Unmanned aerial vehicle cluster-oriented flight path planning method for rapidly covering key disaster area

Abstract

The invention discloses a flight path planning method for rapidly covering a key disaster area by an unmanned aerial vehicle cluster. The method comprises the following steps: constructing a disaster-affected area tile map for gradually analyzing disaster-affected conditions in a single tile map area; information obtained by a sensor, such as GPS position and image data of an unmanned aerial vehicle, is transmitted into an image segmentation network based on deep learning, and the type and the influence range of a disaster area are extracted; establishing a disaster degree calculation model in a single tile map, and calculating whether a current block area has a key disaster area or not; the disaster situation distribution condition of the key disaster-affected area is established, a disaster situation estimation model is generated to detect whether there are remaining undetected tile maps in the key area, and a detection route is planned for an unmanned aerial vehicle cluster by using an unmanned aerial vehicle course guidance function based on disaster situation evolution characteristics in subsequent route planning. According to the method, the coverage detection efficiency of the unmanned aerial vehicle cluster on key disaster areas is improved, the complexity of flight path planning of the unmanned aerial vehicle cluster system is reduced, and the method can be widely applied to the fields of unmanned aerial vehicle technology and artificial intelligence.

Description

Technical Field

[0001] The present invention relates to the technical fields of artificial intelligence and unmanned aerial vehicle technology, and particularly relates to a path planning method for a swarm of unmanned aerial vehicles to quickly cover key disaster areas. Background Art

[0002] Disaster events are closely related to the lives and safety of the people, social security and stability, and people's livelihood development. In recent years, unmanned aerial reconnaissance systems represented by unmanned aerial vehicles have been widely used in reconnaissance tasks at various disaster sites due to their advantages such as wide field of vision, low cost, easy organization, and strong flexibility. A multi-unmanned aerial vehicle cluster detection system can also shorten the time for reconstructing the disaster situation and significantly improve the efficiency of rescue decision-making.

[0003] In the prior art, for disaster reconnaissance tasks where the situation of the disaster area is unknown and key disaster areas need to be explored, multiple unmanned aerial vehicles are usually used to conduct full-coverage searches of the entire disaster area, so as to obtain sequence aerial images through the image sensors carried by the unmanned aerial vehicles, infrared imaging information obtained by infrared sensors, data of trapped people detected by vital sign detectors, etc. In order to quickly obtain specific information about the disaster area and plan the rescue deployment in the disaster area, it is usually necessary to give priority to detecting key disaster areas, such as buildings, crowded places, areas where dangerous goods exist, etc. How to optimize the path planning of multiple unmanned aerial vehicles to cover key disaster areas within the disaster area has received extensive attention at home and abroad.

[0004] In existing published literature, for example, in the existing invention patent application document "A Multi-UAV Disaster Detection Method and System" with the publication number CN115016540A, the existing method includes: constructing the problem of multi-UAV path planning to maximize the disaster detection effect; decoupling the problem of multi-UAV path planning into a problem of maximizing the detection point positions based on global trajectory planning and a problem of maximizing the detection effect based on local path planning; respectively solving the problem of maximizing the detection point positions based on global trajectory planning and the problem of maximizing the detection effect based on local path planning to obtain the target solution of the problem of multi-UAV path planning; and performing motion control on the multi-UAV according to the target solution to complete disaster detection. And in the existing invention patent application document "A Multi-UAV Coverage Search Planning Method and System Based on Estimated Flight Range" with the publication number CN118131816A, the existing method includes: initial planning processing: setting the task area and converting it into a convex polygon area; calculating coverage parameters to obtain coverage parameters such as the heading field of view size, radial field of view size, aerial photography interval, and flight strip interval; coverage task allocation: determining the flight strip direction; multi-aircraft area allocation, estimating the total length of the straight-line segments and turning trajectories of the UAVs, and combining the estimated flight range to allocate task sub-areas to each UAV; coverage trajectory planning: calculating sequential trajectory points, and through the intersection detection of the straight-line segment trajectory and the convex polygon, cyclically expanding to calculate sequential trajectory points; determining the starting points of multi-aircraft tasks, combining the estimated flight range to determine the task starting points of each UAV, and planning the coverage search trajectories of each UAV. From the specific implementation content of the foregoing existing technologies, it can be seen that the current research represented by the foregoing existing solutions mainly has the following deficiencies: the actual needs of disaster detection and post-disaster rescue are not fully considered, the disaster area is not modeled and key areas are not segmented, resulting in the problem that key areas are not covered and it is impossible to carry out applications in the actual disaster environment; the method based on reinforcement learning has the problem of insufficient generalization, and it is impossible to ensure that the trained network is applicable to multiple UAV systems and multiple disaster environments; in some cases, only the solution of the local optimal problem is considered, which may lead to problems such as too long final coverage convergence time. Summary of the Invention

[0005] To solve the above technical problems, the purpose of the present invention is to provide a multi-UAV trajectory planning method for rapid disaster detection and preferentially covering key disaster areas.

[0006] The technical problem to be solved by the present invention lies in: how to solve the technical problems that the UAV cluster in the existing technology cannot quickly cover the key disaster areas, the UAV system cannot autonomously judge the key disaster areas, and the lack of technical means for cluster trajectory planning based on the data obtained by the UAV sensors.

[0007] The present invention solves the above technical problems by adopting the following technical solutions: The trajectory planning method for the UAV cluster to quickly cover the key disaster areas includes: Construct a method for acquiring and analyzing sensor information of an unmanned aerial vehicle (UAV) cluster, and establish the initial distribution of the overall disaster area, specifically including: S1. Construct a tile map of the disaster area for gradually analyzing the disaster situation within a single tile map area; S2. Input information obtained by sensors, such as UAV GPS positions and image data, into an image segmentation network based on deep learning to extract the types and affected ranges of the disaster areas; S3. Establish a calculation model for the degree of disaster within a single tile map to calculate whether there are key disaster areas in the current segmented area. If so, the key areas need to be covered; S4. Establish the disaster situation distribution of key areas and generate a disaster situation estimation model to detect whether there are remaining undetected tile maps in the key areas. If so, route planning for detection will be carried out in subsequent flight path planning.

[0008] In a more specific technical solution, step S1 includes: S11. Preset the regional boundary for the disaster situation detection of the UAV cluster, where the setting of the regional boundary can be input by manually entering the longitude and latitude of boundary points; S12. Construct a tile map of the disaster area. According to the boundary range set in S11, divide the tile map according to the interval longitude and interval latitude. The size of each tile map is determined by the interval longitude and interval latitude, and its area size is equal to the result of multiplying the interval longitude and interval latitude after conversion to meters; In a more specific technical solution, step S2 includes: S21. Collect images of the disaster area from the UAV image sensor, annotate them as an image segmentation data set, and send the data set into an image segmentation network based on a pyramid pooling module for training; In a more specific technical solution, step S21 includes: S211. First, input the image into a convolutional neural network to extract the feature map; S212. Pass the feature map output by the convolutional neural network through a compression and excitation network to perform dynamic channel feature recalibration; S213. Input the feature map into the pooling pyramid module for multiple feature extraction, and obtain and merge the feature maps output by each convolutional block; S214. Through a fully convolutional neural network, connect the outputs in S212 and S213, and merge them again as the output feature, so that the features of the original image are retained to improve the receptive field of the neural network model.

[0009] S22. Based on the neural network weights obtained through training in S21, identify the foreground area and background area in the images collected by the drone, and identify the contour ranges of each area in the disaster-stricken area according to the annotation of the segmented images, so as to calculate the area size and disaster degree of the disaster-stricken area.

[0010] It should be noted that during the data annotation process, in order to minimize the difference between the image segmentation results identified by the neural network model and the data annotation, the background areas that do not need to be identified in the images should also be annotated and marked as the background class, so as to improve the recognition effect of the neural network model.

[0011] In a more specific technical solution, step S3 includes: S31. Establish a calculation model for the disaster degree within a single tile map:

[0012] In the formula, is the pixel area of the disaster-stricken area is the damage intensity of the disaster-stricken area, is the weight represented by the type of the disaster-stricken area, is the weight represented by the type of the disaster-stricken area, is the final disaster degree value; S32. Judge whether there is a key disaster-stricken area in the current tile map according to the result identified by the image segmentation model, and calculate the disaster degree according to the disaster degree calculation model described in S31; In a more specific technical solution, step S32 includes: S321. If there is a key disaster-stricken area in the current tile map, construct a distribution fitting model for the distribution of the key disaster-stricken area. The constructed distribution fitting model is obtained by linearly weighting and approximating multiple normal distribution functions. Assuming that the current tile map belongs to a certain key disaster-stricken area, construct a Gaussian function based on the disaster degree of the current tile map:

[0013] In the formula, is the disaster intensity distribution function, is the disaster weight represented by the current tile map, is the Gaussian function, is the number of the current tile map, is the expected value of the disaster intensity, is the variance of the disaster intensity.

[0014] S322. For the overall tile map, it is necessary to construct an overall distribution function of the disaster situation:

[0015] In the formula, is the overall disaster situation distribution function, K is the number of key disaster areas, is the weight of the th key disaster area, is the probability density function, is the Gaussian distribution of the th key disaster area;

[0016] In a more specific technical solution, step S4 includes: S41. Calculate the tile map area covered by the key disaster area according to the disaster situation estimation model described in S3, and use the constructed disaster situation estimation model to simulate the disaster. S42. Check the undetected map parts in the covered tile map area. Among them, it is necessary to determine the level and range of the tile map to be loaded according to the geographical coordinates and range of this area, and the tile accuracy level determines the resolution of the map; S43. Record the undetected tile map parts, and mark the undetected tile map parts separately in the overall tile map for subsequent unified planning of the re-detection route.

[0017] The present invention solves the above technical problems by adopting the following technical solutions: The method for path planning of an unmanned aerial vehicle (UAV) cluster to quickly cover key disaster areas further includes: Establish a path planning model for the UAV cluster according to the distribution of the affected degree of the disaster area, specifically including: S1. A method for representing a two-layer map model divides the overall tile map information of the disaster area into two levels: the bottom layer map and the top layer map, and each level carries different information and functions.

[0018] Furthermore, in order to enable the two-layer map model to assist the UAV cluster in achieving more efficient disaster detection path planning and information collection, it is necessary to collect detailed geospatial data, including terrain, landform, buildings, roads, etc. In the above-mentioned method for acquiring and analyzing the sensor information of the UAV cluster, the data collected by the UAV cluster is processed and analyzed, and useful information needs to be extracted and classified. According to needs, the map information is divided into two levels: the bottom layer and the top layer. The bottom layer map contains detailed geospatial information, while the top layer map is an abstraction and simplification of the bottom layer map. For the convenience of rescue personnel to view and use, the method for representing the two-layer map model given by the present invention also needs to use GIS software or map rendering tools to render and visualize the two-layer map.

[0019] The two-layer map model given by the present invention has the following characteristics; Complex geospatial information can be organized according to different levels, enabling rescue personnel to select and view map information at different levels as needed. The double-layer map model can be extended or modified according to requirements to adapt to different application scenarios and needs. By abstracting and simplifying the information of the underlying map, rescue personnel can more easily understand the spatial relationships and functional layouts.

[0020] S2. A UAV heading guidance function based on the characteristics of disaster evolution, used for generating detection waypoints for UAV clusters; It should be noted that common disaster types (such as earthquakes, floods, wildfires, forest fires, etc.) have obvious evolution characteristics, which are related to the spread speed, influence range, change trend, etc. of the disasters. Through historical data analysis and real-time monitoring data, a mathematical model or prediction model of disaster evolution can be established. For example, the distribution of the affected degree in the disaster area mentioned above, a UAV heading guidance function based on the characteristics of disaster evolution; S3. Construct the UAV cluster trajectory planning problem to achieve coverage and information collection of key disaster areas, for generating the flight routes of UAV clusters for detecting key disaster areas; In a more specific technical solution, step S2 includes: S21. Divide the target area into several sub-areas according to the severity and distribution range of the disaster. Each sub-area can have different priorities and monitoring requirements; S22. Based on the disaster evolution model, predict the possible future diffusion paths and influence ranges of the disaster, and generate the initial flight waypoints of the UAVs; S23. During the flight, update the disaster information according to the real-time monitoring data, and dynamically adjust the flight waypoints of the UAVs to cope with sudden changes in the disaster situation; S24. The designed UAV heading guidance function based on the characteristics of disaster evolution has the specific form of:

[0021] In the formula, respectively represent the three-dimensional coordinates and height of the disaster point, and represent the source strength and the disaster point diffusion speed factor, and represent their probability distributions in two main directions.

[0022] In a more specific technical solution, step S3 includes: S31. Use the double-layer map model to construct the global and local distribution information for the disaster area, and plan the navigation points for dispersion for the UAVs in the upper-layer map; Among them, in order to model the tile map as a disaster distribution probability map, it is also necessary to initialize the probability density function of the disaster distribution center. As the detection process progresses, the probability density function will also change accordingly.

[0023] S32. Integrate the image segmentation model with the disaster intensity assessment module, process the image information obtained by the UAV sensor, and establish a Gaussian mixture model for key disaster areas for local trajectory planning; S33. Map the map information updated by the sensor in the lower-level map to the upper-level map to perform rolling updates on the navigation points of the UAV system, and finally obtain a set of routes that conform to the priority detection of key disaster areas; The present invention has the following advantages compared with the prior art: The present invention is applicable to any selected disaster area, and the boundary points of the disaster area can be manually selected to generate a polygon area. Combining the image segmentation model based on deep learning and the distribution of key disaster areas to determine the distribution of the overall disaster area, and performing preliminary trajectory planning according to the UAV heading guidance function of the disaster evolution characteristics. When it is detected that the disaster area spreads or new disasters occur, the flight waypoints of the UAV are dynamically adjusted. Therefore, the present invention has the advantages of universality in multiple disaster scenarios and high coverage of key disaster areas.

[0024] The present invention solves the problems of weak pertinence and omission of key area coverage in the existing technology for UAV cluster detection in disaster areas, and provides a trajectory planning method for UAV clusters to quickly cover key disaster areas in disaster areas with disaster evolution characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the basic steps of the trajectory planning method for UAV clusters to quickly cover key disaster areas in Embodiment 1 of the present invention; Figure 2 It is a representation of tile maps of different sizes in Embodiment 1 of the present invention; Figure 3 It is the initial waypoint position in Embodiment 1 of the present invention; Figure 4 It is the steps of extracting the disaster area type and influence range after image segmentation and recognition of the disaster area in Embodiment 1 of the present invention; Figure 5 It is the double-layer map expression in Embodiment 1 of the present invention; Figure 6 It is the initial disaster distribution in Embodiment 1 of the present invention; Figure 7 It is the maximum tile map range and the corresponding grid area in Embodiment 1 of the present invention; Figure 8The detection model of the visual sensor carried by the UAV cluster involved in Embodiment 1 of the present invention when detecting the disaster area; Figure 9 The double - layer map update process of Embodiment 1 of the present invention; Figure 10 The UAV heading guidance function based on the disaster evolution characteristics of Embodiment 1 of the present invention; Figure 11 The disaster evolution process of Embodiment 1 of the present invention; Detailed implementation manners

[0026] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] It should be noted that the present invention uniformly converts the initially given task area into a convex polygon area and selects the direction perpendicular to the width of the polygon as the flight path direction, which has fewer turning times and total flight distances.

[0028] In response to the application requirements of UAV wide - area reconnaissance situation awareness, the present invention comprehensively considers the coupling relationship between optoelectronic image stitching and flight path planning, calculates coverage planning parameters such as field - of - view size, frame extraction interval, and flight path width, and can achieve precise sub - area planning in irregular areas.

[0029] Embodiment 1

[0030] As Figure 1 shown, the flight path planning method for the UAV cluster to quickly cover the key disaster areas provided by the present invention includes the following basic steps: Step S1: Initialize the overall tile map of the disaster area according to the boundary of the disaster area; Preferably, for the generation of the tile map, the map data can be rendered hierarchically according to the zoom level of the map. At a lower zoom level, relatively rough map information is displayed; at a higher zoom level, more detailed map information is displayed. Using the slicing function of GIS software, the rendered map is cut into multiple square tiles according to a certain size. In this embodiment, the minimum tile map unit is 256x256 pixels; It should be pointed out that during the slicing process, the operator can select different slicing schemes according to needs to optimize the loading speed and display effect of the tiles; S2: Load the information obtained by the sensor, such as the UAV GPS position and image data, onto the tile map; Specifically, for selecting a suitable map projection method, such as Web Mercator projection, to accurately represent the longitude and latitude information of the Earth's surface on a two-dimensional plane As Figure 2 shown, in this embodiment, the images collected by the UAV are sequentially loaded into the corresponding tile map spaces. To facilitate the scaling and storage of the tile map, the maximum size of all tile maps does not exceed 2048x2048; According to the characteristics and display requirements of the disaster situation data, a suitable color scheme, layer style, and symbol system should be set for the map; As Figure 3 shown, in this embodiment, the initial target waypoints of the UAV cluster are scattered in the disaster area; during the flight of the UAV cluster, corresponding image data will be continuously collected, and GPS information will be saved. The collected sensor data will be filtered and corrected to ensure the accuracy and integrity of the data; S3. Input the images of the disaster area collected by the UAV into an image segmentation network based on deep learning to extract the types and affected ranges of the disaster areas; As Figure 4 shown, in this embodiment, after the image segmentation and recognition of the disaster area, step S3 of extracting the types and affected ranges of the disaster areas further includes the following specific steps; S31. Image preprocessing, including enhancing contrast and image denoising; S32. Image segmentation area recognition, assigning values to the disaster intensity distributions of each disaster area according to predefined disaster area parameters; S33. Calculate the affected degree of the area within a single tile map; S34. Construct a two-layer map model of the disaster area based on the overall tile map. As Figure 5 shown, divide the overall tile map information of the disaster area into two levels: the bottom layer map and the top layer map.

[0031] In this embodiment, according to the basic requirements for the initialization and construction of the two-layer map model, set the UAV sensor data acquisition sites and solve the disaster coverage parameters; In this embodiment, the parameters that need to be set for the construction of the two-layer map model include: the flight speed of the UAV cluster, the flight altitude, the field of view angle of the visual payload carried, the maximum size of the tile map, the minimum size of the tile map, the initial disaster situation distribution function of the top layer map in the two-layer map model, the grid size of the bottom layer map in the two-layer map model, the weight of the disaster situation distribution type, and the weight of the disaster situation distribution area; As Figure 6 shown, in this embodiment, the initial state of the disaster situation distribution is initialized in the top layer map of the two-layer map model; As Figure 7As shown, in this embodiment, in the double-layer map model, the maximum tile map range and the corresponding grid area are initialized in the underlying map; As Figure 8 shown, in this embodiment, the UAV swarm carries the same image acquisition sensor. The detection model of the disaster area is given in the figure, and the time interval for image acquisition is 1 s; S35. As Figure 9 shown, in this embodiment, after the double-layer map model completes the initialization construction of the top layer map and the bottom layer map, it also needs to be updated as the detection progress changes. The update process includes: S351. Single-sensing model construction In this subsection, the disaster area map is modeled as a double-layer map structure. The upper layer map is an implicit information entropy map, and different probability distribution fields are constructed for the ground object centers in the disaster area. Its influence radius and probability intensity are determined by the ground object classification results and are used for UAV global navigation planning. The lower layer map is an explicit occupancy grid map, which is used for UAV local trajectory planning.

[0032] After the upper layer map is established, this paper establishes a probability field map to describe the disaster impact intensity of different regions. The probability field map uses the classification information of ground objects in the disaster area and combines the calculated value of information entropy to characterize the impact range and intensity of the disaster in this region. In the lower layer map it is characterized by using an occupancy grid map for the disaster area, where each grid contains three states: free, occupied, and unknown. The occupancy probability of the unknown grid is inherited from the probability of the upper layer map. The lower layer map consists of grids, and the state of each grid is identified by the following function:

[0033] In the formula, is the grid occupancy value

[0034] For the classical occupancy grid map, the occupancy value of its grid is between (0, 1), and the grid value is independently taken and not affected by other grids. In the double-layer map model of this paper, the occupancy value used in the lower layer map is inherited from the disaster distribution model of the upper layer map and is used for UAV local trajectory planning.

[0035] S352. Disaster distribution value estimation

[0036] In the classical occupancy grid map, a random variable is used to record the observed value of the map, is the measurement value from the start time to time, is the UAV from the start time to The path sequence at a moment, for The posterior probability estimation of the map at a moment is denoted as:

[0037] The classical occupancy grid map updates the map using the Inverse Sensor Model (ISM), which is commonly used in the simultaneous mapping and localization problem of robots to detect map grids using sensors such as depth cameras or lidar. For each established map grid, there are generally multiple estimated values. Since map values are usually stored as integer data or floating-point data, directly multiplying the estimated probabilities for grids with estimated probabilities close to 0 will result in unstable calculation results. Therefore, it is converted to the logarithmic occupancy probability representation method:

[0038] In the formula, the logarithmic occupancy probability has a value range of (-inf, inf); S353. Update of the underlying map grid status Assume the map grid The neighborhood The neighborhood points in The observed values are completely known or partially known. Define the forward sensor model To represent The predicted value of, which is related to the Euclidean distance Of the neighborhood point Related: Assume the map grid Needs to be observed multiple times, and the evolution of the disaster situation intensity belongs to a Bayesian process

[86] , and the update of its observed values satisfies:

[0039] For the map grid Observed by multiple drones, it is considered that the observed values of each drone are independent of each other:

[0040] For the neighborhood-related map grid And , the observation of its disaster situation value is related, that is, when Is observed, its The disaster situation value estimation and probability distribution of the grid points within Will all be affected.

[0041] Through the above assumptions, the model establishment and status update description of the double-layer map Can be completed.

[0042] S4. The present invention also provides a UAV heading guidance function based on the characteristics of disaster evolution, which is used for generating detection waypoints for UAV clusters; A UAV heading guidance function based on the characteristics of disaster evolution, such as the distribution of the affected degree in the disaster area described above; In this embodiment, the characteristics of disaster evolution need to be selected according to different types of disasters; As Figure 10 shown, in this embodiment, the optional characteristics of disaster evolution target spreading disasters and diffusing disasters; In this embodiment, considering the different types of disasters, affected by different weather conditions, wind speeds, temperatures, and disaster environment obstacles, the UAV heading guidance functions and internal parameters to be selected in different scenarios are also different; In this embodiment, by comprehensively considering the actual characteristics of the estimated scenario and the distribution guiding points of the UAV cluster, the most suitable detection waypoints are generated for different UAV individuals, which can effectively reduce the computational amount of real-time operation and has better applicability for relatively harsh disaster reconnaissance and detection environments; S411. The designed UAV heading guidance function based on the characteristics of spreading disaster evolution has the specific form of:

[0043] In the formula, respectively represent the three-dimensional coordinates and height of the disaster point, and represent the source strength and the disaster point diffusion speed factor, and represent their probability distributions in two main directions; S412. The designed UAV heading guidance function based on the characteristics of diffusing disaster evolution has the specific form of:

[0044] Among them, is the field strength, is the influence radius, is the point to the Euclidean distance of the source point.

[0045] S42. During the flight, update the disaster information according to the real-time monitoring data

[0046] S43. According to the disaster distribution value obtained from the grid points in the underlying map, adjust the flight waypoints of the UAVs to preferentially cover the severely affected areas to cope with the sudden changes of the disaster; As Figure 11 shown, the waypoints of the UAV cluster have changed with the spread of the disaster, transferring from area A to area B; S5. Construct the UAV swarm trajectory planning problem, establish the disaster situation distribution in the key area to achieve the coverage and information collection of the key disaster areas, and plan the coverage routes of the UAV swarm for the key disaster areas, specifically including: S51. Use the map initialization function to initialize the simulation parameters, including the top-level map information, bottom-level map information, disaster situation distribution probability function, UAV path tree, and bottom-level map occupancy grid information. In a finite time step, perform trajectory planning for each UAV flight.

[0047] S52. During the trajectory planning process, first obtain the information of the sampling points and navigation points through the top-level map, and obtain the neighbor grid information of the current UAV in the bottom-level map for planning the local trajectory. After obtaining all the neighbor grid point information, use the image perception module to calculate the feasible waypoints, where is the image matching threshold.

[0048] S53. Through the path conflict check module, after ensuring that the trajectory from the current position to the next waypoint position satisfies the conflict-free path constraint, add the new path to the path tree as the updated result of the local trajectory. Through the disaster situation fitting model, the disaster situation intensity distribution in the known area can be fitted into multiple independent Gaussian distributions, and update the disaster situation intensity distribution and the occupancy value of the grid in the bottom-level map in the top-level map and the bottom-level map respectively to complete the calculation within the current simulation step.

[0049] The present invention plans the initial detection waypoints based on the initial parameters of the UAV swarm and the disaster situation distribution in the disaster area, and conducts the detection task of the key disaster areas in different simulated disaster scenarios. The method proposed by the present invention can effectively achieve the coverage detection of the key disaster areas, thereby quickly completing the aerial detection task of the entire disaster area.

[0050] The present invention obtains the detection sites to be detected in the key disaster areas by establishing a UAV heading guidance function based on the characteristics of disaster situation evolution and planning the fast coverage routes of the UAV swarm, combines the disaster situation distribution fitting function, guides the UAVs to move from the current disaster area to the next disaster area, reduces unnecessary distances and detection time, and preferentially completes the detection task of covering the key disaster areas, thereby realizing efficient UAV swarm trajectory planning.

[0051] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for tracking planning of drone clusters to quickly cover key disaster areas, characterized in that: The method comprises: S1. construct a tile map of the disaster-affected area to gradually analyze the disaster situation in a single tile map area; S2, passing the information obtained by the sensor, such as the GPS location and image data of the drone, into the image segmentation network based on deep learning to extract the type of disaster-affected area and the scope of impact; S3, establishing a disaster degree calculation model in a single tile map, calculating whether there is a key disaster-stricken area in the current block area, and if so, it is necessary to cover the key area; S4. Establish the disaster distribution in key areas and establish a disaster estimation model to detect whether there are any remaining undetected tile maps in the key areas. If so, detection route planning will be carried out in subsequent trajectory planning.

2. The method of constructing a disaster area tile map according to claim 1, for gradually analyzing the disaster situation in a single tile map area, characterized in that: The step S1 comprises: S11, presetting the regional boundary for the drone cluster to detect disasters, wherein the regional boundary can be set by manually inputting the longitude and latitude of the boundary point as input; S12, construct a tile map of the disaster-stricken area, and divide the tile map according to the interval longitude and interval latitude based on the boundary range set in S11. The size of each tile map is determined by the interval longitude and interval latitude, and its area size is equal to the multiplication result of the interval longitude and interval latitude converted into meters.

3. According to claim 1, the information obtained by the sensor, such as the GPS position and image data of the drone, is transmitted to the image segmentation network based on deep learning to extract the type of disaster-affected area and the scope of impact, characterized in that: The step S2 comprises: S21, collecting images of the disaster area from the drone image sensor, annotating them into an image segmentation dataset, and sending the dataset to an image segmentation network based on a pyramid pooling module for training; S22. According to the weights of the neural network trained in S21, the foreground area and the background area in the image collected by the drone are identified, and the contour range of each area in the disaster-stricken area is identified according to the annotations of the segmented image, so as to calculate the size and degree of disaster of the disaster-stricken area.

4. According to claim 3, the information obtained by the sensor, such as the GPS position and image data of the drone, is transmitted to the image segmentation network based on deep learning to extract the type of disaster-affected area and the scope of impact, characterized in that: The step S21 comprises: S211, first input the image into the convolutional neural network to extract the feature map; S212, subjecting the feature map output by the convolutional neural network to compression and excitation network to perform dynamic channel feature recalibration; S213, inputting the feature map into the pooling pyramid module to extract multiple features, and obtaining the feature maps output by each convolution block for merging; S214, through the fully convolutional neural network, connect the outputs in S212 and S213, and merge them again as output features, so that the features of the original image are retained to improve the receptive field of the neural network model.

5. According to claim 1, the disaster degree calculation model in a single tile map is established to calculate whether there is a key disaster-stricken area in the current block area, characterized in that: The step S3 comprises: S31. Establish a disaster degree calculation model within a single tile map: In the formula, The disaster-affected area The pixel area, is the damage intensity of the affected area, is the weight represented by the disaster-affected area type, is the final disaster severity value; S32. Determine whether there is a key disaster-stricken area in the current tile map according to the result identified by the image segmentation model, and calculate the disaster degree of the model according to the disaster degree as described in S31.

6. The method of determining whether there is a key disaster-stricken area in the current tile map according to the result of image segmentation model recognition according to claim 5, characterized in that: The step S32 comprises: S321. If there is a key disaster-stricken area in the current tile map, a distribution fitting model for the key disaster-stricken area is constructed. The constructed distribution fitting model is obtained by linear weighted approximation of multiple normal distribution functions. Assuming that the current tile map belongs to a key disaster-stricken area, a Gaussian function is constructed based on the disaster degree of the current tile map: In the formula, is the disaster intensity distribution function, The disaster weight of the current tile map representative team. is a Gaussian function, is the current tile map number, is the expected value of disaster intensity, is the variance of disaster intensity; S322. For the overall tile map, it is necessary to construct the overall disaster distribution function: In the formula, is the overall disaster distribution function, K is the number of key disaster areas, For the The weight of each key disaster area, is the probability density function, For the Gaussian distribution of key disaster areas; S322, setting different K values. According to the Gaussian distribution of different tile map areas, calculate the overall distribution function and covariance loss of the current disaster situation, and estimate whether there are any missed key disaster areas.

7. The method of establishing the disaster distribution of key areas according to claim 1, and establishing a disaster estimation model to detect whether there are any remaining undetected tile maps in the key areas, characterized in that: The step S4 comprises: S41, calculating the tile map area covered by the key disaster area according to the disaster estimation model described in S3, and simulating the disaster using the constructed disaster estimation model; S42, checking the undetected map portion in the covered tile map area, wherein the level and range of the tile map to be loaded need to be determined according to the geographic coordinates and range of the area, and the tile accuracy level determines the resolution of the map; S43, recording the undetected tile map portion, and marking the undetected tile map portion separately in the overall tile map for subsequent unified planning of re-detection routes.

8. The UAV cluster trajectory planning model established according to the distribution of disaster severity in the disaster area is characterized by: The method comprises: S1. A two-layer map model representation method divides the overall tile map information of the disaster area into two levels: the bottom map and the top map, each level carries different information and functions; S2, a UAV heading guidance function based on disaster evolution characteristics, used for UAV cluster detection waypoint generation; S3. Construct the UAV cluster trajectory planning problem to achieve coverage and information collection of key disaster areas, and generate UAV cluster routes for detecting key disaster areas.

9. The UAV heading guidance function based on disaster evolution characteristics according to claim 8 is characterized in that: The step S2 comprises: S21. Divide the target area into several sub-areas according to the severity and distribution of the disaster, each of which may have different priorities and monitoring requirements; S22. Based on the disaster evolution model, predict the possible future diffusion path and impact range of the disaster and generate the initial flight waypoints of the drone; S23. During the flight, the disaster information is updated according to the real-time monitoring data, and the flight waypoints of the drone are dynamically adjusted to cope with sudden changes in the disaster situation; S24. The designed UAV heading guidance function based on the disaster evolution characteristics is in the form of: In the formula, Respectively represent the three-dimensional coordinates and height of the disaster point, and represents the source strength and the diffusion speed factor of the disaster point, and Represents its probability distribution in two main directions.

10. The method for generating a cluster of drone routes for detecting key disaster-stricken areas according to claim 7, characterized in that: The step S3 comprises: S31. Use a two-layer map model to construct global and local distribution information for the disaster area, and plan dispersed navigation points for drones in the upper map; In order to model the tile map as a disaster distribution probability map, it is also necessary to initialize the probability density function of the disaster distribution center. As the detection process progresses, the probability density function will change accordingly; S32, integrating the image segmentation model with the disaster intensity assessment module, processing the image information obtained by the drone sensor, and establishing a mixed Gaussian model of the key disaster area for local trajectory planning; S33, mapping the map information updated by the sensor in the lower map to the upper map to perform rolling updates on the navigation points of the drone system, and finally obtaining a set of routes that meet the priority detection requirements of key disaster areas.

Images