Intelligent surveying and mapping data analysis and management method and system based on Internet of Things

Abstract

The invention relates to the technical field of land surveying and mapping, in particular to a surveying and mapping data intelligent analysis and management method and system based on the Internet of Things. The method comprises the following steps: collecting multi-source surveying and mapping data of a target area through the Internet of Things, and carrying out area perception fusion to obtain geological survey original fusion data; carrying out space-semantic-attribute surveying and mapping feature space reconstruction on the geodesic survey original fusion data to obtain a land semantic unit set; performing significant land element chain right confirmation on each space object in the land semantic unit set to obtain a mapping chain identification block; extracting land element association features according to the identification blocks on the surveying and mapping chain to obtain a land map embedded vector; and acquiring real-time target area multi-source sensing data, and performing land surveying and mapping task intelligent scheduling on the land map embedded vector by using the real-time target area multi-source sensing data so as to obtain a land surveying and mapping instruction network. According to the invention, the response speed, the scheduling efficiency and the task execution quality of land surveying and mapping can be improved.

Description

Technical Field

[0001] The present invention relates to the field of land surveying and mapping technology, and in particular to an Internet of Things-based surveying and mapping data intelligent analysis and management method and system. Background Art

[0002] As technologies such as Geographic Information Systems (GIS), Remote Sensing (RS), and Global Navigation Satellite Systems (GNSS) continue to mature, the means of acquiring surveying and mapping data are becoming increasingly diverse, the data dimensions are constantly expanding, and the data volume is also growing exponentially. Traditional surveying and mapping operations primarily rely on manual inspections, field measurements, and static data processing. These issues include long data acquisition cycles, poor real-time performance, high labor costs, and low information fusion efficiency. These issues make it difficult to meet the current demand for intelligent analysis and efficient management of surveying and mapping data in high-frequency, high-precision, and dynamically changing environments.

[0003] The rapid development of the Internet of Things (IoT) provides a new technological path for the intelligent collection and dynamic perception of surveying and mapping data. By deploying various sensor nodes at surveying and mapping sites or geospatial target areas and integrating them with cloud computing platforms and edge computing nodes, real-time acquisition and remote perception of multi-source data, including spatial geographic information, environmental parameters, and equipment operating status, can be achieved. However, despite the increasing adoption of IoT technology in surveying and mapping, existing land surveying and management methods still have many limitations.

[0004] Traditional surveying and mapping data management methods often rely on distributed storage and static database management, lacking the ability to intelligently integrate and correlate dynamic, heterogeneous, and spatiotemporally coupled data. This makes it difficult to generate systematic knowledge output from massive amounts of surveying and mapping data after collection. Furthermore, the process of accessing heterogeneous data from multiple sensor sources presents significant challenges, including inconsistent data formats, difficulty in time synchronization, and missing anomalous data. These issues severely restrict the overall utilization efficiency and processing quality of surveying and mapping data. Furthermore, traditional methods often rely on fixed rules or manual experience for data cleaning and modeling, lacking adaptive learning and semantic understanding capabilities, making it difficult to flexibly adjust and optimize accuracy for different surveying and mapping scenarios. Summary of the Invention

[0005] Based on this, it is necessary for the present invention to provide a method and system for intelligent analysis and management of surveying and mapping data based on the Internet of Things to solve at least one of the above technical problems.

[0006] To achieve the above objectives, a method for intelligent analysis and management of surveying and mapping data based on the Internet of Things includes the following steps:

[0007] Step S1: Collect multi-source surveying and mapping data of the target area through the Internet of Things, and perform regional perception fusion based on the multi-source surveying and mapping data of the target area to obtain the original fusion data of the ground survey;

[0008] Step S2: Perform multimodal semantic deconstruction on the original fused land survey data to obtain a modality-aligned semantic vector; perform spatial-semantic-attribute mapping feature space reconstruction based on the modality-aligned semantic vector to obtain a land semantic unit set;

[0009] Step S3: Concentrate the land semantic units on each spatial object and perform blockchain data rights confirmation, thereby constructing a surveying and mapping data index chain; perform on-chain rights confirmation of significant land elements on the surveying and mapping data index chain to obtain an identification block on the surveying and mapping chain;

[0010] Step S4: construct a geographic element topology map based on the identification blocks on the surveying and mapping chain, and use the land element topology map to extract the land element correlation features to obtain the land map embedding vector;

[0011] Step S5: Acquire real-time multi-source sensor data of the target area, and use the real-time multi-source data of the target area to intelligently schedule land surveying and mapping tasks on the land map embedding vector, thereby obtaining a land surveying and mapping instruction network; upload the land surveying and mapping instruction network to the Internet of Things to execute the scheduling instruction replacement.

[0012] Optionally, step S1 specifically includes:

[0013] Step S11: collecting multi-source surveying and mapping data of the target area including spatial positioning data, remote sensing image data, surface elevation data, soil moisture data, and meteorological parameter data through the Internet of Things;

[0014] Step S12: performing multi-source data format parsing and unified coding processing on the multi-source surveying and mapping data of the target area, and performing spatial registration and time series alignment based on a unified coordinate reference and time reference to generate a standard surveying and mapping data matrix;

[0015] Step S13: performing noise identification and anomaly elimination processing on the standard surveying and mapping data matrix, and extracting a stable and effective surveying and mapping observation sequence to generate a high-confidence surveying and mapping data set;

[0016] Step S14: construct a multimodal information channel based on the data type dimension of the high-confidence surveying and mapping dataset, and perform weighted mapping on the multimodal information channel to obtain a node space perception map;

[0017] Step S15: Perform regional feature clustering and land type recognition based on the node spatial perception map, extract high-order regional perception feature vectors, and obtain original ground measurement fusion data.

[0018] Optionally, step S13 is specifically as follows:

[0019] Step S131: discriminating abnormal surveying and mapping noise samples based on the temporal variation trend, spatial gradient and numerical distribution characteristics in the standard surveying and mapping data matrix to obtain an initial noise mask set;

[0020] Step S132: performing fine-grained classification on abnormal noise samples in the standard surveying and mapping data matrix based on the initial noise mask set, and outputting a labeled noise removal index table;

[0021] Step S133: performing regional weighted completion and credible interval residual correction on the abnormal points marked in the labeled noise removal index table to obtain a multi-dimensional observation repair matrix;

[0022] Step S134: extracting an observation data subset based on the multidimensional observation repair matrix that simultaneously satisfies the following conditions: temporal continuity ≤ 1 hour, spatial continuity ≤ 7 meters, change rate ≤ 8% within the past three time points, and sensor channel value difference ≤ 0.08;

[0023] Step S135: Calculate the confidence score of each data point in the three dimensions of accuracy, stability and continuity in the observation data subset, and remove the data points whose average confidence score of the three dimensions is lower than the preset confidence score threshold to obtain a high-confidence mapping data set.

[0024] Optionally, step S15 is specifically as follows:

[0025] Step S151: performing initial regional feature division based on the structural edge weights and spatial embedding relationships of the node spatial perception graph to obtain a set of spatial clustering units;

[0026] Step S152: combining the multimodal node attributes in the spatial clustering unit set to construct a land class candidate attribute vector, and performing a priori remote sensing interpretation on the land class candidate attribute vector to determine the initial land class type, thereby obtaining a preliminary land class judgment map;

[0027] Step S153: performing local feature compensation learning on the boundary area in the preliminary land class judgment map to generate a boundary optimized land class map;

[0028] Step S154: extracting the structural center vector and characteristic statistical index of each land class unit according to the boundary optimized land class map, thereby constructing a regional land class aggregation feature set;

[0029] Step S155: Perform high-order spatial semantic fusion and redundant feature compression on the regional land classification aggregation feature set to generate original land survey fusion data.

[0030] Optionally, the multimodal semantic deconstruction in step S2 is specifically as follows:

[0031] The original fusion data of ground survey is divided into spatial positioning sequence, remote sensing image channel, surface elevation model unit and ground object environment parameter set according to the data source type;

[0032] Image texture features are extracted based on remote sensing image channels, and combined with the slope, aspect, and height difference attributes in the surface elevation model unit to construct a primary spatial semantic feature set.

[0033] The trajectory distribution pattern in the spatial positioning sequence is used to perform spatial overlay analysis with land cover change points, extracting the semantic labels of the spatial overlay positions, and matching them with the preset geographic entity coding system to generate a candidate land class annotation set;

[0034] Semantic feature alignment and fusion are performed on the primary spatial semantic feature set, the ground object environment parameter set and the candidate land class annotation set to generate a modality-aligned semantic vector.

[0035] Optionally, the spatial-semantic-attribute mapping feature space reconstruction in step S2 is specifically as follows:

[0036] Mapping the spatial positioning information in the modality-aligned semantic vector to a unified coordinate base for geocoding and back-projection, thereby constructing a basic spatial unit grid.

[0037] A semantic mapping grid is constructed using the basic spatial unit grid as the basic unit, and a location-semantic-attribute triple set is constructed based on the spatial texture characteristics, land use labels and environmental attribute indicators of the semantic mapping grid;

[0038] Vectorized land class expression feature encoding is performed on the location-semantics-attribute triple set to obtain a set of spatial unit triples. Dimensionality reduction and reconstruction are then performed on the high-dimensional space to extract the principal component semantic distribution factor and generate a spatial-semantic feature matrix.

[0039] Combining the terrain adjacency relationship, land cover similarity and environmental factor correlation between spatial units in the basic spatial unit grid, a surveying and mapping attribute association graph structure is constructed, and the attributes of each spatial unit are supplemented and the association is strengthened to obtain a surveying and mapping attribute enhancement structure;

[0040] The spatial-semantic feature matrix and the mapping attribute enhancement structure are fused through high-dimensional embedding to obtain the land semantic unit set.

[0041] Optionally, the on-chain confirmation of the significant land elements in step S3 is specifically as follows:

[0042] Based on the land semantic units in the surveying and mapping data index chain, the land class code, spatial boundary and area index are extracted for each spatial unit to construct the land feature attribute metadata table;

[0043] Perform multidimensional significance evaluation on the land feature attribute metadata table, screen out the land feature sets with average significance greater than [0.6, 0.85], and thus construct the significant land feature set;

[0044] Mapping the collection of significant land features to the surveying and mapping data index chain structure, performing ownership confirmation, data source verification, and timestamp registration for each significant land feature, and generating an on-chain candidate pool for title confirmation transactions;

[0045] The land elements in the candidate pool of on-chain title confirmation transactions are processed through a multi-node consensus verification mechanism to generate a surveying and mapping title confirmation certificate block;

[0046] The surveying and mapping right confirmation certificate block is chain-bound with the surveying and mapping data index chain, the right confirmation status and identification index are marked, and an identification block on the surveying and mapping chain is generated.

[0047] Optionally, step S4 is specifically:

[0048] Step S41: Based on the boundary coordinates of the spatial objects in the identification blocks on the surveying and mapping chain and the ownership confirmation relationship, a topological adjacency relationship between the spatial units is established to generate an initial topological map structure of the geographic elements;

[0049] Step S42: introducing land element attributes in the land element attribute metadata table into the nodes in the initial topological graph structure of the geographic elements, and assigning spatial coupling weights to the edges in the graph to generate a geographic element topological graph;

[0050] Step S43: performing structural learning on the geographic element topology map, extracting the high-dimensional spatial-semantic embedding representation of each land element, and generating an initial node embedding vector set;

[0051] Step S44: performing spectral clustering and aggregate convolution operations on the initial node embedding vector set to obtain a land-class aggregated embedding vector map;

[0052] Step S45: Perform position normalization and edge weight optimization reordering operations based on the land class aggregation embedded vector map to generate a land map embedded vector.

[0053] Optionally, step S5 is specifically as follows:

[0054] Step S51: Acquire real-time multi-source sensor data of the target area, perform regional sensing fusion, and generate a real-time mapping sensing data packet;

[0055] Step S52: using the real-time mapping perception data packet to input the pre-trained dynamic graph attention scheduling model, performing node weight update and edge weight adjustment on the land map embedding vector to generate a task-sensitive map embedding vector;

[0056] Step S53: Based on the task-sensitive graph embedding vector and in combination with the acquired current surveying and mapping task list, a surveying and mapping task-area-resource mapping graph model is constructed;

[0057] Step S54: performing scheduling path optimization and task priority sorting on the surveying and mapping task-area-resource mapping model to generate a land surveying and mapping instruction set;

[0058] Step S55: Encode the land surveying and mapping instruction set into a standardized land surveying and mapping instruction network, and send it to the corresponding operation unit through the IoT edge gateway node to execute the scheduling instruction replacement task.

[0059] This invention, by introducing an Internet of Things (IoT) perception system and a graph expression method, enables full-process processing, modeling, and task scheduling of multi-source surveying and mapping data, significantly improving the intelligence and processing efficiency of land surveying and mapping operations. Through spatial registration and temporal alignment of regional multi-source data, the multidimensional consistency of surveying and mapping data is ensured, forming a standard surveying and mapping data matrix. This effectively avoids data distortion caused by inconsistent coordinate references and time delays. The introduction of a unified coordinate reference and time reference significantly improves the spatiotemporal coordination capabilities of cross-source data. The generation of high-confidence surveying and mapping datasets uses thresholds such as temporal continuity ≤ 1 hour, spatial continuity ≤ 7 meters, rate of change ≤ 8%, and channel difference ≤ 0.08 to accurately screen stable and reliable data points, enhancing the robustness of subsequent feature extraction and mapping. During multimodal semantic deconstruction, spatial semantic labels are constructed by fusing remote sensing textures, DEM elevation differences, and environmental parameters, enhancing the expressive power of feature category recognition. The high-dimensional fusion of semantic vectors and attribute triples not only enables refined expression of spatial unit semantics but also provides a structured semantic foundation for subsequent mapping and scheduling. An on-chain ownership confirmation mechanism ensures the credibility and immutability of surveying and mapping data in multi-node collaborative scenarios. Land features with a significance greater than [0.6, 0.85] are confirmed, which not only improves on-chain load efficiency but also focuses on key geographic units, increasing the business value density of data ownership confirmation. The construction of geographic feature topology maps and aggregated graph learning effectively capture the complex relationships between regional spatial distribution, semantic labels, and attribute coupling, enhancing contextual understanding for land class identification and task assignment. A dynamic graph attention scheduling structure dynamically updates node and edge weights based on real-time task lists and sensor data, enabling adaptive matching of scheduling paths to task status, resource capabilities, and regional characteristics, enhancing the timeliness and resource adaptability of task paths. Finally, through priority sorting and instruction set encapsulation, a closed-loop system for land surveying and mapping task scheduling is constructed, achieving consistent structural distribution of task instructions and precise control of work units, significantly improving the responsiveness, scheduling efficiency, and task execution quality of the land surveying and mapping system in dynamic environments.

[0060] Optionally, this specification also provides a surveying and mapping data intelligent analysis and management system based on the Internet of Things, which is used to execute the surveying and mapping data intelligent analysis and management method based on the Internet of Things as described above. The surveying and mapping data intelligent analysis and management system based on the Internet of Things includes:

[0061] The regional perception fusion module is used to collect multi-source surveying and mapping data of the target area through the Internet of Things, and perform regional perception fusion based on the multi-source surveying and mapping data of the target area to obtain the original fusion data of the ground survey;

[0062] The semantic reconstruction module is used to perform multimodal semantic deconstruction on the original fusion data of land surveying to obtain the modal alignment semantic vector; based on the modal alignment semantic vector, the spatial-semantic-attribute surveying feature space is reconstructed to obtain the land semantic unit set;

[0063] The blockchain rights confirmation module is used to centralize the land semantic units into various spatial objects for blockchain data rights confirmation, thereby building a surveying and mapping data index chain; the surveying and mapping data index chain is used to perform on-chain rights confirmation of significant land elements to obtain an identification block on the surveying and mapping chain;

[0064] The topology analysis module is used to construct a geographic element topology map based on the identification blocks on the surveying and mapping chain, and use the land element topology map to extract the associated features of the land elements to obtain the land map embedding vector;

[0065] The instruction generation module is used to obtain real-time multi-source sensor data of the target area, and use the real-time multi-source data of the target area to intelligently schedule land surveying and mapping tasks on the land map embedding vector, thereby obtaining a land surveying and mapping instruction network; and uploading the land surveying and mapping instruction network to the Internet of Things to execute the scheduling instruction replacement.

[0066] The surveying and mapping data intelligent analysis and management system based on the Internet of Things of the present invention can implement any one of the surveying and mapping data intelligent analysis and management methods based on the Internet of Things of the present invention, and is used to combine the operation and signal transmission medium between various modules to complete the surveying and mapping data intelligent analysis and management method based on the Internet of Things. The internal modules of the system cooperate with each other, thereby improving the response speed, scheduling efficiency and task execution quality of the land surveying and mapping system in a dynamic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0068] Figure 1 This is a schematic diagram of the steps of the method for intelligent analysis and management of surveying and mapping data based on the Internet of Things of the present invention;

[0069] Figure 2 Detailed step flow diagram of step S1 in the present invention;

[0070] Figure 3 Detailed flowchart of step S13 in the present invention;

[0071] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0072] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0073] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0074] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0075] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for intelligent analysis and management of surveying and mapping data based on the Internet of Things, the method comprising the following steps:

[0076] Step S1: Collect multi-source surveying and mapping data of the target area through the Internet of Things, and perform regional perception fusion based on the multi-source surveying and mapping data of the target area to obtain the original fusion data of the ground survey;

[0077] In this example, a surveying and mapping sample area located in the Yellow River Delta agricultural reclamation area was selected. An IoT surveying and mapping node system, including an RTK high-precision positioning terminal, a multispectral remote sensing imager mounted on an unmanned aerial vehicle (UAV), a lidar device, a surface temperature and humidity composite sensor, and an automatic weather station for wind speed and direction, was deployed to collect multi-source data within the area for 72 hours. The collected spatial positioning data frequency was set to 10 Hz, the remote sensing image resolution was 0.5 meters, and the lidar point cloud density was no less than 4 points per square meter. This multi-source data was input into the edge data preprocessing platform and standardized using the WGS-84 spatial coordinate system and UTC time base. A five-dimensional channel fusion structure was then used to construct a five-channel tensor map of spatial positioning, elevation, humidity, imagery, and wind speed. After evaluating feature gain using a channel mutual information evaluation function, a channel-weighted entropy fusion strategy was used to generate a regional spatial density field distribution map. This density field was based on a 100×100 grid unit, forming the original fused ground survey data.

[0078] Step S2: Perform multimodal semantic deconstruction on the original fused land survey data to obtain a modality-aligned semantic vector; perform spatial-semantic-attribute mapping feature space reconstruction based on the modality-aligned semantic vector to obtain a land semantic unit set;

[0079] In this embodiment, the constructed spatial density field distribution map is decomposed into a positioning sequence matrix A (dimensions 100×100×3), a remote sensing texture matrix B (100×100×7), a terrain elevation unit matrix C (100×100×1), and an environmental parameter tensor D (100×100×5) according to channel attributes. First, the Gabor texture response features in the texture matrix B are combined with the height gradient in C to form a primary texture landform label set L1. Subsequently, the movement trajectory fitting curve in A is spatially superimposed with L1. The superimposed area is spatially semantically matched based on the segmented centroid position. This is then compared with the standard land class labels in the national land class coding library to obtain a label matching matrix M (containing preliminary labels for 34 land class categories). L1, D, and M are then multimodally fused. Modal alignment embedding is performed under the conditions of spatial coordinate position consistency weight and channel saliency weight of 0.35 and 0.65, respectively, to ultimately generate a modal alignment semantic vector tensor of the form 100×100×15. Next, principal component analysis (PCA) was performed on the semantic vector tensor within a unified spatial grid, retaining the top five principal factors to construct a semantic embedding matrix V (100 × 100 × 5). Simultaneously, a land class triple structure (x-coordinate, y-coordinate, label number) was constructed. Based on the spatial overlap coefficient and attribute correlation between land use types, a land attribute adjacency matrix W (100 × 100, asymmetric sparse structure) was generated. V and W were then fused to form a set of land semantic units.

[0080] Step S3: Concentrate the land semantic units on each spatial object and perform blockchain data rights confirmation, thereby constructing a surveying and mapping data index chain; perform on-chain rights confirmation of significant land elements on the surveying and mapping data index chain to obtain an identification block on the surveying and mapping chain;

[0081] In this example, an attribute data table is constructed for each unit in the land semantic unit set, including the land class code, boundary polygon vertex coordinates, and area value. This data is stored in a structured record format (JSON + GeoJSON combination). Each record in this data table is digested with a SHA-256 hash and uploaded to the blockchain node system. Based on the consortium chain architecture built on Fabric version 2.2, the channel structure is used to segment the surveying and mapping data into an index chain by region. The chain structure adopts a Merkle tree structure, with every 50 spatial units as a leaf node batch. A significance assessment is then performed. The significance score is composed of land class representativeness (weight 0.3), area proportion (0.25), topological centrality (0.25), and attribute dissimilarity (0.2). Min-Max normalization is used to form a significance matrix S (N×4 dimensions). Records with a mean score falling between [0.6, 0.85] are filtered to form a significant candidate set. The significant candidate set is then subject to the Fabric smart contract to execute the ownership confirmation transaction, including registering the owner ID (using ECDSA signature verification), verifying the data source hash, and recording timestamp. Each title confirmation transaction constructs a title confirmation block structure containing 5 fields: identification ID, original hash, signature, registration time and status code, and is submitted to the main ledger in the chain to form an identification block on the surveying and mapping chain.

[0082] Step S4: construct a geographic element topology map based on the identification blocks on the surveying and mapping chain, and use the land element topology map to extract the land element correlation features to obtain the land map embedding vector;

[0083] In this embodiment, the spatial boundary coordinates recorded in the on-chain identification block are read, and the topological relationships between adjacent boundaries are calculated, including three adjacency types: common boundary, intersection, and inclusion. An adjacency matrix G (N×N Boolean values) is constructed, and G is then converted into a weighted graph, where the weights are determined by the adjacency type and the sum of the spatial overlap area. Based on this, an initial geographic element topological map is constructed, and its node attributes are populated with fields from the aforementioned significant land element attribute table (a total of 12-dimensional attribute vectors, including land class code, area, slope, vegetation index, etc.). Subsequently, a graph structure embedding module is introduced, using a four-layer spatial semantic embedding structure: the first layer performs adjacency aggregation (including linear combination and activation function ReLU), the second layer performs layer regularization and residual fusion, the third layer performs feature compression (output dimension is reduced from 12 to 4), and the fourth layer constructs the final node representation vector. After aggregation, spectral clustering is performed to form a land class embedding vector map (node ​​number N, vector dimension 4). On this basis, graph structure optimization is performed, and edges are reordered according to the standard deviation of edge weights, and the final land map embedding vector matrix T (N×4) is output.

[0084] Step S5: Acquire real-time multi-source sensor data of the target area, and use the real-time multi-source data of the target area to intelligently schedule land surveying and mapping tasks on the land map embedding vector, thereby obtaining a land surveying and mapping instruction network; upload the land surveying and mapping instruction network to the Internet of Things to execute the scheduling instruction replacement.

[0085] In this example, a real-time collection cycle of 5 minutes is set. Sensor data includes surface temperature and humidity (recorded every 0.5 minutes), light intensity, real-time rainfall, and air particulate matter indicators. Real-time data is integrated and uploaded to a local processing server using edge acquisition terminals. Based on this, a real-time time series tensor Z (size: 12×100×100×5, representing time×space×space×channel) is constructed. This data is input into a pre-trained graph attention model, which consists of an input projection layer (with a dimensionality change from 5 to 8), a two-layer attention computation (key-value-query structure with 3 heads), and an output update layer (output dimension 8). Based on the land map embedding vector T, the model performs node importance weight adjustment and edge connection optimization to generate a task sensitivity map T' (N×8). The current task list (based on fields such as task number, task priority, required land type, and required area) is then matched with T' to construct a ternary mapping graph M: task node - spatial location - mapping resource. A weight W is calculated for each path (taking into account geographic cost, resource accessibility, and task urgency), and the path with the minimum cost is selected. A scheduling table is generated based on the path and structured into a surveying and mapping task instruction set (including fields such as task ID, target location coordinates, task type, estimated operation time, required resources, etc.), which is uniformly encoded into a JSON structure and encapsulated into a land surveying and mapping instruction network. Finally, it is sent to the field automatic operation equipment through the MQTT protocol of the edge gateway to execute task replacement.

[0086] Optionally, step S1 specifically includes:

[0087] Step S11: collecting multi-source surveying and mapping data of the target area including spatial positioning data, remote sensing image data, surface elevation data, soil moisture data, and meteorological parameter data through the Internet of Things;

[0088] In this example, a 5-kilometer-radius farmland area within the Yellow River Delta agricultural reclamation area was selected as the target area for mapping. Multi-source mapping data was collected collaboratively by deploying LoRaWAN IoT nodes and drone-mounted remote sensing equipment. Spatial positioning data was acquired using GNSS equipment at a sampling frequency of 10 Hz. Remote sensing images were acquired using a multispectral camera with a resolution of 0.5 meters. Surface elevation data was derived from a LiDAR point cloud with a point density of no less than 4 points per square meter. Soil moisture data was acquired in real time by buried sensors at 15-minute intervals. Meteorological parameters, including temperature, humidity, wind speed, and air pressure, were provided by a local micrometeorological station. All sensors pushed raw data packets to the edge server via the MQTT protocol, forming a unified, multi-source mapping raw database.

[0089] Step S12: performing multi-source data format parsing and unified coding processing on the multi-source surveying and mapping data of the target area, and performing spatial registration and time series alignment based on a unified coordinate reference and time reference to generate a standard surveying and mapping data matrix;

[0090] In this embodiment, the GNSS trajectories, remote sensing images, point cloud data, soil moisture curves, and meteorological time series data from the multi-source surveying and mapping data of the target area are converted to a unified format, uniformly encoded into five types of coded data streams, and converted to the WGS-84 spatial reference and UTC time reference. During the spatial registration process, the WGS-84 coordinate system is used as the spatial reference, and the five-point interpolation method is used to align the images and points. The time alignment is set to a time step of 5 seconds and a maximum tolerance error of 0.3 seconds to ensure that the observation sequences from different sensors are temporally and spatially consistent. The resulting standard surveying and mapping data is expressed as a five-dimensional tensor, with the dimensions corresponding to data type, time, spatial position, numerical intensity, and confidence level.

[0091] Step S13: performing noise identification and anomaly elimination processing on the standard surveying and mapping data matrix, and extracting a stable and effective surveying and mapping observation sequence to generate a high-confidence surveying and mapping data set;

[0092] In this embodiment, the reliability of the unified five-dimensional standard surveying and mapping data is tested. First, a noise recognition threshold is set. For example, in the surface elevation data, if the height difference between adjacent pixels exceeds 5 meters, it is marked as an anomaly; in the meteorological data, if the humidity changes suddenly by more than 15% or the wind speed changes by more than 10m / s, it is also regarded as an abnormal fluctuation. The statistical sliding window mechanism is used to eliminate such mutation data. For the remote sensing image part, contrast enhancement and edge strength analysis are used to remove pixel blocks affected by the occluded area. All screened surveying and mapping sequences must pass the minimum confidence coefficient screening, and those below 0.7 are all eliminated. The final high-confidence surveying and mapping data set is represented in the form of a three-dimensional data table, where the columns represent the surveying and mapping dimensions, the rows represent the time steps, and the values ​​in the table represent the observation data intensity.

[0093] Step S14: construct a multimodal information channel based on the data type dimension of the high-confidence surveying and mapping dataset, and perform weighted mapping on the multimodal information channel to obtain a node space perception map;

[0094] In this embodiment, the high-confidence data set is divided into different data channels according to types such as remote sensing images, positioning point clouds, elevation profiles, and meteorological records, and each channel inputs the corresponding data subset. Through a hierarchical weighting strategy, the remote sensing image is given the highest weight of 0.35, the terrain elevation and spatial positioning are set to 0.25 respectively, and the remaining data channels share the remaining weights. An adjacency matrix is ​​established, in which if the spatial distance between two data points is less than 50 meters and the time interval is less than 10 seconds, it is considered that there is a connection relationship and the value is assigned to 1, otherwise it is 0. A node space perception graph is constructed based on the adjacency matrix, in which the nodes represent spatial points, and the weight of each edge comprehensively considers the weight distance function of the time difference and the data type difference. Finally, the graph is stored in the form of a sparse matrix for subsequent spatial aggregation and semantic extraction.

[0095] Step S15: Perform regional feature clustering and land type recognition based on the node spatial perception map, extract high-order regional perception feature vectors, and obtain original ground measurement fusion data.

[0096] In this embodiment, regional aggregation analysis is performed on the constructed node spatial perception map. First, a feature density-based aggregation strategy is adopted to classify nodes with a spatial distance less than 30 meters and a semantic information similarity greater than 0.8 into the same cluster. Within each cluster, 9-dimensional spatial statistical features are extracted based on remote sensing image attributes such as texture direction, terrain slope change, and soil moisture gradient to form a primary regional feature vector. Further, high-order relationship modeling is performed on each cluster feature to construct a set of fusion vectors \mathbf{F}=[f_1,f_2,...,f_n], where each f_i represents the aggregated perception state of a region, and the vector length is 128. This perception vector is used to represent the overall geodetic semantics of the corresponding region, and the final output is geodetic raw fusion data, providing a data foundation for subsequent semantic recognition and surveying and mapping task scheduling.

[0097] Optionally, step S13 is specifically as follows:

[0098] Step S131: discriminating abnormal surveying and mapping noise samples based on the temporal variation trend, spatial gradient and numerical distribution characteristics in the standard surveying and mapping data matrix to obtain an initial noise mask set;

[0099] In this embodiment, five main data sources in the standard surveying and mapping data matrix are selected: remote sensing image brightness value, DEM elevation value, soil moisture percentage, wind speed (m / s), and surface temperature (℃). The continuous observation data within 24 hours are subjected to preliminary feature extraction, including temporal variation trend, spatial gradient and numerical distribution characteristics. In each sensing channel, a time series difference curve is constructed with a sampling step of five minutes to identify gradient mutation points, and the neighborhood gradient field is calculated in combination with the local spatial difference. If the absolute value of the gradient is greater than two standard deviations of the average level, it is marked as an abnormal trend. At the same time, the empirical distribution function of each data is calculated, and the upper and lower 5% quantiles are set as abnormal distribution thresholds. All data points that meet any of the three conditions of temporal mutation, spatial anomaly and distribution extremes are classified as potential anomalies, and the initial noise mask set is output. Its structure is a 0-1 Boolean mask matrix with the same dimension as the standard matrix.

[0100] Step S132: performing fine-grained classification on abnormal noise samples in the standard surveying and mapping data matrix based on the initial noise mask set, and outputting a labeled noise removal index table;

[0101] In this embodiment, the initial noise mask set is read, and a fine classification process is introduced for each data point marked as abnormal. Each suspected noise point is located with a three-dimensional identification index (spatial coordinates x, y, timestamp t), and is divided into a severe disturbance class (deviation greater than 3 times the standard deviation) according to the degree of deviation between its original value and the average value of the adjacent space-time points, a slowly varying dislocation class (the change trend is discontinuous but the amplitude is small), and a structural distortion class (such as edge blur and shadow interference in remote sensing images). The classification results are stored in a dictionary structure index. Each record contains fields such as index number, data type, abnormal category, offset value, channel to which it belongs, original value, etc., and is output as a labeled noise removal index table for reference in subsequent repair operations.

[0102] Step S133: performing regional weighted completion and credible interval residual correction on the abnormal points marked in the labeled noise removal index table to obtain a multi-dimensional observation repair matrix;

[0103] In this embodiment, a repair method is set for each anomaly category based on the outliers recorded in the noise removal index table. For the severe disturbance category, a regional weighting strategy is adopted: a weighted average is calculated within a 5×5 spatial neighborhood, and the weight is set based on the Euclidean distance from the center point, w_{ij} = \exp(-d_{ij}^2 / 2\sigma^2), where d_{ij} is the Euclidean distance between repair point i and the jth point in its neighborhood, and \sigma = 2.5. Slowly varying dislocations use linear trend extrapolation. Structural distortions introduce confidence interval residual correction, performing local regression fitting on the values ​​within the three previous and subsequent time points, calculating a 95% confidence interval, and replacing the original outlier with the center value. The repaired observations are then added to the original matrix, and the repair method and residual are labeled for each repair point to form a new multidimensional observation repair matrix. Its structure retains the dimensions of the original surveying matrix and adds a new dimension to represent the repair status and residual error.

[0104] Step S134: extracting an observation data subset based on the multidimensional observation repair matrix that simultaneously satisfies the following conditions: temporal continuity ≤ 1 hour, spatial continuity ≤ 7 meters, change rate ≤ 8% within the past three time points, and sensor channel value difference ≤ 0.08;

[0105] In this embodiment, a high-quality observation subset that meets specific constraints is extracted from the repaired mapping matrix. Each data point is traversed in turn to calculate whether the interval between it and the previous time point is less than 1 hour, and whether the Euclidean distance between it and the nearest valid neighbor point in space is less than 7 meters; then further determine whether the value change amplitude of the point in the past three time points does not exceed 8% continuously, such as the wind speed change is within 0.3m / s and the soil moisture change is less than 2%. At the same time, the difference coefficient of the data point between each sensor channel is calculated, for example, the temperature and humidity channel difference is <0.08. All those who meet the four conditions at the same time are included in the observation data subset. The subset is stored with the index number of the valid point screened out in the original matrix, and is accompanied by a time continuity flag and a spatial proximity score.

[0106] Step S135: Calculate the confidence score of each data point in the three dimensions of accuracy, stability and continuity in the observation data subset, and remove the data points whose average confidence score of the three dimensions is lower than the preset confidence score threshold to obtain a high-confidence mapping data set.

[0107] In this embodiment, for the extracted subset of observation data, confidence scores of three dimensions are calculated for each data point. The precision score is normalized to [0,1] using the inverse of its repair residual as the score; the stability score is linearly reverse mapped to [0,1] according to its standard deviation of the rate of change; the continuity score is based on the average length of the observation interval in the last 5 moments, and the inverse of the ratio of the set standard interval (such as once every 10 minutes) forms a score. The weighted average of the three is used to obtain the total confidence score, with the weights set to 0.4 (precision), 0.3 (stability), and 0.3 (continuity) respectively. Samples with an average score lower than 0.75 are judged as low-confidence data and eliminated. The data points finally retained constitute a high-confidence mapping data set, which is output as a three-dimensional tensor structure with the dimensions of data type × spatial position × time step, and contains a scoring field for subsequent screening and review.

[0108] Optionally, step S15 is specifically as follows:

[0109] Step S151: performing initial regional feature division based on the structural edge weights and spatial embedding relationships of the node spatial perception graph to obtain a set of spatial clustering units;

[0110] In this embodiment, the target area is initially divided into spatial features based on the edge weights between the nodes in the node space perception graph and their corresponding spatial embedding coordinate information. The edge weight represents the spatial correlation between the nodes, and the value range is set to [0,1]. The spatial distance d_ij and the attribute similarity sim_ij are normalized and weighted, and the formula is w_ij = exp(-d_ij / σ_d) × sim_ij, where σ_d is the spatial scale adjustment factor, set to 50 meters. The embedding relationship is represented by a two-dimensional vector (x_i, y_i), and each node is projected onto a unified spatial feature plane through principal component analysis. Subsequently, the density peak clustering method is adopted, and the edge weight and spatial position are used as the clustering judgment criteria to output the cluster label vector C = {c1, c2, ..., cn}, where each label corresponds to a spatial clustering unit. The final set of spatial clustering units is stored in polygon GeoJSON format, providing a spatial segmentation basis for subsequent land attribute recognition.

[0111] Step S152: combining the multimodal node attributes in the spatial clustering unit set to construct a land class candidate attribute vector, and performing a priori remote sensing interpretation on the land class candidate attribute vector to determine the initial land class type, thereby obtaining a preliminary land class judgment map;

[0112] In this embodiment, after the spatial clustering unit set is constructed, five modal attributes are extracted from each unit: remote sensing band reflectance vectors (e.g., B3, B4, B8, B11), NDVI values, terrain slope, mean elevation, and mean soil moisture. These attributes are then structured as a set of candidate land class attributes, A_i = [r1, r2, ..., r5]. Each vector is normalized to the interval [0, 1] and scaled before being fed into a priori land class interpretation system. This system uses a remote sensing expert knowledge base to set threshold rules for different attribute combinations and makes classification decisions. For example, if NDVI > 0.4 and B11 < 0.2, it is classified as forest land, while if slope > 15° and NDVI < 0.2, it is classified as bare rock. This process does not rely on deep neural networks but is instead performed through a cross-search of threshold intervals. The final result for each clustering unit is the initial land class classification, which is output as a two-dimensional grid map with a cell resolution of 10m × 10m, forming a preliminary land class judgment map.

[0113] Step S153: performing local feature compensation learning on the boundary area in the preliminary land class judgment map to generate a boundary optimized land class map;

[0114] In this embodiment, considering that the boundary areas in the preliminary land classification map often have mixed pixels and blurred transitions between land features, local feature resampling and residual compensation are performed on the nodes within the 5-meter buffer zone of the cluster unit boundary. A sliding window (window size is 3×3) is used to extract the attribute difference Δa_ij of adjacent pixels for the boundary pixels, and the main attribute value of the boundary pixel is corrected by the weighted mean. For the attribute cross-fuzzy section, normalization is performed by constructing the feature difference residual matrix R_k=|A_i-A_j|, and the residual threshold θ=0.1 is set for difference fusion. If it is lower than the threshold, it is judged as the same type of boundary completion, and if it is higher, the boundary is retained. The compensated boundary area is re-labeled with land class attributes, and the updated boundary information is merged into the main map to generate a boundary optimized land class map, which has higher boundary consistency and classification stability.

[0115] Step S154: extracting the structural center vector and characteristic statistical index of each land class unit according to the boundary optimized land class map, thereby constructing a regional land class aggregation feature set;

[0116] In this embodiment, based on the land class map after boundary optimization, the centroid coordinate set (X_i, Y_i) of all units in each land class is counted, and the structural center vector C_i = [mean(X_i), mean(Y_i)] of each type of land feature is calculated. The statistical characteristics of the unit in terms of area, shape complexity (expressed as the ratio of the square of the perimeter to the area), elevation mean, vegetation index mean, etc. are further counted to form a structural statistical vector S_i = [area_avg, shape_comp, elev_avg, ndvi_avg]. By merging C_i and S_i, an aggregated feature record F_i = [C_i, S_i] is formed for each land class, and the feature set F = {F1, F2, ..., Fn} corresponding to all land class types is summarized in a tabular form. These feature data are used to represent the stable center position and attribute feature trend of the spatial distribution of each land class, providing a quantitative basis for subsequent high-order semantic modeling and feature fusion.

[0117] Step S155: Perform high-order spatial semantic fusion and redundant feature compression on the regional land classification aggregation feature set to generate original land survey fusion data.

[0118] In this embodiment, after constructing the regional land class aggregate feature set F, a high-order spatial semantic fusion process is performed on it. The similarity between land classes is measured by constructing a semantic correlation matrix M_ij = cosine_similarity(F_i, F_j), and feature fusion is performed based on the similarity matrix. First, land class types with redundancy greater than 0.85 are removed, and their feature vectors are merged into the nearest land class. Second, principal component compression is performed on the feature vectors in the feature set F' that remains after redundant feature removal in the regional land class aggregate feature set F. Only the first three principal components with a cumulative variance explanation rate greater than 95% are retained as the fused high-order semantic vectors H_i. Ultimately, a unified land survey raw fusion data structure is formed, with H = {H1, H2, ..., Hm} representing the fused semantic representation of all regions. Each vector dimension is uniformly 3, corresponding to structural spatiality, surface attributes, and multimodal matching, respectively, to support subsequent semantic deconstruction and spatial management tasks.

[0119] Optionally, the multimodal semantic deconstruction in step S2 is specifically as follows:

[0120] The original fusion data of ground survey is divided into spatial positioning sequence, remote sensing image channel, surface elevation model unit and ground object environment parameter set according to the data source type;

[0121] In this embodiment, the raw fused ground survey data is split into four information channels based on the data source type. The spatial positioning sequence includes a set of GPS track points based on the WGS-84 coordinate system, with data recorded at 0.5-second intervals. The remote sensing image channel is a multispectral remote sensing image with a spatial resolution of 0.5 meters. The surface elevation model unit is a 1-meter resolution DEM data grid, containing slope, aspect, and adjacent height difference fields for each pixel. The ground feature environmental parameter set includes meteorological data such as soil moisture, temperature, and wind speed. The data is linearly interpolated from hourly sampling to every 5 minutes to ensure synchronization with the spatial positioning sequence. After division, all data is stored in a unified data matrix set T = {L, I, D, E}, corresponding to the four data channels, and recording the corresponding timestamps and region block numbers.

[0122] Image texture features are extracted based on remote sensing image channels, and combined with the slope, aspect, and height difference attributes in the surface elevation model unit to construct a primary spatial semantic feature set.

[0123] In this embodiment, the red, green, blue and near-infrared bands are selected from the remote sensing image channel I to perform a 3x3 neighborhood sliding window traversal, and the gray-level co-occurrence matrix (GLCM) is calculated for each sliding window area to extract texture statistical features such as contrast, energy, entropy, and homogeneity to generate an image texture vector set V_img; then, from the DEM unit D, the slope, slope direction and adjacent maximum height difference of each pixel are calculated using the eight-neighborhood analysis method to form a terrain structure feature matrix M_terrain∈R^(h×w×3), where h and w are grid dimensions. Finally, V_img and M_terrain are spliced ​​at the corresponding regional positions and mean pooled by regional blocks to form a primary spatial semantic feature set S0={s1,s2,...,s n}, each s i ∈R 15 The basic spatial representation of a surface area block.

[0124] The trajectory distribution pattern in the spatial positioning sequence is used to perform spatial overlay analysis with land cover change points, extracting the semantic labels of the spatial overlay positions, and matching them with the preset geographic entity coding system to generate a candidate land class annotation set;

[0125] In this embodiment, the trajectory data in the spatial positioning sequence L is read, and the speed, steering angle and time continuity index are calculated for each trajectory; the change intensity distribution map G∈R^(x×y) is generated by combining the historical trajectory overlay map, where each position pixel G(i,j) represents the trajectory change density of the point in the past 72 hours. The map is overlapped with the land cover change point raster map generated by remote sensing images, and the overlapping area is used as the candidate change position set R; for each pixel in R, the corresponding geographic entity coding system is queried according to its latitude and longitude coordinates, for example, using the GB / T20261 standard, the location point is classified into a predefined geographic category number, such as 010101 (arable land), 030201 (urban construction land), etc., and finally a candidate land class annotation set C_tag={c1,c2,...,c_k} is formed, and each c i Contains location coordinates and matching class number.

[0126] Semantic feature alignment and fusion are performed on the primary spatial semantic feature set, the ground object environment parameter set and the candidate land class annotation set to generate a modality-aligned semantic vector.

[0127] In this embodiment, the primary spatial semantic feature set S0, the ground object environment parameter set E and the candidate land class label set C_tag are mapped one by one according to the spatial region number to construct a multi-source semantic fusion structure Q = {q1, q2, ..., q n}, where each q i Contains the eigenvector s i 、Meteorological environment vector e i and land class label c i Based on the principle of unified feature dimension, each vector is normalized to the interval [0,1], and cosine similarity is used for feature reconstruction: for each group {s i ,e i}, construct weighted fusion expression f i =α·s i +β·e i , α and β are set to 0.6 and 0.4 respectively to enhance the spatial texture expression. Then, the fusion feature f i Align with the land class label vector, and form a modality alignment semantic vector set F_align={f1',f2',...,f n '}, each vector has uniform dimension R 30 The expression structure is used for subsequent semantic space reconstruction tasks.

[0128] Optionally, the spatial-semantic-attribute mapping feature space reconstruction in step S2 is specifically as follows:

[0129] Mapping the spatial positioning information in the modality-aligned semantic vector to a unified coordinate base for geocoding and back-projection, thereby constructing a basic spatial unit grid.

[0130] In this embodiment, the spatial positioning subvector in the modal alignment semantic vector is converted into the WGS-84 coordinate standard, and a three-element coordinate matrix L = [l1, l2, ..., ln] containing longitude, latitude and elevation information is constructed. Each l i is the three-dimensional coordinate vector (x i ,y i ,z i ), and use the back-projection function R_proj to calculate the geographic coordinates of the pixel-level coordinates in the remote sensing image. Then, according to the unified resolution standard, the spatial cell size is set to 20m×20m, and the study area is divided into a regularized basic spatial grid set G = {g1, g2, ..., gm}, each g j With geographic coordinates j One-to-one correspondence. All semantic information is assigned to the mapped grid cells according to their spatial orientation, establishing a direct mapping relationship from semantic vectors to underlying spatial locations.

[0131] A semantic mapping grid is constructed using the basic spatial unit grid as the basic unit, and a location-semantic-attribute triple set is constructed based on the spatial texture characteristics, land use labels and environmental attribute indicators of the semantic mapping grid;

[0132] In this embodiment, the constructed basic spatial grid is used as a unit to calculate the texture features within each grid, including the contrast, uniformity and entropy value in the gray-level co-occurrence matrix, and the corresponding land use labels (such as farmland, forest land, water body) and environmental attributes (such as humidity, temperature, wind speed, etc.) are extracted to form a position-semantic-attribute triplet t i =(p i ,s i ,a i ). where p i Indicates g i The geographic center coordinates, s i is the grid semantic label, a i is a three-dimensional attribute vector. The entire set of regional triples, T = {t1, t2, ..., tn}, is stored in a structured form for subsequent processing. To enhance the spatial consistency of the semantic mapping grid, a grid-level adjacency analysis is performed on T to ensure that the continuity of semantic and attribute distributions between adjacent cells is at least 80% stable.

[0133] Vectorized land class expression feature encoding is performed on the location-semantics-attribute triple set to obtain a set of spatial unit triples. Dimensionality reduction and reconstruction are then performed on the high-dimensional space to extract the principal component semantic distribution factor and generate a spatial-semantic feature matrix.

[0134] In this embodiment, each element in the triple set T is vector-encoded, and the semantic label is converted into a one-hot encoding form s i '∈R^k (k is the number of semantic label types), attribute vector a i Retain the original dimension and form the spatial unit semantic vector e i =[p i ,s i ',a i ]∈R^d. Integrate all e i A high-dimensional semantic expression matrix E∈R^{n×d} is formed. PCA is then used to reduce the dimensionality of E and extract the first r principal component vectors (r is typically 5–10) to construct a low-dimensional main semantic feature factor matrix F∈R^{n×r}, where each row represents the embedded representation of a spatial unit in the main semantic distribution direction. This matrix is ​​used for subsequent spatial relationship modeling and semantic fusion.

[0135] Combining the terrain adjacency relationship, land cover similarity and environmental factor correlation between spatial units in the basic spatial unit grid, a surveying and mapping attribute association graph structure is constructed, and the attributes of each spatial unit are supplemented and the association is strengthened to obtain a surveying and mapping attribute enhancement structure;

[0136] In this embodiment, a spatial graph structure G'=(V, E, A) is constructed based on the divided basic spatial units, where V is a set of spatial grids, E is a set of edges between adjacent grids, and A is an attribute association matrix. i ,v j ), calculate its terrain adjacency d_topo(v i ,v j ), land cover similarity d_cover(v i ,v j ) and the correlation between environmental factors d_env(v i ,v j ), using the comprehensive weighted distance metric function D(v i ,v j ) = α·d_topo+β·d_cover+γ·d_env (α, β, and γ are empirical weights, satisfying α+β+γ=1) to establish edge weights. By traversing the graph structure, attribute completion is performed based on the mean and covariance of adjacent nodes for nodes with missing or inconsistent attributes in the spatial unit, and an enhanced attribute graph A'∈R^{n×f} is constructed, where each row represents the enhanced attribute vector of the unit.

[0137] The spatial-semantic feature matrix and the mapping attribute enhancement structure are fused through high-dimensional embedding to obtain the land semantic unit set.

[0138] In this embodiment, the spatial-semantic feature matrix F∈R^{n×r} and the mapping attribute enhancement structure A'∈R^{n×f} are concatenated row by row to obtain the fusion expression matrix Z∈R^{n×(r+f)}. By constructing a shallow semantic embedding structure, each unit feature vector z i ∈R^{r+f} is processed by the linear mapping W1∈R^{(r+f)×h} and the nonlinear activation function ReLU to generate an intermediate vector h i ∈R^h, and then mapped to the final semantic space through the fully connected layer W2∈R^{h×m}, outputting the land semantic unit y i ∈R^m, where m is the semantic dimension of the preset land class space. All y i The composed set is the land semantic unit set Y = {y1, y2, ..., yn}, which has a unified semantic embedding structure and high-dimensional expression capability, and is suitable for subsequent property rights confirmation, index chain construction and task scheduling.

[0139] Optionally, the on-chain confirmation of the significant land elements in step S3 is specifically as follows:

[0140] Based on the land semantic units in the surveying and mapping data index chain, the land class code, spatial boundary and area index are extracted for each spatial unit to construct the land feature attribute metadata table;

[0141] In this embodiment, the land semantic unit set carried by the surveying and mapping data index chain is used to extract the land class code, boundary polygon, and area index corresponding to each spatial unit one by one. The land class code is based on the 2021 version of the land use classification standard of the Ministry of Natural Resources, and is coded with a three-level land class code structure. For example, "0111" represents the paddy field type; the spatial boundary is recorded in GeoJSON format as a polygon coordinate array P i =[(x1,y1),(x2,y2),...,(x n ,y n )], and ensure the continuity of the boundary through spatial closure verification; the area index uses the equal area conversion formula to uniformly convert the boundary projection area into square meters in the UTM coordinate system, and construct the area vector A=[a1,a2,...,a n ]. Finally, the above information is integrated to form the land feature attribute metadata table M∈R^{n×3}, in which each row records the land class code, boundary vector and area value of the spatial unit, which is used for subsequent ownership confirmation and significance analysis.

[0142] Perform multidimensional significance evaluation on the land feature attribute metadata table, screen out the land feature sets with average significance greater than [0.6, 0.85], and thus construct the significant land feature set;

[0143] In this embodiment, a significance analysis is performed on all records in the land element attribute metadata table M, and a three-dimensional index system is used to evaluate the importance of each element. Specifically, it includes the independence score S1 of the land classification, the boundary complexity score S2 (calculated based on the convex hull area ratio of the boundary polygon), and the area proportion score S3 (calculated as the proportion of the current unit area to the total area of ​​the area). The three are weighted and calculated to obtain a comprehensive significance score S i =w1·S1+w2·S2+w3·S3, where w1=0.3, w2=0.4, and w3=0.3 are empirical settings. Set the significance selection interval to [0.6, 0.85] and select the feature set F={f1,f2,...,f k}, constructing a collection of significant land features. This collection is indexed and associated with the original spatial unit number through the data identification field, ensuring that subsequent operations can be traced and have structural continuity.

[0144] Mapping the collection of significant land features to the surveying and mapping data index chain structure, performing ownership confirmation, data source verification, and timestamp registration for each significant land feature, and generating an on-chain candidate pool for title confirmation transactions;

[0145] In this embodiment, the selected significant land element set F is mapped back to the node index table of the surveying and mapping data index chain structure, and its original data source path, affiliation agency identifier and timeliness information are matched one by one. i , perform three registration operations: first, match the ownership rights by comparing the land classification code with the spatial zoning database and confirm the ownership unit code; second, verify whether the source data comes from a legal collection chain, such as drone images, GNSS measurement records or remote sensing inversion results, and record the data source identifier DID; third, add a standard UNIX timestamp t to each registration information i , into the time registration index pool. Construct the right confirmation transaction candidate pool C = {c1,c2,...,c k}, where each c i =(f i ,DID i ,t i ), which serves as the input basis for subsequent processing on the chain.

[0146] The land elements in the candidate pool of on-chain title confirmation transactions are processed through a multi-node consensus verification mechanism to generate a surveying and mapping title confirmation certificate block;

[0147] In this embodiment, for each transaction in the candidate pool C of the confirmation transaction, the multi-node consensus mechanism of the preset geographic chain platform is called for verification. Assume that there is a set of verification nodes N = {n1, n2, ..., n_m}, and each node verifies the transaction c. iThe consistency of spatial boundary recalculation, data source hash value consistency, and timestamp validity are independently verified. The GeoChain's multi-node consensus mechanism is implemented by building a distributed network structure consisting of multiple surveying and mapping verification nodes. Each node holds an independent surveying and mapping verification copy and key pair, which are used to concurrently review and sign land rights confirmation transactions. When a land rights confirmation transaction (such as the boundaries, code, and data source of a particular plot of land) is proposed, it is first packaged into a structured candidate block containing the plot's spatial ID, timestamp, attribute hash, and data source identifier, and broadcast to all participating nodes. Each node performs hash verification, boundary spatial consistency verification, and historical state comparison on each block, generating a signed response upon successful verification. By aggregating these signatures, a consensus strategy such as BFT or PoA is employed to determine whether a predetermined signature ratio (e.g., more than 67% of nodes have signed) is met as the basis for reaching consensus. Once consensus is reached, the candidate block is confirmed as a "surveying and mapping confirmation certificate block." Its structure includes the plot ID, multi-node signature set, confirmation timestamp, and hash pointer to the previous block. Ultimately, this certificate block is bound to the chain, becoming part of the immutable surveying and mapping data index chain. The verified transaction is signed and approved by more than 2 / 3 of the nodes (the minimum consensus threshold is set to 66.7%), and the surveying and mapping right confirmation certificate structure B is generated. i =(Hash(c i ),Sign(n1)...Sign(n_t)), where t≥floor(2m / 3)+1. All B generated by consensus verification i The collection constitutes the surveying and mapping right confirmation certificate block B={B1,B2,...,B k}, using the Merkle tree structure to construct the block header for hash summary, ensuring that each ownership record is traceable and tamper-resistant.

[0148] The surveying and mapping right confirmation certificate block is chain-bound with the surveying and mapping data index chain, the right confirmation status and identification index are marked, and an identification block on the surveying and mapping chain is generated.

[0149] In this embodiment, for each certificate structure B in the certificate block B, i , insert and bind operations according to the index chain rules to build the identification block on the surveying and mapping chain. The binding process first matches B i The corresponding land element f i Position h(f in the index chain i ), then add a new identification field ID to the linked list structure i =Hash(f i +B i ) and set the status flag to "Confirmed". Each chain record contains a five-tuple record format R i =(ID i ,f i ,Bi ,h(f i ),state), and add it to the next valid block position of the index chain to achieve state update and structure extension. Finally, a stable structure of the mapping chain identification block set L = {R1, R2, ..., R k}, used for subsequent land title confirmation result query, title dispute review and blockchain compliance supervision services.

[0150] It is particularly important that the multidimensional significance evaluation is specifically:

[0151] A multi-dimensional indicator system including land type representativeness, spatial topological importance, area proportion and attribute difference is set to obtain a set of significance indicators;

[0152] In this embodiment, when setting the significance index system, quantitative modeling is performed from four dimensions: (1) Land class representativeness is calculated by integrating land class frequency statistics with geographic dictionary coverage, and the threshold is set to land classes with a frequency of more than 1.5‰ in the entire region; (2) Spatial topological importance is based on the topological map constructed by the Delaunay triangulation, and the degree centrality of each spatial unit is calculated. Units with a value higher than the average value of the entire map are marked as high importance; (3) Area proportion is calculated by counting the proportion of each type of land area to the total area, and the screening lower limit is set to 3%; (4) Attribute difference is based on the Euclidean distance between attribute feature vectors, and the distinction threshold is set to 0.2. The above four types of indicators are uniformly organized into a significance index set according to the land parcel code, forming a multidimensional feature set with the structure S = {s_i1, s_i2, s_i3, s_i4}, where s_ij represents the original significance score of the i-th spatial unit in the j-th indicator dimension.

[0153] Using the significance index set, each significance index in the land element attribute metadata table is numerically normalized to generate a standardized significance feature matrix;

[0154] In this embodiment, when normalizing the set of significance indicators, different normalization methods are used for different indicators. For land type representativeness and area proportion, maximum and minimum value normalization is adopted to map the values ​​to the [0,1] interval; Z-score normalization is used for topological importance; and the attribute difference introduces the exponential mapping rule to enhance the influence of outliers. The unified mapping formula is: s_ij'=(s_ij-min_j) / (max_j-min_j). After normalization, a standardized significance feature matrix M_s∈R^{n×4} is formed, where n is the number of spatial units and 4 is the indicator dimension. Each row in the matrix represents the four-dimensional standard significance feature vector of a spatial unit, which is used for subsequent fusion score calculations.

[0155] Perform weighted linear combination on the standardized saliency feature matrix of each spatial unit to fuse various saliency indicators and calculate the saliency score value;

[0156] In this example, when weighted fusion is performed on the standardized significance feature matrix M_s, a weight vector W = [w1, w2, w3, w4] is set for each significance dimension, where w1 = 0.3 for land class representativeness, w2 = 0.25 for topological importance, w3 = 0.2 for area proportion, and w4 = 0.25 for attribute diversity. The fusion operation is performed according to the linear combination rule, and the significance score S_i = w1s_i1' + w2s_i2' + w3s_i3' + w4s_i4'. The scores of all spatial units are aggregated to generate a significance score vector S∈R^n. The score value range is controlled in the interval [0, 1] for the next screening process.

[0157] The threshold of the significance score valid interval is set to [0.6, 0.85], and the spatial units with significance scores below the lower limit or above the upper limit are eliminated to construct the candidate set of significant land features.

[0158] In this embodiment, when screening the significance score interval, the spatial unit index positions with scores below 0.6 or above 0.85 in the score vector S are removed to retain the candidate plots with scores between [0.6, 0.85]. The removal operation is implemented in the form of a Boolean mask, that is, a mask vector B∈{0,1}^n is constructed, satisfying B_i=1 if and only if 0.6≤S_i≤0.85. The mask is used to filter out the corresponding spatial units from the land feature attribute metadata table to construct a candidate set of significant land features. The candidate set is structured as a set F={f_1,f_2,...,f_m}, where each f_i contains metadata such as spatial unit code, coordinate boundary, and score value.

[0159] The significant land feature candidate set is marked with significance levels and scoring values ​​to generate a significant land feature set.

[0160] In this example, when assigning a grade to a candidate set of significant land features, the scores are first divided into three significance levels within the interval [0.6, 0.85]: 0.6–0.68 is labeled "low significance," 0.68–0.76 is labeled "medium significance," and 0.76–0.85 is labeled "high significance." Each candidate parcel is assigned a corresponding grade label and recorded as a triplet f_i = <code, significance level, significance score>. Once all labeling is complete, the set of significant land features, F_tagged = {f_1,...,f_m}, is aggregated for subsequent on-chain land rights registration or land classification analysis. This set can be visualized or statistically analyzed based on spatial distribution or grade intensity.

[0161] Optionally, step S4 is specifically:

[0162] Step S41: Based on the boundary coordinates of the spatial objects in the identification blocks on the surveying and mapping chain and the ownership confirmation relationship, a topological adjacency relationship between the spatial units is established to generate an initial topological map structure of the geographic elements;

[0163] In this embodiment, first, based on the spatial object boundary coordinates recorded in the identification block on the surveying and mapping chain, a polygon contour point set matching method is used to extract the boundary node sequence of each land unit, and its envelope box is calculated to simplify the adjacency judgment. The minimum bounding box intersection test is performed on the boundary polygon through the spatial index method to determine whether there is boundary overlap or adjacency. For spatial units that meet the adjacency relationship, a set of directed edges are constructed and the directionality is marked to support subsequent topological direction analysis. At the same time, the title confirmation associated field of each spatial object is parsed from the identification block on the chain, and logical connection edges are established for pairs of objects with an associated relationship to form an initial spatial object relationship graph. Finally, the spatial unit is used as a node in the graph, and the boundary adjacency relationship and the title confirmation connection relationship are used as graph edges. The output form is G0 = (V0, E0), where V0 is the spatial unit node set and E0 is the adjacency and title confirmation connection edge set.

[0164] Step S42: introducing land element attributes in the land element attribute metadata table into the nodes in the initial topological graph structure of the geographic elements, and assigning spatial coupling weights to the edges in the graph to generate a geographic element topological graph;

[0165] In this embodiment, based on the G0 structure, the land attribute metadata table of each node introduces the land type code, area, utilization intensity, coverage type and other attribute values, and organizes them into a node attribute matrix X∈R nxd , n is the number of nodes, and d is the attribute dimension. For each pair of spatial units in edge E0, based on their similarity in attribute space and the degree of geographic coupling, the spatial coupling weight w_ij = exp(-λ1Δs_ij - λ2Δa_ij) is calculated, where Δs_ij is the spatial distance difference, Δa_ij is the attribute difference, and λ is a control parameter. After integrating all nodes and weighted edges, a structured geographic feature topology graph G1 = (V0, E1, X, W) is generated, where W is the weighted adjacency matrix. This graph provides a foundation for the joint representation of attribute and structural information for subsequent feature learning.

[0166] Step S43: performing structural learning on the geographic element topology map, extracting the high-dimensional spatial-semantic embedding representation of each land element, and generating an initial node embedding vector set;

[0167] In this embodiment, structural learning is performed on the G1 graph structure, and a three-layer graph neural structure learning framework is adopted. Each layer performs structural aggregation and attribute update respectively. The initial input is the node attribute matrix X, combined with the adjacency matrix W, and the structural context information of each node is obtained by weighted adjacency propagation. The calculation form of each layer structure update is: H^(l+1)=σ(WH^(l)Θ^(l)), where H^(0)=X, Θ is the transformation matrix to be learned, and σ is the nonlinear activation function. After three layers of propagation, a high-dimensional node representation vector H∈R is obtained. nxd ', d' is the output dimension, which represents the semantic embedding expression of each land feature under the coupling of spatial structure and attributes. Finally, the initial node embedding vector set H_node is generated for spatial clustering and aggregate mapping.

[0168] Step S44: performing spectral clustering and aggregate convolution operations on the initial node embedding vector set to obtain a land-class aggregated embedding vector map;

[0169] In this embodiment, spectral clustering is performed on the node embedding vector set H_node, a graph Laplacian matrix is ​​constructed based on L=DW, and the eigenvectors corresponding to the first k smallest eigenvalues ​​are extracted to form a spectral space mapping Z∈R nxk , where k is the number of clusters. Using Z as the embedding representation, we perform K-means spatial partitioning on all nodes to generate land-class cluster labels. We then embed the clustering results into a convolutional aggregation structure, constructing a graph convolutional aggregation function: H' = ∑(i∈N(v))W_ij·H_node(i). We integrate features of each cluster center node to improve intra-cluster consistency and output a new aggregated embedding vector atlas H_clustered, which reflects the joint distribution characteristics of land classes in both spatial and semantic dimensions.

[0170] Step S45: Perform position normalization and edge weight optimization reordering operations based on the land class aggregation embedded vector map to generate a land map embedded vector.

[0171] In this embodiment, the clustered embedding vector map H_clustered is mapped to a standardized spatial reference coordinate system, and all node positions are normalized so that all node positions p_i satisfy p_i∈[0,1] 2region, ensuring spatial comparability. A reordering-based optimization operation is performed on the edge weights Wij in the graph. A joint evaluation metric, Sij = α·cos_sim(Hi,Hj)+β·(1 / dij)+γ·path_strength(i,j), is constructed based on the spatial proximity between nodes, the cosine similarity of the embedding vectors, and the topological path tightness. Hi and Hj are the embedding vectors of nodes i and j, and cosine similarity is used to represent semantic proximity. dij is the Euclidean distance between nodes i and j, in meters. Path_strength(i,j) represents the path tightness from node i to j, defined as a metric reflecting the number of common neighbors between nodes in the topological structure. Parameter weights are set to α = 0.5, β = 0.3, and γ = 0.2, selected through cross-validation to balance semantic, spatial, and topological factors. The adjacent edge set is reordered by the S value, retaining the top eight strong edges and removing the remaining semantically significant edges. Weight normalization is then performed. Finally, a land map embedding vector set H_land with spatial consistency, semantic expression ability and structural compactness is generated, which serves as the input basis for subsequent land type identification and surveying map output.

[0172] Optionally, step S5 is specifically as follows:

[0173] Step S51: Acquire real-time multi-source sensor data of the target area, perform regional sensing fusion, and generate a real-time mapping sensing data packet;

[0174] In this embodiment, multiple types of IoT edge nodes deployed in the measurement area collect sensor data from sources such as GNSS receivers, remote sensing image sensors, ground-based laser radar (LiDAR), and weather stations. The sampling frequency of each sensor is set to update once every 5 seconds, and all collected data is uploaded to the local edge fusion node via the MQTT protocol. The edge node performs timestamp alignment, spatial coverage matching, and outlier removal on the multi-source data, and constructs a perception fusion tensor D based on the weighted spatiotemporal fusion function. f ∈R m×n×t , where m is the number of spatial grids, n is the dimension of the sensing indicator, and t is the number of time slices. The final output fusion tensor is encoded as a "real-time mapping perception data packet" for subsequent task map update processing.

[0175] Step S52: using the real-time mapping perception data packet to input the pre-trained dynamic graph attention scheduling model, performing node weight update and edge weight adjustment on the land map embedding vector to generate a task-sensitive map embedding vector;

[0176] In this embodiment, the constructed perception data packet is input into an attention scheduling framework based on a dynamic graph structure. The framework consists of a 3-layer temporal perception graph convolution structure and a 1-layer fully connected output structure. Each layer contains a dynamic node attention update function.i ∈R 128 and edge set weight w ij ∈R is the input, combined with the regional environment change index in the perception data tensor, the weight θ of each node is updated i , and recalculate the associated value of each edge The final task-sensitive graph embedding vector set H′∈R n×128 , dynamically adapt to the priority areas of surveying and mapping tasks under real-time environmental conditions.

[0177] Step S53: Based on the task-sensitive graph embedding vector and in combination with the acquired current surveying and mapping task list, a surveying and mapping task-area-resource mapping graph model is constructed;

[0178] In this embodiment, the current task list is imported from the surveying and mapping command center database, and the task set T={t1, t2, ..., t k Each task includes spatial scope, job type and priority index; at the same time, the spatial node set R={r1,r2,...,r k}, and combined with the available resource matrix A∈R of the surveying and mapping team k×p , construct a task-area-resource ternary mapping relationship graph. The graph model is based on triples (t i ,r j ,a k ) is the structural unit of the edge, and the node features include task embedding vector, region embedding vector and resource capability vector a k ∈R 16 , forming a three-layer graph network structure. The model supports adaptive adjustment of relationship strength based on task complexity and regional environmental changes to support downstream scheduling processes.

[0179] Step S54: performing scheduling path optimization and task priority sorting on the surveying and mapping task-area-resource mapping model to generate a land surveying and mapping instruction set;

[0180] In this embodiment, in the constructed three-layer task-area-resource mapping model, according to the task priority vector P∈R k , using inter-layer projection compression to map the graph structure to the scheduling relationship matrix M∈R k×u , where each row represents the path efficiency score between task t_i and u resource units. Further combined with the task urgency factor λ t∈[0.1,1.0] and resource response time threshold settings (for example, the maximum response delay does not exceed 8 minutes), filter out non-optimal paths in the M matrix, and select the shortest scheduling path set by score sorting. This path set is organized into a surveying and mapping task instruction structure, which contains information such as task ID, instruction target, execution node, time window, and geographic coverage, and is output as a land surveying and mapping instruction set.

[0181] Step S55: Encode the land surveying and mapping instruction set into a standardized land surveying and mapping instruction network, and send it to the corresponding operation unit through the IoT edge gateway node to execute the scheduling instruction replacement task.

[0182] In this embodiment, during the instruction issuing phase, the land surveying and mapping instruction set is translated into a unified format land surveying and mapping instruction network structure, which is defined as a directed graph G I =(V,E), where node V represents each task execution unit (such as a surveying vehicle, drone, or survey station), and edge E represents the task scheduling command link, complete with the command ID, target region code, and time window metadata. LoRaWAN and 5G dual-channel edge gateway nodes deployed at the region boundary transmit scheduling information from the command network to each execution unit via a lightweight TLS channel. Upon receiving the command, the execution unit switches tasks based on timestamps and priorities, returning status codes in real time. Using status feedback, the command network topology is dynamically modified, achieving an efficient closed-loop scheduling of surveying and mapping tasks.

[0183] It is particularly important that step S54 is specifically as follows:

[0184] Extract the task node, region node, and resource node information from the surveying and mapping task-region-resource mapping graph model, and calculate the edge weight attributes of each extracted node to obtain the task scheduling weight matrix;

[0185] In this embodiment, for the three-layer mapping diagram of surveying and mapping tasks, regions and resources, the task node set T = {t1, t2, ..., t n}、Region node set R={r1,r2,...,r m} and resource node set A={a1,a2,...,a p Each task node is accompanied by a surveying type (such as ownership survey, land classification update), a time window, and an operation accuracy requirement; each regional node records the center point coordinates, coverage area, and land classification attributes; and the resource node contains the equipment type, remaining operation capacity, and response delay. In the three-layer graph structure, the edge connection represents the scheduling possibility, and the edge weight W i,j Calculated based on three dimensions: job suitability ω1, geographic adjacency coefficient ω2 and equipment availability coefficient ω3, through the weighted linear combination form W i,j=α·ω1+β·ω2+γ·ω3, where α=0.4, β=0.3, γ=0.3 are empirical parameters. Finally, the task scheduling weight matrix M∈R is formed n×m×p , where each element represents the overall scheduling weight from a task to a region and then to a resource.

[0186] Perform task path search based on the task scheduling weight matrix to generate the optimal path set for task-region-resource scheduling;

[0187] In this embodiment, based on the three-dimensional task scheduling weight matrix M∈R generated in the previous step n×m×p , through multi-level path search, for each task node t i Select the corresponding maximum weight path (t i →r j →a k ), and obtain the path set with the best scheduling efficiency. Path selection is based on the following processing logic: Prioritize the weight value W i,j The largest regional node r j , and further screen the resource nodes a in this area whose response time is less than 5 minutes and whose operation capacity is greater than the current task requirements k Each final selected path is represented by a triplet π i =(t i ,r j ,a k ) represents and records the path scheduling score S i =W i,j,k , and the path scheduling set Π={π1,π2,...,π n}, this set serves as the basis for subsequent sorting and instruction generation.

[0188] Read the priority settings in the current surveying and mapping task list, perform task weighting sorting on the task-area-resource scheduling optimal path set, and obtain a surveying and mapping task priority list;

[0189] In this embodiment, the task node t corresponding to each task path is read from the scheduling path set π. i The priority vector P={p1,p2,...,p n}, the priority value range is [1,5], the smaller the value, the higher the priority. Combined with the path scheduling score S i Execute task weight sorting, and set the sorting rule to R i =λ·(1 / p i )+(1-λ)·S i , where λ = 0.6 is the task urgency weight factor. According to the weighted score R iReorder the path set to generate a priority-ordered task path list Π′={π′1,π′2,...,π′ n The first few paths in the list will be prioritized in the task scheduling process. The system will also dynamically adjust the order of the list based on the task load balance to ensure both fairness and efficiency in scheduling.

[0190] According to the surveying and mapping task priority list, task scheduling instructions are assigned to each path in the task-region-resource scheduling optimal path set to obtain a structured land surveying and mapping task instruction set;

[0191] In this embodiment, in the priority sorted path set Π′, a corresponding scheduling instruction structure is generated for each path, and each structure includes: a task identifier (such as T0003), a target area code (such as GZ-R112), a resource unit identifier (such as UAV-A17), a start time, a task duration, an operation accuracy requirement (centimeter level or sub-meter level) and an operation equipment configuration description. Each scheduling instruction is encapsulated using a structured JSON format and uniformly converted into a key-value pair format and stored in a Redis distributed cache. The structured task instruction set is indexed and arranged according to the task identifier when stored, which facilitates fast retrieval and incremental updates, and is seamlessly connected to the scheduling status management system. The system retains the original path and priority sorting records for the generation process of each scheduling instruction for scheduling optimization backtracking.

[0192] The structured land surveying and mapping task instruction set is encapsulated in a consistent instruction set structure to obtain a land surveying and mapping instruction set.

[0193] In this embodiment, the structured task instruction set is encapsulated and normalized to generate a land surveying and mapping instruction set in a unified format. Each instruction set is grouped in units of task days, and a hierarchical ordered structure is adopted internally. The first is the task metadata segment (task day, total number of operation areas, total execution time), followed by the scheduling details segment (including all scheduling paths and their time arrangements), and finally the device instruction segment (task assignment list for each device). The entire instruction set is constructed in a serializable instruction network format, which is logically expressed as a graph structure G = (V, E), where the node is the operation unit and the edge represents the instruction dependency or serial order. This format is convenient for batch distribution through the MQTT gateway or 5G edge device, and the execution status, response time and feedback data of each instruction can be tracked after distribution to ensure closed-loop management of tasks.

[0194] Optionally, this specification also provides a surveying and mapping data intelligent analysis and management system based on the Internet of Things, which is used to execute the surveying and mapping data intelligent analysis and management method based on the Internet of Things as described above. The surveying and mapping data intelligent analysis and management system based on the Internet of Things includes:

[0195] The regional perception fusion module is used to collect multi-source surveying and mapping data of the target area through the Internet of Things, and perform regional perception fusion based on the multi-source surveying and mapping data of the target area to obtain the original fusion data of the ground survey;

[0196] The semantic reconstruction module is used to perform multimodal semantic deconstruction on the original fusion data of land surveying to obtain the modal alignment semantic vector; based on the modal alignment semantic vector, the spatial-semantic-attribute surveying feature space is reconstructed to obtain the land semantic unit set;

[0197] The blockchain rights confirmation module is used to centralize the land semantic units into various spatial objects for blockchain data rights confirmation, thereby building a surveying and mapping data index chain; the surveying and mapping data index chain is used to perform on-chain rights confirmation of significant land elements to obtain an identification block on the surveying and mapping chain;

[0198] The topology analysis module is used to construct a geographic element topology map based on the identification blocks on the surveying and mapping chain, and use the land element topology map to extract the associated features of the land elements to obtain the land map embedding vector;

[0199] The instruction generation module is used to obtain real-time multi-source sensor data of the target area, and use the real-time multi-source data of the target area to intelligently schedule land surveying and mapping tasks on the land map embedding vector, thereby obtaining a land surveying and mapping instruction network; and uploading the land surveying and mapping instruction network to the Internet of Things to execute the scheduling instruction replacement.

[0200] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0201] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent analysis and management of surveying and mapping data based on the Internet of Things, characterized in that: The following steps are involved: Step S1: Collect multi-source surveying and mapping data of the target area through the Internet of Things, and perform regional perception fusion based on the multi-source surveying and mapping data of the target area to obtain the original fusion data of the ground survey; Step S2: Perform multimodal semantic deconstruction on the original fused land survey data to obtain a modality-aligned semantic vector; perform spatial-semantic-attribute mapping feature space reconstruction based on the modality-aligned semantic vector to obtain a land semantic unit set; Step S3: Concentrate the land semantic units into various spatial objects for blockchain data rights confirmation, thereby building a surveying and mapping data index chain; The surveying and mapping data index chain is used to confirm the ownership of significant land elements on the chain and obtain the identification block on the surveying and mapping chain; Step S4: construct a geographic element topology map based on the identification blocks on the surveying and mapping chain, and use the land element topology map to extract the land element correlation features to obtain the land map embedding vector; Step S5: Acquire real-time multi-source sensor data of the target area, and use the real-time multi-source data of the target area to intelligently schedule land surveying and mapping tasks on the land map embedding vector, thereby obtaining a land surveying and mapping instruction network; upload the land surveying and mapping instruction network to the Internet of Things to execute the scheduling instruction replacement.

2. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 1, characterized in that: Step S1 is specifically as follows: Step S11: collecting multi-source surveying and mapping data of the target area including spatial positioning data, remote sensing image data, surface elevation data, soil moisture data, and meteorological parameter data through the Internet of Things; Step S12: performing multi-source data format parsing and unified coding processing on the multi-source surveying and mapping data of the target area, and performing spatial registration and time series alignment based on a unified coordinate reference and time reference to generate a standard surveying and mapping data matrix; Step S13: performing noise identification and anomaly elimination processing on the standard surveying and mapping data matrix, and extracting a stable and effective surveying and mapping observation sequence to generate a high-confidence surveying and mapping data set; Step S14: construct a multimodal information channel based on the data type dimension of the high-confidence surveying and mapping dataset, and perform weighted mapping on the multimodal information channel to obtain a node space perception map; Step S15: Perform regional feature clustering and land type recognition based on the node spatial perception map, extract high-order regional perception feature vectors, and obtain original ground measurement fusion data.

3. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 2, characterized in that: Step S13 is specifically as follows: Step S131: discriminating abnormal surveying and mapping noise samples based on the temporal variation trend, spatial gradient and numerical distribution characteristics in the standard surveying and mapping data matrix to obtain an initial noise mask set; Step S132: performing fine-grained classification on abnormal noise samples in the standard surveying and mapping data matrix based on the initial noise mask set, and outputting a labeled noise removal index table; Step S133: performing regional weighted completion and credible interval residual correction on the abnormal points marked in the labeled noise removal index table to obtain a multi-dimensional observation repair matrix; Step S134: extracting an observation data subset based on the multidimensional observation repair matrix that simultaneously satisfies the conditions of temporal continuity ≤ 1 hour, spatial continuity ≤ 7 meters, change rate ≤ 8% within the past three time points, and sensor channel value difference ≤ 0.08; Step S135: Calculate the confidence score of each data point in the three dimensions of accuracy, stability and continuity in the observation data subset, and remove the data points whose average confidence score of the three dimensions is lower than the preset confidence score threshold to obtain a high-confidence mapping data set.

4. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 2, characterized in that: Step S15 is specifically as follows: Step S151: performing initial regional feature division based on the structural edge weights and spatial embedding relationships of the node spatial perception graph to obtain a set of spatial clustering units; Step S152: combining the multimodal node attributes in the spatial clustering unit set to construct a land class candidate attribute vector, and performing a priori remote sensing interpretation on the land class candidate attribute vector to determine the initial land class type, thereby obtaining a preliminary land class judgment map; Step S153: performing local feature compensation learning on the boundary area in the preliminary land class judgment map to generate a boundary optimized land class map; Step S154: extracting the structural center vector and characteristic statistical index of each land class unit according to the boundary optimized land class map, thereby constructing a regional land class aggregation feature set; Step S155: Perform high-order spatial semantic fusion and redundant feature compression on the regional land classification aggregation feature set to generate original land survey fusion data.

5. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 1, characterized in that: The multimodal semantic deconstruction in step S2 is specifically as follows: The original fusion data of ground survey is divided into spatial positioning sequence, remote sensing image channel, surface elevation model unit and ground object environment parameter set according to the data source type; Image texture features are extracted based on remote sensing image channels, and combined with the slope, aspect, and height difference attributes in the surface elevation model unit to construct a primary spatial semantic feature set. The trajectory distribution pattern in the spatial positioning sequence is used to perform spatial overlay analysis with land cover change points, extracting the semantic labels of the spatial overlay positions, and matching them with the preset geographic entity coding system to generate a candidate land class annotation set; Semantic feature alignment and fusion are performed on the primary spatial semantic feature set, the ground object environment parameter set and the candidate land class annotation set to generate a modality-aligned semantic vector.

6. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 5 is characterized in that: The spatial-semantic-attribute mapping feature space reconstruction in step S2 is specifically as follows: Mapping the spatial positioning information in the modality-aligned semantic vector to a unified coordinate base for geocoding and back-projection, thereby constructing a basic spatial unit grid. A semantic mapping grid is constructed using the basic spatial unit grid as the basic unit, and a location-semantic-attribute triple set is constructed based on the spatial texture characteristics, land use labels and environmental attribute indicators of the semantic mapping grid; Vectorized land class expression feature encoding is performed on the location-semantics-attribute triple set to obtain a set of spatial unit triples. Dimensionality reduction and reconstruction are then performed on the high-dimensional space to extract the principal component semantic distribution factor and generate a spatial-semantic feature matrix. Combining the terrain adjacency relationship, land cover similarity and environmental factor correlation between spatial units in the basic spatial unit grid, a surveying and mapping attribute association graph structure is constructed, and the attributes of each spatial unit are supplemented and the association is strengthened to obtain a surveying and mapping attribute enhancement structure; The spatial-semantic feature matrix and the mapping attribute enhancement structure are fused through high-dimensional embedding to obtain the land semantic unit set.

7. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 1, characterized in that: The specific confirmation of the significant land element chain in step S3 is as follows: Based on the land semantic units in the surveying and mapping data index chain, the land class code, spatial boundary and area index are extracted for each spatial unit to construct the land feature attribute metadata table; Perform multidimensional significance evaluation on the land feature attribute metadata table, screen out the land feature sets with average significance greater than [0.6, 0.85], and thus construct the significant land feature set; Mapping the collection of significant land features to the surveying and mapping data index chain structure, performing ownership confirmation, data source verification, and timestamp registration for each significant land feature, and generating an on-chain candidate pool for title confirmation transactions; The land elements in the candidate pool of on-chain title confirmation transactions are processed through a multi-node consensus verification mechanism to generate a surveying and mapping title confirmation certificate block; The surveying and mapping right confirmation certificate block is chain-bound with the surveying and mapping data index chain, the right confirmation status and identification index are marked, and an identification block on the surveying and mapping chain is generated.

8. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 1, characterized in that: Step S4 is specifically as follows: Step S41: Based on the boundary coordinates of the spatial objects in the identification blocks on the surveying and mapping chain and the ownership confirmation relationship, a topological adjacency relationship between the spatial units is established to generate an initial topological map structure of the geographic elements; Step S42: introducing land element attributes in the land element attribute metadata table into the nodes in the initial topological graph structure of the geographic elements, and assigning spatial coupling weights to the edges in the graph to generate a geographic element topological graph; Step S43: performing structural learning on the geographic element topology map, extracting the high-dimensional spatial-semantic embedding representation of each land element, and generating an initial node embedding vector set; Step S44: performing spectral clustering and aggregate convolution operations on the initial node embedding vector set to obtain a land-class aggregated embedding vector map; Step S45: Perform position normalization and edge weight optimization reordering operations based on the land class aggregation embedded vector map to generate a land map embedded vector.

9. The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 1, characterized in that: Step S5 is specifically as follows: Step S51: Acquire real-time multi-source sensor data of the target area, perform regional sensing fusion, and generate a real-time mapping sensing data packet; Step S52: using the real-time mapping perception data packet to input the pre-trained dynamic graph attention scheduling model, performing node weight update and edge weight adjustment on the land map embedding vector to generate a task-sensitive map embedding vector; Step S53: Based on the task-sensitive graph embedding vector and in combination with the acquired current surveying and mapping task list, a surveying and mapping task-area-resource mapping graph model is constructed; Step S54: performing scheduling path optimization and task priority sorting on the surveying and mapping task-area-resource mapping model to generate a land surveying and mapping instruction set; Step S55: Encode the land surveying and mapping instruction set into a standardized land surveying and mapping instruction network, and send it to the corresponding operation unit through the IoT edge gateway node to execute the scheduling instruction replacement task.

10. An intelligent analysis and management system for surveying and mapping data based on the Internet of Things, characterized in that: The method for intelligent analysis and management of surveying and mapping data based on the Internet of Things according to claim 1 is used to execute the method, the intelligent analysis and management system for surveying and mapping data based on the Internet of Things comprises: The regional perception fusion module is used to collect multi-source surveying and mapping data of the target area through the Internet of Things, and perform regional perception fusion based on the multi-source surveying and mapping data of the target area to obtain the original fusion data of the ground survey; The semantic reconstruction module is used to perform multimodal semantic deconstruction on the original fusion data of land surveying to obtain the modal alignment semantic vector; based on the modal alignment semantic vector, the spatial-semantic-attribute surveying feature space is reconstructed to obtain the land semantic unit set; The blockchain rights confirmation module is used to centralize the land semantic units into various spatial objects for blockchain data rights confirmation, thereby building a surveying and mapping data index chain; the surveying and mapping data index chain is used to perform on-chain rights confirmation of significant land elements to obtain an identification block on the surveying and mapping chain; The topology analysis module is used to construct a geographic element topology map based on the identification blocks on the surveying and mapping chain, and use the land element topology map to extract the associated features of the land elements to obtain the land map embedding vector; The instruction generation module is used to obtain real-time multi-source sensor data of the target area, and use the real-time multi-source data of the target area to intelligently schedule land surveying and mapping tasks on the land map embedding vector, thereby obtaining a land surveying and mapping instruction network; and uploading the land surveying and mapping instruction network to the Internet of Things to execute the scheduling instruction replacement.

Images