Garden plant periodic pruning intelligent reminding and management system based on internet of things and growth prediction

Abstract

The application discloses a kind of based on internet of things and growth prediction's garden plant periodic pruning intelligent reminding and management system.System includes: multi-modal fusion perception and plant state diagnosis subsystem, data are collected by multi-source sensor and based on graph convolution network output plant health state vector;Growth prediction and pruning knowledge graph subsystem, use time series knowledge graph and prediction model simulate future growth trajectory;Pruning scheme generation subsystem, adopt multi-agent reinforcement learning to generate personalized pruning strategy;Man-machine object collaborative scheduling subsystem, dynamically schedule robot, unmanned aerial vehicle and AR auxiliary personnel collaborative operation;Effect evaluation and self-evolution subsystem, through multidimensional evaluation feedback drive system optimization.The application realizes the transformation of garden pruning from experience-driven to data intelligent driving, from single-point operation to man-machine object collaboration, and improves the scientificity and intelligent level of pruning operation.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance technology for garden plants, specifically to an intelligent reminder and management system for periodic pruning of garden plants based on the Internet of Things and growth prediction. Background Technology

[0002] As a crucial component of modern urban construction and ecological management, the sophistication and intelligence of landscaping maintenance directly impact landscape sustainability and ecosystem services. Plant pruning is a core operation in landscaping maintenance, aiming to regulate plant growth patterns, promote healthy development, and maintain aesthetic value. However, with the expansion of urban green spaces, the diversification of plant species, and the public's increasing demands for landscape quality, traditional landscaping plant pruning management methods are no longer sufficient to meet the demands of efficient, precise, and scientific modern maintenance.

[0003] Currently, pruning of garden plants mainly relies on cyclical plans developed through human experience, and on-site judgment and operation by maintenance personnel. Although the Internet of Things and image recognition technologies have begun to be applied to garden monitoring in recent years, existing technical solutions still suffer from the following systemic defects and unresolved technical bottlenecks in the specific scenario of pruning management: 1. Monitoring dimensions are limited, lacking in-depth integrated diagnosis of plant physiological states. Existing technologies mostly use visible light cameras for timed photography or deploy single soil moisture and temperature sensors. These solutions can only acquire superficial images of plants or isolated environmental parameters, failing to simultaneously acquire key information reflecting internal plant physiological activity (such as leaf surface temperature, water stress index, photosynthetic characteristics, etc.), and further failing to establish models of physiological relationships between organs. This results in a significant lag and low accuracy in diagnosing early diseases and latent stresses. For example, changes in leaf color alone are insufficient to distinguish between water deficiency, nutrient deficiency, and disease infection, easily leading to misdiagnosis.

[0004] 2. Rigid decision-making processes prevent dynamic planning based on individual growth predictions. Current so-called "intelligent" pruning systems mostly trigger reminders based on fixed schedules or simple threshold rules, such as "prune once per quarter" or "remind when plant height exceeds a certain threshold." This model completely ignores the individual genetic differences of plants, their real-time health status, and the dynamic impact of future climate on their growth rate. The same tree species grows at vastly different rates under different site conditions. A fixed pruning cycle inevitably leads to over-pruning of some plants and under-pruning of others, wasting maintenance resources and potentially causing long-term damage to plant health.

[0005] 3. The optimization objectives are fragmented, lacking the technical means to automatically generate optimal solutions under multiple constraints. Ideal pruning requires a comprehensive balance of multiple objectives, including landscape morphology, plant physiological resilience, and operational costs. Existing technologies either focus only on a single objective (such as shape) or rely entirely on subjective trade-offs based on human experience, lacking a systematic method that can combine aesthetic rules, plant physiological models, and operations research optimization algorithms to automatically solve for multi-objective Pareto optimal solutions. This makes it difficult to scientifically quantify pruning schemes, often resulting in compromises.

[0006] 4. Execution units are isolated, failing to form a human-machine-object collaborative operation system adapted to complex scenarios. While automated equipment such as gardening robots or pruning drones have emerged in the market, they are mostly independent working units with limited functions and cannot cope with the challenges of complex terrain, numerous obstacles, and unstructured tasks in gardens. Existing systems lack a top-level scheduling hub capable of dynamically coordinating autonomous robots, aerial drones, and human maintenance personnel to ensure safe and efficient collaborative operations based on real-time task requirements (such as accuracy, altitude, and intensity). This results in low utilization rates of automated equipment, hindering widespread adoption.

[0007] 5. The system lacks a closed-loop mechanism, preventing models and knowledge bases from continuously evolving through practice. Most monitoring and management systems only complete an open-loop process of "data collection - simple alarm," lacking objective and quantitative evaluation of the effectiveness of pruning operations and whether they have achieved the expected goals. The system lacks the ability to automatically iterate and optimize its identification models, prediction algorithms, and decision rules using operational feedback data. Therefore, system performance tends to solidify over time, failing to adapt to different regions, newly introduced tree species, or changing climatic conditions, resulting in poor long-term applicability.

[0008] 6. The management process lacks forward-looking simulation and risk prevention capabilities. Before implementing a pruning plan, managers cannot predict its long-term effects; during operations, it is also difficult to predict in real time the potential conflicts and risks arising from the collaboration of multiple devices. Existing technologies lack pre-running and verification processes based on virtual simulation and digital twins, and management decisions largely remain in an "experience-driven, post-event remedial" mode, which carries certain blind spots and safety risks.

[0009] To address the aforementioned technological bottlenecks, while recent research has attempted to introduce individual advanced technologies (such as deep learning image recognition and single drone applications) into the landscaping field, these efforts have largely resulted in isolated breakthroughs. They have failed to construct a complete technological chain at the system architecture level, integrating "deep fusion perception, individual growth prediction, multi-objective optimization decision-making, heterogeneous intelligent agent collaboration, and closed-loop evaluation and evolution." Therefore, the field urgently needs a comprehensive solution to fundamentally revolutionize the pruning and management of landscaping plants, enabling a paradigm shift from experience-based management to data-driven intelligence. Summary of the Invention

[0010] The intelligent pruning decision-making and collaborative management system for garden plants provided by this invention, based on multimodal perception and dynamic knowledge graph, comprehensively utilizes cutting-edge technologies such as the Internet of Things, artificial intelligence, multi-agent collaboration and digital twins to construct a closed-loop intelligent management architecture that integrates perception, decision-making, execution and evaluation.

[0011] On the one hand, this application provides an intelligent reminder and management system for the periodic pruning of garden plants based on the Internet of Things and growth prediction. This system performs intelligent pruning decisions and collaborative management of garden plants based on multimodal perception and dynamic knowledge graphs, including interconnected communication connections: The multimodal fusion sensing and real-time diagnosis subsystem for plant physiological status is used to simultaneously collect multi-scale morphological data and multi-dimensional physiological time-series data of plants through a visible-infrared spectral collaborative imaging array, a near-ground microenvironment sensing network and a UAV mobile observation node. Based on graph convolutional networks, it models the topological and functional relationships between plant organs and outputs a comprehensive plant health status vector that integrates morphological abnormalities, physiological stress and potential diseases in real time. The plant growth dynamic prediction and pruning knowledge graph subsystem coupled with environmental perturbations constructs a dynamic knowledge graph with individual plants as entities, pruning operations as intervention events, and environmental factors as influence edges. It also integrates a time series prediction model based on attention mechanism to simulate the future growth trajectory and morphological evolution path of plants under different environmental perturbations and historical pruning interventions. The multi-objective game-based adaptive generation subsystem for pruning schemes is used to receive the health state vector and future growth trajectory, and use the pruning rules and aesthetic constraints in the dynamic knowledge graph as prior knowledge to construct an optimization model with the game objectives of maximizing landscape aesthetic consistency, optimizing plant physiological recovery, and minimizing maintenance resource consumption. The model is solved by a multi-agent reinforcement learning algorithm and outputs a personalized pruning strategy and execution timing that matches the current plant state and environmental scenario. The intelligent task decomposition and dynamic scheduling subsystem for human-machine-object collaboration is used to parse the personalized pruning strategy into an executable atomic task sequence, and dynamically schedule three heterogeneous intelligent agents—autonomous pruning robots, drone-assisted pruning devices, and maintenance personnel equipped with augmented reality interactive interfaces—to work together based on the task's requirements for accuracy, load, and intelligence. At the same time, it makes online adjustments to the task sequence and resource allocation based on real-time feedback. The closed-loop driven quantitative evaluation and self-evolution subsystem for pruning effect is used to re-collect data through the multimodal fusion perception subsystem after the pruning operation is completed, and compare the 3D point cloud model, hyperspectral features and microenvironment parameters before and after pruning to perform multi-dimensional quantitative evaluation of the pruning effect and generate evaluation results. The evaluation results are used to drive the iterative update of rules and weights in the dynamic knowledge graph, and also serve as new training samples input into the time series prediction model and optimization model to achieve continuous self-optimization of the entire system.

[0012] As a preferred embodiment of this application, in the multimodal fusion perception and real-time diagnosis subsystem for plant physiological state, the input of the graph convolutional network is a topological graph constructed based on the three-dimensional point cloud model of the plant. The node features include the color, temperature and spectral reflectance of the corresponding part. The edge weights are jointly determined by the structural connection strength and the correlation of physiological signal transmission. Through multi-layer message passing, the global impact assessment and root cause tracing of local abnormal states are realized.

[0013] As a preferred embodiment of this application, in the plant growth dynamic prediction and pruning knowledge graph subsystem coupled with environmental disturbances, the dynamic knowledge graph adopts a temporal knowledge graph structure, allowing the same entity to have different state attributes and relationships in different time slices; the time series prediction model is a Transformer architecture that integrates external environment memory units, which uses historical meteorological data and soil data sequences as external memory and performs cross-modal attention calculations with the plant's own growth sequence to predict the potential impact of environmental mutations on growth trends.

[0014] As a preferred embodiment of this application, in the multi-objective game-based adaptive generation subsystem for pruning schemes, the multi-agent reinforcement learning algorithm includes three types of agents: aesthetic agents, physiological agents, and resource agents, each focusing on learning to achieve its corresponding game objective; the system encodes the environmental state through a central coordinator and guides each agent to reach an equilibrium strategy in competition and cooperation, and this equilibrium strategy is the output pruning scheme.

[0015] As a preferred embodiment of this application, in the intelligent task decomposition and dynamic scheduling subsystem for human-machine-object collaboration, the dynamic scheduling process specifically includes: for fine pruning tasks requiring high-precision positioning, priority is given to scheduling autonomous pruning robots equipped with visual servoing; for operations on tall trees or inaccessible areas, drone-assisted pruning devices equipped with robotic arms are scheduled for high-altitude collaboration; for scenarios requiring complex judgment or protective pruning, virtual pruning guide lines, tissue health status heat maps, and suggested cutting angles are overlaid on the maintenance personnel through an augmented reality interactive interface to achieve human-machine hybrid intelligent decision-making.

[0016] As a preferred embodiment of this application, in the closed-loop driven quantitative evaluation of pruning effect and system self-evolution subsystem, the multi-dimensional quantitative evaluation includes: Morphological compliance assessment based on 3D point cloud difference calculation measures the degree of conformity between the actual pruning form and the expected aesthetic goal. Assessment of physiological function recovery based on changes in the ratio of chlorophyll fluorescence to photosynthetically active radiation absorption; Assessment of habitat microclimate optimization based on microenvironment sensor data (such as canopy ventilation and light transmittance); The aforementioned evaluation indicators are integrated into a comprehensive benefit score, which is used to weight and adjust the confidence of the corresponding pruning rules in the dynamic knowledge graph.

[0017] On the other hand, this application also provides a smart reminder and management method for periodic pruning of garden plants based on the Internet of Things and growth prediction, which is executed by the aforementioned system and includes the following steps: S1: Simultaneously collect multi-scale morphological and physiological time-series data of the target plant through a multimodal collaborative sensing network; S2: The collected data is modeled based on graph convolutional networks to diagnose and output a fusion vector of plant health status; S3: Query historical and rule information in the dynamic knowledge graph and use time series prediction models to simulate the growth trajectory of plants under current health status and future environmental disturbances; S4: Taking landscape aesthetics, plant physiology and resource consumption as game objectives, multi-agent reinforcement learning is used to generate the optimal personalized pruning strategy. S5: Decompose the pruning strategy into atomic tasks and dynamically schedule heterogeneous agents to perform them collaboratively; S6: After pruning, the system re-perceives the effects through multi-dimensional comparison and quantitative evaluation, and feeds the evaluation results back to the knowledge graph and prediction model to drive the system's self-evolution.

[0018] As a preferred embodiment of this application, in step S4, when generating the pruning strategy, a virtual pruning inference module based on generative adversarial networks is also introduced to quickly generate visual previews of possible results of multiple pruning schemes in virtual space, assisting managers in making final decisions.

[0019] As a preferred embodiment of this application, in step S5, when performing collaborative operations, digital twins of all participating intelligent agents are established, real-time simulation and collision detection of task execution are carried out in virtual space, and the predicted risks are mapped to the physical world in advance for early warning and intervention.

[0020] Thirdly, the present invention provides an electronic device including a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described above.

[0021] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the method described above.

[0022] Compared with existing technologies, the implementation of this invention can effectively solve the technical problems in garden plant pruning management, such as reliance on experience, delayed response, inefficient resource scheduling, and lack of adaptive capabilities, and has achieved the following significant technical progress and beneficial effects: 1. Significantly improves the depth and accuracy of plant health status monitoring and diagnosis. Traditional methods mainly rely on visible light images or manual inspections, making it difficult to effectively quantify and analyze the internal physiological state and abnormal correlations of plants. This invention achieves simultaneous acquisition of multidimensional physiological time-series data of plants by deploying a visible light-infrared collaborative imaging array and a near-ground microenvironment sensing network. Furthermore, it uses graph convolutional networks (GCNs) to model the topological and functional relationships between plant organs, enabling the system to not only identify apparent morphological abnormalities but also analyze the propagation paths and potential causes of physiological stress at the mechanistic level. This technique overcomes the shortcomings of existing technologies in detecting latent diseases and complex environmental stresses, providing a reliable data foundation for early warning and precise intervention.

[0023] 2. This invention represents a fundamental shift from fixed-cycle pruning to dynamic, personalized pruning based on growth prediction. Addressing the limitations of existing pruning plans, which are rigid and unable to adapt to individual differences and dynamic environmental changes, this invention constructs a growth simulation subsystem that integrates a temporal knowledge graph and an attention mechanism prediction model. This system can dynamically simulate the growth trajectory of a specific plant by combining historical data, real-time health status, and predictions of future environmental disturbances (such as extreme weather). Based on this, pruning decisions no longer rely on a fixed schedule but are scientifically generated according to the actual growth needs and future morphological evolution trends of each plant. This effectively avoids growth inhibition or resource waste caused by inappropriate pruning timing, thus improving the scientific rigor and foresight of plant maintenance.

[0024] 3. This invention solves the technical challenge of automatically generating multi-objective optimization pruning schemes. Landscape pruning needs to consider multiple objectives, including aesthetics, ecology, and cost, which traditional methods struggle to quantify and balance. This invention creatively introduces a decision-making framework based on multi-agent reinforcement learning. By establishing agents focused on landscape aesthetics, plant physiological restoration, and minimizing resource consumption, and designing a central coordinator to guide their game, the final output achieves a pruning scheme that reaches Pareto optimality or Nash equilibrium. This technical solution is the first in the field of landscape maintenance to achieve automated and optimal solutions for pruning strategies under multiple constraints, providing a systematic technical solution for complex decision-making scenarios.

[0025] 4. A highly efficient and flexible human-machine-object collaborative operation system has been constructed, significantly improving operational capabilities and safety in complex scenarios. To address the challenges of unstructured garden environments and diverse task types, this invention proposes a heterogeneous intelligent agent dynamic scheduling method. This system can automatically decompose tasks and optimally match execution units such as autonomous robots, drones, or AR-assisted personnel based on the task's requirements for accuracy, accessibility, and intelligence. Simultaneously, by establishing a digital twin model of the operation process for real-time simulation and collision warning, it achieves advanced risk perception and proactive intervention in the physical operation process. This collaborative system breaks through the capability boundaries of single automated equipment, forming a complementary operation mode that significantly reduces operational risks in dangerous scenarios such as high altitudes and confined spaces while improving operational efficiency and coverage.

[0026] 5. A closed-loop feedback and self-evolution mechanism for continuous self-optimization of the driving system was established. Unlike most open-loop management systems, this invention designs a self-evolutionary subsystem centered on multi-dimensional quantitative evaluation. This subsystem objectively and quantitatively evaluates the pruning effect by comparing high-precision 3D point clouds, spectral features, and microenvironment parameters before and after pruning. The evaluation results are not only used for acceptance testing but also fed back as training data to the knowledge graph (updating rule weights) and the prediction model (for incremental learning). This mechanism enables the entire system to continuously learn and improve from practice, adapting to new plant varieties, regional climates, and maintenance standards, greatly enhancing the system's long-term applicability, robustness, and intelligence.

[0027] 6. By employing virtual simulation and visualization pre-visualization technologies, the predictability and controllability of management decisions are enhanced. This invention integrates a virtual pruning and inference module based on Generative Adversarial Networks (GANs) into the decision chain, which can visualize the expected effects of different solutions before execution, assisting managers in making scientific comparisons. Combined with digital twin technology for full-process simulation of operations, it achieves "pre-judgment" of potential risks and "a priori verification" of solution feasibility. These technological applications elevate traditional experience-based, trial-and-error management to a data-driven, simulation-optimized, and precise management model, effectively reducing decision-making error rates and project implementation risks. Attached Figure Description

[0028] 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.

[0029] Figure 1 This is a structural block diagram of the intelligent reminder and management system for periodic pruning of garden plants based on the Internet of Things and growth prediction, as described in this invention. Figure 2 This is a flowchart illustrating the intelligent reminder and management method for periodic pruning of garden plants based on the Internet of Things and growth prediction, as described in this invention. Detailed Implementation

[0030] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0031] Example 1 like Figure 1 As shown, this application provides an intelligent reminder and management system for the periodic pruning of garden plants based on the Internet of Things and growth prediction. This system performs intelligent pruning decisions and collaborative management of garden plants based on multimodal perception and dynamic knowledge graphs, including interconnected systems: 1. A multimodal fusion sensing and real-time diagnosis subsystem for plant physiological status is used to simultaneously collect multi-scale morphological data and multi-dimensional physiological time-series data of plants through a visible-infrared spectral collaborative imaging array, a near-ground microenvironment sensor network and a UAV mobile observation node. Based on graph convolutional networks, it models the topological and functional relationships between plant organs and outputs a comprehensive plant health status vector that integrates morphological abnormalities, physiological stress and potential diseases in real time. The input of the graph convolutional network is a topological graph constructed based on the three-dimensional point cloud model of the plant. The node features include the color, temperature and spectral reflectance of the corresponding part. The edge weights are jointly determined by the structural connection strength and the correlation of physiological signal transmission. Through multi-layer message passing, the global impact assessment and root cause tracing of local abnormal states are realized.

[0032] The implementation process of the multimodal fusion sensing and real-time diagnosis subsystem for plant physiological status includes: This subsystem serves as the system's data entry point and state awareness layer. During implementation, a visible-infrared dual-spectrum imaging array (fixed nodes) and a near-ground microenvironment sensor network (monitoring soil temperature, humidity, light intensity, etc.) are first deployed according to the plan within the target garden area. Simultaneously, mobile observation nodes equipped with autonomous path planning capabilities (UAVs) are configured as a supplement. All nodes synchronize time and transmit data through an IoT gateway.

[0033] When the system starts its monitoring cycle, the fixed imaging array synchronously acquires visible light and infrared images of the target plant at a preset frequency (e.g., once per hour). Unmanned aerial vehicle (UAV) nodes, triggered by system commands or preset rules (e.g., after detecting a local anomaly), fly to the designated area to collect supplementary data from multiple angles and at close range. Microenvironment sensors continuously collect time-series data.

[0034] The acquired raw data (dual-light images, sensor readings) is transmitted to the edge computing server. The server first performs data preprocessing and spatiotemporal alignment: using feature point matching and sensor fusion algorithms, the visible light, infrared images, and sensor data at the same time are aligned at the pixel level and spatial position.

[0035] The aligned data was used to construct a 3D point cloud topology map of the plant: a 3D point cloud model of the plant was generated using binocular vision or LiDAR (if configured) data, and the point cloud was segmented into "super nodes" corresponding to organs such as the trunk, main branches, and leaf clusters. Feature vectors were extracted for each node, including the average color (RGB, HSV) from the visible light image, the surface temperature distribution from the infrared image, and the spectral reflectance index (such as NDVI, PRI) retrieved from hyperspectral data (if available). The "edges" between nodes were assigned initial weights based on the connectivity relationships in 3D space (structural connections) and the correlation coefficients of temporal sensor data changes (physiological signal transduction correlations).

[0036] This topological graph serves as input to a pre-trained graph convolutional network (GCN). This GCN model, through a multi-layer message passing mechanism, allows node features to aggregate and update along edges. For example, the high temperature and low pigmentation characteristics exhibited by a local leaf node will transmit signals of "heat stress" and "decreased activity" to its connected branch nodes via edge weights. After multi-layer propagation, the network outputs the "abnormal probability" of each node and a global "health state vector." This vector is a multi-dimensional feature that quantifies the overall state of the entire plant in terms of morphological symmetry, water stress index, and potential disease probability. The output of this vector provides accurate and structured initial state input for subsequent growth prediction and decision-making.

[0037] 2. A plant growth dynamic prediction and pruning knowledge graph subsystem coupled with environmental perturbations. It constructs a dynamic knowledge graph with individual plants as entities, pruning operations as intervention events, and environmental factors as influence edges. It also integrates a time series prediction model based on attention mechanism to simulate the future growth trajectory and morphological evolution path of plants under different environmental perturbations and historical pruning interventions. The dynamic knowledge graph adopts a temporal knowledge graph structure, which allows the same entity to have different state attributes and relationships in different time slices. The time series prediction model is a Transformer architecture that integrates external environment memory units. It uses historical meteorological data and soil data sequences as external memory and performs cross-modal attention calculation with the plant's own growth sequence to predict the potential impact of environmental mutations on growth trends.

[0038] The implementation process of the plant growth dynamics prediction and pruning knowledge graph subsystem coupled with environmental perturbations includes: This subsystem serves as the system's "brain" and knowledge base. Its implementation begins with the construction of a dynamic knowledge graph. The graph uses individual plants as uniquely identified core entities, whose attributes include species, age, historical pruning records, and historical disease records. Around this entity, entities such as "pruning" events (recording time, method, and executor), "meteorological" environments (temperature, precipitation, and light), and "soil" environments are created. Entities are connected through semantic relationship edges such as "exerting influence," "occurring at," and "causing." The key innovation lies in employing a temporal knowledge graph structure, assigning a valid timestamp to each entity-relationship relationship, thereby recording dynamic knowledge such as "Plant A was sensitive to water stress in the summer of 2023."

[0039] Upon receiving the current health status vector from subsystem 1, the system first retrieves historical similar status segments and records of the effects of pruning interventions at that time, using the plant entity as a starting point. Simultaneously, it retrieves the predicted data sequence of its growing environment for the next 7-15 days from external meteorological and soil databases.

[0040] This information, along with the current health status vector, is input into a Transformer prediction model that incorporates external environment memory units. The model's encoder uses the plant's own historical growth sequence (e.g., height, crown width changes) as the primary sequence, while simultaneously using external environmental sequences (temperature, precipitation) as external memory. At the attention computation layer, the model not only calculates the temporal dependencies within the growth sequence but also, through a cross-modal attention mechanism, calculates the potential impact weights of future environmental conditions (e.g., "high temperature tomorrow") on the plant's growth state (e.g., "branch elongation rate"). The model decoder ultimately outputs the predicted trajectory of the plant's key morphological indicators (e.g., height, crown width, leaf area index) over a future period (e.g., the next 30 days), as well as simulations of multi-branch evolutionary paths under different hypothetical pruning interventions. This predicted trajectory and evolutionary path provide crucial future scenario data support for the next subsystem's decision on "when to prune and how much to prune."

[0041] 3. A multi-objective game-based adaptive generation subsystem for pruning schemes is used to receive the health state vector and future growth trajectory, and use the pruning rules and aesthetic constraints in the dynamic knowledge graph as prior knowledge to construct an optimization model with the game objectives of maximizing landscape aesthetic consistency, optimizing plant physiological recovery, and minimizing maintenance resource consumption. The model is solved by a multi-agent reinforcement learning algorithm and outputs a personalized pruning strategy and execution timing that matches the current plant state and environmental scenario. The multi-agent reinforcement learning algorithm includes three types of agents: aesthetic agents, physiological agents, and resource agents, each focusing on learning to achieve its corresponding game objective. The system encodes the environmental state through a central coordinator and guides each agent to reach an equilibrium strategy through competition and cooperation. This equilibrium strategy is the output pruning scheme.

[0042] The implementation process of the adaptive generation subsystem for pruning schemes based on multi-objective game theory includes: This subsystem is the core of the system's decision-making process. Its implementation is a dynamic optimization solution process. The system receives the future growth trajectory from subsystem 2 (especially the "natural growth" trajectory without intervention) and the current health state vector from subsystem 1.

[0043] First, the system retrieves pruning rules and aesthetic constraints related to the plant species and current morphology from the dynamic knowledge graph (such as "trees should maintain a prominent trunk" and "hedges should have a flat top"), and converts them into quantifiable mathematical constraints.

[0044] Then, the system constructs a multi-objective optimization problem. There are three objective functions: F1 (aesthetic): based on the predicted trajectory, calculate and minimize the difference between the post-pruning morphology and the target aesthetic template (such as a specific geometric shape); F2 (physiological): based on the health state vector, evaluate and maximize the gain of the proposed pruning location (such as removing diseased or weak branches) on the overall physiological recovery (such as improving ventilation and light penetration); F3 (resource): estimate and minimize the time, energy consumption, and labor costs required to complete the pruning plan.

[0045] To address this multi-objective game problem, a multi-agent reinforcement learning environment is deployed. The environment state is encoded by the current plant health status, predicted growth trajectory, and available resources. Three agents (aesthetic agent, physiological agent, and resource agent) each possess independent policy networks, aiming to learn and maximize their respective long-term rewards (F1, F2, F3). A central coordinator observes the global state and generates a shared "latent state" signal, guiding agents to consider not only their own objectives but also the possible reactions of other agents when taking action (i.e., suggesting the specific location and intensity of pruning).

[0046] Through extensive offline and online simulation training, the agents learn to cooperate in competition. When faced with a specific decision request, the three agents output their respective strategies based on their current states. A central coordinator synthesizes these strategies and outputs a final, balanced pruning plan through an arbitration mechanism (such as weighted voting or a Nash equilibrium solver). This plan specifies the exact location of the pruning (e.g., "prune 20% of the branches and leaves below the third main branch on the southeast side"), the recommended pruning amount, and the suggested execution time window. This plan is the single, explicit set of instructions connecting the upper-level decision-making with the lower-level execution.

[0047] 4. A human-machine-object collaborative intelligent task decomposition and dynamic scheduling subsystem is used to parse the personalized pruning strategy into an executable atomic task sequence, and dynamically schedule three heterogeneous intelligent agents—autonomous pruning robots, drone-assisted pruning devices, and maintenance personnel equipped with augmented reality interactive interfaces—to work together based on the task's requirements for accuracy, load, and intelligence. At the same time, the task sequence and resource allocation are adjusted online based on real-time feedback. The dynamic scheduling process specifically includes: for fine pruning tasks requiring high-precision positioning, priority is given to scheduling autonomous pruning robots equipped with visual servo systems; for operations on tall trees or inaccessible areas, drone-assisted pruning devices equipped with robotic arms are scheduled for high-altitude collaboration; for scenarios requiring complex judgment or protective pruning, virtual pruning guide lines, tissue health status heat maps, and suggested cutting angles are overlaid on the maintenance personnel through an augmented reality interactive interface to achieve human-machine hybrid intelligent decision-making.

[0048] The implementation process of the intelligent task decomposition and dynamic scheduling subsystem for human-machine-object collaboration includes: This subsystem is the system's "commander" and "dispatch center." Its implementation begins with receiving customized pruning schemes from subsystem 3.

[0049] First, the task decomposition module parses the solution into a series of atomic operations, such as: "Locate to coordinates (X,Y,Z)", "Identify target branch B", "Perform a cut of length L", and "Clean up fallen branches and leaves C". Each atomic task is labeled with its required execution accuracy (millimeter level, centimeter level), load requirements (torque magnitude), and required level of intelligence (whether real-time visual judgment is required).

[0050] Subsequently, the resource management and scheduling module begins operation. The system maintains a real-time updated heterogeneous intelligent agent resource pool, including: autonomous pruning robots in standby or operation (high positioning accuracy, large load capacity), drone-assisted pruning devices (high mobility, high accessibility), and maintenance personnel who log in to the system via mobile terminals (strong judgment, high flexibility).

[0051] The scheduling engine performs dynamic matching and allocation based on the tags of atomic tasks: For tasks involving "precisely locating and cutting branches with a diameter of <2cm", the task is assigned to the nearest autonomous pruning robot, and a work instruction containing three-dimensional coordinates is sent via wireless network.

[0052] For tasks such as "removing dead branches at the top of tree canopies (more than 5 meters above the ground)," a drone equipped with a robotic arm is dispatched. The system will plan a safe approach path and operating posture for the drone.

[0053] For the task of “identifying and pruning complex areas where diseased and healthy tissues are intertwined,” the system assigns the task to the maintenance personnel’s augmented reality (AR) terminal. The terminal screen will display an “tissue health status heat map” (disease areas are highlighted in red) generated by subsystem 1 and a “virtual pruning guide line” generated by subsystem 3.

[0054] The scheduling process is dynamic. For example, when a robot encounters an unexpected obstacle during its operation, it will report a "task blocked" signal. The scheduling center can immediately reschedule, potentially reassigning the remaining tasks to drones or personnel. The status, location, and task progress of all agents are visible in real time on a digital twin control panel, ensuring global controllability. This process ensures that optimal decisions are safely, accurately, and efficiently translated into actions in the physical world.

[0055] 5. A closed-loop driven quantitative evaluation and self-evolution subsystem for pruning effect is used to re-collect data through the multimodal fusion perception subsystem after the pruning operation is completed, and compare the 3D point cloud model, hyperspectral features and micro-environment parameters before and after pruning to perform multi-dimensional quantitative evaluation of the pruning effect and generate evaluation results. The evaluation results are used to drive the iterative update of rules and weights in the dynamic knowledge graph, and also serve as new training samples input into the time series prediction model and optimization model to achieve continuous self-optimization of the entire system.

[0056] The multi-dimensional quantitative assessment includes: Morphological compliance assessment based on 3D point cloud difference calculation measures the degree of conformity between the actual pruning form and the expected aesthetic goal. Assessment of physiological function recovery based on changes in the ratio of chlorophyll fluorescence to photosynthetically active radiation absorption; Assessment of habitat microclimate optimization based on microenvironment sensor data (such as canopy ventilation and light transmittance); The aforementioned evaluation indicators are integrated into a comprehensive benefit score, which is used to weight and adjust the confidence of the corresponding pruning rules in the dynamic knowledge graph.

[0057] The implementation process of the closed-loop driven quantitative evaluation of pruning effect and the system self-evolution subsystem includes: This subsystem acts as the system's "quality inspector" and "learning engine," ensuring continuous system improvement. Its implementation begins after the pruning process is complete.

[0058] First, the system triggers a post-pruning specialized sensing operation. All relevant sensing nodes (fixed imaging array, UAV) are instructed to perform a comprehensive data collection on the newly pruned plants, following the same process as subsystem 1. This yields a new 3D point cloud model, a new infrared thermal image, and new microenvironment data after pruning.

[0059] The evaluation module then initiates a multi-dimensional comparative analysis: Morphological compliance assessment: The trimmed 3D point cloud is automatically registered and differentially calculated with the expected aesthetic target model generated by subsystem 3 to quantify the accuracy of volume removal and the consistency of the overall shape.

[0060] Physiological function recovery assessment: By comparing hyperspectral data before and after pruning, we analyzed the changes in chlorophyll fluorescence parameters (Fv / Fm) and photosynthetically active radiation absorption ratio (APAR) to assess whether pruning effectively alleviated stress and promoted the recovery of photosynthetic capacity.

[0061] Habitat microclimate optimization assessment: Compare the sensor data of wind speed and light intensity in the canopy before and after pruning to assess whether permeability has been improved.

[0062] The above evaluation results are used by a fusion algorithm to calculate a comprehensive benefit score. This score, along with a detailed evaluation report, is first provided to management personnel as the basis for work acceptance.

[0063] More importantly, this score serves as a reward signal for reinforcement learning and is fed back into the multi-agent reinforcement learning environment of subsystem 3 to update the policy networks of each agent in the corresponding state, enabling them to make better decisions in similar scenarios in the future.

[0064] Simultaneously, the "rule effectiveness" information revealed in the evaluation results (e.g., "For a certain tree species, moderate pruning under a specific health condition yields the highest physiological recovery score") was used to update the dynamic knowledge graph in subsystem 2, specifically by adjusting the confidence weights of relevant pruning rules. Furthermore, the complete data chain of "initial state - intervention - final effect" from this operation was used as new training samples, added to the training datasets of the prediction model in subsystem 2 and the optimization model in subsystem 3, achieving iterative updates of the models through incremental learning. Thus, the system completed a complete closed loop from perception to learning and back to perception, achieving true intelligent self-evolution.

[0065] Compared with existing technologies, the implementation of this invention can effectively solve the technical problems in garden plant pruning management, such as reliance on experience, delayed response, inefficient resource scheduling, and lack of adaptive capabilities, and has achieved the following significant technical progress and beneficial effects: 1. Significantly improves the depth and accuracy of plant health status monitoring and diagnosis. Traditional methods mainly rely on visible light images or manual inspections, making it difficult to effectively quantify and analyze the internal physiological state and abnormal correlations of plants. This invention achieves simultaneous acquisition of multidimensional physiological time-series data of plants by deploying a visible light-infrared collaborative imaging array and a near-ground microenvironment sensing network. Furthermore, it uses graph convolutional networks (GCNs) to model the topological and functional relationships between plant organs, enabling the system to not only identify apparent morphological abnormalities but also analyze the propagation paths and potential causes of physiological stress at the mechanistic level. This technique overcomes the shortcomings of existing technologies in detecting latent diseases and complex environmental stresses, providing a reliable data foundation for early warning and precise intervention.

[0066] 2. This invention represents a fundamental shift from fixed-cycle pruning to dynamic, personalized pruning based on growth prediction. Addressing the limitations of existing pruning plans, which are rigid and unable to adapt to individual differences and dynamic environmental changes, this invention constructs a growth simulation subsystem that integrates a temporal knowledge graph and an attention mechanism prediction model. This system can dynamically simulate the growth trajectory of a specific plant by combining historical data, real-time health status, and predictions of future environmental disturbances (such as extreme weather). Based on this, pruning decisions no longer rely on a fixed schedule but are scientifically generated according to the actual growth needs and future morphological evolution trends of each plant. This effectively avoids growth inhibition or resource waste caused by inappropriate pruning timing, thus improving the scientific rigor and foresight of plant maintenance.

[0067] 3. This invention solves the technical challenge of automatically generating multi-objective optimization pruning schemes. Landscape pruning needs to consider multiple objectives, including aesthetics, ecology, and cost, which traditional methods struggle to quantify and balance. This invention creatively introduces a decision-making framework based on multi-agent reinforcement learning. By establishing agents focused on landscape aesthetics, plant physiological restoration, and minimizing resource consumption, and designing a central coordinator to guide their game, the final output achieves a pruning scheme that reaches Pareto optimality or Nash equilibrium. This technical solution is the first in the field of landscape maintenance to achieve automated and optimal solutions for pruning strategies under multiple constraints, providing a systematic technical solution for complex decision-making scenarios.

[0068] 4. A highly efficient and flexible human-machine-object collaborative operation system has been constructed, significantly improving operational capabilities and safety in complex scenarios. To address the challenges of unstructured garden environments and diverse task types, this invention proposes a heterogeneous intelligent agent dynamic scheduling method. This system can automatically decompose tasks and optimally match execution units such as autonomous robots, drones, or AR-assisted personnel based on the task's requirements for accuracy, accessibility, and intelligence. Simultaneously, by establishing a digital twin model of the operation process for real-time simulation and collision warning, it achieves advanced risk perception and proactive intervention in the physical operation process. This collaborative system breaks through the capability boundaries of single automated equipment, forming a complementary operation mode that significantly reduces operational risks in dangerous scenarios such as high altitudes and confined spaces while improving operational efficiency and coverage.

[0069] 5. A closed-loop feedback and self-evolution mechanism for continuous self-optimization of the driving system was established. Unlike most open-loop management systems, this invention designs a self-evolutionary subsystem centered on multi-dimensional quantitative evaluation. This subsystem objectively and quantitatively evaluates the pruning effect by comparing high-precision 3D point clouds, spectral features, and microenvironment parameters before and after pruning. The evaluation results are not only used for acceptance testing but also fed back as training data to the knowledge graph (updating rule weights) and the prediction model (for incremental learning). This mechanism enables the entire system to continuously learn and improve from practice, adapting to new plant varieties, regional climates, and maintenance standards, greatly enhancing the system's long-term applicability, robustness, and intelligence.

[0070] 6. By employing virtual simulation and visualization pre-visualization technologies, the predictability and controllability of management decisions are enhanced. This invention integrates a virtual pruning and inference module based on Generative Adversarial Networks (GANs) into the decision chain, which can visualize the expected effects of different solutions before execution, assisting managers in making scientific comparisons. Combined with digital twin technology for full-process simulation of operations, it achieves "pre-judgment" of potential risks and "a priori verification" of solution feasibility. These technological applications elevate traditional experience-based, trial-and-error management to a data-driven, simulation-optimized, and precise management model, effectively reducing decision-making error rates and project implementation risks.

[0071] Example 2 This invention also provides a method for intelligent reminders and management of periodic pruning of garden plants based on the Internet of Things and growth prediction. This method is based on multimodal perception and dynamic knowledge graph for intelligent pruning decision-making and collaborative management of garden plants, and is executed by the system described in Embodiment 1 above. It includes the following steps: S1: Simultaneously collect multi-scale morphological and physiological temporal data of the target plant through a multimodal collaborative sensing network; S2: The collected data is modeled based on graph convolutional networks to diagnose and output a fusion vector of plant health status; S3: Query historical and rule information in the dynamic knowledge graph and use time series prediction models to simulate the growth trajectory of plants under current health status and future environmental disturbances; S4: Taking landscape aesthetics, plant physiology and resource consumption as game objectives, multi-agent reinforcement learning is used to generate the optimal personalized pruning strategy. S5: Decompose the pruning strategy into atomic tasks and dynamically schedule heterogeneous agents to perform them collaboratively; S6: After pruning, the system re-perceives the effects through multi-dimensional comparison and quantitative evaluation, and feeds the evaluation results back to the knowledge graph and prediction model to drive the system's self-evolution.

[0072] In a preferred embodiment, in step S4, when generating the pruning strategy, a virtual pruning deduction module based on generative adversarial networks is also introduced to quickly generate visual previews of possible results of multiple pruning schemes in virtual space, assisting managers in making final decisions.

[0073] In a preferred embodiment, in step S5, when performing collaborative work, digital twins of all participating agents are established, real-time simulation and collision detection of task execution are performed in virtual space, and the predicted risks are mapped to the physical world in advance for early warning and intervention.

[0074] The steps in this embodiment are performed to implement the functions of the subsystems in Embodiment 1, and will not be described in detail here.

[0075] Example 3 The present invention also provides an electronic device, including: a processor, a transmitting device, an input device, an output device, and a memory. The processor may be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory may be implemented using a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), and is used to store computer program code. The computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any of the above possible implementation methods.

[0076] Example 4 The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0077] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0078] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A smart reminder and management system for periodic pruning of garden plants based on the Internet of Things and growth prediction, characterized in that, Including interconnected communication connections: The multimodal fusion sensing and real-time diagnosis subsystem for plant physiological status is used to simultaneously collect multi-scale morphological data and multi-dimensional physiological time-series data of plants through a visible-infrared spectral collaborative imaging array, a near-ground microenvironment sensing network and a UAV mobile observation node. Based on graph convolutional networks, it models the topological and functional relationships between plant organs and outputs a comprehensive plant health status vector that integrates morphological abnormalities, physiological stress and potential diseases in real time. The plant growth dynamic prediction and pruning knowledge graph subsystem coupled with environmental perturbations constructs a dynamic knowledge graph with individual plants as entities, pruning operations as intervention events, and environmental factors as influence edges. It also integrates a time series prediction model based on attention mechanism to simulate the future growth trajectory and morphological evolution path of plants under different environmental perturbations and historical pruning interventions. The multi-objective game-based adaptive generation subsystem for pruning schemes is used to receive the health state vector and future growth trajectory, and use the pruning rules and aesthetic constraints in the dynamic knowledge graph as prior knowledge to construct an optimization model with the game objectives of maximizing landscape aesthetic consistency, optimizing plant physiological recovery, and minimizing maintenance resource consumption. The model is solved by a multi-agent reinforcement learning algorithm and outputs a personalized pruning strategy and execution timing that matches the current plant state and environmental scenario. The intelligent task decomposition and dynamic scheduling subsystem for human-machine-object collaboration is used to parse the personalized pruning strategy into an executable atomic task sequence, and dynamically schedule three heterogeneous intelligent agents—autonomous pruning robots, drone-assisted pruning devices, and maintenance personnel equipped with augmented reality interactive interfaces—to work together based on the task's requirements for accuracy, load, and intelligence. At the same time, it makes online adjustments to the task sequence and resource allocation based on real-time feedback. The closed-loop driven quantitative evaluation and self-evolution subsystem for pruning effect is used to re-collect data through the multimodal fusion perception subsystem after the pruning operation is completed, and compare the 3D point cloud model, hyperspectral features and microenvironment parameters before and after pruning to perform multi-dimensional quantitative evaluation of the pruning effect and generate evaluation results. The evaluation results are used to drive the iterative update of rules and weights in the dynamic knowledge graph, and also serve as new training samples input into the time series prediction model and optimization model to achieve continuous self-optimization of the entire system.

2. The system according to claim 1, characterized in that, In the multimodal fusion perception and real-time diagnosis subsystem for plant physiological state, the input of the graph convolutional network is a topological graph constructed based on the three-dimensional point cloud model of the plant. The node features include the color, temperature and spectral reflectance of the corresponding part. The edge weights are jointly determined by the structural connection strength and the correlation of physiological signal transmission. Through multi-layer message passing, the global impact assessment and root cause tracing of local abnormal states are realized.

3. The system according to claim 1, characterized in that, In the plant growth dynamic prediction and pruning knowledge graph subsystem coupled with environmental perturbation, the dynamic knowledge graph adopts a temporal knowledge graph structure, which allows the same entity to have different state attributes and relationships in different time slices. The time series prediction model is a Transformer architecture that integrates external environment memory units. It uses historical meteorological data and soil data sequences as external memory and performs cross-modal attention calculation with the plant's own growth sequence to predict the potential impact of environmental mutations on growth trends.

4. The system according to claim 1, characterized in that, In the adaptive generation subsystem of pruning scheme based on multi-objective game theory, the multi-agent reinforcement learning algorithm includes three types of agents: aesthetic agents, physiological agents, and resource agents, each focusing on learning to achieve its corresponding game objective. The system encodes the environmental state through a central coordinator and guides each agent to reach an equilibrium strategy through competition and cooperation. This equilibrium strategy is the output pruning scheme.

5. The system according to claim 1, characterized in that, In the aforementioned intelligent task decomposition and dynamic scheduling subsystem for human-machine-object collaboration, the dynamic scheduling process specifically includes: for fine pruning tasks requiring high-precision positioning, priority is given to scheduling autonomous pruning robots equipped with visual servoing; for operations on tall trees or inaccessible areas, drone-assisted pruning devices equipped with robotic arms are scheduled for high-altitude collaboration; for scenarios requiring complex judgment or protective pruning, virtual pruning guide lines, tissue health status heat maps, and suggested cutting angles are overlaid on the maintenance personnel through an augmented reality interactive interface to achieve human-machine hybrid intelligent decision-making.

6. The system according to claim 1, characterized in that, In the closed-loop driven quantitative evaluation of pruning effect and system self-evolution subsystem, the multi-dimensional quantitative evaluation includes: Morphological compliance assessment based on 3D point cloud difference calculation measures the degree of conformity between the actual pruning form and the expected aesthetic goal. Assessment of physiological function recovery based on changes in chlorophyll fluorescence and photosynthetically active radiation absorption ratio; Assessment of habitat microclimate optimization based on microenvironment sensor data (such as canopy ventilation and light transmittance); The aforementioned evaluation indicators are integrated into a comprehensive benefit score, which is used to weight and adjust the confidence of the corresponding pruning rules in the dynamic knowledge graph.

7. A method for intelligent reminders and management of periodic pruning of garden plants based on the Internet of Things and growth prediction, characterized in that, Performed by the system according to any one of claims 1-6, the method includes the following steps: S1: Simultaneously collect multi-scale morphological and physiological temporal data of the target plant through a multimodal collaborative sensing network; S2: The collected data is modeled based on graph convolutional networks to diagnose and output a fusion vector of plant health status; S3: Query historical and rule information in the dynamic knowledge graph and use time series prediction models to simulate the growth trajectory of plants under current health status and future environmental disturbances; S4: Taking landscape aesthetics, plant physiology and resource consumption as game objectives, multi-agent reinforcement learning is used to generate the optimal personalized pruning strategy. S5: Decompose the pruning strategy into atomic tasks and dynamically schedule heterogeneous agents to perform them collaboratively; S6: After pruning, the system re-perceives the effects through multi-dimensional comparison and quantitative evaluation, and feeds the evaluation results back to the knowledge graph and prediction model to drive the system's self-evolution.

8. The method according to claim 7, characterized in that, In step S4, when generating the pruning strategy, a virtual pruning inference module based on generative adversarial networks is also introduced to quickly generate visual previews of possible results of various pruning schemes in virtual space, assisting managers in making final decisions.

9. The method according to claim 7, characterized in that, In step S5, when performing collaborative operations, digital twins of all participating agents are established, and real-time simulation and collision detection of task execution are carried out in virtual space. The predicted risks are mapped to the physical world in advance for early warning and intervention.

10. An electronic device comprising a processor, a memory, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 7 to 9.

Images