A multi-unmanned aerial vehicle path planning method fusing greedy strategy and improved ant colony algorithm

Abstract

The application discloses a kind of multi-unmanned aerial vehicle path planning methods of fusing greedy strategy and improved ant colony algorithm, comprising: first, the detection value of each to be measured node is calculated based on remote sensing image and environmental sensor data;Subsequently, using greedy strategy, according to preset benefit function, assign detection node sequence for multiple unmanned aerial vehicles to generate initial path;Then, according to node detection value, initialize path pheromone, and in improved ant colony algorithm, adopt heuristic function of fusion benefit item and uncovered area guide factor to search path, simultaneously, pheromone is dynamically updated by coverage-time double adaptive mechanism;Finally, according to task execution result, dynamically update node detection value, realize multi-batch closed loop planning.The application effectively improves the path planning efficiency of unmanned aerial vehicle vegetation anomaly detection in complex environment, global coverage and anomaly detection rate, while significantly reducing flight energy consumption and invalid coverage.

Description

Technical Field

[0001] This invention belongs to the field of UAV inspection and path planning technology, and particularly relates to a multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm. Background Technology

[0002] In agricultural production, forestry protection, and ecological environment monitoring, vegetation health status detection is a crucial link in ensuring crop yield and ecological balance. Factors such as pests and diseases, nutrient imbalances, or water stress can cause abnormal symptoms in vegetation, such as yellowing, wilting, and spotting. Failure to detect and address these issues promptly can lead to economic losses and ecological degradation. Traditional manual inspection methods suffer from low detection efficiency, high labor intensity, and poor safety in complex terrains such as large areas of forest, hills, and terraced fields, making them unsuitable for meeting the needs of rapid and accurate vegetation anomaly monitoring.

[0003] In recent years, drones have been widely used for vegetation health monitoring due to their advantages such as maneuverability and clear imaging. By equipping them with RGB cameras, multispectral or thermal imaging sensors, and combining them with image analysis algorithms, large-scale identification and location of vegetation diseases can be achieved. However, many challenges remain when performing tasks in complex environments: First, drones need to ensure comprehensive coverage of target areas in complex terrain and obstacle environments. Second, drones have limited battery life, requiring path planning to consider whether they can return within a specified time and explore high-value areas within that limited time. Finally, the actual flight environment is dynamic; factors such as changes in lighting and terrain can affect drone flight safety and detection accuracy.

[0004] In the field of UAV path planning, commonly used planning methods mainly include the following categories: (1) Rasterization or partitioned coverage method. This type of method divides the target area into several regular grids or sub-regions and uses fixed scanning patterns such as parallel lines or spirals to achieve full coverage. This method is simple to implement, but it lacks consideration for terrain and obstacle distribution, resulting in redundant flight paths, low efficiency, and the inability to dynamically adjust the path.

[0005] (2) Classical path planning algorithms, such as Dijkstra's algorithm. Graph theory-based methods such as Floyd-Warshall can generate shortest paths and are suitable for finding optimal single paths in static environments. However, these methods are computationally intensive, poorly adaptable to dynamically changing environments, and prone to issues such as repeated visits or missed regions in coverage tasks.

[0006] (3) Intelligent optimization algorithms. These include genetic algorithms (GA), particle swarm optimization (PSO), and ant colony optimization (ACO). These algorithms achieve path optimization through group cooperation or by simulating natural behavior. Among them, ant colony optimization has high applicability in the field of path planning due to its distributed computing, adaptive optimization, and multi-objective optimization capabilities. However, existing ant colony optimization-based schemes mostly focus on the shortest path or single-object detection, and lack a comprehensive balance between coverage, detection value, and flight energy consumption. In complex environments, it is difficult to balance efficiency and detection quality. This makes it difficult to fully leverage the advantages of UAVs in achieving accurate detection in various fields.

[0007] (4) Path planning method based on detection priority. Multispectral imagery and vegetation indices (such as NDVI and SAVI) are used to analyze diseased areas, and UAVs are prioritized to cover high-risk areas. This type of method often separates detection and path planning, which is not strongly coupled and lacks support for multi-UAV collaboration, obstacle avoidance, and dynamic route adjustment.

[0008] In summary, existing technologies for coverage path planning in drone-based vegetation anomaly detection in complex environments have the following problems: 1. The detection value model is crude and weakly coupled. Existing coverage path planning mainly focuses on geometric full coverage, separating detection from path planning, which lacks strong coupling. At the same time, it lacks multi-dimensional quantitative indicators, fails to convert real-time environmental factors into the probability of successful scanning, and does not consider scenarios where the target is abnormal but may fail to be detected, thus leading to increased invalid coverage and wasted energy.

[0009] 2. The multi-drone task allocation is not precise enough, and the quality of the initial trajectory is low. Common initial path generation methods ignore the actual costs of drone flight, such as turning time, detection dwell time, and climb cost. They also lack explicit constraints on the uniform take-off and landing points and return-to-home safety margins, which can easily lead to unbalanced tasks, over-limit endurance, and return-to-home risks.

[0010] 3. In the process of using ant colony algorithm for path planning, most methods will assign the same pheromone size to all navigation points on the path at the initial moment. This does not fit the complex and diverse real-world scenarios well, resulting in a lack of directionality in the early stages of the search, making it easy to get trapped in local optima or produce a long cold start delay.

[0011] 4. In the path planning process using the ant colony algorithm, the problem of avoiding getting trapped in local optima was not considered. This problem has two aspects. First, there is the issue of pheromone evaporation mechanism. Traditional pheromone evaporation rates are usually fixed or only depend on time, without adaptive adjustment based on coverage progress. This can easily lead to the model converging quickly in the early stages, failing to reach the global optimum. Second, existing heuristic functions mostly focus on single indicators such as distance or time, without simultaneously incorporating multiple factors such as flight time, detection value, terrain risk, and uncovered areas. This can easily lead to duplicate coverage, making it difficult to guarantee high coverage rates, thus getting trapped in local optima. Summary of the Invention

[0012] To address the aforementioned technical problems, this invention proposes a multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm, thereby resolving the issues present in the existing technologies.

[0013] To achieve the above objectives, this invention provides a multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm, comprising: Based on remote sensing image data and environmental sensor data of the area to be tested, the detection value of each node to be tested is calculated. Based on the detection value of each node to be tested and the preset revenue function, a greedy strategy is used to assign a sequence of nodes to be tested to multiple drones and generate an initial path. The path pheromone is initialized based on the detection value of each node to be tested; A heuristic function is constructed based on the profit function and the uncovered area guiding factor. Based on the initial path, the heuristic function is used to search the path during the ant colony algorithm iteration process. After each iteration, the path pheromone is updated using a coverage-time dual adaptive pheromone evaporation mechanism. After completing path planning and executing detection tasks based on iteratively updated path pheromones, the detection value of each node is dynamically updated according to the execution results of the detection tasks. Execute subsequent tasks based on the updated node detection value to achieve closed-loop path planning for multiple batches of tasks.

[0014] Optionally, the process of calculating the detection value of each node to be tested includes: calculating the visible light vegetation index of each node based on the orthophoto RGB image; obtaining scene weights based on the statistical characteristics of the vegetation index of each node and the visible light vegetation index; calculating the success probability of scanning based on the light intensity and relative height information of the node; and fusing the scene weights and the success probability of scanning to obtain the node detection value.

[0015] Optional, the preset payoff function is: ; in, For time consumption items, ; The typical linear velocity of a drone's flight. The typical angular velocity for drone flight, Indicates the straight-line flight time of the drone. This indicates the time taken for the drone to turn. Fixed time consumption for node detection The detection value of the point j to be tested. The climb penalty term indicates whether a flight from node k to node j requires climbing to a safe altitude. , Let j be the safe height of the point to be measured. This represents the actual height of the current point.

[0016] Optionally, the process of obtaining the initial path includes: Initialize all drone task queues to empty, cumulative revenue to zero, and the set of assigned nodes to empty; In each round of allocation, node selection begins with the drone with the lowest current cumulative profit. Candidate nodes that meet the time constraints are selected from the unallocated test nodes. The profit function value of the drone moving from its current position to each candidate node is calculated. The candidate node with the highest profit function value is selected as the next target node. The target node is added to the task queue of the UAV, and the cumulative revenue, cumulative flight time and the set of assigned nodes of the UAV are updated until the termination condition is met, and the next UAV is assigned. When all UAVs have completed task assignment or triggered the return condition, the initial path of each UAV is obtained.

[0017] Optionally, the time constraint is that the sum of the current cumulative flight time, the time to move to the candidate node, and the estimated time to return from the candidate node to the starting point does not exceed the maximum safe flight time of the UAV.

[0018] Optionally, initializing the pheromone on the path includes: calculating the initial pheromone value of the path based on the detection values ​​of the nodes at both ends of the path; and normalizing the initial pheromone value of all paths.

[0019] Optionally, constructing a heuristic function based on the revenue function and the uncovered area guiding factor includes: normalizing the revenue function value to obtain a standardized revenue term; setting an uncovered area guiding factor based on whether the node has been visited; and multiplying the standardized revenue term by the uncovered area guiding factor to obtain the heuristic function value.

[0020] Optionally, the coverage-time dual adaptive pheromone evaporation mechanism includes: calculating a time-adaptive baseline evaporation coefficient based on the iteration rounds, wherein the baseline evaporation coefficient decays exponentially with the number of iterations; obtaining the coverage progress based on the ratio of the number of times each path has been visited in historical iterations to the theoretical maximum number of visits; calculating the actual evaporation rate of each path based on the baseline evaporation coefficient and the coverage progress; and updating the pheromone of each path based on the actual evaporation rate.

[0021] Optionally, based on the detection results of the node in the current task batch, a corresponding gain-attenuation factor is calculated, wherein a gain factor is used for nodes that are successfully detected, and an attenuation factor related to the number of consecutive failures is used for nodes that fail to be detected; a new detection value is calculated based on the gain-attenuation factor and the current detection value of the node; the new detection value is subjected to interval pruning to ensure that it is within a preset range; and the record of the number of consecutive failures of the node is updated.

[0022] Compared with the prior art, the present invention has the following advantages and technical effects: This invention uses node-level detection value as a priori for task allocation and path generation. Compared with existing methods that focus on geometric coverage and separate detection from planning, it can guide the path for detection tasks in the initial stage, significantly improving effective coverage and anomaly detection rate.

[0023] Under the constraints of a shared takeoff and landing point and multi-drone collaboration, this invention generates UAV mission sequences through greedy allocation and ensures the executability of the planned path with a flight time safety threshold. Furthermore, this invention initializes pheromones based on node value, which, compared to traditional uniform initialization schemes, allows for faster entry into the high-quality solution domain and shortens convergence time.

[0024] This invention introduces coverage-time dual adaptive pheromone evaporation and a heuristic function guided by gains and uncovered areas to prevent premature convergence of the algorithm. Compared with existing methods that use a fixed evaporation rate or a single distance heuristic function, this approach can effectively reduce path overlap and ineffective round trips, and exhibits better robustness and global coverage balance under complex terrain and environmental disturbances.

[0025] By using the node value transfer function between different task batches, a closed-loop self-learning mechanism of "detection-planning-re-detection" is formed, which adaptively focuses on high-value and high-uncertainty areas, reduces the repeated coverage of low-value areas, and adapts to the actual operation needs of multi-round and multi-machine collaboration. Attached Figure Description

[0026] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 A flowchart illustrating the task allocation process using the greedy strategy in an embodiment of the present invention. Detailed Implementation

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0029] Example 1 like Figures 1-2 As shown, this embodiment provides a multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm, including: The first step is to initialize the detection value of the node to be tested, as follows: (1) Calculate scene weights. First, the work area is rasterized and a set of nodes to be tested is generated. Then, a multi-rotor UAV equipped with an RGB camera, RTK positioning module and light sensor conducts a full-domain aerial survey of the work area to complete orthorectification and image mosaicking, and constructs a global orthorectified RGB image for subsequent node attribute calculation and value assessment.

[0030] On the preprocessed orthophoto, the visible vegetation index (VARI, preferred) is calculated to generate a pixel-level vegetation vitality raster, which serves as a component of the scene weights. The VARI index is as follows: ; in, 、 、 For nodes after radiometric calibration and topographic correction The reflectance at that location. The mean VARI value for all nodes is: ; The anomaly score of a node is defined as follows: ; in, The standard deviation of VARI for all nodes in the graph. This is used to measure the "abnormality" of a node relative to the global situation; a value greater than 0 indicates low vegetation vitality and has detection value. Subsequently, [further details are needed]. Perform interval reduction and linear mapping to ensure that the final value is between 0 and 1. The closer it is to 1, the higher the degree of abnormality.

[0031] ; Based on the degree of anomaly, the node to be tested is obtained. The scenario weights are as follows, where, This represents the vegetation anomaly weighting coefficient.

[0032] ; (2) Calculate at the node to be tested Success rate of scanning at the location Considering factors such as lighting and safe altitude, the success rate of scanning may be reduced when the lighting is too dim or the relative altitude is outside the safe range. Real-time readings for the light sensor With light saturation value The ratio. If the current drone's altitude relative to the ground... If it is within the safe zone, then Set to 0 otherwise set to 1. and These are the illumination gain coefficient and the height penalty coefficient, respectively. This is a baseline constant. After weighting and summing all terms, the final success probability of the scan is calculated using the sigmoid function.

[0033] ; (3) Multiply the success rate of scanning by the scene weight to obtain the final detection value of the node to be tested.

[0034] .

[0035] The second step involves assigning the nodes to be detected by each drone using a greedy algorithm, based on maximizing the reward function. The reward function for moving from the current position k to the point j to be detected is defined as follows: ; in, This is a time efficiency weighting coefficient. To detect the value weighting coefficient, The terrain complexity penalty coefficient is between (0,1) and is assigned empirically based on the focus of the task.

[0036] For time-consuming items: ; in, The typical linear velocity of a drone's flight. The typical angular velocity for drone flight, Indicates the straight-line flight time of the drone. This indicates the time taken for the drone to turn. A fixed time is allocated for node detection.

[0037] The detection value of the point j to be tested.

[0038] The climb penalty term represents whether a flight from node k to node j requires climbing to a safe altitude, and is expressed as follows: ; in, Let j be the safe height of the point to be measured. This represents the actual height of the current point.

[0039] The specific process is as follows: Initially, the task queue of drone k... Cumulative income Task time Allocated node set .

[0040] Each round starts by selecting the next test point from the drone with the lowest current cumulative profit. For each unassigned test point... Remove nodes that exceed the time budget, i.e., nodes that satisfy the following formula: ; in, This is the maximum safe threshold time estimated based on aircraft energy consumption. The estimated time for the return trip from point j to the starting point b is... .

[0041] Subsequently, the overall benefit for each selectable test point is calculated. Select the node with the highest profit As the next target point for the drone. Add to task queue In the middle, and put As the latest location of the drone, simultaneously update the cumulative revenue of drone k. Total task time and the set of allocated nodes : ; ; ; Proceed to the next round of node allocation until either of the following conditions is triggered, at which point the drone returns to base: (1) There are no available test nodes for the current drone. (2) All test nodes have been added to the set. In the middle. Once all drones have returned to base, the initial task assignment is complete, and each drone has generated an initial exploration path.

[0042] For each drone, the nodes that need to be traversed are already determined, and the problem is transformed into a general path planning problem.

[0043] The third step involves initializing the pheromones for each path based on the previously obtained node detection values. This guides the UAV to search for paths that are easy to detect, easy to fly, and low in energy consumption, while preserving the global guidance capabilities of nodes with high detection values ​​to prevent it from getting stuck in locally optimal paths in the early stages of the search. Assume the detection values ​​of nodes i and j are... , Then the pheromone on path ij is initialized as follows: ; Normalize all pheromones: ; At this point, pheromone initialization is complete.

[0044] The fourth step is to utilize the benefits obtained in the second step. The improved heuristic function in the traditional ant colony algorithm, along with the uncovered area guiding factor, better addresses the path planning problem. The key steps of the improved ant colony algorithm are shown below.

[0045] Each drone corresponds to an ant colony for optimal path search. The ant colony starts from the starting point and randomly selects nodes according to the state transition probability function. An iteration is considered complete when all ants have searched all the nodes assigned to them.

[0046] In each iteration, the state transition probability from node i to node j is: ; in, Let be the pheromone of path ij in the t-th iteration. This represents the heuristic function value of path ij in the t-th iteration. Let represent the product of the pheromone concentration and heuristic function value of node i and all its neighboring nodes j. For pheromone weights, To heuristically weight information, pheromones are controlled separately. and inspirational information The degree to which these two parameters affect the ants' path selection will be dynamically adjusted within an empirical range based on actual results: The empirical range is (0.5, 5), and the commonly used initial value is 1; The empirical range is (1, 10), and the commonly used initial value is 2. The definition is as follows: ; in, The result after normalizing the revenue term: ; It is a minimum value used to prevent the denominator from being 0.

[0047] For uncovered areas, This indicates that node j has not been visited yet; otherwise, it is 0. This is an exploration enhancement coefficient. This design, while ensuring a certain level of benefit, guides drones into unexplored blind spots, effectively improving overall coverage.

[0048] Fifth, after completing one iteration of the ant path using the state transition probability formula, it is necessary to update the pheromones on each path. This invention designs a "coverage-time dual adaptive" pheromone evaporation mechanism for pheromone updating. This mechanism reflects the dynamic characteristics of pheromones during the search process and can effectively enhance the adaptability and optimization performance of the algorithm. Specific details are as follows.

[0049] In the initial stage of the iteration, the pheromone evaporation coefficient is... Setting it to a relatively large value effectively promotes the ant colony's exploration ability. As time progresses, the volatile coefficient gradually decreases, causing the model to converge and stabilize. That is, at the beginning of each iteration, the volatile coefficient is at its maximum value; after each iteration, it decreases based on the current iteration round. Calculate the global baseline volatility coefficient Until the volatility coefficient reaches the preset minimum value: ; The global baseline volatility coefficient was calculated. Then, this coefficient needs to be dynamically adjusted based on the current coverage progress. First, the coverage rate of each edge needs to be calculated. : ; in, This represents the cumulative number of times edge ij has been visited up to the current round t. This represents the total number of ants. Therefore... This can be viewed as the theoretical maximum number of times an edge is visited in historical iterations. Based on edge coverage... and global baseline volatility coefficient The volatile rate on edge ij can be obtained: ; To cover influencing factors, the extent to which coverage affects volatiles is determined. The larger the value, the greater the decrease in evaporation rate under low coverage; when At lower levels, the edges The pheromone evaporation rate decreases, and more pheromones are retained. When it is higher, the edge The pheromone evaporation rate is closer to the baseline evaporation rate. The advantages of this approach are: early high volatility can quickly eliminate poor paths, and coverage adjustment ensures that pheromones on less popular routes are not depleted too quickly, preserving exploration potential; in the later stages, high-coverage edges evaporate normally, and pheromones concentrate on truly high-quality routes, accelerating convergence.

[0050] After obtaining the pheromone evaporation rate of edge ij, the pheromone is updated: ; in, It is the edge Updated pheromones. It's an ant. In the path The increase in pheromones left on the trail is usually inversely proportional to the length of the flight path. The calculation formula is: ; in, Let be a constant representing the importance of pheromones. Indicates the current ant The length of the path traveled. Therefore... This indicates that in the t-th iteration, on the path... The sum of the pheromone increments left on the surface. This completes the pheromone update after one iteration, and the next iteration begins. Steps four and five are repeated until the preset number of iterations is reached, at which point the ant colony algorithm terminates. At this point, the shortest path traversed by all ants is considered the globally optimal path, and the task is complete.

[0051] The sixth step is to design a node detection value transfer function to update the node detection value after all drones have completed a mission and returned, so as to adapt to scenarios with multiple batches of missions.

[0052] This step ensures that the detection value of each node under test is not static, but dynamically adjusted based on historical detection results: if a node fails to detect a target, its detection value is appropriately reduced, with the reduction increasing with the number of detections; if a node detects a target, its detection value is significantly increased. The drone completes the first step... The gain-attenuation factor design for the secondary task is as follows: ; in, This is the positive gain coefficient when the detection is successful. The attenuation coefficient when detection fails. Representing the Secondary task node The test result was 1 successful. Representing the deadline During the next task, the number of consecutive failed checks is counted; once a check is successful, The update formula for resetting to zero is as follows: ; Ultimately, the value of node detection The updated formula is as follows, which guarantees that... It will not exceed the preset range: ; Before the next drone takeoff, tasks are redistributed using a greedy strategy based on the new node detection value.

[0053] Node detection value determination method: Based on rasterized partitioning, the cell is used as the smallest decision node. The vegetation index VARI is calculated using orthorectified RGB imagery, and scene weights are generated. The probability of successful scanning is evaluated by combining node illumination intensity and relative height. The scene weights and the probability of successful scanning are fused to form the node detection value, which is used as a priori input for subsequent task allocation and path planning.

[0054] 1. In the initial full-map scanning phase of the task, a method for determining the detection value of nodes is established as an important reference for subsequent task allocation and path planning. Here, a node refers to a detectable cell. The method includes: (1) Scene weight calculation: Calculate the visible light vegetation index based on the airborne RGB image to obtain the initial detection value of the node. The above vegetation index can also be replaced by other visible light vegetation indices. (2) Scan success probability assessment: Combine the node illumination intensity and relative height information, consider terrain occlusion and sensor imaging constraints, and determine and quantify the scanning success probability of the node. (3) Weight and fuse the node scene weight and scanning success probability obtained above to obtain the detection value of the node, which is used to guide the subsequent multi-UAV task allocation and path planning.

[0055] 2. Under the constraint of multiple UAVs sharing the same takeoff and landing point, a greedy selection-based method for task allocation and initial path generation is proposed. Initially, a unified takeoff and landing point location, a set of nodes to be detected, and the detection value of each node are determined. Parameters such as the maximum flight endurance, return time, detection time, and average flight speed of the UAVs are set. Subsequently, greedy selection is used for task allocation among the multiple UAVs. Based on the principle of maximizing profit, a profit function is established by comprehensively considering time efficiency factors (straight-line flight time, turning maneuver time, detection dwell time), detection profit factors (detection value of nodes), and safety penalties (such as the climb requirement triggered when the camera is too close to the ground and the resulting energy consumption penalty). In each round, the path is greedily selected starting from the UAV with the lowest current cumulative profit. The node with the highest current profit is calculated and added to the path set of that UAV, while its cumulative profit is updated. When the total flight time of a UAV reaches a certain threshold, the UAV triggers a return, and task allocation for that UAV stops. When all UAVs trigger a return or there are no nodes to be detected, task allocation ends, and the UAV path initialization is completed.

[0056] 3. This embodiment proposes a pheromone initialization scheme based on node detection value to improve the quality of the initial solution, focusing attention on high-yield regions in the early search phase. Prior weights based on node detection value are introduced to provide positive guidance for pathways with high potential returns, ease of detection, and low energy consumption across the entire domain.

[0057] 4. Design a novel heuristic function that balances benefit and coverage guidance. This function considers flight time consumption, node detection value, and terrain risk during ant colony state transitions, and explicitly overlays guidance factors for uncovered areas to enhance attractiveness to blind spots and edge regions. This design can improve effective coverage and anomaly detection rate within given endurance and safety constraints.

[0058] 5. A coverage-time dual adaptive pheromone evaporation mechanism is proposed, which dynamically adjusts the evaporation intensity based on the current coverage progress of the pheromone region and the time status of the search process: in the early exploration stage or when coverage is insufficient, evaporation is increased to suppress path solidification and encourage exploration; when coverage is nearly sufficient or the convergence stage is reached, evaporation is reduced to stabilize high-quality pathways. This dual adaptive strategy can avoid overly rapid convergence or long-term oscillations caused by relying solely on time or fixed parameters, thereby achieving higher global coverage efficiency and a more robust optimization process in complex environments.

[0059] 6. Set up a node detection value transfer function to deal with scenarios where multiple batches of tasks are executed: After all drones complete a round of tasks, the node detection value is dynamically adjusted based on the detection results of each node. The value of areas where abnormal vegetation is detected is increased, while the value of areas where no abnormalities or low effectiveness are detected continuously is decreased. In this way, key areas are gradually focused on and ineffective coverage is reduced in multiple rounds of operation.

[0060] Multi-drone greedy task allocation and safe return mechanism at the same take-off and landing point: Under the constraint of a known number of drones and the same take-off and landing point, the task sequence of each drone is generated by comprehensively considering the time consumption of straight flight, turning maneuver, detection dwell and climb, node detection value and terrain risk according to the principle of "maximizing benefits". At the same time, a flight time safety threshold is set to trigger the return, so as to ensure the one-time execution rate of the path and flight safety.

[0061] Pheromones initialization method based on prior knowledge: Using the node detection value obtained in the greedy phase as a prior, pheromone initialization is performed on each edge to shorten the cold start time and avoid early undirected search.

[0062] Based on a heuristic function guided by benefits and uncovered areas: During state transition, flight time consumption, node detection value and terrain risk are jointly considered, and an uncovered area guiding factor is explicitly added to increase the attractiveness to blind spots and edge areas, suppress path overlap, and improve the overall effective coverage and anomaly detection rate.

[0063] Coverage-time dual adaptive pheromone evaporation mechanism: The pheromone evaporation intensity is simultaneously regulated by search time and coverage progress. When coverage is insufficient or in the early stages of the search, the evaporation rate is increased to encourage exploration; when coverage is nearly sufficient or in the later stages of the search, the evaporation rate is reduced to stabilize high-quality pathways, achieving a dynamic balance between exploration breadth and convergence speed.

[0064] Cross-batch node detection value transfer function: After each round of tasks, the node detection value is dynamically transferred based on the detection results, increasing the value of detected abnormal areas and incrementally decreasing the value of areas with continuous non-detection or low effectiveness. This enables adaptive adjustment of detection value across rounds, achieving a closed loop between planning and detection.

[0065] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-UAV path planning method integrating a greedy strategy and an improved ant colony algorithm, characterized in that, Includes the following steps: Based on remote sensing image data and environmental sensor data of the area to be tested, the detection value of each node to be tested is calculated. Based on the detection value of each node to be tested and the preset revenue function, a greedy strategy is used to assign a sequence of nodes to be tested to multiple drones and generate an initial path. The path pheromone is initialized based on the detection value of each node to be tested; A heuristic function is constructed based on the profit function and the uncovered area guiding factor. Based on the initial path, the heuristic function is used to search the path during the ant colony algorithm iteration process. After each iteration, the path pheromone is updated using a coverage-time dual adaptive pheromone evaporation mechanism. The coverage-time dual adaptive pheromone evaporation mechanism includes: calculating a time-adaptive baseline evaporation coefficient based on the iteration rounds, wherein the baseline evaporation coefficient decays exponentially with the number of iterations; obtaining the coverage progress based on the ratio of the number of times each path is visited in historical iterations to the theoretical maximum number of visits; calculating the actual evaporation rate of each path based on the baseline evaporation coefficient and the coverage progress; and updating the pheromone of each path based on the actual evaporation rate. After completing path planning and executing detection tasks based on iteratively updated path pheromones, the detection value of each node is dynamically updated according to the execution results of the detection tasks. Execute subsequent tasks based on the updated node detection value to achieve closed-loop path planning for multiple batches of tasks.

2. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 1, characterized in that, The process of calculating the detection value of each node to be tested includes: calculating the visible light vegetation index of each node based on the orthophoto RGB image; obtaining scene weights based on the statistical characteristics of the vegetation index of each node and the visible light vegetation index; calculating the success probability of scanning based on the light intensity and relative height information of the node; and fusing the scene weights and the success probability of scanning to obtain the node detection value.

3. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 1, characterized in that, The preset profit function is: ； in, For time consumption items, ; The typical linear velocity of a drone's flight. The typical angular velocity for drone flight, Indicates the straight-line flight time of the drone. This indicates the time taken for the drone to turn. Fixed time consumption for node detection Point to be measured The detection value, The climbing penalty term represents the penalty from the node. Fly to the node Is it necessary to climb to a safe height? , Point to be measured Safe altitude, This represents the actual height of the current point. This is a time efficiency weighting coefficient. To detect the value weighting coefficient, The terrain complexity penalty coefficient is between (0,1).

4. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 1, characterized in that, The process of obtaining the initial path includes: Initialize all drone task queues to empty, cumulative revenue to zero, and the set of assigned nodes to empty; In each round of allocation, node selection begins with the drone with the lowest current cumulative profit. Candidate nodes that meet the time constraints are selected from the unallocated test nodes. The profit function value of the drone moving from its current position to each candidate node is calculated. The candidate node with the highest profit function value is selected as the next target node. The target node is added to the task queue of the UAV, and the cumulative revenue, cumulative flight time and the set of assigned nodes of the UAV are updated until the termination condition is met, and the next UAV is assigned. When all UAVs have completed task assignment or triggered the return condition, the initial path of each UAV is obtained.

5. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 4, characterized in that, The time constraint is that the sum of the current cumulative flight time, the time to move to the candidate node, and the estimated time to return from the candidate node to the starting point shall not exceed the maximum safe flight time of the UAV.

6. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 1, characterized in that, Initializing pheromones on a path includes: calculating the initial pheromone value of the path based on the detection values ​​of the nodes at both ends of the path; and normalizing the initial pheromone values ​​of all paths.

7. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 1, characterized in that, Constructing a heuristic function based on the revenue function and the uncovered area guiding factor includes: normalizing the revenue function value to obtain a standardized revenue term; setting an uncovered area guiding factor based on whether a node has been visited; and multiplying the standardized revenue term by the uncovered area guiding factor to obtain the heuristic function value.

8. The multi-UAV path planning method that integrates a greedy strategy and an improved ant colony algorithm according to claim 1, characterized in that, Based on the detection results of the node in the current task batch, calculate the corresponding gain-attenuation factor, where the node that is successfully detected uses the gain factor, and the node that fails to be detected uses the attenuation factor related to the number of consecutive failures; calculate the new detection value based on the gain-attenuation factor and the node's current detection value; perform interval pruning on the new detection value to ensure it is within a preset range; update the node's consecutive failure count record.

Images