A national land space monitoring system based on remote sensing data modeling

Abstract

The present invention discloses a land space monitoring system based on remote sensing data modeling. The system includes modules for data collection, remote sensing feature modeling, temporal change detection, ecological security assessment, and comprehensive monitoring. The present invention belongs to the field of remote sensing data modeling and analysis technology. The system collects multi-source remote sensing data in a standardized manner, constructs a structured model of land features, and adopts improved causal change detection and multi-scale ecological index assessment methods to achieve dynamic perception and assessment of surface changes and ecological risks. Innovations include a differentiable feature selection mechanism for deep geographic perception coding, a dual-stream change detection model, and ecological risk circuit simulation technology. This system constructs a highly integrated, data-driven, fully closed-loop monitoring platform, improving monitoring accuracy and response efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing data modeling and analysis, and specifically relates to a land space monitoring system based on remote sensing data modeling. Background Art

[0002] The National Land and Space Monitoring System, based on remote sensing data modeling, is a comprehensive platform that uses multi-source remote sensing data (such as satellite and aerial data) to automatically monitor and analyze surface changes, ecological environments, and land use conditions. Through data collection, feature modeling, temporal change detection, and evaluation and analysis, the system enables continuous observation and precise identification of national land and space dynamics, providing scientific decision-making support for land planning, resource management, and ecological protection.

[0003] However, in the existing national land space monitoring system, there are technical problems such as scattered monitoring elements, low data coupling, and fragmented analysis process; in the existing remote sensing feature modeling process, there are technical problems such as high reliance on manual feature engineering and difficulty in expressing spatial heterogeneity; in the existing time series change detection process, there are technical problems such as insufficient change identification accuracy and difficulty in tracing the causes of changes; in the existing ecological security assessment process, there are technical problems such as a single indicator system and delayed static response. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a land space monitoring system based on remote sensing data modeling. In view of the technical problems of scattered monitoring elements, low data coupling and fragmented analysis process in the existing land space monitoring system, this solution creatively adopts a comprehensive land space monitoring process that combines remote sensing feature modeling, time series change detection and ecological security assessment, and realizes closed-loop management of the entire process from data collection, feature extraction, change detection to ecological assessment, and constructs a unified monitoring architecture dominated by remote sensing data and collaborative assessment of ecology and utilization, which significantly improves the overall integration and response efficiency of the monitoring system; in view of the technical problems of high reliance on manual feature engineering and difficulty in expressing spatial heterogeneity in the existing remote sensing feature modeling process, this solution creatively adopts a differentiable feature selection method combined with geographic perception deep feature coding to perform remote sensing feature modeling, and realizes the automated, efficient and interpretable transformation of remote sensing raw data, which not only greatly reduces the cost of manual intervention, but also effectively improves the adaptability and generalization ability of the model in complex terrain and multi-source data background. To address the technical issues of insufficient change identification accuracy and difficulty in tracing the causes of changes in existing temporal change detection processes, this proposal creatively adopts a differentiable feature selection method combined with geographic perception deep feature encoding to conduct remote sensing feature modeling, constructing a detection mechanism that parallels macro and micro change flows, and incorporating causal relationship maps and violation pattern matching logic. This enables multi-scale, multi-dimensional dynamic monitoring and causal tracing of surface cover and utilization changes, effectively improving the sensitivity of sudden change detection and the scientific nature of legitimacy analysis. To address the technical issues of a single indicator system and lagging static response in existing ecological security assessment processes, this proposal creatively adopts an ecological risk propagation simulation method combined with an improved multi-layer perception ecological index to conduct ecological security assessments. This achieves the integration of multidimensional ecological indicators from pressure, state, to response layers, and leverages resistivity modeling and circuit theory to spatially simulate the ecological risk diffusion process. This significantly enhances the dynamic expression and spatial propagation perception capabilities of ecological security assessments, providing high-precision, systematic quantitative support for national land space ecological risk warning and governance.

[0005] The technical solution adopted by the present invention is as follows: the present invention provides a land space monitoring system based on remote sensing data modeling, including a data collection module, a remote sensing feature modeling module, a time series change detection module, an ecological security assessment module and a land space monitoring module;

[0006] The data collection module is used for data collection, obtaining spatiotemporal reference remote sensing data through data collection, and sending the spatiotemporal reference remote sensing data to the remote sensing feature modeling module;

[0007] The remote sensing feature modeling module is used for remote sensing feature modeling, obtains plot feature structured modeling data through remote sensing feature modeling, and sends the plot feature structured modeling data to the temporal change detection module and the ecological security assessment module;

[0008] The temporal change detection module is used for temporal change detection, obtains land temporal change detection reference data through temporal change detection, and sends the land temporal change detection reference data to the land space monitoring module;

[0009] The ecological security assessment module is used for ecological security assessment, obtains ecological security index reference data through ecological security assessment, and sends the ecological security index reference data to the national land space monitoring module;

[0010] The national land space monitoring module is used for national land space monitoring, and obtains national land space comprehensive monitoring data through national land space monitoring.

[0011] Furthermore, the data collection is used to standardize the collection and preprocessing of multi-source heterogeneous remote sensing data, specifically to obtain spatiotemporal benchmark remote sensing data through multi-source heterogeneous remote sensing data collection;

[0012] The time-space reference remote sensing data includes satellite remote sensing data, aerial remote sensing data and auxiliary data.

[0013] Furthermore, the remote sensing feature modeling is used to convert the original remote sensing data into feature vectors, specifically using a differentiable feature selection method combined with geographic perception deep feature encoding to perform remote sensing feature modeling to obtain plot feature structured modeling data, including the following steps: multi-scale grid division, physical constraint feature generation, spatiotemporal context encoding, geographic deep feature encoding and differentiable feature selection;

[0014] The multi-scale grid division constructs a three-level feature extraction network, performs feature extraction based on the spatiotemporal benchmark remote sensing data, and obtains plot feature structured modeling data;

[0015] The three-level feature extraction network includes a macro grid layer, a middle grid layer and a micro grid layer;

[0016] The land feature structured modeling data includes band reflectance features and digital elevation features;

[0017] The physical constraint feature generation is specifically to obtain physical constraint feature data by extracting terrain features and spectral features based on the structured modeling data of the land parcel features;

[0018] The physical constraint feature data includes terrain feature data and spectral feature data;

[0019] The spatiotemporal context coding is specifically performed by aggregating neighborhood features using GraphSAGE and introducing seasonal periodic coding, and combining the physical constraint feature data to perform spatiotemporal context coding to obtain spatiotemporal coding data;

[0020] The geographic deep feature coding specifically adopts a feature cross-fusion method combined with an improved geographic weighted multi-head attention to perform feature selection optimization, perform geographic deep feature coding, and obtain geographic deep coding data;

[0021] The feature cross-fusion method combined with the improved geographically weighted multi-head attention includes a geographically weighted attention sub-block and a feature cross-fusion sub-block;

[0022] The differentiable feature selection is specifically to construct a differentiable selection layer based on the geographic deep coding data, retain the first fifty key features, and obtain the plot feature structured modeling data.

[0023] Furthermore, the temporal change detection is used to identify spatiotemporal changes in land cover and utilization. Specifically, the temporal change detection is performed using a dual-stream adaptive change detection method improved by combining change causal reasoning to obtain reference data for land temporal change detection. The method includes the following steps: improved spatiotemporal causal graph construction, violation pattern matching, macro-interannual change detection, micro-sudden change detection, dynamic dual-stream adaptive change weight fusion, and temporal change detection.

[0024] The temporal change detection is used to identify spatiotemporal changes in land cover and utilization. Specifically, it uses an improved dual-stream adaptive change detection method combined with change causal reasoning to perform temporal change detection and obtain reference data for land temporal change detection. The method includes the following steps: improved spatiotemporal causal graph construction, violation pattern matching, macro-interannual change detection, micro-sudden change detection, dynamic dual-stream adaptive change weight fusion, and temporal change detection.

[0025] The improved spatiotemporal causal graph construction specifically comprises constructing spatiotemporal nodes by treating the change data of land cover and utilization as change events, and constructing spatiotemporal causal graph data by introducing the standard Granger causality test as the edge node;

[0026] The spatiotemporal node includes the time when the change event occurs, the coordinate position of the change, and the change intensity value characteristics;

[0027] The violation pattern matching is specifically to build a violation pattern matching database, perform violation pattern matching based on the spatiotemporal causal graph data, and obtain violation pattern matching reference data;

[0028] The macro-interannual variation detection is specifically to construct a change detection channel for processing interannual variation data, specifically using a standard window t-test to perform macro-interannual variation detection and obtain macro-interannual variation detection reference data;

[0029] The micro-burst change detection specifically involves building a change detection channel for processing burst change event data, specifically using KL divergence-based anomaly detection to perform micro-burst change detection and obtain micro-burst change detection reference data;

[0030] The dynamic dual-stream adaptive change weight fusion is specifically performed by constructing a dynamic weighted fusion mechanism based on the macro-interannual change detection reference data and the micro-sudden change detection reference data, and obtaining a dynamic dual-stream adaptive change weight fusion model through model training;

[0031] The time series change detection is specifically to perform time series change detection based on the dynamic dual-stream adaptive change weight fusion to obtain land time series change detection reference data.

[0032] Furthermore, the ecological security assessment is used to quantify the regional ecological health status and risks. Specifically, the ecological security assessment is conducted by using an ecological risk propagation simulation method combined with an improved multi-layer perception ecological index to obtain ecological security index reference data, including the following steps: constructing a pressure layer ecological index, constructing a state layer ecological index, constructing a response layer ecological index, constructing a resistivity model, risk diffusion simulation, and comprehensive ecological security assessment;

[0033] The construction of the pressure layer ecological index specifically involves calculating the human activity intensity index based on the night light index data and the road density weighted score data to obtain reference data for the pressure layer ecological index;

[0034] The construction of the state-level ecological index is specifically to calculate the weighted score based on the biodiversity proxy indicator data and the landscape pattern index data to obtain the ecosystem integrity assessment data;

[0035] The construction of the response layer ecological index specifically involves constructing a natural resilience model based on soil moisture time series data, calculating the natural recovery response layer ecological index, and obtaining reference data for the response layer ecological index;

[0036] The resistivity model is constructed by constructing a standard resistivity model to simulate ecological risk propagation and obtain a basic ecological risk propagation model;

[0037] The risk diffusion simulation is specifically to conduct risk diffusion simulation based on the basic model of ecological risk propagation and adopt circuit theory to obtain risk diffusion simulation reference data;

[0038] The circuit theory abstracts the ecosystem into a circuit network model, treating each regional unit as a node in a resistance network. The ecological resistivity between different regions is used as the weight of the connecting edge. By constructing an ecological resistance network diagram based on the standard resistivity model, the transmission path and intensity of ecological risks between regions are simulated.

[0039] The comprehensive ecological security assessment is specifically to conduct a comprehensive ecological security assessment in combination with the risk diffusion simulation reference data to obtain ecological security index reference data.

[0040] Furthermore, the land space monitoring is used for comprehensive analysis and decision support, specifically based on the land time series change detection reference data and the ecological security index reference data, by combining multi-criteria decision analysis to conduct comprehensive land space monitoring and obtain comprehensive land space monitoring data.

[0041] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0042] (1) In response to the technical problems of scattered monitoring elements, low data coupling, and fragmented analysis processes in the existing land and space monitoring system, this solution creatively adopts a comprehensive land and space monitoring process that combines remote sensing feature modeling, temporal change detection, and ecological security assessment. It achieves closed-loop management of the entire process from data collection, feature extraction, change detection to ecological assessment, and constructs a unified monitoring architecture dominated by remote sensing data and coordinated assessment of ecology and utilization, significantly improving the overall integration and response efficiency of the monitoring system.

[0043] (2) In response to the technical problems of high reliance on manual feature engineering and difficulty in expressing spatial heterogeneity in the existing remote sensing feature modeling process, this solution creatively adopts a differentiable feature selection method combined with geographic perception deep feature encoding to perform remote sensing feature modeling, achieving an automated, efficient, and interpretable transformation of remote sensing raw data. This not only significantly reduces the cost of manual intervention, but also effectively improves the adaptability and generalization ability of the model in complex terrain and multi-source data backgrounds.

[0044] (3) In response to the technical problems of insufficient change recognition accuracy and difficulty in tracing the causes of changes in the existing temporal change detection process, this solution creatively adopts a differentiable feature selection method combined with geographic perception deep feature coding to perform remote sensing feature modeling, construct a parallel detection mechanism for macro and micro change flows, and integrates causal relationship maps and violation pattern matching logic to achieve multi-scale and multi-dimensional dynamic monitoring and causal tracing of surface cover and utilization changes, effectively improving the sensitivity of sudden change detection and the scientific nature of legitimacy analysis;

[0045] (4) In response to the technical problems of a single indicator system and delayed static response in the existing ecological security assessment process, this plan creatively adopts an ecological risk propagation simulation method combined with an improved multi-layer perception ecological index to conduct ecological security assessment, realizing the fusion of multi-dimensional ecological indicators from pressure, state to response layer, and using resistivity modeling and circuit theory to perform spatial simulation of the ecological risk diffusion process, significantly enhancing the dynamic expression ability and spatial propagation perception ability of ecological security assessment, and providing high-precision and systematic quantitative support for national land space ecological risk warning and governance. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of the structure of a land space monitoring system based on remote sensing data modeling provided by the present invention;

[0047] Figure 2 A flowchart of the steps performed by the system;

[0048] Figure 3 A flowchart of the steps performed by the remote sensing feature modeling module;

[0049] Figure 4 A flowchart illustrating the steps performed by the timing change detection module;

[0050] Figure 5 Flowchart showing the steps performed by the ecological safety assessment module.

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0053] Example 1, see Figure 1 The present invention provides a land space monitoring system based on remote sensing data modeling, which includes a data collection module, a remote sensing feature modeling module, a time series change detection module, an ecological security assessment module and a land space monitoring module;

[0054] The data collection module is used for data collection, obtaining spatiotemporal reference remote sensing data through data collection, and sending the spatiotemporal reference remote sensing data to the remote sensing feature modeling module;

[0055] The remote sensing feature modeling module is used for remote sensing feature modeling, obtains plot feature structured modeling data through remote sensing feature modeling, and sends the plot feature structured modeling data to the temporal change detection module and the ecological security assessment module;

[0056] The temporal change detection module is used for temporal change detection, obtains land temporal change detection reference data through temporal change detection, and sends the land temporal change detection reference data to the land space monitoring module;

[0057] The ecological security assessment module is used for ecological security assessment, obtains ecological security index reference data through ecological security assessment, and sends the ecological security index reference data to the national land space monitoring module;

[0058] The national land space monitoring module is used for national land space monitoring, and obtains national land space comprehensive monitoring data through national land space monitoring.

[0059] The data collection is used to collect and pre-process multi-source heterogeneous remote sensing data in a standardized manner, specifically to obtain spatiotemporal benchmark remote sensing data through multi-source heterogeneous remote sensing data collection;

[0060] The temporal and spatial reference remote sensing data include satellite remote sensing data, aerial remote sensing data and auxiliary data;

[0061] The satellite remote sensing data includes multispectral reflectance data, thermal infrared radiation observation data and radar remote sensing scattering coefficient data;

[0062] The aerial remote sensing data includes lidar structured point cloud data and hyperspectral continuous spectrum reflectance curve data;

[0063] The auxiliary data includes surface elevation matrix data, meteorological observation data and administrative division topology verification vector data;

[0064] Preferably, the multi-source heterogeneous remote sensing data acquisition is optimized through standardization processing, quality optimization and data storage in sequence; the standardization processing includes data coordinate system conversion, time information standardization and numerical range standardization operations; the quality optimization includes manual identification of numerical quality, data source processing and spatial data topology verification operations; the data storage specifically adopts column storage and constructs a spatial search index to optimize data time series retrieval.

[0065] By performing the above operations, in order to address the technical problems in the existing land space monitoring system, such as scattered monitoring elements, low data coupling, and fragmented analysis process, this solution creatively adopts a comprehensive land space monitoring process that combines remote sensing feature modeling, temporal change detection, and ecological security assessment. It realizes closed-loop management of the entire process from data collection, feature extraction, change detection to ecological assessment, and constructs a unified monitoring architecture dominated by remote sensing data and coordinated assessment of ecology and utilization, which significantly improves the overall integration and response efficiency of the monitoring system.

[0066] Example 2: This example is based on the above example. Figure 1 、 Figure 2 and Figure 3 , the remote sensing feature modeling is used to convert the original remote sensing data into feature vectors, specifically using a differentiable feature selection method combined with geographic perception deep feature encoding to perform remote sensing feature modeling to obtain plot feature structured modeling data, including the following steps: multi-scale grid division, physical constraint feature generation, spatiotemporal context encoding, geographic deep feature encoding and differentiable feature selection;

[0067] The multi-scale grid division constructs a three-level feature extraction network, performs feature extraction based on the spatiotemporal benchmark remote sensing data, and obtains plot feature structured modeling data;

[0068] The three-level feature extraction network includes a macro grid layer, a middle grid layer and a micro grid layer;

[0069] The land feature structured modeling data includes band reflectance features and digital elevation features;

[0070] The macro grid layer specifically adopts a 1km×1km level basic grid and constructs a geographic grid coding system;

[0071] The middle grid layer specifically subdivides the macro grid layer into 10×10 layers by a quadtree recursive algorithm to obtain a 100m×100m basic grid;

[0072] The micro-grid layer is specifically generated by a sliding window method based on the middle-level grid layer to obtain a 10m×10m level basic network;

[0073] The physical constraint feature generation is specifically to obtain physical constraint feature data by extracting terrain features and spectral features based on the structured modeling data of the land parcel features;

[0074] The physical constraint feature data includes terrain feature data and spectral feature data;

[0075] The terrain features are calculated using the Horn algorithm to obtain the terrain feature data. The calculation formula is:

[0076]

[0077] Where S is the terrain feature data, arctan(·) is the hyperbolic tangent function, z is the surface elevation data, x is the east-west horizontal elevation data, and y is the north-south vertical elevation data;

[0078] The calculation formula of the spectral characteristics is:

[0079] T={NDVI,ISA,EVI,NDWI,CelluLose,T c ,PT};

[0080] Where, T is the spectral characteristic data, NDVI is the normalized difference vegetation index, which is used to reflect the vegetation coverage, ISA is the impervious surface index, which is used to indicate the degree of surface hardening increase, EVI is the vegetation sensitivity index, which is used to reflect the growth status of plants in the forest area, NDWI is the normalized water index, which is used to indicate the turbidity of water distribution, CelluLose is the cellulose characteristic peak parameter, which is used to monitor cellulose through the ratio of short-wave infrared bands, and T c It is the suspended matter concentration index of water bodies, which is used to indicate the change of water turbidity. PT is the index of the degree of sudden change in surface cover, which is used to indicate whether the use of cultivated land has changed.

[0081] The spatiotemporal context coding is specifically performed by aggregating neighborhood features using GraphSAGE and introducing seasonal periodic coding, and combining the physical constraint feature data to perform spatiotemporal context coding to obtain spatiotemporal coding data;

[0082] The calculation formula for the GraphSAGE-aggregated neighborhood features is:

[0083]

[0084] Where, is the aggregation domain feature vector value of the jth neighbor node corresponding to the l+1th layer, which is used to represent the aggregation domain feature vector, σ(·) is a nonlinear activation function, and W (l) is the weight of the lth layer, CONCAT(·) is the vector concatenation function, i is the graph node index, l is the network level index, j is the adjacent node index, is the set of neighboring nodes of graph node i, is the aggregated domain feature vector value of the jth neighbor node corresponding to the lth layer;

[0085] The introduction of seasonal periodic coding specifically refers to feature coding optimization through time series periodic coding, and the calculation formula is:

[0086]

[0087] Where P(·) is the seasonal period encoding vector used to optimize the temporal periodicity of the domain feature vector, and t is the time index;

[0088] The geographic deep feature coding specifically adopts a feature cross-fusion method combined with an improved geographic weighted multi-head attention to perform feature selection optimization, perform geographic deep feature coding, and obtain geographic deep coding data;

[0089] The feature cross-fusion method combined with the improved geographically weighted multi-head attention includes a geographically weighted attention sub-block and a feature cross-fusion sub-block;

[0090] The geographically weighted attention sub-block is used to enhance the attention of geographical features. The calculation formula is:

[0091]

[0092] Where Attention(·) is the attention function, Q is the query vector, K is the key vector, V is the value vector, softmax(·) is the normalized classification function, and d k is the key vector dimension value, λ is the geographical weight coefficient, and D is the node spacing parameter;

[0093] The feature cross-fusion sub-block is used to optimize the fusion of spectral features and terrain features. The calculation formula is:

[0094] X cross =[S☉T,||ST||2,exp(-||ST|| 2 )];

[0095] Where, X cross is the geographic depth encoding data, S is the terrain feature data, T is the spectral feature data, and ⊙ is the Hadamard product operator;

[0096] The differentiable feature selection is specifically to construct a differentiable selection layer based on the geographic depth encoding data, retain the first fifty key features, and obtain the plot feature structured modeling data;

[0097] The differentiable selection layer includes a sampling generation layer and a dynamic pruning layer;

[0098] The sampling generation layer is used to generate feature selection probability masks, and the calculation formula is:

[0099]

[0100] Where m m is the selection probability of the mth feature, π m is the weight of the mth feature, g mis a disturbance variable that obeys the Gumbel (0,1) distribution, D is the total number of feature dimensions, n is the feature dimension index, τ is the smoothing control weight, π n is the weight of the feature of feature dimension n, g n is the disturbance variable corresponding to the feature of feature dimension n;

[0101] Specifically, during the model training process, the dynamic pruning layer removes features whose feature selection probability mask is lower than the selection threshold every five training cycles, and retains the top fifty key feature data.

[0102] By performing the above operations, this solution creatively adopts a differentiable feature selection method combined with geographic perception deep feature encoding to perform remote sensing feature modeling, addressing the technical problems of high reliance on manual feature engineering and difficulty in expressing spatial heterogeneity in the existing remote sensing feature modeling process. It realizes the automated, efficient and interpretable conversion of remote sensing raw data, which not only greatly reduces the cost of manual intervention, but also effectively improves the adaptability and generalization ability of the model in complex terrain and multi-source data backgrounds.

[0103] Example 3: This example is based on the above example. Figure 1 、 Figure 4 ,The temporal change detection is used to identify the spatiotemporal changes of land cover and utilization. Specifically, a dual-stream adaptive change detection method improved by ,combining change causal reasoning is adopted to perform temporal change detection and obtain ,reference data for land temporal change detection. This ,process includes the following steps: improved spatiotemporal causal graph construction, violation pattern matching, macro-interannual change detection, micro-burst change detection, dynamic dual-stream adaptive change weight fusion, and temporal change detection;

[0104] The improved spatiotemporal causal graph construction specifically comprises constructing spatiotemporal nodes by treating the change data of land cover and utilization as change events, and constructing spatiotemporal causal graph data by introducing the standard Granger causality test as the edge node;

[0105] The spatiotemporal node includes the time when the change event occurs, the coordinate position of the change, and the change intensity value characteristics;

[0106] The violation pattern matching is specifically to build a violation pattern matching database, perform violation pattern matching based on the spatiotemporal causal graph data, and obtain violation pattern matching reference data;

[0107] Preferably, Table 1 is a sample table of the violation pattern matching reference data, such as Table 1. The violation pattern matching database specifically includes violation types, threshold rules, spatial features, and temporal features; the violation types include illegal mining, illegal construction, deforestation and reclamation, illegal lake filling, garbage dumping, illegal sand mining, and non-grain use of cultivated land;

[0108] Table 1 Sample table of reference data for violation pattern matching

[0109]

[0110] The macro-interannual variation detection is specifically to construct a change detection channel for processing interannual variation data, specifically using a standard window t-test to perform macro-interannual variation detection and obtain macro-interannual variation detection reference data;

[0111] The micro-burst change detection specifically involves building a change detection channel for processing burst change event data, specifically using KL divergence-based anomaly detection to perform micro-burst change detection and obtain micro-burst change detection reference data;

[0112] The dynamic dual-stream adaptive change weight fusion is specifically performed by constructing a dynamic weighted fusion mechanism based on the macro-interannual change detection reference data and the micro-sudden change detection reference data, and obtaining a dynamic dual-stream adaptive change weight fusion model through model training;

[0113] The calculation formula of the dynamic weighted fusion mechanism is:

[0114] P t =σ(W α [h t-1 ,P(t)]+b α )·M t +(1-σ(W α [h t-1 ,P(t)]+b α ))·m t ;

[0115] Where, P t is the output of the dynamic two-stream adaptive change weight fusion model, σ(·) is the nonlinear activation function, W α is the long short-term memory neural network weight, h t-1 is the LSTM network output hidden state, P(·) is the seasonal cycle encoding vector, t is the time index, b α is the bias term of the long short-term memory neural network, M t is the reference data for detecting macro-interannual changes, m t It is the reference data for detecting microscopic sudden changes;

[0116] The time series change detection is specifically to perform time series change detection based on the dynamic dual-stream adaptive change weight fusion to obtain land time series change detection reference data.

[0117] By performing the above operations, in order to address the technical problems of insufficient change recognition accuracy and difficulty in tracing the causes of changes in the existing time-series change detection process, this solution creatively adopts a differentiable feature selection method combined with geographic perception deep feature coding to perform remote sensing feature modeling, construct a parallel detection mechanism for macro and micro change flows, and incorporates causal relationship maps and violation pattern matching logic to achieve multi-scale, multi-dimensional dynamic monitoring and causal tracing of surface cover and utilization changes, effectively improving the sensitivity of sudden change detection and the scientific nature of legitimacy analysis.

[0118] Example 4: This example is based on the above example. Figure 1 、 Figure 5 The ecological security assessment is used to quantify the regional ecological health status and risks. Specifically, the ecological security assessment is conducted by using an ecological risk propagation simulation method combined with an improved multi-layer perception ecological index to obtain ecological security index reference data, including the following steps: constructing an ecological index for the pressure layer, constructing an ecological index for the state layer, constructing an ecological index for the response layer, constructing a resistivity model, risk diffusion simulation, and comprehensive ecological security assessment;

[0119] The construction of the pressure layer ecological index specifically involves calculating the human activity intensity index based on the night light index data and the road density weighted score data to obtain reference data for the pressure layer ecological index;

[0120] The calculation formula for the human activity intensity index is:

[0121] HAI=w1·NTL std +w2·w2·RD norm ;

[0122] Where HAI is the human activity intensity index, w1 is the night light index weight, NTL std is the night light index data, w2 is the road density weight, RD norm is the normalized road density weighted score data;

[0123] Preferably, the specific value of the night light index weight w1 is 0.6, and the specific value of the road density weight w2 is 0.4;

[0124] The construction of the state-level ecological index is specifically to calculate the weighted score based on the biodiversity proxy indicator data and the landscape pattern index data to obtain the ecosystem integrity assessment data;

[0125] The calculation formula for the biodiversity proxy indicator data is:

[0126]

[0127] Where SVH is the biodiversity proxy indicator data, C is the total number of species habitat grid cells, c is the grid cell index, and ρ c is the species density index, is the mean species density;

[0128] The landscape pattern index data includes a spread index and a diversity index. The calculation formula of the spread index is:

[0129]

[0130] Where CONTAG is the landscape spread index, m′ is the total number of surface landscape types, i′ is the landscape type index, j′ is the second index of the landscape type, P i′ is the proportion of features in the i′th type of landscape, g i′j′ is the spatial adjacency frequency of the first landscape i′ and the second landscape j′;

[0131] The calculation formula of the diversity index is:

[0132]

[0133] Where SHDI is the Shannon diversity index of landscape, R is the total number of surface features, P r is the proportion of the rth type of land feature, r is the surface feature index;

[0134] The calculation formula for the weighted score calculation is:

[0135] EI=α·SVH+β·CONTAG+γ·SHDI;

[0136] Where, EI is the ecosystem integrity assessment data, α is the biodiversity weight, β is the landscape spread index weight, and γ is the landscape diversity weight;

[0137] Preferably, the specific value of the biodiversity weight α is 0.5, the specific value of the landscape spread index weight β is 0.3, and the specific value of the landscape diversity weight γ is 0.2;

[0138] The construction of the response layer ecological index specifically involves constructing a natural resilience model based on soil moisture time series data, calculating the natural recovery response layer ecological index, and obtaining reference data for the response layer ecological index;

[0139] The calculation formula for the ecological index of the natural recovery response layer is:

[0140]

[0141] Where R(t) is the natural resilience parameter, which is used to represent the reference data of the ecological index of the response layer, T′ is the total time, t is the time index, k is the natural recovery moment index in months, θ(k) is the time weight attenuation function, and NDVI is the time weight attenuation function. max is the historical maximum normalized difference vegetation index, NDVI(k) is the normalized difference vegetation index at the kth moment;

[0142] The resistivity model is constructed to simulate the difficulty of ecological risk propagation, specifically, a standard resistivity model is constructed to simulate ecological risk propagation to obtain a basic ecological risk propagation model;

[0143] The calculation formula of the standard resistivity model is:

[0144]

[0145] Where R ab is the probability value of ecological risk transmission simulation, d ab is the geographical distance between propagation area a and target area b, a is the propagation area identifier, b is the target area identifier, EI a is the ecosystem integrity assessment data of the dissemination area a, EI b is the ecosystem integrity assessment data of target area b, HAI a is the human activity intensity index of the propagation area a, HAI b is the human activity intensity index of target area b;

[0146] The risk diffusion simulation is specifically to conduct risk diffusion simulation based on the basic model of ecological risk propagation and adopt circuit theory to obtain risk diffusion simulation reference data;

[0147] The circuit theory abstracts the ecosystem into a circuit network model, treating each regional unit as a node in a resistance network. The ecological resistivity between different regions is used as the weight of the connecting edge. By constructing an ecological resistance network diagram based on the standard resistivity model, the transmission path and intensity of ecological risks between regions are simulated.

[0148] The comprehensive ecological security assessment is specifically to conduct a comprehensive ecological security assessment in combination with the risk diffusion simulation reference data to obtain ecological security index reference data;

[0149] The calculation formula for the comprehensive ecological security assessment is:

[0150] ESI=100-(40·HAI+30·(1-EI)+30·(1-Risk));

[0151] Where ESI is the ecological security index reference data, HAI is the human activity intensity index, EI is the ecosystem integrity assessment data, and Risk is the risk diffusion simulation reference data.

[0152] By performing the above operations, in response to the technical problems of a single indicator system and lagging static response in the existing ecological security assessment process, this plan creatively adopts an ecological risk propagation simulation method combined with an improved multi-layer perception ecological index to conduct ecological security assessment, realizing the fusion of multidimensional ecological indicators from pressure, state to response layer, and using resistivity modeling and circuit theory to spatially simulate the ecological risk diffusion process, significantly enhancing the dynamic expression ability and spatial propagation perception ability of ecological security assessment, and providing high-precision and systematic quantitative support for national land space ecological risk warning and governance.

[0153] Example 5: This example is based on the above example. Figure 1 The land space monitoring is used for comprehensive analysis and decision support, specifically, based on the land time series change detection reference data and the ecological security index reference data, combined with early warning decision analysis, to conduct comprehensive land space monitoring to obtain comprehensive land space monitoring data;

[0154] Preferably, Table 2 is a schematic table of the early warning decision analysis. As shown in the table, when the value of the ecological security index reference data is [80, 100], conventional monitoring and maintenance management measures are implemented; when the value of the ecological security index reference data is [60, 80), development degree restriction management measures are implemented; when the value of the ecological security index reference data is [0, 60), ecological restoration control management measures are implemented.

[0155] Table 2 Schematic diagram of early warning decision analysis

[0156]

[0157] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include," "comprise," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process or method comprising a set of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process or method.

[0158] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0159] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A land space monitoring system based on remote sensing data modeling, characterized by: It includes data collection module, remote sensing feature modeling module, temporal change detection module, ecological security assessment module and land space monitoring module; The data collection module is used for data collection, obtaining spatiotemporal reference remote sensing data through data collection, and sending the spatiotemporal reference remote sensing data to the remote sensing feature modeling module; The remote sensing feature modeling module uses a differentiable feature selection method combined with geographic perception deep feature coding to perform remote sensing feature modeling, obtains plot feature structured modeling data, and sends the plot feature structured modeling data to the temporal change detection module and the ecological security assessment module, including the following steps: multi-scale grid division, physical constraint feature generation, spatiotemporal context coding, geographic deep feature coding and differentiable feature selection; The temporal change detection module uses an improved dual-stream adaptive change detection method combined with change causal reasoning to perform temporal change detection, obtain land temporal change detection reference data, and send the land temporal change detection reference data to the land space monitoring module, including the following steps: improved spatiotemporal causal graph construction, violation pattern matching, macro-interannual change detection, micro-sudden change detection, dynamic dual-stream adaptive change weight fusion and temporal change detection; The ecological security assessment module is used for ecological security assessment, obtains ecological security index reference data through ecological security assessment, and sends the ecological security index reference data to the national land space monitoring module; The national land space monitoring module is used for national land space monitoring, and obtains national land space comprehensive monitoring data through national land space monitoring.

2. The land space monitoring system based on remote sensing data modeling according to claim 1, characterized in that: The data collection is used to collect and pre-process multi-source heterogeneous remote sensing data in a standardized manner, specifically to obtain spatiotemporal benchmark remote sensing data through multi-source heterogeneous remote sensing data collection; The time-space reference remote sensing data includes satellite remote sensing data, aerial remote sensing data and auxiliary data.

3. The land space monitoring system based on remote sensing data modeling according to claim 2, characterized in that: Remote sensing feature modeling is used to convert raw remote sensing data into feature vectors. Specifically, it adopts a differentiable feature selection method combined with geographic perception deep feature encoding to perform remote sensing feature modeling and obtain plot feature structured modeling data. The method includes the following steps: multi-scale grid division, physical constraint feature generation, spatiotemporal context encoding, geographic deep feature encoding and differentiable feature selection.

4. The land space monitoring system based on remote sensing data modeling according to claim 3 is characterized by: The multi-scale grid division constructs a three-level feature extraction network, performs feature extraction based on the spatiotemporal benchmark remote sensing data, and obtains plot feature structured modeling data; The three-level feature extraction network includes a macro grid layer, a middle grid layer and a micro grid layer; The land feature structured modeling data includes band reflectance features and digital elevation features; The physical constraint feature generation is specifically to obtain physical constraint feature data by extracting terrain features and spectral features based on the structured modeling data of the land parcel features; The physical constraint feature data includes terrain feature data and spectral feature data; The spatiotemporal context coding is specifically performed by aggregating neighborhood features using GraphSAGE and introducing seasonal periodic coding, and combining the physical constraint feature data to perform spatiotemporal context coding to obtain spatiotemporal coding data; The geographic deep feature coding specifically adopts a feature cross-fusion method combined with an improved geographic weighted multi-head attention to perform feature selection optimization, perform geographic deep feature coding, and obtain geographic deep coding data; The feature cross-fusion method combined with the improved geographically weighted multi-head attention includes a geographically weighted attention sub-block and a feature cross-fusion sub-block; The differentiable feature selection is specifically to construct a differentiable selection layer based on the geographic deep coding data, retain the first fifty key features, and obtain the plot feature structured modeling data.

5. The land space monitoring system based on remote sensing data modeling according to claim 4 is characterized in that: The temporal change detection is used to identify spatiotemporal changes in land cover and utilization. Specifically, it uses an improved dual-stream adaptive change detection method combined with change causal reasoning to perform temporal change detection and obtain reference data for land temporal change detection. The method includes the following steps: improved spatiotemporal causal graph construction, violation pattern matching, macro-interannual change detection, micro-sudden change detection, dynamic dual-stream adaptive change weight fusion, and temporal change detection. The improved spatiotemporal causal graph construction specifically comprises constructing spatiotemporal nodes by treating the change data of land cover and utilization as change events, and constructing spatiotemporal causal graph data by introducing the standard Granger causality test as the edge node; The spatiotemporal node includes the time when the change event occurs, the coordinate position of the change, and the change intensity value characteristics; The violation pattern matching is specifically to build a violation pattern matching database, perform violation pattern matching based on the spatiotemporal causal graph data, and obtain violation pattern matching reference data; The macro-interannual variation detection is specifically to construct a change detection channel for processing interannual variation data, specifically using a standard window t-test to perform macro-interannual variation detection and obtain macro-interannual variation detection reference data; The micro-burst change detection specifically involves building a change detection channel for processing burst change event data, specifically using KL divergence-based anomaly detection to perform micro-burst change detection and obtain micro-burst change detection reference data; The dynamic dual-stream adaptive change weight fusion is specifically performed by constructing a dynamic weighted fusion mechanism based on the macro-interannual change detection reference data and the micro-sudden change detection reference data, and obtaining a dynamic dual-stream adaptive change weight fusion model through model training; The time series change detection is specifically to perform time series change detection based on the dynamic dual-stream adaptive change weight fusion to obtain land time series change detection reference data.

6. The land space monitoring system based on remote sensing data modeling according to claim 5, characterized in that: The ecological security assessment is used to quantify the regional ecological health status and risks. Specifically, an ecological risk propagation simulation method combined with an improved multi-layer perception ecological index is used to conduct an ecological security assessment to obtain ecological security index reference data. The method includes the following steps: constructing a pressure layer ecological index, constructing a state layer ecological index, constructing a response layer ecological index, constructing a resistivity model, risk diffusion simulation, and comprehensive ecological security assessment.

7. The land space monitoring system based on remote sensing data modeling according to claim 6, characterized in that: The construction of the pressure layer ecological index specifically involves calculating the human activity intensity index based on the night light index data and the road density weighted score data to obtain the pressure layer ecological index reference data; The construction of the state-level ecological index is specifically to calculate the weighted score based on the biodiversity proxy indicator data and the landscape pattern index data to obtain the ecosystem integrity assessment data; The construction of the response layer ecological index specifically involves constructing a natural resilience model based on soil moisture time series data, calculating the natural recovery response layer ecological index, and obtaining reference data for the response layer ecological index; The resistivity model is constructed by constructing a standard resistivity model to simulate ecological risk propagation and obtain a basic ecological risk propagation model; The risk diffusion simulation is specifically to conduct risk diffusion simulation based on the basic model of ecological risk propagation and adopt circuit theory to obtain risk diffusion simulation reference data; The circuit theory abstracts the ecosystem into a circuit network model, treating each regional unit as a node in a resistance network. The ecological resistivity between different regions is used as the weight of the connecting edge. By constructing an ecological resistance network diagram based on the standard resistivity model, the transmission path and intensity of ecological risks between regions are simulated. The comprehensive ecological security assessment is specifically to conduct a comprehensive ecological security assessment in combination with the risk diffusion simulation reference data to obtain ecological security index reference data.

8. The land space monitoring system based on remote sensing data modeling according to claim 7, characterized in that: The land space monitoring is used for comprehensive analysis and decision support. Specifically, based on the land time series change detection reference data and the ecological security index reference data, comprehensive land space monitoring is carried out by combining multi-criteria decision analysis to obtain comprehensive land space monitoring data.

Images