Unmanned aerial vehicle mounted integrated air pollution detection method

Abstract

The invention relates to the technical field of unmanned aerial vehicle intelligent environment monitoring, in particular to an unmanned aerial vehicle mounted integrated air pollution detection method, which comprises the following steps of: constructing a shared global pollution probability model by using a multi-machine communication network and Gaussian process regression, and generating an initial collaborative waypoint sequence for an unmanned aerial vehicle cluster according to information entropy distribution; the unmanned aerial vehicle collects multi-dimensional monitoring data in the process of executing a flight mission, and carries out space-time lag compensation according to the response characteristics of the sensor and the flight speed, thereby obtaining accurate space-time calibration data, and determining and estimating the position of a pollution source near the local maximum value of the concentration. A multi-machine communication network gathers calibration data and pollution source estimation results in real time, online training and parameter updating are carried out on a shared pollution probability model, a system finally analyzes the updated model, a high-uncertainty area is extracted according to prediction variance, a dynamic position covering waypoint is generated for an unmanned aerial vehicle which does not trigger re-planning, and the unmanned aerial vehicle is subjected to dynamic position covering. And multi-machine cooperative self-adaptive pollution monitoring is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent environmental monitoring technology for unmanned aerial vehicles (UAVs), and more particularly to an integrated air pollution detection method mounted on a UAV. Background Technology

[0002] With the continuous growth in the number of industrial parks, chemical plants, and urban solid waste treatment facilities, sudden pollution emissions and covert illegal discharges are characterized by high randomness, high diffusion, and high hazard. Traditional fixed-point monitoring is insufficient to capture the dynamics of pollution plumes in a timely manner. Unmanned aerial vehicles (UAVs), as highly mobile and rapidly deployable monitoring platforms, have been increasingly used for air quality inspections and pollution source location. However, single-unit flight time is limited, spatial coverage is insufficient, and it is difficult to maintain model continuity and detection efficiency when facing transient pollution diffusion processes. To address the frequent changes in pollution plume morphology with wind fields, it is necessary to introduce an intelligent monitoring mechanism that can continuously update environmental knowledge and adaptively adjust its path during flight. Simultaneously, dynamic supplementary measurements of uncertain areas need to be achieved at the multi-unit collaborative level to obtain high-resolution, high-reliability spatial distribution data of the pollution field.

[0003] Most existing drone air patrol solutions rely on pre-planned fixed grid tracks, lacking the ability to utilize real-time observation information. Even when pollution concentration suddenly increases, they cannot interrupt the predetermined route for in-depth tracking, resulting in low efficiency in locating pollution sources. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an integrated air pollution detection method mounted on a drone, aiming to improve the problem that most existing drone air patrol schemes rely on pre-planned fixed grid tracks and lack the ability to utilize real-time observation information.

[0005] This invention provides the following technical solution: an integrated air pollution detection method mounted on a drone, comprising the following steps: S1. Establish a multi-machine communication network and construct a shared global contamination probability model based on Gaussian process regression. Allocate non-overlapping initial cooperative waypoint sequences to the UAV cluster according to the information entropy distribution. S2. Control the UAV cluster to execute the flight mission corresponding to the initial cooperative waypoint sequence, and collect raw multidimensional monitoring data including three-dimensional position, meteorological parameters and pollutant concentration; S3. Based on the preset sensor response time constant and the UAV flight speed, perform spatiotemporal lag compensation on the original multidimensional monitoring data to generate spatiotemporal calibration monitoring data; S4. Calculate the spatial change rate of the spatiotemporal calibration monitoring data to obtain the concentration gradient vector. When the magnitude of the concentration gradient vector is greater than a preset threshold, generate a path replanning trigger command. S5. In response to the path replanning trigger command, control the single UAV that triggered the command to interrupt the original route, generate a tracking waypoint based on the direction of the concentration gradient vector and the wind direction data in the meteorological parameters, and determine the estimated pollution source coordinates when the monitoring data shows a local maximum. S6. By aggregating the spatiotemporal calibration monitoring data and the estimated pollution source coordinates through the multi-machine communication network, the shared global pollution probability model is trained and updated online to obtain the updated shared global pollution probability model. S7. Analyze the updated shared global pollution probability model to extract the coordinates of regions where the prediction variance exceeds a preset uncertainty threshold, and generate dynamic collaborative waypoints for members of the UAV cluster that have not triggered replanning.

[0006] Preferably, in step S1, the process of establishing a multi-machine communication network and constructing a shared global contamination probability model based on Gaussian process regression specifically includes the following steps: Configure the self-organizing network communication protocol parameters and time synchronization benchmark for each node of the drone swarm; The squared exponential kernel function was selected as the covariance function to describe the spatial correlation of pollutant concentrations, and a constant mean function was set. Initialize a hyperparameter set containing signal variance and length scale, establish a Gaussian process regression prior distribution that maps three-dimensional spatial coordinates to pollutant concentration, and use it as a shared global pollution probability model.

[0007] Preferably, in step S2, controlling the UAV cluster to execute the flight mission corresponding to the initial cooperative waypoint sequence specifically includes the following steps: The coordinate data in the initial cooperative waypoint sequence is input into the UAV flight control system to generate attitude control commands for the drive power unit; Data acquisition is performed by synchronously triggering the satellite positioning module, airborne anemometer, and gas sensor according to the preset sampling frequency. The collected latitude, longitude, and altitude data, wind speed and direction data, and pollutant concentration readings are linked and bound to the current system timestamp to form raw multidimensional monitoring data.

[0008] Preferably, in step S3, the step of performing spatiotemporal lag compensation on the original multidimensional monitoring data to generate spatiotemporal calibration monitoring data specifically includes the following steps: Extract the recording time, three-dimensional position coordinates, and real-time flight speed vector from the original multidimensional monitoring data; The spatial lag displacement vector is obtained by multiplying the real-time flight velocity vector with the preset sensor response time constant. Calculate the vector difference between the three-dimensional position coordinates and the spatial hysteresis displacement vector to obtain the calibrated actual sampling position coordinates; Establish a mapping relationship between pollutant concentration and the coordinates of the actual sampling location to generate spatiotemporal calibration monitoring data.

[0009] Preferably, in step S4, calculating the spatial change rate of the spatiotemporal calibration monitoring data to obtain the concentration gradient vector specifically includes the following steps: Extract the position coordinates and concentration values ​​of the current sampling point and the previous sampling point that are continuous in time from the spatiotemporal calibration monitoring data; Calculate the displacement vector of the current sampling point relative to the previous sampling point, and calculate the concentration difference between the two points; Calculate the ratio of the concentration difference to the displacement vector magnitude, use this ratio as the modulus, and construct a concentration gradient vector with the direction of the displacement vector as the reference direction.

[0010] Preferably, in step S5, generating tracking waypoints based on the direction of the concentration gradient vector and the wind direction data in the meteorological parameters specifically includes the following steps: The concentration gradient vector is normalized to obtain the gradient direction vector, and the headwind vector is calculated based on meteorological parameters. The gradient direction vector and the headwind vector are weighted and summed using preset weighting coefficients to obtain the fused search direction vector; Starting from the current position of the UAV, extend the fusion search direction vector by a preset search step length, and the calculated target position coordinates are the tracking waypoints.

[0011] Preferably, in step S6, the step of aggregating the spatiotemporal calibration monitoring data and the estimated pollution source coordinates through the multi-machine communication network specifically includes the following steps: The spatiotemporal calibration monitoring data and estimated pollution source coordinates are encapsulated into a status update data package containing source node identifiers and generation timestamps according to a preset data format. The status update data packet is sent to other nodes in the UAV cluster using the broadcast protocol of the multi-machine communication network; The receiving node performs integrity verification on the status update data packet. After the verification is successful, it parses and extracts the data content and stores it in the local synchronization dataset.

[0012] Preferably, in step S7, the step of parsing the updated shared global contamination probability model to extract the coordinates of regions where the prediction variance exceeds a preset uncertainty threshold specifically includes the following steps: Input the coordinates of discretized grid points within the preset monitoring area into the updated shared global pollution probability model, and calculate the posterior variance value corresponding to each grid point; Each posterior variance value is compared with a preset uncertainty threshold, and high-entropy grids with variance values ​​greater than the threshold are selected. Extract the geometric center coordinates of the high-entropy grid and use them as the coordinates of the areas that need to be focused on detection.

[0013] The present invention has the following beneficial effects: 1. In this invention, by introducing a perception-based path replanning mechanism, the UAV can immediately interrupt its original route and switch to plume tracking mode when it detects an abnormal increase in the concentration gradient. This realizes the transformation from fixed-route patrol to real-time adaptive detection, thereby significantly shortening the response cycle from the discovery of pollution anomalies to the locating of potential sewage outlets, and effectively improving the timeliness and pertinence of emergency law enforcement.

[0014] 2. In this invention, a multi-machine collaborative sampling strategy based on information entropy is adopted, which enables the UAV swarm to share a unified pollution probability model and automatically schedule maneuvers to areas with high uncertainty in the model. This achieves intelligent spatial allocation of monitoring resources, thereby avoiding the problems of insufficient coverage by a single machine and redundant operations by the formation. Higher quality pollution field estimation results are obtained with fewer flight sorties, effectively improving the comprehensiveness and efficiency of air quality monitoring.

[0015] 3. In this invention, by constructing a unified data sharing and model synchronization mechanism within the cluster, each UAV can obtain a globally consistent pollution probability field at any time, thereby avoiding monitoring blind spots and decision-making biases caused by information lag in traditional distributed operations, and significantly improving the cluster's global situational awareness and collaborative autonomous decision-making level in dynamic and complex pollution environments. Attached Figure Description

[0016] Figure 1 This is a flowchart of an integrated air pollution detection method mounted on a drone, as proposed in this invention. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] This invention provides an integrated air pollution detection method mounted on a drone, such as... Figure 1 As shown, it includes the following steps: S1. Establish a multi-machine communication network and construct a shared global contamination probability model based on Gaussian process regression. Allocate non-overlapping initial cooperative waypoint sequences to the UAV cluster according to the information entropy distribution. Furthermore, in S1, the establishment of a multi-machine communication network and the construction of a shared global pollution probability model based on Gaussian process regression specifically include the following steps: Configure the self-organizing network communication protocol parameters and time synchronization benchmark for each node of the drone swarm; The squared exponential kernel function was selected as the covariance function to describe the spatial correlation of pollutant concentrations, and a constant mean function was set. Initialize a hyperparameter set containing signal variance and length scale, establish a Gaussian process regression prior distribution that maps three-dimensional spatial coordinates to pollutant concentration, and use it as a shared global pollution probability model.

[0019] Specifically, before the mission begins, the system first triggers multiple drones to enter the network initialization state through short-range wireless link commands, enabling them to automatically complete node discovery and topology establishment. Each drone loads communication protocol parameters according to the preset channel access method, neighbor scanning cycle and network handshake process, and after receiving the synchronization pulse, it uses the clock drift compensation mechanism to adjust the local clock to the central node time reference, so that the time error between nodes is kept within one millisecond, thereby ensuring the consistency of the time stamp of subsequent monitoring data.

[0020] After completing the construction of the multi-machine communication network, the system establishes a shared probability model of pollution concentration field based on the spatial characteristics of the air pollution monitoring task. The system treats the monitoring area as a continuous three-dimensional spatial field. Combining the diffusion and smooth change characteristics of pollutants in the atmosphere, a Gaussian process regression model is used to describe the spatial correlation of pollutant concentrations. The Gaussian process model is in the form of: ; in, A three-dimensional coordinate vector at any spatial location; The mean function is taken as a constant in this scheme. This indicates the overall average pollution level of the region. The system uses the squared exponential kernel function to determine spatial correlation, and its expression is as follows: ; in, and These are the three-dimensional coordinate vectors of two monitoring points in space, in meters; Let be the Euclidean distance between the two points; The signal variance reflects the true magnitude of the spatial variation in pollutant concentration. It is a length scale used to describe the range of correlation between pollutant concentration decay and spatial distance.

[0021] Based on historical pollution monitoring data, the system initializes the above hyperparameters using the maximum likelihood method, enabling the model to conform to the actual pollutant diffusion characteristics of the monitoring area. Once completed, the Gaussian process prior model is synchronized to all UAVs via the communication network for subsequent prediction and path decision-making.

[0022] After the model is built, the system will perform three-dimensional meshing of the monitoring area at a fixed resolution, forming a set of mesh points: ; in, For the number of grid points, For the first The system inputs the three-dimensional coordinates of each grid point into a Gaussian process model, calculates its prediction variance, and uses this variance to characterize the uncertainty in pollution concentration prediction. For any prediction point... have: ; in, The prior covariance of the prediction point itself; This is the covariance vector between the predicted point and all sampled points; This is the covariance matrix between the sampled points; To measure the noise variance; The larger the prediction variance value, the higher the uncertainty of the model at that location, indicating a lack of effective sampling information.

[0023] For regional priority assessment, the system normalizes all prediction variances by maximizing and minimizing them. Let the maximum prediction variance for the entire region be... The minimum value is Then the first The uncertainty weights for each grid point are: ; in, The weight is used to represent the potential contribution of the location to improving the overall accuracy of the model. The higher the weight, the more critical the point is to the modeling of the pollution field.

[0024] The system sorts the grid points from highest to lowest according to their uncertainty weight, and then further sorts them according to the number of drones. The high-weighted elements after ranking are clustered based on spatial distance. The system selects the grid point with the highest weight or its geometric center in each subregion as the initial cooperative waypoint for the corresponding UAV, forming a set of waypoints: ; in, To be assigned to the The system obtains the three-dimensional spatial waypoint coordinates of the UAVs and distributes the waypoints to each UAV through the communication network, enabling them to automatically navigate to different high-uncertainty areas at the initial stage of the mission, thereby achieving the decentralization of UAV detection paths and maximizing information gain.

[0025] Through this step, the UAV swarm can obtain a stable communication link and a consistent pollution probability model foundation before the mission begins. The uncertainty distribution constructed based on the Gaussian process prediction variance can accurately indicate the location of missing information in the monitoring area. Waypoint allocation based on uncertainty weights allows each UAV to autonomously enter different high uncertainty areas at the beginning of the mission, thereby avoiding duplicate sampling, improving monitoring coverage, accelerating the update speed of the pollution field model, reducing the overall prediction error, and significantly improving the collaborative detection efficiency and modeling accuracy of pollution monitoring missions.

[0026] S2. Control the UAV cluster to execute the flight mission corresponding to the initial coordinated waypoint sequence, and collect raw multidimensional monitoring data including three-dimensional position, meteorological parameters and pollutant concentration; Furthermore, in S2, controlling the UAV swarm to execute the flight mission corresponding to the initial cooperative waypoint sequence specifically includes the following steps: The coordinate data in the initial cooperative waypoint sequence is input into the UAV flight control system to generate attitude control commands for the drive power unit; Data acquisition is performed by synchronously triggering the satellite positioning module, airborne anemometer, and gas sensor according to the preset sampling frequency. The collected latitude, longitude, and altitude data, wind speed and direction data, and pollutant concentration readings are linked and bound to the current system timestamp to form raw multidimensional monitoring data.

[0027] Specifically, after generating the initial cooperative waypoint sequence, the system controls the UAV cluster to execute the flight mission corresponding to the waypoint sequence to collect raw multidimensional monitoring data. The system first inputs the three-dimensional coordinate data in the initial cooperative waypoint sequence into the UAV flight control system, and then converts the continuous waypoint coordinates into attitude control commands for driving the power unit through the navigation path planning module. Let the waypoint sequence of the j-th UAV be represented as: ; in This represents the number of waypoints for the drone. Given the three-dimensional coordinates of the k-th waypoint, the flight control system generates attitude commands based on the UAV's dynamic model, using the following control variables: ; in Let t be the actual position of the UAV at time t. For velocity vector, The coordinates of the current target waypoint The output attitude control commands, including throttle, roll angle, pitch angle and yaw angle control values, are used to drive the power unit via ESC and servo motors to make the UAV fly along the planned trajectory.

[0028] During flight, the drone performs sensor synchronization triggering according to the sampling frequency set by the system. The sampling period is defined as: ; in The sampling frequency is expressed in Hertz. At the beginning of each sampling period, the system simultaneously triggers the satellite positioning module, airborne anemometer, and gas sensor to acquire data. The satellite positioning module outputs latitude, longitude, and altitude data. Let its output at time t be: ; in Latitude Longitude At altitude, the airborne anemometer outputs wind speed and direction data, denoted as: ; in Wind speed, measured in meters per second. The wind direction angle, in degrees, is the pollutant concentration reading output by the gas sensor, denoted as: ; in The unit is micrograms per cubic meter or the concentration unit corresponding to the type of pollutant. The system reads the current system timestamp during data acquisition. The timestamp is associated with all the above-mentioned monitoring quantities, and a single original multidimensional monitoring data is generated. The multidimensional data structure is represented as follows: ; in This is the complete monitoring data vector collected by the system at time t. The system continuously collects data at sampling periods throughout the entire flight of the UAV. Repeat the above data collection process and store all generated data vectors in the UAV's local cache in chronological order, or transmit them back to the mission center in real time via the communication link.

[0029] This enables the simultaneous acquisition of multi-source information such as the UAV's spatial location, meteorological parameters, and pollutant concentration during the initial collaborative waypoint flight mission, generating original multi-dimensional monitoring data with a unified structure and strict temporal correlation, providing a complete and accurate input basis for subsequent pollution field model updates and path optimization.

[0030] S3. Based on the preset sensor response time constant and UAV flight speed, perform spatiotemporal lag compensation on the original multidimensional monitoring data to generate spatiotemporal calibration monitoring data. Furthermore, in S3, the process of performing spatiotemporal lag compensation on the original multidimensional monitoring data to generate spatiotemporal calibration monitoring data specifically includes the following steps: Extract the recording time, three-dimensional position coordinates, and real-time flight speed vector from the original multidimensional monitoring data; The spatial lag displacement vector is obtained by multiplying the real-time flight velocity vector with the preset sensor response time constant. Calculate the vector difference between the three-dimensional position coordinates and the spatial hysteresis displacement vector to obtain the calibrated actual sampling position coordinates; Establish a mapping relationship between pollutant concentration and actual sampling location coordinates to generate spatiotemporal calibration monitoring data.

[0031] Specifically, after obtaining the raw multidimensional monitoring data, the system performs spatiotemporal lag compensation on the raw data based on a preset sensor response time constant and the real-time flight speed of the UAV to generate spatiotemporal calibration monitoring data. The system first extracts the recording time, three-dimensional position coordinates, and real-time flight speed vector from the raw multidimensional monitoring data. Let the system at time... The collected raw monitoring data vector is represented as follows: ; in , , These are longitude, latitude, and altitude, respectively. , which represents its corresponding three-dimensional position coordinates. For pollutant concentration readings, To record timestamps, the real-time flight velocity vector of the UAV used in this step is represented as: ; in , , These are the instantaneous velocity components along the three coordinate axes, in meters per second.

[0032] Since gas sensors typically have a limited response time, the actual sampled gas corresponding to their output concentration value comes from the sensor's inhalation location over a past period. The system presets a sensor response time constant. The unit is seconds, and the original sampled position is compensated based on the UAV's displacement within this time constant. The system first calculates the product of the real-time flight velocity vector and the time constant to obtain the spatial lag displacement vector: ; in This indicates the sampling position offset caused by sensor response delay, expressed in meters.

[0033] The system then calibrates the original three-dimensional position coordinates based on the spatial lag displacement vector, and calculates the compensated actual sampled position coordinates using the position vector difference: ; in The actual sampling location after calibration represents the pollutant concentration output by the sensor. The calibration model described above can reflect the coupling relationship between the UAV's motion and the sensor response at the corresponding real sampling location, so that the measured value matches the real sampling point in space.

[0034] Based on the calibrated sampling location coordinates, the system establishes a mapping relationship between pollutant concentration and actual sampling location at each time point, forming a single spatiotemporal calibration monitoring data point: ; in The data vector after spatiotemporal lag compensation includes calibration location, meteorological parameters, pollutant concentration and time information. The system performs the above compensation process on each piece of raw data collected by the UAV throughout the process to form a complete spatiotemporal calibration monitoring data sequence, which is then used to update the Gaussian process model of the pollution field.

[0035] This effectively eliminates sensor response delay and sampling position offset caused by the high-speed movement of drones, ensuring that pollutant concentration readings are consistent with the corresponding spatial sampling positions, improving the spatial accuracy of multidimensional monitoring data, and thus enhancing the accuracy of pollution field modeling and prediction processes.

[0036] S4. Calculate the spatial change rate of the spatiotemporal calibration monitoring data to obtain the concentration gradient vector. When the magnitude of the concentration gradient vector is greater than the preset threshold, generate a path replanning trigger command. Further, in step S4, calculating the spatial rate of change of the spatiotemporal calibration monitoring data to obtain the concentration gradient vector specifically includes the following steps: Extract the position coordinates and concentration values ​​of the current sampling point and the previous sampling point that are continuous in time from the spatiotemporal calibration monitoring data; Calculate the displacement vector of the current sampling point relative to the previous sampling point, and calculate the concentration difference between the two points; Calculate the ratio of the concentration difference to the displacement vector magnitude, use this ratio as the magnitude, and construct the concentration gradient vector with the direction of the displacement vector as the reference direction.

[0037] Specifically, after generating the spatiotemporal calibration monitoring data, the system further calculates the pollutant concentration gradient vector based on the spatial rate of change between sampling points, and determines whether to trigger path replanning based on the comparison between its magnitude and a preset threshold. The system first extracts the information of the current sampling point and the previous sampling point that are continuous in time from the spatiotemporal calibration monitoring data sequence. Let the system at time... The obtained spatiotemporal calibration monitoring data vector is represented as follows: ; in , , These are the calibrated 3D sampling position coordinates, in meters; The pollutant concentration measurement value at this sampling point is expressed in micrograms per cubic meter or the corresponding physical concentration unit of the pollutant. The timestamp recorded by the system, the previous sampling time. The calibration data vector is represented as: ; The meanings and units of each symbol are consistent with the definitions above.

[0038] The system reflects the actual spatial motion of the UAV within two sampling periods by calculating the displacement vector of the current sampling point relative to the previous sampling point, and obtains: ; in To calibrate the sampling position coordinates, the magnitude of the displacement vector is: ; The module length represents the spatial distance between two consecutive sampling points of the drone, in meters.

[0039] The system further calculates the difference in pollutant concentration between the two sampling times: ; in This indicates the change in pollutant concentration caused by the movement of the drone, with units equal to... same.

[0040] To construct the concentration gradient vector, the system uses the rate of change of concentration as the gradient modulus and the direction of the displacement vector as the gradient direction. First, the concentration gradient modulus is obtained by the ratio of the concentration difference to the displacement modulus. ; This quantity represents the rate of change of concentration per unit length along the direction of the drone's movement, in units of concentration per meter. Simultaneously, the displacement vector is normalized to obtain its direction vector. ; The unit vector represents the direction of the UAV's motion in space. Based on the above two components, the concentration gradient vector is constructed as follows: ; in The unit is concentration per meter, used to characterize the rate of change of pollutant concentration in space along the direction of the drone's movement.

[0041] The system further calculates the magnitude of the concentration gradient vector: ; The modulus reflects the intensity of concentration changes, and the system has a preset concentration gradient trigger threshold. Its unit and Consistency is used to determine whether there are significant spatial abrupt changes in pollutants, when the following conditions are met: ; The system automatically generates path replanning trigger commands, enabling the drone to travel to areas with large concentration gradients for intensive sampling, thereby improving the spatial resolution capability of pollution boundaries and diffusion areas.

[0042] This enables accurate quantification of the spatial changes of pollutants along the direction of drone movement and immediate path adjustment when concentrations change rapidly, effectively improving the ability to resolve local details and the overall modeling accuracy of pollution field detection.

[0043] S5. The response path replanning trigger command control interrupts the original route of the single UAV that triggers the command. Based on the direction of the concentration gradient vector and the wind direction data in the meteorological parameters, the tracking waypoint is generated, and the estimated pollution source coordinates are determined when the monitoring data shows local maxima. Furthermore, in S5, the generation of tracking waypoints based on the direction of the concentration gradient vector and wind direction data in meteorological parameters specifically includes the following steps: The concentration gradient vector is normalized to obtain the gradient direction vector, and the headwind vector is calculated based on meteorological parameters. The gradient direction vector and the headwind vector are weighted and summed using preset weight coefficients to obtain the fused search direction vector; Starting from the current position of the UAV, extend the fusion search direction vector by a preset search step length, and the calculated target position coordinates are the tracking waypoints.

[0044] Specifically, after the system generates a path replanning trigger command based on the concentration gradient magnitude exceeding a preset threshold, the UAV that triggers the command immediately interrupts its original route and generates the next tracking waypoint based on the currently calculated concentration gradient vector direction and the wind direction data in the meteorological parameters, so that the UAV can move in the direction that may lead to the pollution source. When the pollutant concentration collected by the UAV during the tracking flight has a local maximum value in space, that is, the current concentration value is higher than the concentration value of the adjacent waypoint or the concentration value before or after, the system will determine the coordinates of this position as the estimated spatial location of the pollution source.

[0045] When generating tracking waypoints, the system first uses the concentration gradient vector obtained in the previous step. Let the system at time... The calculated concentration gradient vector is: ; in , , This represents the components of the concentration gradient along the three coordinate axes, with units of concentration per meter. It is used to represent the rate of change of concentration along each direction, and is a vector. Modulus length: : This represents the intensity of concentration variation in space, expressed in units of concentration per meter. The system normalizes the gradient vector to extract pure directional information, resulting in the gradient direction vector: ; in It is a dimensionless vector used to indicate the direction of the fastest concentration increase.

[0046] The system extracts the wind direction angle from the meteorological parameters recorded at the current sampling point to obtain the headwind direction. Let the wind direction angle be... The unit is degrees or radians, used to represent the angle at which the wind blows towards the drone. The system uses this to construct a counter-wind unit vector in the horizontal plane: ; This vector is dimensionless, where the negative sign indicates that the direction is reversed to obtain the headwind direction, and the third component being zero indicates that the headwind search direction is processed only in the horizontal plane, which is consistent with the physical characteristics of pollutants mainly spreading horizontally under the influence of the wind field.

[0047] To comprehensively consider the influence of concentration change trends and wind field on pollutant transport paths, the system weights and fuses the gradient direction vector and the headwind vector, setting the gradient direction weight coefficient as follows: The weighting coefficient for the headwind direction is Both are dimensionless quantities and satisfy: ; Based on this, the system constructs a fusion search direction vector: ; in As a dimensionless vector, it comprehensively represents the next search direction of the UAV. The system normalizes it and uses it to generate tracking waypoints.

[0048] Assume the drone is at time The current position coordinates are: ; Three of these are calibrated spatial coordinates, all in meters, and the system's set search step size. The unit is meters, used to control the UAV's forward distance in this search step. The system generates tracking waypoints by extending the step size along the fusion search direction from the current position, and their coordinates are defined as follows: ; in To track the three-dimensional coordinates of waypoints, in meters, the system uses them as target waypoints for the next stage of flight and issues corresponding control commands to make the UAV move along that path.

[0049] As the drone continues to advance along the tracking waypoint, the system records the concentration value in real time. When the concentration value reaches a local maximum in space, that is, when it is higher than the sampling value at the positions before and after it, the point is considered to be close to the pollution source. Based on this, the system determines the estimated coordinates of the pollution source and realizes the location of the pollution source.

[0050] This allows the drone to quickly enter an active search mode when a sudden change in concentration occurs, and generates tracking waypoints based on the fusion calculation of concentration gradient and wind direction, thereby significantly improving the drone's approach efficiency and positioning accuracy to pollution sources.

[0051] S6. By aggregating spatiotemporal calibration monitoring data and estimating pollution source coordinates through a multi-machine communication network, the shared global pollution probability model is trained and updated online to obtain the updated shared global pollution probability model. Furthermore, in S6, the aggregation of spatiotemporal calibration monitoring data and estimation of pollution source coordinates through a multi-machine communication network specifically includes the following steps: The spatiotemporal calibration monitoring data and estimated pollution source coordinates are encapsulated into a status update data package containing source node identifiers and generation timestamps according to a preset data format. The status update data packets are sent to other nodes in the drone swarm using the broadcast protocol of the multi-drone communication network; The receiving node performs integrity verification on the status update data packet. After the verification is successful, it parses and extracts the data content and stores it in the local synchronized dataset.

[0052] Specifically, after obtaining the spatiotemporal calibration monitoring data of each individual UAV in the UAV cluster and estimating the coordinates of pollution sources, the system achieves cross-node data aggregation through a multi-UAV communication network. The system then uses this aggregated data to train and update the parameters of a shared global pollution probability model online, enabling the model to gradually improve its prediction accuracy of the spatial distribution of pollution fields and the diffusion trend of pollution sources as the monitoring data increases.

[0053] The system first encapsulates the latest monitoring data generated by a single drone, assuming the drone is at time... The generated spatiotemporal calibration monitoring data vector is: ; in , , These are calibrated three-dimensional spatial coordinates, all in meters. These are measured values ​​of pollutant concentration, expressed in physical quantities. The timestamp is recorded by the system in seconds. If the drone records a local maximum concentration and estimates the coordinates of the pollution source during the pollution source tracking step, then the pollution source coordinates are: ; Each item represents the estimated three-dimensional location coordinates of the pollution source, in meters.

[0054] The system constructs a status update data packet based on the above monitoring data and pollution source coordinates according to a preset format. The data packet contains the source node identifier and the generation timestamp to ensure accurate identification of the data source and generation sequence in a multi-machine communication network. Let the status update data packet be represented as: ; in The identifier of the drone node that sent the data packet. and The meanings are monitoring data vector and pollution source estimated coordinates, respectively. If no pollution source is detected in the current period, the pollution source coordinates are marked as null.

[0055] The system utilizes the wireless multi-hop communication network broadcast protocol adopted by the UAV swarm to send the status update data packet to the entire swarm, enabling all other UAVs to obtain the monitoring data of this node in real time. After receiving the data packet, the receiving node first performs integrity verification on the data packet, using a cyclic redundancy check (CRC) code to verify the byte sequence of the data packet. After the verification passes, the data packet content is parsed and the monitoring data vector and pollution source coordinates are extracted and stored in the local synchronization dataset. Let the synchronization dataset of the receiving node be represented as: ; in This represents the number of active nodes in the current cluster. and They are nodes The calibration monitoring data and estimated pollution source coordinates.

[0056] After data aggregation is completed, the system trains the shared global pollution probability model online based on the synchronized dataset. The system discretizes the pollution field space using a regular grid, with the grid index set to . And set grid cells At any moment The probability of pollution is This represents the probability that a pollutant exists in the area or that it is near a pollution source, with a value ranging from zero to one. The system uses a Bayesian update strategy to correct the pollution probability online. Let the observation information extracted from the dataset be in the grid cell. The likelihood value is The likelihood value is determined by the pollutant concentration measurement and the estimated pollution source coordinates. Therefore, the online update formula for the pollution probability is: ; in Indicates the system at time 10:00 The resulting updated pollution probability value is achieved by a method that enables grid cells to rapidly increase their pollution probability when supported by increased observation data, while simultaneously reducing the probability of areas without pollution evidence, thus forming a continuously refined pollution probability distribution model.

[0057] To enhance the spatial continuity of the model, the system further performs spatial smoothing on the updated probability field, assuming it is related to the grid cells. The adjacent grid set is The average contamination probability of adjacent grids is: ; in Given the number of neighboring grid cells, the system fuses the local average probability and the online update probability using a balance coefficient to obtain the final updated global contamination probability: ; in , which is a dimensionless equilibrium coefficient, is used to stabilize the spatial variation of the probability field and improve the robustness of the model in large-area pollution fields.

[0058] This enables the pollution probability distribution model to be updated in real time based on multi-source spatiotemporal monitoring data from UAV swarms. The shared global pollution probability model gradually improves its overall ability to characterize pollution diffusion trends and source area features during continuous monitoring, thereby providing a more accurate global environmental understanding for subsequent collaborative search strategies and trajectory planning of UAV swarms.

[0059] S7. Analyze the updated shared global pollution probability model to extract the coordinates of regions where the prediction variance exceeds the preset uncertainty threshold, and generate dynamic collaborative waypoints for members in the UAV cluster that have not triggered replanning.

[0060] Furthermore, in S7, the process of parsing the updated shared global contamination probability model to extract the coordinates of regions where the prediction variance exceeds a preset uncertainty threshold specifically includes the following steps: Input the coordinates of discretized grid points within the preset monitoring area into the updated shared global pollution probability model, and calculate the posterior variance value corresponding to each grid point; Each posterior variance value is compared with a preset uncertainty threshold, and high-entropy grids with variance values ​​greater than the threshold are selected. Extract the geometric center coordinates of the high-entropy grid and use them as the coordinates of the areas that need to be focused on for detection.

[0061] Specifically, after completing the online update and spatial smoothing of the pollution probability field, the system inputs the updated shared global pollution probability model into the calculation process to analyze the estimation uncertainty of each spatial location within the monitoring area, and generates dynamic collaborative waypoints for UAVs that have not triggered track replanning based on the analysis results.

[0062] The system first discretizes the monitoring area according to rules to obtain the spatial coordinates of the geometric center of all grid cells. The center coordinates of grid cell j are represented as: ; in , , These represent the spatial coordinates of the grid center in the east, north, and vertical directions, respectively, all in meters. The system will use coordinate vectors... Inputting the data into the updated shared global contamination probability model yields the result at time [time value missing]. Final pollution probability value: ; This probability value represents the likelihood that a pollutant exists or is near a pollution source within the area of ​​grid cell j, and the value ranges from zero to one.

[0063] Based on the property that the pollution probability follows a Bernoulli statistical distribution, the system calculates the value of grid cell j at time [time value missing]. The corresponding posterior prediction variance is calculated as follows: ; in It is a dimensionless measure of the uncertainty in pollution probability estimation. A larger value indicates that the model lacks stable predictive support at that location.

[0064] The system sets an uncertainty threshold: ; This threshold is set based on the measurement accuracy of the UAV-borne sensors, the stability of model training in previous missions, and the complexity of the monitoring scenario. The system performs uncertainty determination on all grid cells, and the determination criteria are as follows: ; When the above inequality holds, the system will identify grid cell j as a high-entropy grid, indicating that the prediction uncertainty of this region in the current model exceeds the acceptable range, and the UAV needs to perform supplementary exploration.

[0065] The system extracts the coordinates of the center of all grids that satisfy the inequality. The coordinates of these coordinates are used to form a set of key detection areas. Then, the system generates dynamic collaborative waypoints for UAV members in the cluster that have not yet triggered their own track replanning, enabling UAVs to actively fly to areas with high uncertainty to perform supplementary observations, thereby providing more sufficient monitoring data support for the next cycle of pollution probability updates.

[0066] This allows for the analysis of high-uncertainty regions from the updated shared global pollution probability model and drives the drone swarm to perform automatic replacement detection, enabling the pollution probability field to gradually converge during continuous tasks and improving the stability and reliability of pollution source location and pollution diffusion distribution estimation.

[0067] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An integrated air pollution detection method mounted on a drone, characterized in that, Includes the following steps: S1. Establish a multi-machine communication network and construct a shared global contamination probability model based on Gaussian process regression. Allocate non-overlapping initial cooperative waypoint sequences to the UAV cluster according to the information entropy distribution. S2. Control the UAV cluster to execute the flight mission corresponding to the initial cooperative waypoint sequence, and collect raw multidimensional monitoring data including three-dimensional position, meteorological parameters and pollutant concentration; S3. Based on the preset sensor response time constant and the UAV flight speed, perform spatiotemporal lag compensation on the original multidimensional monitoring data to generate spatiotemporal calibration monitoring data. S4. Calculate the spatial change rate of the spatiotemporal calibration monitoring data to obtain the concentration gradient vector. When the magnitude of the concentration gradient vector is greater than a preset threshold, generate a path replanning trigger command. S5. In response to the path replanning trigger command, control the single UAV that triggered the command to interrupt the original route, generate a tracking waypoint based on the direction of the concentration gradient vector and the wind direction data in the meteorological parameters, and determine the estimated pollution source coordinates when the monitoring data shows a local maximum. S6. By aggregating the spatiotemporal calibration monitoring data and the estimated pollution source coordinates through the multi-machine communication network, the shared global pollution probability model is trained and updated online to obtain the updated shared global pollution probability model. S7. Analyze the updated shared global pollution probability model to extract the coordinates of regions where the prediction variance exceeds a preset uncertainty threshold, and generate dynamic collaborative waypoints for members of the UAV cluster that have not triggered replanning.

2. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S1, the steps of establishing a multi-machine communication network and constructing a shared global pollution probability model based on Gaussian process regression specifically include the following steps: Configure the self-organizing network communication protocol parameters and time synchronization benchmark for each node of the drone swarm; The squared exponential kernel function was selected as the covariance function to describe the spatial correlation of pollutant concentrations, and a constant mean function was set. Initialize a hyperparameter set containing signal variance and length scale, establish a Gaussian process regression prior distribution that maps three-dimensional spatial coordinates to pollutant concentration, and use it as a shared global pollution probability model.

3. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S2, controlling the UAV cluster to execute the flight mission corresponding to the initial cooperative waypoint sequence specifically includes the following steps: The coordinate data in the initial cooperative waypoint sequence is input into the UAV flight control system to generate attitude control commands for the drive power unit; Data acquisition is performed by synchronously triggering the satellite positioning module, airborne anemometer, and gas sensor according to the preset sampling frequency. The collected latitude, longitude, and altitude data, wind speed and direction data, and pollutant concentration readings are linked and bound to the current system timestamp to form raw multidimensional monitoring data.

4. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S3, the process of performing spatiotemporal lag compensation on the original multidimensional monitoring data to generate spatiotemporal calibration monitoring data specifically includes the following steps: Extract the recording time, three-dimensional position coordinates, and real-time flight speed vector from the original multidimensional monitoring data; The spatial lag displacement vector is obtained by multiplying the real-time flight velocity vector with the preset sensor response time constant. Calculate the vector difference between the three-dimensional position coordinates and the spatial hysteresis displacement vector to obtain the calibrated actual sampling position coordinates; Establish a mapping relationship between pollutant concentration and the coordinates of the actual sampling location to generate spatiotemporal calibration monitoring data.

5. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S4, calculating the spatial change rate of the spatiotemporal calibration monitoring data to obtain the concentration gradient vector specifically includes the following steps: Extract the position coordinates and concentration values ​​of the current sampling point and the previous sampling point that are continuous in time from the spatiotemporal calibration monitoring data; Calculate the displacement vector of the current sampling point relative to the previous sampling point, and calculate the concentration difference between the two points; Calculate the ratio of the concentration difference to the displacement vector magnitude, use this ratio as the modulus, and construct a concentration gradient vector with the direction of the displacement vector as the reference direction.

6. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S5, generating tracking waypoints based on the direction of the concentration gradient vector and the wind direction data in the meteorological parameters specifically includes the following steps: The concentration gradient vector is normalized to obtain the gradient direction vector, and the headwind vector is calculated based on meteorological parameters. The gradient direction vector and the headwind vector are weighted and summed using preset weighting coefficients to obtain the fused search direction vector; Starting from the current position of the UAV, extend the fusion search direction vector by a preset search step length, and the calculated target position coordinates are the tracking waypoints.

7. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S6, the step of aggregating the spatiotemporal calibration monitoring data and the estimated pollution source coordinates through the multi-machine communication network specifically includes the following steps: The spatiotemporal calibration monitoring data and estimated pollution source coordinates are encapsulated into a status update data package containing source node identifiers and generation timestamps according to a preset data format. The status update data packet is sent to other nodes in the UAV cluster using the broadcast protocol of the multi-machine communication network; The receiving node performs integrity verification on the status update data packet. After the verification is successful, it parses and extracts the data content and stores it in the local synchronization dataset.

8. The integrated air pollution detection method mounted on a drone according to claim 1, characterized in that, In step S7, the step of parsing the updated shared global contamination probability model to extract the coordinates of regions where the prediction variance exceeds a preset uncertainty threshold specifically includes the following steps: Input the coordinates of discretized grid points within the preset monitoring area into the updated shared global pollution probability model, and calculate the posterior variance value corresponding to each grid point; Each posterior variance value is compared with a preset uncertainty threshold, and high-entropy grids with variance values ​​greater than the threshold are selected. Extract the geometric center coordinates of the high-entropy grid and use them as the coordinates of the areas that need to be focused on detection.

Images