Large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and unmanned aerial vehicle

Abstract

The invention discloses a large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and unmanned aerial vehicles, which comprises an unmanned aerial vehicle scheduling system and a data analysis system, and is characterized in that the unmanned aerial vehicle scheduling system is mainly used for realizing full-coverage and high-efficiency image data acquisition for any complex terrain. The system has the beneficial effects that the unmanned aerial vehicle scheduling system and the data analysis system are matched to collect and analyze large-scale agriculture, forestry and animal husbandry data, the data collection efficiency is remarkably improved through cooperative operation of multiple unmanned aerial vehicles, and an intelligent scheduling algorithm ensures optimal utilization of electric quantity; meanwhile, during data acquisition, the system can adapt to complex terrain environments and support various agriculture, forestry and animal husbandry application scenes; furthermore, the data analysis quality is improved, seamless coverage is realized through an image merging technology, and the large-scale data processing efficiency is improved through dividing and conquering processing; in addition, repeated calculation and recognition are avoided through intelligent result merging, the operation cost is reduced, and the resource utilization efficiency is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of large-scale agriculture, forestry and animal husbandry data analysis, and in particular to a large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and unmanned aerial vehicles. Background Art

[0002] With the development of agricultural modernization, the collection and analysis of large-scale agricultural, forestry and animal husbandry data has become increasingly important. It is the only way to achieve digital and intelligent transformation of agriculture, forestry and animal husbandry, which helps to reduce the workload of farmers and herdsmen, increase safety management efficiency, and assist government decision-making. Typical application scenarios include forest fire prevention, harvest estimation, livestock positioning, etc. Automated large-scale data analysis requires full coverage, high resolution, and high-precision data collection of complex target terrain.

[0003] The existing technology mainly has the following problems:

[0004] 1. Traditional manual data collection methods are time-consuming and labor-intensive, with limited coverage and potential safety hazards;

[0005] 2. Solutions based on satellite images have the limitations of low resolution and difficulty in updating data in real time;

[0006] 3. Existing data analysis methods are mainly targeted at a single scenario and lack flexibility and versatility;

[0007] 4. There are efficiency issues such as repeated calculations and repeated target identification during the data analysis process. Summary of the invention

[0008] The purpose of the present invention is to provide a large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones in order to solve the above problems.

[0009] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0010] A large-scale agricultural, forestry and animal husbandry data analysis system based on artificial intelligence and drones, including a drone dispatching system and a data analysis system, wherein the drone dispatching system is mainly used to achieve full coverage and high-efficiency image data collection for any complex terrain; the data analysis system is mainly based on the data collected by the drone to achieve high-precision and high-flexibility data extraction.

[0011] Furthermore, the information that needs to be input into the drone dispatch system includes: a map of the data collection target area, including the boundaries, orientation, size, etc. of the area; the location of the drone dispatch station; drone performance, including speed, endurance, charging speed, etc.; the number and ID of drones.

[0012] Furthermore, the output of the drone dispatching system includes: the total duration and total number of trips of the entire data collection process; the drone ID, take-off time, and return time of each trip; and the flight route of each trip.

[0013] Furthermore, the workflow of the drone dispatching system is as follows:

[0014] Process 1: Discretization of target area points. According to the definition requirements of the drone photos, determine the flight altitude of the drone and the coverage area of ​​each photo. According to the coverage area, divide the target area into multiple evenly distributed shooting points to ensure that the area represented by all points can cover the entire target area.

[0015] Process 2: Point connection graph construction, connecting adjacent target points to form a dense graph. According to the altitude difference between the points, obstacle distribution, etc., the flight distance between the points is obtained as the weight of the edge in the graph, and finally an undirected fully connected graph is formed;

[0016] Process 3: Split the point connection graph into several non-overlapping subgraphs, so that the sum of the path lengths in each subgraph is slightly less than the longest flight distance that the drone can support (excluding the shortest flight distance between the drone dispatch station and the area). Each subgraph corresponds to a drone flight trip.

[0017] Process 4: Itinerary path planning. For each trip, the drone aims to traverse all nodes in the subgraph corresponding to the trip in the shortest time. This problem is equivalent to the traveling salesman problem and can be solved by methods such as greedy method and dynamic programming. After the algorithm solves it, the specific flight path corresponding to the trip is generated. Because the sum of all paths in each subgraph is guaranteed to be less than the longest flight distance of the drone during the segmentation process, the flight path of each trip can also be guaranteed to be completed by a fully charged drone.

[0018] Process 5: Dynamic scheduling of itineraries. The specific itinerary schedule is dynamically determined during the actual operation of the system. At the initial time t0, all itineraries are "pending". The drone dispatch station sends out drones in turn. Each drone selects a itinerary to execute and sets the corresponding itinerary to the "executing" state. Itineraries in the "executing" state will not be selected repeatedly. After a drone itinerary ends, it enters the dispatch station for charging. A normally completed itinerary is marked as "completed", and an abnormally completed itinerary is marked as "failed", and the route is re-planned for the itinerary. After the drone is fully charged, the next "pending" or "failed" itinerary is selected for execution until all itineraries are completed.

[0019] Further, the workflow of the data analysis system is as follows:

[0020] Process 1: Global image merging. First, based on the fact that each photo collected by the drone should contain GPS positioning information, the positioning information is matched with the map of the target area to obtain photos of each location point in the target area. Secondly, by searching for the possible rotation angle and small translation distance of each photo, the optimal matching method of adjacent photos is found so that the content of the overlapping parts of each pair of adjacent photos is consistent. Finally, the overlapping parts of the adjacent photos are deleted to obtain a global image of the target terrain.

[0021] Process 2: Divide and conquer intelligent processing. The global image of the target terrain is segmented into multiple non-overlapping blocks, and each block is input into a multimodal large model for processing. For example, for a forest fire prevention scenario, each image is used as input, and a prompt word "Is there fire or smoke in the image?" is added. The image question answering large model (a pre-trained model that receives images and text questions as input and outputs answers) outputs the result. For different application scenarios (such as crop monitoring and animal behavior analysis), a professional prompt word library is used to improve the accuracy of image question answering large models.

[0022] Process three: intelligent merging of results. The results generated by the model for each block can be merged to obtain the global results for the target area. Different merging methods are required for different types of data. For example, for disaster warning analysis, the results of each block are directly spliced ​​together, and the location and type of the alarm are finally output. For data statistics analysis (such as harvest estimation), the results of each block are accumulated and the global statistical value is output. For target tracking and counting problems (such as livestock positioning), considering that the target to be analyzed may move between different blocks, it is necessary to call the large model again, and input the shooting time and analysis results of the photos of adjacent blocks into the model. The model determines whether there is repeated identification of the target object between adjacent blocks, and deduplicates the analysis results based on this.

[0023] Furthermore, the data analysis system ultimately outputs data analysis results that fully cover the target area and are non-redundant; for scenarios with time-series changes (such as crop growth analysis or fire change capture), dynamic data change trends can be obtained by calling the data collection and data analysis processes of this system multiple times.

[0024] Furthermore, the hardware configuration of the drone dispatching system adopts commercial drones with GPS positioning and high-definition cameras; the dispatching station is set up by setting multiple charging stations according to the size of the target area; the communication system adopts 4G / 5G network to achieve real-time data transmission; the scheduling algorithm parameter settings include: flight altitude (determined according to the shooting clarity requirements), photo overlap rate (not less than 30%), and charging threshold (return to charging when the remaining power is 20%).

[0025] Furthermore, the server configuration of the data analysis system adopts a high-performance GPU server; the image processing parameters include: image resolution (not less than 4K), image stitching accuracy (pixel-level alignment), and large model selection (using a large language model that supports image and text understanding, such as the qwen-vl model).

[0026] The beneficial effects of the present invention are:

[0027] The present invention collects and analyzes large-scale agriculture, forestry and animal husbandry data through the cooperation of the UAV dispatching system and the data analysis system. Not only does the collaborative operation of multiple UAVs significantly improve the data collection efficiency, but the intelligent dispatching algorithm also ensures the optimal use of electricity. At the same time, it can adapt to complex terrain environments during data collection and support a variety of agriculture, forestry and animal husbandry application scenarios. It also improves the quality of data analysis, achieves seamless coverage through image merging technology, and improves the efficiency of large-scale data processing through divide-and-conquer processing. In addition, intelligent result merging avoids repeated calculations and identifications, which not only reduces operating costs but also optimizes resource utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a workflow diagram of a drone dispatching system for a large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones according to the present invention;

[0029] Figure 2 This is an analysis flow chart of a data analysis system in a large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones described in the present invention. DETAILED DESCRIPTION

[0030] A large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and unmanned aerial vehicles comprises an unmanned aerial vehicle dispatching system and a data analysis system, wherein the unmanned aerial vehicle dispatching system is mainly used to realize full coverage and high-efficiency image data collection for any complex terrain; the data analysis system is mainly based on the data collected by the unmanned aerial vehicle to realize high-precision and high-flexibility data extraction, and a large number of high-definition photos of different coordinate positions in the target area can be obtained through the cruise shooting of the unmanned aerial vehicle, and the data analysis system can obtain the desired analysis results by batch processing the photos of the target area. The data analysis system adopts the method of image merging-divide-and-conquer processing-result merging to solve the problems of repeated processing and target dynamic processing in large-scale image data analysis.

[0031] In this embodiment, the information that needs to be input into the drone dispatching system includes: a map of the data collection target area, including the boundaries, orientation, size, etc. of the area; the location of the drone dispatching station; drone performance, including speed, endurance, charging speed, etc.; the number and ID of drones.

[0032] In this embodiment, the content output by the drone scheduling system includes: the total duration and total number of trips of the entire data collection process; the drone ID, take-off time, and return time of each trip; and the flight route of each trip.

[0033] In this embodiment, the workflow of the drone dispatching system is as follows:

[0034] Process 1: Discretization of target area points. According to the definition requirements of the drone photos, determine the flight altitude of the drone and the coverage area of ​​each photo. According to the coverage area, divide the target area into multiple evenly distributed shooting points to ensure that the area represented by all points can cover the entire target area.

[0035] Process 2: Point connection graph construction, connecting adjacent target points to form a dense graph. According to the altitude difference between the points, obstacle distribution, etc., the flight distance between the points is obtained as the weight of the edge in the graph, and finally an undirected fully connected graph is formed;

[0036] Process 3: Split the point connection graph into several non-overlapping subgraphs, so that the sum of the path lengths in each subgraph is slightly less than the longest flight distance that the drone can support (excluding the shortest flight distance between the drone dispatch station and the area). Each subgraph corresponds to a drone flight trip.

[0037] Process 4: Itinerary path planning. For each trip, the drone aims to traverse all nodes in the subgraph corresponding to the trip in the shortest time. This problem is equivalent to the traveling salesman problem and can be solved by methods such as greedy method and dynamic programming. After the algorithm solves it, the specific flight path corresponding to the trip is generated. Because the sum of all paths in each subgraph is guaranteed to be less than the longest flight distance of the drone during the segmentation process, the flight path of each trip can also be guaranteed to be completed by a fully charged drone.

[0038] Process 5: Dynamic scheduling of itineraries. The specific itinerary schedule is dynamically determined during the actual operation of the system. At the initial time t0, all itineraries are "pending". The drone dispatch station sends out drones in turn. Each drone selects a itinerary to execute and sets the corresponding itinerary to the "executing" state. Itineraries in the "executing" state will not be selected repeatedly. After a drone itinerary ends, it enters the dispatch station for charging. A normally completed itinerary is marked as "completed", and an abnormally completed itinerary is marked as "failed", and the route is re-planned for the itinerary. After the drone is fully charged, the next "pending" or "failed" itinerary is selected for execution until all itineraries are completed.

[0039] In this embodiment, the workflow of the data analysis system is as follows:

[0040] Process 1: Global image merging. First, based on the fact that each photo collected by the drone should contain GPS positioning information, the positioning information is matched with the map of the target area to obtain photos of each location point in the target area. Secondly, by searching for the possible rotation angle and small translation distance of each photo, the optimal matching method of adjacent photos is found so that the content of the overlapping parts of each pair of adjacent photos is consistent. Finally, the overlapping parts of the adjacent photos are deleted to obtain a global image of the target terrain.

[0041] Process 2: Divide and conquer intelligent processing. The global image of the target terrain is segmented into multiple non-overlapping blocks, and each block is input into a multimodal large model for processing. For example, for a forest fire prevention scenario, each image is used as input, and a prompt word "Is there fire or smoke in the image?" is added. The image question answering large model (a pre-trained model that receives images and text questions as input and outputs answers) outputs the result. For different application scenarios (such as crop monitoring and animal behavior analysis), a professional prompt word library is used to improve the accuracy of image question answering large models.

[0042] Process three: intelligent merging of results. The results generated by the model for each block can be merged to obtain the global results for the target area. Different merging methods are required for different types of data. For example, for disaster warning analysis, the results of each block are directly spliced ​​together, and the location and type of the alarm are finally output. For data statistics analysis (such as harvest estimation), the results of each block are accumulated and the global statistical value is output. For target tracking and counting problems (such as livestock positioning), considering that the target to be analyzed may move between different blocks, it is necessary to call the large model again, and input the shooting time and analysis results of the photos of adjacent blocks into the model. The model determines whether there is repeated identification of the target object between adjacent blocks, and deduplicates the analysis results based on this.

[0043] In this embodiment, the data analysis system ultimately outputs data analysis results that fully cover the target area and are non-redundant; for scenarios with time-series changes (such as crop growth analysis or fire change capture), dynamic data change trends can be obtained by calling the data collection and data analysis processes of this system multiple times.

[0044] In this embodiment, the hardware configuration of the drone dispatching system adopts a commercial drone with GPS positioning and a high-definition camera; the dispatching station is set to set multiple charging stations according to the size of the target area; the communication system adopts 4G / 5G network to achieve real-time data transmission; the scheduling algorithm parameter settings include: flight altitude (determined according to the shooting clarity requirements), photo overlap rate (not less than 30%), and charging threshold (return to charging when the remaining power is 20%).

[0045] In this embodiment, the server configuration of the data analysis system is to use a high-performance GPU server; the image processing parameters include: image resolution (not less than 4K), image stitching accuracy (pixel-level alignment), and large model selection (using a large language model that supports image and text understanding, such as the qwen-vl model).

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A large-scale agricultural, forestry and animal husbandry data analysis system based on artificial intelligence and drones, characterized by: It includes a drone dispatching system and a data analysis system, wherein the drone dispatching system is mainly used to achieve full coverage and high-efficiency image data collection for any complex terrain; the data analysis system is mainly based on the data collected by the drone to achieve high-precision and high-flexibility data extraction.

2. According to claim 1, a large-scale agricultural, forestry and animal husbandry data analysis system based on artificial intelligence and drones is characterized by: The information that needs to be input into the drone dispatch system includes: a map of the data collection target area, including the boundaries, orientation, size, etc. of the area; the location of the drone dispatch station; drone performance, including speed, endurance, charging speed, etc.; and the number and ID of drones.

3. According to claim 2, a large-scale agricultural, forestry and animal husbandry data analysis system based on artificial intelligence and drones is characterized by: The output of the drone dispatching system includes: the total duration and total number of trips of the entire data collection process; the drone ID, take-off time, and return time of each trip; and the flight route of each trip.

4. A large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones according to any one of claims 1 to 3, characterized in that: The workflow of the drone dispatching system is as follows: Process 1: Discretization of target area points. According to the definition requirements of the drone photos, determine the flight altitude of the drone and the coverage area of ​​each photo. According to the coverage area, divide the target area into multiple evenly distributed shooting points to ensure that the area represented by all points can cover the entire target area. Process 2: Point connection graph construction, connecting adjacent target points to form a dense graph. According to the altitude difference between the points, obstacle distribution, etc., the flight distance between the points is obtained as the weight of the edge in the graph, and finally an undirected fully connected graph is formed; Process 3: Split the point connection graph into several non-overlapping subgraphs, so that the sum of the path lengths in each subgraph is slightly less than the longest flight distance that the drone can support (excluding the shortest flight distance between the drone dispatch station and the area). Each subgraph corresponds to a drone flight trip. Process 4: Itinerary path planning. For each trip, the drone aims to traverse all nodes in the subgraph corresponding to the trip in the shortest time. This problem is equivalent to the traveling salesman problem and can be solved by methods such as greedy method and dynamic programming. After the algorithm solves it, the specific flight path corresponding to the trip is generated. Because the sum of all paths in each subgraph is guaranteed to be less than the longest flight distance of the drone during the segmentation process, the flight path of each trip can also be guaranteed to be completed by a fully charged drone. Process 5: Dynamic scheduling of itineraries. The specific itinerary schedule is dynamically determined during the actual operation of the system. At the initial time t0, all itineraries are "pending". The drone dispatch station sends out drones in turn. Each drone selects a itinerary to execute and sets the corresponding itinerary to the "executing" state. Itineraries in the "executing" state will not be selected repeatedly. After a drone's itinerary ends, it enters the dispatch station for charging. A normally completed itinerary is marked as "completed", and an abnormally completed itinerary is marked as "failed". The route is re-planned for the itinerary. After the drone is fully charged, the next "pending" or "failed" itinerary is selected for execution until all itineraries are completed.

5. The large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones according to claim 1 is characterized by: The workflow of the data analysis system is as follows: Process 1: Global image merging. First, based on the fact that each photo collected by the drone should contain GPS positioning information, the positioning information is matched with the map of the target area to obtain photos of each location point in the target area. Secondly, by searching for the possible rotation angle and small translation distance of each photo, the optimal matching method of adjacent photos is found so that the content of the overlapping parts of each pair of adjacent photos is consistent. Finally, the overlapping parts of the adjacent photos are deleted to obtain a global image of the target terrain. Process 2: Divide and conquer intelligent processing. The global image of the target terrain is segmented into multiple non-overlapping blocks, and each block is input into a multimodal large model for processing. For example, for a forest fire prevention scenario, each block of the image is used as input, and a prompt word "Is there fire or smoke in the image?" is added. The image question answering large model (a pre-trained model that receives images and text questions as input and outputs answers) outputs the result. For different application scenarios (such as crop monitoring and animal behavior analysis), a professional prompt word library is used to improve the accuracy of the image question answering large model processing. Process three: intelligent merging of results. The results generated by the model for each block can be merged to obtain the global results for the target area. Different merging methods are required for different types of data. For example, for disaster warning analysis, the results of each block are directly spliced ​​together, and the location and type of the alarm are finally output. For data statistics analysis (such as harvest estimation), the results of each block are accumulated and the global statistical value is output. For target tracking and counting problems (such as livestock positioning), considering that the target to be analyzed may move between different blocks, it is necessary to call the large model again, and input the shooting time and analysis results of the photos of adjacent blocks into the model. The model determines whether there is repeated identification of the target object between adjacent blocks, and deduplicates the analysis results based on this.

6. The large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones according to claim 5 is characterized by: The data analysis system ultimately outputs data analysis results that fully cover the target area and are non-redundant; for scenarios with time-series changes (such as crop growth analysis or fire change capture), dynamic data change trends can be obtained by calling the data collection and data analysis processes of this system multiple times.

7. The large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones according to claim 1 is characterized by: The hardware configuration of the drone dispatch system adopts a commercial drone with GPS positioning and a high-definition camera; the dispatch station is set up to set up multiple charging stations according to the size of the target area; The communication system uses 4G / 5G network to achieve real-time data transmission; the scheduling algorithm parameter settings include: flight altitude (determined according to the shooting clarity requirements), photo overlap rate (not less than 30%), and charging threshold (return to charging when the remaining power is 20%).

8. The large-scale agriculture, forestry and animal husbandry data analysis system based on artificial intelligence and drones according to claim 1 is characterized by: The server configuration of the data analysis system is a high-performance GPU server; the image processing parameters include: image resolution (not less than 4K), image stitching accuracy (pixel-level alignment), and large model selection (using a large language model that supports image and text understanding, such as the qwen-vl model).

Images