Landscaping maintenance monitoring and early warning method and system

Abstract

The invention discloses a landscaping maintenance monitoring and early warning method and system, and relates to the technical field of garden irrigation, the system comprises a sensing module, an execution module and a communication module, the sensing module is used for collecting image and video data, the execution module is used for executing irrigation and fertilization operation, and the communication module is used for data interaction inside and outside the early warning system. The early warning system further comprises an image recognition module and a decision intelligent agent, the sensing module comprises image acquisition equipment and video acquisition equipment which are arranged on a fixed monitoring point and a mobile platform, and the sensing module is used for acquiring visual data of a plant growth state; the system has the advantages that plant physiological status and pest and disease damage signs are automatically analyzed through the image recognition module, simulation deduction is performed by fusing multi-source environmental data in the digital twinborn body through the decision agent, and optimized decision instructions such as scientific water and fertilizer ratio, irrigation scheduling and pest and disease damage intervention strategies can be generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of garden irrigation, in particular to a method and system for monitoring and early warning of garden maintenance. BACKGROUND

[0002] Traditional garden maintenance work highly depends on regular manual inspection and experience-based judgment, which has strong subjectivity, low efficiency and delayed response, etc. Maintenance personnel are difficult to grasp the physiological state changes of vegetation in real time and comprehensively, and the discovery of pests and diseases is often in the middle and late stages, missing the best prevention opportunity, resulting in high maintenance cost and poor effect. Although there are some automatic irrigation equipment, their operation is mostly based on preset time or simple soil moisture threshold, lacking precise insight into the actual needs of plants, and unable to achieve on-demand maintenance.

[0003] In recent years, although some researches have attempted to use image sensors and other devices for plant monitoring, they are mostly limited to simple analysis of visual information, without deep fusion with environmental data, lacking effective decision support mechanism. These methods usually process data in isolation, with one-sided analysis results, and are difficult to construct and simulate the whole growth process of plants in digital space, making it difficult to form scientific, systematic and forward-looking maintenance decisions, with limited intelligent level, and unable to meet the needs of modern fine garden management. SUMMARY

[0004] The purpose of the present application is to provide a method and system for monitoring and early warning of garden maintenance.

[0005] To achieve the above purpose, the present application provides the following technical solution: a garden maintenance monitoring and early warning system, comprising a perception module, an execution module and a communication module, the perception module is used for collecting image and video data, the execution module is used for performing irrigation and fertilization operations, and the communication module is used for data interaction within and outside the early warning system, the early warning system further comprises an image recognition module and a decision-making agent, the perception module comprises image acquisition equipment and video acquisition equipment arranged on fixed monitoring points and mobile platforms, which are used to obtain visual data of plant growth state, and further comprises soil sensors and meteorological sensors for collecting non-visual environmental data, the image recognition module processes the visual data, identifies plant physiological state and pest and disease signs through feature extraction and semantic understanding, and outputs structured recognition results, the decision-making agent generates optimized decision-making instructions including water and fertilizer ratio, irrigation scheduling and pest and disease intervention strategy based on the structured recognition results output by the image recognition module and the non-visual environmental data collected by the soil sensors and meteorological sensors, and simulates in a digital twin containing soil properties, plant growth state, meteorological conditions and time dimension parameters, the output end of the decision-making agent is connected with the control end of the execution module, and the optimized decision-making instructions are sent to the execution module to perform corresponding operations.

[0006] As a further scheme of the present application: the image acquisition device comprises a multispectral imager and a hyperspectral imager, the multispectral imager captures the spectral reflectance characteristics of plants in multiple specific discrete wave bands, and the hyperspectral imager acquires fine spectral information of plants in a continuous narrow wave band range.

[0007] As a further scheme of the present application: the image recognition module adopts a convolutional neural network model constructed based on a deep learning framework, the convolutional neural network model extracts color distribution features, surface morphology texture features, and crown layer three-dimensional spatial structure features of plant leaves through multiple layers of convolution and pooling operations, and fuses and classifies the color distribution features, surface morphology texture features, and crown layer three-dimensional spatial structure features using a full connection layer and a classifier.

[0008] As a further scheme of the present application: the image recognition module is connected with a historical image database storing plant image data in historical periods, and the image recognition module identifies the dynamic change trend and abnormal performance mode of plants in the entire growth cycle by performing time sequence comparison and difference analysis on real-time collected visual data and historical visual data.

[0009] As a further scheme of the present application: the image recognition module performs real-time processing and analysis on video stream data, and extracts time sequence features representing plant growth dynamics or pest movement behavior by using optical flow method and inter-frame difference method to analyze pixel-level changes between consecutive image frames.

[0010] As a further scheme of the present application: the early warning system further comprises a model optimization module, the model optimization module receives artificial labeling data generated after professional personnel verify image recognition results, and performs supervised training and parameter optimization on the deep learning model in the image recognition module using these data with accurate labels.

[0011] As a further scheme of the present application: the communication module adopts a low-latency network protocol designed for large-scale image and video data transmission, the low-latency network protocol includes data compression, error control, and traffic shaping functions.

[0012] As a further scheme of the present application: in the simulation deduction process of generating optimized decision instructions, the key input variables of the structured recognition results output by the image recognition module make the visual evidence data affect the parameter settings and deduction logic in the simulation model.

[0013] A garden greening maintenance monitoring and early warning method, the early warning method comprising the following steps:

[0014] Step one, synchronously collecting visual data containing image and video sequence and non-visual environmental data collected by soil sensor and meteorological sensor through the perception module;

[0015] Step two, processing the visual data by the image recognition module, outputting structured recognition results quantitatively representing plant physiological state and pest signs;

[0016] Step three, constructing or updating digital twin based on the structured recognition results and the non-visual environmental data;

[0017] Step four, the decision-making agent runs simulation in the digital twin, generating optimized decision-making instructions;

[0018] Step five, the optimized decision-making instructions are issued to the execution module to drive its operation, and the feedback visual data after operation is collected for optimizing the model parameters of the image recognition module and the decision-making agent

[0019] Compared with the prior art, the beneficial effects of the present application are as follows:

[0020] 1. The present application automatically analyzes plant physiological state and pest signs through the image recognition module, and simulates and deduces through the decision-making agent in the digital twin by fusing multi-source environmental data, which can generate scientific water and fertilizer ratio, irrigation scheduling and pest intervention strategies, and automatically drive the execution mechanism to complete the operation, effectively overcoming the uncertainty of human judgment, realizing closed-loop automatic management from perception to execution, greatly reducing the consumption of human resources, avoiding resource waste through precise measures, and significantly promoting the healthy growth of plants;

[0021] 2. The present application not only processes static images, but also has real-time dynamic analysis capability for video stream, which can capture the continuous change of plant growth and the migration behavior of pests, and combine with historical image database for time sequence comparison and trend analysis, which can sensitively identify slow-growing abnormal growth or early weak pest signs, much earlier than the stage that can be observed by naked eye, this early diagnosis capability wins valuable time for taking intervention measures, can suppress the problem in the embryonic state, greatly reduces the difficulty and cost of later treatment, avoids the risk of large-scale disease outbreak;

[0022] 3、The application can continuously train and optimize the deep learning model by using the labeled data fed back by professionals through the special model optimization module, so that the model can continuously learn and adapt to the performance characteristics of different seasons, different climates and different growth stages of plants, thereby effectively overcoming the recognition challenges brought by environmental complexity, enhancing the generalization ability and robustness of the model, and at the same time, the special low-latency communication protocol guarantees the integrity and real-time performance of the massive visual data transmission, providing a reliable data foundation for the back-end analysis, which enables the whole system to run stably for a long time and provide continuous and accurate maintenance decision support. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The garden water and fertilizer irrigation system flowchart in the embodiment of the application. DETAILED DESCRIPTION

[0024] The specific embodiments of the application will be further described below with reference to the accompanying drawings, and it should be noted that the description of these embodiments is used to help understand the application, but does not constitute a limitation on the application.

[0025] In addition, the technical features involved in each of the embodiments of the application described below can be combined with each other as long as they do not conflict with each other.

[0026] Please refer to the accompanying Figure 1 The application is a garden greening maintenance monitoring and early warning system, which comprises a perception module, an execution module and a communication module, the perception module is used for collecting image and video data, the execution module is used for executing irrigation and fertilization operations, and the communication module is used for data interaction within and outside the early warning system, the early warning system further comprises an image recognition module and a decision-making agent, the perception module comprises image acquisition equipment and video acquisition equipment arranged on fixed monitoring points and mobile platforms, which are used for obtaining visual data of plant growth state, and further comprises soil sensors and meteorological sensors for collecting non-visual environmental data, the image recognition module processes the visual data, identifies plant physiological state and pest and disease signs through feature extraction and semantic understanding, and outputs structured recognition results, the decision-making agent simulates in a digital twin containing soil properties, plant growth state, meteorological conditions and time dimension parameters based on the structured recognition results output by the image recognition module and the non-visual environmental data collected by the soil sensors and the meteorological sensors, generates optimization decision instructions containing water and fertilizer ratio, irrigation scheduling and pest and disease intervention strategy, and the output end of the decision-making agent is connected with the control end of the execution module, so as to send the optimization decision instructions to the execution module to perform corresponding operations.

[0027] In an embodiment of the application: the adaptive learning module is a core support module for realizing self-optimization and long-term stable operation of the system, and mainly undertakes the key functions of data storage, model optimization and parameter iteration.

[0028] In terms of data storage, the adaptive learning module automatically receives and stores various types of data collected by the perception module at 23:59 every day, including soil moisture, meteorological parameters, visual data of plant growth, water and fertilizer concentration data, and operation feedback data of the execution module. All data are classified and stored in data units to build a historical database covering all elements of gardens, providing a solid data foundation for subsequent model training and trend analysis.

[0029] In terms of model optimization mechanism, the gradient descent algorithm will be launched every Sunday at 24:00 to optimize and adjust the hidden layer weights of the deep learning model in the decision agent and image recognition module. Through continuous iterative calculation, the model prediction error is reduced by ≥2%, and the optimized parameter package is generated and pushed to the relevant modules of the system to ensure that the model can always adapt to the dynamic changes of the garden environment and improve the accuracy of decision-making.

[0030] Meanwhile, the adaptive learning module also has a data preprocessing function. Before the main module calls the neural network algorithm at 06:30 every day, it will perform preprocessing operations such as cleaning, normalization and feature extraction on the raw data of the past 6 hours to provide high-quality input data for the neural network algorithm and ensure the accuracy of the adaptation coefficient matrix generation.

[0031] In one embodiment of the present invention: the remote monitoring module is an important module for realizing human-computer interaction and global control of system status, and is mainly responsible for basic data configuration during system startup and status monitoring during operation.

[0032] During the system initialization phase, after the main controller wakes up each module, the remote monitoring module will first receive the readiness notification from each module to confirm that the core components such as the perception module, execution module, and image recognition module have started normally. Subsequently, the operator can enter garden zoning information through the interactive interface of the remote monitoring module, and at the same time enter the soil type data of each zoning, including key parameters such as soil texture, basic soil fertility value, and initial soil pH value. These basic data will serve as the initial input conditions for the construction of the digital twin.

[0033] During system operation, the remote monitoring module receives various data and status information transmitted by the main controller in real time. It dynamically displays soil moisture curves, real-time meteorological data, plant growth status identification results, and operation records of execution modules in each garden zone through a visual interface. When the system triggers the early warning mechanism, the remote monitoring module will receive the early warning signal immediately and notify the operator through sound and light alarms, pop-up prompts, etc. At the same time, it supports the operator to manually intervene and adjust the system parameters, realizing the organic combination of automation and manual control.

[0034] In an embodiment of the present application: the real-time calibration module is a key auxiliary module to ensure the operation accuracy of the execution module, through multi-dimensional parameter monitoring and dynamic correction, to ensure the accuracy of key execution parameters such as water and fertilizer ratio and irrigation amount;

[0035] The real-time calibration module establishes real-time communication connection with the core devices such as screw pump and variable frequency water pump in the execution module, continuously collects temperature data, soil and water and fertilizer mixture conductivity data, and device cumulative running time information during device operation, based on the preset parameter standard range, real-time analyzes the deviation degree of each monitoring data and standard value;

[0036] When the deviation of key indicators such as pH value and trace element concentration is ≥±3% during water and fertilizer mixing, the real-time calibration module will immediately send adjustment instructions to the execution module to accurately control the speed of the screw pump until the deviation is controlled within the allowed range. In the irrigation operation, the module dynamically calibrates the irrigation amount by monitoring the water pump running power, pipeline pressure and other parameters, ensures the effective execution of the scheduling instructions such as accurately reducing the irrigation amount in B area to 70% and accurately increasing the water pump power in C area to 80%, and guarantees the accuracy and reliability of the maintenance operation.

[0037] Example one,

[0038] Urban comprehensive park maintenance scene

[0039] System deployment overview:

[0040] This embodiment is applied to the core landscape area of a city park with an area of 3500m 2 , which is divided into three functional subareas:

[0041] Arboretum viewing area (A area): area 1500m 2 , soil type is loam, pH value is 6.5-7.2, mainly planting landscape trees such as ginkgo and magnolia;

[0042] Flower border display area (B area): area 1200m 2 , soil type is sandy loam, pH value is 6.0-6.8, mainly root flowers such as hydrangea and daylily;

[0043] Leisure lawn area (C area): area 800m 2 , soil type is clay loam, pH value is 7.0-7.5, planting four seasons green lawn;

[0044] Hardware configuration:

[0045] Sensing module: 4 fixed monitoring points are set in each partition, equipped with multispectral imager, hyperspectral imager and high-definition video camera, 6 soil moisture sensors and 6 conductivity sensors, 5 weather sensors, and mobile inspection vehicles collect dynamic visual data at 9:00 and 15:00 every day according to the preset route;

[0046] Execution module: Screw pump and frequency conversion water pump (3kW) are configured in A area, pipeline mixer, multi-stage mixing device and intelligent spraying equipment are set in B area, and automatic irrigation valve group and pressure sensor are installed in C area.

[0047] Communication and storage: LoRa wireless communication network is adopted, low latency protocol supports data compression ratio of 8:1, 8TB data unit stores historical image database and environmental monitoring data.

[0048] System operation and data details

[0049] Initialization configuration:

[0050] The operator enters the basic data through the remote monitoring module:

[0051] Partition boundary coordinates and soil parameters, initial soil moisture in A area is 25%, B area is 22%, and C area is 30%;

[0052] Plant growth characteristics data;

[0053] Daily operation data

[0054] Data collection: The sensing module collects data every 20 minutes, the soil moisture in A area is 26%, the air humidity in B area is 60%, and the leaf spectral reflectance in C area is 70%, and the video stream extracts the dynamic change characteristics of the leaves through the optical flow method;

[0055] Image recognition: Convolutional neural network extracts leaf texture features in B area, identifies aphid risk of 62%, and finds abnormal leaf color change rate of 8% compared with historical database;

[0056] Decision generation: The main module calls neural network algorithm at 06:30 every day, inputs 6 hours of preprocessed data after 1100 iterations, generates adaptive coefficient matrix, and adjusts the phosphorus ratio in B area to 1:2 by decision-making agent;

[0057] Execution control: The sub-controller starts the frequency conversion water pump, the fertilizer in B area is delivered to the multi-stage mixing device through the pipeline mixer, the water injection pressure is 0.3MPa, the trace elements are added, and the pH is adjusted to 6.3;

[0058] Abnormal processing: Rainwater storage is 28%, irrigation amount in B area is reduced from 45m 3 / day to 31.5m 3 , air water extraction equipment is started synchronously, and daily average water production is 4m 3 ;

[0059] Model optimization: daily data is stored at 23:59, including irrigation amount, fertilizer amount and pest control records, and the model is optimized every Sunday at 24:00 using gradient descent algorithm, with the error reduced from 4.9% to 2.6% in the first week;

[0060] Implementation effect:

[0061] Irrigation accuracy is improved by 28%, saving 320m of water per month 3 ;

[0062] Early identification rate of pests and diseases is 90%, and prevention cost is reduced by 38%;

[0063] Manual inspection workload is reduced by 55%, and plant survival rate is improved to 97%.

[0064] Example two,

[0065] Industrial park green maintenance scenario

[0066] System deployment overview:

[0067] This example is applied to an industrial park green area of 2800m 2 , which is divided into three functional zones:

[0068] Arboretum area (A area): 1000m 2 , soil type is loam, pH value is 6.8-7.5, and camphor and magnolia grandiflora are planted;

[0069] Green belt shrub area (B area): 1000m 2 , soil type is sandy loam, pH value is 6.2-6.8, and holly and red leaf plum are planted;

[0070] Leisure area lawn (C area): 800m 2 , soil type is clay loam, pH value is 7.2-7.8, and tall fescue lawn is planted;

[0071] Hardware configuration:

[0072] Sensing module: 3 fixed monitoring points are set in each zone, equipped with multispectral / hyperspectral imager and video acquisition equipment, 5 soil sensors, 4 weather sensors, and unmanned aerial vehicle panoramic data acquisition every week at 14:00;

[0073] Execution module: A area is equipped with screw pump and variable frequency water pump, B area is equipped with multi-stage mixing device and timing spray equipment, and C area is equipped with intelligent irrigation system and power monitor;

[0074] Communication and storage: LoRa network full coverage, low latency protocol supports error control function; 6TB data unit stores historical data and model parameters;

[0075] System operation and data details:

[0076] Data acquisition: the perception module collects data every 30 minutes, the soil conductivity in area A is 1.5mS / cm, the temperature in area B is 26℃, the canopy height in area C is 4.2cm, and the video stream is analyzed by frame difference method to analyze the plant growth dynamics;

[0077] Image recognition: the convolutional neural network extracts the leaf color distribution features in area A, compares with historical data to identify the growth trend, and the accuracy of disease and pest risk identification in area B is 88%;

[0078] Decision generation: the main module inputs 6 hours of data every day at 06:30, generates a coefficient matrix after 1050 iterations, controls the screw pump speed to 22L / h, and the real-time calibration module corrects the parameters through the device temperature;

[0079] Execution control: the multi-stage mixing device adjusts the water and fertilizer pH to 6.5 with a deviation of 1.5%, and the irrigation in area C is performed in advance during the low period, and the water pump power is increased to 80%;

[0080] Abnormal processing: the pH of water and fertilizer mixing in area B temporarily deviates by 4%, the real-time calibration module adjusts the screw pump speed to 25L / h, and after 30 minutes the deviation decreases to 2%;

[0081] Model optimization: store data to the adaptive learning module every day, optimize the hidden layer weight of the model every Sunday, and the error decreases from 3.8% to 1.7% in the second week;

[0082] Implementation effect:

[0083] Irrigation energy consumption is reduced by 22%, saving 150 degrees of electricity per month;

[0084] The response time of plant growth anomaly identification is shortened to 2.5 hours;

[0085] Water and fertilizer utilization rate is improved by 32%, reducing fertilizer consumption by 90kg per year.

[0086] According to the above two groups of example contents, the system content realizes the closed-loop automatic management from perception to execution, significantly improves the maintenance accuracy, resource utilization rate and early prevention and control ability of diseases and pests, and provides a reliable technical scheme for fine garden management.

[0087] Although the present application is disclosed with reference to the preferred embodiments, it is not intended to limit the present application and any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, any modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application without departing from the technical solutions of the present application, all fall within the protection scope defined by the claims of the present application.

Claims

1. A landscape greening maintenance monitoring and early warning system, comprising a sensing module, an execution module, and a communication module, wherein the sensing module is used to collect image and video data, the execution module is used to perform irrigation and fertilization operations, and the communication module is used for data interaction within and outside the early warning system, characterized in that, The early warning system also includes an image recognition module and a decision-making agent. The perception module includes image acquisition devices and video acquisition devices set at fixed monitoring points and mobile platforms to acquire visual data on plant growth status. It also includes soil sensors and meteorological sensors to collect non-visual environmental data. The image recognition module processes the visual data, identifies plant physiological status and signs of pests and diseases through feature extraction and semantic understanding, and outputs structured recognition results. The decision-making agent, based on the structured recognition results output by the image recognition module and integrating the non-visual environmental data collected by the soil and meteorological sensors, performs simulation in a digital twin containing parameters of soil properties, plant growth status, meteorological conditions, and time dimension, and generates optimized decision instructions including water and fertilizer ratio, irrigation scheduling, and pest and disease intervention strategies. The output end of the decision-making agent is connected to the control end of the execution module to send the optimized decision instructions to the execution module to perform the corresponding operations.

2. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: The image acquisition device includes a multispectral imager and a hyperspectral imager. The multispectral imager captures the spectral reflectance characteristics of plants in multiple specific discrete bands, and the hyperspectral imager acquires fine spectral information of plants in a continuous narrow band range.

3. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: The image recognition module adopts a convolutional neural network model built on a deep learning framework. The convolutional neural network model extracts the color distribution features, surface morphology and texture features, and canopy three-dimensional spatial structure features of plant leaves through multi-layer convolution and pooling operations. It then uses fully connected layers and classifiers to fuse and classify the color distribution features, surface morphology and texture features, and canopy three-dimensional spatial structure features for recognition.

4. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: The image recognition module is connected to a historical image database that stores plant image data from different historical periods. By comparing and analyzing the real-time visual data with the historical visual data over time, the image recognition module identifies the dynamic change trends and abnormal behavior patterns of plants throughout their growth cycle.

5. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: The image recognition module processes and analyzes video stream data in real time. By using optical flow and inter-frame difference methods to analyze pixel-level changes between consecutive image frames, it extracts temporal features that characterize the dynamic process of plant growth or the movement and migration behavior of pests and diseases.

6. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: The early warning system also includes a model optimization module, which receives manually labeled data generated by professionals after verifying the image recognition results, and uses this data with accurate labels to perform supervised training and parameter optimization of the deep learning model in the image recognition module.

7. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: The communication module employs a low-latency network protocol designed for large-scale image and video data transmission, which includes data compression, error control, and traffic shaping functions.

8. The garden greening maintenance monitoring and early warning system according to claim 1, characterized in that: During the simulation and deduction process of generating optimized decision instructions, the key input variables of the structured recognition results output by the image recognition module cause visual evidence data to influence the parameter settings and deduction logic in the simulation model.

9. A method for monitoring and early warning of landscaping maintenance applied to the system described in any one of claims 1-8, characterized in that, The early warning method includes the following steps: Step 1: Simultaneously acquire visual data including image and video sequences, as well as non-visual environmental data acquired by soil and meteorological sensors, through the sensing module. Step 2: Process the visual data using the image recognition module to output structured recognition results that quantitatively characterize the plant's physiological state and signs of pests and diseases; Step 3: Based on the structured recognition results and the non-visual environment data, construct or update the digital twin; Step 4: The decision-making agent runs a simulation in the digital twin to generate optimized decision instructions; Step 5: Send the optimization decision instruction to the execution module to drive it to perform the operation, and collect the feedback visual data after the operation to optimize the model parameters of the image recognition module and the decision agent.

Images