Natural resource dynamic monitoring and evaluation system based on multi-source spatio-temporal data fusion

Abstract

The invention relates to the technical field of natural resource monitoring and evaluation, and discloses a natural resource dynamic monitoring and evaluation system based on multi-source spatio-temporal data fusion, comprising a semantic mapping module for analyzing a business document to generate a constraint rule and constructing a basic spatio-temporal tensor; the perception strategy module monitors state node evolution and generates an acquisition strategy instruction to drive an external sensor to obtain enhanced perception data; the state evolution module deduces an expected legal state sequence and reconstructs an actual physical state sequence in combination with enhanced perception data; the energy quality fingerprint module establishes nonlinear mapping by using the time sequence energy quality data, and calculates a fingerprint consistency index to verify the authenticity of a physical state; and the deviation risk module calculates a space-time semantic distance and a propagation risk value based on the topological network in combination with the consistency index, and generates a resource audit report. According to the invention, through event-driven active perception and cross-modal energy quality implicit verification, the problems of insufficient monitoring data space-time precision and difficult camouflage compliance identification are effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural resource monitoring and evaluation, in particular to a natural resource dynamic monitoring and evaluation system based on multi-source spatio-temporal data fusion. BACKGROUND

[0002] The existing natural resource monitoring and evaluation technical solution usually adopts a satellite remote sensing image and vector data comparison method, and the mainstream technical solution usually uses an optical satellite or an unmanned aerial vehicle to obtain surface images according to a fixed revisit period, extracts a ground feature map spot through an image recognition algorithm, and performs spatial superposition analysis on the extracted ground feature map spot and preset planning data to find a region where the surface state is inconsistent with the planning data. However, the existing technology has a contradiction between monitoring timeliness and data acquisition cost in actual application.

[0003] Due to the passive and fixed frequency monitoring mode, the system cannot perceive the key nodes of the state evolution of the land plot, so that high-resolution data is often missing due to the mismatch of satellite overflight time or cloud cover in the construction mutation period or the key rectification period, and a large amount of redundant data is generated in the long period of stable state of the land plot, so that it is difficult to balance the global coverage and the spatio-temporal accuracy of the key nodes under the limited observation resources.

[0004] In addition, the existing monitoring means excessively relies on the optical visual features of the surface cover to determine the compliance. In actual supervision, there are phenomena of covering the traces of illegal mining or illegal land occupation by means of spraying paint to restore green, laying dustproof nets or transplanting shallow-rooted vegetation. The existing technology only compares the image texture and spectral features, and cannot penetrate the surface camouflage to verify the actual production activity intensity and material flow logic inside the land plot, so that the identification rate of the disguised illegal behavior is low, and false judgments are easily made.

[0005] At the same time, the existing evaluation method is limited to geometric overlap analysis of a single spot, and lacks correlation risk evaluation based on natural geographical processes. The existing system usually regards each monitoring spot as an isolated geometric object, only calculates the illegal area or type of the spot itself, and does not consider the geographical topological relationship such as hydrological flow, atmospheric diffusion or biological migration, so that it is impossible to quantify the collateral ecological risk or cumulative influence of the illegal behavior of a single land plot on the downstream or surrounding areas, and it is difficult to meet the management needs of the natural resource whole-element system governance. SUMMARY

[0006] In view of the deficiencies of the existing technology, the present application provides a natural resource dynamic monitoring and evaluation system based on multi-source spatio-temporal data fusion, which solves the problems of missing data of key spatio-temporal nodes caused by the passive monitoring mode of the existing technology, difficulty in identifying surface camouflage by simply relying on visual features, and lack of collateral risk evaluation based on geographical topology.

[0007] This invention provides a dynamic monitoring and assessment system for natural resources based on multi-source spatiotemporal data fusion. The system includes a semantic mapping construction module, a perception strategy generation module, a state evolution modeling module, an energy quality fingerprint verification module, and a deviation risk analysis module.

[0008] The semantic mapping construction module is configured to access the business license database and multi-source basic geographic data. This module generates a structured set of constraint rules by parsing business license documents and maps the multi-source basic geographic data to a basic spatiotemporal tensor.

[0009] In its implementation, this module uses natural language processing algorithms to parse business license documents, extract spatiotemporal constraint parameters including time windows, spatial red line coordinates, and floor area ratio limits, and then combines these parameters to generate a set of constraint rules. Meanwhile, this module performs spatial discretization and attribute mapping on multi-source basic geographic data, which includes vector redline data and basic remote sensing image rasters, and maps them to a unified four-dimensional sparse tensor space to construct a basic spatiotemporal tensor.

[0010] The perception strategy generation module is configured to monitor the evolution of state nodes output by the state evolution modeling module, and generate acquisition strategy instructions based on the state node evolution to drive the external sensor network to acquire enhanced perception data, and transmit the enhanced perception data to the state evolution modeling module.

[0011] Furthermore, the module monitors the evolution of state nodes in real time and receives the current state judgment confidence level. It identifies key state transition nodes based on the accessed external business state transition template and determines the state ambiguity node when the current state judgment confidence level is lower than the preset safety threshold.

[0012] This module increases the attention weight value of the spatial region corresponding to the above nodes to generate a thermal distribution, and determines the current state deviation type based on the evolution of the state nodes. It then combines the thermal distribution to generate acquisition strategy instructions that include geometric texture acquisition instructions or spectral attribute acquisition instructions.

[0013] The state evolution modeling module is configured to receive a set of constraint rules, a basic spatiotemporal tensor, and enhanced sensing data, and perform dual-track state reconstruction. This module deduces the expected legal state sequence based on the constraint rule set, determines the evolution of state nodes based on the expected legal state sequence and the basic spatiotemporal tensor, and reconstructs the actual physical state sequence by combining the basic spatiotemporal tensor and enhanced sensing data.

[0014] In its implementation, this module maps the parameters in the constraint rule set to the time axis based on a preset business logic template in order to deduce the expected sequence of legal states. Simultaneously, the module updates the enhanced sensing data to the corresponding spatiotemporal voxels of the basic spatiotemporal tensor, generates a state probability distribution vector based on the classification of the basic spatiotemporal tensor, generates an actual physical state sequence using temporal coherence filtering, and calculates the current state interpretation confidence based on the state probability distribution vector.

[0015] The energy quality fingerprint verification module is configured to receive the actual physical state sequence and access the time-series energy quality data, and calculate the fingerprint consistency index by establishing a nonlinear mapping relationship between the actual physical state sequence and the time-series energy quality data.

[0016] Specifically, this module accesses and monitors time-series energy quality data associated with the monitored plots, including industrial electricity consumption, logistics vehicle traffic, signaling density, and production water consumption, and performs normalization processing to construct non-visual data vectors.

[0017] This module extracts actual visual feature vectors from actual physical state sequences, uses a nonlinear mapping model pre-trained based on historical samples to map non-visual data vectors to theoretical visual features, calculates the Euclidean distance between actual visual feature vectors and theoretical visual features, and uses the hyperbolic tangent function to normalize the Euclidean distance, thereby calculating the fingerprint consistency index, which represents the logical consistency between visual features and energy quality data.

[0018] The deviation risk analysis module is configured to receive the expected legal state sequence, the actual physical state sequence, and the fingerprint consistency index, calculate the spatiotemporal semantic distance between the expected legal state sequence and the actual physical state sequence based on the fingerprint consistency index, and calculate the propagation risk value accordingly to generate a resource audit report.

[0019] In its specific implementation, this module performs spatiotemporal semantic distance calculation. Within the evaluation time window, it calculates the difference in state categories and the difference in patch area between the expected legal state sequence and the actual physical state sequence. The differences are then weighted and summed, and the fingerprint consistency index is used as a reverse weighting factor to correct the weighted summation result, thus obtaining the spatiotemporal semantic distance.

[0020] When the spatiotemporal semantic distance exceeds the preset compliance threshold, the module establishes a topological network between land parcels based on the accessed external geographic information system, identifies the set of risk source land parcels that affect downstream or neighboring land parcels, calculates the topological transmission probability value from the risk source land parcels to downstream or neighboring land parcels, and calculates the propagation risk value by combining the initial violation degree value of the risk source land parcels with the distance attenuation factor.

[0021] Ultimately, the module combines land value data obtained from external databases to generate a resource audit report that includes compliance status determination, confidence assessment report, and prediction of potential associated risks.

[0022] This invention provides a dynamic monitoring and assessment system for natural resources based on multi-source spatiotemporal data fusion. It has the following beneficial effects: 1. This invention generates acquisition strategy instructions to drive an external sensor network by using a perception strategy generation module based on the evolution of state nodes output by the state evolution modeling module. This feedback control mechanism changes the traditional passive acquisition mode with a fixed frequency across the entire domain, enabling the automatic acquisition of enhanced perception data in critical time windows or areas with ambiguous data. This ensures the quality of data acquisition for key stages throughout the entire life cycle of natural resources while controlling monitoring costs.

[0023] 2. This invention incorporates time-series energy quality data such as industrial electricity consumption and logistics vehicle traffic through an energy quality fingerprint verification module, and calculates a fingerprint consistency index using a nonlinear mapping model. This mechanism establishes a logical correlation verification between physical state (visual features) and energy quality data (non-visual features). For plots of land that exhibit compliant visual characteristics through surface camouflage (such as revegetation by painting or dust netting) but have abnormal energy quality data, the system can identify logical violation risks through the calculated low fingerprint consistency index, reducing the misjudgment rate caused by relying solely on optical remote sensing imagery.

[0024] 3. This invention calculates the spatiotemporal semantic distance through a deviation risk analysis module and introduces a fingerprint consistency index as a reverse weighting factor to correct deviations, objectively measuring the degree of violation. Simultaneously, the system integrates with an external geographic information system to construct a topological network and calculates the propagation risk value of the risk source plot to downstream or neighboring plots. This assessment method expands independent map patch comparison into regional risk transmission analysis, enabling the output of resource audit reports that include both the degree of direct violation and the indirect propagation impact, providing a quantitative basis for natural resource management. Attached Figure Description

[0025] Figure 1 This is a diagram illustrating the architecture of the natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to the present invention. Figure 2 This is a diagram illustrating the architecture of the energy quality fingerprint verification module of the present invention. Figure 3 This is a diagram of the deviation risk analysis module architecture of the present invention.

[0026] Among them, 100 is the semantic mapping construction module; 200 is the perception strategy generation module; 300 is the state evolution modeling module; 400 is the energy quality fingerprint verification module; and 500 is the deviation risk analysis module. Detailed Implementation

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see the appendix Figure 1 This invention provides a dynamic monitoring and assessment system for natural resources based on multi-source spatiotemporal data fusion, including a semantic mapping construction module 100, a perception strategy generation module 200, a state evolution modeling module 300, an energy quality fingerprint verification module 400, and a deviation risk analysis module 500.

[0029] The semantic mapping construction module 100 serves as the system's data benchmark construction end, establishing communication connections with external business license databases and multi-source basic geographic data sources.

[0030] The semantic mapping construction module 100 is responsible for parsing business license documents to extract spatiotemporal constraint parameters and generating a structured set of constraint rules. Simultaneously, the semantic mapping construction module 100 maps multi-source basic geographic data to a unified four-dimensional space, constructing a basic spatiotemporal tensor. The semantic mapping construction module 100 then transmits the generated set of constraint rules and the basic spatiotemporal tensor to the state evolution modeling module 300.

[0031] The perception strategy generation module 200 serves as the system's feedback control link and is connected to the state evolution modeling module 300 and the external sensor network.

[0032] The perception strategy generation module 200 monitors the evolution of state nodes output by the state evolution modeling module 300 in real time. When a critical state transition node or a state ambiguity node is detected, the perception strategy generation module 200 generates differentiated acquisition strategy instructions based on the state deviation type and drives the external sensor network to acquire enhanced perception data. The perception strategy generation module 200 then sends the acquired enhanced perception data back to the state evolution modeling module 300.

[0033] The state evolution modeling module 300 receives a set of constraint rules and a basic spatiotemporal tensor from the semantic mapping construction module 100, as well as enhanced perception data from the perception policy generation module 200.

[0034] The state evolution modeling module 300 deduces and generates a expected legal state sequence based on the constraint rule set, and reconstructs and generates an actual physical state sequence by combining the basic spatiotemporal tensor and enhanced sensing data. The state evolution modeling module 300 transmits the expected legal state sequence and the actual physical state sequence to the deviation risk analysis module 500, and transmits the actual physical state sequence to the energy quality fingerprint verification module 400.

[0035] The energy quality fingerprint verification module 400 receives the actual physical state sequence from the state evolution modeling module 300 and accesses the time-series energy quality data associated with the land parcel.

[0036] The energy quality fingerprint verification module 400 processes time-series energy quality data to construct a non-visual data vector and extracts actual visual feature vectors from the actual physical state sequence. The energy quality fingerprint verification module 400 executes the fingerprint consistency index formula to calculate the fingerprint consistency index.

[0037] The formula for the fingerprint consistency index is as follows: ; In the formula: The fingerprint consistency index represents the logical consistency between visual features and energy quality data. This is the symbol for the hyperbolic tangent function, used to map the calculation result to a specific interval; This is the sensitivity adjustment coefficient, a constant used to control the magnitude of the effect of the difference value on the index; The square sign of the Euclidean norm, representing the distance metric between vectors; These are actual visual feature vectors, derived from features extracted from remote sensing images; This is a nonlinear mapping function symbol, representing the theoretical mapping relationship from the non-visual data space to the visual feature space; These are non-visual data vectors derived from energy metabolism data.

[0038] The energy quality fingerprint verification module 400 transmits the calculated fingerprint consistency index to the deviation risk analysis module 500.

[0039] The deviation risk analysis module 500 receives the expected legal state sequence and the actual physical state sequence from the state evolution modeling module 300, as well as the fingerprint consistency index from the energy and quality fingerprint verification module 400. The deviation risk analysis module 500 executes the spatiotemporal semantic distance formula to calculate the accumulated spatiotemporal semantic distance.

[0040] The formula for spatiotemporal semantic distance is as follows: ; In the formula: This represents the accumulated spatiotemporal semantic distance; a larger value indicates a higher degree of violation or anomaly. The summation operator is used to accumulate values ​​within a specified range. The total length of the evaluation time window; The weighting coefficients for the differences in state categories; This is the symbol for the state difference indicator function. It outputs one when the two input state categories are inconsistent, and zero otherwise. For at any time The expected state category; For at any time The actual state category; This represents the weighting coefficient for area differences; The absolute value symbol indicates that the value is a non-negative value. For at any time The expected area value of the patch; For at any time The actual area value of the map patch; For at any time The fingerprint consistency index.

[0041] When the calculated spatiotemporal semantic distance exceeds the preset compliance threshold, the deviation risk analysis module 500 executes the topology propagation risk formula based on the topology network between plots to calculate the propagation risk value.

[0042] The formula for topology propagation risk is as follows: ; In the formula: downstream or adjacent plots The cumulative risk value of transmission received; For the land parcel The collection of all upstream or associated risk source sites that have an impact; Risk source plot The initial violation level value; For the plot of land To the plot The topological propagation probability value; It is the base of the natural logarithm constant; This is the distance attenuation coefficient; For the plot of land With the plot of land The topological path distance between them.

[0043] The Deviation Risk Analysis Module 500 combines spatiotemporal semantic distance with propagation risk value to generate and output resource audit reports.

[0044] The semantic mapping construction module 100 acts as a bridge connecting physical space data and business management rules, and is responsible for building the digital benchmarks required for subsequent analysis. The operation process can specifically include the following logical steps: The semantic mapping construction module 100 performs structured extraction of business rules through the text rule parsing unit. This process first involves accessing and traversing the business license database.

[0045] The business license documents stored in the business license database are typically in unstructured or semi-structured form, including but not limited to legal documents such as mining licenses, construction land planning permits, land reclamation plans, and ecological restoration acceptance reports. These documents contain the core legal basis for controlling natural resource land parcels.

[0046] The semantic mapping construction module 100 preprocesses the acquired business license documents. For scanned documents, optical character recognition (OCR) technology is used to convert them into editable text data; for electronic documents, the text content is extracted directly.

[0047] Based on this, text cleaning is performed to remove formatting symbols, headers, footers, and irrelevant official jargon from the document, retaining text paragraphs containing substantive content. The specific algorithmic implementations for OCR recognition and text cleaning are standard techniques in the field of natural language processing and will not be elaborated upon here.

[0048] The semantic mapping construction module 100 utilizes natural language processing algorithms to extract entities and relations from the cleaned text, with the core objective of extracting spatiotemporal constraint parameters. The specific extraction logic is as follows: To address time constraints, the system identifies key time nodes in the document, such as "validity start date," "validity end date," and "construction rectification period," and standardizes them into a unified timestamp format to form time window parameters. To address spatial constraints, the system identifies coordinate lists describing geographical areas in documents, or textual descriptions of specific regions (such as "located 50 meters east of the river centerline"), and converts them into standard spatial redline coordinates. To address the constraints of the indicator dimensions, the system extracts numerical indicators such as "maximum floor area ratio", "maximum mining elevation", and "minimum reclamation and greening rate".

[0049] The semantic mapping construction module 100 associates and combines the extracted time window, spatial redline coordinates and various indicator parameters to generate a structured set of constraint rules.

[0050] This set of constraint rules is stored using a standardized data structure (such as JSON or XML format). Each rule object corresponds to a specific natural resource plot and explicitly marks the physical state constraints that the plot should follow within a specific time period. For example, the constraint rule set can explicitly record that a mining plot should be in the "infrastructure construction" state at a certain period, and its excavation range should not exceed a specific red line coordinate sequence.

[0051] Meanwhile, the semantic mapping construction module 100 performs tensor mapping of physical spatial data through spatiotemporal tensor construction units. This process aims to address the inconsistencies in spatial reference and data structure between multi-source heterogeneous geographic data.

[0052] The semantic mapping construction module 100 receives multi-source basic geographic data. This data includes vector data such as administrative boundaries and basic farmland protection red lines, as well as raster data such as satellite orthophotos and digital elevation models (DEMs). To address the projection differences between data from different sources, the module uses a seven-parameter coordinate transformation model or affine transformation algorithm to uniformly reproject the multi-source data to the system's preset standard geographic coordinate system (such as CGCS2000).

[0053] The semantic mapping construction module 100 performs spatial discretization and attribute mapping. The system uses voxelization technology to divide continuous geographic space into spatial units of fixed size. Specifically, the system sets the planar resolution (e.g., 2 meters × 2 meters) and elevation resolution of the spatial units.

[0054] For vector data, the system uses a scan line filling algorithm or an inclusion detection algorithm to determine whether each spatial unit is located within the vector red line range, thereby assigning the corresponding management attribute label (such as "permitted construction area" or "ecological protection area") to the spatial unit. For basic remote sensing image raster, bilinear interpolation or nearest neighbor interpolation is used to extract the spectral and elevation values ​​at the corresponding coordinates and assign them to the corresponding spatial units.

[0055] Semantic mapping building module 100 in three-dimensional space coordinates Introducing a time axis (based on) The system constructs a unified four-dimensional sparse tensor space. It stacks the processed spatial units with attributes according to a time sequence to generate a basic spatiotemporal tensor. Each element (voxel) in this basic spatiotemporal tensor contains its spatial location information, physical attribute information (such as spectral reflectance and elevation), and administrative attribute information (such as whether it is located within the planned boundary line) at a specific time.

[0056] By constructing sparse tensor structures, data storage space can be effectively compressed, and only spatiotemporal nodes with valid data can be stored and computed.

[0057] Finally, the semantic mapping construction module 100 outputs the generated constraint rule set and basic spatiotemporal tensor. The constraint rule set provides the logical legal basis for subsequent state deduction, while the basic spatiotemporal tensor provides the data-level spatiotemporal framework for subsequent physical state reconstruction. Together, they constitute the benchmark data environment for the system's closed-loop management.

[0058] The perception strategy generation module 200 is used to dynamically adjust the front-end data acquisition strategy based on the state evolution analysis results. The specific operation process and internal logic are as follows: The perception strategy generation module 200 performs real-time calculation and classification of the attention level of the monitoring area through the thermal dynamic reconstruction unit. This thermal dynamic reconstruction unit maintains real-time communication with the state evolution modeling module 300, reading the state machine operating parameters of each monitored plot at the current moment.

[0059] The thermal dynamic reconfiguration unit first executes the capture logic for key state nodes. This unit has a built-in business state transition template, which can identify key state transition nodes in the life cycle of a land parcel.

[0060] A critical state transition node refers to the moment when the physical attributes of a land parcel are about to undergo a qualitative change, such as the node that transitions from "under construction" to "earthwork construction", or from "mining" to "ecological restoration".

[0061] When a plot of land is detected to be within the time window of the aforementioned transition node, the thermal dynamic reconstruction unit extracts the spatial coordinates of the plot and prepares to perform thermal weighting.

[0062] In addition, the thermal dynamic reconstruction unit also executes the capture logic for fuzzy state nodes. When the confidence level of the current state judgment of a certain plot is lower than the preset safety threshold, the plot is marked as a fuzzy state node.

[0063] Here, the current state judgment confidence is calculated by the physical trajectory reconstruction unit in the state evolution modeling module 300 when performing time coherence filtering.

[0064] Specifically, when the physical trajectory reconstruction unit classifies the basic spatiotemporal tensor at the current moment, it generates a state probability distribution vector (e.g., [woodland: 0.4, bare soil: 0.3, grassland: 0.3]). The confidence level of the current state judgment is the maximum probability value in this probability distribution vector (such as 0.4 in the previous example), or it can be defined as the stability coefficient of the state sequence within the sliding time window (i.e., the frequency ratio of the same state appearing within the window).

[0065] The preset safety threshold is a value pre-set based on the statistical distribution of historical classification accuracy of basic remote sensing data (for example, it is set to 0.75, which means that if the maximum classification probability is less than 75%, the machine interpretation result is considered unreliable).

[0066] Once a critical state transition node or a state ambiguity node is captured, the thermal dynamic reconstruction unit maps the spatial coordinates of these plots to the system's global grid map and increases their corresponding attention weight value (for example, increasing the default weight of 1.0 to a high priority weight of 5.0).

[0067] The thermal dynamic reconstruction unit performs spatial smoothing on the attention weights of all plots across the entire region, ultimately generating a thermal distribution (i.e., high-attention thermal zones). This thermal distribution intuitively quantifies the urgency of acquiring enhanced sensing data for each spatial region at the current moment and is transmitted as input to the collaborative scheduling generation unit.

[0068] The perception strategy generation module 200 utilizes the collaborative scheduling generation unit to generate specific control commands based on thermal distribution and deviation types. This collaborative scheduling generation unit receives a list of high-interest areas from the thermal dynamic reconfiguration unit and further analyzes the specific reasons or potential deviation types that lead to increased interest in these areas.

[0069] The collaborative scheduling generation unit executes differentiated sensor scheduling logic. Internally, this unit maintains a sensor capability mapping table, which pre-defines the correspondence between different physical state deviation types and sensor parameters.

[0070] Specifically, if system analysis reveals a potential deviation type of geometric boundary overflow, meaning there is a spatial conflict between the terrain outline and the planning boundary line, the collaborative scheduling generation unit generates a geometric texture acquisition command. This command specifies the mobilization of optical satellites or low-altitude unmanned aerial vehicle systems with high spatial resolution to acquire images of the target area.

[0071] If system analysis indicates potential deviations such as attribute anomalies or abnormal vegetation changes—for example, abnormal fluctuations in vegetation indices detected in areas that should be revegetated, or suspected camouflage netting—then spectral resolution takes precedence over spatial resolution. The collaborative scheduling generation unit then generates a spectral attribute acquisition command. This command instructs the activation of an observation platform equipped with a hyperspectral imager or thermal infrared sensor to acquire multi-band spectral data or thermal radiation data of the target area, in order to utilize spectral characteristics to invert the true surface material.

[0072] The collaborative scheduling generation unit encodes and distributes the generated acquisition strategy instructions. These instructions include precise spatial coordinates (bounding box or polygon vertices), the required time window (e.g., the moment of optimal lighting conditions), and specified sensor configuration parameters (e.g., yaw angle, band combination). These instructions are transmitted via a communication interface to an external sensor network or flight control system.

[0073] After the external sensor network completes data acquisition in response to the above instructions, the perception strategy generation module 200 receives the returned enhanced perception data. This data, after preprocessing, is directly transmitted back to the physical trajectory reconstruction unit of the state evolution modeling module 300.

[0074] In this way, the system can automatically acquire higher-quality data during periods of high risk or ambiguous status, thereby improving the monitoring accuracy and status confirmation capability of key plots without significantly increasing the cost of full-area monitoring.

[0075] The state evolution modeling module 300, as the core deduction engine of the system, is responsible for constructing a dual-track benchmark for compliance comparison: the expected legal state sequence representing the logic of administrative licensing and the actual physical state sequence representing objective facts. This state evolution modeling module 300 achieves the transformation from legal text and physical images to standardized state sequences through forward rule mapping and reverse data reconstruction.

[0076] The state evolution modeling module 300 performs rule-based forward logic deduction through the expected trajectory deduction unit. This expected trajectory deduction unit receives a set of constraint rules from the semantic mapping construction module 100. The constraint rule set includes structured time windows, spatial ranges, and control indicators.

[0077] The expected trajectory deduction unit contains a pre-set business logic template for natural resource land parcels. This template defines the legal state transition paths and triggering conditions for different types of land parcels throughout their entire lifecycle. For example, for a legal open-pit mine land parcel, its lifecycle typically strictly follows a unidirectional or cyclical transition logic of "original landform state, infrastructure stripping state, mining operation state, ecological restoration state, and completion and acceptance state." For construction land parcels, the evolution path follows: "pre-approval land classification, construction with fencing, main construction, and hardening completion".

[0078] The expected trajectory deduction unit maps specific parameters (such as "work commencement permit date" and "repair completion deadline") in the constraint rule set to the business logic template, and instantiates the theoretical state category that should correspond to each discrete time step on the time axis.

[0079] Through this parameterized mapping, the system generates a sequence of expected legal states covering the entire monitoring period. This sequence is a standard set of state labels with time indexes, explicitly specifying the state at any given time. The physical appearance that the target plot should present in terms of law constitutes the logical benchmark for subsequent deviation calculations.

[0080] Meanwhile, the state evolution modeling module 300 performs inverse state reconstruction based on observation data using the physical trajectory reconstruction unit. The data input of the physical trajectory reconstruction unit consists of two parts: one part is the basic spatiotemporal tensor provided by the semantic mapping construction module 100, and the other part is the enhanced perception data returned by the perception policy generation module 200.

[0081] The basic spatiotemporal tensor provides macroscopic background data covering the entire domain, while the enhanced sensing data provides high-frequency, high-resolution detailed data acquired at key state nodes or high-risk areas. The physical trajectory reconstruction unit first performs a multi-source data fusion operation, registering and updating the enhanced sensing data to the corresponding spatiotemporal voxels of the basic spatiotemporal tensor, thereby improving the signal-to-noise ratio and resolution of the voxel physical properties within a specific spatiotemporal domain.

[0082] Based on this, the physical trajectory reconstruction unit processes the fused time-series data using temporal coherence filtering techniques. Specifically, the physical trajectory reconstruction unit employs either a sliding window weighted filtering algorithm or a Bayesian state update algorithm.

[0083] Taking a sliding window as an example, for time... State determination, physical trajectory reconstruction unit selection The time window is used to calculate the frequency distribution of image classification results at each time point within the window.

[0084] If time If the classification result is an isolated mutation state (such as "woodland, bare soil, woodland"), and the proportion of this state in the window is lower than the preset confidence threshold, then it is judged as cloud cover or temporary noise, and corrected according to the dominant state in the window. The system confirms a transition in a physical state only when a certain state remains stable over multiple consecutive time steps and conforms to the laws of physical evolution (e.g., an irreversible transition from "vegetation" to "hardened ground"). After the above processing, the physical trajectory reconstruction unit outputs the corrected sequence of actual physical states. ).

[0085] See appendix Figure 2 The energy quality fingerprint verification module 400, as an implicit verification link in the system, aims to solve the technical problem of difficulty in identifying camouflage behaviors (such as dust netting and spray painting for revegetation) in traditional remote sensing monitoring. By introducing non-visual elements that are strongly correlated with the physical state of the land surface, the energy quality fingerprint verification module 400 constructs a second verification dimension independent of the image spectral features, realizing the logical verification of the authenticity of the "actual physical state sequence".

[0086] The energy quality fingerprint verification module 400 utilizes a non-visual element mapping unit to access and process multi-dimensional time-series energy quality data. Time-series energy quality data refers to time-series data that can indirectly reflect the intensity of land production activities or the state of material flow, specifically including: The data includes industrial electricity consumption data from the substation associated with the target site, which reflects the operating status of large machinery and equipment; logistics vehicle flow data based on location data, which reflects the transportation intensity of ore, earthwork, or building materials; signaling density data obtained from mobile communication base stations, which reflects the concentration of human activity in the area; and production water consumption data.

[0087] Because the above data differ in units, sampling frequency and physical meaning, the non-visual element mapping unit first performs data cleaning and spatiotemporal alignment operations to map the heterogeneous data into the same time window as the remote sensing observation.

[0088] Subsequently, the non-visual element mapping unit performs range transformation or Z-score normalization to eliminate the influence of dimensions, and combines the processed energy quality indicators to construct a non-visual data vector. The energy fingerprint verification module 400 performs cross-modal logical consistency determination through the behavioral confidence evaluation unit. The non-visual feature mapping unit receives a non-visual data vector and a real visual feature vector extracted from the actual physical state sequence.

[0089] Actual visual feature vector ( The data is a high-dimensional feature representation extracted from remote sensing imagery, including visual attributes such as spectral reflectance, texture roughness, and vegetation index (NDVI) of ground features. The behavioral confidence assessment unit uses a nonlinear mapping model to establish causal relationships between non-visual data and visual features.

[0090] This nonlinear mapping model (denoted as) The deep neural network (such as a multilayer perceptron or graph neural network) is generated based on historical samples pre-training. The pre-training process is as follows: collect data of compliant land parcels that have been confirmed within the historical time period, construct training sample pairs containing (non-visual data vectors and actual visual feature vectors), and train the neural network using supervised learning so that it learns the mapping relationship from energy quality data to visual features.

[0091] The function of this nonlinear mapping model is to map the input energy metabolism data to the visual feature space and predict the theoretical visual features that the land parcel should present at that energy level.

[0092] The behavioral confidence assessment unit calculates the Euclidean distance between the visual features actually observed and the theoretical features predicted based on energy quality data, and transforms it into a standardized fingerprint consistency index using the fingerprint consistency index formula.

[0093] The formula for the fingerprint consistency index is as follows: ; In the formula: The fingerprint consistency index represents the logical consistency between visual features and energy quality data. This is the symbol for the hyperbolic tangent function, used to map the calculation result to a specific interval; This is the sensitivity adjustment coefficient, a constant used to control the magnitude of the effect of the difference value on the index; The square sign of the Euclidean norm, representing the distance metric between vectors; These are actual visual feature vectors, derived from features extracted from remote sensing images; This is a nonlinear mapping function symbol, representing the theoretical mapping relationship from the non-visual data space to the visual feature space; These are non-visual data vectors derived from energy metabolism data.

[0094] The fingerprint consistency index output by the behavioral confidence assessment unit will be passed as a confidence weight to the subsequent deviation risk analysis module 500.

[0095] when When the value approaches 1, it indicates that the visual state of the Earth's surface is logically consistent with its internal energy and matter activities, and the observation results are reliable. when A significant decrease indicates a risk of feigned compliance or data anomalies, and the system will amplify the regulatory weight of the land parcel in subsequent deviation calculations.

[0096] See appendix Figure 3 The Deviation Risk Analysis Module 500, serving as the system's decision-making and auditing hub, is responsible for comprehensively evaluating the intermediate results generated by the preceding modules. It transforms technical-level state differences and confidence indicators into compliance judgments and risk assessments with administrative significance. This Deviation Risk Analysis Module 500 integrates multiple analysis mechanisms, including spatiotemporal semantic comparison, implicit feature weighting, and geographic topology deduction, achieving a leap from single-point monitoring to regional risk management.

[0097] The deviation risk analysis module 500 performs core compliance calculations through the state deviation measurement unit. This state deviation measurement unit receives the expected legal state sequence and the actual physical state sequence output by the state evolution modeling module 300, as well as the fingerprint consistency index output by the energy quality fingerprint verification module 400.

[0098] The state deviation measurement unit is not limited to static comparison at a single moment, but is based on the dynamic time warping algorithm (DTW) or sliding window mechanism to calculate the cumulative difference between two state sequences within a continuous time window.

[0099] The calculation logic considers not only whether the state categories are consistent (qualitative difference) but also the magnitude of the deviation in patch area (quantitative difference), and introduces the fingerprint consistency index as a reverse weighting factor (penalty term).

[0100] The core purpose of this design is to: when energy fingerprint verification shows a low confidence level in the observation results (i.e., When the difference in visual state appears small (indicating potential spoofing), the system will amplify the final semantic distance value through the denominator effect, thereby triggering a high-level violation warning and effectively preventing false compliance behavior from evading supervision.

[0101] The state deviation metric unit specifically uses a spatiotemporal semantic distance formula to quantify this comprehensive deviation. The formula is as follows: ; In the formula: This represents the accumulated spatiotemporal semantic distance; a larger value indicates a higher degree of violation or anomaly. The summation operator is used to accumulate values ​​within a specified range. The total length of the evaluation time window; The weighting coefficients for the differences in state categories; This is the symbol for the state difference indicator function. It outputs one when the two input state categories are inconsistent, and zero otherwise. For at any time The expected state category; For at any time The actual state category; This represents the weighting coefficient for area differences; The absolute value symbol indicates that the value is a non-negative value. For at any time The expected area value of the patch; For at any time The actual area value of the map patch; For at any time The fingerprint consistency index.

[0102] When the calculated cumulative spatiotemporal semantic distance exceeds a preset compliance threshold, the deviation risk analysis module 500 activates the neighborhood propagation analysis unit to conduct a consequential risk assessment. The compliance threshold is preset based on the region's historical statistical distribution. This neighborhood propagation analysis unit constructs a topological network between land parcels based on a geographic information system.

[0103] Specifically, the neighborhood propagation analysis unit treats each plot as a node in the graph and determines directed edges between nodes based on hydrological flow direction, prevailing wind direction data, or biological migration corridor data calculated using a digital elevation model (DEM). If a plot... Located on the plot Upstream or upwind, then establish from point to The directed edges are assigned weights, and the propagation probabilities calculated based on distance and medium properties are assigned to the weights of these edges, thus constructing a directed weighted topology network.

[0104] Based on the aforementioned topological network, the neighborhood propagation analysis unit calculates the potential impact of the risk source plot on downstream or related nodes. This calculation follows the distance decay law and the propagation probability mechanism, meaning that the closer the distance and the smoother the propagation channel, the higher the received risk value.

[0105] By applying the topological propagation risk formula, the system can quantify the cumulative impact of a single site violation on the entire regional ecosystem.

[0106] The formula is as follows: ; In the formula: downstream or adjacent plots The cumulative risk value of transmission received; For the land parcel The collection of all upstream or associated risk source sites that have an impact; Risk source plot The initial violation level value; For the plot of land To the plot The topological propagation probability value; It is the base of the natural logarithm constant; This is the distance attenuation coefficient; For the plot of land With the plot of land The topological path distance between them.

[0107] Finally, the deviation risk analysis module 500 summarizes the above calculation results through the value and loss calculation unit. This value and loss calculation unit combines the basic natural resource value table of the land parcel (including economic parameters such as farmland productivity and ecosystem service value) with abstract technical indicators ( and This is converted into a specific profit and loss statement.

[0108] The value and loss calculation unit generates a resource audit report that includes a compliance status determination (such as "compliant", "non-compliant", "suspected fraud"), a confidence level assessment report, and a prediction of potential associated risks. This report is output to the management terminal through the system interface, providing quantitative basis for natural resource law enforcement, ecological compensation calculation, and adjustments to land and space planning.

[0109] In summary, the natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion provided in this embodiment of the invention, by integrating a semantic mapping construction module 100, a perception strategy generation module 200, a state evolution modeling module 300, an energy quality fingerprint verification module 400, and a deviation risk analysis module 500, constructs an intelligent regulatory system with adaptive feedback and multi-dimensional logical mutual verification capabilities.

[0110] The system utilizes the perception strategy generation module 200 to achieve a shift from passive fixed-frequency monitoring to event-driven active perception. Based on key nodes of state evolution, sensor resources are dynamically scheduled, improving the spatiotemporal accuracy and targeting of monitoring.

[0111] Meanwhile, the system incorporates non-visual energy and mass data strongly correlated with the surface physical state through the energy and mass fingerprint verification module 400, establishing a cross-modal implicit verification mechanism. This effectively addresses the technical bottlenecks of traditional remote sensing monitoring, which is susceptible to surface camouflage interference and struggles to identify logical violations. Combined with dual-track state deviation calculation and topology propagation risk analysis, the system can output audit results that include confidence assessment and associated risk prediction, thereby achieving precise control and dynamic closed-loop auditing of the entire lifecycle of natural resources.

Claims

1. A dynamic monitoring and assessment system for natural resources based on multi-source spatiotemporal data fusion, characterized in that, include: The semantic mapping construction module is used to access the business license database and multi-source basic geographic data, parse the business license document to generate a set of constraint rules, and map the multi-source basic geographic data to a basic spatiotemporal tensor. The perception strategy generation module is used to monitor and generate acquisition strategy instructions based on the evolution of state nodes to drive the external sensor network to acquire enhanced perception data. The state evolution modeling module is used to receive the constraint rule set, the basic spatiotemporal tensor and the enhanced sensing data, deduce the expected legal state sequence based on the constraint rule set, determine the evolution of the state nodes based on the expected legal state sequence and the basic spatiotemporal tensor, and reconstruct the actual physical state sequence by combining the basic spatiotemporal tensor and the enhanced sensing data. The energy quality fingerprint verification module is used to receive the actual physical state sequence and access the time-series energy quality data, and calculate the fingerprint consistency index by establishing a nonlinear mapping relationship between the actual physical state sequence and the time-series energy quality data. The deviation risk analysis module is used to receive the expected legal state sequence, the actual physical state sequence and the fingerprint consistency index, calculate the spatiotemporal semantic distance between the expected legal state sequence and the actual physical state sequence, and calculate the propagation risk value based on the spatiotemporal semantic distance to generate a resource audit report.

2. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The semantic mapping construction module includes: The text rule parsing unit is used to parse the business license documents in the business license database using natural language processing algorithms, extract spatiotemporal constraint parameters, and associate and combine the spatiotemporal constraint parameters to generate the constraint rule set. The spatiotemporal tensor construction unit is used to receive the multi-source basic geographic data, perform spatial discretization and attribute mapping on the multi-source basic geographic data, and map it to a unified four-dimensional sparse tensor space to construct the basic spatiotemporal tensor.

3. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The state evolution modeling module includes: The expected trajectory deduction unit is used to receive the constraint rule set, map the parameters in the constraint rule set to the time axis according to the preset business logic template, generate the expected legal state sequence, and determine the evolution of the state nodes based on the expected legal state sequence and the basic spatiotemporal tensor. The physical trajectory reconstruction unit is used to receive the basic spatiotemporal tensor and the enhanced sensing data, update the enhanced sensing data to the corresponding spatiotemporal voxels of the basic spatiotemporal tensor, generate a state probability distribution vector based on the classification of the basic spatiotemporal tensor, generate the actual physical state sequence using temporal coherence filtering, and calculate the current state interpretation confidence based on the state probability distribution vector. The preset business logic template is pre-constructed based on the legal status transfer path of natural resource land parcels.

4. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 3, characterized in that, The perception strategy generation module includes: The thermal dynamic reconstruction unit is used to monitor the evolution of the state node in real time and receive the current state judgment confidence level. It identifies key state transition nodes in the evolution of the state node based on the accessed external service state flow template. When the current state judgment confidence level is lower than a preset safety threshold, it determines the state fuzzy node and increases the attention weight value of the spatial region corresponding to the key state transition node and the state fuzzy node to generate a thermal distribution. A collaborative scheduling generation unit is used to receive the thermal distribution, determine the current state deviation type based on the state node evolution, and generate the acquisition strategy instruction based on the current state deviation type to drive the external sensor network to acquire the enhanced sensing data. The preset safety threshold is set in advance based on the statistical distribution of historical classification accuracy.

5. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The energy quality fingerprint verification module includes: The non-visual element mapping unit is used to access the time-series energy quality data associated with the monitored plots and to normalize the time-series energy quality data to construct a non-visual data vector. The behavioral confidence assessment unit is used to extract the actual visual feature vector from the actual physical state sequence, establish the causal relationship between the non-visual data vector and the actual visual feature vector using a pre-trained nonlinear mapping model, and execute the fingerprint consistency index formula to calculate the fingerprint consistency index. The nonlinear mapping model is obtained by pre-training based on historical samples containing the non-visual data vectors and the actual visual feature vectors.

6. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 5, characterized in that, The behavioral confidence assessment unit, when executing the fingerprint consistency index formula, specifically includes: The nonlinear mapping model is used to map the non-visual data vectors into theoretical visual features; Calculate the Euclidean distance between the actual visual feature vector and the theoretical visual feature vector; The Euclidean distance is normalized and mapped using the hyperbolic tangent function, and the fingerprint consistency index, which represents the logical consistency between visual features and energy quality data, is calculated.

7. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The deviation risk analysis module includes: The state deviation measurement unit is used to execute the spatiotemporal semantic distance formula, calculate the cumulative difference between the expected legal state sequence and the actual physical state sequence, and obtain the spatiotemporal semantic distance. The state deviation measurement unit, when executing the spatiotemporal semantic distance formula, specifically includes: The difference in state category and the difference in patch area between the expected legal state sequence and the actual physical state sequence are calculated. The difference in state category and the difference in patch area are weighted and summed. The fingerprint consistency index is used as a reverse weighting factor to correct the result of the weighted summation, and finally the spatiotemporal semantic distance is obtained.

8. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 7, characterized in that, The deviation risk analysis module also includes: The neighborhood propagation analysis unit is used to establish a topological network between land parcels based on a geographic information system, and to execute a topological propagation risk formula to calculate the propagation risk value when the spatiotemporal semantic distance exceeds a preset compliance threshold. The neighborhood propagation analysis unit, when executing the topology propagation risk formula, specifically includes: Based on the topology network, a set of risk source plots that affect downstream or neighboring plots is determined, the topological transmission probability value from the risk source plots to the downstream or neighboring plots is calculated, and the propagation risk value is calculated by combining the initial violation degree value of the risk source plots with the distance attenuation factor. The preset compliance threshold is pre-set based on the historical statistical distribution of the region, and the geographic information system is an external system accessed by the deviation risk analysis module.

9. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 8, characterized in that, The deviation risk analysis module also includes: A value and loss calculation unit is used to receive the spatiotemporal semantic distance and the propagation risk value, and generate the resource audit report by combining the land value data; The resource audit report includes a compliance status determination, a confidence level assessment report, and a prediction of potential associated risks; The land value data is obtained from an external database based on a basic natural resource value table.

10. The natural resource dynamic monitoring and assessment system based on multi-source spatiotemporal data fusion according to claim 4, characterized in that, The acquisition strategy instructions include: Geometric texture acquisition commands are used to schedule optical satellites or low-altitude unmanned aerial vehicle systems to acquire images of a target area. The spectral attribute acquisition command is used to schedule observation platforms equipped with hyperspectral imagers or thermal infrared sensors to acquire multi-band data.

Images