Air-ground cooperative termite detection adaptive path planning method based on artificial intelligence

Abstract

The invention discloses an air-ground cooperative termite detection adaptive path planning method based on artificial intelligence. The method comprises the following steps: S1, collecting an environment data set of a target detection area; s2, constructing a fractal space model of the detection area by using a fractal modeling method; s3, generating a priority distribution diagram of the target detection area; s4, initializing a population position and a population scale of a sparrow search algorithm; s5, generating a preliminary detection path covering the detection area; s6, generating a priority distribution diagram according to the preliminary detection path in combination with fractal modeling; s7, preferentially covering a high-priority area in the detection area; and S8, planning a scheme according to the adjusted path. The dynamic path optimization mechanism has excellent performance in the aspects of path conflict avoidance and priority area coverage, so that the path planning efficiency is improved by more than 20%, and the resource waste is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of termite detection, and in particular to an adaptive path planning method for air-ground collaborative termite detection based on artificial intelligence. Background Art

[0002] With the rapid development of artificial intelligence and unmanned system technologies, intelligent tools have gradually been introduced into the field of environmental monitoring for pest detection and control. As a highly concealed and dispersed pest, termites pose a serious threat to buildings, the ecological environment, and infrastructure. Therefore, the precise detection and efficient control of termite activities have become a research hotspot in the field of environmental monitoring.

[0003] Traditional termite detection mainly relies on manual inspections or single devices, usually inferring the activity range of termites by detecting local areas. However, due to the dispersion and concealment of termite activities, traditional methods often cannot achieve full coverage, resulting in a high missed detection rate. In addition, manual detection is inefficient and labor-intensive, and the difficulty of manual detection increases significantly in complex terrains or high-risk areas.

[0004] In recent years, the collaborative technology of drones and ground robots has begun to be applied to the pest detection scenario, expanding the detection range and improving the detection efficiency through air-ground collaboration. However, most existing air-ground collaborative detection methods adopt fixed path planning or simple full-coverage strategies, ignoring the priority differences of different areas within the detection area and the dynamic distribution of obstacles, which not only increases resource consumption but also may lead to insufficient detection of high-priority areas. In addition, existing methods lack the ability of dynamic optimization in path planning, making it difficult to achieve efficient collaboration between drones and ground robots, resulting in uneven task allocation and path conflict problems.

[0005] In summary, the existing technologies have significant deficiencies in the full coverage of termite detection, the accuracy of priority area detection, dynamic path planning, and air-ground collaborative optimization, and are difficult to meet the requirements of efficient and precise termite detection in complex environments. The above technical defects directly affect the efficiency and accuracy of pest detection, and there is an urgent need for a method that can make full use of artificial intelligence and collaborative optimization technologies to solve the above problems. Summary of the Invention

[0006] An object of the present invention is to propose an adaptive path planning method for air-ground collaborative termite detection based on artificial intelligence. The dynamic path optimization mechanism of the present invention performs excellently in path conflict avoidance and priority area coverage, improving the path planning efficiency by more than 20% and effectively reducing resource waste.

[0007] An adaptive path planning method for air-ground collaborative termite detection based on artificial intelligence according to an embodiment of the present invention includes the following steps:

[0008] S1. Collect the environmental data set of the target detection area;

[0009] S2. Based on the collected environmental data set, use the fractal modeling method to construct the fractal space model of the detection area;

[0010] S3. According to the fractal space model, divide the priority of the target detection area, determine the priority weight, and generate the priority distribution map of the target detection area;

[0011] S4. According to the priority distribution map of the detection area, initialize the population position and population size of the sparrow search algorithm;

[0012] S5. Using the foraging strategy of the sparrow search algorithm, perform global path planning in the fractal space model through an aerial platform to generate a preliminary detection path covering the detection area;

[0013] S6. According to the preliminary detection path and the priority distribution map generated by fractal modeling, use the local optimization strategy of the sparrow search algorithm to optimize the path planning for high-priority areas;

[0014] S7. Based on the optimized detection path, adaptively adjust the detection paths of the aerial platform and the ground platform. The updated detection path avoids obstacle conflicts and preferentially covers high-priority areas within the detection area;

[0015] S8. According to the adjusted path planning scheme, the aerial platform is responsible for performing the global scan of the target detection area, and the ground platform is responsible for performing the detection of the target detection area.

[0016] Optionally, S1 includes the following specific steps:

[0017] S11. Use multi-modal sensors to collect the environmental data of the target detection area, including terrain information, multi-modal feature data of termite activities, thermal imaging data, and obstacle distribution data;

[0018] S12. Digitally process the terrain information to generate a digital elevation model T(x, y) representing the terrain features of the detection area, where T(x, y) represents the elevation value of any point (x, y) in the target detection area and is used to mark the undulating features of the detection area;

[0019] S13. Collect the multi-modal feature data of termite activities, generate the two-dimensional image data set V using the vision sensor, and generate the temperature distribution data H(x, y) using the thermal imaging sensor;

[0020] S14. Generate an obstacle distribution matrix O(x, y) using sensors, where O(x, y) represents whether there is an obstacle at point (x, y), and the value is a binary variable: O(x, y) = 1 indicates the presence of an obstacle, and O(x, y) = 0 indicates the absence of an obstacle;

[0021] S15. Fuse the terrain information T(x, y), the two-dimensional image dataset V of termite activities, the temperature distribution data H(x, y), and the obstacle distribution data O(x, y) to construct an environmental dataset D:

[0022] D = {T(x, y), V, H(x, y), O(x, y)}.

[0023] Optionally, the S2 includes the following specific steps:

[0024] S21. Dynamically partition the target detection area based on the environmental dataset D, and use an adaptive grid division method based on the thermal feature gradient and obstacle distribution to divide the detection area into several non-uniform sub-regions R i , i = 1, 2,..., n, where the grid size of the sub-region is dynamically adjusted according to the change rate of the thermal distribution gradient and the obstacle density gradient

[0025]

[0026] where, Δx, Δy represent the side lengths of the sub-region grids;

[0027] S22. Introduce a dynamically adjusted grid scale ∈ k , k = 1, 2,..., m:

[0028] ∈ k = ∈ min ·(1 + α k ·G h + β k ·G o );

[0029] where, ∈ min is the minimum grid scale, α k and β k are the influence coefficients of the thermal feature and the obstacle feature;

[0030] Calculate the effective coverage grid number N(∈ k ) at each grid scale ∈ k and obtain the fractal dimension:

[0031]

[0032] S23. Classify the complexity of sub-regions according to the fractal dimension D f (R i ) to classify the complexity of sub-regions, and divide the detection area into three categories: high-complexity area, medium-complexity area, and low-complexity area. The classification threshold is determined by clustering analysis of the statistical distribution of the fractal dimension;

[0033] S24. Introduce the probability weight W based on the fractal dimension for high-complexity regions i for priority assignment:

[0034]

[0035] where W i represents the priority weight of sub-region R i in the entire detection area;

[0036] S25. Map the priority weight W i to the task assignment matrix M in combination with the task execution characteristics of the aerial platform and the ground platform:

[0037] M = {(R i , P k ): R i →P k , where P k ∈{aerial platform, ground platform}};

[0038] where P k represents the execution entity of the air-ground collaborative detection, and the assignment in the matrix is completed based on the priority weight and the task adaptability of the platform;

[0039] S26. Output the fractal space model F, which includes the fractal dimension, complexity classification, and task assignment matrix of the sub-regions.

[0040] Optionally, the S3 includes the following specific steps:

[0041] S31. Normalize the fractal dimension D i (R f ) of each sub-region R i based on the fractal space model F;

[0042] S32. Calculate the priority weight of each sub-region according to the normalized fractal dimension ;

[0043] S33. Map the priority weight to the spatial distribution of the detection area in combination with the priority weight and the detection task requirements, and generate the priority distribution map P(x, y) of the target detection area:

[0044]

[0045] Among them, P(x, y) represents the priority value of the point (x, y) in the target detection area. is the indicator function of the sub-region R i which takes the value of 1 when the point (x, y) belongs to the sub-region R i and 0 otherwise;

[0046] S34. Classify the priority distribution map P(x, y), and divide the detection area into high-priority area, medium-priority area and low-priority area according to the priority value. The classification basis is to set the threshold P high and P low for division:

[0047] When P(x, y) ≥ P high , the point (x, y) belongs to the high-priority area;

[0048] When P low ≤ P(x, y) < P high , the point (x, y) belongs to the medium-priority area;

[0049] When P(x, y) < P low , the point (x, y) belongs to the low-priority area.

[0050] Optionally, the S4 includes the following specific steps:

[0051] S41. Divide the target detection area into several priority areas Q k , and the priority area Q k contains all spatial points (x, y) with the same priority:

[0052]

[0053] Among them, and are the priority value ranges of the k-th priority area;

[0054] S42. According to the distribution characteristics of the priority area Q k , initialize the population size N k of the sparrow search algorithm:

[0055]

[0056] Among them, N total represents the total population size of the sparrow search algorithm;

[0057] S43. For each priority area Q k , according to the distribution density ρ of the points in the areak Initialize the positions of the population individuals. The initial position X of the population individuals i =(x i , y i ) is distributed according to the following density control principle:

[0058]

[0059] where |Q k | represents the number of points in the priority area Q k . The population individuals are preferentially distributed in the areas with high point density;

[0060] S44. Initialize the attributes of the population individuals, including the distribution status of detection resources for aerial platforms and ground platforms. The detection resource attributes include platform type and the initial resource quantity of the platform;

[0061] S45. Output the initialized population information as the initial condition for path planning of the sparrow search algorithm, including population size, population position, platform type, and resource status.

[0062] Optionally, the S5 includes the following specific steps:

[0063] S51. Construct a foraging model of the sparrow search algorithm based on the initialized population information;

[0064] S52. Assign initial positions to each population individual according to the priority area Q k and the priority weight W k . The initial positions need to be located in the high-priority sub-areas within the priority area;

[0065] S53. Update the positions of the population individuals according to the foraging strategy of the sparrow search algorithm

[0066]

[0067] where and represent the current global optimal position and the worst position respectively, α and β are adjustment parameters, r is a random number, and P c is the control threshold;

[0068] S54. On the basis of the foraging strategy, evaluate the positions of the population individuals through the fitness function f(X i ):

[0069] f(X i ) = w 1 ·Coverage(X i ) - w 2 ·Energy(Xi ) + w 3 ·Priority(X i );

[0070] Among them, Coverage(X i ) represents the area of the detection region covered by an individual, Energy(X i ) is the energy consumption of the path, Priority(X i ) is the priority weighted value of the area covered by an individual path, and w 1 , w 2 , w 3 are weight factors;

[0071] S55. Select the current optimal path node set according to the fitness function value, and generate a preliminary detection path P global :

[0072] P global ={X start , X 1 , X 2 ,…, X end};

[0073] Among them, X start and X end are the initial node and the target node respectively, and the path nodes are dynamically updated through the sparrow foraging strategy;

[0074] S56. Control the aerial platform to perform real-time data collection along the preliminary detection path P global , including multi-modal environmental data and termite activity characteristic data of the target detection region.

[0075] Optionally, the S6 includes the following specific steps:

[0076] S61. Extract the set R global of high-priority regions covered by the preliminary detection path based on the preliminary detection path P high :

[0077] R high ={(x, y) | P(x, y) ≥ P high and (x, y) ∈ P global};

[0078] Among them, P high is the high-priority threshold;

[0079] S62. Use the set R high of high-priority regions as the local optimization target region, and initialize the population individual positions of the sparrow search algorithm in R highUniformly distributed inside, and the position initialization rule is:

[0080]

[0081] S63. Update the positions of the population individuals according to the local optimization strategy of the sparrow search algorithm

[0082] Among them, and respectively represent the current local optimal position and the worst position, γ and δ are local optimization adjustment parameters, r is a random number, and P c is the control threshold;

[0083] S64. Define the local optimization fitness function f local (X i ), and conduct a comprehensive evaluation by combining the fractal dimension and the detection area priority value:

[0084]

[0085] Among them, P(x i ,y i ) is the priority value of the current individual position, w 4 and w 5 are weight factors;

[0086] S65. According to the fitness function value, select individuals with high fitness as local path nodes to generate an optimized high-priority area path P local :

[0087] P local ={X start ,X h1 ,X h2 ,…,X end};

[0088] Among them, X h1 ,X h2 ,… are optimized path nodes in the high-priority area.

[0089] Optionally, the above S7 includes the following specific steps:

[0090] S71. Based on the optimized high-priority area path P local and combine the obstacle distribution matrix O(x,y) to detect obstacles in the path, and define the conflict detection function:

[0091]

[0092] Among them, C(X) is the path conflict volume. If C(X) > 0, there is an obstacle conflict in the path, and the path needs to be readjusted.

[0093] S72. For the set of path nodes X with obstacle conflicts conflict , redefine the set of feasible path nodes X feasible ;

[0094] S73. On the generated set of feasible path nodes X feasible , reassign the path priority weights according to the priority distribution map P(x, y);

[0095] S74. Regenerate the optimized paths of the aerial platform and the ground platform according to the updated priority weights:

[0096] P updated = {X start , X u1 , X u2 , …, X end};

[0097] Among them, X u1 , X u2 , … are the updated feasible path nodes;

[0098] S75. Based on the updated optimized paths of the aerial platform and the ground platform P updated , allocate the path nodes to the aerial platform and the ground platform, and determine the detection range of each platform through the following task allocation function M(X, P k ):

[0099]

[0100] Among them, P k is the type of the detection platform;

[0101] S76. Output the updated path P updated and the platform allocation result M(X, P k ), so that the path avoids obstacle conflicts and preferentially covers high-priority areas.

[0102] The beneficial effects of the present invention are:

[0103] (1) The present invention uses the fractal modeling method to calculate the fractal dimension of the detection area, establishes a complexity characterization model of the target area, can dynamically determine the priority weights of high-complexity areas and low-complexity areas through the priority division mechanism, and preferentially processes high-priority areas in path planning. It can dynamically adjust the priority distribution map according to the real-time collected data, making the detection of key areas in complex environments more targeted, and significantly improving the comprehensiveness of detection coverage and detection accuracy.

[0104] (2) The present invention introduces the sparrow search algorithm in path planning. By means of the global foraging strategy, a preliminary detection path is generated, and combined with the local optimization strategy, the layout of path nodes is further optimized within the high-priority area, ensuring efficient cooperation between the aerial platform and the ground platform during task execution. The dynamic path optimization mechanism performs excellently in path conflict avoidance and coverage of priority areas, increasing the path planning efficiency by more than 20% and effectively reducing resource waste.

[0105] (3) By introducing the obstacle detection and dynamic avoidance strategy, the present invention redistributes the path nodes in combination with the functional characteristics of the aerial platform and the ground platform, and adjusts the detection path in real time through the obstacle distribution matrix and the priority distribution map to avoid path conflicts and preferentially cover high-priority areas. At the same time, through the task allocation function, the high-priority areas are assigned to the aerial platform and the low-priority areas are assigned to the ground platform, realizing the rational utilization of resources and the maximization of the collaborative work efficiency. Description of the Drawings

[0106] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0107] Figure 1 is a flowchart of an adaptive path planning method for aerial-ground collaborative termite detection based on artificial intelligence proposed by the present invention. Detailed Embodiment

[0108] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0109] Refer to Figure 1 , an adaptive path planning method for aerial-ground collaborative termite detection based on artificial intelligence, comprising the following steps:

[0110] S1. Collect the environmental data set of the target detection area;

[0111] S2. Based on the collected environmental data set, use the fractal modeling method to construct the fractal space model of the detection area;

[0112] S3. According to the fractal space model, perform priority division on the target detection area, determine the priority weight, and generate the priority distribution map of the target detection area;

[0113] S4. According to the priority distribution map of the detection area, initialize the population position and population size of the sparrow search algorithm;

[0114] S5. Utilize the foraging strategy of the Sparrow Search Algorithm to perform global path planning in the fractal space model through an aerial platform, generating a preliminary detection path that covers the detection area.

[0115] S6. Based on the preliminary detection path and the priority distribution map generated by fractal modeling, utilize the local optimization strategy of the Sparrow Search Algorithm to optimize the path planning for high-priority areas.

[0116] S7. Based on the optimized detection path, adaptively adjust the detection paths of the aerial platform and the ground platform. The updated detection path avoids obstacle conflicts and preferentially covers high-priority areas within the detection area.

[0117] S8. According to the adjusted path planning scheme, the aerial platform is responsible for performing a global scan of the target detection area, and the ground platform is responsible for performing detections in the target detection area.

[0118] In this embodiment, S1 includes the following specific steps:

[0119] S11. Use multimodal sensors to collect environmental data of the target detection area, including terrain information, multimodal feature data of termite activities, thermal imaging data, and obstacle distribution data.

[0120] S12. Digitally process the terrain information to generate a digital elevation model T(x, y) representing the terrain features of the detection area, where T(x, y) represents the elevation value of any point (x, y) in the target detection area and is used to mark the undulating features of the detection area.

[0121] S13. Collect multimodal feature data of termite activities, generate a two-dimensional image dataset V using a vision sensor, and generate temperature distribution data H(x, y) using a thermal imaging sensor.

[0122] S14. Use the sensor to generate an obstacle distribution matrix O(x, y), where O(x, y) represents whether there is an obstacle at point (x, y), and the value is a binary variable: O(x, y) = 1 indicates the presence of an obstacle, and O(x, y) = 0 indicates no obstacle.

[0123] S15. Fuse the terrain information T(x, y), the two-dimensional image dataset V of termite activities, the temperature distribution data H(x, y), and the obstacle distribution data O(x, y) to construct an environmental dataset D:

[0124] D = {T(x, y), V, H(x, y), O(x, y)}.

[0125] In this embodiment, S2 includes the following specific steps:

[0126] S21. Dynamically partition the target detection area based on the environmental data set D, and divide the detection area into several non-uniform sub-regions R by using an adaptive grid division method based on thermal feature gradients and obstacle distributions i , i = 1, 2, …, n, where the grid size of the sub-region is based on the change rate of the thermal distribution gradient and the obstacle density gradient and is dynamically adjusted. The grid division rule is as follows:

[0127]

[0128] where Δx and Δy represent the side lengths of the sub-region grid;

[0129] S22. Introduce a dynamically adjusted grid scale ∈ k , k = 1, 2, …, m:

[0130] ∈ k = ∈ min ·(1 + α k ·G h + β k ·G o );

[0131] where ∈ min is the minimum grid scale, and α k and β k are the influence coefficients of the thermal feature and the obstacle feature;

[0132] Calculate the number of effective coverage grids N(∈ k ) for each grid scale ∈ k and obtain the fractal dimension:

[0133]

[0134] S23. Classify the complexity of the sub-regions according to the fractal dimension D f (R i ) and divide the detection area into three categories: high-complexity areas, medium-complexity areas, and low-complexity areas. The classification threshold is determined by clustering analysis of the statistical distribution of the fractal dimension;

[0135] S24. Introduce a probability weight W i based on the fractal dimension for priority allocation in the high-complexity areas:

[0136]

[0137] where W i represents the priority weight of the sub-region R i in the entire detection area;

[0138] S25. Map the priority weight W to the task assignment matrix M according to the task execution characteristics of the aerial platform and the ground platform: i Map to the task assignment matrix M:

[0139] M = {(R i , P k ): R i →P k , where P k ∈ {aerial platform, ground platform}};

[0140] Among them, P k represents the execution entity of the air-ground collaborative detection, and the assignment in the matrix is completed according to the priority weight and the task adaptability of the platform;

[0141] S26. Output the fractal space model F, and the fractal space model includes the fractal dimension of the sub-region, complexity classification, and task assignment matrix.

[0142] In this embodiment, S3 includes the following specific steps:

[0143] S31. Perform normalization processing on the fractal dimension D i (R f ) of each sub-region R i ;

[0144] S32. Calculate the priority weight of each sub-region according to the normalized fractal dimension ;

[0145] S33. Map the priority weight to the spatial distribution of the detection region in combination with the priority weight and the detection task requirements, and generate the priority distribution map P(x, y) of the target detection region:

[0146]

[0147] Among them, P(x, y) represents the priority value of the point (x, y) in the target detection region, is the indicator function of the sub-region R i , and takes the value of 1 when the point (x, y) belongs to the sub-region R i , otherwise takes the value of 0;

[0148] S34. Classify the priority distribution map P(x, y), and divide the detection region into high-priority regions, medium-priority regions, and low-priority regions according to the priority value. The classification basis is to set thresholds P high and P low through statistical distribution characteristics of the priority value P(x, y) for division:

[0149] When P(x, y) ≥ P high the point (x, y) belongs to the high-priority area;

[0150] When P low ≤ P(x, y) < P high the point (x, y) belongs to the medium-priority area;

[0151] When P(x, y) < P low the point (x, y) belongs to the low-priority area.

[0152] In this embodiment, S4 includes the following specific steps:

[0153] S41. Divide the target detection area into several priority areas Q k based on the priority distribution map P(x, y), and the priority area Q k contains all spatial points (x, y) with the same priority:

[0154]

[0155] wherein, and are the priority value ranges of the k-th priority area;

[0156] S42. Initialize the population size N k of the sparrow search algorithm according to the distribution characteristics of the priority area Q k :

[0157]

[0158] wherein, N total represents the total population size of the sparrow search algorithm;

[0159] S43. For each priority area Q k , initialize the positions of the population individuals according to the distribution density ρ k of the points in the area, and the initial positions X i = (x i , y i ) of the population individuals follow the following density control principle:

[0160]

[0161] wherein, |Q k | represents the number of points in the priority area Q k , and the population individuals are preferentially distributed in the areas with high point density;

[0162] S44. Initialize the attributes of the population individuals, including the distribution status of detection resources for aerial platforms and ground platforms. The detection resource attributes include platform types and the initial resource amounts of the platforms.

[0163] S45. Output the initialized population information as the initial condition for path planning in the sparrow search algorithm, including population size, population location, platform types, and resource status.

[0164] In this embodiment, S5 includes the following specific steps:

[0165] S51. Construct a foraging model of the sparrow search algorithm based on the initialized population information.

[0166] S52. Assign initial positions to each population individual according to the priority area Q k and the priority weight W k The initial positions need to be located in the high-priority sub-areas within the priority area.

[0167] S53. Update the positions of population individuals according to the foraging strategy of the sparrow search algorithm

[0168]

[0169] Among them, and respectively represent the current global optimal position and the worst position, α and β are adjustment parameters, r is a random number, and P c is the control threshold;

[0170] S54. On the basis of the foraging strategy, evaluate the positions of population individuals through the fitness function f(X i ):

[0171] f(X i ) = w 1 ·Coverage(X i ) - w 2 ·Energy(X i ) + w 3 ·Priority(X i );

[0172] Among them, Coverage(X i ) represents the area of the detection region covered by the individual, Energy(X i ) is the path energy consumption, Priority(X i ) is the priority weighted value of the area covered by the individual's path, and w 1 , w 2 , w 3 are weight factors;​

[0173] S55. Select the current optimal path node set according to the fitness function value to generate a preliminary detection path P covering the detection area global :

[0174] P global ={X start ,X 1 ,X 2 ,…,X end};

[0175] Among them, X start and X end They are the initial node and the target node respectively, and the path nodes are dynamically updated through the sparrow foraging strategy;

[0176] S56. Control the aerial platform along the preliminary detection path P global Perform real-time data collection, including multi-modal environmental data and termite activity characteristic data of the target detection area.

[0177] In this implementation, S6 includes the following specific steps:

[0178] S61. Based on the preliminary detection path P global And the priority distribution map P(x,y) is used to extract the high priority area set R covered by the preliminary detection path high :

[0179] R high ={(x,y)|P(x,y)≥P high And (x,y)∈P global};

[0180] Among them, P high is the high priority threshold;

[0181] S62. Set the high priority area R high As the local optimization target area, initialize the population individual positions of the sparrow search algorithm In R high The internal distribution is uniform, and the position initialization rule is:

[0182]

[0183] S63. Update the position of individuals in the population according to the local optimization strategy of the sparrow search algorithm

[0184] in, and Represent the current local optimal position and the worst position respectively, γ and δ are local optimization adjustment parameters, r is a random number, Pc is the control threshold;

[0185] S64. Define the local optimization fitness function f local (X i ), and conduct a comprehensive evaluation by combining the fractal dimension and the detection area priority value:

[0186]

[0187] where P(x i , y i ) is the priority value of the current individual position, and w 4 and w 5 are weight factors;

[0188] S65. According to the fitness function value, select the high-fitness individuals as local path nodes to generate the optimized high-priority area path P local :

[0189] P local = {X start , X h1 , X h2 , …, X end};

[0190] where X h1 , X h2 , … are the optimized path nodes in the high-priority area.

[0191] In this embodiment, S7 includes the following specific steps:

[0192] S71. Based on the optimized high-priority area path P local , conduct obstacle conflict detection on the path by combining the obstacle distribution matrix O(x, y), and define the conflict detection function:

[0193]

[0194] where C(X) is the path conflict amount. If C(X)>0, there is an obstacle conflict in the path and the path needs to be readjusted;

[0195] S72. For the set of path nodes X conflict with obstacle conflicts, redefine the set of feasible path nodes X feasible ;

[0196] S73. Reassign the path priority weights on the generated set of feasible path nodes X feasible according to the priority distribution map P(x, y);

[0197] S74. Regenerate the optimized paths of the aerial platform and the ground platform according to the updated priority weights:

[0198] P updated = {X start , X u1 , X u2 , …, X end};

[0199] Among them, X u1 , X u2 , … are the updated feasible path nodes;

[0200] S75. Based on the updated optimized path P updated between the aerial platform and the ground platform, allocate the path nodes to the aerial platform and the ground platform, and determine the detection range of each platform through the following task allocation function M(X, P k ):

[0201]

[0202] Among them, P k is the type of the detection platform;

[0203] S76. Output the updated path P updated and the platform allocation result M(X, P k ), so that the path avoids obstacle conflicts and preferentially covers high-priority areas.

[0204] Example 1:

[0205] In a field test in August 2024, the research team conducted a termite detection test at an agricultural experimental base on the edge of a tropical rainforest in the south. The area of the base is about 3,000 square meters, including a wooden warehouse, an orchard, and multiple irrigation ditches. Previously, the base manager suspected that termites had invaded the warehouse foundation and damaged the fruit tree roots, but traditional manual detection methods failed to find the specific termite activity locations. The experimental team decided to use the method of the present invention for detection to verify its actual feasibility and effectiveness.

[0206] At 6 o'clock, the research team began to deploy the detection equipment. A drone equipped with a thermal imaging sensor and a lidar (aerial platform) and two ground robots equipped with micro cameras and vibration sensors (ground platforms) were placed in the base respectively. By collecting the initial environmental data, the thermal imaging data in the flight coverage area of the drone generated a thermal distribution map, marking the temperature anomaly points, including the north of the orchard and near the east foundation of the warehouse.

[0207] Based on the thermal imaging data, the system generated a fractal space model using fractal modeling analysis. The results showed that the fractal dimension of the foundation on the east side of the warehouse was 2.47, the fractal dimension of the northern area of the orchard was 2.32, and the fractal dimension near the irrigation ditch was only 2.10. The priority distribution map further divided the foundation on the east side of the warehouse into a high-priority area, the northern part of the orchard into a medium-priority area, and the irrigation ditch into a low-priority area.

[0208] The drone first started to execute the task along the generated preliminary path. At 9:45 am, the system detected an abnormal temperature fluctuation at a point (coordinates "23.545N, 113.875E") in the northern part of the orchard. Combining real-time thermal imaging analysis, the temperature at this point was 5 degrees Celsius higher than the surrounding area, indicating the possible existence of a termite nest. During the same period, a path node (coordinates "23.543N, 113.876E") on the monitoring path of the foundation on the east side of the warehouse triggered the vibration sensor signal. The data record showed that the vibration frequency of the node abnormally increased to 15 Hz, much higher than the background vibration value of other areas of the foundation. The system automatically generated a preliminary detection report, recording the time, location, and characteristic values of the temperature and vibration abnormal points.

[0209] Subsequently, the system performed local path optimization on the high-priority area according to the optimized path. At 10:30 am, the drone completed the refined detection after path adjustment on the foundation on the east side of the warehouse, and the ground robot simultaneously carried out further exploration near the abnormal point. The detection report showed that within a 1-meter range around the abnormal point, the vibration frequency further increased to 18 Hz, and there was a significant change in the moisture content of the foundation soil, with the humidity more than 3 times higher than the background value.

[0210] In the northern part of the orchard, the drone detected a thermal anomaly path 30 centimeters away from the fruit tree roots. Through the path optimization algorithm, it was confirmed that the change rate of the thermal distribution gradient around this path reached 3 degrees Celsius per meter. The ground robot further explored this area and finally found an active termite nest at the end of the thermal anomaly path (coordinates "23.546N, 113.874E"). The record showed that the depth of the nest was about 15 centimeters, the diameter was about 30 centimeters, and it contained a large number of active termite individuals.

[0211] After the detection task was completed, the system automatically generated a comprehensive detection report, which recorded 9 termite activity points, including 3 main nests and 6 activity trace points. The specific detection data is as follows:

[0212] Foundation on the east side of the warehouse: The detection time was from 9:00 to 10:30. Two active points were found, with vibration frequencies of 18 Hz and 15 Hz respectively, 4 humidity abnormal points, and 1 termite nest, and the positioning error was less than 10 centimeters.

[0213] Northern area of the orchard: The detection time was from 9:30 to 11:00. 3 temperature anomaly points, 1 heat path, and 2 termite nests were found. The error of the heat distribution anomaly points was less than 8 cm.

[0214] Irrigation ditch area: The detection time was from 11:00 to 11:30. No abnormal activity points were found.

[0215] To verify the effectiveness of the present invention, the team simultaneously used traditional manual detection means to detect the same area, and the comparison results are as follows:

[0216] Index The method of the present invention Traditional method Detection time (hours) 5 8 Detection range coverage rate (%) 98 75 Detection accuracy (nest location error) ≤10 cm ≥30 cm Number of termite activity points found (pcs) 9 4 Resource utilization rate (unmanned aerial vehicle and robot task balance rate) 92 60

[0217] It can be seen from this embodiment that the present invention can achieve efficient and accurate termite detection through dynamic fractal modeling and sparrow search algorithm in a complex environment. It is significantly superior to traditional methods in terms of coverage rate, detection time, accuracy, and resource utilization, further verifying its technical feasibility and application value.

[0218] The present invention uses a fractal modeling method to calculate the fractal dimension of the detection area, establish a complexity characterization model of the target area, and can dynamically determine the priority weights of high-complexity areas and low-complexity areas through a priority division mechanism, and give priority to processing high-priority areas in path planning. It can dynamically adjust the priority distribution map according to the real-time collected data, making the detection of key areas in a complex environment more targeted, and significantly improving the comprehensiveness of detection coverage and detection accuracy.

[0219] The present invention introduces the sparrow search algorithm in path planning, generates a preliminary detection path through a global foraging strategy and combines a local optimization strategy to further optimize the path node layout in high-priority areas, ensuring that the aerial platform and the ground platform can cooperate efficiently when performing tasks. The dynamic path optimization mechanism performs excellently in path conflict avoidance and priority area coverage, improving the path planning efficiency by more than 20% and effectively reducing resource waste.

[0220] The present invention reallocates path nodes by introducing obstacle detection and dynamic avoidance strategies, combines the functional characteristics of the aerial platform and the ground platform, and adjusts the detection path in real time through the obstacle distribution matrix and the priority distribution map to avoid path conflicts and preferentially cover high-priority areas. At the same time, the high-priority areas are assigned to the aerial platform and the low-priority areas are assigned to the ground platform through a task allocation function, realizing the rational use of resources and the maximization of collaborative work efficiency.

[0221] As described above, it is only the preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. An adaptive path planning method for air-ground collaborative termite detection based on artificial intelligence, characterized in that: The steps include: S1. Collect environmental data sets of the target detection area; S2. Based on the collected environmental data set, a fractal space model of the detection area is constructed using a fractal modeling method; S3. Prioritize the target detection area according to the fractal space model, determine the priority weight, and generate a priority distribution map of the target detection area; S4. Initialize the population position and population size of the sparrow search algorithm according to the priority distribution map of the detection area; S5. Using the foraging strategy of the sparrow search algorithm, the aerial platform performs global path planning in the fractal space model to generate a preliminary detection path covering the detection area; S6. Based on the priority distribution map generated by the preliminary detection path combined with fractal modeling, the local optimization strategy of the sparrow search algorithm is used to optimize the path planning of the high priority area; S7. Adaptively adjust the detection paths of the aerial platform and the ground platform based on the optimized detection path, so that the updated detection path avoids obstacle conflicts and preferentially covers high-priority areas within the detection area; S8. According to the adjusted path planning scheme, the aerial platform is responsible for performing a global scan of the target detection area, and the ground platform is responsible for performing detection of the target detection area.

2. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S1 comprises the following specific steps: S11. Use multimodal sensors to collect environmental data of the target detection area, including terrain information, multimodal characteristic data of termite activities, thermal imaging data, and obstacle distribution data; S12. Digitally process the terrain information to generate a digital elevation model T(x, y) representing the terrain features of the detection area, where T(x, y) represents the elevation value of any point (x, y) in the target detection area, and is used to mark the undulating features of the detection area; S13. Collect multimodal feature data of termite activities, use a two-dimensional image data set V generated by a visual sensor, and use temperature distribution data H(x,y) generated by a thermal imaging sensor; S14. Generate an obstacle distribution matrix O(x, y) using the sensor, where O(x, y) indicates whether there is an obstacle at the point (x, y), and the value is a binary variable: O(x, y) = 1 indicates the existence of an obstacle, and O(x, y) = 0 indicates no obstacle; S15. The terrain information T(x, y), the two-dimensional image dataset V of termite activities and the temperature distribution data H(x, y), and the obstacle distribution data O(x, y) are integrated to construct the environment dataset D: D={T(x,y),V,H(x,y),O(x,y)}.

3. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S2 comprises the following specific steps: S21. Dynamically partition the target detection area based on the environmental data set D, and use an adaptive grid partitioning method based on thermal feature gradient and obstacle distribution to divide the detection area into several non-uniform sub-areas R i , i = 1, 2, ..., n, where the grid size of the sub-region depends on the rate of change of the thermal distribution gradient and obstacle density gradient Dynamic adjustment, the grid division rules are: Among them, Δx, Δy represent the side length of the sub-region grid; S22. Introduce a dynamically adjusted grid scale ∈ in each sub-region k , k=1,2,…,m: ∈ k =∈ min ·(1+a k ·G h +b k ·G o ); Among them, ∈ min is the minimum grid size, α k and β k is the influence coefficient of thermal characteristics and obstacle characteristics; Calculate each grid size ∈ k The number of effective covering grids N(∈ k ), and obtain the fractal dimension: S23. According to the fractal dimension D f (R i ) classify the sub-regions according to their complexity, and divide the detection region into three categories: high complexity region, medium complexity region and low complexity region. The classification threshold is determined by clustering analysis of the statistical distribution of the fractal dimension; S24. Introducing probability weight W based on fractal dimension for high complexity regions i For priority assignment: Among them, W i Represents sub-region R i Priority weight in the entire detection area; S25. Combine the mission execution characteristics of the air platform and the ground platform to set the priority weight W i Mapped to the task allocation matrix M: M={(R i ,P k ):R i →P k ,where P k ∈{air platform, ground platform}}; Among them, P k It represents the execution subject of air-ground collaborative detection. The allocation in the matrix is ​​completed according to the priority weight and the task adaptability of the platform; S26. Output the fractal space model F, which includes the fractal dimension, complexity classification and task allocation matrix of the sub-region.

4. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S3 comprises the following specific steps: S31. Based on the fractal space model F, each sub-region R i The fractal dimension D f (R i ) is normalized; S32. According to the normalized fractal dimension Calculate the priority weight of each sub-region; S33. Combine the priority weights and the detection task requirements to map the priority weights to the spatial distribution of the detection area, and generate a priority distribution map P(x, y) of the target detection area: Among them, P(x,y) represents the priority value of the midpoint (x,y) in the target detection area. Sub-region R i The indicator function is that when the point (x, y) belongs to the subregion R i The value is 1 when it is, otherwise the value is 0; S34. The priority distribution map P(x, y) is classified and processed, and the detection area is divided into a high priority area, a medium priority area and a low priority area according to the priority value. The classification is based on setting a threshold P by statistically analyzing the distribution characteristics of the priority value P(x, y). high and P low To divide: When P(x,y)≥P high When , the point (x, y) belongs to the high priority area; When P low ≤P(x,y) <P high When , the point (x, y) belongs to the medium priority area; When P(x,y) <P low , the point (x, y) belongs to the low priority area.

5. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S4 comprises the following specific steps: S41. Divide the target detection area into several priority areas Q based on the priority distribution map P(x, y) k , priority area Q k Contains all spatial points (x,y) of the same priority: in, and is the priority value range of the kth priority area; S42. According to the priority area Q k The distribution characteristics of the initialization sparrow search algorithm population size N k : Among them, N total Represents the total population size of the sparrow search algorithm; S43. For each priority area Q k , according to the distribution density of points in the region ρ k Initialize the position of the population individuals, the initial position of the population individuals X i =(x i ,y i ) follows the following density control principle: Among them, |Q k | indicates the priority area Q k The number of points within the population, the individuals of the population are preferentially distributed in the area with high point density; S44. Initialize the properties of the individuals in the population, including the distribution status of the detection resources of the air platform and the ground platform. The detection resource properties include the platform type and the initial resource amount of the platform; S45. Output the initialized population information as the initial condition for the path planning of the sparrow search algorithm, including the population size, population location, platform type and resource status.

6. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S5 comprises the following specific steps: S51. Constructing a foraging model of the sparrow search algorithm based on the initialized population information; S52. According to the priority area Q k and priority weight W k Assign an initial position to each individual in the population The initial position must be in a high-priority sub-area within the priority area; S53. Update the position of individual populations according to the foraging strategy of the sparrow search algorithm in, and Represent the current global optimal position and worst position respectively, α and β are adjustment parameters, r is a random number, P c To control the threshold; S54. Based on the foraging strategy, the fitness function f(X i ) evaluates the position of individuals in the population: f(X i )=w1·Coverage(X i )-w2·Energy(X i )+w3·Priority(X i ); Among them, Coverage(X i ) represents the detection area covered by the individual, Energy(X i ) is the path energy consumption, Priority(X i ) is the priority weighted value of the area covered by the individual path, w1, w2, w3 are weight factors; S55. Select the current optimal path node set according to the fitness function value to generate a preliminary detection path P covering the detection area global : P global {X start ,X1,X2,…,X end }; Among them, X start and X end They are the initial node and the target node respectively, and the path nodes are dynamically updated through the sparrow foraging strategy; S56. Control the aerial platform along the preliminary detection path P global Perform real-time data collection, including multi-modal environmental data of the target detection area and termite activity characteristic data.

7. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S6 comprises the following specific steps: S61. Based on the preliminary detection path P global And the priority distribution map P(x,y) is used to extract the high priority area set R covered by the preliminary detection path high : R high = {(x, y) | P(x, y) ≥ P high and (x, y) ∈ P global}; Among them, P high is the high priority threshold; S62. Set the high priority area R high As the local optimization target area, initialize the population individual positions of the sparrow search algorithm In R high The internal distribution is uniform, and the position initialization rule is: S63. Update the position of individuals in the population according to the local optimization strategy of the sparrow search algorithm in, and Represent the current local optimal position and the worst position respectively, γ and δ are local optimization adjustment parameters, r is a random number, P c To control the threshold; S64. Define the local optimization fitness function f local (X i ), combined with the fractal dimension and the detection area priority value for comprehensive evaluation: Among them, P(x i ,y i ) is the priority value of the current individual position, w4 and w5 are weight factors; S65. According to the fitness function value, select high fitness individuals as local path nodes to generate optimized high priority regional path P local : P local ={X start ,X h1 ,X h2 ,…,X end }; Among them, X h1 ,X h2 ,…are the optimized path nodes in the high priority area.

8. The method for adaptive path planning for air-ground collaborative termite detection based on artificial intelligence according to claim 1 is characterized in that: The S7 comprises the following specific steps: S71. Based on the optimized high priority area path P local Combined with the obstacle distribution matrix O(x,y), the path is subjected to obstacle conflict detection, and the conflict detection function is defined as: Where C(X) is the path conflict amount. If C(X)>0, there is an obstacle conflict on the path and the path needs to be readjusted. S72. Path node set X with obstacle conflicts conflict , redefine the feasible path node set X feasible ; S73. In the generated feasible path node set X feasible Redistribute the path priority weights according to the priority distribution graph P(x,y); S74. Regenerate the optimized paths of the air platform and the ground platform according to the updated priority weights: P updated ={X start ,X u1 ,X u2 ,…,X end }; Among them, X u1 ,X u2 ,…are the updated feasible path nodes; S75. Optimize the path P between the updated air platform and the ground platform updated Based on this, the path nodes are allocated to the aerial platform and the ground platform through the following task allocation function M(X,P k )Determine the detection scope of each platform: Among them, P k is the detection platform type; S76. Output the updated path P updated and platform allocation result M(X,P k ), so that the path avoids obstacle conflicts and preferentially covers high-priority areas.

Images