A dynamic release sub-machine and data fusion reconnaissance method and system

Abstract

The application relates to the field of rescue reconnaissance, and discloses a reconnaissance method and system of a dynamic release sub-machine and data fusion, which comprises the following steps: when a mother ship unmanned aerial vehicle carries a sub-machine and flies to a disaster area, macroscopic reconnaissance load is used to generate macroscopic situation information by scanning; three-dimensional modeling load is used to generate a three-dimensional model of a scene by scanning, and three-dimensional environment information is output; a target area to be approached and reconnoitered is determined according to the macroscopic situation information; the mother ship unmanned aerial vehicle calculates a sub-machine release time according to the three-dimensional environment information and the characteristics of the target area, generates a sub-machine reconnaissance track, carries the sub-machine and flies to a release position, and releases the sub-machine; the released sub-machine flies to the target area according to the sub-machine reconnaissance track, performs fine reconnaissance, and returns fine information obtained through the reconnaissance to the mother ship unmanned aerial vehicle in real time for fusion. The application has the advantages that the overall endurance time can be prolonged, and seamless connection from macroscopic situation perception to microscopic detail identification can be realized.

Description

Technical Field

[0001] This invention relates to the field of rescue reconnaissance, specifically to a reconnaissance method and system that combines dynamic release of submachines with data fusion. Background Technology

[0002] For reconnaissance of forest fires, manual reconnaissance has drawbacks such as high risk, low efficiency, and difficulty in reaching the core disaster area. With the rapid development of technology, drones have been widely used in forest fire reconnaissance.

[0003] When using drones for reconnaissance, a single large drone faces a fundamental contradiction between endurance and flexibility, and cannot achieve wide-area coverage and detailed identification; while using multiple independent drones in coordination presents problems such as complex scheduling and difficulties in airspace management. Therefore, existing disaster reconnaissance methods typically have the following drawbacks: short flight time, making it difficult to complete large-area disaster scanning; fixed payloads cannot simultaneously handle wide-area search and fine identification; inability to adjust reconnaissance strategies in real time according to changes in the disaster situation; and communication interruptions at single nodes due to complex terrain. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a reconnaissance method and system that dynamically releases sub-units and integrates data, thereby solving the problems of short flight time, difficulty in completing large-scale disaster scanning, inability of fixed payloads to balance wide-area search and fine identification, inability to adjust reconnaissance strategies in real time according to changes in disaster conditions, and communication interruption of single nodes due to complex terrain.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a reconnaissance method for dynamically releasing submachines and data fusion, comprising: Step S1: When the UAV carrier-carried aircraft arrives at the disaster area, the macro-reconnaissance payload of the UAV carrier performs a large-scale scan to generate macro-situation information; the 3D modeling payload of the UAV carrier scans to generate a 3D model of the site and outputs 3D environmental information. Step S2: Based on the macro situation information, determine the target area that needs to be closely reconnoitered. The UAV carrier calculates the timing of the deployment of the drones based on the three-dimensional environmental information and the characteristics of the target area, and generates the drone reconnaissance trajectory. The hierarchical decision-making and optimization strategy includes four levels: ROI hotspot area identification, drone deployment necessity assessment, drone reconnaissance trajectory planning, and deployment timing calculation. Step S3: The unmanned aerial vehicle carrier carries the sub-drones to the deployment location and deploys the sub-drones according to the deployment timing; Step S4: The deployed drones fly to the target area according to their reconnaissance flight paths to conduct detailed reconnaissance and transmit the detailed information they have collected back to the UAV mothership in real time. The UAV mothership integrates the detailed information from the drones to generate complete disaster area situation information.

[0006] Secondly, the present invention provides a reconnaissance system for dynamically releasing submachines and data fusion, the system comprising: The scanning and modeling module is used to generate macroscopic situational information by using the macroscopic reconnaissance payload of the UAV carrier to perform large-scale scanning when the UAV carrier's carrier-based aircraft arrive at the disaster area; and to generate a 3D model of the site by using the 3D modeling payload of the UAV carrier, and output 3D environmental information. The hierarchical decision-making and optimization module is used to determine the target area that needs to be closely reconnoitered based on macro-situation information. The UAV carrier calculates the timing of the deployment of the drones based on the three-dimensional environmental information and the characteristics of the target area, and generates the drone reconnaissance trajectory. The hierarchical decision-making and optimization strategy includes four levels: ROI hotspot area identification, drone deployment necessity assessment, drone reconnaissance trajectory planning, and deployment timing calculation. The sub-aircraft deployment module is used to transport sub-aircraft from the unmanned aerial vehicle (UAV) carrier to the deployment location and deploy the sub-aircraft according to the deployment timing. The data fusion module is used by the deployed drones to conduct detailed reconnaissance of the target area according to their reconnaissance flight paths, and to transmit the detailed information obtained back to the drone mothership in real time. The drone mothership then integrates the detailed information from the drones to generate complete disaster area situation information.

[0007] This invention utilizes a drone carrier to provide endurance (4 to 6 hours), allowing the carrier to hover over disaster areas for reconnaissance. Based on the reconnaissance results, it intelligently releases drones to conduct detailed surveys of hotspots, achieving a seamless connection from macro-level situational awareness to micro-level detail identification. Simultaneously, it employs multi-dimensional information fusion, combining the aerial view from the drone carrier with data collected by the drones at close range to create a complete disaster situation report. Furthermore, the drones act as communication relays, extending the coverage of the drone carrier and constructing a resilient communication network. Therefore, the technical solution of this invention possesses at least the following beneficial technical effects: 1. Improve overall battery life; 2. The coverage area can be expanded from single-point reconnaissance to large-area dynamic focusing, achieving full coverage; 3. The reconnaissance accuracy has been upgraded from fixed resolution to multi-level zoom, which can be adaptively improved from 1 meter level to centimeter level; 4. The response speed has been upgraded from manual scheduling to AI real-time decision-making, achieving a response time within minutes; 5. Cost control is reduced from multiple large machines to one mothership plus multiple submachines, which can reduce costs by more than 60%; 6. The ability to adapt to complex environments has been upgraded from being prone to disconnection to being able to form a relay network of slave units, enabling flexible communication; 7. The sub-aircraft is an electric multi-rotor reconnaissance sub-aircraft with vertical take-off and landing and hovering capabilities. After release, it can fly directly along the planned trajectory without the need for gliding trajectory optimization, which can simplify the deployment decision and improve the mission response speed. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the execution process of a reconnaissance method for dynamically releasing submachines and fusing data, as described in this invention. Figure 2 This is a schematic diagram of the structure of a reconnaissance system for dynamically releasing submachines and data fusion according to the present invention; Figure 3 This is a hardware system framework diagram used in this invention. Detailed Implementation

[0009] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] Prior to this, the hardware system framework used in this invention includes: Figure 3 As shown, the entire hardware system structure includes an unmanned aerial vehicle (UAV) carrier, a macroscopic reconnaissance payload mounted on the UAV carrier, a rapid 3D modeling payload mounted on the UAV carrier, an edge computing unit located on the UAV carrier, several sub-units mounted on the UAV carrier via a sub-unit mounting and release mechanism, and payloads mounted on each sub-unit. Each sub-unit is an electrically powered multi-rotor reconnaissance sub-unit with vertical takeoff and landing and hovering capabilities. Each sub-unit can carry different payloads such as a high-definition optical zoom camera, an infrared thermal imager, a gas sensor, a temperature and humidity sensor, and a macro camera to perform close-range, high-precision reconnaissance missions. The macroscopic reconnaissance payload is used to perform large-scale disaster scanning to obtain macroscopic situational information about the disaster area. The rapid 3D modeling payload is used to perform rapid oblique photography or laser scanning of the disaster area, generating a 3D model of the disaster site and transmitting it back to the command center in real time. The edge computing unit is used to implement edge computing functions. Example 1

[0011] like Figure 1 As shown, this embodiment provides a reconnaissance method for dynamically releasing submachines and data fusion, the reconnaissance method including the following steps: Step S1: When the UAV carrier-borne drones arrive at the disaster area, the macroscopic reconnaissance payload of the UAV carrier performs a large-scale scan to generate macroscopic situational information; the 3D modeling payload of the UAV carrier scans to generate a 3D model of the site and outputs 3D environmental information, specifically the output 3D environmental information. Including point cloud data Digital Elevation Model and signed distance field This is for use in subsequent hierarchical decision-making and optimization strategies; Step S2: Based on the macro situation information, determine the target area that needs to be closely reconnoitered. The UAV carrier calculates the timing of the deployment of the drones based on the three-dimensional environmental information and the characteristics of the target area, and generates the drone reconnaissance trajectory. The hierarchical decision-making and optimization strategy includes four levels: ROI hotspot area identification, drone deployment necessity assessment, drone reconnaissance trajectory planning, and deployment timing calculation. Step S3: The unmanned aerial vehicle carrier carries the sub-drones to the deployment location and deploys the sub-drones according to the deployment timing; Step S4: The deployed sub-drones fly to the target area according to their reconnaissance flight paths to conduct detailed reconnaissance. Specifically, the payloads carried by the sub-drones can be used to conduct reconnaissance and data collection on the target area, and the detailed information obtained can be transmitted back to the UAV mothership in real time. The UAV mothership integrates the detailed information from the sub-drones to generate complete disaster area situation information.

[0012] In step S2, the ROI hotspot region identification specifically involves: using RANSAC plane fitting and Euclidean distance clustering to identify ROI hotspot regions and outputting a set of ROIs containing bounding boxes; The necessity assessment of the dispatch of sub-machines is specifically as follows: the dispatch value is calculated based on information gain and comprehensive dispatch cost, and the optimal allocation is solved by the Hungarian algorithm to obtain the allocation scheme. The sub-aircraft reconnaissance trajectory planning is specifically as follows: according to the allocation scheme, the RRT*-Constrained algorithm is used to plan the trajectory, generating a sub-aircraft reconnaissance trajectory from the deployment point to the target hotspot area (i.e., the target area) in the ROI hotspot area set, and ensuring that the generated sub-aircraft reconnaissance trajectory meets the requirements of communication maintenance, energy constraints and obstacle avoidance. The obstacle avoidance requirements are assessed by a risk function that integrates three-dimensional environmental information and disaster dynamic evolution information, and the target inspection is judged by a distance threshold. The specific calculation of the deployment timing is as follows: the deployment point is calculated by the position of the UAV mothership, the attitude matrix of the UAV mothership and the mounting position of the sub-units, the optimal deployment timing is solved based on the dynamic characteristics of the electric multi-rotor (i.e. the sub-units do not need initial velocity), and communication maintenance and energy are used as constraints.

[0013] More specifically, the ROI hotspot region identification includes: A11. Input the point cloud data from the 3D environment information; A12. The RANSAC plane fitting algorithm is used for ground segmentation to separate the point cloud data into ground point sets and non-ground point sets. A13. Use the Euclidean distance clustering algorithm to divide spatially continuous non-ground point sets into different clusters; A14. Extract multi-dimensional features for each cluster, including at least the coefficient of variation of height, normal vector dispersion, temperature anomaly index or color anomaly degree, to quantify the disaster characteristics. A15. Calculate the ROI hotspot region score for each cluster, filter the ROI hotspot regions according to the set ROI threshold, and output the set of filtered ROI hotspot regions.

[0014] in: Specifically, A11 refers to: point cloud data from the input three-dimensional environment information. Where N is the total number of points in the point cloud; For the first The UTM coordinates of each point are obtained by LiDAR or oblique photogrammetry. For the first The UTM coordinates of each point are obtained by LiDAR or oblique photogrammetry. For the first The RGB color values ​​of each point are obtained by a visible light camera; A12 specifically includes: A121. Initialization parameters: including the number of iterations. Interior point distance threshold Optimal number of interior points and optimal plane parameters ,in, That is, the maximum number of iterations of the RANSAC algorithm, which is 1000; That is, the maximum distance to the ground point is determined, and the value is 0.1 m; That is, the number of interior points of the current optimal plane is initialized to 0; The plane parameter with the most interior points is initialized to empty.

[0015] A122. Iterative generation of candidate planes For each iteration : 1. Random sampling: from point cloud data Three non-collinear points are randomly selected from the data. ; 2. Calculate the candidate plane parameters: Construct vectors: ; Normal vector (cross product): ; Normalization: , This represents the normalized plane normal vector. Representing the normal vector Quantity, Representing the normal vector Quantity, Representing the normal vector Quantity; Planar parameters: ; 3. Count the number of points within the area: right For all points, calculate their distances to the plane: ;because Normalized, denominator Therefore ; Interior point determination and statistics: ,in, This is an indicator function; it returns 1 if the condition is met, and 0 otherwise. 4. Update the optimal plane: like ,but: ; A123. Determine ground points and non-ground points: Using optimal plane parameters Classify all points to form a ground point set. Non-ground point set , here This represents the set difference operation, i.e., from the entire set... Remove belonging to The elements, non-ground points are used for subsequent target recognition.

[0016] Specifically, A13 includes: A131. Initialization parameters: including clustering distance threshold. Minimum number of cluster points Clustering sets and unvisited point set ,in, That is, the maximum distance between points within the same cluster, which is 0.5m; That is, the minimum number of points required for an effective cluster, with a value of 50; That is, the set of all valid clusters is initialized to empty; Points that have not yet been classified are initialized to... ; A132. Construct a KD-Tree index structure to support fast indexing of any point. Nearest neighbor search; A133, Iterative Clustering: 1. Selecting a seed point: When When not empty, randomly select seed points. ; 2. Initialize the current cluster ,queue ; 3. Breadth-first expansion: When the queue If not empty, retrieve the head point of the queue. The query distance is less than the cluster distance threshold. All neighboring points , express A point in the array; for each neighboring point ,like If not yet visited, add it to the cluster. and queue and will from Remove from; 4. When the queue is empty, if If so, retain the cluster and assign it the cluster number. ;like If the cluster is not found, discard it; output the cluster set. ,in, The effective number of clusters, each cluster Include One point.

[0017] Specifically, A14 includes: A141, Computational Clustering Coefficient of variation of height for all points within: Among them, the average height Standard deviation of height , For point Height; A142, Computational Clustering Discreteness of normal vectors at each point within the interior: ,in, Let be the normal vector of each point. The cluster center normal vector, Represents the vector dot product. This indicates that the dot product is converted to degrees (radians). A143, Computational Clustering Internal temperature anomaly index: ,in, The maximum temperature for clustering, i.e., clustering The highest temperature at all points inside can be obtained using an infrared thermal imager; The ambient temperature, i.e. the background temperature of the disaster area, can be obtained through meteorological data; This is the temperature threshold, which is the minimum temperature difference required to determine an anomaly. It can be preset as needed. A144, Computational Clustering Color aberration compared to normal environment: ;in, For clustering The average color vector, The color vector for the normal environment;

[0018] Output each cluster eigenvectors .

[0019] Specifically, A15 includes: A151. For each cluster Calculate ROI hotspot region scores: ;in, The weight for the coefficient of variation is 0.3. This is the weight for the dispersion of the normal vector, with a value of 0.2. This is the weight of the temperature anomaly index, with a value of 0.4; This is the weight for color anomaly, with a value of 0.1; A152. Based on the set ROI threshold By filtering ROI hotspot areas, a set of ROI hotspot areas is obtained: And sorted in descending order of rating: , ,in, This represents the total number of ROI hotspot areas. Indicates the first ROI hotspot regions (corresponding to the original cluster) ), For the first Scoring of ROI hotspot areas; A153. Output the set of ROI hotspot regions: And each All contain attributes ,in, With the center coordinates, For bounding box, For the equivalent radius, The disaster level is determined by the information entropy. .

[0020] In step S2, more specifically, the assessment of the necessity of dispatching the slave unit includes: A21. Input the data for the ROI hotspot region set, the data for the sub-machines, and the data for the unmanned aerial vehicle carrier; A22. Calculate information gain: Specifically, this includes calculating the current information entropy based on discrete probability distribution, calculating the observation likelihood, and calculating the posterior entropy, expected posterior entropy, and information gain to quantify the information value of the submachine gun's reconnaissance of ROI hotspot areas. A23. Based on the data of the ROI hotspot area set, the data of the sub-machines and the data of the UAV mothership, calculate the Euclidean distance, flight energy consumption, path risk and overall dispatch cost; A24. Calculate the dispatch value based on information gain and overall dispatch cost, use the Hungarian algorithm to solve for the optimal allocation, and output the allocation scheme.

[0021] In A21, the input submachine data includes the first... Frame machine The probability of correct detection by the sensor False alarm probability of sensors UTM coordinates of the slave unit mounting location Sensor type Current remaining battery power wait, and Derived from load performance parameters, Database sourced from the drone carrier Derived from the design parameters of the unmanned aerial vehicle carrier, Originating from task configuration, The data originates from the battery management system; the input data for the UAV carrier includes the UTM coordinates of the UAV carrier's current position. Attitude matrix wait, Obtained from GNSS / fusion estimation. Sourced from IMU output.

[0022] A22 specifically includes: A221. Calculate the current information entropy: Let the current information entropy be calculated for the set of ROI hotspot regions. The perception is a discrete probability distribution , Then the current information entropy ; A222. Calculate the observed likelihood: The probability of a sensor making a correct detection during a disaster: False alarm probability of sensors in the absence of disaster: Probability of missed detection: ; Probability of correct rejection: , For observation results; A223. Calculate the posterior entropy, expected posterior entropy, and information gain: For each possible observation The posterior probability is obtained by updating using Bayesian methods. , For the first Frame machine; Calculate the posterior entropy: ; Calculate the expected posterior entropy: ; in, This represents a probability measure (strict mathematical probability), with a value range of [0,1]. Calculate information gain: ; Output information gain matrix , That is to say The matrix.

[0023] A23 specifically includes: A231. Calculate the Euclidean distance: ,in, For the submachine UTM coordinates of the mount location In the set of ROI hotspot areas The center's UTM coordinates, The UTM coordinates of the current position of the drone carrier. The attitude matrix of the unmanned aerial vehicle carrier. The mounting location for the submachine; A232. Calculate flight energy consumption: ,in, The hovering power of the submachine gun. The hovering time of the submachine gun. The cruise power of the submachine gun. The submachine gun's cruising speed; A233. Constructing risk fields and calculating path risks: Based on 3D environment information Constructing a spatial risk field: ,in, Represents static geometric risk; This represents dynamic risk, the initial value of which is obtained through simulation prediction; Calculate path risk: ,in, The planned path from the mount point to the ROI hotspot area is derived from simplified estimation or pre-planning. For position The signed distance field is obtained through the output of three-dimensional environment information. The 3D spatial location on the path; The predicted location of the disaster epicenter is derived from physical simulation. This is the square of the variance in the spread of the disaster, which is derived from physical simulation. Geometric risk and dynamic risk weights are derived from disaster type configurations. To avoid dividing by zero for small quantities, the setting can be adjusted according to actual needs; For along the path The integral parameter, Represents the path element; A234. Calculate the overall dispatch cost: ,in, This is the distance weight (normalized weight), with a value of 1 / 3; The energy consumption weight (normalized weight) is set to 1 / 3. The risk weight (normalized weight) is 1 / 3. The maximum flight speed of the sub-machine is derived from its performance parameters. This represents the maximum battery life of the slave unit, derived from its performance parameters. The battery capacity (full charge energy) of the handset is derived from the battery specifications; As a reference for maximum risk, it is a risk normalization factor that can be preset as needed; Output cost matrix .

[0024] The A24 specifically includes: A241. Calculate the dispatch value based on information gain and overall dispatch cost: ,in, This is the cost penalty coefficient; A242. Solving for the optimal allocation: Building a value matrix ; The optimal allocation is solved using the Hungarian algorithm. The constraints are: Submachine uniqueness: ; Uniqueness of ROI hotspot areas: ; Binary decision-making: The optimal allocation matrix is ​​obtained using the Hungarian algorithm. Determine which submachine to execute the corresponding hot task; Output allocation scheme .

[0025] In step S2, during the reconnaissance trajectory planning of the sub-aircraft, obstacle avoidance requirements are assessed using a risk function (i.e., the risk field in A233) that integrates three-dimensional environmental information and disaster dynamic evolution information. The risk function is a weighted sum of geometric risk and dynamic risk. Geometric risk is calculated by using a signed distance field to determine the distance from the location to the obstacle, while dynamic risk is predicted by disaster physical simulation. The weights are adaptively configured according to the disaster type. Target inspection uses a distance threshold for judgment.

[0026] The sub-aircraft reconnaissance trajectory planning specifically includes: A31. Planning initialization, the purpose of which is to initialize the allocated submachines. Generate target hotspot areas from the delivery point to the ROI hotspot area set. Reference flight paths; specifically including: A311. Determine the initial state Since trajectory planning occurs before the release of the drone, it belongs to the pre-planning stage. The state variables used in the planning are predicted values, derived from the database and real-time status of the drone mothership. Let the initial state variables be... ,in, For the submachine The UTM coordinates of the mount location can be obtained from Calculated; It is the initial heading angle, which usually points towards the target hotspot area. The center UTM coordinates; A312, Accurate Constraint Parameters: Communication constraint parameter: Maximum communication distance Communication quality threshold Communication quality ,in, That is, the minimum acceptable quality. That is, the slave machine in node state Communication quality; Energy constraint parameter: battery capacity Safety Reserve Climbing energy consumption Calculated by multiplying the height difference by the unit energy consumption; Obstacle avoidance constraint parameters: safe distance That is, the minimum distance to the obstacle; Risk constraint parameter: Geometric risk weight and dynamic risk weights (Same as A233); Altitude constraint parameters: safe flight altitude Terrain height Flight altitude constraints (That is, the planned waypoints need to meet this constraint). That is, location Elevation.

[0027] A32, RRT*-Constrained algorithm solution, specifically including: A321. Algorithm initialization, including initializing the search tree. Parent node mapping and cumulative cost ,in, That is, the set of explored state points, with an initial value of ; That is, the parent node of each node, with an initial value of ; That is, the cost from the starting point to this node, with an initial value of ; A322, Iterative expansion (looping until convergence). The following process is executed in each iteration: Sampling: based on probability From target hotspot areas Sampling, in probability Sampling points are obtained by randomly sampling from the global space. ; Nearest neighbor: ; Extended: ,in, To extend the step size; Course Update: represent and exist Distance at coordinates represent and exist Distance between coordinates; Constraint check: If (i.e., location) (The signed distance field is less than the safe distance) or And there is no relay, or the energy consumption exceeds the limit (i.e., the remaining battery energy). If ), then discard it; Determine the parent node: Define the neighborhood radius ,in, Adjust the parameters (the specific settings can be determined based on experience); For the state dimension; express The power root is used to achieve dimension normalization. Find the neighborhood: ; For each Calculate the cost of the line segment: ; in, Indicates from node arrive The cost of the line segment; Current node state (position + heading) (), which originates from the search tree Nodes in; This is the state of the new sampling node, which is generated through RRT* sampling; The state space distance between the two nodes is the Euclidean distance (including heading weighting). For the path parameters on the line segment (from 0 to the line segment length); For position The expected speed at the location is set according to the complexity of the environment; For position Communication quality at the location; To avoid small constants that divide by zero, they can be set according to actual needs; For position Risk value at the location (same as the calculation of risk for route A233); and The weighting coefficients for speed, communication, and risk can be set based on experience. Select the parent node: ; renew: ; Rewiring: for each ,like: If the path satisfies the constraints, then update. Recursive update The cumulative cost of all descendant nodes; Target check: If , backtracking to find the path Record the current optimal value, where, The target tolerance threshold; Termination condition: Reaching the maximum number of iterations Or the cost converges; Output submachine reconnaissance track ,in, The sequence represents the flight path.

[0028] In step S2, the calculation of the delivery timing specifically includes: A41. Establish an optimization problem: Decision variables include the timing of deployment. and the location of the drone carrier at the time of deployment ; A42. Calculation of Delivery Points: ,in, for The attitude matrix of the unmanned aerial vehicle carrier at all times. For the submachine Mounting position (i.e., the mounting position of the submachine gun in the mothership's coordinate system). For the deployment height; A43. Energy Consumption Estimation: Total Task Energy Consumption Among them: climbing energy consumption , For the quality of the submachine, It is the acceleration due to gravity. For the height difference, For motor efficiency, Working height , The center terrain elevation of the ROI hotspot area is derived from the output 3D environmental information; cruise energy consumption. Horizontal distance This refers to the horizontal cruising distance between the target's horizontal position and the drop point's horizontal position. Indicates cruise power. Indicates cruising speed; hovering energy consumption. , The task hovering time. Indicates hovering power; A44. Optimization Solution Objective function: ,in, The cost of maneuvering the drone carrier to its deployment location. These are weight values; Communication constraints: ,in, For the slave machine in state Communication quality; Discretization time window Iterate through the evaluations and select: ,in, Indicates the current moment. Indicates the prediction time domain, Indicates the latest deployment time considered for optimization; A45, Output delivery command ,in, For a moment, The target location of the drone carrier. This is the initial state of the slave machine. The heading is the point at which the course begins.

[0029] In step S4, the process of the UAV mothership fusing the detailed information of the sub-drones to generate complete disaster area situation information specifically involves: using a multimodal spatiotemporal alignment and data fusion algorithm to spatiotemporally align and overlay the detailed information of the sub-drones with the macroscopic situation information of the UAV mothership and the on-site 3D model to generate complete disaster area situation information; simultaneously, combining Kriging spatial interpolation to update the risk field, and using the updated risk field for the generation of undeparted sub-drones in this round, new ROI hotspot areas, or the next round of missions.

[0030] Furthermore, the detailed information of the unmanned aerial vehicle (UAV) mothership, which integrates the data from its sub-drones to generate complete disaster area situational information, specifically includes: B1. Time synchronization, specifically, time synchronization is performed after the sub-unit completes reconnaissance; Time synchronization uses GNSS timing PPS signals, i.e., the mothership timestamp is... The slave machine timestamp is Time deviation ,like If so, calibration is triggered; otherwise, it is uniformly marked as... ; B2. The complete transformation chain for mapping the pixel coordinates of the sub-machine image to the world UTM coordinate system through spatial registration is as follows: ; in, These are the homogeneous coordinates of the image pixels, derived from the images captured by the slave camera. Invert the camera intrinsic matrix (3×3) to transform pixel coordinates to the normalized image plane (factory calibration). The camera extrinsic calibration matrix (4×4) is used to transform the camera coordinate system to the sub-body coordinate system (factory calibration). The submachine real-time pose matrix (4×4) consists of GNSS real-time position and IMU real-time attitude, and is used to transform the submachine body coordinate system to the UTM world coordinate system; The steps are as follows: (Camera coordinate system); (Subbody coordinate system); (UTM World Coordinate System).

[0031] After spatial registration is completed, the detailed information of the sub-unit is spatiotemporally aligned and superimposed with the macro-situational information of the UAV mothership and the on-site 3D model to generate complete disaster area situational information.

[0032] B3, Kriging spatial interpolation B31, Variation function fitting The experimental variogram, which is the variogram value calculated from the discrete sample data collected from the sub-computer: ; in, The experimental variability function value, i.e., the distance is The empirical semivariance of all sample point pairs; Lag distance is the spatial distance between two sample points, typically discretized into intervals. ; Indicates distance as The set of sample point pairs, that is, all point pairs that satisfy the condition. The number of sample point pairs at this distance, i.e., the set. The number of elements (dot-logarithms); To traverse all distances The points are correct, and the points are correct. Belongs to set ; For position The measured value at that location, i.e., the submachine at Sensor readings (such as temperature) at the location; For position The measured value at that location, i.e., the submachine at Sensor readings at the location; The squared difference between observations at two points reflects the degree of spatial variability; the larger the difference, the stronger the spatial heterogeneity. The semivariance coefficient is used to ensure the theoretical relationship between the variogram and the covariance function. It is used in the variogram calculation of the above experiment. (Sample observations), used as known data in subsequent Kriging spatial interpolation, are then processed using optimal weights. Weighted estimation of unknown locations The value of .

[0033] The theoretical model (spherical), i.e., a continuous variogram model described by an analytical function: ; in, The lag distance is the spatial distance between two sample points. The value of a nugget, i.e. The value of the variability function over time, which is micro-variance plus measurement error, reflects small-scale randomness; The arch height, i.e., the difference between the sill value and the nugget value, is the structural variation caused by spatial correlation. The sill value is the value at which the variogram reaches a stationary state, representing the total degree of variation and reflecting the overall dispersion of the data. The range is the distance at which the variogram reaches the sill value, representing the spatial correlation range beyond which the variogram ceases to be correlated.

[0034] Parameter estimation: obtained by weighted least squares fitting .

[0035] The experimental variability function is approximated / fitted using a theoretical model, and the specific process is as follows: (1) Calculate experimental values: Obtain discrete points from sample data ; (2) Parameter estimation: Adjusting the parameters of the theoretical model This ensures that the model curve best matches the experimental points. (3) Weighted least squares: ; In summary: the experimental variogram is the "observed reality" (discrete, noisy, finite), while the theoretical model is the "mathematical ideal" (continuous, smooth, infinite extrapolation). The relationship between the two is that the theoretical model is used to fit experimental data, thereby achieving: noise reduction and smoothing; continuous spatial prediction; ensuring the Kriging equations have solutions; and extracting physically interpretable parameters (range). (Reflects the scope of impact).

[0036] B32. Solving the Kriging equations For prediction points : ; in, For the sample covariance, its elements ; Let the sample-predicted point covariance have its elements ; For the optimal weight vector, each element Indicates the first The contribution weight of each sample point to the prediction point; For Lagrange multipliers (mean estimate); Estimated value: ; in, This is the vector of sample observations, i.e., the measured data of all sample points; For the first The observed values ​​of each sample point, i.e., the submachine at the location Sensor readings at the location.

[0037] Estimated variance: .

[0038] B33, Closed-loop feedback Using prediction points Update the risk field (i.e., update the risk field of A233), and the updated risk field. Used for: For aircraft that did not depart in this round, when the triggering conditions are met, they will return to A32 with the current predicted state as the new starting point to replan their reconnaissance flight path; New ROI hotspot areas are generated, if , If the threshold corresponding to the observed information is used, a new ROI hotspot region is added; In the next round of tasks, as a priori risk field, return to step A24 to update the dispatch cost.

[0039] The above scheme effectively forms a closed loop of "prediction-planning-execution-measurement-update-re-prediction".

[0040] This invention utilizes a drone carrier to provide endurance (4 to 6 hours), allowing the carrier to hover over disaster areas for reconnaissance. Based on the reconnaissance results, it intelligently releases drones to conduct detailed surveys of hotspots, achieving a seamless connection from macro-level situational awareness to micro-level detail identification. Simultaneously, it employs multi-dimensional information fusion, combining the aerial view from the drone carrier with data collected by the drones at close range to create a complete disaster situation report. Furthermore, the drones act as communication relays, extending the coverage of the drone carrier and constructing a resilient communication network. Therefore, the technical solution of this invention possesses at least the following beneficial technical effects: 1. Improve overall battery life; 2. The coverage area can be expanded from single-point reconnaissance to large-area dynamic focusing, achieving full coverage; 3. The reconnaissance accuracy has been upgraded from fixed resolution to multi-level zoom, which can be adaptively improved from 1 meter level to centimeter level; 4. The response speed has been upgraded from manual scheduling to AI real-time decision-making, achieving a response time within minutes; 5. Cost control is reduced from multiple large machines to one mothership plus multiple submachines, which can reduce costs by more than 60%; 6. The ability to adapt to complex environments has been upgraded from being prone to disconnection to being able to form a relay network of slave units, enabling flexible communication; 7. The sub-aircraft is an electric multi-rotor reconnaissance sub-aircraft with vertical take-off and landing and hovering capabilities. After release, it can fly directly along the planned trajectory without the need for gliding trajectory optimization, which can simplify the deployment decision and improve the mission response speed.

[0041] Based on the same inventive concept, this application also provides a system corresponding to the method in Embodiment 1, as detailed in Embodiment 2.

[0042] Example 2 like Figure 2As shown, this embodiment provides a reconnaissance system that dynamically releases submachines and fuses data. The system includes: The scanning and modeling module is used to generate macroscopic situational information by using the macroscopic reconnaissance payload of the UAV carrier to perform large-scale scanning when the UAV carrier's carrier-based aircraft arrive at the disaster area; and to generate a 3D model of the site by using the 3D modeling payload of the UAV carrier, and output 3D environmental information. The hierarchical decision-making and optimization module is used to determine the target area that needs to be closely reconnoitered based on macro-situation information. The UAV carrier calculates the timing of the deployment of the drones based on the three-dimensional environmental information and the characteristics of the target area, and generates the drone reconnaissance trajectory. The hierarchical decision-making and optimization strategy includes four levels: ROI hotspot area identification, drone deployment necessity assessment, drone reconnaissance trajectory planning, and deployment timing calculation. The sub-aircraft deployment module is used to transport sub-aircraft from the unmanned aerial vehicle (UAV) carrier to the deployment location and deploy the sub-aircraft according to the deployment timing. The data fusion module is used by the deployed drones to conduct detailed reconnaissance of the target area according to their reconnaissance flight paths, and to transmit the detailed information obtained back to the drone mothership in real time. The drone mothership then integrates the detailed information from the drones to generate complete disaster area situation information.

[0043] It should be noted that: in this embodiment, the specific functions implemented by the scanning modeling module are exactly the same as step S1 in the method; the specific functions implemented by the hierarchical decision and optimization module are exactly the same as step S2 in the method; the specific functions implemented by the sub-machine deployment module are exactly the same as step S3 in the method; and the specific functions implemented by the data fusion module are exactly the same as step S4 in the method. For details, please refer to the detailed description of Embodiment 1. Here, we will not go into detail about each module.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A reconnaissance method for dynamically releasing submachines and fusing data, characterized in that, include: Step S1: When the UAV carrier-carried aircraft arrives at the disaster area, the macro-reconnaissance payload of the UAV carrier performs a large-scale scan to generate macro-situation information; the 3D modeling payload of the UAV carrier scans to generate a 3D model of the site and outputs 3D environmental information. Step S2: Based on the macro situation information, determine the target area that needs to be closely reconnoitered. The UAV carrier calculates the timing of the deployment of the drones based on the three-dimensional environmental information and the characteristics of the target area, and generates the drone reconnaissance trajectory. The hierarchical decision-making and optimization strategy includes four levels: ROI hotspot area identification, drone deployment necessity assessment, drone reconnaissance trajectory planning, and deployment timing calculation. Step S3: The unmanned aerial vehicle carrier carries the sub-drones to the deployment location and deploys the sub-drones according to the deployment timing; Step S4: The deployed drones fly to the target area according to their reconnaissance flight paths to conduct detailed reconnaissance and transmit the detailed information they have collected back to the UAV mothership in real time. The UAV mothership integrates the detailed information from the drones to generate complete disaster area situation information.

2. The reconnaissance method for dynamically releasing submachines and data fusion according to claim 1, characterized in that, The ROI hotspot region identification specifically involves: using RANSAC plane fitting and Euclidean distance clustering to identify ROI hotspot regions and outputting a set of ROIs containing bounding boxes; The necessity assessment of the dispatch of sub-machines is specifically as follows: the dispatch value is calculated based on information gain and comprehensive dispatch cost, and the optimal allocation is solved by the Hungarian algorithm to obtain the allocation scheme. The sub-aircraft reconnaissance trajectory planning is specifically as follows: according to the allocation scheme, the RRT*-Constrained algorithm is used to plan the trajectory, generate the sub-aircraft reconnaissance trajectory from the deployment point to the target hotspot area in the ROI hotspot area set, and ensure that the generated sub-aircraft reconnaissance trajectory meets the requirements of communication maintenance, energy constraint and obstacle avoidance. The specific calculation of the deployment timing is as follows: the deployment point is calculated by the position of the UAV mothership, the attitude matrix of the UAV mothership and the mounting position of the sub-aircraft, the optimal deployment timing is solved based on the dynamic characteristics of the electric multi-rotor, and communication maintenance and energy efficiency are taken as constraints.

3. The reconnaissance method for dynamically releasing submachines and data fusion according to claim 2, characterized in that, The ROI hotspot region identification specifically includes: A11. Input the point cloud data from the 3D environment information; A12. The RANSAC plane fitting algorithm is used for ground segmentation to separate the point cloud data into ground point sets and non-ground point sets. A13. Use the Euclidean distance clustering algorithm to divide spatially continuous non-ground points into different clusters; A14. Extract multi-dimensional features for each cluster, including at least the coefficient of variation of height, normal vector dispersion, temperature anomaly index or color anomaly degree, to quantify the disaster characteristics. A15. Calculate the ROI hotspot region score for each cluster, filter the ROI hotspot regions according to the set ROI threshold, and output the set of filtered ROI hotspot regions.

4. The reconnaissance method for dynamically releasing submachines and data fusion according to claim 2, characterized in that, The assessment of the necessity of dispatching submachines specifically includes: A21. Input the data for the ROI hotspot region set, the data for the sub-machines, and the data for the unmanned aerial vehicle carrier; A22. Calculate information gain: This includes calculating the current information entropy based on the discrete probability distribution, calculating the observation likelihood, and calculating the posterior entropy, expected posterior entropy, and information gain. A23. Based on the data of the ROI hotspot area set, the data of the sub-machines and the data of the UAV mothership, calculate the Euclidean distance, flight energy consumption, path risk and overall dispatch cost; A24. Calculate the dispatch value based on information gain and overall dispatch cost, use the Hungarian algorithm to solve for the optimal allocation, and output the allocation scheme.

5. The reconnaissance method for dynamically releasing submachines and data fusion according to claim 2, characterized in that, In the reconnaissance trajectory planning of the sub-aircraft, obstacle avoidance requirements are evaluated by a risk function that integrates three-dimensional environmental information and disaster dynamic evolution information. The risk function is a weighted sum of geometric risk and dynamic risk. Geometric risk is calculated by using a signed distance field to calculate the distance from the position to the obstacle, and dynamic risk is predicted by disaster physical simulation. The weights are adaptively configured according to the disaster type. Target inspection is judged by a distance threshold.

6. The reconnaissance method for dynamically releasing submachines and data fusion according to claim 1, characterized in that, The process of fusing detailed information from the unmanned aerial vehicle (UAV) carrier with that of the sub-drones to generate complete disaster area situation information involves the following steps: a multimodal spatiotemporal alignment and data fusion algorithm is used to spatiotemporally align and overlay the detailed information of the sub-drones with the macroscopic situation information of the UAV carrier and the on-site 3D model to generate complete disaster area situation information; at the same time, the risk field is updated by combining Kriging spatial interpolation, and the updated risk field is used for sub-drones that have not yet departed in this round, the generation of new ROI hotspot areas, or the next round of missions.

7. A reconnaissance system that dynamically releases submachines and fuses data, characterized in that, The system includes: The scanning and modeling module is used to generate macroscopic situational information by using the macroscopic reconnaissance payload of the UAV carrier to perform large-scale scanning when the UAV carrier's carrier-based aircraft arrive at the disaster area; and to generate a 3D model of the site by using the 3D modeling payload of the UAV carrier, and output 3D environmental information. The hierarchical decision-making and optimization module is used to determine the target area that needs to be closely reconnoitered based on macro-situation information. The UAV carrier calculates the timing of the deployment of the drones based on the three-dimensional environmental information and the characteristics of the target area, and generates the drone reconnaissance trajectory. The hierarchical decision-making and optimization strategy includes four levels: ROI hotspot area identification, drone deployment necessity assessment, drone reconnaissance trajectory planning, and deployment timing calculation. The sub-aircraft deployment module is used to transport sub-aircraft from the unmanned aerial vehicle (UAV) carrier to the deployment location and deploy the sub-aircraft according to the deployment timing. The data fusion module is used by the deployed drones to conduct detailed reconnaissance of the target area according to their reconnaissance flight paths, and to transmit the detailed information obtained back to the drone mothership in real time. The drone mothership then integrates the detailed information from the drones to generate complete disaster area situation information.

Images