Task allocation and conflict resolution system and method for cooperative operation of multiple unmanned aerial vehicles

Abstract

The invention relates to the technical field of unmanned aerial vehicle control, in particular to a task allocation and conflict resolution system and method for multi-unmanned aerial vehicle collaborative operation, and provides the following scheme: dividing an initial operation area and generating a response weight by constructing a crop growth state map and a three-dimensional plot model; based on path planning and resource adaptation, a flight route is dynamically generated, and the crop state and the unmanned aerial vehicle state are monitored in real time; when adjustment conditions are met, a multi-dimensional dynamic task evaluation model is constructed, and task migration and conflict decoupling are completed in combination with particle swarm optimization and an autonomous negotiation mechanism. The method is suitable for a precision agriculture scene, the unmanned aerial vehicle path dynamic scheduling in the operation area and the high-priority area precision coverage are realized, and the operation efficiency and the resource cooperation capability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle control, and in particular to a task allocation and conflict resolution system and method for multi-unmanned aerial vehicle cooperative operation. BACKGROUND

[0002] In the application of existing multi-unmanned aerial vehicle cooperative operation schemes in the field of agriculture, problems such as high overlap rate of operation paths, delayed response of task scheduling, and insufficient operation coverage of high-priority areas often occur. In actual execution, multiple unmanned aerial vehicles are prone to unreasonable task area division, uneven resource load, and frequent flight route conflicts, resulting in decreased overall operation efficiency. At the same time, in the face of crop state mutations or local abnormal conditions, existing methods cannot achieve timely and effective task migration and path adjustment, causing delayed operation response in hot areas and even the problem of missing unmanned areas. In addition, due to insufficient task priority identification capability, some high-value crop areas are not fully covered, affecting the precision operation effect and agronomic output quality.

[0003] To solve the above problems, the present application designs a task allocation and conflict resolution system and method for multi-unmanned aerial vehicle cooperative operation. SUMMARY

[0004] The technical problem to be solved by the present application is to address the deficiencies of the prior art by providing a task allocation and conflict resolution system and method for multi-unmanned aerial vehicle cooperative operation. By constructing a crop growth state map and a three-dimensional field model, the initial operation area is divided and a response weight is generated. Based on path planning and resource adaptation, the flight route is dynamically generated and the crop state and unmanned aerial vehicle state are monitored in real time. When the adjustment conditions are met, a multi-dimensional dynamic task evaluation model is constructed, and the task migration and conflict decoupling are completed by combining the particle swarm optimization and the autonomous negotiation mechanism. This method is suitable for precision agriculture scenarios, realizes dynamic scheduling of unmanned aerial vehicle paths in operation areas and precise coverage of high-priority areas, and improves operation efficiency and resource collaboration capability.

[0005] To achieve the above purpose, the present application provides the following technical solutions:

[0006] A task allocation and conflict resolution method for multi-unmanned aerial vehicle cooperative operation, the method comprising:

[0007] Obtaining a three-dimensional field model of a target area, and planning an unmanned aerial vehicle flight route based on the topographic features and distribution information of the three-dimensional field model through an optimization algorithm;

[0008] Allocating the unmanned aerial vehicle flight route to the corresponding unmanned aerial vehicle, and monitoring the growth state information of the crop area and the unmanned aerial vehicle state information during the operation of the unmanned aerial vehicle to determine whether the conditions for entering the adjustment phase are met;

[0009] If the condition is met, a region to be adjusted is obtained from the UAV flight route, and a UAV performing a corresponding task in the region to be adjusted is re-assigned according to a crop growth state map of the region to be adjusted.

[0010] The UAV flight route is planned by an optimization algorithm, including:

[0011] According to the three-dimensional land model and the number of unmanned aerial vehicles to be operated, a plurality of initial operation regions are divided, wherein the plurality of initial operation regions do not overlap with each other;

[0012] According to the priority of each initial operation region, the operation amount and the current state of the unmanned aerial vehicle to be operated, the initial operation region is assigned to the corresponding unmanned aerial vehicle;

[0013] According to the initial operation region corresponding to each unmanned aerial vehicle, a complete flight plan is generated in combination with a preset crop growth state map, wherein the crop growth state map represents a crop growth activity index, and the crop growth state map is constructed according to historical crop data.

[0014] The complete flight plan is generated in combination with the preset crop growth state map, including:

[0015] Boundary information and terrain features of a crop region are extracted from the initial operation region to determine the operation range and flight limitation of the unmanned aerial vehicle;

[0016] The crop region is divided into regular grids, wherein the size of each grid satisfies the effective coverage area of one-time shooting of the unmanned aerial vehicle, and each grid is labeled with a crop response weight associated with the crop growth state map, and the crop response weight is determined according to the growth state of the crop region in the crop growth state map;

[0017] A set of vertices of the boundary of the regular grid is extracted, a minimum circumscribed rectangle is calculated according to the set of vertices, and the long side direction of the rectangle is taken as the operation direction of the unmanned aerial vehicle;

[0018] Parallel paths are generated along the operation direction, and the start point and end point coordinates of each path are marked;

[0019] Obstacle detection is performed on the generated paths, if there is an obstacle grid in the path, the path is re-planned in the region corresponding to the obstacle grid by A* algorithm, the start point and end point of the path after bypassing the obstacle are re-connected, if there is no obstacle grid in the path, the path is added to the path set, and the paths are repeatedly traversed until all the paths are added to the path set;

[0020] A multi-objective evaluation function is established, wherein the objectives include path length, turning frequency, coverage repetition rate, energy consumption, and crop response weight, an evaluation score of each path in the path set is calculated according to the multi-objective evaluation function, a path corresponding to the highest evaluation score is taken as an optimal path scheme, and a flight route of the unmanned aerial vehicle is generated according to the path nodes, turning points, and obstacle avoidance parameters of the optimal path scheme.

[0021] Parallel paths are generated along the working direction, and the starting point and the ending point coordinates of each path are marked, including:

[0022] The starting point of the path is determined according to the crop response weight of the regular grid;

[0023] The ending point of the path is determined according to the remaining power of the unmanned aerial vehicle, the wind direction, and the return route;

[0024] The starting points of each parallel path are different from each other.

[0025] If there are multiple unmanned aerial vehicles to be worked in the initial working area, the method further includes:

[0026] The path set and the evaluation scores corresponding to the paths in the path set are obtained, and the paths are prioritized;

[0027] According to the current positions of the multiple unmanned aerial vehicles to be worked, and according to the descending order of the priorities of the paths, the unmanned aerial vehicle that meets the flight requirements and is closest to the starting point of the current path is selected as the execution unmanned aerial vehicle of the current path in sequence until there is no unmanned aerial vehicle to be worked.

[0028] The conditions for entering the adjustment stage include at least one of the following:

[0029] The variation amplitude of the crop growth state index of any sub-region in the crop area is greater than or equal to a preset generation threshold;

[0030] An abnormal area with spatial aggregation characteristics of pests and diseases, high temperature, or low humidity is detected in any sub-region in the crop area, wherein the high temperature and the low humidity can be determined by a preset high temperature threshold and a low humidity threshold, if the temperature corresponding to the sub-region is greater than or equal to the high temperature threshold, the abnormal area is a high temperature abnormal area; if the humidity corresponding to the sub-region is less than or equal to the low humidity threshold, the abnormal area is a low humidity abnormal area;

[0031] The power of the unmanned aerial vehicle is less than or equal to a preset flight safety threshold during the execution of the work;

[0032] The unmanned aerial vehicle deviates, the task is interrupted, or the path execution is timed out during the execution of the work;

[0033] The flight paths of the multiple unmanned aerial vehicles exist spatial intersection, and the safety interval is less than a preset tolerance threshold.

[0034] redistributing the unmanned aerial vehicle performing the corresponding task of the region to be adjusted according to the crop growth state map of the region to be adjusted, comprising:

[0035] constructing a multi-dimensional dynamic task evaluation model of the region to be adjusted, the multi-dimensional dynamic task evaluation model comprising a crop growth response dimension, an operation urgency degree dimension, a historical operation effect dimension and a task migration cost dimension;

[0036] calculating a task migration evaluation score of the region to be adjusted according to the multi-dimensional dynamic task evaluation model, the task migration evaluation score being used to determine a task migration priority of the region to be adjusted;

[0037] generating an unmanned aerial vehicle candidate task set according to the task migration evaluation score and a spatial distribution feature of the region to be adjusted;

[0038] obtaining an unmanned aerial vehicle performing the unmanned aerial vehicle candidate task set through a particle swarm optimization algorithm based on a remaining power, a flight distance, a current load, a historical operation efficiency and a position state of a remaining unmanned aerial vehicle, and assigning a corresponding task in the unmanned aerial vehicle candidate task set to the corresponding unmanned aerial vehicle.

[0039] the obtaining an unmanned aerial vehicle performing the unmanned aerial vehicle candidate task set, comprising:

[0040] calculating a resource fitness of each unmanned aerial vehicle performing the unmanned aerial vehicle candidate task set according to a resource fitness evaluation function preset in the particle swarm optimization algorithm;

[0041] determining an initial task allocation scheme of the unmanned aerial vehicle according to the resource fitness;

[0042] performing secondary allocation on the initial task allocation scheme of the unmanned aerial vehicle through autonomous negotiation logic between unmanned aerial vehicles according to a conflict situation in the initial task allocation scheme of the unmanned aerial vehicle, wherein the conflict situation comprises unmanned aerial vehicle path intersection, time-space overlap and task load imbalance.

[0043] the autonomous negotiation logic comprises conflict resolution between unmanned aerial vehicles according to a game theory strategy, wherein:

[0044] each unmanned aerial vehicle broadcasts negotiation information to other unmanned aerial vehicles according to its own state, and updates an order of performing tasks of the unmanned aerial vehicles through negotiation information interaction results until a balance state of minimizing path conflicts between the unmanned aerial vehicles and maximizing overall task benefits is reached.

[0045] A task allocation and conflict resolution system for multi-unmanned aerial vehicle cooperative operation, the system comprising:

[0046] The modeling module is configured to acquire a three-dimensional plot model of a target area, and construct a work area according to a terrain feature, obstacle information and a crop growth state atlas;

[0047] The path planning module is configured to generate an initial flight route of the multiple unmanned aerial vehicles based on the work area model;

[0048] The task allocation module is configured to perform dynamic task allocation on the flight path based on a crop response weight, a task priority and an unmanned aerial vehicle state parameter;

[0049] The conflict resolution module is configured to perform path adjustment, time window dislocation or flight height reconstruction when there is a path conflict, and output an updated flight control instruction.

[0050] Compared with the prior art, the application has the following beneficial effects:

[0051] The application realizes accurate identification and preferential coverage of a high-response area in a work area by task value modeling based on a crop growth state atlas and multi-target path optimization logic, improves the work pertinence and resource allocation efficiency of the unmanned aerial vehicle group, can respond to crop state changes and unmanned aerial vehicle resource fluctuations in real time during task execution, realizes flexible reconstruction of tasks and resolution of path conflicts, improves the stability and adaptability of multi-unmanned aerial vehicle collaborative work, and reduces work redundancy and omission. BRIEF DESCRIPTION OF DRAWINGS

[0052] Other features, objects and advantages of the application will become more apparent after reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0053] Figure 1 An exemplary application scenario of an embodiment of the application is shown in the following figure:

[0054] Figure 2 A flowchart of a task allocation and conflict resolution method for multi-unmanned aerial vehicle collaborative work of an embodiment of the application is shown in the following figure:

[0055] Figure 3 A crop identification network structure diagram of an embodiment of the application is shown in the following figure:

[0056] Figure 4 A flowchart of generating a complete flight plan of an embodiment of the application is shown in the following figure. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments of the application.

[0058] Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. Those skilled in the art will recognize that the embodiments described herein can be combined with one another.

[0059] The task allocation and conflict resolution method for multi-UAV cooperative operation proposed in the present application is applicable to a UAV operation system with requirements for multi-autonomous unit cooperation, real-time environmental state changes, and dynamic reconstruction of operation scenarios. The method is particularly suitable for the following types of application scenarios: complex agricultural operation environments in which the growth state of crops presents spatial non-uniformity, the division of task areas has boundary ambiguity, or path congestion, task abnormalities, or state imbalance may occur during UAV operation.

[0060] The present application does not rely on rigid plot division of task areas, nor does it require prior division of labor or fixed flight path templates between UAVs. Instead, by constructing a crop growth state map, a UAV resource parameter model, and a path conflict detection logic, the present application completes task reconstruction and scheduling optimization for dynamically changing target operation areas.

[0061] Application scenarios include, but are not limited to, the following:

[0062] In precision agriculture, multiple UAVs are involved in spraying, fertilizing, or sowing tasks on large areas of farmland, where there are significant differences in the growth state of crop areas;

[0063] Multiple UAVs simultaneously enter high-response crop areas to perform tasks, which may have risks such as power attenuation, operation failure, or path intersection;

[0064] During task execution, the plot environment, climate conditions, or crop health state change unexpectedly, requiring local reconstruction and redistribution of flight paths;

[0065] There are some high-value operation points or hotspots in the crop area, and their priorities need to be dynamically adjusted during task planning.

[0066] It should be noted that the related logic of scheduling and conflict resolution in the present application is not limited to agricultural operation scenarios, but can also be applied to:

[0067] In urban building group flight detection tasks, there are environments with obstructions and limited airspace resources;

[0068] In forest fire patrol operations, the distribution of monitoring targets is uneven, and the path is limited by wind direction and obstacles;

[0069] Coordinated inspection tasks in large industrial areas or traffic networks, where path assignment needs to be dynamically optimized based on target abnormal priority and device state.

[0070] The present application is particularly suitable for systems with one or more of the following characteristics:

[0071] There are heterogeneous response areas within the work area, and the spatial distribution of task density is uneven.

[0072] There are real-time differences in the operating states of multiple UAVs, such as different power reserves, work speeds, or historical loads.

[0073] The system does not have high-frequency global command and dispatching capabilities, and needs to be locally self-reassigned based on task weight and resource matching relationship.

[0074] Multiple UAV paths exist dynamic crossing risk, need to achieve multi-target path adaptive planning under the premise of no collision.

[0075] Please refer to Figure 1 , which is a schematic diagram of an exemplary application scenario provided by an embodiment of the present application.

[0076] As Figure 1 shown, multiple UAVs are deployed above the farmland area to perform coordinated work tasks, including UAV one, UAV two, and UAV three. The farmland area is divided by regular grid to clearly define the spatial boundaries of different work sub-areas. The crop symbols drawn in each grid represent the actual planting distribution of crops, where the lower right area is a target area with strong crop response, i.e., the area to be adjusted, which is marked with a dot matrix shading in the figure, indicating that this area needs to be prioritized for work or enter the task reconstruction scheduling phase.

[0077] Figure 1 The three UAVs shown in the middle are in different work states and flight stages, where the dashed line represents the work state. The UAV discovers a response hotspot area in the work area, which needs to be cooperated by other UAVs. The solid arrows between UAVs represent the dynamic migration path of the work task, where the curved line and arrow dashed line represent the adjustment process of the current flight path.

[0078] Figure 1 The schematic scenario shown is applicable to the following cases: the growth state of the crop area is uneven, the UAV path has overlapping or vacant areas, and the system dynamically adjusts the task assignment result based on the crop state map and re-plans the UAV flight path to maximize resource utilization efficiency and minimize path conflict risk.

[0079] It can be understood that when a UAV cannot complete a task, the method of the present application is also applicable.

[0080] Next, the multi-unmanned aerial vehicle cooperative operation task allocation and conflict resolution method provided by the embodiment of the application is introduced in combination with the drawings. Figure 2 The method shown includes the following S1-S3, and the specific steps are as follows:

[0081] S1: Obtain a three-dimensional land model of a target area, and plan an unmanned aerial vehicle flight route according to the topographic features and distribution information of the three-dimensional land model through an optimization algorithm;

[0082] In the embodiment, the target area can be a crop planting area, an economic forest area or a greenhouse extension farmland, and the land model is obtained through multi-view image modeling or laser point cloud reconstruction, without limitation on specific modeling means. The three-dimensional model contains data such as slope, height difference and boundary, which are used to determine the workable area and obstacle boundary. In order to improve work efficiency and reduce flight redundancy, the minimum turning route set generated based on the land geometric contour and crop row distribution direction is further combined with work priority, plant height map and regional avoidance attribute to perform flight path constraint optimization, and finally a flight route set satisfying the constraint condition is output. The optimization algorithm of this step does not depend on static land division, has flexible adaptation capability to complex planting areas, and is helpful to improve the route quality and reduce the path overlap problem caused by irregular areas.

[0083] S2: Assign the unmanned aerial vehicle flight route to the corresponding unmanned aerial vehicle, monitor the growth state information of the crop area and the unmanned aerial vehicle state information during the unmanned aerial vehicle operation, and determine whether the conditions for entering the adjustment stage are met;

[0084] In the embodiment, the initial allocation of the flight route considers the current starting position, remaining power and effective load of the unmanned aerial vehicle, and is realized through a flight path priority and resource matching degree calculation strategy. The growth state information can generate a response layer based on historical multi-spectral atlas, red edge index map or remote sensing time series data, reflecting the health degree distribution, abnormal hot spot and possible pest risk points of the current target area. The unmanned aerial vehicle state information includes but is not limited to battery power, heading stability, task execution delay and task failure times. The collection method of the unmanned aerial vehicle state information is obvious to those skilled in the art, as long as it can at least meet the basis for judging whether to enter the adjustment stage.

[0085] S3: If the conditions are met, obtain the adjustment area from the unmanned aerial vehicle flight route, and reassign the unmanned aerial vehicle performing the task corresponding to the adjustment area according to the crop growth state atlas of the adjustment area;

[0086] In this embodiment, the reallocation process considers the task migration cost function and the availability of UAV resources, combined with the remaining range, current task density and path offset cost, to optimize the matching of UAVs and task paths through a particle swarm algorithm. At the same time, in order to avoid multiple UAVs concentrating in the hot area at the same time, causing path conflict or resource overload, path conflict detection and dynamic conflict resolution logic are executed, specifically including time window staggered scheduling, height layer redistribution and path micro-offset scheme. Ensure that when the task is concentrated in the local area with high density, the system can still maintain path decoupling, reasonable allocation of resources and stable overall operation efficiency.

[0087] Before expanding the specific implementation logic corresponding to each step, the embodiments of the present application need to emphasize again:

[0088] In the multi-UAV cooperative operation scenario, not all operation areas have the characteristics of clear structure, uniform load and stable response. Taking agricultural application as an example, in large or complex crop areas, there are often the following uncertain factors:

[0089] First, the crop growth conditions have significant spatio-temporal heterogeneity, for example, the humidity gradient, disease and pest spread area or uneven fertility area in the same field may change rapidly over time;

[0090] Second, the running state of the UAV group is not absolutely synchronized, for example, power attenuation, path deviation, communication link delay, etc. will cause the decrease of operation consistency;

[0091] Third, due to factors such as flight airspace, obstacle distribution and vegetation height, some areas have resource coverage edge or path overlap risk in the initial path generation stage.

[0092] It can be understood that the traditional path allocation method is mostly based on fixed area blocking or cost matrix matching. When facing the above disturbance factors, it can only be fault-tolerant by increasing redundant flight or setting a buffer zone, which is easy to cause low resource utilization or lag in high response area operation. In addition, the existing method mostly separates the path conflict recognition and task adjustment, that is, first allocation, then conflict recognition, and then backtracking correction, which fails to consider the dynamic trade-off of path intersection possibility and priority value at the initial stage of allocation, resulting in the decline of global path efficiency.

[0093] The task allocation and conflict resolution method proposed in this application does not simply try to optimize the flight path itself, but models the path allocation problem as a multi-objective cooperative resource scheduling game scenario:

[0094] The operation path not only carries the spatial information of the crop state, but also implies multiple factors such as task priority, resource availability, and flight path interference potential. In actual execution, the task coordination between unmanned aerial vehicles is no longer regarded as a static resource allocation problem, but as a process balance that is constantly adjusted in the dynamic resource and target adaptation process.

[0095] Specifically, in task allocation, a task scoring model is constructed by fusing the response weight in the crop growth state map, the time window value, and the unmanned aerial vehicle state itself; then, a multi-objective optimization algorithm is used to dynamically search and match the optimal path allocation strategy. Compared with the traditional idea of path planning and then local correction, the conflict potential is introduced as a constraint variable in the early stage of task allocation in the present application, ensuring that even when unmanned aerial vehicles enter the high response area, safety isolation can be achieved through time scheduling, flight height layering, or path misplacement.

[0096] Further, in the conflict resolution process, a completely centralized instruction issuing mode is not adopted, but each unmanned aerial vehicle is allowed to broadcast its resource state and execution intention to realize local autonomous negotiation type path redistribution, based on the cooperative and equal bidding model in game theory, the intersection path segment is voted for weight value, and the operation order or control parameter is adjusted accordingly. Without relying on a central unified scheduling system, the overall scalability and robustness of the system are enhanced, which is particularly suitable for typical scenarios such as farmland, mountainous and forest areas where communication links are unstable or environmental restrictions are strong.

[0097] Next, the part of the present application method about the three-dimensional land model is further expanded.

[0098] Before performing the task on the target area, it is necessary to collect multi-angle images of the target area by the unmanned aerial vehicle, generate three-dimensional point cloud data using feature matching and structure light technology, and construct a complete land model. The three-dimensional model provides accurate terrain and boundary information, which is helpful for subsequent flight route planning and regional analysis.

[0099] In one example, the construction of the three-dimensional land model includes the following four aspects:

[0100] First, the image sensor carried by the unmanned aerial vehicle collects multi-angle images of the target area, and the multi-angle images are spliced by a feature matching algorithm to generate continuous images.

[0101] Specifically, the target area is photographed from different angles and heights to collect multi-angle images. These images are input into a feature matching algorithm such as ORB algorithm to extract key feature points, and continuous images are generated through image stitching technology. The collection of multi-angle images ensures the integrity of the coverage area, and the feature matching and stitching technology can integrate discrete images into continuous images to provide a comprehensive description of the target area.

[0102] In a second aspect, a three-dimensional point cloud data of the target region is generated by a structured light technique according to the continuous image, and the three-dimensional point cloud data is processed to extract the terrain feature;

[0103] Specifically, the spatial position of each pixel point is calculated by using the light intersection principle of multi-view images to generate a dense point cloud. The generation of the dense point cloud can refer to the specific technology of image point cloud, which is not repeated herein. The dense point cloud is filtered to remove noise, and parameters such as terrain elevation, slope and slope direction are extracted from the point cloud data. The average elevation difference and slope direction change rate of each cell, i.e., the terrain feature, are calculated by grid division.

[0104] In a third aspect, a crop recognition network is constructed, the continuous image is taken as an input parameter of the crop recognition network, the input parameter is processed by the crop recognition network, and crop region distribution information is output.

[0105] In this embodiment, the crop recognition network is a semantic segmentation model based on deep learning, the input is a continuous image, and the output is a crop region distribution map. The network uses an improved DeepLabv3+ structure, the backbone network is a lightweight ResNet, and a dilated convolution pyramid pooling module is combined to enhance the multi-scale feature extraction capability. During model training, a labeled data set is used to cover crop images in different growth stages, terrain conditions and light conditions. The network optimizes the ability to distinguish crop regions and backgrounds (such as soil, weeds, etc.) through context feature fusion and spatial attention. Specifically, the model automatically outputs accurate distribution information of the crop region, reducing the error of manual labeling and judgment. Through deep learning, the segmentation capability for complex boundaries and small area regions of the crop region is enhanced.

[0106] Please refer to Figure 3 The crop recognition network structure diagram of the embodiment of the present application.

[0107] In an optional embodiment, the crop recognition network comprises:

[0108] A residual layer is configured to expand the plane dimension of the input parameter by bilinear difference, and a residual network is taken as a backbone network to extract deep features and shallow features, wherein the deep features are used to extract semantic information, and the shallow features are used to restore boundary detail information.

[0109] Specifically, the input image is preprocessed by bilinear interpolation to expand the planar dimension, so that the feature extraction process can better retain the detailed information in the input image. At the same time, a residual network is used as the backbone structure to extract multi-level features. Deep features are extracted through multiple convolutional layers and nonlinear activation functions to extract global semantic information of the image. These features are mainly used to identify the overall distribution pattern and shape features of the crop region. The shallow features retain the boundary and texture information of the image through fewer convolutional operations, which are used to repair the details of the crop region boundary.

[0110] The region recognition layer is configured to fuse the deep features and the shallow features to generate a feature map, process the feature map through a convolution kernel and an attention mechanism to extract region semantic information, calculate label weights according to the region semantic information, classify the label weights through a Softmax classifier, and determine the class information of the region.

[0111] Specifically, the deep features and the shallow features are combined by pixel-by-pixel weighted addition to generate a feature map that contains both global semantic information and boundary details. Multiple convolution kernels are applied to the fused feature map to further extract spatial correlation information of the features. The convolution kernel size uses a combination of 1x1, 3x3 and 5x5 to capture spatial features of different scales. Channel attention mechanism and spatial attention mechanism are used to dynamically adjust the response values of important regions in the feature map, focusing on key features of the crop region and reducing the interference of non-target regions.

[0112] The output layer is configured to segment the crop region according to the classification result of the region recognition layer, perform pixel-level statistics on the crop region, calculate the crop coverage area and distribution characteristics, and generate crop region distribution information.

[0113] Specifically, based on the classification result output by the region recognition layer, the crop region pixels are separated and a binary crop region segmentation map is generated. Based on the segmentation result, the shape features (such as aspect ratio and compactness) and spatial distribution features (such as Voronoi diagram area distribution) of the crop region are calculated. The generation and statistical calculation of the segmentation map visualize the spatial distribution information of the crop region, and the statistical results such as coverage area provide intuitive reference for agricultural management. Through distribution feature extraction, the planting rules and density distribution of crops can be further analyzed.

[0114] In a fourth aspect, a three-dimensional field model is constructed based on the crop region distribution information and the terrain features.

[0115] Specifically, the crop area distribution information is fused with the terrain features, and a complete three-dimensional plot model is generated using a three-dimensional modeling tool. During the fusion process, the crop area is cropped based on the point cloud data, and the point cloud of the non-target area is removed; the boundary area is smoothed by a weighted interpolation method to ensure the continuity and accuracy of the model. The three-dimensional plot model includes the boundary shape of the crop area, the terrain undulation and the elevation difference, forming a basic model for flight path planning and data analysis.

[0116] Next, the part of the method of the present application related to the optimization algorithm is further expanded.

[0117] In one example, the specific steps of S1 are as follows:

[0118] S1.1: According to the three-dimensional plot model and the number of unmanned aerial vehicles to be operated, a plurality of initial operation areas are divided, wherein the plurality of initial operation areas do not overlap with each other;

[0119] Specifically, at the beginning of task scheduling, in order to ensure that multiple unmanned aerial vehicles can cover the target area in the same operation period and the operation paths are independent of each other, the overall area needs to be spatially divided based on the constructed three-dimensional plot model.

[0120] In this embodiment, the initial operation area division is completed by using a regular partition strategy. The area division can be realized based on grid division, area main axis direction division, or clustering segmentation strategy based on operation density. In some optional embodiments, the main axis direction scanning slice method can be used in combination with the main direction of the plot and the crop row-column structure to divide the target area into strip-shaped regions according to the convenient operation direction, and the boundary line is corrected in cooperation with the terrain obstacle buffer to make each initial operation area executable and safe.

[0121] Further, in the division process, considering the unmanned aerial vehicle operation radius, coverage bandwidth and flight safety buffer setting, it is ensured that each sub-region can be independently completed by a certain unmanned aerial vehicle in space, and the edge region has a minimum overlapping boundary. The initial operation area obtained by division meets the spatial geometric requirements of no intersection between each other and complete closure of the coverage range.

[0122] Further, the number of divided areas corresponds to the number of available unmanned aerial vehicles. The division strategy is not a simple equal-area division, but is optimized in combination with the slope information in the three-dimensional model to avoid forcibly dividing a region with too large slope or obvious obstacles into a single region, ensuring the coherence and realizability of subsequent flight path planning. The reasonable division of the initial task space provides a high-resolution and clear boundary input basis for the subsequent task allocation process, avoiding the overlap and intersection of unmanned aerial vehicle operation areas and the generation of flight blind area.

[0123] In some optional embodiments, the number of divided work areas can be less than the number of available UAVs, in which case multiple UAVs are allowed to cooperatively execute the same work area to improve the processing capacity for large areas and high priority areas. When cooperatively executed, the work area is further divided into multiple sub-paths that can be executed in parallel in the path planning stage, and is distributed. In contrast, to reduce the complexity of path switching and scheduling delay, it is not recommended in this application to assign a single UAV to multiple initial work areas at the beginning of task allocation, i.e. the number of initial work areas cannot be greater than the number of available UAVs, to avoid wasting air resources and scheduling confusion due to task switching, and to maintain the continuity and logical consistency of the UAV flight path.

[0124] S1.2: According to the priority of each initial work area, the work amount and the current state of the UAV to be worked, the initial work area is allocated to the corresponding UAV;

[0125] Specifically, after completing the initial area division, path scheduling and resource matching need to be performed according to the complexity of the work task and the carrying capacity of the UAV. The priority of the work area can be determined by the historical UAV work results. The UAV state parameters include real-time remaining power, starting position distance, maximum range, historical work coverage rate and remaining load capacity.

[0126] In this embodiment, a weighted greedy allocation strategy is used for preliminary matching. The non-work flight distance of the UAV from the starting point to the target area is also considered to ensure that the resource scheduling achieves the optimal compromise between coverage efficiency and route cost, so that the initial work area is allocated to the corresponding UAV according to the matching result.

[0127] S1.3: According to the initial work area corresponding to each UAV, a complete flight plan is generated in combination with a preset crop growth state atlas, wherein the crop growth state atlas represents a crop growth activity index, and is constructed according to historical crop data;

[0128] Specifically, after completing task allocation, in order to ensure that each UAV can execute precise work strategies according to the crop state differences in the target area in actual work, path refinement configuration needs to be performed according to the crop growth state atlas. The crop growth state atlas can be constructed from multi-temporal UAV remote sensing data or ground sensor feedback data. The response indicators in the crop growth state atlas can include a normalized vegetation index, a spectral shift threshold and a crop growth speed gradient, and the atlas adopts a three-dimensional heat map form. For each initial work area matched by a UAV, the response data matrix corresponding to the area in the crop growth state atlas is extracted as a response feature reference for the flight task.

[0129] Referring to Figure 4 , Figure 4A flowchart for generating a complete flight plan for an embodiment of the present application.

[0130] In one example, the generation of the complete flight plan in combination with the preset crop growth state map includes:

[0131] S1.3.1: Extracting boundary information and terrain features of the crop area from the initial operation area, and determining the operation range and flight restriction conditions of the unmanned aerial vehicle;

[0132] Specifically, in order to achieve efficient coverage of crops in the target area and path executability assessment, key spatial information related to the task space needs to be extracted before the flight task is generated.

[0133] In the present embodiment, the boundary information includes the outer boundary contour of the initial operation area and the distribution range of the internal crop planting area parsed from the land three-dimensional model. The boundary information can include a set of vertex coordinates, a combination of boundary line segments, and a closed verification logic. The extraction process of the terrain features includes identifying slope information, elevation differences, local mutation areas, and obstacle occupation areas from the point cloud model or structured light reconstruction model. By fusing the above boundary information and terrain features, a grid layer for route constraint and flight passability judgment is formed, and the flyable range, minimum flight height, flight buffer area, and obstacle avoidance strategy applicable area of the unmanned aerial vehicle are determined accordingly.

[0134] It can be understood that the setting of the operation range is not only the area that can fly, but also needs to be constrained by the effective response interval of the crops to avoid arranging redundant paths on the low-response, shielding, or vegetation-free areas at the edge of the farmland. The flight restriction conditions are superimposed on the operation map in the form of a spatial mask to realize real-time discrimination during the path generation process, ensuring that the finally generated path does not touch high obstacles, stay away from slope mutation points, and avoid edge flight error areas.

[0135] S1.3.2: Dividing the crop area into regular grids, wherein the size of each grid meets the effective coverage area of one shot of the unmanned aerial vehicle, and labeling the crop response weight associated with each grid through the crop growth state map, wherein the crop response weight is determined according to the growth state of the crop area in the crop growth state map;

[0136] Specifically, in order to realize the fine response of the unmanned aerial vehicle to the growth of crops during navigation and dynamic task weight management, the crop area needs to be divided into regular grids, and each grid is given a quantifiable growth response index.

[0137] In this embodiment, after the crop area is divided into regular grids, in order to realize the fine level of flight task value calibration, it is necessary to establish an independent crop response weight for each grid. The crop response weight is not a direct pixel value or physical measurement value in the atlas, but a task priority index calculated by combining the spatial state, growth trend, abnormal factor and regional contrast.

[0138] Further, the response data matrix corresponding to each grid unit of the crop growth state atlas is first extracted. The response data matrix contains multiple spectral channel information at the current time phase, and part of the optional environmental factor layers, including but not limited to temperature and humidity, vegetation height model and historical drought record. The response data matrix is mapped in the form of a two-dimensional array inside the grid area.

[0139] Further, the calculation of the crop response weight is not a simple average or maximum value processing, but is calculated by constructing a regional response function. The regional response function considers the following four types of indicators as input variables:

[0140] The first type is the local response intensity index, that is, the mean, peak and standard deviation of the spectral image in the grid, which reflects the fluctuation amplitude and density of the vegetation state;

[0141] The second type is the change trend factor, which is calculated based on historical multi-time phase atlas to identify the instability or potential risk of the growth state. The response slope, inflection point and fluctuation frequency of the region in the past several periods are used to identify the instability or potential risk of the growth state;

[0142] The third type is the boundary neighborhood feature, which introduces the response value gradient, boundary contrast and directionality entropy between the grid and its adjacent grid to identify the abnormal transition area or disease spread edge that may exist;

[0143] The fourth type is the regional importance factor, such as whether the grid is close to a known traffic artery, water channel, monitoring node or edge high-contrast area, which is an external marker for management attention in actual deployment;

[0144] Further, the calculation of the crop response weight is weighted according to the aforementioned four types of indicators. Each type of indicator can select a different weighting scheme according to the type of work. The specific weighting scheme and weighting weight can be determined through a large number of experiments, which will not be described herein.

[0145] In some optional embodiments, the aforementioned four types of indicators can also be normalized by gradient-based regional scoring logic to enhance the recognition ability of local fluctuations or abnormal hotspots through a nonlinear function, thereby forming a comprehensive evaluation of the degree of priority of the grid for work. The final generated crop response weight is a quantitative index between a preset interval.

[0146] S1.3.3: Extract a vertex set of the rule grid boundary, calculate a minimum circumscribed rectangle according to the vertex set, and take the long side direction of the rectangle as the operation direction of the unmanned aerial vehicle;

[0147] Specifically, in order to construct the operation flight direction closely fitted with the shape of the land plot, it is necessary to extract effective geometric features from the boundary of the rule grid to determine the flight path direction conforming to the shape optimal fitting principle. The traditional path planning often directly takes the land plot coordinate system direction as the default operation direction, but this method often causes problems such as increase in path number, insufficient edge cutting, and the like when the shape of the land plot is diagonal, strip-shaped, narrow, or irregular contour, thereby affecting the operation efficiency. The present application constructs a minimum circumscribed rectangle to obtain the main direction features of the rule grid boundary in two-dimensional space. The construction of the minimum circumscribed rectangle is based on a boundary point set, and a rotation puzzle method or a direction projection extreme scanning method is used to traverse the boundary projection point set at multiple rotation angles, and finally the minimum envelope rectangle and its direction angle are obtained.

[0148] In the present embodiment, the long side direction of the minimum circumscribed rectangle is taken as the operation direction, which can maximize the extension of the flight path along the crop row direction without changing the actual structure of the land plot, thereby reducing the number of turns and shortening the flight path length of the non-operation section. In addition, the operation direction has strong spatial stability, avoiding direction deviation caused by map coordinate rotation or multi-time phase map resampling, and ensuring the consistency and generalization ability of path planning on different land plot scales.

[0149] Further, after the operation direction is determined, a set of direction reference base lines can be generated as the input basis of subsequent path parallel generation, edge alignment control, operation band overlap rate adjustment, and the like algorithm modules.

[0150] S1.3.4: Generate parallel paths along the operation direction, and mark the start point and end point coordinates of each path, wherein:

[0151] The start point of the path is determined according to the crop response weight of the rule grid;

[0152] The end point of the path is determined according to the remaining power of the unmanned aerial vehicle, the wind direction, and the return route;

[0153] The start points of each parallel path are different from each other;

[0154] Specifically, after obtaining the task direction, the path generation part needs to develop based on the determined direction line to obtain a path set. To improve the uniformity of path distribution and task efficiency, the application adopts the strategy of generating parallel paths at equal intervals along the task direction. The spacing between each path is determined by the UAV task width, image overlap requirement and path coverage overlap control strategy. The path itself is a continuous line segment, and the starting and ending points need to be dynamically set according to multiple factors, without using the uniform direction or fixed point take-off mode.

[0155] In the selection of path starting point, the application performs crop response aggregation analysis on the grid covered by the parallel path based on the gradient sorting logic of crop response weight. If the starting end covered by a path has higher crop response, the starting point of the path is set to the end, so that the UAV preferentially performs the task in the area with higher response intensity, forming a task flow sequence from high priority area to low priority area. It can effectively control the task hot area priority response problem and reduce the response area omission or processing delay caused by sudden crop stress reaction.

[0156] The determination of the path end point considers the current power of the UAV, the total length of the path, the influence of wind speed and direction on the task speed, the energy consumption difference caused by the change of climbing height during flight, and the estimated energy consumption of the flight return path. Specifically, it includes flight energy consumption budget, energy consumption simulation and prediction for each path. If the prediction result shows that the path cannot guarantee the safe return of the aircraft to the starting point or the task gathering point after execution, the end point is set in advance, and the path breakpoint is labeled for subsequent task inheritance scheduling. The wind direction factor has a significant impact on the end of the path. If the end of the flight path is opposite to the wind direction, the end point distance needs to be adjusted accordingly to adapt to the flight control safety strategy.

[0157] The flight control safety strategy can be understood as a set of decision-making and control criteria for ensuring flight safety and task recoverability under the constraints of UAV flight path planning and energy consumption. Its core goal is to ensure that the UAV has controllable, returnable, interruptible and recoverable capabilities when executing path tasks in complex environments such as wind field interference, energy attenuation and weather changes.

[0158] Further, in order to prevent the problem of air resource congestion caused by path overlap or starting point overlap of multiple UAVs at the initial take-off stage, the starting points of each path in the application are different, and the execution time axis of adjacent paths is separated by using staggered path start sequence or staggered activation method. It is beneficial to construct a path rhythm staggered structure in the high-density task area, and improve the flight path space distribution quality and execution fault tolerance ability from the scheduling dimension.

[0159] S1.3.5: obstacle detection is performed on the generated path, if there is an obstacle grid in the path, the path is re-planned around the obstacle grid through the A* algorithm, the start and end points of the path after bypassing the obstacle are re-connected, if there is no obstacle grid in the path, the path is added to the path set, and the path is repeatedly traversed until all paths are added to the path set;

[0160] Specifically, the generated flight path is a theoretical execution route, which must pass environmental constraint verification before being officially issued to the unmanned aerial vehicle for execution, mainly including static obstacle detection and flight safety boundary review. The application extracts the obstacle region from the elevation point cloud and structure texture information in the three-dimensional model of the land plot to form an obstacle grid mask, performs spatial overlap calculation on the path grid set, and identifies whether there is a section crossing the obstacle section in the path section. If the identification result is no, that is, the whole path is in a passable area, the path is directly added to the path set; if the identification result is yes, the obstacle grid needs to be repaired.

[0161] In this embodiment, the classic heuristic path planning algorithm A* is used as the core algorithm skeleton, the start point and the end point are defined as the source and target nodes of the heuristic search, the path direction is used as the constraint to build the path search space, the search process is allowed to bypass the obstacle grid, and the parameters of flight path smoothness, turning cost and flight stability are considered in the search process.

[0162] S1.3.6: a multi-objective evaluation function is established, wherein the objectives include path length, turning number, coverage repetition rate, energy consumption and crop response weight, the evaluation score of each path in the path set is calculated according to the multi-objective evaluation function, the path corresponding to the highest evaluation score is taken as the optimal path scheme, and the unmanned aerial vehicle flight route is generated according to the path node, turning point and obstacle avoidance parameter of the optimal path scheme;

[0163] Specifically, in order to select the path execution scheme with the optimal comprehensive value under the current operation condition, each path in the path set needs to be quantitatively evaluated. The application considers five key influencing factors through a multi-objective path evaluation function:

[0164] First, the path length, as an important indicator of flight time and operation cost, the shorter the path, the better;

[0165] Second, the turning number, the more turns mean the increase of flight control complexity and energy consumption, and a penalty factor is given to the continuous winding path during evaluation;

[0166] Third, the coverage repetition rate, by analyzing the overlap ratio between the path coverage area and the historical operation record, the waste of resources in the repeated area is avoided;

[0167] Fourth, the energy consumption budget, the total energy consumption of the path is simulated combined with the path length, height change and unmanned aerial vehicle parameters;

[0168] The fifth is the crop response weight, i.e., the average crop value density of the path coverage area, which reflects the operation priority of the path.

[0169] In the embodiment, the evaluation function is not in the equal weight weighting manner, but constructs a dynamic weight based on the task type. If the task target is rapid intervention of pests and diseases, the crop response weight factor weight is increased. If the task target is full coverage detection, the coverage integrity and path smoothness factor weights are increased. After each path is calculated by the evaluation function, a total score value is output, and the path with the highest score is taken as the priority execution path. The path node, turning point information and obstacle avoidance strategy parameters corresponding to the path are extracted together to form the optimal flight control instruction set of the unmanned aerial vehicle in the current operation period.

[0170] It can be understood that if there are multiple unmanned aerial vehicles to be operated in the initial operation area, the method further comprises:

[0171] Obtaining the path set and the evaluation scores corresponding to the paths in the path set, and prioritizing the paths;

[0172] According to the current positions of the multiple unmanned aerial vehicles to be operated, and according to the priority descending order of the paths, a unmanned aerial vehicle that meets the flight requirements and has the smallest distance to the starting point of the current path is selected as the execution unmanned aerial vehicle of the current path in sequence until there is no unmanned aerial vehicle to be operated.

[0173] Next, the part of the method of the application about the adjustment stage is further expanded.

[0174] In one example, the growth state information and the unmanned aerial vehicle state information in S2 are obtained from the real-time report of the unmanned aerial vehicle.

[0175] It can be understood that the growth state information is collected by a remote sensing device carried by the unmanned aerial vehicle. The remote sensing device can include a multispectral camera, a thermal infrared imager, a high-resolution visible light camera, a laser ranging device, or a narrowband filter assembly for vegetation reflection monitoring. Different devices are selected according to the operation target. In the embodiment, the normalized vegetation index, the red edge index and the blue-green waveband reflectivity are preferentially selected as the basic monitoring indicators of the growth state information, which are used to judge the growth intensity, density change and stress response of the crop area.

[0176] In some optional embodiments, vegetation elevation data and temperature layers can also be combined for joint analysis to determine whether there is a wilting, water loss or lesion spread trend in the local area. The foregoing information will be structured into growth state data blocks after collection, each block corresponding to a certain area below or around the current unmanned aerial vehicle, and constantly dynamically updated along the flight trajectory to generate an approximately continuous crop growth state.

[0177] In one example, the condition of entering the adjustment stage comprises at least one of the following:

[0178] The variation range of the crop growth state index of any sub-region in the crop region is greater than or equal to a preset generation threshold;

[0179] An abnormal area with spatial aggregation characteristics of pests and diseases, high temperature, or low humidity is detected in any sub-region in the crop region, wherein the high temperature and the low humidity can be determined by a preset high temperature threshold and a low humidity threshold, and if the temperature corresponding to the sub-region is greater than or equal to the high temperature threshold, the sub-region is a high temperature abnormal area; if the humidity corresponding to the sub-region is less than or equal to the low humidity threshold, the sub-region is a low humidity abnormal area.

[0180] The generation threshold, the high temperature threshold, and the low humidity threshold can be determined by a person skilled in the art through a large number of experiments.

[0181] The power of the unmanned aerial vehicle is less than or equal to a preset flight safety threshold during the execution of the task;

[0182] The unmanned aerial vehicle has a yaw, a task interruption, or a path execution timeout during the execution of the task;

[0183] The flight paths of the plurality of unmanned aerial vehicles exist spatial intersection, and the safety interval is less than a preset tolerance threshold.

[0184] It can be understood that the aforementioned thresholds are derived from experimental calibration, which is not described herein.

[0185] In one example, the specific steps of S3 are as follows:

[0186] S3.1: Construct a multi-dimensional dynamic task evaluation model of the region to be adjusted, wherein the multi-dimensional dynamic task evaluation model comprises a crop growth response dimension, a task urgency degree dimension, a historical task effect dimension, and a task migration cost dimension;

[0187] Specifically, during the execution of the unmanned aerial vehicle task, some regions may become regions to be adjusted due to crop state mutation, environmental interference, or unmanned aerial vehicle resource changes. In order to determine whether these regions need to be rescheduled and which unmanned aerial vehicle should take over, a multi-dimensional dynamic task evaluation model needs to be established to quantify the scheduling value of each region to be adjusted.

[0188] In the embodiment, the multi-dimensional dynamic task evaluation model is composed of four main dimensions. Among them, the crop growth response dimension is based on the NDVI, red edge index, spectral abnormal gradient or drought stress index of the current regional crop, measures the current agronomic importance of the region, and is the direct basis for determining whether the work must be completed; the work urgency dimension is evaluated by combining the crop cycle, the number of delayed work and the response index change slope, reflecting the sensitivity of the region to time, which belongs to the scheduling priority dimension; the historical work effect dimension is calculated by the coverage of the current work, the density of unmanned aerial vehicle work trajectory and the completeness of image acquisition. If there is a record of insufficient coverage, trajectory deviation or image acquisition failure in the historical work, the region needs to be further subdivided into sub-tasks in the redistribution; the task migration cost dimension is used to measure whether it is reasonable to transfer the current task to other unmanned aerial vehicles, mainly referring to the switching scheduling distance between unmanned aerial vehicles, the path reconstruction complexity, and the transfer cost between the work interruption point and the adjacent unmanned aerial vehicle trajectory.

[0189] S3.2: calculating a task migration evaluation score of the region to be adjusted according to the multi-dimensional dynamic task evaluation model, the task migration evaluation score being used to determine the task migration priority of the region to be adjusted;

[0190] The task migration evaluation score is calculated by weighting the four dimensions. The essence of the task migration evaluation score is to convert the comprehensive importance of whether each region should be migrated into an ordered numerical result, thereby supporting the downstream links of task sorting, resource allocation and path control.

[0191] S3.3: generating a candidate task set of unmanned aerial vehicles according to the task migration evaluation score and the spatial distribution characteristics of the region to be adjusted;

[0192] Specifically, after the task migration evaluation score is calculated, the high-score regions need to be clustered and analyzed and the task structure needs to be generated in combination with the relative position relationship, shape boundary and connectivity of the regions in the geographical space. Directly assigning high-score regions to unmanned aerial vehicles may cause path fragmentation, flight route jumping and sharp rise in energy consumption, so a reasonable candidate task set needs to be constructed to merge regions that are logically adjacent, have similar work characteristics and have similar scheduling costs into a scheduling unit for matching unmanned aerial vehicle resources.

[0193] In the embodiment, the candidate task set generation takes the task migration evaluation score as the main sorting index, and is supplemented by spatial clustering for region grouping. Spatial clustering can select common clustering algorithms, which are not described herein.

[0194] S3.4: obtaining a UAV to execute the UAV candidate task set based on the residual power, flight distance, current load, historical operation efficiency and location state of the residual UAVs by a particle swarm optimization algorithm, and assigning a corresponding task in the UAV candidate task set to the corresponding UAV;

[0195] Specifically, after obtaining the UAV candidate task set, it is necessary to perform optimal matching between the UAVs and the UAV candidate task set in combination with the current schedulable UAV resource situation. Due to the multiple matching target dimensions and complex resource constraints, it is difficult for conventional greedy matching, local heuristic allocation and other methods to simultaneously meet the triple requirements of matching efficiency, resource balance and path executability. The embodiment adopts a method based on a particle swarm optimization algorithm, takes the UAVs and the UAV candidate task set as a particle matching unit, and optimizes the scheduling score objective function through a population parallel search strategy.

[0196] It can be understood that each particle represents an allocation scheme, the dimension of the particle is composed of the number of candidate task sets, and each dimension takes the UAV number allocated to the task set as the value. The particle fitness function includes multiple scheduling indicators:

[0197] The first is the matching degree between the residual power of the UAV and the predicted operation energy consumption;

[0198] The second is the flight distance cost between the current location of the UAV and the task set centroid;

[0199] The third is the load ratio of the current UAV load and the task workload;

[0200] The fourth is the historical task execution efficiency (such as unit time coverage rate, path deviation rate);

[0201] The fifth is the path connection continuity and return trip cost estimation.

[0202] By setting these fitness function components, the algorithm can continuously optimize the UAV task allocation combination in the iteration process and find the optimal matching structure that takes into account efficiency and stability.

[0203] Further, after completing the allocation, the optimal matching structure output by the particle swarm algorithm is subjected to feasibility review. If there are problems such as overloading of the UAV, overlapping between paths, or unconnectable flight paths in a certain allocation scheme, the scheme is returned and the search direction is re-directed. If there are multiple near-optimal solutions, a multi-branch decision model can be enabled to generate a backup scheduling scheme, facilitating quick switching in case of task interruption or resource failure.

[0204] In an optional embodiment, the obtaining a UAV to execute the UAV candidate task set includes:

[0205] According to a preset resource fitness evaluation function in the particle swarm optimization algorithm, resource fitness of each UAV executing the UAV candidate task set is calculated.

[0206] The resource fitness evaluation function can be determined by weighting and synthesizing energy consumption, task benefit, time constraint, load state and environmental factors when the UAV executes the task.

[0207] Specifically, the resource fitness evaluation function can normalize indexes such as task completion rate, path energy consumption, flight time, load utilization rate and wind speed and direction, and perform linear or nonlinear weighted calculation according to a preset weight, so as to obtain a resource fitness value reflecting the overall matching degree of the UAV executing the task.

[0208] According to the resource fitness, an initial task allocation scheme of the UAV is determined.

[0209] According to a conflict situation in the initial task allocation scheme of the UAV, the initial task allocation scheme of the UAV is secondarily allocated through autonomous negotiation logic between the UAVs, wherein the conflict situation includes path intersection, time-space overlap and task load imbalance of the UAVs.

[0210] In one example, the autonomous negotiation logic includes conflict resolution between the UAVs according to a game theory strategy, wherein:

[0211] Each UAV broadcasts negotiation information to other UAVs according to its own state, and updates the order of the UAV executing the task through negotiation information interaction results until a balance state of minimizing path conflict and maximizing overall benefit of the task between the UAVs is reached.

[0212] In the embodiment, the core idea of the autonomous negotiation logic is to introduce a Nash equilibrium game model in the game theory as a task conflict resolution framework. Specifically, it is assumed that each UAV is an independent participant, the strategy space is a set of selectable task paths, the utility function is the task execution benefit minus the path conflict penalty factor and the energy consumption prediction cost. The UAVs push the current path plan, state parameters and priority execution right information to the cluster network through the broadcast mechanism, and receive similar information from other UAVs. After each round of information interaction, each UAV calculates the conflict cost of the selected task path in the overall cluster according to the received path intersection point, time overlap window and spatial adjacency relationship, and adjusts the task selection based on the local optimal response function to minimize the total conflict and improve the utility.

[0213] It can be understood that the negotiation process can be regarded as a discrete-time game process, and each unmanned aerial vehicle approaches the equilibrium state under multi-party game according to its own strategy and collective feedback in a limited strategy set. In order to prevent path shock or negotiation failure, a tolerance range is set for each iteration, and when the change range of task selection is lower than the preset disturbance threshold for two consecutive rounds, it is determined that the negotiation is completed. If it does not converge within the specified number of steps, a forced conflict resolution mechanism (such as time window mispositioning execution, Z-axis height mislayering, etc.) is triggered to schedule the bottom.

[0214] Further, after the negotiation is completed, the output result is a set of unmanned aerial vehicle and task path mapping sets, wherein each path is the minimum conflict path collectively accepted by the cluster after negotiation, and the space-time staggering, resource balancing and priority difference have been considered. Compared with the centralized matching strategy based on particle swarm alone, the risk of path overlap, task crosstalk and operation failure caused by multiple unmanned aerial vehicles entering adjacent high-response areas at the same time can be reduced, and the system has the ability of collaborative self-adaptation under complex conditions such as stable local communication link and frequent task graph update.

[0215] In one example, the present application provides a task allocation and conflict resolution system for multi-unmanned aerial vehicle cooperative operation, which comprises:

[0216] A modeling module is configured to obtain a three-dimensional land model of a target area, and construct a work area according to the terrain features, obstacle information and crop growth state graph;

[0217] A path planning module is configured to generate initial flight routes for multiple unmanned aerial vehicles based on the work area model;

[0218] A task allocation module is configured to perform dynamic task allocation for the flight path based on the crop response weight, task priority and unmanned aerial vehicle state parameters;

[0219] A conflict resolution module is configured to perform path adjustment, time window mispositioning or flight height reconstruction when there is path conflict, and output updated flight control instructions.

[0220] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for task allocation and conflict resolution in multi-UAV cooperative operation, characterized in that, The method comprises: acquiring a three-dimensional field model of a target area, and planning a UAV flight route by an optimization algorithm according to terrain features and distribution information of the three-dimensional field model; allocating the UAV flight route to a corresponding UAV, monitoring crop growth state information and UAV state information in a UAV operation process, and determining whether a condition for entering an adjustment phase is met; if the condition is met, acquiring a region to be adjusted from the UAV flight route, and re-allocating a UAV performing a corresponding task of the region to be adjusted according to a crop growth state atlas of the region to be adjusted.

2. The method of claim 1, wherein, The UAV flight route is planned by an optimization algorithm, which comprises: dividing a plurality of initial operation regions according to the three-dimensional field model and the number of UAVs to be operated, wherein the plurality of initial operation regions do not overlap with each other; allocating the initial operation regions to corresponding UAVs in combination with the priority of each initial operation region, the operation amount, and the current state of the UAVs to be operated; generating a complete flight plan according to the initial operation region corresponding to each UAV in combination with a preset crop growth state atlas, wherein the crop growth state atlas represents a crop growth activity index, and is constructed according to historical crop data.

3. The method of claim 2, wherein, The complete flight plan is generated in combination with the preset crop growth state atlas, which comprises: extracting boundary information and terrain features of a crop region from the initial operation region to determine the operation range and flight limitation condition of the UAV; dividing the crop region into regular grids, wherein the size of each grid meets the effective coverage area of one UAV shot, and the crop response weight associated with each grid is labeled by the crop growth state atlas, and the crop response weight is determined according to the growth state of the crop region in the crop growth state atlas; extracting a vertex set of the boundary of the regular grid, calculating a minimum circumscribed rectangle according to the vertex set, and taking the long side direction of the rectangle as the operation direction of the UAV; generating parallel paths along the operation direction, and marking the start point and end point coordinates of each path; detecting obstacles for the generated paths, if there are obstacle grids in the paths, re-planning the paths in the region corresponding to the obstacle grids by A* algorithm, re-connecting the start point and end point of the paths after bypassing the obstacles, if there are no obstacle grids in the paths, adding the paths to a path set, and repeating the iteration of the paths until all the paths are added to the path set; establishing a multi-objective evaluation function, wherein the objectives include path length, turning frequency, coverage repetition rate, energy consumption, and crop response weight, calculating the evaluation score of each path in the path set according to the multi-objective evaluation function, taking the path corresponding to the highest evaluation score as the optimal path scheme, and generating a UAV flight route according to the path nodes, turning points, and obstacle avoidance parameters of the optimal path scheme.

4. The method of claim 3, wherein, Generating parallel paths along the operation direction, and marking the start point and end point coordinates of each path, comprises: The start point of the path is determined according to the crop response weight of the regular grid; The end point of the path is determined according to the remaining power of the UAV, the wind direction, and the return route; The start points of each parallel path are different from each other.

5. The method of claim 3, wherein, If multiple unmanned aerial vehicles (UAVs) to be operated exist in the initial operation area, the method further comprises: obtaining the path set and evaluation scores corresponding to paths in the path set, and prioritizing the paths; according to the current positions of the multiple UAVs to be operated, and in descending order of the priorities of the paths, selecting, in sequence, a UAV that meets flight requirements and has the minimum distance to a starting point of a current path as an execution UAV of the current path until no UAV to be operated exists.

6. The method of claim 1, wherein, The conditions for entering the adjustment stage include at least one of the following: a variation amplitude of a crop growth state indicator of any sub-area in the crop area is greater than or equal to a preset generation threshold; a disease, pest, high-temperature, or low-humidity abnormal area with a spatial aggregation feature is detected in any sub-area in the crop area, wherein the high-temperature and low-humidity can be determined by preset high-temperature and low-humidity thresholds, and if a temperature corresponding to a sub-area is greater than or equal to the high-temperature threshold, the sub-area is a high-temperature abnormal area; if a humidity corresponding to a sub-area is less than or equal to the low-humidity threshold, the sub-area is a low-humidity abnormal area; a power of the UAV is less than or equal to a preset flight safety threshold during execution of the operation; a yaw, task interruption, or path execution timeout occurs during execution of the operation by the UAV; flight paths of multiple UAVs exist spatially intersected, and a safety interval is less than a preset tolerance threshold.

7. The method of claim 1, wherein, redistributing the UAVs that execute tasks corresponding to the to-be-adjusted area according to a crop growth state atlas of the to-be-adjusted area, comprising: constructing a multi-dimensional dynamic task evaluation model of the to-be-adjusted area, wherein the multi-dimensional dynamic task evaluation model comprises a crop growth response dimension, an operation urgency degree dimension, a historical operation effect dimension, and a task migration cost dimension; calculating a task migration evaluation score of the to-be-adjusted area according to the multi-dimensional dynamic task evaluation model, wherein the task migration evaluation score is used to determine a task migration priority of the to-be-adjusted area; generating a UAV candidate task set according to the task migration evaluation score and a spatial distribution feature of the to-be-adjusted area; obtaining the UAVs that execute the UAV candidate task set by a particle swarm optimization algorithm based on remaining power, flight distance, a current load, historical operation efficiency, and a position state of the remaining UAVs, and assigning corresponding tasks in the UAV candidate task set to corresponding UAVs.

8. The method of claim 7, wherein, The obtaining the UAVs that execute the UAV candidate task set comprises: calculating resource fitness of each UAV executing the UAV candidate task set according to a preset resource fitness evaluation function in the particle swarm optimization algorithm; determining a UAV initial task allocation scheme according to the resource fitness; performing secondary allocation on the UAV initial task allocation scheme by autonomous negotiation logic between the UAVs according to a conflict situation in the UAV initial task allocation scheme, wherein the conflict situation includes path intersection, time-space overlap, and task load imbalance between the UAVs.

9. The method of claim 8, wherein, The autonomous negotiation logic includes conflict resolution between the UAVs according to a game theory strategy, wherein: Each unmanned aerial vehicle broadcasts negotiation information to other unmanned aerial vehicles according to its own state, and the order of the unmanned aerial vehicles in executing tasks is updated through negotiation information interaction results until a balance state of minimizing path conflicts among the unmanned aerial vehicles and maximizing overall task benefits is reached.

10. A multi-UAV cooperative task allocation and conflict resolution system for implementing the multi-UAV cooperative task allocation and conflict resolution method according to any one of claims 1-9, characterized in that, The system comprises: a modeling module configured to acquire a three-dimensional land model of a target area, and construct a work area based on terrain features, obstacle information and a crop growth state atlas; a path planning module configured to generate initial flight routes of multiple unmanned aerial vehicles based on the work area model; a task allocation module configured to perform dynamic task allocation for the flight routes based on a crop response weight, a task priority and an unmanned aerial vehicle state parameter; a conflict resolution module configured to perform path adjustment, time window misplacement or flight height reconstruction when there is a path conflict, and output updated flight control instructions.

Images