An unmanned aerial vehicle based weed monitoring system and method

Abstract

The present application relates to the field of agricultural automation and intelligent control technology, especially relates to a kind of weeding monitoring system and method based on unmanned aerial vehicle, the present application utilizes the multispectral camera and microwave sensor carried by unmanned aerial vehicle, the spectral reflectance data and microwave internal structure data of target area are collected;System generates standardized perception data through data fusion module, and combines feature extraction, adaptive fuzzy clustering and multi-model collaborative learning technology, to carry out accurate classification and identification of weeds;Based on the spatial distribution map of weeds, the system predicts the weed spread trend through graph convolution network, and generates the optimal weeding strategy combined with environmental data;Unmanned aerial vehicle executes weeding task through optimized flight path, and monitors weeding effect in real time, continuously optimizes strategy, ensures the efficient execution of weeding task.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural automation and intelligent control, and particularly relates to a weed monitoring system and method based on unmanned aerial vehicles. BACKGROUND

[0002] Unmanned aerial vehicles are increasingly used in agriculture, particularly in weed monitoring, which is of great significance. Traditional weed control methods usually rely on manual or mechanical patrols and operations, which are inefficient and labor-intensive. Weed control systems based on unmanned aerial vehicles can use remote sensing technology and sensors to monitor, identify, and evaluate weeds in farmland in real time, thereby improving the accuracy and efficiency of weed control. This automated monitoring method not only reduces dependence on chemical herbicides, but also reduces operating costs and improves crop yields and environmental friendliness.

[0003] Existing unmanned aerial vehicle weed monitoring technology (Chinese invention patent, publication number: CN118247643B, name: A weed monitoring and evaluation method and system based on unmanned aerial vehicles) mainly relies on computer vision technology and deep learning models, using a monitoring method based on visual and behavioral trajectory semantics. However, this technology has many shortcomings in practical application. Existing technology often requires multiple data collections for repeated analysis, making it difficult to capture the dynamic process of weed spread in real time, resulting in a lag in weed control strategies. In addition, existing systems have limited integration of environmental data, making it difficult to adapt to changes in weed spread under different environmental conditions. SUMMARY

[0004] To address the many problems with existing technology, the present application provides a weed monitoring system and method based on unmanned aerial vehicles. The present application uses an unmanned aerial vehicle equipped with a multispectral camera and a microwave sensor to collect spectral reflectance data and microwave internal structure data of the target area. Through data fusion, feature extraction, convolutional neural networks, and recurrent neural networks, the system can accurately identify the spatial distribution and spread trend of weeds, generate optimal weed control strategies, and optimize flight paths. The system can monitor weed control effects in real time and adjust strategies based on feedback, thereby achieving precise weed management in farmland.

[0005] A weed monitoring system based on unmanned aerial vehicles, comprising the following steps:

[0006] An unmanned aerial vehicle platform equipped with a multispectral camera and a microwave sensor for collecting spectral reflectance data and microwave internal structure data of the target area;

[0007] A data fusion module that receives the spectral reflectance data and microwave internal structure data and generates standardized perception data including spectral reflectance characteristics and internal structure information;

[0008] The feature extraction module is used to extract features from the standardized sensing data, generate spectral feature data and microwave feature data respectively, expand the data dimension through nonlinear transformation, and classify the feature data using adaptive fuzzy clustering to generate preliminary feature classification data.

[0009] The multi-model collaborative learning system receives the preliminary feature classification data, extracts spatial feature data through a convolutional neural network, performs time series feature analysis in conjunction with a recurrent neural network, generates fused feature data, and optimizes through a feature reconstruction network to output the final weed classification data.

[0010] The spatial distribution map construction module constructs a spatial distribution map of weeds based on the final weed classification data, and uses a graph convolutional network to analyze the spatial relationship between weed nodes, predict the weed diffusion trend, generate diffusion trend data, and perform density analysis on the weed distribution in the region to generate density data.

[0011] The weeding strategy generation module combines the diffusion trend data, density data, and real-time environmental data to generate preliminary weeding strategy data, optimize the drone's flight path, execute the weeding task, monitor and generate weeding effect data in real time, and optimize the strategy based on the weeding effect data.

[0012] Preferably, the data fusion module includes:

[0013] The processing unit receives spectral reflectance data and microwave internal structure data. This unit generates standardized sensing data using a weighted average method. The calculation expression for the standardized sensing data is as follows:

[0014]

[0015] Where S represents standardized sensing data; R represents spectral reflectance data; M represents microwave internal structure data; w1 represents the weighting coefficient of the spectral data; and w2 represents the weighting coefficient of the microwave data.

[0016] Preferably, the feature extraction module includes:

[0017] A nonlinear transformation unit is used to perform nonlinear function transformation on the standardized sensing data to expand the data dimension and generate expanded feature data;

[0018] An adaptive fuzzy clustering unit is used to perform fuzzy clustering analysis on the expanded feature data to generate preliminary feature classification data. The clustering formula for the fuzzy clustering is:

[0019]

[0020] Where, μ ij d represents the membership degree of the j-th sample in the i-th cluster;ij denoted as , where is the distance between the sample and the cluster center; C is the number of clusters; and m is the fuzziness index, used to control the degree of fuzziness in clustering.

[0021] Preferably, the multi-model collaborative learning system includes:

[0022] The convolutional neural network module is used to process the preliminary feature classification data, and the extracted spatial feature data reflects the shape, color and other visual features of the plant;

[0023] The recurrent neural network module is used to analyze time series feature data and identify plant growth change trends and cycles.

[0024] Preferably, the fused feature data is generated by weighted fusion of the spatial feature data and the time series feature data, and the fused feature data is used to improve the classification accuracy of the final weed classification data.

[0025] Preferably, the spatial distribution map construction module is used to construct a spatial distribution map of weeds based on the node information in the final weed classification data. This spatial distribution map is used to visually display the distribution of weeds in the target area.

[0026] Preferably, the graph convolutional network is used to analyze weed nodes in the spatial distribution map, and the generated diffusion trend data includes the diffusion rate and diffusion direction of each weed node, which is used to guide subsequent weeding strategies.

[0027] Preferably, the weeding strategy generation module combines the diffusion trend data, density data, and real-time environmental data to generate preliminary weeding strategy data using a decision tree algorithm. The decision tree algorithm makes weeding decisions based on weed density and diffusion potential.

[0028] Preferably, the weeding strategy generation module further includes a real-time monitoring submodule, used to monitor the execution of the weeding task and generate real-time weeding effect data, which is used to evaluate the weeding effect and optimize subsequent weeding strategies.

[0029] A method for implementing the drone-based weeding monitoring system includes the following steps:

[0030] Collect spectral reflectance data and microwave internal structure data of the target area;

[0031] The system receives the spectral reflectance data and microwave internal structure data, and generates standardized sensing data that includes spectral reflectance characteristics and internal structure information.

[0032] Feature extraction is performed on the standardized sensing data to generate spectral feature data and microwave feature data, respectively. The data dimension is expanded by nonlinear transformation, and the feature data is classified by adaptive fuzzy clustering to generate preliminary feature classification data.

[0033] The system receives the preliminary feature classification data, extracts spatial feature data through a convolutional neural network, performs time series feature analysis in conjunction with a recurrent neural network, generates fused feature data, and optimizes it through a feature reconstruction network to output the final weed classification data.

[0034] Based on the final weed classification data, a spatial distribution map of weeds is constructed, and a graph convolutional network is used to analyze the spatial relationship between weed nodes, predict the weed diffusion trend, generate diffusion trend data, and perform density analysis on the weed distribution in the region to generate density data.

[0035] By combining the diffusion trend data, density data, and real-time environmental data, preliminary weeding strategy data is generated, the flight path of the drone is optimized, and the weeding task is executed. Weeding effect data is generated in real time, and the strategy is optimized based on the weeding effect data.

[0036] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0037] This invention achieves accurate identification and real-time monitoring of weeds by combining multispectral camera and microwave sensor technology. The system generates standardized sensing data, including spectral reflectance characteristics and internal structure information, through a data fusion module, effectively improving the ability to analyze plant health status and weed spread trends.

[0038] This invention addresses the shortcomings of existing technologies in real-time dynamic monitoring of weed spread through adaptive fuzzy clustering technology and a multi-model collaborative learning system, thereby improving the accuracy of weed identification and the response speed of weed control strategies.

[0039] This invention significantly improves the efficiency and accuracy of weeding operations by optimizing the flight path of drones and combining it with real-time environmental data, thereby reducing the use of pesticides and protecting the ecological environment. Attached Figure Description

[0040] Figure 1 This is a structural block diagram of the system of the present invention;

[0041] Figure 2 This is a schematic diagram of the feature extraction and learning process in this invention;

[0042] Figure 3 This is a schematic diagram of the spatial distribution and diffusion analysis process in this invention;

[0043] Figure 4This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0044] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0045] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0046] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0047] like Figure 1 As shown, a drone-based weeding monitoring system includes the following steps:

[0048] The drone platform is equipped with a multispectral camera and microwave sensors to collect spectral reflectance data and microwave internal structure data of the target area;

[0049] In this invention, the unmanned aerial vehicle (UAV) platform is the core component. It is equipped with a multispectral camera and a microwave sensor, primarily used to collect spectral reflectance data and microwave internal structure data of the target area. This configuration is based on remote sensing technology, which captures and analyzes the optical characteristics of plants and their surrounding environment to obtain real-time information on vegetation health and weed distribution.

[0050] Multispectral cameras capture reflectance spectral information using light of different wavelengths, and information in each band can reveal the physiological state of a plant. A plant's spectral reflectance varies depending on factors such as its chlorophyll content, water content, and leaf structure. By analyzing this reflectance data, the health status of plants can be assessed. For example, healthy plants will show higher reflectance in specific bands (such as red and near-infrared light), while weeds or damaged plants may exhibit different reflectance characteristics. Combined with the generation of standardized sensor data, the system can effectively distinguish different plant types, identify the presence of weeds, and assess their growth status.

[0051] Microwave sensors, by emitting microwave signals and receiving their echoes, can deeply probe the internal structure of plants, providing information about soil moisture and root distribution. The advantage of microwave technology lies in its ability to penetrate the soil and plant internal structure; this characteristic allows it to operate stably under varying soil conditions and is unaffected by light and climate changes. By combining data obtained from microwave sensors, the system can more comprehensively assess the growth environment and potential spread trends of weeds.

[0052] By combining data from these two sensors, the drone platform can generate standardized perception data of the target area in real time, forming a high-resolution plant distribution map. This monitoring method significantly improves the efficiency of weed management, allowing both crop growth and weed growth dynamics to be reflected in real-time data analysis. Ultimately, this helps farmers develop more precise weeding strategies, reduce pesticide use, and improve the economic benefits and ecological sustainability of operations.

[0053] For example, in one implementation, drones fly over wheat fields, collecting spectral reflectance and microwave data which are processed in real time. Image recognition technology is then used to analyze the distribution of wheat and weeds. If the system analysis reveals a high weed density in a particular area, the decision support system will recommend that farmers prioritize mechanical or chemical weeding in that area. Through this intelligent monitoring and management, farmers can more effectively control weed growth, ensuring healthy wheat growth and ultimately achieving higher yields and economic benefits.

[0054] The data fusion module receives the spectral reflectance data and microwave internal structure data, and generates standardized sensing data including spectral reflectance characteristics and internal structure information.

[0055] Preferably, the data fusion module includes:

[0056] The processing unit receives spectral reflectance data and microwave internal structure data. This unit generates standardized sensing data using a weighted average method. The calculation expression for the standardized sensing data is as follows:

[0057]

[0058] Where S represents standardized sensing data; R represents spectral reflectance data; M represents microwave internal structure data; w1 represents the weighting coefficient of the spectral data; and w2 represents the weighting coefficient of the microwave data.

[0059] In this invention, the data fusion module is a key component. Its main function is to receive spectral reflectance data and microwave internal structure data, and generate standardized sensing data that includes spectral reflectance characteristics and internal structure information. This module is based on data fusion technology, aiming to combine data collected from different sensors to improve the overall understanding and analysis of the plant conditions in the target area.

[0060] Specifically, the data fusion module's processing unit generates standardized sensing data using a weighted average method. This weighted average method ensures comprehensive consideration of different data sources, thus establishing a reasonable connection between multispectral and microwave data. In this process, spectral reflectance data (R) provides information on the surface characteristics of plants, while microwave internal structure data (M) provides deeper information about plants and soil. The calculation expression for standardized sensing data is shown above. Through these calculation expressions, the system can dynamically adjust the influence of each data source in the final output, ensuring that the fusion result more accurately reflects the actual situation of the target area.

[0061] For example, in monitoring a specific farmland, spectral reflectance data collected by a drone during flight can effectively identify the plant's growth status, but this data alone cannot provide in-depth information about soil moisture or root conditions. Microwave sensors, however, can obtain this information through the soil. Through a data fusion module, the combination of spectral reflectance data and microwave internal structure data results in standardized sensing data that includes comprehensive information on vegetation health and its growing environment. This integration of information not only improves the accuracy and reliability of the data but also provides a solid foundation for subsequent weed identification and weed control strategy generation.

[0062] Through such data fusion, the system can achieve more intelligent decision support. For example, when the system detects that the weed density in a certain area is high, and combines microwave data analysis of the soil moisture conditions in that area, it can suggest targeted weeding measures to maximize resource utilization efficiency and weeding effectiveness.

[0063] The feature extraction module is used to extract features from the standardized sensing data, generate spectral feature data and microwave feature data respectively, expand the data dimension through nonlinear transformation, and classify the feature data using adaptive fuzzy clustering to generate preliminary feature classification data.

[0064] like Figure 2 As shown, preferably, the feature extraction module includes:

[0065] A nonlinear transformation unit is used to perform nonlinear function transformation on the standardized sensing data to expand the data dimension and generate expanded feature data;

[0066] An adaptive fuzzy clustering unit is used to perform fuzzy clustering analysis on the expanded feature data to generate preliminary feature classification data. The clustering formula for the fuzzy clustering is:

[0067]

[0068] Where, μ ij d represents the membership degree of the j-th sample in the i-th cluster; ij denoted as , where is the distance between the sample and the cluster center; C is the number of clusters; and m is the fuzziness index, used to control the degree of fuzziness in clustering.

[0069] In this invention, the feature extraction module is a crucial component, responsible for extracting useful features from standardized sensing data for subsequent analysis and decision-making. This module works by combining nonlinear transformation and adaptive fuzzy clustering to accurately extract spectral and microwave feature data, thereby enabling effective identification and classification of vegetation and weeds within the target area.

[0070] The nonlinear transformation unit in the feature extraction module transforms standardized perceptual data using nonlinear functions. The purpose of this process is to expand the dimensionality of the data, making it richer and more complex in a multidimensional space. This transformation allows for better capture of latent features in the data, especially those that cannot be simply expressed by linear transformations. For example, when processing spectral reflectance data, the reflectance characteristics of certain plants at specific wavelengths may change due to light conditions and health status. Nonlinear transformation helps extract these changing features, thus providing a more detailed data foundation for subsequent classification.

[0071] The adaptive fuzzy clustering unit performs fuzzy clustering analysis on the extended feature data after nonlinear transformation. The advantage of fuzzy clustering is that it allows a sample to have different degrees of membership in multiple clusters, which is particularly effective for handling complex plant species and growth states. In this way, the system can more flexibly identify weeds and crops in mixed vegetation areas.

[0072] For example, suppose a field is interspersed with weeds and wheat. The feature extraction module can classify wheat and weeds using fuzzy clustering based on different features of spectral and microwave data. The system might label plant feature data in a certain area as belonging to both the "weeds" and "wheat" clusters, but reflecting their primary characteristics under different membership levels. This flexibility allows the system to provide farmers with more accurate weed identification information, thereby enabling them to develop more effective weed control strategies.

[0073] The multi-model collaborative learning system receives the preliminary feature classification data, extracts spatial feature data through a convolutional neural network, performs time series feature analysis in conjunction with a recurrent neural network, generates fused feature data, and optimizes through a feature reconstruction network to output the final weed classification data.

[0074] Preferably, the multi-model collaborative learning system includes:

[0075] The convolutional neural network module is used to process the preliminary feature classification data, and the extracted spatial feature data reflects the shape, color and other visual features of the plant;

[0076] The recurrent neural network module is used to analyze time series feature data and identify plant growth change trends and cycles.

[0077] In this invention, a multi-model collaborative learning system is a key intelligent analysis component, designed to efficiently process preliminary feature classification data and extract valuable information to support decision-making. This system includes a convolutional neural network module and a recurrent neural network module, used to process spatial feature data and time-series feature data, respectively.

[0078] The convolutional neural network (CNN) module is specifically designed to process the extracted preliminary feature classification data, which includes plant shape, color, and other visual features. Through its hierarchical structure, the CNN can automatically extract key features from images, such as identifying the edges, textures, and shapes of different plants. For farmland monitored by drones, this means the system can effectively distinguish between weeds and crops. For example, through training, the CNN can learn to recognize the features of wheat and common weeds, thereby generating detailed plant distribution maps during monitoring. This accuracy in spatial feature extraction significantly improves the accuracy of weed identification, providing reliable data support for subsequent weeding strategies.

[0079] The recurrent neural network module is responsible for analyzing time-series feature data to identify trends and cycles in plant growth. Because plant growth is dynamic, recurrent neural networks can process sequential data and capture its temporal dependencies, thus effectively analyzing plant growth cycles and trends. This is crucial for agricultural management, as understanding changes in plant growth status helps farmers weed or fertilize at the appropriate time. For example, the system can monitor weed growth rates and predict future growth trends, combining environmental and historical data to provide farmers with suggestions, such as implementing mechanical weeding during peak weed growth periods.

[0080] Combining the outputs of these two modules, the system can form a comprehensive plant growth analysis model, enabling precise monitoring and management of weeds and crops. Through this intelligent analysis method, the drone monitoring system not only improves the efficiency of crop management but also effectively reduces the use of chemical herbicides, promoting sustainable agricultural development. In summary, the multi-model collaborative learning system, through the effective combination of convolutional neural networks and recurrent neural networks, provides powerful data processing and analysis capabilities for drone weed control monitoring, helping farmers better address weed problems and optimize agricultural production.

[0081] Preferably, the fused feature data is generated by weighted fusion of the spatial feature data and the time series feature data, and the fused feature data is used to improve the classification accuracy of the final weed classification data.

[0082] In this invention, the generation of fused feature data is a crucial step, aiming to improve the accuracy of the final weed classification data. This fused feature data is generated by weighted fusion of spatial feature data and time-series feature data. This process is based on the idea of ​​multi-source information integration, aiming to fully utilize the advantages of different data sources.

[0083] Spatial feature data, extracted by convolutional neural networks, reflects the shape, color, and other visual characteristics of vegetation, while time-series feature data, analyzed through recurrent neural networks, reveals the trends and cycles of plant growth changes. Both types of feature data provide important information: spatial feature data accurately captures the static characteristics of plants, while time-series feature data reflects their dynamic changes. By weighted and fused together, a more comprehensive feature set can be generated, thereby improving the ability to classify weeds and crops.

[0084] In practical applications, the generation process of fused feature data typically involves setting weighting coefficients to better reflect the importance of various features in the classification task. For example, if the growth status of weeds in a specific area is significantly affected by seasonal changes, the weight of time-series feature data can be appropriately increased during the fusion process to ensure that dynamic information is fully utilized in the classification process. Through this weighted fusion, the system can comprehensively consider the current state and historical changes of plants to generate more representative feature data.

[0085] For example, when monitoring a wheat field, the system first extracts spatial feature data for the area, potentially showing the distribution of wheat and weeds. Then, by combining this with time-series data, the system can identify the growth trend of weeds within the area, forming a deeper understanding of the weed growth cycle. Through weighted fusion, the resulting fused feature data not only includes the visual characteristics of the weeds but also incorporates information related to their growth dynamics, thus significantly improving the accuracy of the final weed classification.

[0086] The fusion of feature data plays a crucial role in drone weeding monitoring systems. By effectively integrating different features, the system can make more accurate weed identification and classification decisions in complex environments.

[0087] The spatial distribution map construction module constructs a spatial distribution map of weeds based on the final weed classification data, and uses a graph convolutional network to analyze the spatial relationship between weed nodes, predict the weed diffusion trend, generate diffusion trend data, and perform density analysis on the weed distribution in the region to generate density data.

[0088] The spatial distribution map construction module processes the final weed classification data to create a spatial distribution map of the weeds. This process converts the classification data into a graphical representation in a spatial coordinate system, allowing the location and category of each weed node to be visually displayed on the map. Such spatial graphics not only clearly show the distribution of weeds but also help farmers identify the weed growth status in different areas, thereby enabling them to formulate corresponding management strategies.

[0089] The core of this module is analyzing the spatial relationships between weed nodes using graph convolutional networks (GCNNs). GCNNs are powerful deep learning tools specifically designed for processing graph-structured data. In this application, they capture the interactions and proximity relationships between weed nodes, effectively identifying weed spread patterns. For example, if weeds are growing rapidly and densely in a certain area, the GCNN can analyze the relationship between this area and its surrounding nodes to infer the temporal spread trend of the weeds, generating spread trend data. Such predictive information is crucial for developing weed control strategies, enabling farmers to take preventative measures to prevent weed spread.

[0090] The module also performs density analysis on the distribution of weeds within the area, generating density data. By combining the number of weed nodes with their coverage area, the system can calculate the weed density value, thus providing farmers with quantitative information about the intensity of weed growth. For example, in a field, if the density analysis shows that the average weed density exceeds a predetermined threshold, farmers can take timely measures, such as mechanical weeding or spraying herbicides, to control weed growth.

[0091] like Figure 3As shown, preferably, the spatial distribution map construction module is used to construct a spatial distribution map of weeds based on the node information in the final weed classification data. This spatial distribution map is used to visually display the distribution of weeds in the target area.

[0092] In this invention, the spatial distribution map construction module plays a crucial role, primarily used to construct a spatial distribution map of weeds based on the node information in the final weed classification data. The core principle of this module is to visualize the weeds in the target area by collecting and processing weed node information, thereby providing farmers with accurate information on weed distribution.

[0093] Specifically, the spatial distribution map construction module first extracts the location information and category of each weed node from the final weed classification data. This node information is usually obtained through the previous feature extraction and classification process. Each node represents weed information within a certain area, including its type, quantity, and growth status. By mapping this node information to a geographic coordinate system, the module can construct a distribution map of weeds within that area. This process typically involves using GIS (Geographic Information System) technology and data visualization tools to present the weed distribution graphically.

[0094] This spatial distribution map not only clearly shows the distribution of weeds, but also uses different colors and symbols to represent the density and type of weeds. For example, higher weed density might be marked in red, while lower density might be marked in green. This visualization allows farmers to easily identify areas that require focused treatment, thus enabling them to develop more effective weed control strategies.

[0095] For example, when monitoring a field, spatial distribution maps can show the growth of weeds in different areas, helping farmers identify which areas have abundant weeds and which areas are relatively clean. This information allows farmers to take targeted measures based on the actual situation, such as prioritizing mechanical weeding or spraying herbicides in areas with high weed density to achieve better management results.

[0096] Preferably, the graph convolutional network is used to analyze weed nodes in the spatial distribution map, and the generated diffusion trend data includes the diffusion rate and diffusion direction of each weed node, which is used to guide subsequent weeding strategies.

[0097] In this invention, Graph Convolutional Networks (GCNs) play a crucial role, specifically designed to analyze weed nodes in spatial distribution maps and generate diffusion trend data. This process leverages the powerful characteristics of GCNs to effectively capture the spatial relationships between weed nodes, thereby assessing the weed diffusion rate and direction, and providing a scientific basis for subsequent weed control strategies.

[0098] Graph Convolutional Networks (GCNs) are deep learning models capable of processing graph-structured data. Their principle is to learn the feature representation of nodes by aggregating information from neighboring nodes. In this invention, the weed nodes and their connections in a spatial distribution map constitute a graph structure. GCNs extract the comprehensive features of each node within its neighborhood by performing convolution operations on the features of these nodes. These features include not only basic information about the weeds (such as type and density) but also the characteristics of neighboring weed nodes, enabling the model to more comprehensively understand the growth environment of the weeds.

[0099] In its implementation, the graph convolutional network (GCN) first encodes the features of each weed node and propagates information based on adjacency relationships. For example, if a node has a high density and its neighboring nodes also have a strong diffusion trend, the GCN can learn that the node has a high diffusion potential. Through such feature aggregation and propagation, the system can generate diffusion trend data for each weed node, including diffusion rate and diffusion direction. The diffusion rate reflects the growth speed of the weeds, while the diffusion direction indicates the direction in which the weeds may expand outwards. This information is crucial for developing precise weed control measures.

[0100] For example, during the monitoring of a cornfield, if graph convolutional network analysis shows that the weed spread rate in a certain area is high and the spread direction is towards adjacent crop rows, farmers can take timely measures based on this information, such as applying targeted herbicides in that area to prevent further threats to crops from the weeds. Such applications not only improve the accuracy of weed control but also reduce the amount of chemical agents used, thereby achieving more sustainable agricultural management.

[0101] The weeding strategy generation module combines the diffusion trend data, density data, and real-time environmental data to generate preliminary weeding strategy data, optimize the drone's flight path, execute the weeding task, monitor and generate weeding effect data in real time, and optimize the strategy based on the weeding effect data.

[0102] In this invention, the weeding strategy generation module is a key decision support component. It is responsible for combining diffusion trend data, density data, and real-time environmental data to generate preliminary weeding strategy data and optimize the drone's flight path. The principle of this module is to formulate effective weeding strategies through data fusion and intelligent algorithms, thereby improving the efficiency and accuracy of weed management.

[0103] Diffusion trend data provides information on weed growth dynamics, including the rate of weed spread and possible directions of spread. Density data shows the concentration of weeds in different areas, enabling the system to accurately identify weed hotspots. Meanwhile, real-time environmental data (such as climate conditions and soil moisture) directly impacts weed growth and is therefore a crucial basis for developing weed control strategies. By combining these data, the weed control strategy generation module can comprehensively analyze the weed status of the target area.

[0104] When generating initial weed control strategy data, the system typically employs machine learning algorithms and optimization techniques, comprehensively considering multiple factors, including weed type, density, spread trend, and environmental conditions. For example, the system can apply decision trees or other intelligent algorithms to evaluate the effectiveness of different weed control methods in order to recommend the most suitable treatment for each area. If an area has a high weed density and a fast growth rate, the system may recommend mechanical weeding, while in areas with fewer weeds, it may recommend chemical treatment.

[0105] Optimizing the drone's flight path is another important function of this module. By analyzing the generated preliminary weeding strategy, the system can calculate the most efficient flight route to maximize coverage of the target area while reducing overlapping operations and unnecessary flights. Drone path optimization not only improves operational efficiency but also reduces energy consumption.

[0106] After performing a weeding task, the system monitors weeding effectiveness data in real time. This data is used to evaluate the effectiveness and execution of the weeding strategy. For example, the system can re-collect spectral and density data to determine changes in weed growth status. If it finds that weeds are still growing vigorously in certain areas, the system will optimize the initial weeding strategy based on this real-time feedback, adjusting subsequent weeding methods and flight paths to ensure the best possible weeding effect.

[0107] Through this closed-loop control, the weeding strategy generation module can continuously learn and improve, making the drone more intelligent and efficient when carrying out weeding tasks.

[0108] Preferably, the weeding strategy generation module combines the diffusion trend data, density data, and real-time environmental data to generate preliminary weeding strategy data using a decision tree algorithm. The decision tree algorithm makes weeding decisions based on weed density and diffusion potential.

[0109] In this invention, the weeding strategy generation module combines diffusion trend data, density data, and real-time environmental data, and uses a decision tree algorithm to generate preliminary weeding strategy data. The core of this process lies in integrating information from multiple data sources to form a scientifically sound weeding decision and optimize farmland management.

[0110] Decision tree algorithms are a common machine learning method that can make classifications and decisions by analyzing input features (such as weed density, spread potential, and environmental conditions). In this invention, the algorithm is first trained on the input feature data to establish a model of the relationship between weed growth and environmental factors. Specifically, spread trend data provides information on how weeds expand in time and space, density data describes the concentration of weeds in a specific area, and real-time environmental data (such as temperature, humidity, and precipitation) has a direct impact on weed growth. Through the decision tree algorithm, the system can identify which features are most important to weed growth and formulate corresponding weeding strategies based on these features.

[0111] For example, in a specific monitoring case, suppose an area has a very high weed density and weather conditions are favorable for weed growth. The decision tree algorithm would take this information as input and generate a decision node indicating that the area should prioritize more effective mechanical weeding measures, while in areas with relatively fewer weeds, the system might suggest using milder chemical herbicides. In this way, through the structure of the decision tree, farmers can clearly understand the specific actions to take in different situations.

[0112] The generated preliminary weeding strategy data can not only guide farmers in choosing appropriate weeding methods but also help optimize drone operation plans. After the strategy is implemented, the data feedback on the weeding effect will be used to update and improve the decision tree model, further enhancing the accuracy of subsequent decisions. This intelligent decision-making process makes weeding management more scientific and precise, effectively reducing labor costs and the use of chemical agents, thereby promoting the development of sustainable agriculture.

[0113] Preferably, the weeding strategy generation module further includes a real-time monitoring submodule, used to monitor the execution of the weeding task and generate real-time weeding effect data, which is used to evaluate the weeding effect and optimize subsequent weeding strategies.

[0114] In this invention, the weeding strategy generation module is not only responsible for formulating a preliminary weeding plan, but also includes a real-time monitoring submodule. This module is used to monitor the execution of the weeding task and generate real-time weeding effect data. This mechanism is designed to ensure the effectiveness of weeding measures and provide data support for the optimization of subsequent strategies.

[0115] The core principle of the real-time monitoring submodule lies in continuously tracking various indicators during the weeding process, collecting relevant data through sensors mounted on the drone. For example, during monitoring, multispectral cameras may be used to acquire real-time spectral reflectance data, analyzing the health status of plants and the growth status of weeds within the target area. Real-time monitoring can also be combined with microwave sensors to obtain information on the state of soil and plant roots. This data not only reflects the current weeding effect but also assesses whether the weeding measures have achieved the expected results, such as whether weed density has been significantly reduced and whether crops have been affected.

[0116] The generated real-time weed control data provides farmers with a dynamic feedback mechanism, enabling them to quickly assess the effectiveness of their current strategies. For example, if monitoring data shows that weed growth is not as expected during a particular application, the system can adjust its weed control strategy based on real-time data, such as increasing the dosage of the herbicide, changing the spraying area, or using other weed control methods. This flexibility allows farmers to respond to changes instantly, improving the success rate and efficiency of weed control operations.

[0117] Through continuous monitoring and feedback, the real-time monitoring submodule can also provide valuable data for optimizing subsequent weeding strategies. By analyzing historical real-time monitoring data, the system can identify which weeding measures are more effective under specific environmental conditions, thereby guiding farmers to develop more precise strategies for future weeding tasks.

[0118] For example, in a weeding task in a cornfield, if real-time monitoring reveals that the weed density in a certain area remains high, and environmental data indicates that the current climatic conditions are favorable for the volatilization and diffusion of pesticides, the system may suggest that the farmer reapply pesticides in that area and continue to evaluate the effectiveness of the improvements based on real-time data. This dynamic adjustment mechanism enables the drone weeding monitoring system to not only execute preset tasks but also to perform intelligent management based on actual conditions, significantly improving the scientific nature and efficiency of agricultural management.

[0119] like Figure 4 As shown, a method for implementing the drone-based weeding monitoring system includes the following steps:

[0120] The system collects spectral reflectance data and microwave internal structure data of the target area. The drone weeding monitoring system is mainly based on multispectral and microwave sensors, and collects spectral reflectance data and microwave internal structure data of the target area through the drone platform.

[0121] The system receives the spectral reflectance data and microwave internal structure data, and generates standardized sensing data that includes both spectral reflectance characteristics and internal structure information. The spectral reflectance data reveals the surface features of the plant, while the microwave internal structure data provides in-depth analysis of the plant's internal growth, such as root structure and soil moisture. These two types of data are integrated through a data fusion module to generate standardized sensing data that includes both spectral reflectance characteristics and internal structure information. This step improves the overall analyzability and accuracy of the data.

[0122] Feature extraction is performed on the standardized sensing data to generate spectral and microwave feature data. The data dimensionality is expanded through nonlinear transformation, and adaptive fuzzy clustering is used to classify the feature data, generating preliminary feature classification data. The feature extraction module processes this standardized data to generate spectral and microwave feature data, and expands the data dimensionality through nonlinear transformation to ensure the system can capture more detailed vegetation features. Adaptive fuzzy clustering is used to classify these features, generating preliminary feature classification data, which enables the system to distinguish between crops and weeds.

[0123] The system receives the preliminary feature classification data, extracts spatial feature data using a convolutional neural network (CNN), performs time-series feature analysis using a recurrent neural network (RNN), generates fused feature data, and optimizes it through a feature reconstruction network to output the final weed classification data. After receiving the preliminary feature classification data, the system uses a CNN in a multi-model collaborative learning system to extract spatial features of plants, while a RNN processes time-series features. The advantage of CNNs lies in their ability to accurately extract spatial features such as plant shape and color, while RNNs are used to analyze plant growth trends. This information is fused to generate more accurate weed classification data, which is further optimized through a feature reconstruction network.

[0124] Based on the final weed classification data, a spatial distribution map of weeds is constructed. A graph convolutional network is used to analyze the spatial relationships between weed nodes, predict weed spread trends, and generate spread trend data. Furthermore, density analysis is performed on the weed distribution within the region to generate density data. Based on this weed classification data, the system constructs a spatial distribution map that not only displays the current distribution of weeds but also analyzes the spatial relationships between weed nodes using a graph convolutional network to predict weed spread trends. This spread trend data is used to generate weed density data within the region, providing a basis for subsequent weeding strategies.

[0125] Combining the diffusion trend data, density data, and real-time environmental data, preliminary weeding strategy data is generated, the drone's flight path is optimized, and the weeding task is executed. Weeding effect data is monitored in real time, and the strategy is optimized based on this data. The system combines diffusion trend data, density data, and real-time environmental data to generate an optimized weeding strategy, using the drone for path optimization and task execution. During execution, the system continuously adjusts and optimizes the strategy by monitoring the weeding effect in real time to ensure the final weeding task achieves the best results.

[0126] This invention, through multi-source data fusion and an intelligent learning system, can monitor the spread of weeds in real time, accurately analyze their growth dynamics, and provide efficient weeding strategies. This significantly reduces the use of chemical agents, improves operational efficiency, and reduces damage to crops, resulting in significant agricultural management and ecological benefits.

[0127] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0128] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An unmanned aerial vehicle based weed monitoring system characterized in that, The method comprises the following steps: An unmanned aerial vehicle platform carries a multispectral camera and a microwave sensor for collecting spectral reflectance data and microwave internal structure data of a target area; A data fusion module receives the spectral reflectance data and microwave internal structure data and generates standardized perception data including spectral reflectance characteristics and internal structure information; A feature extraction module extracts features from the standardized perception data, generates spectral feature data and microwave feature data respectively, and expands the data dimension through nonlinear transformation, classifies the feature data using adaptive fuzzy clustering, and generates preliminary feature classification data; A multi-model collaborative learning system receives the preliminary feature classification data, extracts spatial feature data through a convolutional neural network, analyzes time series feature data using a recurrent neural network, generates fusion feature data, and outputs final weed classification data through feature reconstruction network optimization; The multi-model collaborative learning system includes a convolutional neural network module for processing the preliminary feature classification data and extracting spatial feature data reflecting the shape and color of plants, and a recurrent neural network module for analyzing time series feature data and identifying plant growth trends and cycles; A spatial distribution map construction module constructs a spatial distribution map of weeds based on the final weed classification data, analyzes the spatial relationships between weed nodes using a graph convolution network, predicts weed spread trends, generates spread trend data, and analyzes the density of weed distribution in the area to generate density data; The spatial distribution map construction module constructs a spatial distribution map of weeds based on node information in the final weed classification data, which is used to visually display the distribution of weeds in the target area; A weeding strategy generation module generates preliminary weeding strategy data by combining the spread trend data, density data, and real-time environmental data, optimizes the flight path of the unmanned aerial vehicle, and executes the weeding task, monitors the weeding effect data in real time, and optimizes the strategy based on the weeding effect data; The weeding strategy generation module generates preliminary weeding strategy data using a decision tree algorithm by combining the spread trend data, density data, and real-time environmental data, and the decision tree algorithm makes weeding decisions based on weed density and spread potential; The weeding strategy generation module also includes a real-time monitoring submodule for monitoring the execution of the weeding task and generating real-time weeding effect data, which is used to evaluate the weeding effect and optimize subsequent weeding strategies.

2. The unmanned aerial vehicle based weed monitoring system of claim 1, wherein, The data fusion module includes: a processing unit receiving the spectral reflectance data and the microwave internal structure data, the processing unit generating normalized perception data using a weighted average method, wherein a computational expression of the normalized perception data is: wherein, is the normalized perception data; is the spectral reflectance data; is the microwave internal structure data; is a weight coefficient of the spectral data; is a weight coefficient of the microwave data.

3. The unmanned aerial vehicle based weed monitoring system of claim 1, wherein, The feature extraction module includes: A nonlinear transformation unit for nonlinear function transformation of the standardized perception data to expand the data dimension and generate expanded feature data; An adaptive fuzzy clustering unit is used to perform fuzzy clustering analysis on the expanded feature data to generate preliminary feature classification data. The clustering formula for the fuzzy clustering is: in, For the first The sample at the th Membership degree in each cluster; The distance between the sample and the cluster center; This represents the number of clusters. This is a fuzziness index used to control the degree of fuzziness in clustering.

4. The unmanned aerial vehicle based weed monitoring system of claim 1, wherein, The fusion feature data is generated by weighting the spatial feature data and time series feature data, and the fusion feature data is used to improve the classification accuracy of the final weed classification data.

5. The unmanned aerial vehicle based weed monitoring system of claim 1, wherein, The graph convolution network is used to analyze the weed nodes in the spatial distribution map, and the generated spread trend data includes the spread rate and direction of each weed node, which is used to guide the subsequent weeding strategy.

6. A method for performing the drone-based weed monitoring system of any of claims 1-5.

Images