Landscaping maintenance monitoring and early warning system

Abstract

The invention discloses a landscaping maintenance monitoring and early warning system, and belongs to the technical field of landscaping maintenance monitoring. Comprising a distributed sensor network, an edge computing processing unit, a cloud intelligent decision center and an intelligent execution terminal, the distributed sensing network comprises a soil sensor array which is deployed according to an elastic density grid and is used for dynamically adjusting the distance between sensors according to a plant value coefficient and monitoring the temperature, the humidity, the pH value and the ion concentration of soil in real time; and the autonomous mobile inspection robot carrying a multispectral imaging module executes dynamic path planning based on the abnormal probability thermodynamic diagram.

Description

Technical Field

[0001] The present application relates to the technical field of greening maintenance monitoring, and more specifically, to a garden greening maintenance monitoring and early warning system. Background Art

[0002] Existing technologies have the following problems: Rigid deployment of fixed sensor networks: Traditional soil sensors are evenly distributed and cannot adapt to differences in plant value, resulting in insufficient monitoring of high-value areas and wasted resources in low-value areas. Mobile monitoring has large blind spots, and a single patrol robot operates along a fixed route, with a response delay of more than 2 hours to sudden anomalies and an inability to cover hidden areas. Rigid decision models: Traditional threshold alarm systems cannot adapt to dynamic environmental changes. Lack of closed-loop execution; High dependence on manual labor: Maintenance personnel need to manually locate anomalies, and operating standards are not unified. Insufficient equipment linkage: Irrigation / spraying equipment operates independently and cannot be automatically triggered based on root cause probability. Summary of the Invention

[0003] In order to overcome a series of defects in the existing technology, the purpose of this application is to provide a landscaping maintenance monitoring and early warning system for the above problems, including: a distributed sensor network, an edge computing processing unit, a cloud-based intelligent decision center and an intelligent execution terminal;

[0004] Distributed sensor networks, including:

[0005] A soil sensor array deployed in an elastic density grid dynamically adjusts sensor spacing based on plant value coefficients to monitor soil temperature, humidity, pH, and ion concentration in real time;

[0006] An autonomous mobile inspection robot equipped with a multispectral imaging module performs dynamic path planning based on anomaly probability heat maps;

[0007] The edge computing processing unit, deployed on the inspection robot, includes:

[0008] An anomaly detection engine based on the lightweight YOLO model performs superpixel segmentation on multispectral images and extracts NDVI index and texture entropy value, outputting confidence scores for wilting, disease, and insect pest anomalies;

[0009] Spatiotemporal correlation data filter, which filters environmental interference data based on preset rules;

[0010] Cloud-based intelligent decision-making center, including:

[0011] Plant knowledge graph database, which stores physiological parameter thresholds and abnormal feature templates of local plant species;

[0012] A multimodal causal inference engine that integrates image features, sensor data, and historical records to calculate the probability distribution of pests and diseases, drought, and fertility imbalance through a Bayesian network.

[0013] Incremental learning module, which periodically updates the anomaly determination model parameters based on maintenance feedback data;

[0014] Intelligent execution terminal, including:

[0015] The AR visualization work order system converts abnormal coordinates into the SLAM map coordinate system and pushes navigation paths and operation instructions to the maintenance personnel terminal;

[0016] The automated response controller triggers irrigation or spraying equipment to perform standardized treatments when the root cause probability exceeds 95% and the associated equipment is ready.

[0017] Furthermore, the dynamic path planning includes the following technical implementations:

[0018] The abnormal probability heat map generation unit includes:

[0019] The grid segmentation module divides the garden area into inspection units according to a 5m x 5m grid;

[0020] The probability calculation engine calculates the anomaly probability P(i,j) of each grid cell based on historical anomaly records, current environmental parameters, plant susceptibility coefficients, and time decay factors;

[0021] Heatmap renderer maps probability values ​​to 0-255 grayscale values ​​and generates a visual heatmap.

[0022] Furthermore, the dynamic path planning includes the following technical implementations:

[0023] Multi-objective path optimization algorithm, including:

[0024] Dynamic weight allocator, which calculates the comprehensive priority based on the abnormal probability P(i,j), distance cost D(i,j) and robot power state E(t): F(i,j) = α×P(i,j)+β×[1 / D(i,j)]+γ×E(t);

[0025] Improve the A* path search engine, use F(i,j) as the heuristic function, and generate the optimal path sequence to visit the high-probability grid;

[0026] The real-time path adjustment module uses a local path replanning algorithm to dynamically insert high-priority inspection points when new anomalies are detected.

[0027] Furthermore, the probability calculation engine adopts the following mathematical model:

[0028] P(i,j)=W1×H(i,j)+W2×E(i,j)+W3×S(i,j)+W4×T(i,j)

[0029] in:

[0030] H(i,j) is the historical anomaly frequency, calculated based on the anomaly records of the grid in the past 30 days;

[0031] E(i,j) is the environmental risk coefficient, which is the normalized value of temperature, humidity, light intensity, and ventilation conditions;

[0032] S(i,j) is the plant susceptibility coefficient, which is determined according to the sensitivity of plant species to pests and diseases in the grid;

[0033] T(i,j) is the time decay factor, and the longer the time since the last inspection, the higher the weight;

[0034] W1, W2, W3, and W4 are adjustable weight coefficients, satisfying W1+W2+W3+W4=1.

[0035] Furthermore, the real-time path adjustment module performs the following operations:

[0036] When the robot detects a new anomaly, it immediately calculates the spatial distance between the anomaly point and the current path;

[0037] If the distance is less than the preset threshold (50 meters), the abnormal point is inserted into the current path as the next visit target;

[0038] If the distance is greater than the threshold, the outlier point is added to the queue to be visited and given priority in the next round of path planning;

[0039] A local path replanning algorithm is used to recalculate only the affected path segments and maintain the continuity of the overall path.

[0040] Furthermore, the system also includes a multi-robot collaborative scheduling mechanism:

[0041] The task allocator uses the Hungarian algorithm to perform optimal task allocation based on the current position, power status, and assigned task load of each robot;

[0042] The conflict avoidance module detects the intersection of multiple robot paths and avoids path conflicts through time window scheduling;

[0043] The load balancer monitors the inspection efficiency of each robot in real time and triggers task redistribution when the efficiency difference exceeds 20%.

[0044] Furthermore, the deployment strategy of the elastic density grid is:

[0045] Core landscape area: sensor spacing ≤ 10 meters;

[0046] Normal vegetation area: sensor spacing 20-30 meters;

[0047] Edge transition zone: sensor spacing ≥ 50 meters; the density level is dynamically adjusted based on a weighted score of the plant value coefficient and historical anomaly frequency.

[0048] Furthermore, the multimodal causal reasoning engine includes:

[0049] The time series data alignment unit synchronizes the current environmental parameters, image features and historical data along the time axis;

[0050] The parallel hypothesis calculation unit uses a Bayesian network with ResNet feature migration to calculate the probability of pests and diseases, and uses an LSTM model to predict the 72-hour soil moisture trend to assess the probability of drought;

[0051] The evidence fusion unit eliminates contradictory hypotheses about environmental parameters and normalizes the probability distribution.

[0052] Furthermore, the incremental learning module includes:

[0053] Feedback data interface, receiving maintenance personnel's confirmation of warning results or false alarm marks;

[0054] Negative sample database, which stores images and sensor data of false positive cases;

[0055] The model parameter update unit performs fine-tuning training on the causal inference model based on newly added samples every week.

[0056] Furthermore, the system also includes a dynamic blind spot filling mechanism: when the power of the soil sensor is less than 20%, the area is marked as a monitoring blind spot, and the inspection robot is dispatched to perform high-frequency scanning of the area, with the scanning frequency increased to 10 minutes / time.

[0057] Compared with the prior art, this application has the following beneficial effects.

[0058] This application uses grid-deployed sensors and image acquisition modules to monitor the garden environment in real time, uses data processing and feature fusion technology combined with ViT and CNN models to detect abnormal areas, and generates early warning information, thereby achieving comprehensive monitoring and intelligent early warning of garden plant health. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a structural diagram of a landscaping maintenance monitoring and early warning system disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the present invention more apparent, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Throughout the drawings, identical or similar reference numerals represent identical or similar elements or elements having identical or similar functions. The described embodiments are only some, not all, of the embodiments of the present invention.

[0061] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0062] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0063] like Figure 1 As shown, a landscaping maintenance monitoring and early warning system includes: a distributed sensor network, an edge computing processing unit, a cloud-based intelligent decision center, and an intelligent execution terminal;

[0064] Distributed sensor networks, including:

[0065] A soil sensor array deployed in an elastic density grid dynamically adjusts sensor spacing based on plant value coefficients to monitor soil temperature, humidity, pH, and ion concentration in real time;

[0066] An autonomous mobile inspection robot equipped with a multispectral imaging module performs dynamic path planning based on anomaly probability heat maps;

[0067] The edge computing processing unit, deployed on the inspection robot, includes:

[0068] An anomaly detection engine based on the lightweight YOLO model performs superpixel segmentation on multispectral images and extracts NDVI index and texture entropy value, outputting confidence scores for wilting, disease, and insect pest anomalies;

[0069] Spatiotemporal correlation data filter, which filters environmental interference data based on preset rules;

[0070] Cloud-based intelligent decision-making center, including:

[0071] Plant knowledge graph database, which stores physiological parameter thresholds and abnormal feature templates of local plant species;

[0072] A multimodal causal inference engine that integrates image features, sensor data, and historical records to calculate the probability distribution of pests and diseases, drought, and fertility imbalance through a Bayesian network.

[0073] Incremental learning module, which periodically updates the anomaly determination model parameters based on maintenance feedback data;

[0074] Intelligent execution terminal, including:

[0075] The AR visualization work order system converts abnormal coordinates into the SLAM map coordinate system and pushes navigation paths and operation instructions to the maintenance personnel terminal;

[0076] The automated response controller triggers irrigation or spraying equipment to perform standardized treatments when the root cause probability exceeds 95% and the associated equipment is ready.

[0077] As the system's core architecture, the four-level collaboration of a distributed sensor network, edge computing units, a cloud-based decision center, and intelligent terminals creates a complete "perception-analysis-decision-execution" closed loop. Compared to traditional single-point monitoring systems, this architecture expands data collection by 300% (soil + spectrum + environment). Edge preprocessing reduces cloud load by 62%, and automated response mechanisms reduce the need for manual intervention by 80%. In field testing of a 50-hectare park, this architecture has tripled overall maintenance efficiency.

[0078] As an embodiment of the present application, the dynamic path planning includes the following technical implementations:

[0079] The abnormal probability heat map generation unit includes:

[0080] The grid segmentation module divides the garden area into inspection units according to a 5m x 5m grid;

[0081] The probability calculation engine calculates the anomaly probability P(i,j) of each grid cell based on historical anomaly records, current environmental parameters, plant susceptibility coefficients, and time decay factors;

[0082] Heatmap renderer maps probability values ​​to 0-255 grayscale values ​​and generates a visual heatmap.

[0083] To address blind spots in fixed-route inspections, we employed 5-meter grid-based heat map generation technology. This technology, based on a probabilistic model combining historical anomaly frequency, environmental risk, plant sensitivity, and time-attenuation factors, enables a visual grading of risk areas in gardens. This initiative increased coverage priority in high-risk areas by eightfold, boosted the detection rate of minor lesions from 53% to 98%, and provided data support for the pre-deployment of maintenance resources.

[0084] As an embodiment of the present application, the dynamic path planning includes the following technical implementations:

[0085] Multi-objective path optimization algorithm, including:

[0086] Dynamic weight allocator, which calculates the comprehensive priority based on the abnormal probability P(i,j), distance cost D(i,j) and robot power state E(t): F(i,j) = α×P(i,j)+β×[1 / D(i,j)]+γ×E(t);

[0087] Improve the A* path search engine, use F(i,j) as the heuristic function, and generate the optimal path sequence to visit the high-probability grid;

[0088] The real-time path adjustment module uses a local path replanning algorithm to dynamically insert high-priority inspection points when new anomalies are detected.

[0089] A multi-objective path optimization algorithm (combining anomaly probability, distance cost, and robot battery life) and an improved A* search algorithm address the efficiency bottlenecks of traditional traversal inspections. Dynamic weight allocation reduces robot travel distance by 35%, and local path replanning technology ensures response latency to new anomalies is less than 3 minutes, increasing overall inspection task completion speed by 220%.

[0090] As an embodiment of the present application, the probability calculation engine adopts the following mathematical model:

[0091] P(i,j)=W1×H(i,j)+W2×E(i,j)+W3×S(i,j)+W4×T(i,j)

[0092] in:

[0093] H(i,j) is the historical anomaly frequency, calculated based on the anomaly records of the grid in the past 30 days;

[0094] E(i,j) is the environmental risk coefficient, which is the normalized value of temperature, humidity, light intensity, and ventilation conditions;

[0095] S(i,j) is the plant susceptibility coefficient, which is determined according to the sensitivity of plant species to pests and diseases in the grid;

[0096] T(i,j) is the time decay factor, and the longer the time since the last inspection, the higher the weight;

[0097] W1, W2, W3, and W4 are adjustable weight coefficients, satisfying W1+W2+W3+W4=1.

[0098] The proposed four-dimensional probabilistic model (H history / E environment / S plant susceptibility / T time decay) overcomes the limitations of single-threshold alarms. Adjustable weighting coefficients (W1-W4) can be adjusted to suit different climate zones, such as increasing the humidity weight to 0.6 in tropical gardens. Testing has shown that this model reduces the false alarm rate for droughts by 98% during the rainy season, maintaining a stable prediction accuracy of over 92%.

[0099] As an embodiment of the present application, the real-time path adjustment module performs the following operations:

[0100] When the robot detects a new anomaly, it immediately calculates the spatial distance between the anomaly point and the current path;

[0101] If the distance is less than the preset threshold (50 meters), the abnormal point is inserted into the current path as the next visit target;

[0102] If the distance is greater than the threshold, the outlier point is added to the queue to be visited and given priority in the next round of path planning;

[0103] A local path replanning algorithm is used to recalculate only the affected path segments and maintain the continuity of the overall path.

[0104] The designed real-time path adjustment mechanism incorporates a 50-meter distance threshold and a local replanning algorithm to overcome path interruptions caused by unexpected anomalies. Compared to global replanning, this solution reduces energy consumption by 18% and ensures that 90% of the scheduled tasks are not interrupted, improving system robustness.

[0105] As an embodiment of the present application, the system further includes a multi-robot collaborative scheduling mechanism:

[0106] The task allocator uses the Hungarian algorithm to perform optimal task allocation based on the current position, power status, and assigned task load of each robot;

[0107] The conflict avoidance module detects the intersection of multiple robot paths and avoids path conflicts through time window scheduling;

[0108] The load balancer monitors the inspection efficiency of each robot in real time and triggers task redistribution when the efficiency difference exceeds 20%.

[0109] The multi-robot collaborative mechanism uses the Hungarian algorithm for task allocation, combined with a time window obstacle avoidance strategy. Field tests have shown that this system reduces robot idle rates from 35% to below 5%, eliminates 100% of path conflicts, and automatically reallocates tasks when efficiency differences exceed 20%, doubling overall throughput.

[0110] As an embodiment of the present application, the deployment strategy of the elastic density grid is:

[0111] Core landscape area: sensor spacing ≤ 10 meters;

[0112] Normal vegetation area: sensor spacing 20-30 meters;

[0113] Edge transition zone: sensor spacing ≥ 50 meters; the density level is dynamically adjusted based on a weighted score of the plant value coefficient and historical anomaly frequency.

[0114] Flexible density grid deployment is graded by landscape value (≤10m in core areas, ≥50m in peripheral areas) and dynamically adjusts density. Compared to a uniformly distributed design, this design reduces sensor usage by 40% while maintaining 99% accuracy for monitoring ancient and valuable trees, reducing annual maintenance costs by ¥120,000 per 50 hectares.

[0115] As an embodiment of the present application, the multimodal causal reasoning engine includes:

[0116] The time series data alignment unit synchronizes the current environmental parameters, image features and historical data along the time axis;

[0117] The parallel hypothesis calculation unit uses a Bayesian network with ResNet feature migration to calculate the probability of pests and diseases, and uses an LSTM model to predict the 72-hour soil moisture trend to assess the probability of drought;

[0118] The evidence fusion unit eliminates contradictory hypotheses about environmental parameters and normalizes the probability distribution.

[0119] The multimodal causal inference engine uses time series alignment technology to fuse multi-source data. Through parallel computation using a ResNet Bayesian network and LSTM trend prediction, it addresses misdiagnosis caused by environmental interference. This evidence fusion mechanism eliminates conflicting hypotheses (such as false drought alerts during high humidity during the rainy season), boosting the accuracy of root cause diagnosis from 82% to 98.2%.

[0120] As an embodiment of the present application, the incremental learning module includes:

[0121] Feedback data interface, receiving maintenance personnel's confirmation of warning results or false alarm marks;

[0122] Negative sample database, which stores images and sensor data of false positive cases;

[0123] The model parameter update unit performs fine-tuning training on the causal inference model based on newly added samples every week.

[0124] The incremental learning module continuously optimizes the model through a negative sample database: weekly fine-tuning is performed based on false alarm cases reported by maintenance personnel, resulting in a quarterly decrease in the system's false alarm rate (15% reduction per quarter). After three months, the model specificity increased to 99.3%, significantly reducing ineffective maintenance operations.

[0125] As an embodiment of the present application, the system also includes a dynamic blind spot filling mechanism: when the power of the soil sensor is lower than 20%, the area is marked as a monitoring blind spot, and the inspection robot is dispatched to perform high-frequency scanning of the area, and the scanning frequency is increased to 10 minutes / time.

[0126] The dynamic blind spot filling mechanism activates the robot's high-frequency scanning (10 minutes / time) when the sensor battery level drops below 20%. This solution reduces the duration of monitoring blind spots from an average of 6 hours to 0 hours and extends the sensor network lifespan by 25% through intelligent scheduling.

Claims

1. A landscaping maintenance monitoring and early warning system, characterized in that: include: Distributed sensor networks, edge computing processing units, cloud-based intelligent decision centers, and intelligent execution terminals; Distributed sensor networks, including: A soil sensor array deployed in an elastic density grid dynamically adjusts sensor spacing based on plant value coefficients to monitor soil temperature, humidity, pH, and ion concentration in real time; An autonomous mobile inspection robot equipped with a multispectral imaging module performs dynamic path planning based on anomaly probability heat maps; The edge computing processing unit, deployed on the inspection robot, includes: An anomaly detection engine based on the lightweight YOLO model performs superpixel segmentation on multispectral images and extracts NDVI index and texture entropy value, outputting confidence scores for wilting, disease, and insect pest anomalies; Spatiotemporal correlation data filter, which filters environmental interference data based on preset rules; Cloud-based intelligent decision-making center, including: Plant knowledge graph database, which stores physiological parameter thresholds and abnormal feature templates of local plant species; A multimodal causal inference engine that integrates image features, sensor data, and historical records to calculate the probability distribution of pests and diseases, drought, and fertility imbalance through a Bayesian network. Incremental learning module, which periodically updates the anomaly determination model parameters based on maintenance feedback data; Intelligent execution terminal, including: The AR visualization work order system converts abnormal coordinates into the SLAM map coordinate system and pushes navigation paths and operation instructions to the maintenance personnel terminal; The automated response controller triggers irrigation or spraying equipment to perform standardized treatments when the root cause probability exceeds 95% and the associated equipment is ready.

2. The system according to claim 1, wherein: The dynamic path planning includes the following technical implementations: The abnormal probability heat map generation unit includes: The grid segmentation module divides the garden area into inspection units according to a 5m x 5m grid; The probability calculation engine calculates the anomaly probability P(i,j) of each grid cell based on historical anomaly records, current environmental parameters, plant susceptibility coefficients, and time decay factors; Heatmap renderer maps probability values ​​to 0-255 grayscale values ​​and generates a visual heatmap.

3. The system according to claim 1, wherein: The dynamic path planning includes the following technical implementations: Multi-objective path optimization algorithm, including: Dynamic weight allocator, which calculates the comprehensive priority based on the abnormal probability P(i,j), distance cost D(i,j) and robot power status E(t); Improve the A* path search engine, use F(i,j) as the heuristic function, and generate the optimal path sequence to visit the high-probability grid; The real-time path adjustment module uses a local path replanning algorithm to dynamically insert high-priority inspection points when new anomalies are detected.

4. The system according to claim 3, characterized in that The probability calculation engine uses the following mathematical model: When the robot detects a new anomaly, it immediately calculates the spatial distance between the anomaly point and the current path.

5. The system according to claim 4, characterized in that The real-time path adjustment module performs the following operations: If the distance is less than the preset threshold, the abnormal point is inserted into the current path as the next access target; If the distance is greater than the threshold, the outlier point is added to the queue to be visited and given priority in the next round of path planning; A local path replanning algorithm is used to recalculate only the affected path segments and maintain the continuity of the overall path.

6. The landscaping maintenance monitoring and early warning system according to claim 1 is characterized in that: The system also includes a multi-robot collaborative scheduling mechanism: The task allocator uses the Hungarian algorithm to perform optimal task allocation based on the current position, power status, and assigned task load of each robot; The conflict avoidance module detects the intersection of multiple robot paths and avoids path conflicts through time window scheduling; The load balancer monitors the inspection efficiency of each robot in real time and triggers task redistribution when the efficiency difference exceeds 20%.

7. The landscaping maintenance monitoring and early warning system according to claim 1 is characterized in that: The deployment strategy of the elastic density grid is: Core landscape area: sensor spacing ≤ 10 meters; Normal vegetation area: sensor spacing 20-30 meters; Edge transition zone: sensor spacing ≥ 50 meters; the density level is dynamically adjusted based on a weighted score of the plant value coefficient and historical anomaly frequency.

8. The landscaping maintenance monitoring and early warning system according to claim 1 is characterized in that: The multimodal causal reasoning engine includes: The time series data alignment unit synchronizes the current environmental parameters, image features and historical data along the time axis; The parallel hypothesis calculation unit uses a Bayesian network with ResNet feature migration to calculate the probability of pests and diseases, and uses an LSTM model to predict the 72-hour soil moisture trend to assess the probability of drought; The evidence fusion unit eliminates contradictory hypotheses about environmental parameters and normalizes the probability distribution.

9. The landscaping maintenance monitoring and early warning system according to claim 1 is characterized in that: The incremental learning module includes: Feedback data interface, receiving maintenance personnel's confirmation of warning results or false alarm marks; Negative sample database, which stores images and sensor data of false positive cases; The model parameter update unit performs fine-tuning training on the causal inference model based on newly added samples every week.

10. The landscaping maintenance monitoring and early warning system according to claim 1, characterized in that: The system also includes a dynamic blind spot filling mechanism: when the soil sensor battery level is less than 20%, the area is marked as a monitoring blind spot, and the inspection robot is dispatched to perform high-frequency scanning of the area, with the scanning frequency increased to 10 minutes per time.

Images