An intelligent garden management method and system based on artificial intelligence

Abstract

This invention relates to the field of landscape management technology, specifically to an artificial intelligence-based intelligent landscape management method and system. The method includes: collecting plant growth data, soil data, and environmental data through a multispectral imager, soil sensor network, and weather station to form a multi-source fusion dataset; performing fusion analysis on the multi-source data based on a hybrid neural network model to generate plant growth status indicators and soil health indicators; using a Transformer-based time-series prediction model to predict future growth trends and identify potential risks, establishing a prediction result database and labeling confidence levels; generating an initial management plan based on a strategy reasoning engine and a landscape management knowledge base, optimizing it using a non-dominated sorting genetic algorithm to form a dynamic management strategy including irrigation, fertilization, and pruning plans, and then distributing it to equipment for execution. This invention achieves high-precision perception, intelligent prediction, and dynamic decision-making of the landscape ecological state, improving the automation and ecological sustainability of landscape management.

Description

Technical Field

[0001] This invention relates to the field of landscape management technology, and in particular to a smart landscape management method and system based on artificial intelligence. Background Technology

[0002] With the continuous advancement of urban ecological and refined management concepts, landscaping, as an important component of the urban ecosystem, plays a crucial role in regulating climate, improving air quality, and enhancing the urban landscape. Traditional landscaping management mainly relies on manual inspections and experience-based judgment for irrigation, fertilization, and pruning. Its data sources are limited, and decision-making is lagging, making it difficult to meet the requirements of modern cities for ecological efficiency and management precision.

[0003] Existing smart garden management systems still suffer from several technical bottlenecks: First, the multi-source data collection process suffers from uneven spatial distribution and temporal asynchrony, resulting in a lack of unified correlation between plant, soil, and environmental information; second, most systems rely on a single model for state analysis, making it difficult to dynamically predict plant growth trends and soil health status and identify risks; third, management strategies are mostly based on experience or static rules, lacking comprehensive optimization of water and fertilizer consumption and ecological benefits; furthermore, the strategy execution level generally lacks feedback loops and adaptive adjustment mechanisms, making it impossible to dynamically correct based on real-time environmental changes. These problems keep garden ecological management at a stage of partial automation, making it difficult to achieve truly intelligent, systematic, and sustainable management. Summary of the Invention

[0004] This invention provides an intelligent garden management method and system based on artificial intelligence, which realizes high-precision perception, intelligent prediction and dynamic decision-making of garden ecological status, and improves the automation and ecological sustainability of garden management.

[0005] An artificial intelligence-based smart garden management method includes the following steps: S1: Collect multi-source data of the garden, including plant growth data, soil data and environmental data; S2: Based on the multi-source data, use an artificial intelligence model to perform data fusion analysis and generate plant growth status indicators and soil health indicators; S3: Based on the plant growth status indicators and soil health indicators, predict the future growth trend and potential risks of garden plants; S4: Based on the future growth trend and potential risks, generate a dynamic management strategy, which includes an irrigation plan, a fertilization plan and a pruning plan.

[0006] Optionally, S1 includes: S11: Collect image data of the plant canopy by using visible light cameras and multispectral imagers deployed in the garden area, and preprocess and extract features from the image data to obtain plant growth data including leaf area index and leaf color distribution data. S12: Collect soil physicochemical parameters through a sensor network deployed in garden soil, and standardize the physicochemical parameters to obtain soil data including soil moisture, pH value, nitrogen, phosphorus and potassium content and microbial community abundance. S13: Collect atmospheric environmental parameters by deploying meteorological stations in the garden, and calibrate and format the atmospheric environmental parameters to obtain environmental data including temperature, humidity, light intensity and precipitation data; S14: The plant growth data, soil data, and environmental data are time-aligned and spatially registered to form the multi-source data for subsequent analysis.

[0007] Optionally, in S11, the multispectral imager acquires image data including at least the red light band, near-infrared band, and green light band, and obtains the leaf area index and leaf color distribution data by calculating the vegetation index on the image data.

[0008] Optionally, in S12, the sensor network includes a soil moisture sensor, a soil pH sensor, a soil EC sensor, and a soil microbial detection sensor; the soil microbial detection sensor obtains the microbial community abundance data by detecting DNA markers of specific microorganisms in the soil.

[0009] Optionally, S2 includes: The artificial intelligence model is a hybrid neural network model, which includes a convolutional neural network for processing the plant growth data, a recurrent neural network for processing the soil data and environmental data, a feature fusion layer for fusing features, and a fully connected layer for generating the final index. S21: Input the plant growth data from the multi-source data into a convolutional neural network to extract spatial features, and simultaneously input the soil data and environmental data from the multi-source data into a recurrent neural network to extract temporal features, thereby obtaining plant spatial features and soil environmental temporal features respectively. S22: The spatial features of the plants and the temporal features of the soil environment are fused at the feature level, and the feature weights in the fusion process are adaptively adjusted through an attention mechanism to form multi-source fused features; S23: Input the multi-source fusion features into the fully connected layer for regression analysis, and simultaneously output quantitative plant growth status indicators and soil health indicators, wherein the plant growth status indicators include growth vitality index and nutrient imbalance index, and the soil health indicators include microbial activity index and fertility sustainability index.

[0010] Optionally, the convolutional neural network in S21 adopts the ResNet50 architecture and performs transfer learning on weights pre-trained on the ImageNet dataset to extract the plant spatial features from the plant growth data. The recurrent neural network in S21 adopts a three-layer LSTM structure with an input time window length of 30 days to extract the soil environmental temporal features from the soil and environmental data. The attention mechanism in S22 is a multi-head self-attention mechanism, which generates the multi-source fusion features by calculating the correlation matrix between the plant spatial features and the soil environmental temporal features. The fully connected layer in S23 contains two independent output branches. The first output branch uses the Sigmoid activation function to generate the plant growth status index, and the second output branch uses the Tanh activation function to generate the soil health index.

[0011] Optionally, S3 includes: S31: Input the plant growth status index and soil health index into the time series prediction model, and output the prediction sequence of plant growth status index and soil health index for a specified future time period by analyzing the pattern of historical index sequences. S32: Based on the plant growth status index prediction sequence and soil health index prediction sequence, combined with the preset plant growth threshold and soil health threshold, calculate the degree of deviation of each index within a specified future time period, and generate future growth trends and quantified risk probabilities. S33: Perform pattern recognition and risk classification on the quantified risk probability, classify risks exceeding the preset probability threshold as potential risks, including soil degradation risk and plant physiological stress risk, and use the future growth trend and potential risks as the final output of S3.

[0012] Optionally, the time-series prediction model in S31 adopts a Transformer-based encoder-decoder architecture, wherein the encoder is used to encode the features of historical plant growth status indicators and historical soil health indicators, and the decoder is used to generate prediction sequences of plant growth status indicators and soil health indicators for a specified future time period. The plant growth thresholds in S32 include a growth vitality index threshold and a nutrient imbalance index threshold, and the soil health thresholds include a microbial activity index threshold and a fertility sustainability index threshold. The degree of deviation is determined by calculating the Euclidean distance between the prediction sequences of plant growth status indicators and soil health indicators and the corresponding thresholds.

[0013] Optionally, S4 includes: S41: Based on the future growth trend and potential risks, the strategy reasoning engine is used to analyze and generate a set of management requirements that include irrigation requirements, fertilization requirements and pruning requirements; S42: Based on the set of management needs, call the preset garden management knowledge base to perform strategy matching and generate an initial irrigation plan, an initial fertilization plan and an initial pruning plan; S43: Perform multi-objective optimization on the initial irrigation plan, initial fertilization plan and initial pruning plan, comprehensively consider resource consumption, execution efficiency and ecological benefits, and generate the final dynamic management strategy that includes specific execution parameters and timing arrangements. The dynamic management strategy includes the irrigation plan, fertilization plan and pruning plan.

[0014] An AI-based smart garden management system, used to implement the aforementioned AI-based smart garden management method, includes the following modules: Multi-source data acquisition module: used to collect multi-source data of the garden, including plant growth data, soil data and environmental data; Data fusion and analysis module: Based on the multi-source data, it uses an artificial intelligence model to perform data fusion and analysis to generate plant growth status indicators and soil health indicators; Growth trend prediction module: used to predict the future growth trend and potential risks of garden plants based on the plant growth status indicators and soil health indicators; Dynamic strategy generation module: used to generate dynamic management strategies based on the future growth trends and potential risks, including irrigation plans, fertilization plans and pruning plans; Strategy Execution Management Module: Used to distribute the dynamic management strategy to garden automation equipment and monitor its execution status.

[0015] The beneficial effects of this invention are: This invention achieves multi-dimensional acquisition of plant growth, soil, and environmental data by deploying visible light cameras, multispectral imagers, soil sensor networks, and weather stations within a garden area. A unified multi-source fusion dataset is formed through temporal alignment and spatial registration. Combining a hybrid neural network model, ResNet50 is used to extract spatial features of plants, and a three-layer LSTM is employed to extract temporal features of the soil environment. Feature-level fusion is achieved through a multi-head self-attention mechanism, effectively eliminating spatiotemporal differences and noise interference between different data modalities. This scheme significantly improves the fusion accuracy and feature extraction capability of garden ecological data, enabling high-precision quantitative characterization of plant growth status and soil health, providing a reliable data foundation for subsequent prediction and management.

[0016] This invention uses a Transformer-based encoder-decoder time-series prediction model to predict future trends of plant growth status and soil health indicators. It calculates the quantitative risk probability of each indicator through sliding window standardization and logistic regression risk modeling. Combining deviation analysis, DBSCAN density clustering, and decision tree classification algorithms, the system can automatically identify and classify potential risks, including water stress risk, nutrient deficiency risk, and pest and disease risk. The judgment criteria for each risk type are quantitatively defined using statistical methods, ensuring objectivity and repeatability in risk identification. Furthermore, the introduction of a confidence level labeling mechanism and a structured prediction result database enables dynamic adjustment and traceable management of the prediction results, forming a complete intelligent risk assessment system capable of providing early warnings of horticultural ecological problems and guiding prevention and control.

[0017] This invention constructs a closed-loop intelligent decision-making system from risk assessment to management execution. The system, based on a rule-based strategy reasoning engine combined with plant growth and soil dynamics models, automatically generates multi-dimensional management requirements for irrigation, fertilization, pruning, soil improvement, and plant protection. It then uses a garden management knowledge base for strategy matching to generate initial plans that conform to regional policies and ecological standards. Furthermore, a non-dominated sorting genetic algorithm is employed for multi-objective optimization, using water consumption, fertilizer utilization, and canopy light transmittance as comprehensive optimization objectives to form the optimal dynamic management strategy. Finally, the system distributes the optimization results to irrigation, fertilization, and pruning equipment and monitors the execution status in real time, achieving intelligent closed-loop control of "prediction-decision-execution-feedback." This solution significantly improves the efficiency of water and fertilizer resource utilization and the automation level of ecological maintenance in gardens, exhibiting high adaptability and scalability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this 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 for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system flow according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0022] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0023] like Figure 1 As shown, an artificial intelligence-based smart garden management method includes the following steps: S1: Collect multi-source data from the garden, including plant growth data, soil data, and environmental data, specifically: S11: Multiple visible light cameras and multispectral imagers are deployed within the garden area to collect image data of the plant canopy. The visible light cameras employ high-definition image sensors with a resolution of at least 1920×1080 pixels, positioned at a height of 1.5 to 2.0 meters above the ground to capture the overall morphology of the plant canopy from a top-down perspective. The multispectral imagers operate in the 450nm to 900nm wavelength range, simultaneously acquiring reflectance data in the red, green, blue, and near-infrared bands. The acquired image data is first processed using a median filtering algorithm to remove image noise, eliminating random interference introduced by lighting variations and the image sensor. Then, a watershed algorithm is used for image segmentation, and the Canny operator is used for edge detection to accurately extract leaf regions. Based on the image segmentation results, the leaf area index is calculated using pixel statistics, and the RGB color model is converted to an HSV color model through color space conversion to extract leaf color distribution data. Finally, the processed leaf area index and leaf color distribution data together constitute plant growth data, reflecting the growth status and spectral characteristics of the plant canopy.

[0024] S12: Deploy a sensor network in the garden soil to collect soil physicochemical parameters. The sensor network includes a soil moisture sensor, a soil pH sensor, a soil EC sensor, and a soil microbial detection sensor. The soil moisture sensor uses the time-domain reflectometry principle with a sampling frequency of 1 Hz to detect soil volumetric water content. The soil pH sensor uses the glass electrode method, with a measurement range of 0 to 14, to characterize soil acidity and alkalinity. The soil EC sensor detects soil conductivity to reflect changes in nitrogen, phosphorus, and potassium ion concentrations. The soil microbial detection sensor is based on quantitative real-time PCR (qPCR) technology, obtaining microbial community abundance data by detecting the concentration of DNA markers of specific microorganisms. The collected physicochemical parameters are standardized using the Z-Score standardization method, converting each parameter into a dimensionless value with a mean of 0 and a standard deviation of 1, eliminating differences caused by different dimensions. After standardization, soil data containing soil moisture, pH, nitrogen, phosphorus, and potassium content, and microbial community abundance data are generated, providing a unified scale basis for subsequent model input.

[0025] S13: Deploy a weather station within the garden to collect atmospheric environmental parameters. The weather station includes a temperature sensor, a humidity sensor, a light intensity sensor, and a rain gauge. The temperature sensor uses a platinum resistance thermometer (Pt100 type), with a measurement range of -20℃ to 60℃, for accurate monitoring of air temperature; the humidity sensor uses a capacitive humidity measuring element, with a measurement range of 0% to 100% relative humidity, to reflect changes in air humidity; the light intensity sensor is based on a silicon photodiode design, with a measurement wavelength range of 300nm to 1100nm, for recording solar irradiance; the rain gauge uses a tipping bucket structure to record precipitation. All sensors are connected to the garden data acquisition terminal via a LoRa wireless communication module to achieve remote data transmission and synchronous sampling. The collected atmospheric environmental parameters are first calibrated. The calibration process is as follows: the collected raw parameters are compared with reference data provided by a metrologically certified benchmark meteorological station in the same area, and a linear regression model is used to correct the systematic bias to obtain calibrated data. Then, the data is formatted to convert all parameters into a sequence format indexed by timestamps, forming environmental data containing temperature, humidity, light intensity and precipitation data.

[0026] S14: Perform time alignment and spatial registration on the plant growth data, soil data, and environmental data. In the time alignment step, all data are based on a unified Coordinated Universal Time (UTC) as the time reference, and the sampling records from different devices are uniformly calibrated through a timestamp synchronization mechanism. For data with inconsistent sampling frequencies, a linear interpolation method is used to fill in missing data points to ensure temporal continuity. In the spatial registration step, the geographic coordinate information (using the WGS-84 coordinate system) recorded by the GPS positioning module of each acquisition device during deployment is used to map data from different sources to a unified garden geographic coordinate grid structure. After registration, a multi-source fusion dataset containing plant growth characteristics, soil physicochemical properties, and environmental meteorological elements is formed. This dataset possesses complete spatiotemporal attributes, providing a reliable data foundation for subsequent artificial intelligence models to perform fusion analysis, state identification, and trend prediction.

[0027] S2: Based on the aforementioned multi-source data, an artificial intelligence model is used to perform data fusion analysis to generate plant growth status indicators and soil health indicators, specifically: Before training and inference of the artificial intelligence model, data cleaning and outlier handling are performed on the input multi-source data. First, missing values ​​in the multi-source data are imputed using the K-nearest neighbor algorithm. The K-nearest neighbor algorithm calculates the Euclidean distance between samples, selects the K nearest sample points (K is set to 5) as neighborhood samples, and estimates the missing values ​​based on the average or weighted average of the neighborhood samples, thereby ensuring the continuity and statistical rationality of the data.

[0028] Subsequently, the Isolation Forest algorithm was used to detect and remove outliers from the multi-source data. The Isolation Forest algorithm constructs multiple tree-structured models by randomly selecting features and setting random segmentation thresholds. The degree of anomaly is determined by the length of the isolated path of a sample within each tree; samples with shorter path lengths are considered outliers. After outlier removal, the resulting dataset exhibits high consistency and low noise, providing a stable data foundation for subsequent feature extraction and fusion.

[0029] S21: Input the processed multi-source data into the artificial intelligence model. The artificial intelligence model is a hybrid neural network model, which includes convolutional neural networks, recurrent neural networks, feature fusion layers, and fully connected layers. The steps are as follows: Plant growth data is input into a convolutional neural network (CNN) using the ResNet50 architecture. Transfer learning is performed on weights pre-trained on the ImageNet dataset to reduce training time and improve feature extraction accuracy. The input layer receives plant canopy image data of size 224×224×3. The network structure includes residual blocks, batch normalization layers, and ReLU activation function layers. Through multiple convolutional operations, texture, morphological, and spectral distribution features of the plant canopy are extracted step-by-step. The output feature map is processed by a global average pooling layer to obtain a 2048-dimensional plant spatial feature vector.

[0030] Soil and environmental data are input into a recurrent neural network (RNN) using a three-layer LSTM structure. The input time window is 30 days long, and each time step includes eight parameters: soil moisture, pH, nitrogen, phosphorus, and potassium content, microbial community abundance, temperature, humidity, light intensity, and precipitation. Each LSTM layer contains 256 hidden neurons, and temporal dependencies are captured using a backpropagation algorithm. The hidden state vector of the output layer is normalized and used as the temporal feature representation of the soil environment.

[0031] Ultimately, the S21 output contains two types of high-dimensional feature representations: plant spatial features and soil environmental temporal features, which serve as inputs for subsequent fusion.

[0032] S22: The spatial features of the plants and the temporal features of the soil environment are input into the feature fusion layer, and a multi-head self-attention mechanism is used for feature-level fusion. This mechanism achieves cross-modal association modeling by calculating the correlation matrix between the two feature vectors.

[0033] Specifically, the spatial characteristics of plants are represented as a matrix. The temporal characteristics of the soil environment are denoted as a matrix. and By calculating the attention weight matrix: ; in, The feature dimension is constant. Eight parallel attention heads are used, each independently calculating its parallel attention weights before concatenation to obtain a multi-source fusion feature vector. This fusion mechanism can dynamically allocate the weight ratios between different data modalities, enabling plant growth characteristics and soil environmental characteristics to form a synergistic response relationship in spatial and temporal dimensions, thereby enhancing the model's comprehensive representation ability of the garden ecosystem. The fused multi-source feature dimension is 4096.

[0034] S23: The multi-source fusion features are input into the fully connected layer for regression analysis to generate plant growth status indicators and soil health indicators. The fully connected layer contains two independent output branches: The first output branch outputs plant growth status indicators through the Sigmoid activation function, specifically including the growth vitality index and the nutrient imbalance index. The second output branch outputs soil health indicators through the Tanh activation function, specifically including the microbial activity index and the fertility sustainability index.

[0035] To ensure comparability of different indicators under a unified dimension, the output results are standardized. A min-max standardization method is used to map all indicator values ​​to the [0,1] interval, ensuring consistency across different indicators in subsequent evaluations. The standardization formula is: ; in, These are the original index values. The standardized index value, and These represent the minimum and maximum values ​​of the corresponding indicators in the sample set. The standardized results form a set of normalized indicators, providing a unified input for generating dynamic management strategies.

[0036] S24: Establish associations between standardized plant growth status indicators and soil health indicators and their corresponding multi-source data, and store them in the indicator database. The database structure includes an indicator identifier field, a timestamp field, a spatial location information field, and an associated data index field. Add timestamp information (UTC format) and spatial location information (using the WGS-84 coordinate system) to each indicator record to form spatiotemporally associated indicator records.

[0037] This database supports time-series queries and geographic location searches, enabling it to track changes in plant growth status and soil health in garden ecosystems across different times and spatial regions, providing high-precision data support for subsequent trend prediction and management strategy optimization.

[0038] S3: Based on the plant growth status indicators and soil health indicators, predict the future growth trend and potential risks of garden plants, specifically as follows: Before time series forecasting, the plant growth status indicators and soil health indicators are standardized using a sliding window to eliminate the impact of seasonal fluctuations on the model's predictive performance.

[0039] The z-score normalization method was used, selecting plant growth status and soil health indicators from the past 30 days as a sliding window. For each time point within the window, the mean value of the corresponding indicator was calculated. and standard deviation Standardize according to the following formula: ; in, The original indicator value at the current time point. These are the standardized indicator values.

[0040] This sliding window shifts the time series in one-day increments, generating a continuous, standardized data sequence. This process effectively reduces model input bias caused by climate cycle variations or seasonal fertilization differences, providing stable input features for subsequent time-series forecasting models.

[0041] S31: A time-series prediction model using a Transformer-based encoder-decoder architecture. The encoder is responsible for extracting time-dependent features from historical plant growth status indicators and historical soil health indicators, while the decoder is used to generate prediction sequences of plant growth status indicators and soil health indicators for a specified future time period.

[0042] The encoder consists of six stacked self-attention layers and a feedforward network. Each self-attention layer employs a multi-head attention mechanism, with eight attention heads per layer and a hidden dimension of 512. The input vector is first positionally encoded to preserve temporal order information. The decoder has the same structure as the encoder but additionally includes a masked self-attention layer to prevent the model from using future information.

[0043] The input is a standardized index sequence from a sliding window over the past 90 days, and the output is a predicted sequence for the next 30 days. The model is trained using the mean squared error (MSE) loss function. ; in, This represents the true indicator value. These are the model's predicted values; Parameters were updated using the Adam optimization algorithm with an initial learning rate of 1e-4 and a batch size of 64. The prediction results, including plant growth status index prediction sequences and soil health index prediction sequences, served as input for the next stage of analysis.

[0044] S32: Based on the plant growth status index prediction sequence and soil health index prediction sequence obtained from S31, the deviation degree is calculated by combining the preset plant growth threshold and soil health threshold.

[0045] Plant growth thresholds include growth vigor index thresholds and nutrient imbalance index thresholds; soil health thresholds include microbial activity index thresholds and fertility sustainability index thresholds.

[0046] The method for calculating the degree of deviation is as follows: For each predicted time point Calculate the Euclidean distance between the predicted value of the indicator and the corresponding threshold: ; in, For the first The predicted value of each indicator at time t. The threshold corresponding to this indicator. For the number of indicators; The greater the calculated deviation, the more likely the plant growth status or soil health condition is to deviate from the ideal range.

[0047] Based on this, a logistic regression model is used to calculate the probability that each indicator will exceed the safe range at each future time point. The probability output of the logistic regression model is defined as follows: ; in, For time points Quantitative risk probability, and These are the model parameters, obtained through training with historical risk event samples.

[0048] The resulting quantitative risk probability sequence reflects the risk level of the garden ecosystem across different indicator dimensions over the future time period, providing a probabilistic basis for identifying potential risks.

[0049] S33: Pattern recognition analysis is performed on the quantified risk probability. A density-based clustering algorithm is used to cluster the distribution of risk probability over time. By setting a minimum sample size of 5 and a neighborhood radius of 0.3, density clustering is performed on the risk probability curve. In the clustering results, continuous high-risk areas are identified as high-density areas, corresponding to periods of sustained high risk.

[0050] When the average risk probability over a certain period exceeds 0.6, it is marked as a potential risk.

[0051] Subsequently, a risk type discrimination model based on decision trees is used to further classify potential risks. This decision tree model uses predicted sequences of plant growth status indicators and predicted sequences of soil health indicators as input features, and classifies risks according to the feature splitting criterion (maximizing information gain ratio). The specific risk classification criteria are defined as follows: 1. Criteria for determining water stress risk: When the plant growth vitality index decreases significantly and the soil moisture is low, it is determined to be at risk of water stress.

[0052] A significant decline is defined as: the plant growth vigor index falling below its historical mean for three consecutive time windows (denoted as ). Subtract 1.5 times, the standard deviation (denoted as ) ),Right now ; Low moisture content is defined as soil moisture continuously below 90% of the lower limit of the suitable moisture range for more than 48 hours. For example, for lawn plants, when the volumetric moisture content is below 18% (the suitable range is 20% to 35%), it is considered low moisture content.

[0053] When the above two conditions are met, output the water stress risk and record the risk start time and duration.

[0054] 2. Criteria for identifying nutrient deficiency risk: When the nutrient imbalance index is high and the fertility sustainability index is low, it is considered to be at risk of nutrient deficiency.

[0055] An elevated level is defined as: a nutritional imbalance index higher than its historical average (denoted as ). Add 1.0 standard deviation (denoted as ) ),Right now And it remains above this threshold for three consecutive sampling periods; A decline in the fertility sustainability index is defined as follows: the fertility sustainability index shows a monotonically decreasing trend over five consecutive time points, with a relative decrease exceeding 10%, i.e.: ; When the above two conditions are met, the risk of nutrient deficiency is output, and the corresponding indicator deviation rate is recorded in the database.

[0056] 3. Criteria for determining the risk of pests and diseases: When the microbial activity index fluctuates significantly and the growth vitality index decreases at the same time, it is determined to be a risk of pests and diseases.

[0057] Significant fluctuation is defined as the coefficient of variation of the microbial activity index over the past 30 days. ,like If the fluctuation is significant, then it is considered to be significant. The mean, Standard deviation; Simultaneous decline is defined as follows: if the growth vitality index meets the above significant decline conditions and the decline lasts for more than two sampling periods (e.g., it does not recover for 48 consecutive hours), it is judged as "simultaneous decline".

[0058] When the above two conditions are met, output the risk of pests and diseases, and mark the risk as a result of a decline in ecological stability.

[0059] After identifying each type of potential risk, the corresponding risk time period, risk level (based on the average level of quantified risk probability), and triggering indicators are stored in the risk classification record table. The risk level classification standard is as follows: Risk probability It is classified as a high-risk level. It is classified as a medium-risk level. It is classified as a low-risk level.

[0060] Furthermore, after risk classification is completed, the confidence index for each risk event is automatically calculated. The confidence index is determined by both model prediction accuracy and data completeness, and is calculated using a weighted average method, defined as follows: ; in, For the final confidence level, The prediction accuracy of the model on the validation set. This is the percentage of data completeness for the input indicator.

[0061] when A score of 0.85 is marked as a high confidence level. When 0.7 At 0.85, it is marked as a medium confidence level; when At a confidence level of 0.7, it is marked as a low confidence level.

[0062] S4: Based on the aforementioned future growth trends and potential risks, generate a dynamic management strategy, which includes irrigation, fertilization, and pruning plans, specifically: S41: Based on future growth trends and potential risks, the strategy reasoning engine analyzes and generates a set of management requirements, including irrigation, fertilization, pruning, soil improvement, and plant protection needs.

[0063] The strategy reasoning engine adopts a rule-based reasoning system structure, which consists of three parts: a knowledge rule base, a fact database, and an inference engine.

[0064] The knowledge rule base stores the logical rules of plant growth models and soil dynamics models. Plant growth models describe the functional relationship between plant photosynthesis, transpiration, and nutrient absorption rates to infer changes in plant water and nutrient requirements; soil dynamics models describe the interactions between soil moisture, permeability, and fertilizer migration processes.

[0065] The fact database stores future growth trends and potential risk data output by S3, including predicted sequences of plant growth status indicators, predicted sequences of soil health indicators, quantified risk probabilities, and risk types.

[0066] The inference engine performs a forward reasoning process, matching facts with rules, and triggering the corresponding rules to generate management requirements when the conditions are met.

[0067] For example, when the future growth trend shows that the growth vitality index continues to decline and the quantified risk probability is greater than 0.6, the system triggers "increase irrigation demand" according to the rules; when the fertility sustainability index shows a downward trend and the risk of soil degradation is identified, the system triggers "generate soil improvement demand"; when the risk of pests and diseases is identified and the microbial activity index fluctuates significantly, the system triggers "generate plant protection demand".

[0068] The set of management requirements output by the inference process is represented in a structured format, with fields including requirement type (irrigation, fertilization, pruning, soil improvement, plant protection), triggering conditions, requirement intensity level, and time constraints.

[0069] S42: Based on the set of management requirements generated in S41, the system calls the preset garden management knowledge base to perform strategy matching and generate an initial irrigation plan, an initial fertilization plan, and an initial pruning plan.

[0070] The garden management knowledge base consists of the following three databases: Crop water requirement database: records the water requirement per unit time, transpiration coefficient, and climate condition correction coefficient of different plant species at different growth stages; Fertilizer application standard database: includes nutrient requirements, recommended fertilization ratios, fertilization cycles, and soil fertility correction parameters for different plant types; Plant pruning standard database: records the pruning cycle, pruning intensity range and canopy light transmittance requirements for plant varieties.

[0071] During the strategy matching process, the system compares each requirement parameter in the management requirement set with the standard parameters in the database mentioned above.

[0072] For example, for irrigation needs, the system matches the closest water requirement curve from the crop water requirement pattern database based on the plant growth stage, regional evaporation and soil moisture threshold, and calculates the irrigation amount. For fertilization needs, the system selects the corresponding fertilizer type and concentration from the fertilizer application standard database based on the degree of nitrogen, phosphorus and potassium nutrient deficiency and the target fertilizer utilization rate. For pruning needs, the system matches appropriate pruning intensity and shape parameters from the plant pruning specification database based on canopy light transmittance and morphological indicators.

[0073] The strategy matching process also incorporates a compliance filtering mechanism. The system inputs water resource quota policies, fertilizer use regulations, and ecological protection requirements for the area where the garden is located as constraints into the comparison module, retaining only solutions that comply with these policies and standards. The final output initial plan includes: Initial irrigation plan: Preliminarily determine the time, amount, and method of each irrigation session; Initial fertilization plan: Determine the type of fertilizer, fertilization time, fertilization amount and fertilization method; Initial pruning plan: Determine the pruning time, pruning intensity, and pruning shape.

[0074] S43: Perform multi-objective optimization on the initial irrigation plan, initial fertilization plan and initial pruning plan to obtain a dynamic management strategy with optimal comprehensive benefits.

[0075] The optimization process employs a non-dominated sorting genetic algorithm, with optimization objectives including: minimizing water consumption, maximizing fertilizer utilization, and maximizing plant canopy light transmittance.

[0076] The algorithm implementation process includes the following steps: Encoding phase: The main control parameters (irrigation amount, fertilizer amount, pruning intensity) of irrigation, fertilization and pruning plans are encoded into individual genes; Population initialization: Generate an initial population containing 100 candidate solutions; Fitness evaluation: Calculate the fitness value for each candidate solution based on the three optimization objectives; Non-dominated sorting and crowding calculation: Candidate solutions are hierarchically sorted according to Pareto dominance and the distance between solutions is calculated to maintain population diversity; Crossover and mutation operations: New solutions are generated using simulated binary crossover and polynomial mutation operators, with a crossover probability of 0.9 and a mutation probability of 0.1. Iteration Termination: Stop iterating when the rate of change of the optimal solution is less than 1% for ten consecutive generations.

[0077] The final Pareto optimal solution set is automatically filtered by an expert system to generate the final dynamic management strategy.

[0078] The final strategy includes: Irrigation plan: Specific time, amount, and method of each irrigation; Fertilization plan: fertilizer type, fertilization time, fertilization amount and fertilization method; Pruning plan: pruning time, pruning intensity, and pruning shape.

[0079] The optimization results are updated dynamically. When new future growth trend data is input, the system can rerun the optimization process to achieve adaptive adjustment of the strategy.

[0080] S44: The system decomposes the dynamic management strategy generated in S43 into a sequence of instructions that can be directly executed by the equipment, and sends them to the corresponding irrigation equipment, fertilization equipment, and pruning equipment.

[0081] The execution instruction sequence includes the device number, execution time, target parameter value, and control mode fields.

[0082] During execution, the system monitors device status and environmental changes in real time through edge nodes. Monitoring data includes execution progress, device power, ambient temperature and humidity, soil moisture, and light intensity. If an execution anomaly is detected (such as abnormal device power, flow deviation exceeding 10%, or sudden changes in environmental conditions), the system immediately triggers the dynamic management strategy regeneration process, re-running steps S41 to S43 to update the strategy parameters.

[0083] The execution monitoring module also sends the timestamp of each execution result, execution deviation value, and environmental response result back to the central database for model self-learning and policy optimization closed-loop updates.

[0084] like Figure 2 As shown, an AI-based smart garden management system, used to implement the aforementioned AI-based smart garden management method, includes the following modules: Multi-source data acquisition module: used to collect multi-source data of the garden, including plant growth data, soil data and environmental data; Data fusion and analysis module: Based on the multi-source data, it uses an artificial intelligence model to perform data fusion and analysis to generate plant growth status indicators and soil health indicators; Growth trend prediction module: used to predict the future growth trend and potential risks of garden plants based on the plant growth status indicators and soil health indicators; Dynamic strategy generation module: used to generate dynamic management strategies based on the future growth trends and potential risks, including irrigation plans, fertilization plans and pruning plans; Strategy Execution Management Module: Used to distribute the dynamic management strategy to garden automation equipment and monitor its execution status.

[0085] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0086] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent garden management based on artificial intelligence, characterized in that, Includes the following steps: S1: Collect multi-source data of the garden, including plant growth data, soil data and environmental data; S2: Based on the multi-source data, use an artificial intelligence model to perform data fusion analysis and generate plant growth status indicators and soil health indicators; S3: Based on the plant growth status indicators and soil health indicators, predict the future growth trend and potential risks of garden plants; S4: Based on the future growth trend and potential risks, generate a dynamic management strategy, which includes an irrigation plan, a fertilization plan and a pruning plan. 2.The AI-based intelligent garden management method of claim 1, wherein, S1 includes: S11: Collect image data of the plant canopy by using visible light cameras and multispectral imagers deployed in the garden area, and preprocess and extract features from the image data to obtain plant growth data including leaf area index and leaf color distribution data. S12: Collect soil physicochemical parameters through a sensor network deployed in garden soil, and standardize the physicochemical parameters to obtain soil data including soil moisture, pH value, nitrogen, phosphorus and potassium content and microbial community abundance. S13: Collect atmospheric environmental parameters by deploying meteorological stations in the garden, and calibrate and format the atmospheric environmental parameters to obtain environmental data including temperature, humidity, light intensity and precipitation data; S14: The plant growth data, soil data, and environmental data are time-aligned and spatially registered to form the multi-source data for subsequent analysis.

3. The intelligent garden management method based on artificial intelligence according to claim 2, characterized in that, In step S11, the multispectral imager acquires image data including at least the red light band, near-infrared band, and green light band. By calculating the vegetation index from the image data, the leaf area index and leaf color distribution data are obtained.

4. The artificial intelligence-based intelligent garden management method of claim 3, wherein, In step S12, the sensor network includes a soil moisture sensor, a soil pH sensor, a soil EC sensor, and a soil microbial detection sensor; the soil microbial detection sensor obtains the microbial community abundance data by detecting DNA markers of specific microorganisms in the soil.

5. The intelligent garden management method based on artificial intelligence according to claim 4, characterized in that, S2 includes: The artificial intelligence model is a hybrid neural network model, which includes a convolutional neural network for processing the plant growth data, a recurrent neural network for processing the soil data and environmental data, a feature fusion layer for fusing features, and a fully connected layer for generating the final index. S21: Input the plant growth data from the multi-source data into a convolutional neural network to extract spatial features, and simultaneously input the soil data and environmental data from the multi-source data into a recurrent neural network to extract temporal features, thereby obtaining plant spatial features and soil environmental temporal features respectively. S22: The spatial features of the plants and the temporal features of the soil environment are fused at the feature level, and the feature weights in the fusion process are adaptively adjusted through an attention mechanism to form multi-source fused features; S23: Input the multi-source fusion features into the fully connected layer for regression analysis, and simultaneously output quantitative plant growth status indicators and soil health indicators, wherein the plant growth status indicators include growth vitality index and nutrient imbalance index, and the soil health indicators include microbial activity index and fertility sustainability index. 6.The AI-based intelligent garden management method of claim 5, wherein, The convolutional neural network in S21 adopts the ResNet50 architecture and performs transfer learning on weights pre-trained on the ImageNet dataset to extract the plant spatial features from the plant growth data. The recurrent neural network in S21 adopts a three-layer LSTM structure with an input time window of 30 days to extract the soil environmental temporal features from the soil and environmental data. The attention mechanism in S22 is a multi-head self-attention mechanism, which generates the multi-source fusion features by calculating the correlation matrix between the plant spatial features and the soil environmental temporal features. The fully connected layer in S23 contains two independent output branches. The first output branch uses the Sigmoid activation function to generate the plant growth status index, and the second output branch uses the Tanh activation function to generate the soil health index.

7. The artificial intelligence-based intelligent garden management method of claim 6, wherein, S3 includes: S31: Input the plant growth status index and soil health index into the time series prediction model, and output the prediction sequence of plant growth status index and soil health index for a specified future time period by analyzing the pattern of historical index sequences. S32: Based on the plant growth status index prediction sequence and soil health index prediction sequence, combined with the preset plant growth threshold and soil health threshold, calculate the degree of deviation of each index within a specified future time period, and generate future growth trends and quantified risk probabilities. S33: Perform pattern recognition and risk classification on the quantified risk probability, classify risks exceeding the preset probability threshold as potential risks, including soil degradation risk and plant physiological stress risk, and use the future growth trend and potential risks as the final output of S3. 8.The AI-based intelligent garden management method of claim 7, wherein, The time-series prediction model in S31 adopts a Transformer-based encoder-decoder architecture, where the encoder is used to encode the features of historical plant growth status indicators and historical soil health indicators, and the decoder is used to generate prediction sequences of plant growth status indicators and soil health indicators for a specified future time period. The plant growth thresholds in S32 include a growth vitality index threshold and a nutrient imbalance index threshold, and the soil health thresholds include a microbial activity index threshold and a fertility sustainability index threshold. The degree of deviation is determined by calculating the Euclidean distance between the prediction sequences of plant growth status indicators and soil health indicators and the corresponding thresholds. 9.The AI-based intelligent garden management method of claim 8, wherein, S4 includes: S41: Based on the future growth trend and potential risks, the strategy reasoning engine is used to analyze and generate a set of management requirements that include irrigation requirements, fertilization requirements and pruning requirements; S42: Based on the set of management needs, call the preset garden management knowledge base to perform strategy matching and generate an initial irrigation plan, an initial fertilization plan and an initial pruning plan; S43: Perform multi-objective optimization on the initial irrigation plan, initial fertilization plan and initial pruning plan, comprehensively consider resource consumption, execution efficiency and ecological benefits, and generate the final dynamic management strategy that includes specific execution parameters and timing arrangements. The dynamic management strategy includes the irrigation plan, fertilization plan and pruning plan.

10. An artificial intelligence-based smart garden management system for implementing an artificial intelligence-based smart garden management method according to claim 9, characterized in that, Includes the following modules: Multi-source data acquisition module: used to collect multi-source data of the garden, including plant growth data, soil data and environmental data; Data fusion and analysis module: Based on the multi-source data, it uses an artificial intelligence model to perform data fusion and analysis to generate plant growth status indicators and soil health indicators; Growth trend prediction module: used to predict the future growth trend and potential risks of garden plants based on the plant growth status indicators and soil health indicators; Dynamic strategy generation module: used to generate dynamic management strategies based on the future growth trends and potential risks, including irrigation plans, fertilization plans and pruning plans; Strategy Execution Management Module: Used to distribute the dynamic management strategy to garden automation equipment and monitor its execution status.

Images