to claim 1, wherein The structure of the lightweight multi-channel convolutional network is a three-layer convolutional structure, where the size of the convolutional kernel in each layer is , and the number of channels is 16, 32, and 64 respectively; among them, the first layer processes image data, the second layer processes the point cloud map of the lidar, and the third layer combines the thermal imaging channel of the infrared sensor to achieve joint encoding of multi-modal features; Among them, a spatial attention module is added after each convolutional layer to strengthen the regional significant risk features; and the risk score map is smoothed by the spatial gradient regularization constraint to suppress noise interference.

3. The low-altitude UAV traffic operation risk warning method according to claim 2, wherein, The step of identifying high-risk area coordinates in the risk score map to construct a high-risk area coordinate set specifically includes: Regard the risk score map as a high-risk candidate area, and combine the aircraft speed direction and the predicted flight path to determine whether there are high-risk intersection points within the path. If so, record its spatial coordinates and the corresponding risk value as the current high-risk area coordinates.

4. A low-altitude UAV traffic operation risk warning method according to claim 1, characterized in that, The dual-input gated recurrent neural network is composed of two parallel GRU modules, specifically: The first GRU module is used to process the risk map sequence, and the second GRU module processes the multi-variable vector sequence. After outputting the intermediate features, they are spliced and fused, and then the predicted risk map sequence for several future time steps is output through the fusion layer. Among them, the multi-variable vector sequence is a combined flight state, sensor response, and high-risk point density sequence. The first GRU module and the second GRU module are two groups of three-layer 128-unit GRU sub-networks.

5. A low-altitude UAV traffic operation risk warning method according to claim 4, characterized in that The step of generating the risk drift point sequence includes: Obtain the current high-risk point set and the current horizontal speed component of the aircraft, and perform a drift mapping on the current high-risk point set based on the current horizontal speed component of the aircraft to generate a risk drift point sequence.

6. The low-altitude UAV traffic operation risk early warning method according to claim 1, wherein The future regional risk level is calculated as follows: where T is the time window, is the set of grid positions included in the ROI region in the th frame, is the risk prediction value of this point; (x, y) is the current horizontal velocity component of the aircraft; The threat coefficient of the high-risk point , is constructed as follows: Among them, is an indicator function used to count the number of high-risk points falling within the ROI area. has the highest value, indicating that the future flight path has the highest intersection degree with the high-risk area. is the current time window; is the current risk drift point sequence.

7. A method for early warning of low-altitude UAV traffic operation risks according to claim 6, characterized in that, The step of combining the future area risk level, the current speed of the aircraft, and the task urgency to generate a decision recommendation flag specifically includes: Divided by the rule tree method, where is the decision-making advice flag, is the task urgency level: If and the height , then , indicating continued flight; If and , then , it means to maintain the current state and continue monitoring; If or , then: If : , it is recommended to hover in place or wait; If : , it is recommended to plan a detour route.

8. A low-altitude UAV traffic operation risk warning method according to claim 7, characterized in that, The step of generating an executable flight control instruction according to the decision recommendation flag to drive the UAV to execute specifically includes: If , the UAV maintains the current speed vector direction, sets the target acceleration to zero, and only makes speed maintenance and attitude stability adjustments; If , it is the same as , but the high-frequency status monitoring flag is enabled and the high-frequency sampling mode is entered. The system rechecks the risk level every 0.2 s; If , issue a deceleration and altitude lock command; apply a reverse acceleration in the current direction of motion to decay the velocity vector to zero, and at the same time activate the altitude lock module to keep the current altitude unchanged; If , then deviate from the current position of the aircraft in a direction away from the high-risk point by a certain angle, construct an avoidance course vector, and give an adjustment to the acceleration direction.

9. A method for early warning of low-altitude UAV traffic operation risks according to claim 8, characterized in that, Each generated control instruction will be encapsulated into a standard MAVLink control frame and sent to the flight control system through the communication module for execution; the control instruction includes acceleration and avoidance heading vector.

10. A low-altitude UAV traffic operation risk warning system, characterized in that, The system includes: Data acquisition module: used to collect the UAV state data and external environment data, perform data preprocessing, and generate fused environment perception data; among them, the UAV state data includes position, speed, and flight altitude; the external environment data is sensor data, and the sensors include image data, lidar, and infrared sensors; Risk identification module: used to input the fused environment perception data into a lightweight multi-channel convolutional network, combined with a spatial attention module to enhance regional significant risk features, and output a local risk score tensor, representing the risk degree score at the corresponding spatial position at time The dimension is the same as that of the environment map, that is, the local risk score tensor is a risk score map, and the coordinates of high-risk areas are identified in the risk score map to construct a set of high-risk area coordinates; Risk prediction module: used to obtain the historical risk map sequence, historical flight state sequence, and historical sensor dynamic weight sequence according to the risk score map and UAV state data, and combine the current risk score map and the high-risk area coordinate set. Through a dual-input gated recurrent neural network, generate the risk map sequence and risk drift point sequence for the future time period; Decision support module: used to intercept an ROI area starting from the current position of the aircraft and extending in the direction of the current speed vector in the risk map sequence, extract the average risk value of each frame in the ROI area to calculate the future area risk level, and at the same time calculate the minimum distance between all high-risk area coordinates and the current heading trajectory of the aircraft during this period, and count the number of high-risk points entering the ROI area to construct a high-risk point threat coefficient, and combine the future area risk level, the current speed of the aircraft, and the task urgency to generate a decision recommendation flag; Flight control module: used to generate executable flight control instructions according to the decision recommendation flag to drive the UAV to execute.

Images