A Smart Land Consolidation Monitoring Method and System Based on the Internet of Things

Abstract

This application relates to an IoT-based intelligent land remediation monitoring method and system. The monitoring method includes: acquiring satellite remote sensing elevation data and historical soil composition data of the land area to be monitored; dividing the land area to be monitored into multiple grid units and assigning grid identifiers; deploying fixed soil sensor nodes and mobile drone nodes to generate a sensor network topology; collecting real-time soil parameters of each grid unit and performing noise filtering and anomaly detection to generate a soil composition dataset; inputting the soil composition dataset and historical soil composition data into a cloud-based prediction model to output a land degradation risk score and remediation strategy; controlling the execution equipment to perform land remediation operations and monitoring changes in soil parameters after execution in real time; updating the parameters of the cloud-based prediction model and optimizing the sensor network topology based on the deviation between the changes in soil parameters after execution and the expected values. This application improves agricultural production efficiency and soil health.

Description

Technical Field

[0001] This application relates to the field of agricultural information technology, and in particular to an intelligent land consolidation monitoring method and system based on the Internet of Things. Background Technology

[0002] With the increasing demands of modern agriculture and ecological protection, traditional land consolidation and management methods are no longer sufficient to meet the challenges posed by increasingly complex environmental changes and diverse human activities. Especially in areas such as large-scale farmland development, mine reclamation, and desertification control, achieving accurate assessment, dynamic monitoring, and scientific intervention of soil quality has become a critical challenge that urgently needs to be addressed.

[0003] For a long time, land consolidation work has relied heavily on a combination of manual surveys and experience-based judgment. While this approach has some practicality, it suffers from significant shortcomings in terms of timeliness, accuracy, and comprehensiveness. On the one hand, due to the lack of continuous and high-density data support, managers often only obtain fragmented information over time, failing to grasp the overall situation of soil composition, structure, and its evolution trends. On the other hand, when faced with problems such as increased regional soil erosion, declining arable land fertility, and the spread of heavy metal pollution, traditional methods struggle to provide timely warnings and differentiated responses, leading to irrational resource allocation and even secondary ecological damage. Furthermore, at the implementation level, consolidation measures are often uniformly deployed and implemented in a haphazard manner, lacking the ability to customize adjustments based on the specific conditions of each plot. This significantly limits restoration efficiency and the level of sustainable development.

[0004] In recent years, with the development of remote sensing technology, geographic information systems, wireless sensor networks, and artificial intelligence algorithms, some studies have attempted to apply these emerging technologies to land resource monitoring and management, achieving some progress. However, existing systems, due to vague or overly simplified spatial division standards, fail to fully integrate topographic features and soil type distribution, leading to distorted data analysis results. Furthermore, most system models remain at the stage of static rule matching or shallow statistical analysis, making it difficult to rise to the level of making forward-looking judgments on future risk situations. This results in risk assessments and remediation recommendations that are often not accurate or personalized enough, weakening the adaptability and intelligence of the entire system. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides an intelligent land consolidation monitoring method and system based on the Internet of Things.

[0006] Firstly, this application provides an intelligent land consolidation monitoring method based on the Internet of Things, employing the following technical solution:

[0007] A smart land consolidation monitoring method based on the Internet of Things, the monitoring method comprising:

[0008] Acquire satellite remote sensing elevation data and historical soil composition data of the land area to be monitored;

[0009] Based on the satellite remote sensing elevation data and the preset landform segmentation rules, the land area to be monitored is divided into multiple grid units, and a unique grid identifier is assigned to each grid unit.

[0010] Based on the grid identifier, fixed soil sensor nodes and mobile drone nodes are deployed to generate a sensor network topology map;

[0011] Real-time soil parameters for each grid cell are collected through the sensor network.

[0012] The real-time soil parameters are subjected to noise filtering and anomaly detection to generate a soil composition dataset indexed by the grid identifier;

[0013] The soil composition dataset and the historical soil composition data are input into the cloud prediction model, which outputs the land degradation risk score and remediation strategy corresponding to each grid identifier.

[0014] The land remediation operation is carried out by controlling the execution equipment according to the remediation strategy, and the changes in soil parameters after execution are monitored in real time.

[0015] Based on the deviation between the changes in soil parameters after execution and the expected values, the parameters of the cloud-based prediction model are updated and the sensor network topology is optimized.

[0016] By adopting the above-mentioned technical solution, integrating satellite remote sensing, IoT sensor networks, and deep learning models, intelligent and precise monitoring and management of the land consolidation process is achieved. This significantly improves the real-time nature and coverage of soil data collection, and outputs personalized land degradation risk scores and consolidation strategies based on spatiotemporal prediction models, thereby driving automated equipment to perform precise interventions. In practical applications, the technical solution of this application promotes the sustainable use of land resources and ecological protection. Through early warning and dynamic optimization, it not only improves agricultural production efficiency and soil health but also reduces labor costs and resource waste, providing a scalable and adaptive solution for smart agriculture and land governance.

[0017] Secondly, this application provides an intelligent land consolidation monitoring system based on the Internet of Things, which adopts the following technical solution:

[0018] An Internet of Things-based intelligent land consolidation monitoring system, the monitoring system comprising:

[0019] The data acquisition module is used to acquire satellite remote sensing elevation data and historical soil composition data of the land area to be monitored;

[0020] The terrain segmentation module is used to divide the land area to be monitored into multiple grid units according to the satellite remote sensing elevation data and preset terrain segmentation rules, and assign a unique grid identifier to each grid unit.

[0021] The network deployment module is used to deploy fixed soil sensor nodes and mobile drone nodes based on the grid identifier, and generate a sensor network topology map.

[0022] The parameter acquisition module is used to acquire real-time soil parameters of each grid unit through the sensor network;

[0023] The data preprocessing module is used to perform noise filtering and anomaly detection on the real-time soil parameters and generate a soil composition dataset indexed by the grid identifier.

[0024] The risk prediction module is used to input the soil composition dataset and the historical soil composition data into the cloud prediction model and output the land degradation risk score and remediation strategy corresponding to each grid identifier.

[0025] The land remediation control module is used to control the execution equipment to perform land remediation operations according to the remediation strategy, and to monitor changes in soil parameters in real time after execution.

[0026] The model optimization module is used to update the parameters of the cloud prediction model and optimize the sensor network topology based on the deviation between the changes in soil parameters after execution and the expected values.

[0027] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0028] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0029] In summary, this application includes at least one of the following beneficial technical effects: multi-source data fusion enables collaborative monitoring of satellite remote sensing and ground sensors; gridded precise positioning ensures refined assessment of land degradation risks; real-time dynamic prediction can provide early warning of land degradation trends and offer personalized remediation strategies; and a closed-loop feedback optimization mechanism continuously improves the accuracy of prediction models and the deployment efficiency of sensor networks, ultimately achieving scientific, intelligent, and precise land remediation decisions, significantly improving the efficiency of land resource management and the effectiveness of ecological restoration. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the first process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0031] Figure 2This is a schematic diagram of the second process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0032] Figure 3 This is a schematic diagram of the third process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0033] Figure 4 This is a schematic diagram of the fourth process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0034] Figure 5 This is a schematic diagram of the fifth process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0035] Figure 6 This is a schematic diagram of the sixth process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0036] Figure 7 This is a schematic diagram of the seventh process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application.

[0037] Figure 8 This is a schematic diagram of the eighth process of an IoT-based smart land consolidation monitoring method according to one embodiment of this application. Detailed Implementation

[0038] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-8 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0039] This application discloses an intelligent land consolidation monitoring method based on the Internet of Things.

[0040] Reference Figure 1 A smart land consolidation monitoring method based on the Internet of Things, the monitoring method includes:

[0041] Step S101: Obtain satellite remote sensing elevation data and historical soil composition data of the land area to be monitored;

[0042] The system acquires geomorphic information about the target area, such as topographic relief and slope distribution, by utilizing high-resolution satellite imagery or LiDAR. It also integrates historical soil sample data accumulated over a long period, including but not limited to physicochemical indicators such as pH value, organic matter content, nitrogen, phosphorus, and potassium concentrations, and heavy metal residues. This foundational data forms the basis for subsequent spatial division and modeling analysis. For example, elevation data can help identify potential soil erosion risk areas, while historical soil composition is used to establish an ecological baseline within the region, facilitating the assessment of the current soil health.

[0043] Step S102: Based on satellite remote sensing elevation data and preset landform segmentation rules, the land area to be monitored is divided into multiple grid units, and a unique grid identifier is assigned to each grid unit.

[0044] This step employs an object-oriented spatial partitioning strategy. Specifically, by combining a slope threshold (each area greater than 15° is treated as a separate block) with the boundaries of different soil types, the entire land area is subdivided into several standardized rectangular grids (e.g., 100m × 100m), and each sub-region is assigned a unique "grid identifier." This identifier not only serves as a physical spatial location code but also bears the core indexing function connecting all subsequent data collection, transmission, processing, and application. It enables massive amounts of heterogeneous data to be organized according to a unified standard, forming a structured spatial database, greatly improving the system's scalability and query efficiency.

[0045] Step S103: Based on the grid identifier, deploy fixed soil sensor nodes and mobile UAV nodes to generate a sensor network topology map;

[0046] The system comprehensively considers the advantages of both static and dynamic data acquisition modes. Fixed nodes are typically deployed at geologically stable and easily maintained key locations to continuously monitor changes in the local microenvironment; while drones can flexibly adjust their flight trajectories based on historical degradation hotspot maps to carry out high-frequency, wide-coverage supplementary sampling tasks. The hybrid network architecture formed by the collaborative work of both effectively overcomes the blind spot problem existing in traditional single deployment methods.

[0047] Based on this, a complete sensor network topology is constructed by further utilizing GPS positioning technology and communication protocol stack. Each edge represents a communication link between nodes, and each node carries information about its geographical location and service range, ensuring that all subsequent measurement results can be accurately mapped back to the corresponding grid identifier.

[0048] Step S104: Collect real-time soil parameters for each grid cell through a sensor network;

[0049] Real-time soil parameters include at least humidity, pH value, and nitrogen, phosphorus, and potassium content;

[0050] Specifically, impedance hygrometers can be used for rapid response to moisture fluctuations, ion-selective electrodes (ISE) in conjunction with buffer solutions measure pH values, and near-infrared spectroscopy can non-destructively extract NPK nutrient concentrations. Due to the complex and variable conditions in the field, raw signals often contain noise interference; therefore, filtering algorithms must be introduced to improve the signal-to-noise ratio. Furthermore, considering that some extreme events may cause abrupt and abnormal readings, statistical methods must be introduced to screen and remove such readings, ensuring the quality of the final data stored is controllable.

[0051] Step S105: Perform noise filtering and anomaly detection on real-time soil parameters to generate a soil composition dataset indexed by grid identifiers.

[0052] This step primarily employs classic techniques from the field of digital signal processing, such as Kalman filters, moving averages, or wavelet denoising, to remove the error caused by random disturbances. For anomaly identification, a three-sigma criterion based on the normal distribution assumption is introduced. When a time-series observation deviates from the historical mean of the grid by more than three standard deviations, it is identified as an anomaly and added to a specially labeled queue for manual review or model retraining. The resulting standardized dataset not only enhances the learning stability of downstream machine learning models but also provides solid data support for the formulation of subsequent remediation strategies.

[0053] Step S106: Input the soil composition dataset and historical soil composition data into the cloud prediction model, and output the land degradation risk score and remediation strategy corresponding to each grid identifier;

[0054] This step employs an advanced deep learning framework, particularly a spatiotemporal convolutional neural network (CNN-LSTM) structure, to capture the trends and patterns of soil properties evolving over time and space. The model accepts both the latest measured data from each grid cell and historical archives as dual input sources. After multi-level feature extraction, it outputs a risk level assessment score for each grid cell and recommends corresponding intervention combinations, such as applying soil amendments, adjusting irrigation systems, or implementing crop rotation programs. It is noteworthy that this process does not simply apply a generic template but rather fully integrates local climate background, vegetation type, and other factors to customize personalized solutions.

[0055] Step S107: Control the execution equipment to carry out land remediation operations according to the remediation strategy, and monitor the changes in soil parameters after execution in real time;

[0056] Once the remediation strategy is determined, the central control system can issue instructions to field robotic arms, automatic spraying devices, or other specialized agricultural implements to complete a series of remediation actions. Simultaneously, portable monitoring instruments deployed in situ will continue to track the evolution of key indicators, paying particular attention to sensitive variables such as nitrogen migration pathways and salinization processes. For example, periodically measuring nitrate absorption rates using a UV-Vis spectrophotometer, supplemented by tracking electrolyte diffusion behavior with a four-electrode conductivity probe, can provide quantitative evaluation data to help determine whether the remediation efforts have achieved the expected goals.

[0057] Step S108: Based on the deviation between the changes in soil parameters after execution and the expected values, update the parameters of the cloud prediction model and optimize the sensor network topology.

[0058] This step demonstrates the system's self-evolution and continuous improvement. After each governance cycle, the system automatically compares the difference between the two samplings and uses this comparison to infer whether the existing model has cognitive biases or overfitting tendencies, thereby triggering an incremental retraining process to continuously improve prediction accuracy.

[0059] At the same time, for specific areas where deviations significantly exceed the allowable limits, the system will increase the frequency of drone inspections or add new fixed nodes, and redraw a more reasonable topology layout to enhance the robustness and sensitivity of the overall monitoring network.

[0060] The above embodiments integrate satellite remote sensing, IoT sensor networks, and deep learning models to achieve intelligent and precise monitoring and management of the land consolidation process. This significantly improves the real-time nature and coverage of soil data collection, and outputs personalized land degradation risk scores and consolidation strategies based on spatiotemporal prediction models, thereby driving automated equipment to perform precise interventions. In practical applications, the technical solution of this application promotes the sustainable use of land resources and ecological protection. Through early warning and dynamic optimization, it not only improves agricultural production efficiency and soil health but also reduces labor costs and resource waste, providing a scalable and adaptive solution for smart agriculture and land governance.

[0061] Reference Figure 2 As one implementation of step S102, the step of dividing the land area to be monitored into multiple grid cells based on satellite remote sensing elevation data and preset terrain segmentation rules, and assigning a unique grid identifier to each grid cell includes:

[0062] Step S201: Obtain satellite remote sensing elevation data and historical geomorphological feature database of the land area to be monitored; wherein, the historical geomorphological feature database includes soil type boundary coordinate set and slope distribution map;

[0063] Specifically, satellite remote sensing elevation data typically originates from global digital elevation model products such as ASTER GDEM and SRTM, providing the spatial distribution of surface elevation values ​​and serving as a key indicator reflecting topographic relief. The introduction of a historical geomorphological feature database endows the system with semantic awareness capabilities over time, including a "soil type boundary coordinate set" and a "slope distribution map." The former records the spatial boundaries of different geological properties (such as clay, sand, and loam), while the latter quantitatively describes the changing patterns of ground tilt.

[0064] The data fusion at this stage not only meets the basic requirements of spatial modeling but also provides rich criteria for subsequent adaptive gridding. For example, in some mountainous or hilly areas, due to frequent changes in soil structure accompanied by significant geomorphic gradient differences, traditional uniform grids are unable to accurately capture localized, subtle but crucial land condition changes. However, with the support of these multi-source heterogeneous historical data, the sensitivity of the gridding strategy to the complexity of the actual terrain can be effectively improved, thereby enhancing the spatial resolution and representativeness of the entire monitoring system.

[0065] Step S202: Based on the preset landform segmentation rules, calculate the dynamic size parameters of the grid cells according to the surface curvature values ​​in the satellite remote sensing elevation data and the historical landform feature database.

[0066] Specifically, the preset landform segmentation rules are as follows:

[0067] When the difference in slope values ​​between consecutive areas in the slope distribution map is less than 5°, they are merged into the same geomorphic zone.

[0068] Calculate the dynamic dimension parameters for each geomorphic zone using the following formula:

[0069] ;

[0070] In the above formula, S is the grid side length (unit: meters), C is the surface curvature value (unit: m⁻¹), and K is the zoning adjustment coefficient.

[0071] In some embodiments, the zoning adjustment coefficient K is generated by: extracting the soil type boundary coordinate set from the historical geomorphological feature database; if the grid crosses the boundaries of two or more soil types, K=80 is set; if the grid is entirely located in a single soil type area, K=100 is set.

[0072] Surface curvature, as an important topographic factor, reflects the degree of surface bending change. It can be obtained from DEM data through mathematical transformation, and common forms include mean curvature and Gaussian curvature. In this embodiment, the selected surface curvature C is defined as the ratio of the squares of the first derivatives of the rate of elevation change per unit area. Its physical meaning is the change in slope angle per unit length. Combined with the "preset landform segmentation rules," the system first performs landform zoning on the original remote sensing image. The judgment criterion mainly relies on whether the slope difference is less than the critical threshold of 5°. This is because geomorphological studies show that when the slope difference between two adjacent areas is small, it often means that they are in the same or similar landform units and have similar erosion / deposition behaviors. Therefore, classifying them into a unified landform zoning helps maintain consistency within the region and reduces redundant computational burden.

[0073] Subsequently, for each geomorphic zone, a specific formula was used to calculate the appropriate grid side length S for that region. Here, K is an adjustment coefficient influenced by soil type, and its value is set to fully consider the impact of factors such as permeability and shear strength of different soil and rock media on monitoring accuracy. The formula itself follows an inverse proportional function relationship, indicating that under the same geomorphic conditions, the gentler the surface (i.e., the smaller the absolute value of curvature), the larger the area required to cover a single grid; conversely, for areas with steeper and more complex terrain, the grid density should be reduced to ensure sufficient detail representation.

[0074] It should be noted that this method breaks through the limitations of traditional fixed-size grids, realizes true on-demand network deployment, and improves the spatial adaptability and resource utilization efficiency of the overall monitoring network.

[0075] Step S203: The land area is divided into multiple grid cells based on dynamic size parameters, and the vertex coordinates of each grid cell are generated by mapping from the geographic coordinate system.

[0076] In this context, a grid cell refers to a regular quadrilateral tessellation constructed on a two-dimensional projection plane. Although not strictly square, it maintains good regularity and ease of management within a certain range. Each grid cell is enclosed by four vertices, and its position is precisely marked by corresponding geographical coordinates (longitude and latitude).

[0077] To ensure the scientific validity of the mesh generation results, this application employs a progressive meshing algorithm driven by dynamic size parameters. This algorithm prioritizes areas with drastic terrain changes as starting points, then gradually expands outwards to more stable surrounding areas, thus forming a hierarchical and appropriately spaced overall layout. Simultaneously, considering the potential error accumulation caused by the Earth's curvature, a spherical distance correction mechanism is introduced, ensuring that the final rectangular mesh maintains high geometric fidelity even at a large scale. Throughout the process, special attention must be paid to boundary condition handling. For example, when encountering rivers, roads, or other linear obstacles, the mesh cannot be simply cut; instead, it needs to be readjusted using vector boundary constraints to avoid ineffective monitoring blind spots or duplicate sampling.

[0078] In addition, after generating the mesh cells, a mesh optimization verification step is also included: detecting elevation abrupt changes in adjacent mesh cells; if the abrupt change exceeds the threshold, a transition sub-mesh is inserted at the abrupt change boundary, and a derived identifier is assigned to the sub-mesh.

[0079] Step S204: Assign a unique mesh identifier to each mesh cell and bind the mesh identifier with vertex coordinates to the spatial index database.

[0080] The system generates a unique "grid identifier" for each grid cell. This identifier is not just a simple number label, but also carries a variety of implicit information such as geographical location and topological adjacency.

[0081] In the specific implementation process, "geohash" is preferred as the primary key identifier. This is an encoding technology widely used in the LBS service field. The basic idea is to compress latitude and longitude coordinates into a string, which preserves spatial proximity and facilitates fast retrieval and location. On this basis, some auxiliary fields such as creation timestamp and geomorphic zone ID are added to construct a complete identity profile. Next, these identifiers are associated and bound to their corresponding vertex coordinates and written into a specially designed "spatial index database". The so-called spatial index refers to a database index structure specifically optimized for geographic object queries, with typical examples including R-trees, quadtrees, and MBRs. They can significantly improve the retrieval speed of massive grid data, especially when facing typical business requests such as "finding all grids near a certain monitoring station" or "filtering areas with an altitude higher than a certain threshold", demonstrating extremely high execution efficiency.

[0082] Furthermore, this mechanism can also well support the future expansion of functional modules, such as connecting to external IoT devices and automatically issuing command tasks, all of which rely on the support of a stable and efficient identifier-coordinate mapping infrastructure.

[0083] The above implementation transforms the traditional static grid into a dynamic and adjustable one, while deeply integrating multiple natural factors such as topography and soil properties into the decision-making process, greatly improving monitoring accuracy and the effectiveness of resource allocation. Furthermore, by combining the use of surface curvature and slope with a flexible zoning adjustment coefficient mechanism, the problem of information distortion caused by neglecting detailed topographic differences is solved. The later addition of geohashing encoding and spatial index database construction fundamentally ensures the stability and scalability of the system, making it fully compatible with the growing demand for large-scale, high-frequency environmental sensing in fields such as smart agriculture and ecological protection.

[0084] Reference Figure 3 As one implementation of step S105, the step of performing noise filtering and anomaly detection on real-time soil parameters to generate a soil composition dataset indexed by grid identifiers includes:

[0085] Step S301: Receive a real-time soil parameter stream from the sensor network, where each data point in the parameter stream is bound to a corresponding grid identifier;

[0086] The real-time soil parameter stream refers to the collection of environmental variable data continuously uploaded by various sensors deployed in fields or areas to be treated, such as key indicators like soil moisture, pH, electrical conductivity, and organic matter content. "Binding to corresponding grid identifiers" means that each observation data point is accompanied by a geocoding information, i.e., a spatial location tag. This design allows the originally scattered and disordered data to be mapped to specific plot units according to preset spatial division rules, thus providing the prerequisite for subsequent spatial differentiation processing.

[0087] Step S302: Call the historical soil database to generate parameter reference baselines corresponding to each grid identifier;

[0088] The parameter reference baseline is generated as follows: extract parameter records from the historical soil database for the past 30 days under the same grid identifier; calculate the daily parameter mean and fit it to a time-series baseline function. , where α, β, and γ are parameter combinations obtained by independent fitting for different grids. This step aims to establish a baseline model for determining whether the current state deviates from the normal range.

[0089] Specifically, by statistically analyzing the historical records of the same grid over a period of time (e.g., the past 30 days), a trend curve reflecting the typical behavioral patterns of the region is extracted and abstracted into a mathematical expression. This function not only reflects the basic laws of soil property evolution over time but also takes into account the seasonal fluctuations in individual areas caused by natural factors (such as rainfall and fertilization). In this way, each grid is assigned a personalized criterion template, rather than using a uniform threshold to measure changes in all areas, greatly improving the system's sensitivity and generalization ability.

[0090] Step S303: The real-time soil parameters are filtered for noise using a composite filtering algorithm, and the denoised parameter sequence is output.

[0091] The composite filtering strategy comprises two levels of operation: First, a sliding window averaging filter, which sets different window widths based on different terrain types—10 sampling points for plains and 5 sampling points for hilly areas. The rationale is that environmental changes in flat areas are relatively slow, making longer-period smoothing methods suitable for removing high-frequency disturbances; while in hilly areas with greater undulations, more local details need to be preserved, thus the averaging window length should be shortened to avoid excessively blurring the true trend. Second, soft-threshold denoising in the wavelet transform domain. The Haar wavelet is chosen as the basis function due to its good local support and fast convergence performance, making it particularly suitable for preserving the edges of abrupt signal changes. After these two stages of collaborative filtering, the original parameter sequence is purified to state values ​​closer to their true physical meaning, laying a solid foundation for further anomaly identification.

[0092] Step S304: Based on the parameter reference baseline, perform anomaly detection on the denoised parameter sequence and mark abnormal data points and associated grid identifiers;

[0093] The system incorporates the statistical concept of Z-score, calculated using the following formula: , where P t Let B(t) represent the actual observed value at a certain moment, B(t) be the predicted value of the aforementioned constructed reference baseline, and σ represent the standard deviation of the corresponding historical samples in the grid. When the deviation δ exceeds the preset critical value of 3 (equivalent to exceeding the confidence interval by approximately 99.7%), the measurement can be considered to have a significant anomaly and should be marked.

[0094] It should be noted that the "mark" here is not just a Boolean flag, but also includes encapsulating the degree of deviation, the time of occurrence, and the grid number to which it belongs in the data packet metadata field, which facilitates later traceability and manual intervention decision-making. In addition, this process fully demonstrates the advantages of spatial indexing, because only within the same grid can the same distribution background and noise level be shared; otherwise, direct horizontal comparison would increase the probability of misjudgment.

[0095] Step S305: Aggregate normal data points and marked abnormal data points to generate a soil composition dataset indexed by grid identifiers.

[0096] The aggregation process is not simply a merger; it involves multiple operations such as filtering, sorting, and fusion. Values ​​confirmed to be normal are directly included in the final report. For flagged problematic items, a special calibration identifier is added, automatically triggering a remote command to request a laboratory-grade spectrophotometer for secondary verification. Once a new, accurate reading is obtained, it replaces the original erroneous data, completing a full feedback correction cycle. In this way, even if a few unreliable measurements occur initially, the overall data quality will not be affected, while simultaneously prompting the system to continuously optimize and upgrade itself.

[0097] In the above implementation, grid identifiers are introduced as the core anchor point throughout the entire process, achieving seamless integration from raw data collection to high-quality output. A personalized evaluation framework for individual plots is established using a spatial indexing mechanism. A two-stage filtering structure combining static and dynamic methods is adopted to address the challenges posed by complex terrain. A closed-loop control loop driven by software labeling for hardware retesting is set up, enhancing the reliability and engineering practicality of the results. This method not only meets the requirements for continuous soil condition monitoring in fields such as precision agriculture and ecological restoration, but also provides a highly scalable and robust general solution for environmental perception systems in the era of big data.

[0098] Reference Figure 4 As one implementation of step S305, after the step of aggregating normal data points and marked abnormal data points to generate a soil composition dataset indexed by grid identifiers, the method further includes:

[0099] Step S401: Based on the abnormal data points in the soil composition dataset, generate a soil parameter calibration instruction chain including the target grid identifier;

[0100] The system uses a pre-defined rule engine to analyze and classify anomaly markers, identifying specific anomaly types (such as excessive chemical elements or drastic changes in ion concentration), their location information (i.e., the corresponding geographic coordinate grid identifier), and their potential impact. The core logic of command chain generation then begins: first, the spatial mapping processing module converts abstract grid numbers into usable GPS waypoint sequences, transforming latitude and longitude information into a three-dimensional trajectory data format directly usable by the UAV navigation system. This conversion process considers not only the influence of terrain undulations but also factors such as flight safety altitude restrictions, ensuring the feasibility and safety of the path.

[0101] Next is the equipment selection mechanism. Based on the different types of soil parameter requirements, the system automatically matches the most suitable testing instrument for the current task. For example, for the determination of inorganic salts and trace elements, an ion chromatograph is preferred due to its excellent separation capability and high-sensitivity conductivity detection capability; while for conventional nutrients such as nitrogen, phosphorus, and potassium, spectrophotometry is used, utilizing the linear relationship between the absorption intensity of a substance at a specific wavelength and its concentration to complete quantitative analysis. The entire process also includes the application of path optimization algorithms, typically using classic graph theory methods such as Dijkstra's algorithm to calculate the optimal access order, so that multiple target points form a flight path with the minimum total distance and lowest energy consumption. In addition, external interference factors such as wind speed changes and obstacle distribution need to be comprehensively considered to further enhance the robustness of path planning.

[0102] Step S402: According to the soil parameter calibration instruction chain, the verification and testing equipment deployed on the mobile drone node is invoked to collect the verification data of the target grid identifier;

[0103] To ensure the scientific rigor and accuracy of the sampling, the mobile platform must possess highly integrated capabilities, including but not limited to hardware support systems such as autonomous navigation, multi-sensor synchronous control systems, and remote communication interfaces. Once the drone arrives at the designated location, the positioning module first confirms whether the current position meets the sampling conditions (e.g., an error not exceeding ±0.5 meters), and then activates the corresponding detection equipment according to the established protocol to begin operation.

[0104] For example, before operating the spectrophotometer, the system activates the accompanying LED light source array to provide a stable monochromatic illumination environment, which effectively eliminates measurement deviations caused by background noise, especially under uneven lighting or cloudy weather conditions. Before activating the ion chromatograph, its internal tubing undergoes an automatic cleaning procedure to prevent residual liquids from causing cross-contamination and affecting the accuracy of experimental results. Simultaneously, each piece of collected information is appended with complete metadata tags, covering dimensions such as timestamp, device number, and sampling depth, for later traceability and quality assessment.

[0105] More importantly, all actions are closely monitored by the embedded controller. If any non-compliant operation or signs of sensor malfunction are detected, the task will be immediately interrupted and an error code will be transmitted back via wireless network for technicians to troubleshoot and repair.

[0106] Step S403: The verification data is fused with the real-time soil parameters of the corresponding grid identifier using a confidence-weighted method.

[0107] The two sets of data obtained in the early stage came from different sources. One set came from continuous observation records provided by traditional fixed monitoring stations that have been deployed for a long time, while the other set came from first-hand measured data obtained by temporarily dispatching unmanned aerial vehicles carrying precision instruments to the site. There are bound to be differences between the two in terms of timeliness, representativeness and even accuracy.

[0108] Therefore, it is necessary to introduce a rigorous and reasonable fusion framework to integrate these two types of highly complementary but inherently uncertain information sources. Specifically, this application uses a weighted average model derived from Bayesian theory as the basic computational tool, expressed by the following formula:

[0109] ,

[0110] Where P f P represents the final estimated value after fusion. v This is validation data, P t It is a real-time monitoring value; W d and C s These represent the historical performance weighting coefficient and the current state confidence factor of the equipment, respectively. These two parameters are not static but are continuously and dynamically adjusted with changes in time and operating conditions. For example, for a particular model of spectrophotometer, if it has demonstrated high repeatability and stability in numerous past experiments (e.g., mean absolute error (MAE) less than 0.15), then a larger W can be assigned to it. d Values ​​(e.g., 0.9); conversely, if there are frequent errors or large fluctuations in readings recently, the weighting ratio should be appropriately reduced to reflect its unreliable risk tendency. As for C... s The value of relies more on the support of real-time feedback mechanisms. Auxiliary variables such as power supply voltage, temperature and humidity, and vibration frequency are taken into consideration, and the corresponding membership score is derived by using the pre-established empirical function curve.

[0111] It should be noted that when the conclusions from two independent sources differ significantly (exceeding the set tolerance range), an additional conflict resolution submodule needs to be introduced to intervene. This submodule re-estimates the most likely true state by correcting the prior probability distribution, thereby mitigating the risk of overall misjudgment caused by individual failed devices. The resulting synthetic dataset, after this series of refined processing steps, not only better approximates the objective reality but also maintains high stability and generalization ability, laying a solid data foundation for the next stage of model iteration.

[0112] Step S404: If the fusion deviation exceeds the preset deviation threshold, the real-time retraining mechanism of the cloud prediction model is triggered.

[0113] Specifically, when the fusion deviation exceeds the preset allowable upper limit (such as a pH change greater than 0.5 or a N content change exceeding 15%), it can be determined that the existing prediction model is no longer fully applicable to the current scenario and urgently needs to be adjusted and upgraded.

[0114] In this embodiment, to avoid the huge waste of resources caused by blindly restarting the entire training process, a more efficient and flexible incremental learning approach is adopted. Only the parameters of the most affected local network layers are fine-tuned, rather than the entire network is rebuilt. The usual practice is to freeze the shallow convolutional components responsible for image feature extraction, as these have already solidified a large number of general visual patterns, and easily modifying them can disrupt the original cognitive structure. The focus is on the fully connected layers related to high-level semantic understanding and reasoning, iterating repeatedly with a small number of fresh samples until convergence. During this process, the AdamW adaptive gradient optimization algorithm is used to suppress potential oscillations and divergences, and a strict stopping criterion is set: the update cycle ends only when the loss function on the validation set decreases by less than one-thousandth for three consecutive validation sets.

[0115] Step S405: Update the remediation strategy based on the retraining results and send parameter calibration instructions to the execution device.

[0116] The retrained prediction model will then generate a new risk rating for each key area of ​​concern. The system will then compare the latest evaluation results with an internally maintained multi-layered governance strategy database to select the most suitable response for the current situation.

[0117] For example, a target plot that was originally rated as Level II (recommended to increase the application of organic fertilizer) may now be upgraded to Level III (recommended to implement lime improvement combined with soil stabilization engineering). This means that the original simple fertilizer prescription is obviously not enough to solve the problem, and a more powerful comprehensive management plan must be adopted.

[0118] The next step is the specific dosage calculation, which involves deriving the required amounts of various conditioners based on a specific mathematical model. These conditioners include, but are not limited to, compound fertilizers, quicklime powder, and other functional additives. A buffer capacity equation can be used in this process. Where Q represents the required dosage of pesticide per unit area, k is the soil buffer coefficient (assigned according to different soil textures), and S represents the actual coverage area of ​​the area to be treated.

[0119] It is worth noting that the parameters selected here are based on the best-fit coefficients obtained through regression analysis of a large amount of historical archive data, thus possessing strong practical reference value. The final step is to package and encapsulate these customized instructions and then send them to various downstream terminal devices, such as intelligent fertilizer dispensers and automatic sprinkler systems. These devices will receive dedicated control commands, thereby achieving truly precise fertilization and water-saving irrigation functions.

[0120] In the above implementation, a soil parameter calibration instruction chain driven by abnormal data is constructed to achieve accurate identification and classification response to abnormal grids. Mobile drone nodes are used to call professional equipment such as spectrophotometers and conductivity sensors on demand for targeted retesting, improving sampling accuracy and spatial coverage flexibility. Through a credibility-weighted fusion mechanism, combining the historical performance weights of equipment and real-time state confidence factors, the verification data and the original monitoring data are dynamically fused to enhance data reliability and robustness. When the fusion deviation exceeds the threshold, the local incremental retraining mechanism of the cloud prediction model is triggered, which only fine-tunes the high-level semantic network, balancing efficiency and adaptability to ensure that the model continuously adapts to actual environmental changes. Finally, the remediation strategy is dynamically adjusted based on the updated prediction results, and refined parameter calibration instructions are issued to the execution terminal, improving the accuracy and adaptability of soil remediation.

[0121] Reference Figure 5 As one implementation of step S401, the step of generating a soil parameter calibration instruction chain including a target grid identifier based on outlier data points in the soil composition dataset includes:

[0122] Step S501: Extract the grid identifiers and abnormal parameter types of all abnormal data points;

[0123] For each abnormal data point containing a specific indicator name (such as pH, Cd content, etc.), the system introduces a parameter type classifier module. This module has a built-in parameter attribute dictionary, which is divided into chemical parameters (such as heavy metal ion concentration, pH, nutrient element content) and physical parameters (such as electrical conductivity, porosity, bulk density) according to the internationally accepted soil science classification system. This classification operation is not a simple string matching, but rather a logical judgment that comprehensively considers the parameter unit, detection method, and sensitivity to environmental changes.

[0124] For example, if a value is measured in mg / kg, it is generally categorized as a chemical measurement, as it typically originates from laboratory-level atomic absorption or ICP-MS measurements. However, if it is measured in dS / m, it is more likely to fall under the category of rapid on-site measurement, corresponding to the output signal of an electromagnetic induction sensor. This dual-analysis mechanism not only improves the accuracy of anomaly identification but also lays the foundation for subsequent differentiated resource allocation.

[0125] Step S502: Based on the type of abnormal parameters, assign spectrophotometer sampling paths to grid identifiers with abnormal chemical parameters, and assign conductivity sensor retest coordinates to grid identifiers with abnormal physical parameters.

[0126] Specifically, for grids identified as having abnormal chemical parameters, the system decides to invoke a spectrophotometer. As an analytical instrument based on a substance's ability to absorb light at specific wavelengths, the spectrophotometer is well-suited for the quantitative detection of various ion and compound concentrations (i.e., chemical parameters) in soil solutions. By assigning spectrophotometer sampling paths to these grids, the system ensures the scientific rigor and specificity of the verification methods, enabling the direct acquisition of accurate data on the key chemical indicators leading to the anomalies.

[0127] Correspondingly, for anomalies in physical parameters, such as conductivity anomalies which are closely related to soil salinity and total solute content, the system assigns a conductivity sensor. This is because conductivity sensors indirectly reflect the total salt content by measuring the conductivity of the soil solution, which is a direct and efficient method for characterizing physical properties.

[0128] It should be noted that allocating sampling paths and retest coordinates is not just about specifying a destination. It usually integrates path optimization algorithms (such as a variant of the Traveling Salesman Problem (TSP)). It takes into account factors such as the spatial distribution of all outliers, the drone's endurance, and terrain obstacles to calculate a flight route with the shortest total time and highest energy efficiency, thereby integrating multiple isolated retest tasks into an orderly and efficient serialized operation.

[0129] Step S503: Generate an instruction queue containing paths, coordinates, and verification detection devices based on the grid identifiers of all abnormal data points.

[0130] The generation of the instruction queue marks the system's switch from "analysis and decision-making" mode to "action control" mode. This queue is an ordered set, in which each instruction contains three key elements: path (the spatial trajectory connecting the various target grids), coordinates (the GPS location of the specific sampling point within each grid, usually considering a representative area within the grid), and verification and detection equipment (the specific sensor that needs to be activated at that grid point, such as a spectrophotometer or conductivity sensor).

[0131] Understandably, the structured format of the command queue allows the UAV flight control system to seamlessly parse and sequentially access each target grid. Upon reaching the designated coordinates, it automatically invokes and activates the corresponding detection equipment for data acquisition. This not only significantly improves the automation level of fieldwork and reduces errors and delays caused by human intervention, but more importantly, it ensures the spatial consistency and equipment professionalism of calibration data. Specifically, the retest data for each anomaly point originates from its original location and is obtained using the most suitable instrument, providing a high-quality, traceable data source for subsequent data fusion and model correction.

[0132] In the above implementation, spatial and attribute features are accurately extracted from abnormal data, achieving the best match between calibration tasks and professional testing equipment. Based on the optimization algorithm, an efficient command queue is generated, which ultimately drives the UAV platform to complete targeted on-site verification. This greatly improves the accuracy, timeliness, and resource utilization efficiency of data verification, and provides more reliable data support for subsequent decision-making.

[0133] Reference Figure 6 As one implementation of step S106, the steps of inputting the soil composition dataset and historical soil composition data into the cloud prediction model and outputting the land degradation risk score and remediation strategy corresponding to each grid identifier include:

[0134] Step S601: Receive soil composition dataset indexed by grid identifier and historical soil composition data;

[0135] Step S602: Load the pre-trained spatiotemporal fusion prediction model, and match the dimension of the input layer of the spatiotemporal fusion prediction model with the number of grid identifiers;

[0136] In this context, a pre-trained model means that the model has already undergone parameter optimization on a large amount of historical data in the relevant field, and can be directly applied to new task scenarios without retraining before formal deployment. Furthermore, during the construction of the neural network model, it is necessary to ensure that each input sample can be accurately mapped to the corresponding grid location information, thereby ensuring that the differentiated characteristics between different geographical locations are preserved and fully expressed within the model. For example, if 100 grid points are set up in the study area, the corresponding deep learning model should have an input vector length of at least 100 dimensions to accommodate the information combinations from all monitoring points.

[0137] Step S603: Time-series alignment of the soil composition dataset with records of the same grid identifier in the historical soil composition data;

[0138] This step aims to address time misalignment issues caused by inconsistent sampling frequencies or communication delays, ensuring that data from different times but belonging to the same grid cell are correctly arranged on a unified timeline for further processing. "Time alignment" is essentially a dynamic adjustment mechanism that eliminates deviations caused by external interference by performing interpolation, sliding window averaging, or nearest neighbor lookup on the original sequence.

[0139] Specifically, a set of the latest measured values ​​recently collected can be extracted for a specific grid, and all observation results of that location in the past period can be retrieved from the archive. Then, the data can be sorted in chronological order and missing segments can be filled in to form a complete multidimensional time series matrix for use by subsequent modules.

[0140] Step S604: Analyze the time-series aligned data using a spatiotemporal fusion prediction model and output the land degradation risk score corresponding to each grid identifier;

[0141] The spatiotemporal fusion prediction model refers to a composite deep learning architecture that comprehensively considers both spatial distribution patterns and temporal evolution trends. It can effectively capture the cross-scale interaction patterns among complex environmental variables. Typical implementations include, but are not limited to, a hybrid model structure combining convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). The former excels at capturing two-dimensional planar feature patterns like images, while the latter specializes in uncovering hidden state transition paths within one-dimensional signal chains. After sufficient training, this joint framework can automatically identify which types of soil change patterns are more likely to trigger potential ecological crisis events and provide a quantitative evaluation value between 0 and 100 to indicate the likelihood of severe deterioration in the region in the future.

[0142] In this embodiment of the application, the training method of the spatiotemporal fusion prediction model includes: extracting gridded data from a historical soil composition database for N consecutive years, where N is greater than or equal to 5; constructing a feature matrix with grid identifiers as the primary key and containing soil chemical properties (pH value, heavy metal content) and physical properties (porosity, electrical conductivity); using a convolutional neural network (CNN) to extract spatial features and a long short-term memory network (LSTM) to extract temporal features, and mapping the output to a risk score of 0-100 through a fully connected layer.

[0143] Step S605: Based on the land degradation risk score, match the remediation strategies in the preset strategy library to generate a binding relationship table between grid identifiers and remediation strategies.

[0144] The pre-defined strategy library is essentially a set of pre-defined governance strategies, each applicable to different types and levels of problem situations. To establish a reasonable and effective mapping between the two, it is necessary to develop a scientific and reasonable hierarchical judgment standard to guide the selection of actions.

[0145] Specifically, strategy levels can be divided according to the risk score range: score ≤30 matches strategy A (conventional irrigation); score 31-70 matches strategy B (increased application of organic fertilizer); score >70 matches strategy C (engineering reinforcement); and the optimal strategy ID corresponding to the level is assigned to each grid identifier.

[0146] In this embodiment, the risk levels are further divided into low-risk (≤30 points), medium-risk (31~70 points), and high-risk (>70 points) zones based on the score range. Each category has its own customized operation guidelines and technical specifications. After the risk rating of all grid cells is completed, the most suitable remediation scheme ID number can be assigned to each cell according to its level. Finally, a complete and detailed map-level strategy comparison table is compiled for relevant departments to refer to and implement.

[0147] In addition, after generating the binding relationship table, policy conflict resolution can be performed: detect the policy ID difference between adjacent grid identifiers; if the policy level difference between adjacent grids is ≥2 levels, then insert a transition policy in the boundary area, generate a derived policy ID and update the binding relationship table.

[0148] In the above implementation, real-time soil composition datasets and historical data are time-series aligned using grid identifiers and input into a pre-trained spatiotemporal fusion prediction model. This model accurately outputs land degradation risk scores for each grid unit. Combined with grading criteria and matching remediation measures from a pre-defined strategy library, a grid-level governance strategy binding relationship table is generated. This achieves fully automated and spatially refined management of the entire process from data collection and intelligent analysis to decision support. This technical solution effectively improves the accuracy and timeliness of land degradation early warning and enhances the model's ability to capture complex spatiotemporal evolution characteristics.

[0149] Reference Figure 7 As a further implementation of the intelligent land consolidation monitoring method, after the step of generating a binding table between grid identifiers and consolidation strategies, the method further includes:

[0150] Step S701: Calculate the difference in remediation strategy levels between adjacent grid cells based on the grid identifier;

[0151] The system pre-maps the governance strategy associated with each grid identifier to an ordered numerical level based on its intervention intensity, resource input, or expected goals. Then, based on the spatial adjacency relationships of the grids (such as grid pairs sharing a boundary), the system calculates the absolute difference in strategy levels between each pair of adjacent grids, i.e., the governance strategy level difference. This process reveals the inconsistencies of governance schemes at the micro-scale.

[0152] For example, a grid assessed as requiring high-intensity fertilization (Level 1) is adjacent to a grid requiring only monitoring (Level 3), resulting in a policy level difference of 2. The purpose of extracting this level difference is to identify boundary areas where abrupt changes in governance measures could lead to conflicting effects, resource waste, or secondary ecological problems. This signifies an evolution in the system's decision-making logic from independent assessment of individual grids to consideration of inter-grid interactions and overall regional coordination.

[0153] Step S702: When the difference between the remediation strategy levels of any two adjacent grid cells is detected to be greater than or equal to the preset level threshold, a transition strategy is generated based on the difference between the remediation strategy levels and the soil type boundary coordinate set, and a derived strategy identifier is assigned to the transition strategy.

[0154] Specifically, when the difference in strategy levels between adjacent grids exceeds a preset level threshold (e.g., level difference ≥ 2), the system determines that there is a significant governance intensity fault at this point. Directly applying two completely different strategies may cause drastic changes in soil physicochemical properties over a short distance, which is detrimental to ecological stability.

[0155] At this point, the system does not simply select one strategy, but generates a transition strategy. Its logical principle is to combine two key pieces of information for interpolation and optimization: first, the quantified difference in remediation strategy levels, which determines the intensity range and necessity of the transition; and second, the set of soil type boundary coordinates, which provides objective constraints on natural geographical attributes. The system analyzes the transition characteristics of soil types in this boundary area (e.g., a gradual transition zone from clay to sandy loam), and based on this, designs a compromise or gradual transition scheme between the mainstream strategies of two adjacent grids.

[0156] For example, a transitional strategy of "moderate fertilization" can be generated between grids that require heavy fertilization and those that do not require fertilization, and the fertilization formula may take into account the characteristics of both soil types in terms of the fertilization formula. At the same time, the system assigns a unique derived strategy identifier to this newly created strategy, which ensures that the strategy is an independent entity that can be identified, managed and traced by the system, so that it can be accurately invoked and executed in subsequent processes.

[0157] Step S703: Associate the derived strategy identifier with the boundary region of the adjacent grid cell and update the binding relationship table;

[0158] The system needs to associate the newly generated derived policy identifier with a specific spatial entity, namely the boundary region (usually a strip of a certain width defined along its common boundary) of two adjacent grid cells with significantly different policies. This does not modify the original grid's policy, but rather "inserts" a unique policy into this special boundary area. Next, the system updates the binding relationship table, recording the new mapping relationship between "boundary region ID and derived policy identifier".

[0159] Understandably, this step adds a more refined layer to the system's "governance map," enabling the execution equipment to not only identify the main strategy of each grid during operation, but also to identify and execute special transition commands for these boundary areas. This demonstrates the spatial semantic enhancement capabilities of a data-driven governance system.

[0160] Step S704: Based on the updated binding relationship table, control the execution device to perform the land reclamation operation corresponding to the transition strategy in the boundary area;

[0161] In this system, when automated execution equipment (such as variable-rate fertilizer applicators and precision irrigation equipment) operates according to a planned path, its positioning system (such as GPS or BeiDou) determines its location in real time. The control system continuously queries the binding relationship table to determine not only which grid the equipment has entered, but more importantly, whether it has entered a registered boundary area. Once entered, the system immediately switches control commands, changing from executing the mainstream strategy for that grid to executing a transition strategy associated with that boundary area. For example, when a variable-rate fertilizer applicator travels to the boundary transition zone between two grids, the fertilizer application rate will smoothly decrease from a high value on one side to a low value on the other side according to the set transition strategy, rather than experiencing a precipitous change at the boundary line. This achieves a natural and gradual change in the treatment measures in space, improving the overall integrity and eco-friendliness of the remediation project.

[0162] Step S705: Monitor the changes in soil parameters in real time after implementing the transition strategy and calculate the expected deviation value;

[0163] After the transition strategy is implemented, a sensor network (fixed nodes or mobile drones) deployed in the boundary area and adjacent areas will activate an enhanced monitoring mode to monitor changes in key soil parameters in real time (e.g., monitoring spatial gradient changes in nitrogen and phosphorus content after implementing the transition fertilization strategy). The system compares the monitored actual parameter changes with the expected effect model set before the implementation of the transition strategy (e.g., the expected soil pH value should show a linear transition within the boundary area). The process of calculating the expected deviation value is the process of quantifying the difference between the actual effect and the expected target. This deviation value becomes the core indicator for evaluating whether the transition strategy design is reasonable and whether its implementation is accurate, and it also provides crucial data feedback for subsequent system learning and optimization.

[0164] Step S706: Trigger the incremental training process of the cloud prediction model based on the expected deviation value to update the model weight parameters;

[0165] Specifically, feedback data from border area governance practices is used to fine-tune and optimize the core prediction model. When the expected deviation value indicates a significant discrepancy between the actual effect of the transition strategy and the expected outcome, the system determines that the existing cloud-based prediction model's ability to predict such "strategy abrupt boundary" scenarios needs improvement. Instead of initiating a time-consuming full model retraining, the system triggers an incremental training process.

[0166] Specifically, this incremental training process uses the full-cycle data of this border area governance project (including the initial state, the implemented transition strategies, and the monitored soil responses) as a new, highly targeted training sample, inputting it into the model for learning. The training focuses on updating the parts of the model's weight parameters related to spatial relationships and strategy interaction effects, enabling the model to more accurately predict reasonable transition strategies and their effects when encountering similar scenarios in the future. This is equivalent to allowing the model to "remember" the lessons learned from this border governance project, improving the granularity and accuracy of its spatial decision-making.

[0167] Step S707: Based on the output of the incrementally trained model, optimize the node deployment density in the sensor network topology diagram, and increase the inspection frequency of mobile drone nodes in areas where the deviation value exceeds the deviation threshold.

[0168] Based on the simulation and output of the new model, the system can reassess the focus of attention in the entire monitoring area. The underlying principle is to dynamically reallocate sensing resources to match the new cognitive state.

[0169] On the one hand, the system may optimize the node deployment density in the sensor network topology. For example, if the model learns that certain soil boundary areas generally require more refined monitoring, the system may strategically increase the node deployment density in similar areas during future network planning. On the other hand, for specific areas where the deviation value exceeds the deviation threshold in this remediation effort, the system will mark them as "high uncertainty areas" or "key monitoring areas." As an immediate response, the system will instruct the frequency of mobile drone node inspections to enhance the control over the evolution trend of soil parameters in the area through dynamic data acquisition with higher spatiotemporal resolution, providing richer data for the next round of decision-making, thereby continuously improving the adaptability and robustness of the entire monitoring-remediation system.

[0170] In the above implementation, the differences in the levels of remediation strategies between grids are quantitatively analyzed, boundary areas where abrupt changes in remediation may occur are proactively identified, and gradual transition strategies are innovatively generated and implemented, effectively avoiding ecological stress or effect offsetting problems caused by discontinuous measures. By incorporating the transition strategies and their implementation effects into the system's learning loop, the predictive model is driven to undergo targeted incremental training, enabling the model's decision-making capabilities to continuously adapt to complex field conditions. Ultimately, the system can also dynamically optimize the configuration of the sensing network based on learning feedback, prioritizing monitoring resources for areas with large effect deviations or high uncertainties. This technical solution significantly improves the precision, coordination, and adaptability of the intelligent land remediation system in spatial governance, moving from pursuing the optimality of individual grids to achieving overall regional synergistic optimality, and enhancing the ecological rationality and long-term sustainability of land engineering measures in complex environments.

[0171] Reference Figure 8 As one implementation of step S108, the steps of updating the parameters of the cloud prediction model and optimizing the sensor network topology based on the deviation between the changes in soil parameters after execution and the expected values ​​include:

[0172] Step S801: Obtain soil parameter monitoring data and corresponding expected strategy effect values ​​after performing land consolidation operations. The data is indexed by grid identifiers.

[0173] The system requires collecting key physicochemical parameters reflecting the current soil condition from various environmental sensors deployed in the field or the area to be treated. These parameters typically include, but are not limited to, representative indicators such as pH value, nitrogen, phosphorus and potassium content, organic matter level, and moisture content, which together constitute the basic dimensions for assessing the effectiveness of the remediation.

[0174] Meanwhile, the system's front-end planning module outputs an idealized "strategy expected effect value" based on the previously formulated land consolidation plan. This value reflects the ideal state to be achieved after a period of intervention. To facilitate subsequent data comparison and spatial positioning analysis, all collected actual monitoring data must be indexed and labeled according to pre-divided grid units, forming a spatiotemporal data sequence with geographic coordinates. For example, a specific plot of land is divided into multiple square grids, each assigned a unique identifier from Grid_ID_001 to Grid_ID_nnn, and storing the average or weighted average results measured by various sensors within its coverage area. This structured organization not only improves retrieval efficiency but also lays the foundation for the next step of spatial differential deviation calculation.

[0175] Step S802: Calculate the deviation between the soil parameter monitoring data of each grid identifier and the expected effect value of the strategy, and generate a deviation dataset;

[0176] In this step, the system does not simply use a single-variable comparison method, but constructs a multi-dimensional difference vector to comprehensively depict the collaborative change trend between different attributes.

[0177] Specifically, for each grid identifier Grid_ID, the system will extract all available measured data items and corresponding target reference values ​​under that grid, and construct a set of feature vectors composed of several sub-deviations, such as pH offset ΔpH=pH_measured-pH_target, nitrogen content growth rate ΔN%=(N_current-N_initial) / N_initial, etc.

[0178] Subsequently, using the weighted Euclidean distance formula The individual deviations mentioned above are normalized and combined into a unique scalar form, namely the combined deviation value D. Where x i For monitoring values, y i For the expected value, w i w is the property weighting coefficient. i The degree of influence of key ecological factors can be determined based on expert experience or principal component analysis. In this way, even if some minor parameters fluctuate slightly, the main contradictions will not be obscured, thus ensuring that the final deviation set accurately reflects the true state of the local remediation quality.

[0179] Step S803: Input the bias dataset into the parameter update module of the cloud prediction model and iteratively adjust the model weight parameters;

[0180] The cloud-based prediction model referred to in this step is often a hybrid AI component, possibly including a convolutional neural network (CNN) to capture spatial distribution patterns, while also incorporating a long short-term memory network (LSTM) to model temporal evolution. In the early stages of training, the original model has been pre-trained with a large number of historical samples and possesses basic generalization capabilities; however, as on-site conditions continuously change, the original knowledge system may become outdated or even ineffective. Therefore, it is necessary to periodically expose it to the latest error signals for fine-tuning and correction.

[0181] To this end, the system first randomly splits the complete bias dataset into a training set (TrainSet) and a validation set (ValSet) in a 7:3 ratio. The former guides the gradient descent direction and updates the connection strength of each neuron layer by layer using the backpropagation algorithm (BackpropagationAlgorithm), while the latter monitors whether the validation loss (Loss) tends to stabilize to avoid falling into the overfitting trap. Throughout the iteration process, the parameters of the filter kernel matrix in the CNN part and the activation functions of the Forget Gate, Input Gate, and Output Gate inside the LSTM will undergo repeated calibration until the predetermined convergence criterion is met. This process essentially teaches the model to automatically identify which factors lead to large prediction errors and improve its accuracy in judging similar future scenarios accordingly.

[0182] Step S804: Optimize the node deployment configuration in the sensor network topology diagram based on the location information of the grid identifiers whose deviation values ​​exceed the preset deviation threshold;

[0183] Specifically, when the overall deviation value corresponding to a certain Grid_ID exceeds the warning line, it indicates that there may be unforeseen problems in that area or key areas that were not adequately covered by the original design. In this case, the system will not blindly increase the global density, but will instead adopt a precise targeting strategy, focusing attention on the vicinity of those locations where problems actually exist.

[0184] Specifically, high-risk zones are delineated, for example, by extending a certain radius R (e.g., 50 meters) outward from the center of the over-limit grid, and designated as the Zone of Interest. Then, mobile detection devices with mobility are deployed in this area to enhance local perception resolution. Conversely, for low-activity areas that show normal operation three times in a row without significant disturbance, redundant fixed sensor nodes are removed to save resources.

[0185] In addition, the existing communication routing table needs to be modified simultaneously to ensure that newly added devices can successfully connect and establish a reliable link with the central server. This topology adjustment is not a one-time action, but rather a cyclical process that occurs with each new deviation statistical cycle, gradually approaching the optimal resource allocation pattern.

[0186] Step S805: Output the updated cloud prediction model parameter set and the optimized sensor network topology.

[0187] In particular, due to the severe limitations of power supply in field operations, frequent battery replacements are clearly impractical, uneconomical, and environmentally unfriendly. Therefore, it is necessary to introduce an energy consumption balancing and scheduling mechanism to extend the overall lifespan of batteries.

[0188] Specifically, after each round of topology reconfiguration, the system re-estimates the additional power consumption caused by the addition of mobile nodes and adds it to the overall network ledger. Once the total amount approaches the critical limit, the energy-saving plan is immediately activated, prioritizing the continued operation of key monitoring points with larger deviations and higher potential value, while other relatively less important low-priority nodes are decisively shut down. To ensure consistency, all decisions are translated into specific start / stop commands and sent to the terminal controller, while simultaneously updating the status flags of relevant records in the database for future queries.

[0189] In the above implementation, the multidimensional deviations between the measured soil parameters and the expected results of each grid after land consolidation are collected to construct a weighted comprehensive deviation dataset, which drives the cloud prediction model to perform online parameter iteration updates, effectively improving the model's adaptability to dynamic environmental changes and prediction accuracy. At the same time, based on the spatial distribution characteristics of the grids with deviations exceeding the limit, the sensor network topology is intelligently optimized to achieve coordinated management of enhanced monitoring density in high-risk areas and dynamic resource recovery in low-activity areas. This not only improves the spatiotemporal resolution and response capability of the sensing system, but also extends the overall network lifecycle through the node scheduling mechanism of energy consumption sensing.

[0190] This application also discloses an intelligent land consolidation monitoring system based on the Internet of Things.

[0191] An IoT-based intelligent land consolidation monitoring system, the monitoring system comprising:

[0192] The data acquisition module is used to acquire satellite remote sensing elevation data and historical soil composition data of the land area to be monitored;

[0193] The terrain segmentation module is used to divide the land area to be monitored into multiple grid units based on satellite remote sensing elevation data and preset terrain segmentation rules, and assign a unique grid identifier to each grid unit.

[0194] The network deployment module is used to deploy fixed soil sensor nodes and mobile drone nodes based on grid identifiers, and generate a sensor network topology map.

[0195] The parameter acquisition module is used to collect real-time soil parameters of each grid cell through a sensor network.

[0196] The data preprocessing module is used to filter noise and detect anomalies in real-time soil parameters, and generate a soil composition dataset indexed by grid identifiers.

[0197] The risk prediction module is used to input soil composition datasets and historical soil composition data into the cloud prediction model and output land degradation risk scores and remediation strategies corresponding to each grid identifier.

[0198] The land remediation control module is used to control the execution equipment to carry out land remediation operations according to the remediation strategy, and to monitor changes in soil parameters in real time after execution.

[0199] The model optimization module is used to update the parameters of the cloud-based prediction model and optimize the sensor network topology based on the deviation between the changes in soil parameters after execution and the expected values.

[0200] The IoT-based intelligent land consolidation monitoring system of this application embodiment can implement any of the above methods, and the specific working process of each module in the system can refer to the corresponding process in the above method embodiments.

[0201] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0202] This application also discloses a computer-readable storage medium.

[0203] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the IoT-based smart land remediation monitoring methods.

[0204] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0205] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A smart land consolidation monitoring method based on the Internet of Things, characterized in that, The monitoring method includes: Acquire satellite remote sensing elevation data and historical soil composition data of the land area to be monitored; Based on the satellite remote sensing elevation data and the preset landform segmentation rules, the land area to be monitored is divided into multiple grid units, and a unique grid identifier is assigned to each grid unit. Based on the grid identifier, fixed soil sensor nodes and mobile drone nodes are deployed to generate a sensor network topology map; Real-time soil parameters for each grid cell are collected through the sensor network. Receive a real-time soil parameter stream from the sensor network, wherein each data point in the parameter stream is bound to a corresponding grid identifier; The historical soil database is used to generate parameter reference baselines corresponding to each grid identifier; The real-time soil parameters are noise-filtered using a composite filtering algorithm, and the denoised parameter sequence is output. Based on the parameter reference baseline, anomaly detection is performed on the denoised parameter sequence to mark abnormal data points and associated grid identifiers; Aggregate normal data points and labeled abnormal data points to generate a soil composition dataset indexed by grid identifiers; The soil composition dataset and the historical soil composition data are input into the cloud prediction model, which outputs the land degradation risk score and remediation strategy corresponding to each grid identifier. The land remediation operation is carried out by controlling the execution equipment according to the remediation strategy, and the changes in soil parameters after execution are monitored in real time. Based on the deviation between the changes in soil parameters after execution and the expected values, the parameters of the cloud prediction model are updated and the sensor network topology is optimized. After the step of aggregating normal data points and labeled outlier data points to generate a soil composition dataset indexed by grid identifiers, the following steps are also included: Based on the abnormal data points in the soil composition dataset, a soil parameter calibration instruction chain including a target grid identifier is generated; The verification and testing equipment deployed on the mobile drone node is invoked according to the soil parameter calibration instruction chain to collect verification data of the target grid identifier; The verification data is fused with the real-time soil parameters of the corresponding grid identifier using a confidence-weighted fusion. The confidence-weighted fusion adopts a weighted average model based on Bayesian theory extension, which introduces the historical performance weight coefficient of the equipment and the current state confidence factor. The historical performance weight coefficient of the equipment and the current state confidence factor are dynamically adjusted with time and operating conditions. If the fusion deviation exceeds a preset deviation threshold, the real-time retraining mechanism of the cloud prediction model is triggered; wherein, the real-time retraining adopts an incremental learning method, and only fine-tunes the parameters of the local network layers that are most affected. The remediation strategy is updated based on the retraining results, and parameter calibration instructions are sent to the execution equipment.

2. The intelligent land consolidation monitoring method based on the Internet of Things according to claim 1, characterized in that, Based on the satellite remote sensing elevation data and preset terrain segmentation rules, the step of dividing the land area to be monitored into multiple grid cells and assigning a unique grid identifier to each grid cell includes: Acquire satellite remote sensing elevation data and historical geomorphological feature database of the land area to be monitored; wherein, the historical geomorphological feature database includes soil type boundary coordinate set and slope distribution map; Based on the preset landform segmentation rules, the dynamic size parameters of the grid cells are calculated according to the surface curvature values ​​in the satellite remote sensing elevation data and the historical landform feature database. Based on the dynamic size parameters, the land area is divided into multiple grid cells, and the vertex coordinates of each grid cell are generated by mapping from the geographic coordinate system. Each grid cell is assigned a unique grid identifier, and the grid identifier is bound to the vertex coordinates and stored in the spatial index database.

3. The intelligent land consolidation monitoring method based on the Internet of Things according to claim 1, characterized in that, The step of generating a soil parameter calibration instruction chain, including a target grid identifier, based on outlier data points in the soil composition dataset includes: Extract the grid identifiers and anomaly parameter types for all abnormal data points; Based on the abnormal parameter type, a spectrophotometer sampling path is assigned to the grid identifier with abnormal chemical parameters, and a conductivity sensor retest coordinate is assigned to the grid identifier with abnormal physical parameters. Based on the grid identifiers of all abnormal data points, generate an instruction queue containing paths, coordinates, and verification detection equipment.

4. The intelligent land consolidation monitoring method based on the Internet of Things according to claim 2, characterized in that, The steps of inputting the soil composition dataset and historical soil composition data into the cloud-based prediction model and outputting the land degradation risk score and remediation strategy corresponding to each grid identifier include: Receive soil composition datasets and historical soil composition data indexed by the grid identifier; Load a pre-trained spatiotemporal fusion prediction model, wherein the dimension of the input layer of the spatiotemporal fusion prediction model matches the number of grid identifiers; The soil composition dataset is time-series aligned with records of the same grid identifier in historical soil composition data; The spatiotemporal fusion prediction model is used to analyze the time-series aligned data and output the land degradation risk score corresponding to each grid identifier. Based on the land degradation risk score, a table is generated that matches the remediation strategies in the preset strategy library to generate a binding relationship table between grid identifiers and remediation strategies.

5. The intelligent land consolidation monitoring method based on the Internet of Things according to claim 4, characterized in that, Following the step of generating the binding table between grid identifiers and remediation strategies, the following steps are also included: Calculate the difference in remediation strategy levels between adjacent grid cells based on the grid identifier; When the difference in the remediation strategy level between any two adjacent grid cells is detected to be greater than or equal to a preset level threshold, a transition strategy is generated based on the difference in the remediation strategy level and the soil type boundary coordinate set, and a derived strategy identifier is assigned to the transition strategy. Associate the derived strategy identifier with the boundary region of the adjacent grid cell and update the binding relationship table; Based on the updated binding relationship table, the execution device is controlled to perform the land reclamation operation corresponding to the transition strategy in the boundary area; Real-time monitoring of soil parameter changes after implementing the transition strategy, and calculation of expected deviation values; The incremental training process of the cloud-based prediction model is triggered based on the expected deviation value to update the model weight parameters; Based on the model output after incremental training, the node deployment density in the sensor network topology is optimized, and the inspection frequency of mobile drone nodes is increased in areas where the deviation value exceeds the deviation threshold.

6. The intelligent land consolidation monitoring method based on the Internet of Things according to claim 1, characterized in that, The steps of updating the parameters of the cloud-based prediction model and optimizing the sensor network topology based on the deviation between the changes in soil parameters after execution and the expected values ​​include: Acquire soil parameter monitoring data and corresponding expected strategy effect values ​​after land consolidation operations are performed, with the data indexed by grid identifiers; Calculate the deviation between the soil parameter monitoring data of each grid identifier and the expected effect value of the strategy, and generate a deviation dataset; The deviation dataset is input into the parameter update module of the cloud prediction model to iteratively adjust the model weight parameters; Based on the location information of grid identifiers whose deviation values ​​exceed a preset deviation threshold, optimize the node deployment configuration in the sensor network topology. Output the updated cloud-based prediction model parameter set and the optimized sensor network topology.

7. An intelligent land consolidation monitoring system based on the Internet of Things, characterized in that, For implementing the IoT-based intelligent land consolidation monitoring method according to any one of claims 1 to 6, the monitoring system comprises: The data acquisition module is used to acquire satellite remote sensing elevation data and historical soil composition data of the land area to be monitored; The terrain segmentation module is used to divide the land area to be monitored into multiple grid units according to the satellite remote sensing elevation data and preset terrain segmentation rules, and assign a unique grid identifier to each grid unit. The network deployment module is used to deploy fixed soil sensor nodes and mobile drone nodes based on the grid identifier, and generate a sensor network topology map. The parameter acquisition module is used to acquire real-time soil parameters of each grid unit through the sensor network; The data preprocessing module is used to perform noise filtering and anomaly detection on the real-time soil parameters and generate a soil composition dataset indexed by the grid identifier. The risk prediction module is used to input the soil composition dataset and the historical soil composition data into the cloud prediction model and output the land degradation risk score and remediation strategy corresponding to each grid identifier. The land remediation control module is used to control the execution equipment to perform land remediation operations according to the remediation strategy, and to monitor changes in soil parameters in real time after execution. The model optimization module is used to update the parameters of the cloud prediction model and optimize the sensor network topology based on the deviation between the changes in soil parameters after execution and the expected values.

8. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 6.

Images