Method and system for evaluating fertility of tobacco field based on remote sensing of unmanned aerial vehicle

Abstract

The invention discloses a method and a system for evaluating fertility of a tobacco field based on unmanned aerial vehicle remote sensing, and relates to the technical field of surveying and mapping service. Autonomous decision making is carried out at each node based on local information through a distributed game decision-making mechanism, the nodes do not follow a preset fixed path any more, but a flight decision is modeled as a non-cooperative game process, so that the fertility of the tobacco field is evaluated. The nodes are driven to preferentially fly to the grid area with the high information entropy value, when the multiple nodes fly to the same high-value area at the same time, the income reduction caused by mutual action is prejudged through game deduction, so that the nodes spontaneously and dispersedly negotiate and turn to other high-entropy areas which are not explored, and the dynamic path planning based on the real-time game has the advantages that the dynamic path planning efficiency is improved. According to the invention, the cluster behavior is converted from mechanical execution to intelligent planning, so that the non-repeated efficient scanning of the tobacco field range is realized under the condition of no central scheduling, and the problem of stiffness of a fixed path and the efficiency bottleneck of full-coverage acquisition are solved.

Description

Technical Field

[0001] This invention relates to the field of surveying and mapping services, specifically to a method and system for assessing tobacco field fertility based on unmanned aerial vehicle (UAV) remote sensing. Background Technology

[0002] Precision agriculture is a core direction of modern agricultural development, and its key lies in the quantitative management of the spatiotemporal variability of field environmental variables. In tobacco cultivation, soil fertility is the primary factor determining tobacco yield and quality. Traditional methods of manual sampling and laboratory analysis are time-consuming, labor-intensive, costly, and difficult to implement large-scale simultaneous assessments, severely hindering the implementation of precision fertilization. Unmanned aerial vehicle (UAV) remote sensing technology, with its flexibility and efficiency, has become an important means of rapidly acquiring crop growth information. By equipping itself with multispectral sensors, UAVs can indirectly infer the state of vegetation canopy, thus providing a data foundation for assessing the spatial distribution of soil fertility. However, how to utilize this technology to achieve efficient, accurate, and comprehensive assessment of tobacco field fertility still faces a series of technical challenges, and the limitations of existing methods urgently need to be overcome.

[0003] Existing drone-based tobacco field fertility assessment schemes have significant shortcomings in multi-drone collaboration and intelligent path planning. First, the commonly used single-drone operation mode becomes a bottleneck in data collection efficiency when dealing with large-scale tobacco fields, making it difficult to complete full-coverage scanning within the short agricultural time window, resulting in delayed fertility assessment results. Second, even when using multiple drones, their path planning is mostly based on preset, fixed parallel routes, lacking real-time and adaptive capabilities. This rigid mode cannot respond to the natural patchy distribution of fertility within the tobacco field, leading to oversampling in low-variability areas and wasted resources, while undersampling in high-variability areas may affect assessment accuracy. Furthermore, existing collaborative control relies heavily on centralized scheduling systems, with a central controller uniformly calculating and allocating tasks. This architecture carries the risk of single-point failure; if the central node fails due to communication delays, interference, or its own malfunction, the entire cluster's operation will be paralyzed.

[0004] In summary, existing technologies have significant shortcomings in data collection efficiency and path adaptability when processing large-area tobacco field fertility assessments. Therefore, there is an urgent need in this field for an intelligent assessment scheme that can achieve autonomous collaboration, dynamic response, and independence from central nodes, thereby truly meeting agriculture's requirements for data timeliness and accuracy. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for assessing tobacco field fertility based on UAV remote sensing. This system utilizes a distributed game theory decision-making mechanism, allowing each node to make autonomous decisions based on local information. Nodes no longer follow preset fixed paths but model flight decisions as a non-cooperative game process, driving them to prioritize flying towards grid areas with high information entropy values. When multiple nodes simultaneously fly towards the same high-value area, they predict, through game theory deduction, that their actions will lead to a decrease in returns, thus spontaneously and decentralizedly negotiating and turning towards other unexplored high-entropy areas. This dynamic path planning based on real-time game theory transforms cluster behavior from mechanical execution to intelligent planning, achieving efficient, non-repetitive scanning of the tobacco field without centralized scheduling. This solves the rigidity problem of fixed paths and the efficiency bottleneck of full-coverage data collection.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a method for assessing tobacco field fertility based on unmanned aerial vehicle (UAV) remote sensing, the specific steps of which are as follows: S100. The geographical boundary information of the tobacco field to be measured is sent to a drone cluster with multiple drone nodes. Each drone node initializes the same two-dimensional grid map representing the tobacco field and assigns an initial information entropy value to each grid cell. ; S200: Each of the aforementioned drone nodes flies to its initial collection area and begins to execute a cyclical game-theoretic collection task. Within each decision cycle, each drone node, based on a non-cooperative game strategy and its real-time location and perceived local environmental information, calculates the expected payoff for moving to different candidate areas. And independently decide the flight path for the next moment; S300 and each UAV node fly to the target grid according to the strategy decided by S200, and use airborne remote sensing sensors to collect tobacco field spectral data and update the information entropy value of the corresponding grid cell in the two-dimensional grid map. S400, repeat S200 to S300 until the information entropy value of all grid cells in the entire two-dimensional grid map is lower than the preset threshold, indicating that the tobacco field data collection has been fully covered and the drone cluster terminates the mission. The S500 integrates, stitches, and processes the spectral data collected by all UAV nodes, and generates the final tobacco field fertility distribution map through a preset fertility inversion model.

[0007] Furthermore, the initialization of the two-dimensional grid map in S100 specifically includes: The total number of grids in the two-dimensional grid map is divided into: ,in For the number of rows, For column number; Initialize an information entropy value for each grid cell. ,in, and These are the row and column indices of the grid, respectively; In the initial state, set for all mesh cells , The maximum value of the preset constant indicates that the grid has not been collected and has the highest information uncertainty.

[0008] Furthermore, in S200, the non-cooperative game strategy specifically includes: Each drone node is treated as a game participant. Each drone node obtains its own real-time position and uses its remote sensing sensor to perceive the scanning status of the area within its field of view, identifying the first type of scanned area and the second type of unscanned area. Define a common utility function for all drone nodes to maximize the overall data coverage integrity of the entire 2D grid map; The drone node discretizes the second type of region within its field of vision into multiple candidate grids. For each candidate grid, the drone node calculates the expected benefit it can obtain by choosing to fly to the candidate grid based on non-cooperative game theory. Each UAV node selects the candidate grid with the highest expected return from all candidate grids as the target point for the next moment, and generates a flight path from its real-time location to the target point.

[0009] Furthermore, the expected returns ,in Indicates the first Each drone node took action Expected returns at that time Indicates except the first The set of actions taken by other drone nodes in the cluster, excluding the individual node. Target area Inner mesh cell The current information entropy value, The cost coefficient represents the cost per unit flight distance. It is the first Real-time location of each drone node to the target area The Euclidean distance.

[0010] Furthermore, in S300, through Update the information entropy value of the corresponding grid cell in the 2D grid map, where, Represents the updated grid Information entropy value, Represents a grid The initial information entropy value, The information attenuation coefficient, in its physical sense, represents the degree of information uncertainty reduced in a single data acquisition. The number of data collections is the number of times a grid is collected by a single drone node. When the grid is If a node is repeatedly sampled, then The information entropy value update characterizes the grid after it is collected, and its information entropy value decays exponentially, thereby reducing its attractiveness to other nodes in subsequent game decisions.

[0011] Furthermore, the specific steps of S500 are as follows: Based on the fused and stitched full-coverage spectral data, the Normalized Difference Vegetation Index (NDVI) and Soil Modified Vegetation Index (SAVI) were calculated for each grid cell in the tobacco field. For each grid cell, its calculated NDVI value, SAVI value, and geographical location information are combined to construct a feature vector, which serves as the input to the fertility inversion model. The feature vector is input into the preset fertility inversion model, and the fertility inversion model outputs continuous soil fertility index values ​​to characterize the comprehensive fertility level of the grid cell. Based on the preset fertility level threshold, the soil fertility index value of each grid unit is divided into different fertility levels, and a visualized tobacco field fertility level distribution map is generated based on the geographical location information of all grid units.

[0012] Furthermore, the fertility inversion model is a convolutional neural network model based on deep learning, and its construction process is as follows: The multispectral images acquired by the UAV are cropped into fixed image patches as network input, and the corresponding labels are the measured fertility values ​​of the central region of the image patch. The convolutional neural network model includes alternating convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to extract fertility-related spatial-spectral features from multispectral image patches, the pooling layers are used to reduce the feature map size and enhance the translation invariance of the features, and the fully connected layers map the extracted high-level features to the final fertility value output. The network is trained using labeled image patches, and the loss function between the predicted fertility value and the actual fertility value is minimized using the backpropagation algorithm. The image of the entire tobacco field is segmented into overlapping image blocks, which are then input into a convolutional neural network model for prediction. The prediction results are then stitched together to generate a tobacco field fertility distribution map.

[0013] Furthermore, in S500, the fusion, stitching, and processing of spectral data collected by all UAV nodes specifically includes: performing radiometric calibration and atmospheric correction on the data blocks independently collected by each UAV node to eliminate sensor differences and the influence of ambient light; using the POS data of each data block, i.e., position and attitude system data, to perform geometric correction; and using a registration algorithm based on SIFT feature point matching to align and stitch the overlapping areas of adjacent data blocks to form an orthophoto remote sensing image covering the entire tobacco field.

[0014] On the other hand, a system for assessing tobacco field fertility based on unmanned aerial vehicle (UAV) remote sensing includes: The cluster networking and data acquisition module consists of multiple UAV nodes as physical entities. Each UAV node integrates the same remote sensing unit, positioning and navigation unit, and self-organizing network communication unit, which are used to collaboratively perform the physical acquisition of tobacco field spectral data and the interaction of status information between nodes. The game decision and control module performs flight control calculations for each UAV node, which is used to maintain the two-dimensional grid map and information entropy state in real time during flight, perform game deduction, and output strategy commands to control the flight actions of the UAV. The data preprocessing and fusion stitching module includes a radiometric correction unit, a geometric fine correction unit, and a feature matching and registration unit. It is used to standardize and stitch together the original spectral data blocks from different nodes that have overlap, and generate an orthophoto base map covering the entire tobacco field. The fertility inversion and visualization output module has a built-in fertility inversion model, which is used to receive the orthophoto base map, calculate the vegetation index of each pixel and construct a feature vector, calculate the continuous soil fertility index by the fertility inversion model, and generate and render the final tobacco field fertility level distribution map. The task management and human-computer interaction module provides a graphical interface for receiving user-inputted tobacco field geographical boundaries and initialization parameters. It is responsible for distributing the boundary information to the entire UAV cluster, monitoring the status of all nodes in real time, and receiving and displaying the tobacco field fertility level distribution map generated by the fertility inversion and visualization output module.

[0015] Compared with existing technologies, this method for assessing tobacco field fertility based on UAV remote sensing has the following advantages: I. This invention utilizes a distributed game-theoretic decision-making mechanism, enabling each node to make autonomous decisions based on local information. Nodes no longer follow preset fixed paths but instead model flight decisions as a non-cooperative game process, driving nodes to prioritize flying towards grid areas with high information entropy values. When multiple nodes simultaneously fly towards the same high-value area, they predict, through game theory, that their actions will lead to a decrease in returns, thus spontaneously and decentralizedly negotiating and turning to other unexplored high-entropy areas. This dynamic path planning based on real-time game theory transforms cluster behavior from mechanical execution to intelligent planning, thereby achieving efficient scanning of tobacco fields without repetition in the absence of centralized scheduling, solving the rigidity problem of fixed paths and the efficiency bottleneck of full-coverage data collection.

[0016] Second, this invention introduces information entropy to quantify the data value and uncertainty of each grid unit and uses it as a direct driving force for game decision-making. During initialization, the entire field is modeled as a two-dimensional grid map, and each grid is assigned the maximum information entropy value, representing a completely unknown state. This mechanism allows the dynamic changes in information entropy to directly reflect the progress and quality of data collection. During the game, nodes are attracted by high information entropy areas and prioritize exploring areas with the highest uncertainty and value, achieving optimal efficiency in the collection phase. When the information entropy of all grids is below the threshold, it indicates that the spatial variability of fertility in the entire field has been fully captured, thus providing a high-quality data foundation for subsequent fertility inversion and ensuring the accuracy and reliability of the final fertility distribution map from the source.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the operation of a method for assessing tobacco field fertility based on UAV remote sensing. Figure 2 This is a flowchart illustrating the steps of a method for assessing tobacco field fertility based on UAV remote sensing. Detailed Implementation

[0020] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a method for diagnosing transformer faults using multi-channel acoustic imaging and feature enhancement”, “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plurality forms, unless the context clearly indicates otherwise; “plural” generally includes at least two.

[0022] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0023] To address the shortcomings of existing technologies, the present invention first describes the tobacco field fertility assessment scenario. This invention is primarily applied in precision agriculture, particularly in assessing the spatial variability of soil fertility in tobacco cultivation. Traditional manual sampling methods are inefficient, costly, and difficult to implement large-scale simultaneous assessments. Existing UAV remote sensing solutions often employ single-unit operation or fixed-path multi-unit collaboration, resulting in low data collection efficiency, rigid paths, and single-point failures in central scheduling.

[0024] To this end, this application discloses a method and system for assessing tobacco field fertility based on UAV remote sensing. Through distributed game decision-making, information entropy-driven data collection, and intelligent fertility inversion, it achieves efficient, accurate, and adaptive tobacco field fertility assessment.

[0025] Specifically, such as Figure 2 As shown, this invention provides a method for assessing tobacco field fertility based on UAV remote sensing, specifically including the following steps: S100. The geographical boundary information of the tobacco field to be measured is sent to a drone cluster with multiple drone nodes. Each drone node initializes the same two-dimensional grid map representing the tobacco field and assigns an initial information entropy value to each grid cell. ; S200: Each of the aforementioned drone nodes flies to its initial collection area and begins to execute a cyclical game-theoretic collection task. Within each decision cycle, each drone node, based on a non-cooperative game strategy and its real-time location and perceived local environmental information, calculates the expected payoff for moving to different candidate areas. And independently decide the flight path for the next moment; S300 and each UAV node fly to the target grid according to the strategy decided by S200, and use airborne remote sensing sensors to collect tobacco field spectral data and update the information entropy value of the corresponding grid cell in the two-dimensional grid map. S400, repeat S200 to S300 until the information entropy value of all grid cells in the entire two-dimensional grid map is lower than the preset threshold, indicating that the tobacco field data collection has been fully covered and the drone cluster terminates the mission. The S500 integrates, stitches, and processes the spectral data collected by all UAV nodes, and generates the final tobacco field fertility distribution map through a preset fertility inversion model.

[0026] In the specific implementation process, the S100 initializes the drone cluster and grid map. Through the task management and human-machine interaction module, the user inputs the geographical boundary information of the tobacco field to be measured and sets the initialization parameters, including grid size and maximum information entropy. Information such as the number of drone nodes is distributed to each node in the drone swarm via a self-organizing network communication unit. Each drone node initializes the same two-dimensional grid map representing the tobacco field, and the map is divided into... 1 grid cell, of which For the number of rows, For the number of columns, each grid cell It is given an initial information entropy value ,in The preset constant represents the grid that has not been collected, indicating the highest information uncertainty. Information entropy is used here to quantify data value; high-entropy regions represent unexplored or highly variable areas, which are the priority targets for drone collection. The drone swarm adopts a distributed architecture with no central node. Each node shares status information, including location, remaining battery power, and collection progress, through a self-organizing network.

[0027] Each drone node flies to its initial data collection area and begins executing a cyclical game-theoretic data collection task. In each decision cycle, the drone node independently decides its flight path for the next moment based on a non-cooperative game strategy. Specifically, the non-cooperative game strategy includes the following sub-steps: Environmental perception: The UAV uses onboard remote sensing sensors to perceive the scanning status of the area within its field of view, identifying scanned areas (low information entropy) and unscanned areas (high information entropy). At the same time, the UAV obtains its own real-time position through the positioning and navigation unit.

[0028] Candidate region discretization: The unscanned region within the perceptual field of view is discretized into multiple candidate grids, with each grid serving as a potential target point.

[0029] Expected return calculation: for each candidate grid region Expected revenue from drone node computing The payoff function is defined as: ,in, Target area Inner mesh cell The current information entropy value, This is a cost coefficient, representing the cost per unit flight distance, used to balance data collection revenue and flight energy consumption. It is the distance from the current location of the drone node to the target area. The Euclidean distance.

[0030] Decision making and path generation: The UAV selects the candidate grid with the highest expected return as the target point for the next moment and generates the optimal flight path.

[0031] The path generation takes into account obstacle avoidance constraints and flight safety. Through non-cooperative game theory, each node negotiates spontaneously to avoid multiple aircraft repeatedly collecting data in the same area, thereby achieving efficient full coverage.

[0032] The drone swarm repeatedly executes a cycle of decision-making, flight, data collection, and updating. Each drone node independently performs perception, calculation, decision-making, and action in each decision cycle, and asynchronously updates its own map based on the data collection results. Through swarm communication, each node periodically broadcasts the grid it has collected and the updated entropy value, enabling other nodes to synchronously update their own maps and maintain state consistency.

[0033] The system continuously monitors the information entropy value of all grid cells in the entire two-dimensional grid map. When the entropy value of all grid cells is lower than the preset termination threshold, it indicates that the tobacco field data collection has reached the preset full coverage level, and the spatial variability of field fertility has been fully captured. At this time, the task management module sends a termination command to all drone nodes, and the drone cluster terminates the mission and returns to base.

[0034] The raw data collected from the front end is transformed into the final fertility distribution map. Spectral data with POS data collected by all UAV nodes are aggregated into the data preprocessing and fusion stitching module for radiometric calibration and atmospheric correction. This eliminates potential response differences between different sensors and the influence of ambient light variations. The raw DN values ​​are converted into surface reflectance data for geometric correction. High-precision POS data is used to correct each data block, eliminating image distortion caused by flight attitude changes. A registration algorithm based on SIFT feature point matching is employed to accurately align and seamlessly stitch overlapping areas of adjacent data blocks, ultimately forming a high-quality orthophoto remote sensing image covering the entire tobacco field.

[0035] The fertility inversion and visualization output module receives the stitched panoramic image, calculates the Normalized Difference Vegetation Index (NDVI) and Soil-Adjusted Vegetation Index (SAVI) for each pixel in the image, and constructs a feature vector for each grid cell by combining its NDVI value, SAVI value and geographic location information.

[0036] The feature vectors of all the constructed grids are input into the preset fertility inversion model. In this embodiment, the model is a convolutional neural network (CNN) model based on deep learning.

[0037] The orthophoto of the entire tobacco field is spatially divided into multiple image patches. Each image patch and its corresponding geographical location information are fed into a CNN model.

[0038] The convolutional and pooling layers of the CNN model automatically extract deep spatial-spectral features related to soil fertility from multispectral image patches. The fully connected layers map these high-level features to a continuous soil fertility index value, which comprehensively reflects the overall fertility level of the grid cell represented by the central region of the image patch.

[0039] The prediction results of all image patches are stitched together. Based on the preset fertility level threshold, the continuous SFI values ​​of each grid cell are divided into different fertility levels. Combining the geographical location information of all grid cells, a visual and intuitive map of tobacco field fertility level distribution is generated using rendering technology and presented to the user through task management and human-computer interaction interface.

[0040] In its implementation, this method utilizes the intelligent collaboration of drone swarms to efficiently and comprehensively collect data, providing a foundation for accurate assessment. For example... Figure 1 As shown in the figure, the specific implementation steps of the method for assessing tobacco field fertility based on UAV remote sensing proposed in this embodiment are as follows: (1) System initialization and task planning Task parameter settings: Users input the geographical boundary information of the tobacco field to be tested through the human-computer interaction interface and set the task parameters.

[0041] Environmental map initialization: Based on the geographical boundary information, the system logically generates a two-dimensional grid map covering the entire tobacco field.

[0042] Information entropy initialization: Assign an initial information entropy value to each grid cell in the grid map. This value is set to the maximum value, representing that all areas are in an unknown state.

[0043] (2) Cluster self-organizing data collection Cluster deployment and takeoff: Multiple drone nodes fly to their initial positions over the tobacco fields according to the initial allocation plan.

[0044] Local environmental perception: Each UAV node uses its positioning module and remote sensing sensor to obtain its own precise position in real time and perceive the scanning status of the grid within its field of view.

[0045] Game theory decision-making and path planning: Each node discretizes the unscanned region within its field of view into multiple candidate grids.

[0046] Each node independently calculates the expected payoff for flying to each candidate grid based on a non-cooperative game strategy.

[0047] Each node selects the candidate grid that brings the highest expected return as the next target point and plans the flight path.

[0048] Data collection and map updates: The drone node flies to the target grid determined by the decision and uses an airborne multispectral sensor to collect spectral data of the tobacco fields in that area.

[0049] Once grid data is collected, the system immediately updates the information entropy value of that grid, significantly reducing its value to decrease its attractiveness to other nodes and avoid duplicate collection.

[0050] Cyclic data collection and collaborative coverage: The process of repeatedly executing steps from local environmental perception to step data collection and map updates involves each node continuously sensing, making decisions, collecting data, and updating. Through this distributed game, the cluster spontaneously collaborates to gradually cover all unknown areas.

[0051] Mission Termination Determination: The system continuously monitors the entire grid map. When the information entropy value of all grid cells is lower than the preset threshold, it is determined that the tobacco field data collection has achieved full coverage, and the drone swarm is instructed to terminate the mission and return to base.

[0052] (3) Data post-processing and fertility assessment

[0053] Data transmission and stitching: Spectral data collected by all UAV nodes are transmitted back to the ground processing center. The system performs radiometric and geometric corrections on the data, and accurately registers and stitches data blocks with overlapping areas to generate a complete and seamless orthophoto map of the tobacco field.

[0054] Feature extraction and model input: From the stitched orthophoto, vegetation index is calculated for each grid cell, and feature vectors are constructed by combining geographic location information.

[0055] Fertility inversion and mapping: Input the feature vector into the pre-trained fertility inversion model to calculate the continuous soil fertility index for each grid cell.

[0056] Results Visualization: Based on the fertility index values ​​and preset grade thresholds, a final visualized map of tobacco field fertility grade distribution is generated, providing users with intuitive decision support.

[0057] Through the coordinated implementation of the above steps, this invention constructs a complete technology chain from distributed intelligent perception and dynamic collaborative decision-making to precise data inversion. The behavior of the drone swarm transforms from mechanical execution to intelligent emergence, efficiently and adaptively completing the data collection work of a large area of ​​tobacco fields. It also ensures from the source that the final fertility distribution map has high accuracy and high reliability, providing key data support for variable fertilization in precision agriculture.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for assessing tobacco field fertility based on UAV remote sensing, characterized in that, The specific steps of this method are as follows: S100. The geographical boundary information of the tobacco field to be measured is sent to a drone cluster with multiple drone nodes. Each drone node initializes the same two-dimensional grid map representing the tobacco field and assigns an initial information entropy value to each grid cell. ; S200: Each of the aforementioned drone nodes flies to its initial collection area and begins to execute a cyclical game-theoretic collection task. Within each decision cycle, each drone node, based on a non-cooperative game strategy and its real-time location and perceived local environmental information, calculates the expected payoff for moving to different candidate areas. And independently decide the flight path for the next moment; S300 and each UAV node fly to the target grid according to the strategy decided by S200, and use airborne remote sensing sensors to collect tobacco field spectral data and update the information entropy value of the corresponding grid cell in the two-dimensional grid map. S400, repeat S200 to S300 until the information entropy value of all grid cells in the entire two-dimensional grid map is lower than the preset threshold, indicating that the tobacco field data collection has been fully covered and the drone cluster terminates the mission. The S500 integrates, stitches, and processes the spectral data collected by all UAV nodes, and generates the final tobacco field fertility distribution map through a preset fertility inversion model.

2. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 1, characterized in that, The initialization of the two-dimensional grid map in S100 specifically includes: The total number of grids in the two-dimensional grid map is divided into: ,in For the number of rows, For column number; Initialize an information entropy value for each grid cell. ,in, and These are the row and column indices of the grid, respectively; In the initial state, set for all mesh cells , The maximum value of the preset constant indicates that the grid has not been collected.

3. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 1, characterized in that, In S200, the non-cooperative game strategies specifically include: Each drone node is treated as a game participant. Each drone node obtains its own real-time position and uses its remote sensing sensor to perceive the scanning status of the area within its field of view, identifying the first type of scanned area and the second type of unscanned area. Define a common utility function for all drone nodes to maximize the overall data coverage integrity of the entire 2D grid map; The drone node discretizes the second type of region within its field of vision into multiple candidate grids. For each candidate grid, the drone node calculates the expected benefit it can obtain by choosing to fly to the candidate grid based on non-cooperative game theory. Each UAV node selects the candidate grid with the highest expected return from all candidate grids as the target point for the next moment, and generates a flight path from its real-time location to the target point.

4. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 3, characterized in that, The expected returns ,in Indicates the first Each drone node took action Expected returns at that time Indicates except the first The set of actions taken by other drone nodes in the cluster, excluding the individual node. Target area Inner mesh cell The current information entropy value, The cost coefficient represents the cost per unit flight distance. It is the first Real-time location of each drone node to the target area The Euclidean distance.

5. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 1, characterized in that, In S300, through Update the information entropy value of the corresponding grid cell in the 2D grid map, where, Represents the updated grid Information entropy value, Represents a grid The initial information entropy value, The information attenuation coefficient, The number of data collections is the number of times a grid is collected by a single drone node. When the grid is If a node is repeatedly sampled, then .

6. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 1, characterized in that, The specific steps of S500 are as follows: Based on the fused and stitched full-coverage spectral data, the Normalized Difference Vegetation Index (NDVI) and Soil Modified Vegetation Index (SAVI) were calculated for each grid cell in the tobacco field. For each grid cell, its calculated NDVI value, SAVI value, and geographical location information are combined to construct a feature vector, which serves as the input to the fertility inversion model. The feature vector is input into the preset fertility inversion model, and the fertility inversion model outputs continuous soil fertility index values ​​to characterize the comprehensive fertility level of the grid cell. Based on the preset fertility level threshold, the soil fertility index value of each grid unit is divided into different fertility levels, and a visualized tobacco field fertility level distribution map is generated based on the geographical location information of all grid units.

7. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 6, characterized in that, The fertility inversion model is a convolutional neural network model based on deep learning, and its construction process is as follows: The multispectral images acquired by the UAV are cropped into fixed image patches as network input, and the corresponding labels are the measured fertility values ​​of the central region of the image patch. The convolutional neural network model includes alternating convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to extract fertility-related spatial-spectral features from multispectral image patches, the pooling layers are used to reduce the feature map size, and the fully connected layers map the extracted high-level features to the final fertility value output. The network is trained using labeled image patches, and the loss function between the predicted fertility value and the actual fertility value is minimized using the backpropagation algorithm. The image of the entire tobacco field is segmented into overlapping image blocks, which are then input into a convolutional neural network model for prediction. The prediction results are then stitched together to generate a tobacco field fertility distribution map.

8. The method for assessing tobacco field fertility based on UAV remote sensing according to claim 1, characterized in that, In S500, the fusion, stitching and processing of spectral data collected by all UAV nodes specifically includes: performing radiometric calibration and atmospheric correction on the data blocks collected independently by each UAV node; using the POS data of each data block, i.e., position and attitude system data, to perform geometric correction; and using a registration algorithm based on SIFT feature point matching to align and stitch the overlapping areas of adjacent data blocks to form an orthophoto remote sensing image covering the entire tobacco field.

9. A system for assessing tobacco field fertility based on UAV remote sensing, applicable to the method for assessing tobacco field fertility based on UAV remote sensing as described in any one of claims 1-8, characterized in that, The system includes: The cluster networking and data acquisition module consists of multiple UAV nodes as physical entities. Each UAV node integrates the same remote sensing unit, positioning and navigation unit, and self-organizing network communication unit, which are used to collaboratively perform the physical acquisition of tobacco field spectral data and the interaction of status information between nodes. The game decision and control module performs flight control calculations for each UAV node, which is used to maintain the two-dimensional grid map and information entropy state in real time during flight, perform game deduction, and output strategy commands to control the flight actions of the UAV. The data preprocessing and fusion stitching module includes a radiometric correction unit, a geometric fine correction unit, and a feature matching and registration unit. It is used to standardize and stitch together the original spectral data blocks from different nodes that have overlap, and generate an orthophoto base map covering the entire tobacco field. The fertility inversion and visualization output module has a built-in fertility inversion model, which is used to receive the orthophoto base map, calculate the vegetation index of each pixel and construct a feature vector, calculate the continuous soil fertility index by the fertility inversion model, and generate and render the final tobacco field fertility level distribution map. The task management and human-computer interaction module provides a graphical interface for receiving user-inputted tobacco field geographical boundaries and initialization parameters. It is responsible for distributing the boundary information to the entire UAV cluster, monitoring the status of all nodes in real time, and receiving and displaying the tobacco field fertility level distribution map generated by the fertility inversion and visualization output module.

Images