A comprehensive health monitoring method and system for ancient trees based on IoT and AI technologies

Abstract

This invention proposes a comprehensive ancient tree health monitoring method and system based on IoT and AI technologies. The method includes: collecting physiological data of ancient trees through sensors and preprocessing it; extracting disturbance indicators from the millimeter-wave radar point cloud frame data through edge computing to represent the intensity of behavioral disturbances at the current time step; initializing the ancient tree health state variable as the current health state and inputting it into a pre-constructed dynamic recursive model to obtain the predicted ancient tree health state variable; sampling the millimeter-wave radar point cloud frame data once per second and performing pre-clustering processing; constructing a graph neural network based on the target point cloud set to encode and extract the disturbance activation degree representing the current behavioral risk; correcting the disturbance indicator according to the disturbance activation degree and outputting the disturbance amount; calculating a comprehensive risk score and dynamically generating corresponding health response actions based on the comprehensive risk score.

Description

Technical Field

[0001] This invention belongs to the field of comprehensive ancient tree health monitoring, and in particular relates to a comprehensive ancient tree health monitoring method and system based on Internet of Things and AI technologies. Background Technology

[0002] Ancient trees, as a key component of the ecosystem, possess extremely high biological, cultural, and environmental value. However, due to their long growth cycle, fragile structure, and slow physiological regulation mechanisms, they are highly susceptible to irreversible health damage after being subjected to external disturbances or environmental fluctuations. Current ancient tree monitoring mainly relies on manual inspections and single-point environmental sensing devices, such as temperature and humidity recorders and image monitoring. These methods are not only sparse in sampling and have delayed responses, but also fail to depict the physiological evolution processes within the tree, and cannot capture latent decline signals such as decreased water conduction efficiency, obstructed transpiration, or changes in tissue conductivity. Furthermore, in unattended forest environments, illegal approaches, climbing, bark harvesting, and sawing are difficult to identify and intervene in a timely manner. Traditional monitoring systems, which rely heavily on video recognition, are limited by insufficient light in the forest canopy and unstable power supply, hindering large-scale deployment. More importantly, existing systems cannot model the tree's condition and link it to the impact of behavioral changes, resulting in a lack of a causal explanation chain between health changes and external disturbances. This prevents predictive judgment and responsive decision-making, limiting the ability of ancient tree management to transition from passive response to proactive protection.

[0003] Therefore, there is an urgent need for an integrated intelligent monitoring system that is adapted to the complex environment under the forest canopy, can continuously sense multiple physiological parameters, identify abnormal behaviors, has predictive capabilities, and can respond automatically, so as to meet the core needs of early identification, accurate judgment, and rapid response in the protection of ancient trees. Summary of the Invention

[0004] The purpose of this invention is to propose a comprehensive health monitoring method and system for ancient trees based on IoT and AI technologies. By deploying multi-source sensors at different structural levels of the tree to collect key physiological and environmental parameters such as temperature, transpiration humidity, electrical conductivity and light intensity, and combining them with a time-series structural modeling mechanism, the system can predict the evolution trend of the tree before its health status shows obvious degradation.

[0005] To achieve the above objectives, a comprehensive ancient tree health monitoring method based on Internet of Things and AI technologies is provided in a first aspect of the present invention, the method comprising the following steps:

[0006] S1. Collect physiological data of ancient trees through sensors and preprocess them to construct a four-dimensional time-synchronized structured input vector, while simultaneously collecting millimeter-wave radar point cloud frame data;

[0007] Based on the millimeter-wave radar point cloud frame data, disturbance indices are extracted from it through edge computing to represent the intensity of behavioral disturbances at the current time step;

[0008] S2. Initialize the ancient tree health status variable as the current health status, and combine it with the disturbance index to input the pre-constructed dynamic recursive model to obtain the predicted ancient tree health status variable.

[0009] S3. Sample the millimeter-wave radar point cloud frame data once per second and perform pre-clustering processing to obtain a target point cloud; construct a graph neural network based on the target point cloud to encode and extract the perturbation activation degree representing the current behavior risk, and correct the perturbation index according to the perturbation activation degree to output the perturbation amount.

[0010] S4. Combine the predicted ancient tree health status variables, disturbance amount, and current health status to calculate a comprehensive risk score, and dynamically generate corresponding response actions based on the comprehensive risk score.

[0011] Furthermore, the collection of physiological data of ancient trees via sensors includes:

[0012] Thermistor patches are attached to the inside of the bark of the ancient tree trunk to collect bark temperature; transpiration humidity is measured by attaching a moisture-sensitive polymer sensor to the trunk; bark conductivity is collected using a surrounding microelectrode array; and light intensity is collected by a PAR sensor in the canopy layer.

[0013] The preprocessing includes:

[0014] The physiological data of the ancient trees are uniformly converted into standard time steps using linear interpolation, and the resulting aligned vector is used as the input vector for four-dimensional time synchronization structure.

[0015] Furthermore, S2 specifically includes:

[0016] Define the current health status variable of the ancient tree as a scalar, representing the overall health level, with a value range of [0,1], where the smallest value represents the least healthy;

[0017] The update function of the dynamic recursive model is constructed by combining the current ancient tree health status variables, structured input vectors, and perturbation indicators to obtain the predicted ancient tree health status variables.

[0018] The dynamic recursive model is used to perform a time recursive mechanism on the current ancient tree health status variables, the predicted structured input vector, and the perturbation index to obtain the predicted health status result for the next τ steps.

[0019] Furthermore, the predicted structured input vector and perturbation index are obtained by extrapolating environmental trends.

[0020] Furthermore, the step of sampling the millimeter-wave radar point cloud frame data once per second and performing pre-clustering processing to obtain the target point cloud specifically includes:

[0021] The millimeter-wave radar point cloud frame data is pre-clustered based on sampling once per second to obtain the target point cloud set. Each of them This represents the set of points distributed in space for a moving object.

[0022] Furthermore, a graph neural network is constructed based on the target point set to encode and extract the perturbation activation degree representing the current behavioral risk, and the perturbation index is corrected according to the perturbation activation degree to output the perturbation amount, specifically including:

[0023] Construct a spatial-temporal graph neural network G for a point cloud t =(V t E t ), where node V t Gathering at target point O t The centroid of the target point cluster in the middle, edge E t It represents the similarity of trajectories or the continuity of motion between frames at different times, and is used to capture behavioral trends;

[0024] Introducing graph convolutional units (GCNs) to G t Encode the perturbation activations to extract the perturbation activations that represent the risk of the current behavior; wherein the perturbation activations are nonnormalized perturbation scores.

[0025] A health-state-sensitive adjustment function is introduced to map and normalize the perturbation activation, and finally output the perturbation amount; wherein the perturbation amount is obtained by dynamically correcting the health state at the current moment.

[0026] Furthermore, the health status sensitive adjustment function is designed with an adjustment coefficient to dynamically correct the disturbance amount according to the current health status, so as to represent the sensitivity of the health status to the disturbance. The largest adjustment coefficient indicates the worst health.

[0027] Furthermore, S4 specifically includes:

[0028] Get the historical cumulative response, which represents the integral result of responses triggered in the past H hours;

[0029] The downward trend magnitude of the health status is determined based on the current health status and the predicted health status variables of the ancient trees to represent the trend risk; the disturbance risk is determined based on the disturbance amount; the severity of the future status is determined based on the predicted health status variables to represent the severity of the predicted health status; and the response suppression penalty is determined based on the historical response accumulation to prevent repeated responses in a short period of time from causing system overload and misjudgment.

[0030] A comprehensive risk score is determined based on the aforementioned trend risk, disturbance risk, future state severity, and response inhibition penalty.

[0031] The comprehensive risk score is adjusted by vulnerability weighting based on the current health status to obtain the final comprehensive risk score.

[0032] Response actions are triggered based on the final comprehensive risk score.

[0033] Furthermore, the triggering of the response action based on the final comprehensive risk score specifically includes:

[0034] Design a first risk classification threshold and a second risk classification threshold;

[0035] If the final comprehensive risk score is less than the first risk classification threshold, a log recording will be triggered.

[0036] If the final comprehensive risk score is greater than the first risk classification threshold and less than the second risk classification threshold, then the structured risk report is uploaded through the LoRa or cellular network connection platform and a low-power infrared camera is called to perform directional shooting.

[0037] If the final comprehensive risk score is greater than the second risk classification threshold, the buzzer is controlled and the high-brightness LED driver is activated. At the same time, the sensor sampling period is changed from 5 minutes to 1 minute, and the situation is reported in real time.

[0038] A second aspect of the invention provides a comprehensive ancient tree health monitoring system based on Internet of Things and AI technologies, the system comprising:

[0039] The signal acquisition unit is used to collect physiological data of ancient trees through sensors and perform preprocessing to construct a four-dimensional time-synchronized structured input vector, while simultaneously acquiring millimeter-wave radar point cloud frame data.

[0040] Based on the millimeter-wave radar point cloud frame data, disturbance indices are extracted from it through edge computing to represent the intensity of behavioral disturbances at the current time step;

[0041] The health prediction unit is used to initialize the ancient tree health status variables as the current health status, and combine them with the disturbance index, inputting them into a pre-built dynamic recursive model to obtain the predicted ancient tree health status variables.

[0042] The health correction unit is used to sample the millimeter-wave radar point cloud frame data once per second and perform pre-clustering processing to obtain a target point cloud; construct a graph neural network based on the target point cloud to encode and extract the perturbation activation degree representing the current behavior risk, and correct the perturbation index according to the perturbation activation degree and output the perturbation amount.

[0043] A health warning unit is used to calculate a comprehensive risk score by combining the predicted ancient tree health state variables, disturbance quantities, and the current moment health state, and dynamically generate corresponding response actions based on the comprehensive risk score.

[0044] The beneficial technical effects of the present invention are at least as follows:

[0045] The present invention constructs an intelligent monitoring system for ancient trees with predictive health modeling as the core, behavior disturbance identification as the driving force, and dynamic risk assessment and response as a closed loop. By deploying multi-source sensors at different structural levels of the tree body to collect key physiological and environmental parameters such as temperature, transpiration humidity, conductivity, and light, and combining with the time series structure modeling mechanism, the system can predict the evolution trend of the ancient tree before the obvious degradation of its health state.

[0046] Furthermore, this system innovatively introduces a millimeter-wave point cloud behavior perception module to extract potential damage behaviors such as approaching and staying and convert them into disturbance variables to be injected into the health model, enabling external behaviors to quantitatively affect the state judgment process and establishing a response channel between the decline of the tree body and human behaviors. When the health trend is downward or the behavior disturbance increases, the system jointly calculates the risk score based on the current state and the prediction result, and realizes differential and controllable linkage responses through the vulnerability adjustment and response suppression mechanisms, such as triggering image acquisition, sound and light alarms, remote reporting and other operations, so as to achieve a complete closed loop from perception to judgment to action.

[0047] The invention solution打通了数据采集、建模、识别、响应之间的通路,显著提升了古树监测系统的灵敏性、解释性与前瞻性,解决了传统方法中状态难预测、行为难识别、响应难分级的系统性问题。 Brief Description of the Drawings

[0048] The present invention is further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to the following drawings without creative efforts.

[0049] Figure 1 It is a flowchart of an all-round ancient tree health monitoring method based on the Internet of Things and AI technologies of the present invention. Detailed Embodiments

[0050] The embodiments of the present invention are described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention.

[0051] Such as Figure 1As shown in the embodiment of the present invention, a comprehensive ancient tree health monitoring method based on IoT and AI technologies is provided. The method includes:

[0052] S1. Collect physiological data of ancient trees through sensors and preprocess them to construct a four-dimensional time-synchronized structured input vector, while simultaneously collecting millimeter-wave radar point cloud frame data;

[0053] Based on the millimeter-wave radar point cloud frame data, disturbance indices are extracted from it through edge computing to represent the intensity of behavioral disturbances at the current time step.

[0054] Specifically, this step constructs the foundational sensing layer for ancient tree health modeling. The goal is to design a data acquisition scheme that can operate stably in the wild forest environment and provide high-frequency, modelable physiological data, from a practical deployment perspective. Ancient trees, as the target, possess physiological structural characteristics such as slow response, large size, complex bark structure, and inaccessible root systems. Conventional single-point environmental data acquisition methods are insufficient to reflect the true physiological changes of the tree, especially internal signals such as transpiration, water conduction, and electrical conductivity. Therefore, this scheme starts with two key areas: the trunk and the crown, collecting representative physiological and driving data from each. Considering subsequent modeling requirements, all sensor data must be organized into time-step data vectors according to a unified temporal sequence. It also possesses precise spatial deployment logic and synchronization mechanisms.

[0055] Furthermore, the data acquisition scheme of this invention revolves around the ancient tree as the main subject, selecting the following four types of key indicators to construct the time step input vector:

[0056]

[0057] Where t represents the unified time step, in minutes. All indicators must be collected every 5 minutes, with an error tolerance of no more than ±30 seconds to ensure synchronization.

[0058] T t The bark temperature is measured using a thermistor patch, which is attached to the inside of the bark at a distance of 1.3 meters from the trunk of the ancient tree. The installation method involves gently peeling off the surface layer of bark before attaching the patch. The temperature measurement range is -20℃ to 60℃, with an accuracy of ±0.3℃, and the unit is Celsius.

[0059] H t Evapotranspiration humidity was measured using a moisture-sensitive polymer sensor attached 1.5 meters north of the trunk, and the unit is %RH. Data was sampled at a frequency of 1Hz, and the average value over 5 minutes was calculated to generate H2O. t .

[0060] C t Bark electrical conductivity was measured using a surround microelectrode array and obtained via an AC method. The unit is μS / cm. The measurement cycle is 5 minutes, and a stable average value is output.

[0061] I t Light intensity, measured by a PAR sensor installed in the canopy layer, is expressed in μmol / m². 2 / s. Used to reflect the external driving force of evaporation.

[0062] Furthermore, the acquisition results are uniformly converted to a standard time step t using linear interpolation to generate an aligned vector:

[0063]

[0064] Where d i ∈{T,H,C,I}, where Δt = 2 minutes is the time tolerance. Interpolation is used to compensate for time step misalignment caused by differences in sensor sampling periods.

[0065] For example, when the system needs to generate If the latest data for a certain sensor is 10:04 and the next data is 10:06, the system will fill in the sensor's value at 10:05 through linear interpolation.

[0066] In addition, the point cloud data P required for behavior recognition t Real-time data acquisition was performed using millimeter-wave radar, with the device positioned 1 meter outside the base of the ancient tree, covering a 60° field of view, and a sampling frequency of 1Hz. Each frame... t It includes the number, distance, velocity, and spatial distribution characteristics of target reflection points, in the format of a multi-dimensional point cloud matrix, which is standardized for subsequent processing.

[0067] Output: Four-dimensional time-synchronized structured input vectors are used for modeling; P t Millimeter-wave radar point cloud frame data, used for behavior recognition.

[0068] S2. Initialize the ancient tree health status variable as the current health status, and combine it with the disturbance index to input the pre-constructed dynamic recursive model to obtain the predicted ancient tree health status variable.

[0069] Specifically, the core task of this step is to utilize the structured input vector obtained in step one. And point cloud data P t The behavioral perturbation factor δ extracted from t This study aims to construct a modeling system for the health status of ancient trees that is interpretable, physiologically plausible, and capable of responding to external disturbances. Changes in the health status of ancient trees exhibit slow, delayed, and invisible characteristics, making it difficult for traditional methods such as time-series classifiers or simple anomaly detection models to meet practical conservation needs such as early prediction, causal judgment, and continuous modeling.

[0070] Furthermore, to address the aforementioned problems, this invention introduces a low-dimensional state variable x.t (Representing the overall health status of ancient trees) serves as an intermediate representation, and a system combining sensor-driven information is established. and external disturbance δ t The dynamic recursive model M, namely Furthermore, considering the data sparsity and suddenness of human interference in the ancient tree scenario, this invention innovatively incorporates a state prediction mechanism with behaviorally sensitive adjustment terms and physiological fluctuation regularization terms into the model structure, thereby achieving the ability to both predict the downward trend of ancient tree health in advance and cope with state mutations caused by abnormal disturbances.

[0071] This step further obtains P. t The data is millimeter-wave radar point cloud data, from which the disturbance index δ is extracted using edge computing. t Its value is δ t ∈[0,1] represents the intensity of behavioral disturbance at the current time step (such as a weighted index formed by the frequency and duration of behaviors such as approaching and stopping, sawing, etc.).

[0072] Furthermore, the core of ancient tree health modeling is to analyze a continuous hidden state variable x. t The goal is to model and predict physiological dynamics to ensure they reflect real-world physiological dynamics and possess predictive capabilities. This invention employs a recursive neural network structure to achieve this objective, and the following design is implemented:

[0073] (1) Define the health status variable x of ancient trees t This is a scalar value representing the overall health level, with a value range of [0,1]. A smaller value indicates poorer health. The initial value can be normalized by mapping from historical expert scores.

[0074] (2) Introduce an update function M that combines environmental changes, disturbance risks, and health inertia, with the following recursive form:

[0075]

[0076] Where, x t Current health status; Structured input data, a sequence of a time window, enters the GRU (Gated Recurrent Unit), and outputs the hidden state; GRU(·): a 1-layer GRU network, the input is a sequence of length T=12 (i.e., the past 1 hour). Sequence, output 8-dimensional compressed vector; δ t : Disturbance factor, determined by P t Calculation (e.g., continuous stationary radar target presence time > 30 seconds is considered high disturbance); W1, W2, W3: linear weights learned during training; σ(·): Sigmoid function used for normalization; λ is the historical average; λ is the fluctuation penalty coefficient, which controls the repeated fluctuations of the state and reflects the slow-change inertia physiological characteristics that the health of ancient trees should have.

[0077] It is important to note that the innovative aspect of this formula lies in:

[0078] Introducing W3δ t This item enables dynamic modeling of disturbance response-driven state changes, meaning that the system allows x to change when the behavioral disturbance intensifies. t Faster downlink response;

[0079] Increase Regularization terms are used to prevent frequent reversals of state due to slight environmental fluctuations, ensuring that state changes conform to the physiological slow change mechanism and solving the problem of allergic false alarms in the health assessment of ancient trees.

[0080] Past state x t With the characteristics of the current comprehensive environment Joint modeling reflects the evolution trend of the system over time.

[0081] (3) State prediction structure:

[0082] Based on x t The system can construct a predicted value x using a time-recursive mechanism. t+τ :

[0083]

[0084] Where M (τ) This represents the recursive calculation of the model over τ future time steps. and δ [t:t+τ] This serves as the input for estimating future sequences (which can be obtained through extrapolation of environmental trends).

[0085] Output: x t : The estimated health status of ancient trees at the current time step, with normalized indices; x t+τ Predict the health status outcome in the next τ steps (e.g., 3 hours later) for risk scoring and response strategy invocation.

[0086] S3. Sample the millimeter-wave radar point cloud frame data once per second and perform pre-clustering processing to obtain a target point cloud; construct a graph neural network based on the target point cloud to encode and extract the perturbation activation degree representing the current behavior risk, and correct the perturbation index according to the perturbation activation degree to output the perturbation amount.

[0087] Specifically, this step aims to extract the millimeter-wave radar point cloud data P collected in step one. t The perturbation factor δ, which reflects the intensity of human interference, was identified and quantified. tAs input, the health status modeling model in step two is injected. Unlike traditional camera-based identification methods, ancient tree monitoring often takes place in forest areas without electricity, light, or obstruction, making it difficult to deploy or maintain stable long-term operation of visual sensors. This step uses point cloud data as the core input, combining spatial structure, target motion state, and temporal characteristics of behavior to extract disturbance quantities representing human behavior risks. The system needs to identify potential destructive behaviors such as approaching, lingering, felling, shaking, and observing under unattended conditions, and output a continuous value δ. t ∈[0,1], used as external dynamic disturbance input in the health model. To adapt to the uncertainty and ambiguity of point cloud data and behavior, this step designs a disturbance modeling method based on point cloud dynamic graph convolutional network, and constructs a disturbance integral mechanism in combination with time continuity.

[0088] Furthermore, point cloud data P t Pre-clustering is performed based on sampling once per second to obtain the target point cloud. Each of them This invention represents a spatially distributed set of points representing a moving object. It constructs a point cloud spatial-temporal graph structure G. t =(V t E t ), where node V t For O t The centroid of the target point cluster in the middle, edge E t It represents the similarity of trajectories or the continuity of motion between frames at different times, and is used to capture behavioral trends.

[0089] This invention introduces graph convolutional units (GCNs) to G... t Encode the perturbation activation s representing the risk of the current behavior. t :

[0090] s t =ReLU(W g ·GCN(G t-29:t )+b)

[0091] Wherein: GCN(G t-29:t The W represents the feature vector obtained after performing two layers of graph convolution on a 30-second action graph, from which spatial and temporal coherence is extracted; g and b are the weights and bias parameters, learned from the training set; ReLU guarantees a non-negative output, representing the perturbation strength; s t It is a non-normalized perturbation fraction (real value, range indefinite).

[0092] Subsequently, considering that behavioral perturbations may have differentiated effects on ancient trees in different health states (e.g., ancient trees in poor health are more vulnerable to the same behavior), this invention introduces a health-state-sensitive adjustment function f(x) t The perturbation values ​​are mapped and normalized to obtain the final perturbation amount:

[0093] δ t =σ(α·s t ·f(x t ))

[0094] Where σ(·) is the Sigmoid function, mapping the perturbation to [0,1]; f(x t )=1+γ(1-x t ), representing the sensitivity of health status to disturbances; γ is the adjustment coefficient, with a larger amplification coefficient as health deteriorates; α is the disturbance standardization coefficient; x t From step two, used for dynamic correction of s t .

[0095] Understandably, the creativity of this structure is reflected in two aspects: First, behavior recognition does not use explicit classification, but captures changes in behavior patterns using a point cloud structure, which is suitable for unstructured scenarios; second, the calculation of disturbance quantities incorporates health state variables, enabling disturbance feedback to respond to individual health backgrounds and solving the problem of decoupling disturbances from risks in existing systems.

[0096] This step, based on the non-visual, low-power, and temporally discontinuous conditions in ancient tree protection, designs a behavior perturbation modeling mechanism specifically for millimeter-wave point clouds. It models point cloud behavior trends using graph convolution, replacing traditional classification-based recognition, and is structurally more adaptable to the actual forest area scenario; it also utilizes a health status correlation function f(x) t The perturbation intensity was adjusted, and a dynamic correction mechanism for individual sensitivity responses was introduced.

[0097] S4. Combine the predicted ancient tree health status variables, disturbance amount, and current health status to calculate a comprehensive risk score, and dynamically generate corresponding response actions based on the comprehensive risk score.

[0098] Specifically, this step is based on the health status prediction results x obtained in the previous stage. t+τ and disturbance factor δ tThis invention constructs a response mechanism with predictive foresight, behavioral context understanding, dynamic inhibition, and differentiated linkage capabilities. Unlike traditional forestry monitoring methods that rely on threshold triggering and SMS alerts, this patent emphasizes prediction-driven response, multi-source cross-validation, and rapid local intervention. Specifically, the invention designs a structure consisting of a dynamic scoring model, a cumulative response penalty term, and a health vulnerability weighting function. This structure is used to accurately classify risk levels and automatically trigger different levels of response actions when the health of ancient trees is declining and accompanied by highly disruptive behaviors. Input:

[0099] Input for this step:

[0100] x t+τ Model in step two The output predicts the health status value at a future time (e.g., 3 hours later);

[0101] x t The current health status is derived from the model's internal state output.

[0102] δ t The disturbance from step three represents the risk level of human behavior at the current moment, with a value range of [0,1]. The larger the value, the stronger the disturbance.

[0103] h t Historical response accumulation represents the integral result of responses triggered in the past H hours. It is maintained by the system's local buffer, with an initial value of 0, and increments (e.g., +1.0) for each high-level response triggered.

[0104] Furthermore, the key step in this process is to construct a comprehensive risk score R. t And generate response action A t First, the comprehensive risk scoring model is defined as follows:

[0105] R t =w1·(x t -x t+τ ) + +w2·δ t +w3·(1-x t+τ ) 2 -λ1·log(1+h t )

[0106] Among them, w1·(x t -x t+τ ) + For trend risk; w2·δ t For disturbance risk; w3·(1-x t+τ ) 2 For the severity of the future state; λ1·log(1+h) t (x) is a response to suppression penalty; t-x t+τ ) + δ indicates the magnitude of the downward trend in health status, representing the intensity of the impending deterioration; t Indicates the intensity of the current behavioral disturbance; (1-x t+τ ) 2 This indicates the predicted severity of health status (using squared amplification for low-value sensitivity); log(1+h) t To address the historical cumulative penalty term and prevent repeated responses within a short period from causing system overload and misjudgment; w1, w2, and w3 are the weights of the scoring terms, with values ​​such as w1 = 1.2, w2 = 1.0, and w3 = 1.5; λ1 is the weight of the response frequency suppression term, which is generally taken as 0.3 to 0.5 and can be adjusted through self-learning.

[0107] To further adapt to the slowly changing yet fragile nature of health status in ancient tree scenarios, this invention adds a health-sensitive weighting function to R. t Perform vulnerability-weighted adjustments:

[0108]

[0109] Wherein, γ is the health sensitivity amplification factor (e.g., γ = 1.5), ensuring that the system response threshold decreases when the current condition of the ancient tree is low, thereby improving protection sensitivity; (1-x t ) 2 The squared term of the current state of health vulnerability reflects the fact that a low state of health is more sensitive to external risks.

[0110] For example: when x t =0.4,x t+τ =0.2,δ t =0.6,h t =2, then the original risk score is:

[0111] Trend item:

[0112] Perturbation term: w2·δ t =0.6;

[0113] Future severity:

[0114] Cumulative response penalties:

[0115] Get R t =0.24 + 0.6 + 0.96 - 0.33 = 1.47;

[0116] Vulnerability weighting factor: (1-0.4) 2 =0.36, γ·0.36=0.54;

[0117] have to

[0118] Furthermore, the system according to Trigger response action A t The response rules are defined as follows:

[0119]

[0120] Where θ1 = 1.2 and θ2 = 2.0 are risk classification thresholds; remote notifications are sent to the platform via LoRa or cellular network connection and upload structured risk reports (including x) t ,x t+τ ,δ t , Image acquisition uses a low-power infrared camera for directional shooting; sound and light alarm control buzzer + high-brightness LED driver; sampling frequency is doubled, changing the sensor sampling period from 5 minutes to 1 minute, and maintaining this for 15 minutes before resuming.

[0121] Furthermore, the output is: A t The system's response action control command at the current time step; structured actions can invoke built-in system devices for execution. The final weighted risk score serves as a reference result for response decisions and a record value for the platform.

[0122] This invention also provides a comprehensive ancient tree health monitoring system based on IoT and AI technologies, the system comprising:

[0123] The health graph construction module is used to acquire historical drone behavior record datasets, including access frequency, charging duration and task intensity, as well as cabinet resource usage records, and construct a behavior health graph. The health score of the cabinet is calculated through the behavior record dataset, and the behavior dependency weight between the drone and the cabinet is established.

[0124] The dynamic topology generation module is used to obtain the spatial coordinates of the racks and drones and the historical anomaly rate of the racks. Combined with the behavioral health map, it generates a dynamic spatiotemporal map containing multiple time slices. Each time slice corresponds to the calculation of the edge weights of the dynamic spatiotemporal map once. The edge weights are updated in real time to reflect the rack resource pressure and health risks. The dynamic spatiotemporal map is used to support multi-time period prediction.

[0125] The resource prediction module uses the dynamic spatiotemporal map to predict the number of empty racks and the trend of health status changes in the future period.

[0126] The intelligent recommendation module generates a priority recommendation for rack access based on the number of empty racks and the trend of health status changes in the racks within the future time period, combined with the current location of the drone, and sends it to the drone for execution.

[0127] The feedback optimization module dynamically adjusts the health score and behavior dependency weights based on the access result feedback optimization behavior health graph; wherein the access result feedback includes access success rate, actual rack health offset, and access behavior deviation.

[0128] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0129] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0130] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A comprehensive ancient tree health monitoring method based on Internet of Things and AI technology, characterized in that, The method comprises the following steps: S1, collecting physiological data of ancient trees through sensors and preprocessing to construct a four-dimensional time-synchronized structured input vector, while collecting millimeter wave radar point cloud frame data; Based on the millimeter wave radar point cloud frame data, the disturbance index is extracted from the edge calculation, which represents the behavior disturbance intensity of the current time step; S2, initialize the ancient tree health state variable as the current time health state, and input the pre-constructed dynamic recursive model combined with the disturbance index to obtain the predicted ancient tree health state variable; S3, the millimeter wave radar point cloud frame data is sampled once per second and pre-clustered to obtain a target point cloud set; based on the target point cloud set, a graph neural network is constructed to encode and extract a disturbance activation representing the current behavior risk, and the disturbance index is corrected according to the disturbance activation to output a disturbance amount, specifically including: Constructing a point cloud space-time graph neural network where nodes are target point cluster centroids in a target point cloud set and edges represent trajectory similarity or motion continuity between different time frames to capture behavior trends; Introducing graph convolution units To encode to extract a perturbation activation representing a current behavioral risk; wherein the perturbation activation is a non-normalized perturbation score; Introducing a health state sensitive adjustment function based on the disturbance activation to map and normalize the disturbance activation, and finally outputting the disturbance amount; wherein the disturbance amount is dynamically corrected based on the current time health state; S4, combining the predicted ancient tree health state variable, disturbance amount, and current time health state, calculating the comprehensive risk score, and dynamically generating the corresponding response action based on the comprehensive risk score, specifically including: The historical response accumulation is indicative of the integral result of responses triggered in the past hour. Based on the current time health state and the predicted ancient tree health state variable, the downward trend amplitude of the health state is determined to represent the trend risk; based on the disturbance amount, the disturbance risk is determined; based on the predicted ancient tree health state variable, the future state severity is determined to represent the severity of the predicted health state; based on the historical response cumulative amount, the response suppression penalty is determined to prevent system overload and misjudgment caused by repeated responses in a short period of time; According to the trend risk, disturbance risk, future state severity, and response suppression penalty, the comprehensive risk score is determined; According to the current time health state, the comprehensive risk score is weighted and corrected for vulnerability to obtain the final comprehensive risk score; According to the final comprehensive risk score, the response action is triggered. 2.The all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 1, characterized in that, The physiological data of ancient trees collected by sensors includes: Thermistor patch, attached to the inner side of the tree bark to collect the temperature of the tree bark; wet polymer sensor attached to the trunk to measure transpiration humidity; use a ring-shaped microelectrode array to collect the electrical conductivity of the tree bark; PAR sensor in the canopy layer to collect light intensity; The preprocessing includes: The ancient tree physiological data is uniformly linearly interpolated into a standard time step to generate an aligned vector as a four-dimensional time-synchronized structured input vector. 3.The all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 1, characterized in that, S2, specifically including: The current ancient tree health state variable is defined as a scalar, representing the comprehensive health degree, and the value range is wherein the minimum value indicates the least healthy state. The update function of the dynamic recursive model is constructed based on the current ancient tree health state variable, structured input vector, and disturbance index to obtain the predicted ancient tree health state variable; The dynamic recursive model employs a time-recursive mechanism to calculate the current ancient tree health status variables, the predicted structured input vector, and the perturbation index, thereby obtaining a predicted future... The results of each step of health status. 4.The all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 3, characterized in that, The predicted structured input vector and disturbance index are obtained by environmental trend extrapolation. 5.The all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 1, characterized in that, The millimeter wave radar point cloud frame data is sampled once per second and pre-clustered to obtain a target point cloud set, specifically including: The millimeter wave radar point cloud frame data is pre-clustered on a per-sample basis to obtain a target point cloud set wherein each represents a spatial distribution point set of a moving object. 6.The all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 1, characterized in that, The health state sensitive adjustment function dynamically corrects the disturbance amount according to the current time health state by designing an adjustment coefficient, so as to represent the response sensitivity of health state to disturbance, wherein the adjustment coefficient is the worst health. 7.The all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 1, characterized in that, The response action triggered according to the final comprehensive risk score specifically includes: Designing a first risk classification threshold and a second risk classification threshold; If the final comprehensive risk score is less than the first risk classification threshold, a log is triggered; If the final comprehensive risk score is greater than the first risk classification threshold and less than the second risk classification threshold, a LoRa or cellular network connection platform is triggered to upload a structured risk report and call a low-power infrared camera for directional shooting; If the final comprehensive risk score is greater than the second risk classification threshold, a buzzer is controlled and a high-brightness LED driver is started, at the same time, the sensor sampling period is changed from 5 minutes to 1 minute, and the situation is reported in real time.

8. The system for performing the all-around ancient tree health monitoring method based on the Internet of Things and AI technology according to claim 1, characterized in that, The system includes: A signal acquisition unit is configured to acquire physiological data of ancient trees through sensors and pre-process the physiological data to construct a four-dimensional time-synchronized structured input vector, and simultaneously acquire millimeter wave radar point cloud frame data; Based on the millimeter wave radar point cloud frame data, an edge computing is used to extract a disturbance index representing the behavior disturbance intensity of the current time step from the millimeter wave radar point cloud frame data; A health prediction unit is configured to initialize an ancient tree health state variable as a current time health state, and input the disturbance index into a pre-constructed dynamic recursive model to obtain a predicted ancient tree health state variable; A health correction unit is configured to sample the millimeter wave radar point cloud frame data once per second and perform pre-clustering processing to obtain a target point cloud set; a graph neural network is constructed based on the target point cloud set to encode and extract a disturbance activation representing a current behavior risk, and the disturbance index is corrected according to the disturbance activation to output a disturbance amount; A health warning unit is configured to calculate a comprehensive risk score based on the predicted ancient tree health state variable, the disturbance amount, and the current time health state, and dynamically generate a corresponding response action based on the comprehensive risk score.

Images