Detection system for natural land resources

Abstract

The invention discloses a detection system for natural land resources, and relates to the technical field of resource detection, the detection system comprises a data preprocessing module, a fusion space-time diagram construction module, a multi-target enhancement module and a feedback module, an entropy law is combined with a semantic space-time diagram index, quantitative modeling of an ecological law is realized, and the real-time performance of the system is improved. Natural-human collaborative intervention simulation under multiple time scales, time-space semantic query response time improvement, prediction precision improvement and interdisciplinary fusion are supported, the entropy law of ecological economics is combined with time-space diagram index, the limitation that a traditional model only pays attention to physical indexes is broken through, ecological-economic balance is achieved through an entropy weight method and Monte Carlo simulation, and the prediction precision is improved. According to the method, self-adaptive decision making is supported, a perception-decision-execution integrated platform is constructed based on the real-time updating capability of space-time diagram indexes, a full-chain solution from data collection to dynamic optimization can be provided for natural resource intelligent management, a prediction model is combined with entropy increase constraint, and the ecological economic cost is quantified.

Description

Technical Field

[0001] The present invention relates to the technical field of resource detection, and in particular to a detection system for land natural resources. Background Art

[0002] Land is one of the fundamental resources for human survival and development, widely used in agriculture, forestry, urban construction, transportation, and other fields. The rational use and protection of land resources directly impacts ecological balance, economic development, and social stability. With the continuous expansion of human activities, land resources are facing increasing pressure, particularly from ecological and environmental changes, overexploitation, and land degradation. Technological advances, particularly the widespread application of remote sensing, geographic information systems (GIS), big data analytics, and artificial intelligence (AI), have made land resource monitoring and assessment methods more accurate and efficient. These technologies have made dynamic land resource monitoring, resource change analysis, and environmental impact assessments more feasible, providing a scientific basis for decision makers.

[0003] At present, the Chinese invention application with application number CN202311302378.2 discloses a natural resource survey and monitoring data processing system, which constructs a three-dimensional spatiotemporal data model of natural resources and integrates the management of natural resource survey and monitoring data; among them, the multi-source data acquisition module, the data management planning module, the real-time data monitoring module, and the data storage application module are connected in sequence, which is conducive to the efficient management of natural resource data and the sustainable update of natural resource data; however, the existing technology cannot support the simulation of natural-human collaborative intervention under multiple time scales, cannot realize the quantitative modeling of ecological laws, has low efficiency in querying natural resource responses, only focuses on physical indicators, lacks the integration of interdisciplinary detection and prediction of ecological-economic balance, and cannot realize closed-loop management. Summary of the Invention

[0004] The technical problem solved by the present invention is that the existing technology cannot support the simulation of natural-human collaborative intervention under multiple time scales, cannot realize the quantitative modeling of ecological laws, has low efficiency in querying natural resource responses, only focuses on physical indicators, lacks the integration of interdisciplinary detection and prediction of ecological-economic balance, and cannot achieve closed-loop management.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a land natural resource detection system, comprising a data preprocessing module, a fusion spatiotemporal graph construction module, a multi-objective enhancement module, and a feedback module: The data preprocessing module is used to detect terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data, and initialize the ecological spatiotemporal feature keyword association matrix; The fusion spatiotemporal graph construction module is used to initialize the ecological spatiotemporal feature keyword association matrix and construct the index of the ecological spatiotemporal feature keyword association matrix; The multi-objective reinforcement module is used to obtain a land natural resource prediction model by constraining the prediction model with entropy increase based on the index of the ecological spatiotemporal feature keyword association matrix under the dual conditions of initialization of the natural evolution of land ecological resources and intervention of human dynamic data; The feedback module is used to use the land natural resource prediction model to predict the spatiotemporal resource changes of the land to be detected and output a visual prediction result, and analyze and optimize the visual prediction result.

[0006] Preferably, the data preprocessing module includes a data acquisition unit and a semantic encoding unit: The data acquisition unit is used to detect terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data; Topographic factors, including slope, aspect, position, and altitude, were measured using digital elevation models and GNSS technology, and derived indices of terrain relief and slope change rate were calculated using 3D laser scanning. Using multi-parameter integrated sensors, combined with soil physical, chemical, and hydraulic parameters, to monitor soil property data includes soil bulk density, penetration resistance, soil pH, organic matter content, total nitrogen content, soil moisture content, and water potential. Monitoring groundwater dynamics through water level gauges and evaluating hydrological data using continuous observations of runoff fields, including precipitation, runoff flow, runoff depth, flow velocity, and water level; Use GNSS and crack meters to monitor landslide and crack displacement changes in real time and obtain disaster anomaly data; Using drones equipped with multispectral remote sensing cameras to collect vegetation ecological parameters, including vegetation cover and root biomass, combined with standard plot surveys to obtain vegetation community structure data; The semantic data of the currently detected land is collected through big data, and the semantic data includes policy texts, land use planning documents, ecological protection zone boundaries, records of land use changes over the years, and ecological restoration project archives.

[0007] Preferably, the semantic encoding unit includes: The terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data collected by the data collection unit are cleaned, spatiotemporally aligned and feature extracted to initialize the ecological spatiotemporal feature keyword association matrix; Data cleaning includes repairing remote sensing images through GAN to remove cloud occlusion and noise interference, removing duplicate records, filling missing values, and detecting and removing extreme values ​​and erroneous data through the statistical method Z-score.

[0008] Preferably, the spatiotemporal alignment includes: Performing semantic standardization on terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters, and semantic data, wherein the semantic standardization includes performing normalization calculation on the terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters, and semantic data, and using the normalized data as data points; A neural network is used to extract relational features between the real-world meanings represented by data points obtained from big data collection, and an association relationship is defined based on the relational features. The association relationship includes dependency, symbiosis, and competition. The definition logic includes partitioning the normalized values ​​of the relational features by preset thresholds to obtain a first threshold interval, a second threshold interval, and a third threshold interval. The first threshold interval corresponds to a dependency relationship, the second threshold interval corresponds to a symbiotic relationship, and the third threshold interval corresponds to a competitive relationship. The geographical location longitude and latitude of each data point are respectively Geohash-encoded, the data point codes are obtained, and all data point codes are mapped to a unified spatial grid. The data point codes in the unified spatial grid respectively include a time node queue for data collection at the geographical location of the data point. Multi-granularity time coding is performed on all time nodes, all time nodes are unified into a single timestamp standard, and missing data in the time interval is filled by an interpolation method. Feature extraction includes extracting semantic features from the spatiotemporally aligned data point encodings using a pre-trained neural network, encoding the semantic features using a BERT model, and obtaining ecological keyword feature vectors.

[0009] Preferably, the fusion spatiotemporal graph construction module is used to initialize the ecological spatiotemporal feature keyword association matrix and construct the index of the ecological spatiotemporal feature keyword association matrix. The initialization process includes: The spatial dimension of the unified spatial grid is used as the first dimension of the ecological spatiotemporal feature keyword association matrix, the timestamp standard of all time nodes is used as the second dimension of the ecological spatiotemporal feature keyword association matrix, and the semantic features included in the encoding of each data point is used as the third dimension of the ecological spatiotemporal feature keyword association matrix. The ecological spatiotemporal feature keyword association matrix is ​​defined according to the association relationship, and each item of the ecological spatiotemporal feature keyword association matrix represents the relationship between different ecological features at a certain spatiotemporal location.

[0010] Preferably, constructing the index of the ecological spatiotemporal feature keyword association matrix includes: Classifying each item of the ecological spatiotemporal feature keyword association matrix, wherein the classification types include entity nodes, time nodes, space nodes, and feature nodes. The entity nodes include terrain factors, soil property data, hydrological data, disaster anomaly data, vegetation ecological parameters, and semantic data corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The time nodes include timestamps corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The space nodes include data point codes corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The feature nodes include ecological keyword feature vectors corresponding to each item of the ecological spatiotemporal feature keyword association matrix. Performing an associated edge link on each item of the ecological spatiotemporal feature keyword association matrix, wherein the associated edge link includes connecting an entity node with a time corresponding to the entity node, an entity node with a space corresponding to the entity node, and an entity node with a relationship between the entity node and the entity node; Establish cross-time slice associations through timestamp queues.

[0011] Preferably, the multi-objective reinforcement module includes a model optimization unit and an entropy increase constraint unit. The model optimization unit is used to construct a land natural resource prediction model based on the index of the ecological spatiotemporal feature keyword association matrix under the dual conditions of the natural evolution of land ecological resources and human dynamic data intervention. The land natural resource prediction model construction process includes: Identify the resource evolution path through k-step reachability query, where k is a constant. The resource evolution path is from degradation to repair and then to stability. Output the resource state transition probability matrix, where each item in the resource state transition probability matrix is ​​the probability of a data point code transforming according to the resource evolution path. Initializing the LSTM prediction model based on the resource state transition probability matrix and adding the ecological economic cost quantification into the LSTM prediction model, initializing the LSTM prediction model includes: The resource state transition probability matrix is ​​spliced ​​as an additional feature to the end of the ecological spatiotemporal feature keyword association matrix to obtain an input sequence, which is input into the LSTM prediction model for training and analysis, and the probability distribution of future resource states is output; The mathematical expression of the cost function for quantifying ecological and economic costs is: ; in, is the cost function, is the land use cost, For biodiversity loss, is the carbon sink loss, t is the time, is the economic cost weight, is the ecological cost weight, is the climate cost weight, 、 and Dynamic adjustment is made through expert scoring.

[0012] Preferably, the entropy increase constraint unit includes: Calculate the future entropy increase, impose entropy increase constraints on the LSTM prediction model, obtain the land natural resources prediction model, and calculate the mathematical expression of the future entropy increase as follows: ; in, is the change in entropy increase in the future, and To manually adjust the parameters, is the entropy rate of resource consumption, obtained from the ratio of resource consumption to renewable capacity, is the state transition uncertainty, which is obtained by calculating the variance of the probability distribution of LSTM output; Input historical entropy value and resource state transition probability matrix, outputting future entropy increment; The upper limit of entropy increase is integrally constrained, and the mathematical expression of the integral constraint is: ; in, and They are the output data time and the input data time, is the preset maximum entropy increase threshold; The integral constraint is embedded in the prediction process of the LSTM prediction model through the projected gradient method. If the constraint is violated, the prediction step size is shortened.

[0013] Preferably, the feedback module includes a prediction unit and a visualization decision unit: The prediction unit is used to use the land natural resource prediction model to predict the spatiotemporal resource changes of the land to be tested and output a visual prediction result, wherein the visual prediction result includes a heat map showing the high risk area of ​​entropy increase and a land evolution path map, such as the degradation trend of cultivated land; Analyze and optimize the visual prediction results, including: The DBSCAN algorithm is used to divide resource hotspots and reduce entropy in areas with excessive entropy increase. Use the clustering results as partition nodes of the spatiotemporal graph index; Use the LOF local outlier factor algorithm and spatiotemporal scanning statistical method to detect anomalies and locate the source of entropy increase; The index failure probability is estimated through Monte Carlo simulation and redundant paths are constructed.

[0014] Preferably, the visualization decision unit includes: Satellite remote sensing and sensor data update the status of spatiotemporal graph index nodes in real time, and adaptively adjust function weights and intervention paths.

[0015] The beneficial effects of the present invention are as follows: combining the entropy law with the semantic spatiotemporal graph index to realize the quantitative modeling of ecological laws, supporting the simulation of natural-human collaborative intervention at multiple time scales, improving the response time of spatiotemporal semantic queries, improving the prediction accuracy, and cross-disciplinary integration. Combining the entropy law of ecological economics with the spatiotemporal graph index breaks through the limitation of traditional models that only focus on physical indicators, and realizes ecological-economic equilibrium through the entropy weight method and Monte Carlo simulation, supports adaptive decision-making, and builds an integrated "perception-decision-execution" platform based on the real-time update capability of the spatiotemporal graph index. This method can provide a full-chain solution from data collection to dynamic optimization for the intelligent management of natural resources, help realize the sustainable governance of the "mountains, rivers, forests, fields, lakes and grasslands" life community, combine the prediction model with the entropy increase constraint, and quantify the ecological and economic costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the basic flow of a land natural resource detection system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0018] Reference Figure 1 , which is an embodiment of the present invention, provides a land natural resource detection system, including a data preprocessing module, a fusion spatiotemporal graph construction module, a multi-objective enhancement module, and a feedback module: The data preprocessing module is used to detect terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data, and initialize the ecological spatiotemporal feature keyword association matrix; The fusion spatiotemporal graph construction module is used to initialize the ecological spatiotemporal feature keyword association matrix and construct the index of the ecological spatiotemporal feature keyword association matrix; The multi-objective reinforcement module is used to index the ecological spatiotemporal feature keyword association matrix, and to impose entropy increase constraints on the LSTM prediction model under the dual conditions of the natural evolution of land ecological resources and human dynamic data intervention to obtain a land natural resource prediction model. The feedback module is used to use the land natural resource prediction model to predict the spatiotemporal resource changes of the land to be tested and output the visual prediction results, and analyze and optimize the visual prediction results.

[0019] The data preprocessing module includes a data acquisition unit and a semantic encoding unit: The data acquisition unit is used to detect terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data; Topographic factors, including slope, aspect, position, and altitude, were measured using digital elevation models and GNSS technology, and derived indicators of terrain relief and slope change rate were calculated using 3D laser scanning. Soil property data is monitored using multi-parameter integrated sensors, combining soil physical, chemical, and hydraulic parameters. These sensors include the TEROS series capacitance sensors, penetration resistance meters, pressure water level gauges, and negative pressure sensors. Soil property data includes soil bulk density, penetration resistance, soil pH, organic matter content, total nitrogen content, soil moisture content, and water potential. 3D laser scanning is used to obtain derived indicators such as terrain relief and slope change rate, providing high-precision 3D terrain data to better understand terrain changes. Groundwater dynamics are monitored through water level gauges, and hydrological data are evaluated using continuous observations of runoff fields. Hydrological data include precipitation, runoff flow, runoff depth, flow velocity, and water level. Runoff flow refers to the amount of water flowing through a specific section, and runoff depth refers to the amount of runoff per unit area. Use GNSS and crack meters to monitor landslide and crack displacement changes in real time and obtain disaster anomaly data; Using drones equipped with multispectral remote sensing cameras to collect vegetation ecological parameters, including vegetation cover and root biomass, combined with standard plot surveys to obtain vegetation community structure data; The semantic data of the currently detected land is collected through big data. The semantic data includes policy texts, land use planning documents, ecological protection zone boundaries, records of land use changes over the years, and ecological restoration project archives.

[0020] The slope fitting equation is used to optimize the DEM grid, and three-dimensional laser scanning is used to avoid errors in dangerous slope areas.

[0021] Semantic coding units include: The terrain factors, soil characteristics data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data collected by the data collection unit are cleaned, spatiotemporally aligned and feature extracted to initialize the ecological spatiotemporal feature keyword association matrix; Data cleaning includes repairing remote sensing images through GAN to remove cloud occlusion and noise interference, removing duplicate records, filling missing values, and detecting and removing extreme values ​​and erroneous data through the statistical method Z-score.

[0022] Space-time alignment includes: Semantic standardization is performed on terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data. Semantic standardization includes normalizing terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data, and the normalized data are respectively used as data points; A neural network is used to extract relational features between the real-world meanings represented by data points obtained from big data collection. Association relationships are defined based on the relational features. Association relationships include dependency, symbiosis, and competition. The definition logic includes partitioning the normalized values ​​of the relational features by preset thresholds to obtain a first threshold interval, a second threshold interval, and a third threshold interval. The first threshold interval corresponds to a dependency relationship, the second threshold interval corresponds to a symbiotic relationship, and the third threshold interval corresponds to a competitive relationship. The geographic location longitude and latitude of each data point are Geohash-encoded, the data point codes are obtained, and all data point codes are mapped to a unified spatial grid. The codes of each data point in the unified spatial grid include a time node queue for data collection at the geographic location of the data point. Multi-granularity time coding is performed on all time nodes, all time nodes are unified into a single timestamp standard, and missing data in the time interval is filled by interpolation methods. Feature extraction includes extracting semantic features from the spatiotemporally aligned data point encodings using a pre-trained neural network, encoding the semantic features using the BERT model, and obtaining ecological keyword feature vectors.

[0023] The fusion spatiotemporal graph construction module is used to initialize the ecological spatiotemporal feature keyword association matrix and construct the index of the ecological spatiotemporal feature keyword association matrix. The initialization process includes: The spatial dimension of the unified spatial grid is used as the first dimension of the ecological spatiotemporal feature keyword association matrix, the timestamp standard of all time nodes is used as the second dimension of the ecological spatiotemporal feature keyword association matrix, and the semantic features included in the encoding of each data point is used as the third dimension of the ecological spatiotemporal feature keyword association matrix. The ecological spatiotemporal feature keyword association matrix is ​​defined according to the association relationship, and each item of the ecological spatiotemporal feature keyword association matrix represents the relationship between different ecological characteristics at a certain spatiotemporal location.

[0024] The indexes for constructing the ecological spatiotemporal feature keyword association matrix include: Classify each item of the ecological spatiotemporal feature keyword association matrix, and the classification types include entity nodes, time nodes, space nodes and feature nodes. The entity nodes include terrain factors, soil property data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The time nodes include the timestamps corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The space nodes include the data point codes corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The feature nodes include the ecological keyword feature vectors corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The associated edge links are performed on each item of the ecological spatiotemporal feature keyword association matrix. The associated edge links include connecting the time corresponding to the entity nodes and the space corresponding to the entity nodes and the entity nodes having relationships with the entity nodes.

[0025] Connecting the time and space corresponding to the entity nodes can obtain: adjacent spatial distance, time series changes, such as interannual changes in vegetation coverage; Connecting entity nodes with entity nodes that have relationships with entity nodes can obtain policy constraints such as "prohibited development areas" and ecological processes such as "soil erosion → river siltation"; The association across time slices is established through the timestamp queue, and the association across time slices represents the evolution path of current land resource utilization or ecological migration.

[0026] The multi-objective reinforcement module includes a model optimization unit and an entropy increase constraint unit. The model optimization unit is used to construct a land natural resource prediction model based on the index of the ecological spatiotemporal characteristic keyword association matrix under the dual conditions of the natural evolution of land ecological resources and human dynamic data intervention. The land natural resource prediction model construction process includes: Identify the resource evolution path through k-step reachability query, where k is a constant. The resource evolution path is from degradation to restoration and then to stability. Output the resource state transition probability matrix. Each item in the resource state transition probability matrix is ​​the probability of the data point code transforming according to the resource evolution path. For example, the probability of the data point code corresponding to the current land changing from "degradation" to "restoration" is 0.6. Initialize the LSTM prediction model based on the resource state transition probability matrix and add the ecological and economic cost quantification into the LSTM prediction model. Initializing the LSTM prediction model includes: The resource state transition probability matrix is ​​added as an additional feature to the end of the ecological spatiotemporal feature keyword association matrix to obtain the input sequence. The input sequence is input into the LSTM prediction model for training and analysis, and the probability distribution of future resource states is output, such as the range of forest coverage changes in the next five years. The mathematical expression of the cost function for quantifying ecological and economic costs is: ; in, is the cost function, is the land use cost, For biodiversity loss, is the carbon sink loss, t is the time, is the economic cost weight, is the ecological cost weight, is the climate cost weight, 、 and Dynamic adjustment is made through expert scoring.

[0027] Based on the data collected by the data collection unit, land use costs are assessed through market prices and resource scarcity indicators. Biodiversity loss uses the species abundance index to quantify changes in species diversity. Carbon sink loss refers to the decline in carbon storage capacity caused by changes in land resource use, such as deforestation and the conversion of grassland to agricultural land. Carbon sink loss is calculated by comparing carbon emissions from different land use changes, such as the conversion of forests to farmland or urbanization.

[0028] The entropy increase constraint unit includes: Calculate the future entropy increase, constrain the entropy increase of the LSTM prediction model, obtain the land natural resource prediction model, and calculate the mathematical expression of the future entropy increase as follows: ; in, is the change in entropy increase in the future, and To manually adjust the parameters, is the entropy rate of resource consumption, which is obtained by the ratio of resource consumption to renewable capacity, that is, the degree to which resource consumption exceeds the renewable threshold, is the state transition uncertainty, which is obtained by calculating the variance of the probability distribution of LSTM output; Input historical entropy value and the resource state transition probability matrix, outputting the future entropy increment, such as 、 and ; The upper limit of entropy increase is integrally constrained, and the mathematical expression of the integral constraint is: ; in, and They are the output data time and the input data time, is the preset maximum entropy increase threshold; The integral constraint is embedded in the prediction process of the LSTM prediction model through the projected gradient method. If the constraint is violated, the prediction step size is shortened, for example, from 5 years to 3 years.

[0029] The feedback module includes a prediction unit and a visualization decision unit: The prediction unit is used to use the land natural resource prediction model to predict the spatiotemporal resource changes of the tested land and output visual prediction results. The visual prediction results include heat maps showing high-risk areas of entropy increase and land evolution path maps, such as cultivated land degradation trends; Analyze and optimize the visual prediction results, including: The DBSCAN algorithm is used to divide resource hotspots, reduce entropy in areas with excessive entropy increase, identify resource aggregation patterns, and reduce spatial disorder; Use clustering results as partition nodes of spatiotemporal graph index to reduce cross-region query overhead; Use the LOF local outlier factor algorithm and spatiotemporal scanning statistical methods to detect anomalies, locate the source of entropy increase, and promptly repair resource status indexes; Use Monte Carlo simulation to estimate the probability of index failure, build redundant paths, evaluate the impact of data anomalies on index structure, and manually design disaster recovery strategies.

[0030] The visualization decision-making unit includes: Satellite remote sensing and sensor data update the status of spatiotemporal graph index nodes in real time, adaptively adjust function weights and intervention paths, and realize iterative optimization of "monitoring → prediction → intervention → re-monitoring".

[0031] This invention combines the entropy law with the semantic spatiotemporal graph index to realize the quantitative modeling of ecological laws, support the simulation of natural-human collaborative intervention at multiple time scales, improve the response time of spatiotemporal semantic queries, improve the prediction accuracy, and integrate disciplines. It combines the entropy law of ecological economics with the spatiotemporal graph index, breaking through the limitation of traditional models that only focus on physical indicators. It achieves ecological-economic equilibrium through entropy weight method and Monte Carlo simulation, supports adaptive decision-making, and constructs an integrated "perception-decision-execution" platform based on the real-time update capability of the spatiotemporal graph index. This method can provide a full-chain solution from data collection to dynamic optimization for the intelligent management of natural resources, help realize the sustainable governance of the "mountains, rivers, forests, fields, lakes and grasslands" life community, combine the prediction model with the entropy increase constraint, and quantify the ecological and economic costs.

[0032] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0033] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A land natural resource detection system, characterized in that: It includes data preprocessing module, fusion spatiotemporal graph construction module, multi-objective reinforcement module and feedback module: The data preprocessing module is used to detect terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data, and initialize the ecological spatiotemporal feature keyword association matrix; The fusion spatiotemporal graph construction module is used to initialize the ecological spatiotemporal feature keyword association matrix and construct the index of the ecological spatiotemporal feature keyword association matrix; The multi-objective reinforcement module is used to obtain a land natural resource prediction model by constraining the prediction model with entropy increase based on the index of the ecological spatiotemporal feature keyword association matrix under the dual conditions of initialization of the natural evolution of land ecological resources and intervention of human dynamic data; The feedback module is used to use the land natural resource prediction model to predict the spatiotemporal resource changes of the land to be detected and output a visual prediction result, and analyze and optimize the visual prediction result.

2. A land natural resource detection system as claimed in claim 1, characterized in that: The data preprocessing module includes a data acquisition unit and a semantic encoding unit: The data acquisition unit is used to detect terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data; Topographic factors, including slope, aspect, position, and altitude, were measured using digital elevation models and GNSS technology, and derived indices of terrain relief and slope change rate were calculated using 3D laser scanning. Using multi-parameter integrated sensors, combined with soil physical, chemical, and hydraulic parameters, to monitor soil property data includes soil bulk density, penetration resistance, soil pH, organic matter content, total nitrogen content, soil moisture content, and water potential. Monitoring groundwater dynamics through water level gauges and evaluating hydrological data using continuous observations of runoff fields, including precipitation, runoff flow, runoff depth, flow velocity, and water level; Use GNSS and crack meters to monitor landslide and crack displacement changes in real time and obtain disaster anomaly data; Using drones equipped with multispectral remote sensing cameras to collect vegetation ecological parameters, including vegetation cover and root biomass, combined with standard plot surveys to obtain vegetation community structure data; The semantic data of the currently detected land is collected through big data, and the semantic data includes policy texts, land use planning documents, ecological protection zone boundaries, records of land use changes over the years, and ecological restoration project archives.

3. A land natural resource detection system as claimed in claim 2, characterized in that: The semantic encoding unit includes: The terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters and semantic data collected by the data collection unit are cleaned, spatiotemporally aligned and feature extracted to initialize the ecological spatiotemporal feature keyword association matrix; Data cleaning includes repairing remote sensing images through GAN to remove cloud occlusion and noise interference, removing duplicate records, filling missing values, and detecting and removing extreme values ​​and erroneous data through the statistical method Z-score.

4. A land natural resource detection system as claimed in claim 3, characterized in that: Space-time alignment includes: Performing semantic standardization on terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters, and semantic data, wherein the semantic standardization includes performing normalization calculation on the terrain factors, soil characteristic data, hydrological data, disaster anomaly data, vegetation ecological parameters, and semantic data, and using the normalized data as data points; A neural network is used to extract relational features between the real-world meanings represented by data points obtained from big data collection, and an association relationship is defined based on the relational features. The association relationship includes dependency, symbiosis, and competition. The definition logic includes partitioning the normalized values ​​of the relational features by preset thresholds to obtain a first threshold interval, a second threshold interval, and a third threshold interval. The first threshold interval corresponds to a dependency relationship, the second threshold interval corresponds to a symbiotic relationship, and the third threshold interval corresponds to a competitive relationship. The geographical location longitude and latitude of each data point are respectively Geohash-encoded, the data point codes are obtained, and all data point codes are mapped to a unified spatial grid. The data point codes in the unified spatial grid respectively include a time node queue for data collection at the geographical location of the data point. Multi-granularity time coding is performed on all time nodes, all time nodes are unified into a single timestamp standard, and missing data in the time interval is filled by an interpolation method. Feature extraction includes extracting semantic features from the spatiotemporally aligned data point encodings using a pre-trained neural network, encoding the semantic features using a BERT model, and obtaining ecological keyword feature vectors.

5. A land natural resource detection system as claimed in claim 4, characterized in that: The fusion spatiotemporal graph construction module is used to initialize the ecological spatiotemporal feature keyword association matrix and construct the index of the ecological spatiotemporal feature keyword association matrix. The initialization process includes: The spatial dimension of the unified spatial grid is used as the first dimension of the ecological spatiotemporal feature keyword association matrix, the timestamp standard of all time nodes is used as the second dimension of the ecological spatiotemporal feature keyword association matrix, and the semantic features included in the encoding of each data point is used as the third dimension of the ecological spatiotemporal feature keyword association matrix. The ecological spatiotemporal feature keyword association matrix is ​​defined according to the association relationship, and each item of the ecological spatiotemporal feature keyword association matrix represents the relationship between different ecological features at a certain spatiotemporal location.

6. A land natural resource detection system as claimed in claim 5, characterized in that: The index for constructing the ecological spatiotemporal feature keyword association matrix includes: Classifying each item of the ecological spatiotemporal feature keyword association matrix, wherein the classification types include entity nodes, time nodes, space nodes, and feature nodes. The entity nodes include terrain factors, soil property data, hydrological data, disaster anomaly data, vegetation ecological parameters, and semantic data corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The time nodes include timestamps corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The space nodes include data point codes corresponding to each item of the ecological spatiotemporal feature keyword association matrix. The feature nodes include ecological keyword feature vectors corresponding to each item of the ecological spatiotemporal feature keyword association matrix. Performing an associated edge link on each item of the ecological spatiotemporal feature keyword association matrix, wherein the associated edge link includes connecting an entity node with a time corresponding to the entity node, an entity node with a space corresponding to the entity node, and an entity node with a relationship between the entity node and the entity node; Establish cross-time slice associations through timestamp queues.

7. A land natural resource detection system as claimed in claim 6, characterized in that: The multi-objective reinforcement module includes a model optimization unit and an entropy increase constraint unit. The model optimization unit is used to construct a land natural resource prediction model based on the index of the ecological spatiotemporal feature keyword association matrix under the dual conditions of the natural evolution of land ecological resources and human dynamic data intervention. The land natural resource prediction model construction process includes: Identify the resource evolution path through k-step reachability query, where k is a constant. The resource evolution path is from degradation to repair and then to stability. Output the resource state transition probability matrix, where each item in the resource state transition probability matrix is ​​the probability of a data point code transforming according to the resource evolution path. Initializing the LSTM prediction model based on the resource state transition probability matrix and adding the ecological economic cost quantification into the LSTM prediction model, initializing the LSTM prediction model includes: The resource state transition probability matrix is ​​spliced ​​as an additional feature to the end of the ecological spatiotemporal feature keyword association matrix to obtain an input sequence, which is input into the LSTM prediction model for training and analysis, and the probability distribution of future resource states is output; The mathematical expression of the cost function for quantifying ecological and economic costs is: ； in, is the cost function, is the land use cost, For biodiversity loss, is the carbon sink loss, t is the time, is the economic cost weight, is the ecological cost weight, is the climate cost weight, 、 and Dynamic adjustment is made through expert scoring.

8. A land natural resource detection system as claimed in claim 7, characterized in that: The entropy increase constraint unit includes: Calculate the future entropy increase, impose entropy increase constraints on the LSTM prediction model, obtain the land natural resources prediction model, and calculate the mathematical expression of the future entropy increase as follows: ； in, is the change in entropy increase in the future, and To manually adjust the parameters, is the entropy rate of resource consumption, obtained from the ratio of resource consumption to renewable capacity, is the state transition uncertainty, which is obtained by calculating the variance of the probability distribution of LSTM output; Input historical entropy value and resource state transition probability matrix, outputting future entropy increment; The upper limit of entropy increase is integrally constrained, and the mathematical expression of the integral constraint is: ； in, and They are the output data time and the input data time, is the preset maximum entropy increase threshold; The integral constraint is embedded in the prediction process of the LSTM prediction model through the projected gradient method. If the constraint is violated, the prediction step size is shortened.

9. A land natural resource detection system as claimed in claim 8, characterized in that: The feedback module includes a prediction unit and a visualization decision unit: The prediction unit is used to use the land natural resource prediction model to predict the spatiotemporal resource changes of the land to be tested and output a visual prediction result, wherein the visual prediction result includes a heat map showing the high risk area of ​​entropy increase and a land evolution path map, such as the degradation trend of cultivated land; Analyze and optimize the visual prediction results, including: The DBSCAN algorithm is used to divide resource hotspots and reduce entropy in areas with excessive entropy increase. Use the clustering results as partition nodes of the spatiotemporal graph index; Use the LOF local outlier factor algorithm and spatiotemporal scanning statistical method to detect anomalies and locate the source of entropy increase; The index failure probability is estimated through Monte Carlo simulation and redundant paths are constructed.

10. A land natural resource detection system as claimed in claim 9, characterized in that: The visualization decision unit includes: Satellite remote sensing and sensor data update the status of spatiotemporal graph index nodes in real time, and adaptively adjust function weights and intervention paths.

Images