Intelligent stadium lawn maintenance system and method based on AI vision

Abstract

The invention relates to the technical field of intelligent agricultural equipment and nondestructive testing, and discloses a stadium lawn intelligent maintenance system and method based on AI vision, and the system comprises a mobile operation platform, a frequency conversion pneumatic excitation module, a bimodal vision collection module and an edge calculation processing module. The system applies controllable pulsating airflow to a lawn by using a variable-frequency pneumatic excitation module, induces blade resonance and destroys an air boundary layer; and after the airflow stops instantaneously, the bimodal visual acquisition module synchronously acquires free attenuation vibration and a thermal recovery image sequence of the blade. And the edge calculation processing module calculates an equivalent stiffness coefficient, a damping ratio and a thermal time constant of the blade through decoupling analysis, and evaluates the water deficit, disease or growth potential state of the lawn based on multi-dimensional parameter joint distribution. According to the method, the non-contact early warning and accurate identification of the microscopic physiological characteristics of the lawn are realized by actively constructing the mechanical thermal unbalanced state.

Description

Technical Field

[0001] This invention relates to the fields of intelligent agricultural equipment and non-destructive testing technology, specifically to an intelligent maintenance system and method for sports stadium lawns based on AI vision. Background Technology

[0002] Modern high-standard sports stadiums place extremely high demands on the quality of natural turf. The flatness, uniformity, and health of the turf directly affect the fairness of the competition and the safety of the athletes. Traditional turf maintenance mainly relies on manual inspection and experience-based judgment. This method is not only inefficient but also greatly affected by subjective factors, making it difficult to meet the needs of refined management.

[0003] With the development of smart agriculture technologies, passive optical remote sensing technologies based on drones or fixed-point cameras are increasingly being applied to lawn monitoring. These technologies typically use multispectral or hyperspectral cameras to capture reflected light from the lawn and calculate indicators such as the normalized difference vegetation index (NDVI) to infer lawn growth. However, this passive monitoring method has a time lag, as its principles are mostly based on changes in leaf chlorophyll content or surface pigments, which often occur in the middle to late stages of lawn stress. By the time visible yellowing or death symptoms appear on the lawn surface, the internal physiological damage to the plant is often irreversible, missing the optimal window for intervention and treatment.

[0004] Furthermore, existing non-contact detection methods have limitations in identifying the causes of disease, making it difficult to effectively distinguish between different types of stressors such as water deficit and disease infection. Because early water shortages and certain root diseases exhibit similar reflectance changes in their spectral characteristics, relying on a single optical feature can easily lead to misjudgments and incorrect maintenance decisions. For example, over-irrigating areas with root damage caused by disease can actually accelerate the spread of pathogens. Meanwhile, while existing contact devices such as soil moisture sensors are accurate, they are difficult to use for large-area, high-density coverage and can physically damage the turf bed structure. Current technologies generally lack an effective means to actively and non-contactly detect the micromechanical properties of turf (such as stiffness changes caused by cell turgor pressure) and dynamic thermodynamic properties (such as stomatal regulation capacity), making it impossible to achieve early and accurate diagnosis of the physiological state before changes in the turf's appearance occur. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent maintenance system and method for sports stadium turf based on AI vision. This solves the problems of existing passive turf detection technologies being slow to identify early physiological changes and having difficulty accurately distinguishing between water stress and disease infection types under non-contact conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides an AI vision-based intelligent turf maintenance system for sports venues, comprising:

[0008] The mobile work platform is equipped with a positioning and navigation unit and an environmental sensor group to provide spatial coordinate reference and environmental meteorological parameters.

[0009] The variable frequency pneumatic excitation module, installed on a mobile work platform, is used to apply a pulsating airflow with controllable frequency and intensity to the lawn blades. This pulsating airflow can induce mechanical resonance in the lawn blades and disrupt the static air boundary layer on the blade surface, thereby putting the blades in a coupled excited state of mechanical vibration and thermodynamic non-equilibrium.

[0010] The dual-modal vision acquisition module, installed on a mobile work platform, is used to simultaneously acquire the transient response process of lawn blades after the airflow excitation stops, and obtain visible light image sequences and infrared thermal image sequences.

[0011] The edge computing processing module is connected to the variable frequency pneumatic excitation module and the dual-modal vision acquisition module respectively. This module is used to control the variable frequency pneumatic excitation module to perform frequency sweep recognition and resonance locking actions, and to decouple and calculate the mechanical dynamic parameters and transient thermodynamic parameters of the lawn blades based on the acquired image sequence. Then, the physiological state of the lawn is evaluated based on the joint distribution characteristics of the mechanical dynamic parameters and transient thermodynamic parameters.

[0012] In some embodiments, the variable frequency pneumatic excitation module includes an air source component, a high-speed solenoid valve array, and a linear nozzle array; the air source component is connected to the linear nozzle array via the high-speed solenoid valve array; the edge computing processing module sends digital control signals to the high-speed solenoid valve array to drive the high-speed solenoid valve array to open and close at a preset frequency, thereby controlling the linear nozzle array to output linear frequency-modulated airflow or fixed-frequency pulsed airflow; the linear nozzle array is installed at a tangential incident angle to ensure that the output airflow can effectively strip away the static air boundary layer on the surface of the lawn blades and generate forced vibration.

[0013] In some implementations, the dual-modal vision acquisition module includes a visible light imaging unit, an infrared thermal imaging unit, and a synchronous triggering unit. The synchronous triggering unit is connected to the edge computing processing module, the visible light imaging unit, and the infrared thermal imaging unit, respectively. When the edge computing processing module controls the variable frequency pneumatic excitation module to stop outputting airflow, it sends a command to the synchronous triggering unit. In response to the command, the synchronous triggering unit simultaneously sends hardware trigger signals to the visible light imaging unit and the infrared thermal imaging unit, so that both units strictly use the moment when the airflow stops as the time reference to synchronously start image acquisition of the transient response process, thereby ensuring the alignment of mechanical vibration data and thermal recovery data in the time dimension.

[0014] In some implementations, the aerodynamic excitation control logic executed by the edge computing processing module includes: controlling the variable frequency aerodynamic excitation module to output a linearly frequency-modulated airflow with a frequency that increases linearly with time, during which the first-order natural frequency of the lawn blades is identified based on the displacement response of the visible light image sequence; subsequently controlling the variable frequency aerodynamic excitation module to switch to a phase-locked excitation mode, outputting a fixed-frequency pulse airflow with a frequency equal to the first-order natural frequency, and maintaining the lock for a preset time to allow the lawn blades to reach maximum amplitude and form a non-equilibrium temperature field; after the lock time is reached, controlling the variable frequency aerodynamic excitation module to instantaneously cut off the airflow output, allowing the system to enter the free decay and thermal recovery stage.

[0015] In some implementations, the edge computing processing module further includes a visual front-end processing unit for preprocessing the acquired data: for visible light image sequences, the dense optical flow algorithm is applied to calculate the displacement vector field of pixels in the region of interest, and a one-dimensional displacement time-domain signal reflecting the motion state of the leaf group is extracted; for infrared thermal image sequences, the homography transformation matrix is ​​applied to map them to the visible light coordinate system, and the corresponding transient temperature response curve is extracted.

[0016] In some implementations, the edge computing processing module calculates mechanical dynamic parameters based on one-dimensional displacement time-domain signals. These parameters include damping ratio and equivalent stiffness coefficient. The damping ratio is obtained by analyzing the amplitude ratio of adjacent peaks in free decaying vibration using the logarithmic decay method, reflecting the viscous characteristics of the blade fiber structure. The equivalent stiffness coefficient is obtained by using the inversion relationship between the identified first-order natural frequency and the preset unit blade equivalent mass parameter based on a single-degree-of-freedom vibration theoretical model, reflecting the blade's bending resistance and cell turgor pressure state.

[0017] In some implementations, the edge computing processing module calculates transient thermodynamic parameters based on the transient temperature response curve, which are thermal time constants. By constructing a linear regression model based on Newton's law of cooling, the logarithmic temperature difference decay data after the airflow stops is fitted, and the reciprocal of the slope of the fitted line is determined as the thermal time constant. This thermal time constant characterizes the rate of thermal equilibrium recovery of lawn leaves under stomatal regulation and is correlated with stomatal conductance.

[0018] In some implementations, the logic by which the edge computing processing module assesses the physiological state of the turf includes: comparing the calculated equivalent stiffness coefficient with a preset stiffness reference threshold, and comparing the thermal time constant with a preset thermal response threshold; determining that the turf in the current area is in an early water deficit state if and only if the equivalent stiffness coefficient is less than the stiffness reference threshold and the thermal time constant is greater than the thermal response threshold. This joint determination mechanism utilizes the coupling effect of water shortage leading to a decrease in cell turgor pressure (reduced stiffness) and stomatal closure (slower thermal recovery).

[0019] In some implementations, the evaluation logic further includes monitoring the numerical distribution of the damping ratio. When a damping ratio higher than the statistical average of healthy samples is detected, accompanied by a decrease in the equivalent stiffness coefficient, it is determined that the lawn in the current area has tissue necrosis or disease infection. This utilizes the high viscosity and low elasticity characteristics exhibited by diseased tissue due to structural damage.

[0020] A second aspect of this invention provides a method for intelligent maintenance of sports stadium turf based on AI vision, the method comprising the following steps:

[0021] A variable frequency pneumatic excitation module was used to perform a pneumatic frequency sweep on the lawn area to identify the first-order natural frequency of the lawn blades.

[0022] By applying a fixed-frequency pulsed airflow excitation to the lawn blades at the first-order natural frequency, the blades resonate and the surface thermal boundary layer is stripped away, thus constructing a non-equilibrium state of mechanical-thermal coupling.

[0023] The airflow excitation was stopped instantaneously, and the free decay vibration image sequence and thermal recovery image sequence of the blade were simultaneously acquired using a dual-modal vision acquisition module;

[0024] The edge computing processing module decouples and analyzes the acquired image sequence to calculate the damping ratio and equivalent stiffness coefficient that characterize the viscoelastic properties, as well as the thermal time constant that characterizes the heat dissipation properties.

[0025] Based on the multidimensional characteristic distribution of damping ratio, equivalent stiffness coefficient and thermal time constant, the water deficit, disease or growth potential of turf can be determined.

[0026] This invention provides an intelligent sports stadium turf maintenance system and method based on AI vision. It has the following beneficial effects:

[0027] 1. This invention, through the cooperation of a dual-modal visual acquisition module and an edge computing processing module, simultaneously decouples the mechanical dynamic parameters (equivalent stiffness, damping ratio) and thermodynamic parameters (thermal time constant) of lawn leaves. It utilizes the difference in physical mechanisms that water stress mainly leads to a decrease in cell turgor pressure (reduced stiffness) and stomatal closure (slower thermal recovery), while disease infection mainly leads to tissue decay (increased damping). This solves the technical problem that traditional single-spectrum remote sensing technology cannot distinguish between physiological water shortage and pathological damage in lawns, and improves the targeting of maintenance decisions.

[0028] 2. This invention employs a variable frequency pneumatic excitation module to actively induce leaf resonance and disrupt thermal equilibrium. Microscopic physicochemical characteristics are then inverted by analyzing the transient response process after airflow cutoff. Compared to passive optical detection that relies on changes in leaf chlorophyll content or yellowing, this invention can directly sense changes in mechanical stiffness caused by minute fluctuations in cell turgor pressure. This allows for the early identification of potential physiological abnormalities before visible yellowing or death of the lawn occurs, providing a valuable window of opportunity for lawn maintenance.

[0029] 3. This invention utilizes a tangentially mounted linear nozzle array to actively strip away the static air boundary layer on the blade surface, and combines this with benchmark calibration from environmental sensors to construct a controllable local convection environment. This design not only enables large-area non-contact, non-destructive testing, avoiding damage to the lawn from physical contact, but also suppresses interference from environmental factors such as natural wind speed fluctuations and uneven lighting on infrared thermometry data by actively disrupting the thermal equilibrium, ensuring high measurement repeatability of the system in outdoor open venues. Attached Figure Description

[0030] Figure 1 This is a block diagram of the overall system structure of the present invention;

[0031] Figure 2 This is a flowchart illustrating the logic for determining the physiological state of turf in this invention.

[0032] Among them, 10 is the mobile operation platform; 11 is the positioning and navigation unit; 12 is the environmental sensor group; 20 is the variable frequency pneumatic excitation module; 21 is the air source component; 22 is the high-speed solenoid valve array; 23 is the linear nozzle array; 30 is the dual-modal vision acquisition module; 31 is the visible light imaging unit; 32 is the infrared thermal imaging unit; 33 is the synchronous triggering unit; and 40 is the edge computing processing module. Detailed Implementation

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] See attached document Figure 1 The present invention provides a sports stadium turf intelligent maintenance system based on active aerodynamic-thermal coupling response. The system includes: a mobile operation platform 10, a variable frequency aerodynamic excitation module 20, a dual-modal vision acquisition module 30, and an edge computing processing module 40.

[0035] The mobile work platform 10 serves as the system carrier, configured to move or stop on the surface of the lawn area to be inspected. A positioning and navigation unit 11 is installed on the mobile work platform 10, which includes a real-time dynamic differential positioning component and an inertial measurement unit (IMU) for acquiring the three-dimensional position coordinates and attitude data of the mobile work platform 10 in the site coordinate system. An environmental sensor group 12 is installed on the outer sidewall of the mobile work platform 10, away from areas prone to airflow disturbance. The environmental sensor group 12 is used to measure the steady-state parameters of the current environment in real time, including at least the ambient air temperature. and ambient relative humidity The environmental sensor group 12 is electrically connected to the edge computing processing module 40, which will... and The data is transmitted to the edge computing processing module 40 as boundary conditions for subsequent thermodynamic calculations.

[0036] The variable frequency pneumatic excitation module 20 is installed at the front bottom of the mobile work platform 10 and is used to apply controllable aerodynamic excitation to the single area to be tested. The variable frequency pneumatic excitation module 20 includes an air source assembly 21, a high-speed solenoid valve array 22, and a linear nozzle array 23. The air source assembly 21 includes an air compressor and a pressure tank to provide compressed air at a constant pressure. The high-speed solenoid valve array 22 is connected to the edge computing processing module 40 and receives control signals to adjust the air path on / off and opening degree, with the adjustment frequency range covering the theoretical first-order natural frequency range of the lawn blades.

[0037] The linear nozzle array 23 is connected to the high-speed solenoid valve array 22 via a pipeline. The linear nozzle array 23 consists of multiple flat nozzles arranged linearly in the horizontal direction. The air outlet direction of the linear nozzle array 23 forms a preset incident angle with the horizontal ground. The angle of incidence The angle is set between 30 and 60 degrees, configured to cause the output airflow to act on the surface of the lawn blades in a tangential shear manner. The linear nozzle array 23 is configured to output a laminar air curtain, which generates a time-varying airflow on the surface of the lawn blades. Changing external incentives .

[0038] The dual-modal vision acquisition module 30 is rigidly fixed to the cantilever support of the mobile work platform 10, located above the linear nozzle array 23, and its field of view covers the lawn area disturbed by airflow. The dual-modal vision acquisition module 30 includes a visible light imaging unit 31, an infrared thermal imaging unit 32, and a synchronous triggering unit 33. The visible light imaging unit 31 uses a panoramic shutter high-speed industrial camera, configured to acquire visible light image sequences containing the mechanical vibration process of the lawn blades. The sampling frame rate of the visible light imaging unit 31... Satisfying the Nyquist sampling theorem with respect to the highest theoretical natural frequency of lawn blades The requirements, namely, to meet .

[0039] The infrared thermal imaging unit 32 employs a long-wave infrared camera, configured to acquire a sequence of thermal infrared images reflecting the temperature field distribution on the surface of lawn leaves. The optical axis of the infrared thermal imaging unit 32 is aligned with the optical axis of the visible light imaging unit 31 through calibration, ensuring that the centers of the fields of view of the two imaging units coincide. The synchronization trigger unit 33 is electrically connected to the visible light imaging unit 31, the infrared thermal imaging unit 32, and the frequency conversion pneumatic excitation module 20, respectively, and is configured to trigger the synchronous trigger unit 33 when the airflow excitation stops. Send a hardware trigger signal to enable the visible light imaging unit 31 and the infrared thermal imaging unit 32 to start high-frequency synchronous acquisition at the same time reference.

[0040] The edge computing processing module 40 is communicatively connected to the positioning and navigation unit 11, the environmental sensor group 12, the variable frequency aerodynamic excitation module 20, and the dual-modal vision acquisition module 30. The edge computing processing module 40 is configured to perform the following operations: control the variable frequency aerodynamic excitation module 20 to output swept frequency airflow or locked frequency airflow; receive visible light image sequences and calculate blade displacement and damping ratio; receive thermal infrared image sequences and calculate thermal time constant; and generate lawn maintenance instructions based on the calculated physical parameters.

[0041] The edge computing processing module 40 internally stores a preset physical inversion algorithm. In this algorithm, the mechanical motion of a single lawn blade is described using a single-degree-of-freedom damped vibration model, and its motion equation is defined as:

[0042] ;

[0043] in, Indicates the blade tip over time The amount of displacement that changes. This is the equivalent mass of the blade. The damping coefficient is... This is the stiffness coefficient. External excitation force applied to the variable frequency pneumatic excitation module 20.

[0044] The edge computing processing module 40 is also configured to calculate the thermodynamic response of the blades based on the thermal infrared image sequence. For the temperature recovery process after the airflow stops, a modified Newtonian cooling model is used to describe it, and its temperature evolution equation is defined as:

[0045] ;

[0046] in, For a moment The surface temperature of the blades, The moment when the airflow stops The surface temperature of the blades, The ambient air temperature collected by the environmental sensor group 12 The thermal time constant to be inverted is given. The edge computing processing module 40 calculates the thermal time constant characterizing the stomatal conductance and moisture state of the leaf by numerically fitting the above equation. .

[0047] See attached document Figure 1 and attached Figure 2 Before performing active aerodynamic excitation, the system first performs initialization and environmental baseline correction to establish the boundary conditions and spatial baseline for subsequent physical inversion calculations.

[0048] The mobile work platform 10 travels to the target detection point according to the preset path planning data. The positioning and navigation unit 11 calculates the current three-dimensional coordinates and heading angle in real time. The edge computing processing module 40 monitors the variance of the positioning data. The edge computing processing module 40 issues a stop command to control the drive mechanism of the mobile work platform 10 to lock, thereby eliminating the interference of mechanical vibration of the body on the microscopic visual measurement.

[0049] After the system reaches a steady-state standstill, the environmental sensor group 12 is activated to acquire the thermodynamic boundary parameters of the current microenvironment. Considering the instantaneous fluctuations in outdoor wind field and illumination, the environmental sensor group 12 does not use single sampling values, but instead continuously acquires a sequence of ambient air temperature and relative humidity data within a preset time window (e.g., 2 to 5 seconds) at a preset sampling frequency (e.g., 10 Hz). The edge computing processing module 40 receives this data sequence and applies a moving average filtering algorithm to remove extreme noise, calculating the steady-state ambient air temperature at the current moment. and steady-state relative humidity The steady-state ambient air temperature It is locked and stored as an asymptotic lower bound for the Newtonian cooling model in subsequent thermodynamic inversion equations.

[0050] Synchronously, the dual-modal vision acquisition module 30 performs static reference image acquisition. When the variable-frequency aerodynamic excitation module 20 is off, the infrared thermal imaging unit 32 performs thermal radiation imaging on the stationary lawn area within the field of view, generating a reference thermal infrared image frame. The edge computing processing module 40 reads the raw radiation data of this image frame and combines it with the pre-stored emissivity coefficient of the lawn blade surface. (The setting range is typically 0.95 to 0.98) and the currently measured , Atmospheric transport corrections are performed, converting radiance values ​​into a physical temperature matrix. The edge computing processing module 40 performs histogram statistical analysis on this physical temperature matrix, using an adaptive threshold segmentation algorithm (e.g., the Otsu algorithm) to identify and remove pixel regions belonging to bare soil (typically with higher temperatures) or deep shadows (typically with lower temperatures), retaining only the effective pixel regions on the surface of lawn leaves. The system calculates the spatial average temperature of these effective pixel regions and defines it as the initial reference temperature. .like If the temperature difference is less than the preset threshold (e.g., 0.5 degrees Celsius), the system determines that it is currently in thermal equilibrium and meets the conditions for conducting transient thermal response testing.

[0051] In addition, the visible light imaging unit 31 simultaneously acquires a high-resolution static visible light image. The edge computing processing module 40 calls the pre-calibrated homography matrix to perform a multimodal image registration operation. This operation maps and transforms the pixel coordinate space of the reference thermal infrared image to the pixel coordinate space of the visible light image, achieving pixel-level alignment of the two modalities in spatial dimensions. Based on the registered image space, the edge computing processing module 40 delineates a rectangular region of interest (ROI) at the center of the field of view. The physical location of this ROI strictly corresponds to the core impact zone of the output airflow of the linear nozzle array 23, and its physical width is set to cover the width of 3 to 5 lawn blades. All subsequent optical flow field calculations and thermal time constant fitting are restricted to this ROI to eliminate nonlinear interference data generated by the airflow divergence zone at the edge of the field of view.

[0052] After completing the environmental baseline correction, the edge computing processing module 40 initiates a pneumatic frequency sweep sequence, a process designed to identify and utilize the mechanical resonance characteristics of the lawn blades in the current area.

[0053] The waveform generation unit inside the edge computing processing module 40 generates a series of digital control signals, which are transmitted to the high-speed solenoid valve array 22 of the variable frequency pneumatic excitation module 20. The high-speed solenoid valve array 22 responds to the control signals by performing high-frequency opening and closing actions, driving the air source assembly 21 to output a linearly frequency-modulated airflow with pressure pulsations that continuously change over time. The frequency of this linearly frequency-modulated airflow... From the preset starting frequency (e.g., 5Hz) linearly increase to the termination frequency (e.g., 50Hz), scan duration set to .

[0054] The linear nozzle array 23 converts the frequency-varying pressure pulsation into a periodic excitation force acting on the surface of the lawn blades. The time-domain representation of this excitation force is configured as follows:

[0055] ;

[0056] in, The amplitude of the airflow impact force is determined by the regulated output pressure of the air source component 21.

[0057] During the entire time window of the frequency sweeping action performed by the variable frequency pneumatic excitation module 20 Inside, the visible light imaging unit 31 in the dual-modal vision acquisition module 30 continuously acquires an image stream of the lawn area at a constant high frame rate. The image stream is transmitted in real-time to the vision front-end processing unit of the edge computing processing module 40. The vision front-end processing unit does not process the entire image; instead, it calculates the pixel displacement vector field between adjacent frames using a dense optical flow algorithm (e.g., the Farneback algorithm) for the pixels within the region of interest (ROI) defined in the aforementioned steps. The system extracts the arithmetic mean of the vertical components of the displacement vectors of all feature points within the ROI region, constructing a temporal displacement signal reflecting the overall mechanical motion state of the leaf cluster. .

[0058] Edge computing processing module 40 pairs of acquired time-domain displacement signals Perform a windowed Fast Fourier Transform to map the time-domain signal to the frequency domain, generating the amplitude-frequency response spectrum. The edge computing processing module 40 performs a peak search algorithm in the amplitude-frequency response spectrum to find peaks in the interval [missing information]. The maximum amplitude point within the detection area. The frequency corresponding to this maximum amplitude point is identified as the first-order natural frequency of the lawn leaf population in the current detection area. The identification criterion is expressed as:

[0059] ;

[0060] To eliminate low-frequency noise interference introduced by ambient wind or equipment vibration, the edge computing processing module 40 is also equipped with signal-to-noise ratio verification logic, which only confirms the signal when the peak amplitude found exceeds a preset multiple (e.g., 3 times) of the floor noise amplitude. efficient.

[0061] Once the first-order natural frequency Upon confirmation, the edge computing processing module 40 immediately terminates the frequency sweep mode and switches to the phase-locked excitation mode. The edge computing processing module 40 adjusts the frequency of the control signal sent to the high-speed solenoid valve array 22, ensuring it is strictly fixed at the identified frequency. Numerically, the linear nozzle array 23 randomly outputs a constant frequency. The fixed-frequency pulsed airflow, with the continuous excitation time set to... (For example, 500ms to 1000ms). During this locked-in excitation phase, the pulsating frequency of the airflow resonates with the natural frequency of the lawn blades, causing the blade's oscillation amplitude to gradually accumulate and reach its maximum value. This large-amplitude mechanical resonant motion not only provides a higher signal-to-noise ratio for subsequent stiffness measurements but also forces the blade surface to undergo violent relative motion with the surrounding air, thereby maximally stripping away the stationary air boundary layer attached to the blade surface and creating non-equilibrium conditions for subsequent thermodynamic measurements.

[0062] The frequency is maintained at the frequency of the variable frequency pneumatic excitation module 20. During the fixed-frequency excitation period, the physical system enters the thermo-boundary layer stripping stage. At this time, the high-speed tangential airflow continuously output by the linear nozzle array 23 interacts with the lawn blades, which are in a resonant state, through fluid-structure interaction.

[0063] Because the blades are in a reciprocating motion with maximum amplitude, the normal and tangential velocity components on the blade surface reach their peak values. The high-speed airflow is incident at a preset tangential angle. The continuous erosion of the blade's surface tissue causes the Reynolds number of the flow field on the blade surface to increase instantaneously, exceeding the critical Reynolds number. This high Reynolds number flow field disrupts the stable laminar air boundary layer attached to the blade surface in a static state, leading to turbulent transition and physical stripping of this boundary layer. The stripping of the boundary layer reduces the convective heat transfer coefficient on the blade surface. It exhibits a step-like surge, with a value far exceeding the natural convection heat transfer coefficient under still air.

[0064] In convective heat transfer coefficient Under conditions of sudden temperature increase, according to thermodynamic principles, a strong forced convection heat transfer effect occurs between the blade surface and the surrounding fluid medium. The rate at which heat is transferred (or vice versa) from the blade's interior to the external airflow through the epidermis accelerates, disrupting the blade's thermodynamic equilibrium. Specifically, if the initial surface temperature of the lawn blade increases due to sunlight radiation... Higher than the airflow temperature (i.e., ambient air temperature) Forced convection causes rapid heat dissipation from the blade surface, resulting in a rapid drop in surface temperature and a deviation from its normal value. If the initial blade temperature is lower than the airflow temperature, the surface temperature will rise rapidly and deviate from the airflow temperature. .

[0065] This process continues until the incentive is locked. The process aims to artificially create a non-steady-state temperature field that deviates from the environmental equilibrium temperature. During this period, the infrared thermal imaging unit 32 continuously monitors the change in average radiant temperature within the region of interest (ROI), and the edge computing processing module 40 calculates the rate of temperature change in the ROL region in real time. When the edge computing processing module 40 detects that the absolute value of the rate of temperature change begins to decrease or the temperature value tends to a new dynamic equilibrium point under the influence of airflow, the system determines that the thermal boundary layer has been sufficiently stripped, thus completing the thermal excitation of the blade thermodynamic system and meeting the conditions for entering transient recovery acquisition.

[0066] When the pneumatic excitation reaches the preset lock duration The system immediately enters the transient response signal synchronous acquisition stage. The core of this stage is to accurately capture the critical point of the grass blade switching from forced vibration to free decay vibration and the complete process of thermodynamic non-equilibrium state returning to equilibrium state through hardware-level nanosecond-level synchronization.

[0067] The edge computing processing module 40 sends a cutoff command to the variable frequency pneumatic excitation module 20, controlling the high-speed solenoid valve array 22 to perform an instantaneous closing action. The closing response time of the high-speed solenoid valve array 22 is selected to be less than 5 milliseconds to ensure the excitation force applied to the lawn blades. The excitation drops to zero in a step-like manner, thus forming a steep excitation falling edge on the time axis. This falling edge moment is defined as the zero-time reference of the system. .

[0068] Within the same clock cycle of sending the cutoff command, the edge computing processing module 40 sends a high-level pulse signal to the synchronization trigger unit 33 via the GPIO interface. In response to this pulse signal, the synchronization trigger unit 33 simultaneously outputs a TTL (transistor-to-transistor logic) hardware trigger signal to both the visible light imaging unit 31 and the infrared thermal imaging unit 32. This hardware trigger signal forces both imaging units to reset their internal frame counters. Starting from the preset high-frequency sampling rate, synchronous image stream acquisition is initiated.

[0069] Upon receiving the trigger signal, the visible light imaging unit 31 initiates a circular buffer write mode, continuously acquiring visible light image sequences at a rate of no less than 120 frames per second. The acquisition duration is set to... (For example, 1.5 seconds), this duration covers the theoretical time required for the mechanical vibration amplitude of the lawn blades to decay to the noise floor. The acquired image sequence records the free decay vibration trajectory of the blades under the action of their own elastic restoring force and internal structural damping after the loss of external aerodynamic support. Each frame in the image sequence is labeled relative to... Microsecond-level precise timestamps.

[0070] Simultaneously, upon receiving the trigger signal, the infrared thermal imaging unit 32 continuously acquires a stream of raw thermal infrared radiation data at its maximum frame rate (e.g., 30 frames / second or 60 frames / second). This data stream records a complete sequence of thermal images showing the evolution of the blade surface temperature field over time. As the airflow ceases, the forced convection heat transfer conditions on the blade surface instantly disappear, a new laminar air boundary layer is established on the blade surface, and the stomatal regulation mechanism (transpiration) of the blade begins to dominate the transition of surface temperature towards ambient temperature. (or reference temperature) The regression process. Each frame of thermal image recorded by the infrared thermal imaging unit 32 is also marked with a relative value. The timestamps of the same time reference ensure strict alignment of thermodynamic and mechanical motion data in the time dimension, providing phase-difference-free data support for subsequent decoupling of multiphysics parameters. The edge computing processing module 40 temporarily stores the timestamped image data streams acquired from the two channels into high-speed memory, constructing a three-dimensional data cube for subsequent algorithm calls.

[0071] See attached document Figure 2 After acquiring the raw data of the synchronous transient response, the edge computing processing module 40 performs preprocessing and optical flow field calculation, aiming to transform the unstructured video stream into a one-dimensional time-domain signal that can be used for physical modeling.

[0072] For the visible light image sequence output by the visible light imaging unit 31, the edge computing processing module 40 first performs region of interest (ROI) cropping and image enhancement processing. The system applies an adaptive histogram equalization algorithm to the image within the ROI region to enhance the contrast of texture features on the surface of the lawn leaves, ensuring that the leaf edges and background have grayscale differences under different lighting conditions. Subsequently, the system applies a Gaussian smoothing filter to the enhanced image sequence, with the convolution kernel size set to [value missing]. Or To suppress high-frequency shot noise in image sensors.

[0073] After image enhancement, the edge computing processing module 40 uses a dense optical flow algorithm based on polynomial expansion (such as the Farneback algorithm) to calculate the pixel motion vectors between two adjacent frames. The system assumes that within a small time interval... Within the lawn, the grayscale values ​​of pixels on the surface of the blades remain constant, thus satisfying the optical flow constraint equation:

[0074] ;

[0075] in, Indicates at time Pixel coordinates grayscale value at that location and These represent the minute displacements of the pixel in the horizontal and vertical directions, respectively.

[0076] The edge computing processing module 40 solves the above equations for each pixel within the ROI region to obtain a dense optical flow field. Considering that the free decay vibration of lawn blades after the airflow excitation is removed mainly manifests as reciprocating oscillations along the direction of gravity or the direction of blade growth normal, the system focuses on extracting the vertical component. Or the projected component along the main vibration direction. The system calculates the weighted average of the displacement vectors of all valid pixels within the ROI region, thereby synthesizing a one-dimensional displacement time-domain signal reflecting the macroscopic mechanical motion state of the blade group. To eliminate integral drift, the system... Perform detrending processing to correct the vibration equilibrium position to zero.

[0077] For the thermal infrared image sequence output by the infrared thermal imaging unit 32, the edge computing processing module 40 first performs radiometric calibration processing. The system calls the pre-stored radiometric response function to convert the raw digital quantities of the image sensor into physical radiance temperature. Subsequently, the system uses a pre-calibrated homography transformation matrix to map the spatial coordinate system of the thermal infrared image to the visible light image coordinate system, achieving pixel-level spatial registration.

[0078] Based on the registered thermal infrared image, the edge computing processing module 40 extracts the thermal imaging ROI region that strictly corresponds to the spatial location of the visible light ROI region. The system calculates the arithmetic mean of the temperature of all pixels within this region, generating a raw temperature sequence that changes over time. To suppress spatial non-uniformity noise (NUC noise) from the thermal imaging sensor and random environmental disturbances, the system applies a sliding window mean filter or a Kalman filter to the raw temperature sequence, ultimately outputting a smooth transient temperature response curve. This curve precisely records the time from the moment the airflow stops. Initially, the complete trajectory of the evolution of blade surface temperature over time.

[0079] Edge computing processing module 40 processes the one-dimensional displacement time-domain signal extracted in the aforementioned steps. In-depth analysis was conducted to separate the physical parameters characterizing the viscoelastic properties of lawn blades from the trajectory of free decaying vibrations.

[0080] The edge computing processing module 40 first processes the time-domain signal. Perform envelope analysis and peak extraction. The extremum search algorithm running internally within the module (e.g., derivative discrimination based on zero-crossing points) scans the entire... The system identifies all local maxima in the sequence. It then extracts the time coordinates and amplitude values ​​of these maxima to construct a discrete amplitude sequence. ,in Representing the The system calculates the peak amplitude of each vibration cycle. To ensure calculation accuracy, the system automatically removes tail peaks with amplitudes smaller than the signal-noise floor, retaining only the peaks with higher signal-to-noise ratios. One valid peak.

[0081] Based on the extracted amplitude sequence, the edge computing processing module 40 calculates the logarithmic decay rate. This parameter quantifies the energy dissipation of the blade during each vibration cycle. In the calculation, the system selects the amplitudes of two adjacent wave peaks. and or cross Peak amplitude of each cycle and To reduce random error by taking the average value. Logarithmic decay rate. The computational logic is configured as follows:

[0082] ;

[0083] Obtain the logarithmic decay rate Then, the edge computing processing module 40 calculates the damping ratio. For underdamped systems like lawn blades, the damping ratio directly reflects the viscous characteristics of the material and the energy dissipation capacity of the fiber structure. The system is calculated based on the following relationship.

[0084] ;

[0085] when An abnormally high value usually indicates that the internal cell structure of the leaf has ruptured or the tissue has died, resulting in changes in rheological properties.

[0086] Subsequently, the edge computing processing module 40 calculates the equivalent stiffness coefficient of the lawn blades. At this point, the system calls the first-order natural frequency value that was identified and stored during the aerodynamic frequency sweep and resonance locking phase. In addition, the system retrieves the unit blade equivalent mass parameters preset in the database. .Should It can be a fixed constant based on the turf variety, or it can be the average plant height identified in visible light images. Through empirical formulas (in The value is obtained by dynamically estimating the linear density.

[0087] The system is based on single-degree-of-freedom vibration theory and utilizes the physical constraint relationship between natural frequency, stiffness, and mass to inversely calculate the stiffness coefficient. :

[0088] ;

[0089] Stiffness coefficient The bending resistance of the leaves was quantified from a mechanical perspective. The numerical value is positively correlated with the leaf cell turgor pressure and lignification degree, making it a key physical indicator for determining whether early water deficit exists in a lawn. Through the above steps, the system successfully decoupled the complex vibration video stream into two independent dimensionless or physical quantity parameters. and And store these two parameters into the feature database of the current detection point.

[0090] The edge computing processing module 40, based on preset thermodynamic inversion logic, processes the smoothed transient temperature response curve. Numerical analysis was conducted to quantify the rate of thermal equilibrium recovery of turf blades after they are removed from the forced convection environment, which is dominated by the transpiration regulation of their own stomata.

[0091] The edge computing processing module 40 first retrieves the steady-state ambient air temperature obtained during the initialization phase from the memory. and the time of airflow cutoff Corresponding initial surface temperature To eliminate the influence of the direction of ambient temperature fluctuations (i.e., whether the blades are absorbing or dissipating heat) on the computational logic, the system constructs a dimensionless instantaneous temperature difference decay function. Considering that in actual physical processes, the decay of temperature data follows the exponential approximation law described in Example 1, the edge computing processing module 40 performs a logarithmic linearization transformation on this exponential relationship to achieve robust solutions using a linear regression algorithm.

[0092] The system first calculates the current sampling time. The absolute difference between the temperature value and the ambient equilibrium temperature is used to construct a linearized variable by taking its natural logarithm. The linearization equations underlying this transformation process are configured as follows:

[0093] ;

[0094] in, Let be the thermal time constant to be determined, and be the term. This constitutes the slope term of the linear equation. It constitutes the intercept term of the linear equation.

[0095] Edge computing processing module 40 selects time window The sampled data points within the range are used to fit the above linearized equation. Here, a starting delay is set. The end time is set to avoid the nonlinear transition region caused by boundary layer reconstruction at the moment the airflow stops. This is to avoid a decrease in signal-to-noise ratio due to excessively small temperature differences. The system uses the least squares method to analyze discrete data point pairs. Perform linear regression analysis to find the slope of the objective line that minimizes the sum of squared residuals. .

[0096] Obtaining the regression slope Then, the edge computing processing module 40 calculates the thermal time constant through reciprocal operations.

[0097] ;

[0098] The thermal time constant It is a physical quantity with a time dimension. The smaller its value, the faster the lawn leaves recover thermal equilibrium, corresponding to a higher sensible or latent heat exchange efficiency.

[0099] To ensure the reliability of the inversion results, the edge computing processing module 40 simultaneously calculates the determination coefficients of the linear regression. The system has a preset goodness-of-fit threshold (e.g., 0.95), and only when the calculated goodness-of-fit value is obtained... The system only considers a thermodynamic measurement valid when the result exceeds this threshold, and then calculates the result. The data point is stored in the feature database; otherwise, the system will mark it as an outlier and remove or reduce its weight in subsequent decisions. Through this inversion process, the system maps macroscopic temperature change trajectories to microscopic physiological characterization parameters, where... The value is mainly modulated by the stomatal conductance on the leaf surface, thus indirectly reflecting the water metabolism function of the lawn.

[0100] See attached document Figure 1 and attached Figure 2 The edge computing processing module 40 completed the processing of the mechanical dynamic parameters (equivalent stiffness coefficient) for each detection point. Damping ratio First-order natural frequency and thermodynamic parameters (thermal time constant) After the solution is obtained, the system enters the physiological state assessment and decision generation stage. The edge computing processing module 40 aggregates the above discrete physical parameters into a multidimensional feature vector corresponding to the spatial coordinate point. The data is then input into a pre-defined decision logic unit for classification and judgment.

[0101] The decision logic unit first performs a joint determination of the water deficit state. This determination logic is based on the mechano-stomatal coupling effect, that is, in the early stages of water shortage, lawn leaves simultaneously exhibit structural softening due to decreased cell turgor pressure and reduced heat exchange efficiency due to the active closure of stomata to reduce water loss. The edge computing processing module 40 then processes the stiffness coefficients in the feature vector. Compared with the preset stiffness reference threshold Compare and use thermal time constants With preset thermal response threshold Comparison is performed if and only if the detection results meet the conditions. (Indicating a decrease in the physical stiffness of the blade) and When the leaf heat recovery rate slows down and stomatal conductance is low, the system determines that the area is in an early water stress state. Based on this determination, the edge computing processing module 40 generates a fixed-point sprinkler irrigation command containing precise coordinate information. This command is sent to the external irrigation control system or the water replenishment device built into the mobile operation platform 10 to perform localized precise water replenishment operations.

[0102] Secondly, the decision logic unit determines whether a plant is pathologically or structurally damaged. This determination logic is based on the viscoelastic changes in the internal fibrous structure of the leaf. Healthy lawn leaves exhibit predominantly elastic mechanical characteristics, while leaf tissue that has rotted, been infected by fungi, or is physically broken will exhibit viscous characteristics. The edge computing processing module 40 monitors the damping ratio. The numerical distribution, when detected Higher than the statistical mean of healthy samples (e.g., more than 3 standard deviations), and accompanied by stiffness coefficient When the concentration drops drastically, the system determines that there is tissue necrosis or disease infection in the area. The edge computing processing module 40 will highlight the area on the digital map and output a warning message to prompt manual intervention or the application of fungicides.

Claims

1. A smart turf maintenance system for sports venues based on AI vision, characterized in that, include: A mobile work platform, equipped with a positioning and navigation unit and an environmental sensor array; A variable frequency pneumatic excitation module, installed on the mobile working platform, is used to apply a pulsating airflow with controllable frequency and intensity to the lawn blades, inducing mechanical resonance of the lawn blades and disrupting the air boundary layer on their surface. A dual-modal vision acquisition module, installed on the mobile work platform, is used to simultaneously acquire visible light image sequences and infrared thermal image sequences of lawn blades after the airflow excitation stops. The edge computing processing module is used to control the variable frequency pneumatic excitation module to perform frequency sweep and resonance locking actions, and to decouple and calculate the mechanical dynamic parameters and transient thermodynamic parameters of the lawn blades based on the visible light image sequence and the infrared thermal image sequence. Then, it evaluates the physiological state of the lawn based on the joint distribution characteristics of the mechanical dynamic parameters and the transient thermodynamic parameters, and generates corresponding precise maintenance operation instructions based on the evaluation results.

2. The intelligent sports stadium turf maintenance system based on AI vision according to claim 1, characterized in that, The variable frequency pneumatic excitation module includes an air source component, a high-speed solenoid valve array, and a linear nozzle array; The gas source component is connected to the linear nozzle array via the high-speed solenoid valve array; The edge computing processing module is used to send digital control signals to the high-speed solenoid valve array, drive the high-speed solenoid valve array and open and close at a preset frequency, thereby controlling the linear nozzle array to output linear frequency-modulated airflow or fixed frequency pulse airflow. The linear nozzle array is installed at a tangential incident angle, enabling the output airflow to strip away the static air boundary layer on the surface of the lawn blades and generate forced vibrations.

3. The intelligent sports stadium lawn maintenance system based on AI vision according to claim 1, characterized in that, The dual-modal vision acquisition module includes a visible light imaging unit, an infrared thermal imaging unit, and a synchronous triggering unit; The synchronization triggering unit is connected to the edge computing processing module, the visible light imaging unit, and the infrared thermal imaging unit, respectively. The edge computing processing module sends a command to the synchronization triggering unit at the moment when it controls the variable frequency pneumatic excitation module to stop outputting airflow; In response to the instruction, the synchronous triggering unit simultaneously sends hardware trigger signals to the visible light imaging unit and the infrared thermal imaging unit, so that both units synchronously start image acquisition of the transient response process with the moment when the airflow stops as the time reference.

4. The intelligent sports stadium turf maintenance system based on AI vision according to claim 1, characterized in that, The edge computing processing module is used to execute the following aerodynamic excitation control logic: The variable frequency pneumatic excitation module is controlled to output a linearly frequency-modulated airflow whose frequency increases linearly with time. During this period, the first-order natural frequency of the lawn blades is identified based on the displacement response of the visible light image sequence. The variable frequency pneumatic excitation module is controlled to switch to phase-locked excitation mode, outputting a fixed frequency pulse airflow with a frequency equal to the first natural frequency, and locking for a preset time to make the lawn blades reach the maximum amplitude and form a non-equilibrium temperature field. After the lock-in time is reached, the variable frequency pneumatic excitation module is controlled to instantly cut off the airflow output and enter the free decay and thermal recovery stage.

5. The intelligent sports stadium turf maintenance system based on AI vision according to claim 1, characterized in that, The edge computing processing module also includes a visual front-end processing unit, used to perform the following preprocessing on the collected data: For the visible light image sequence, the dense optical flow algorithm is applied to calculate the displacement vector field of the pixels in the region of interest, and a one-dimensional displacement time-domain signal reflecting the motion state of the leaf group is extracted; For the infrared thermal image sequence, a homography transformation matrix is ​​applied to map the infrared thermal image sequence to the visible light coordinate system, and the corresponding transient temperature response curve is extracted.

6. The intelligent sports stadium turf maintenance system based on AI vision according to claim 5, characterized in that, The edge computing processing module calculates the mechanical dynamic parameters based on the one-dimensional displacement time-domain signal, and the mechanical dynamic parameters include the damping ratio and the equivalent stiffness coefficient. The damping ratio is calculated based on the logarithmic decay method, which is obtained by analyzing the amplitude ratio of adjacent wave peaks in free decaying vibration. The equivalent stiffness coefficient is calculated based on a single-degree-of-freedom vibration theory model, using the inversion relationship between the identified first-order natural frequency and the preset unit blade equivalent mass parameter.

7. The intelligent sports stadium lawn maintenance system based on AI vision according to claim 5, characterized in that, The edge computing processing module calculates the transient thermodynamic parameters based on the transient temperature response curve, and the transient thermodynamic parameters are thermal time constants; The edge computing processing module constructs a linear regression model based on Newton's law of cooling, fits the logarithmic temperature difference decay data after the airflow stops, and determines the thermal time constant by the reciprocal of the slope of the fitted line. The thermal time constant is used to characterize the rate of thermal equilibrium recovery of lawn leaves under stomatal regulation.

8. The intelligent sports stadium turf maintenance system based on AI vision according to claim 6, characterized in that, The specific logic by which the edge computing processing module evaluates the physiological state of the lawn includes: The calculated equivalent stiffness coefficient is compared with a preset stiffness reference threshold, and the thermal time constant is compared with a preset thermal response threshold. The lawn in the current area is determined to be in an early water deficit state if and only if the equivalent stiffness coefficient is less than the stiffness reference threshold and the thermal time constant is greater than the thermal response threshold.

9. The intelligent sports stadium lawn maintenance system based on AI vision according to claim 6, characterized in that, The specific logic for the edge computing processing module to evaluate the physiological state of the lawn also includes: The distribution of the damping ratio is monitored. When the damping ratio is found to be higher than the statistical average of healthy samples and accompanied by a decrease in the equivalent stiffness coefficient, it is determined that the lawn in the current area has tissue necrosis or disease infection.

10. A method for intelligent maintenance of sports stadium turf based on AI vision, characterized in that, An AI vision-based intelligent turf maintenance system for sports venues, applicable to any one of claims 1-9, comprises the following steps: A variable frequency pneumatic excitation module was used to perform a pneumatic frequency sweep on the lawn area to identify the first-order natural frequency of the lawn blades. By applying a fixed-frequency pulsed airflow excitation to the lawn blades using the first-order natural frequency, the blades resonate and the surface thermal boundary layer is stripped away. The airflow excitation was stopped instantaneously, and the free decay vibration image sequence and thermal recovery image sequence of the blade were simultaneously acquired using a dual-modal vision acquisition module; The edge computing processing module decouples and analyzes the acquired image sequence to calculate the damping ratio and equivalent stiffness coefficient that characterize the viscoelastic properties, as well as the thermal time constant that characterizes the heat dissipation properties. Based on the multidimensional characteristic distribution of the damping ratio, the equivalent stiffness coefficient, and the thermal time constant, the water deficit, disease, or growth potential of the lawn is determined, and corresponding precise maintenance operation instructions are generated according to the determination results.

Images