A cultivated land resource analysis system based on multi-source spatio-temporal data fusion

Abstract

The application relates to the technical field of cultivated land resource evaluation and geographic information analysis, in particular to a cultivated land resource analysis system based on multi-source space-time data fusion, which comprises the following steps: receiving multi-source space-time data, extracting spatial adjacency relations and multi-dimensional attribute indexes, and constructing a space-time potential energy topology diagram mounted with a tensor; receiving a vector boundary to be occupied, positioning an occupied node subset and zeroing; calculating a tensor cascade attenuation value of a surrounding connected node based on a graph Laplace operator, and generating a potential energy residual matrix; converting a constraint condition into a directional impedance, and solving a supplementary node combination in a backup resource node library by adopting a gradient descent algorithm with constraints; generating a spatial layout scheme containing compensation land block coordinates and outputting a potential energy recovery trajectory diagram; and the cultivated land occupation and compensation are changed from a static area hedging into a topology repair process, so that only the quantity is supplemented and the function is not supplemented.

Description

Technical Field

[0001] This invention relates to the field of farmland resource evaluation and geographic information analysis technology, specifically a farmland resource analysis system based on multi-source spatiotemporal data fusion. Background Technology

[0002] With the development of land space governance, farmland protection supervision and natural resource information technology, farmland resource analysis for balancing the occupation and replenishment of permanent basic farmland has become an important application direction in the field of farmland protection. In order to accurately identify the impact of construction projects on land occupation and to formulate reasonable compensation plans, the modeling, correlation analysis and spatial extrapolation of farmland resources have become particularly important. In the process of reviewing the balance between cultivated land occupation and compensation, factors such as land quality, irrigation and drainage conditions, ecological connectivity, and spatial contiguousness all affect the effectiveness of compensation plans. However, existing analysis methods mainly rely on area statistics, layer overlay, or static land comparison, which can only determine how much has been compensated and are difficult to reflect the cascading impact of the proposed occupied land on the functions of surrounding cultivated land. They are also difficult to combine multi-source spatiotemporal data to comprehensively analyze the attribute matching, spatial recoverability, and policy constraints of the compensated land. When natural resources authorities conduct project reviews, there are still significant difficulties in forming an integrated processing mechanism that can support occupation identification, damage propagation analysis, and compensation layout simulation when faced with data from different regions, time periods, and types. Therefore, it is crucial to improve the scientific, realistic, and interpretable nature of the balance between cultivated land occupation and compensation by integrating existing multi-source spatiotemporal data such as remote sensing images, elevation information, soil sampling, and cultivated land patches, establishing an analytical model that can express the spatial adjacency and functional relationship between cultivated land units, further identifying the functional gap of cultivated land caused by project occupation, and solving the corresponding compensation plot layout scheme under constraints. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a farmland resource analysis system based on multi-source spatiotemporal data fusion. Specifically, the technical solution of this invention includes: The module includes a data mapping module, an event triggering module, a decay calculation module, an optimization solution module, and a decision output module. The data mapping module is used to receive multi-source spatiotemporal data, extract spatial adjacency relationships and multi-dimensional attribute indicators from the multi-source spatiotemporal data, delineate spatial polygons from the multi-source spatiotemporal data, extract the centroids of the spatial polygons to generate a set of nodes, establish an edge set between the set of nodes based on the spatial adjacency relationship, and convert the multi-dimensional attribute indicators into tensors and map them to the set of nodes to construct a spatiotemporal potential energy topology graph. The event triggering module communicates with the data mapping module to receive the proposed occupied vector boundary, locate the occupied node subset that has spatial intersection with the proposed occupied vector boundary in the spatiotemporal potential energy topology map, and set the tensor value of the occupied node subset to zero. The decay calculation module is connected to the event triggering module and is used to calculate the tensor cascade decay value of the occupied node subset to the surrounding connected nodes based on the graph Laplacian operator, and generate the potential energy residual matrix based on the tensor cascade decay value. The optimization solution module and the attenuation calculation module are connected to obtain the preset constraints, convert the constraints into directional impedances, and use a constrained gradient descent algorithm in the backup resource node library composed of unoccupied nodes in the spatiotemporal potential energy topology graph to solve for the supplementary node combination that makes up for the potential energy residual matrix, with the goal of minimizing the potential energy residual matrix. The decision output module communicates with the optimization solution module to generate a spatial layout scheme containing the coordinates of compensated plots based on the combination of supplementary nodes, and outputs a potential energy recovery trajectory map.

[0004] Preferably, the multi-source spatiotemporal data includes remote sensing image data, elevation matrix data, and soil sampling point sequences; the multi-dimensional attribute indicators include quality indicators, ecological indicators, and spatial indicators.

[0005] Preferably, the data mapping module includes a centroid extraction unit, a topology connection unit, and a tensor encoding unit. The centroid extraction unit is used to extract the centroids of spatial polygons from multi-source spatiotemporal data as a set of nodes. The topology connection unit is connected to the centroid extraction unit and is used to extract hydrological flow direction data based on elevation matrix data, extract spatial distance data based on the set of nodes, and establish an edge set based on the hydrological flow direction data and spatial distance data. The tensor encoding unit is connected to the topology connection unit and is used to encode quality indicators, ecological indicators, and spatial indicators into high-dimensional tensors, and attach the high-dimensional tensors to the set of nodes to generate a spatiotemporal potential energy topology map.

[0006] Preferably, the attenuation calculation module includes a diffusion equation solving unit and a state determination unit. The diffusion equation solving unit is used to construct the diffusion equation on the spatiotemporal potential energy topology using the graph Laplacian operator, and obtains the tensor cascade attenuation value by solving the diffusion equation. The state determination unit is connected to the diffusion equation solving unit and is used to mark the corresponding node as a severely damaged node when the tensor cascade attenuation value is greater than a preset attenuation threshold, and to include the tensor cascade attenuation value of the severely damaged node in the potential energy residual matrix. When the tensor cascade attenuation value is not greater than the preset attenuation threshold, the corresponding node is marked as a slightly damaged node, and a preset spatial smoothing filter is applied to filter the tensor cascade attenuation value of the slightly damaged node before including it in the potential energy residual matrix.

[0007] Preferably, the preset constraints include cross-regional compensation reduction coefficients and attribute balance constraints; the optimization solution module includes an impedance mapping unit, a residual input unit, and a gradient descent unit; the impedance mapping unit is used to map the cross-regional compensation reduction coefficients and attribute balance constraints into directed impedances of the spatiotemporal potential energy topology; the residual input unit is used to receive the potential energy residual matrix; the gradient descent unit is connected to the impedance mapping unit and the residual input unit, and is used to perform gradient descent operations along the low-impedance path of the spatiotemporal potential energy topology under the constraint of directed impedances to determine the supplementary node combination.

[0008] Preferably, the gradient descent unit is used to construct the potential energy residual minimization objective function; the potential energy residual minimization objective function includes a compensation term for the potential energy residual matrix and a penalty term for the directed impedance; continuous activation intensities are assigned to each backup resource node in the backup resource node library; and the gradient value of the potential energy residual minimization objective function with respect to the activation intensities of each backup resource node is calculated. The activation intensity of each backup resource node is updated based on the gradient value to determine the combination of supplementary nodes until the objective function of minimizing the potential residual converges.

[0009] Preferably, the decision output module includes a coordinate generation unit and a trajectory reconstruction unit; the coordinate generation unit is used to extract the coordinates of the compensation plots from the supplementary node combination and generate a spatial layout scheme; the trajectory reconstruction unit is connected to the coordinate generation unit and is used to record the tensor recovery time series data of each node in the spatiotemporal potential energy topology map during the solution process of the gradient descent algorithm, and to render the potential energy recovery trajectory map based on the tensor recovery time series data.

[0010] Preferably, the system further includes a tensor computation acceleration module; the tensor computation acceleration module includes a graphics processor or a neural network processor; the data mapping module, event triggering module, decay calculation module and optimization solution module are all communicatively connected to the tensor computation acceleration module; the tensor computation acceleration module is used to provide hardware acceleration support for the construction of the spatiotemporal potential energy topology graph, the calculation of tensor cascade decay values ​​and the solution of supplementary node combinations.

[0011] Preferably, the system also includes a graph database module; the graph database module is communicatively connected to the data mapping module and is used to persistently store the spatiotemporal potential energy topology graph; the graph database module is communicatively connected to the event triggering module and is used to respond to the query requests of the event triggering module and return the topological adjacency data of the occupied node subset.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This system extracts multi-source spatiotemporal data into spatial adjacency relationships and multi-dimensional attribute indicators, and transforms them into tensors that are mounted on nodes to construct a spatiotemporal potential energy topology map. This method integrates and expresses the originally isolated cultivated land patches into a topological network with actual physical connections, avoiding one-sided judgments caused by a single data source and truly reflecting the comprehensive capabilities of the cultivated land system.

[0013] 2. This system uses the graph Laplacian operator to calculate the tensor cascade attenuation value of the occupied node to its surroundings and generates a potential energy residual matrix. It can not only identify the direct loss of the occupied area, but also accurately quantify the damage to the cascade function of the surrounding plots caused by irrigation and drainage blockage, fragmentation, etc., and realize a true characterization of the actual potential energy gap.

[0014] 3. This system transforms constraints such as cross-regional compensation into directed impedances and uses a constrained gradient descent algorithm to solve for the combination of supplementary nodes. It can integrate policy requirements and spatial recoverability into the search process, and prioritize repairing the plots that can best close the gap and have the least implementation impedance, thus avoiding the problem of only satisfying static area statistics in traditional unguided search.

[0015] 4. This system generates a spatial layout scheme based on the combination of supplementary nodes and simultaneously renders and outputs a potential energy recovery trajectory map; it transforms the abstract algorithm results into a business output that can be plotted and reviewed, intuitively showing the process evidence of the gradual repair of damaged links, and improving the traceability and decision interpretability of the farmland compensation scheme.

[0016] 5. This system introduces a tensor computation acceleration module and a graph database module to accelerate topology computation in hardware and persistently store topology data to support local adjacency queries; it improves the computation response rate under large-scale underlying data and high-frequency iteration scenarios, and meets the business continuity requirements of cross-domain joint review and multiple boundary fine-tuning comparisons. Attached Figure Description

[0017] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of the structure of a farmland resource analysis system based on multi-source spatiotemporal data fusion in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] A farmland resource analysis system based on multi-source spatiotemporal data fusion is provided. The system includes: a data mapping module, which receives multi-source spatiotemporal data, extracts spatial adjacency relationships and multi-dimensional attribute indicators from the multi-source spatiotemporal data; delineates spatial polygons from the multi-source spatiotemporal data, extracts the centroids of the spatial polygons to generate a set of nodes, establishes an edge set between the node sets based on spatial adjacency relationships, and converts the multi-dimensional attribute indicators into tensors and maps them to the node sets to construct a spatiotemporal potential energy topology graph. The event triggering module communicates with the data mapping module. It is used to receive the proposed occupied vector boundary, locate the occupied node subset that has spatial intersection with the proposed occupied vector boundary in the spatiotemporal potential energy topology map, and set the tensor value of the occupied node subset to zero. The decay calculation module communicates with the event triggering module and is used to calculate the tensor cascade decay value of the occupied node subset to the surrounding connected nodes based on the graph Laplacian operator, and generate the potential energy residual matrix based on the tensor cascade decay value. The optimization solution module communicates with the attenuation calculation module to obtain preset constraints, transform the constraints into directional impedances, and use a constrained gradient descent algorithm in the backup resource node library composed of unoccupied nodes in the spatiotemporal potential energy topology graph to solve for the supplementary node combination that compensates for the potential energy residual matrix, with the goal of minimizing the potential energy residual matrix. The decision output module, which communicates with the optimization solution module, is used to generate a spatial layout scheme containing the coordinates of compensated plots based on the combination of supplementary nodes, and outputs a potential energy recovery trajectory map. This embodiment provides an implementation mechanism for a farmland resource analysis system based on multi-source spatiotemporal data fusion, such as... Figure 1 As shown; specifically, taking the business process of a provincial natural resources department reviewing the occupation of permanent basic farmland by the expansion project of the North-South Link Expressway as the main scenario, the system forms a complete closed loop around the identification of the land to be occupied—analysis of the spread of damage to surrounding farmland—reverse deduction of supplementary land plots—spatial layout output, in order to avoid the problem of traditional area balancing schemes that only supplement quantity but not function; The specific processing procedure is as follows: In the modeling stage, the data mapping module receives data such as cultivated land patches, irrigation and drainage canal systems, topographic elevation, soil sampling points, and historical remote sensing monitoring results in the county where the project is located; the system does not directly treat these data as isolated layers overlay, but first regards each cultivated land patch as a basic unit with agricultural production and ecological connections, and extracts the centroid of the corresponding polygon as a node; the edges between nodes not only represent geometrical adjacency, but can also represent actual connections such as upstream to downstream irrigation, farmland shelterbelt connection, and field accessibility; The system encodes attributes such as farmland quality, ecological stability, and spatial contiguousness at each node and attaches them to the node to form a spatiotemporal potential energy topology map. The potential energy here can be understood as the comprehensive capacity of the land parcel to undertake production capacity, conservation, connectivity, and policy adaptation in the existing regional farmland system. After the business side enters the proposed occupancy vector boundary, the event triggering module locates the nodes that intersect with the boundary in the topology map. For example, the three patches covered by the proposed occupancy range can be denoted as patch 1, patch 2, and patch 3. If they overlap with the proposed occupancy vector boundary entered by the business side, these nodes are identified as a subset of occupied nodes. The system sets the tensor values ​​corresponding to these nodes to zero. The physical meaning is not simply deleting the patches, but rather that these plots can no longer provide the original functions such as cultivation, irrigation, and ecological buffering. The attenuation calculation module not only calculates the directly lost area, but also further analyzes its cascading propagation impact on surrounding connected plots. For example, if the second plot is located between a field irrigation canal and a downstream contiguous paddy field, its occupation may lead to a decrease in arable land for adjacent but not directly occupied plots Q1 and Q2, even if they retain their plot boundaries, due to the interruption of the irrigation path. The system models the propagation of this topological disturbance based on the graph Laplacian operator, obtains the tensor cascading attenuation values ​​of surrounding nodes, and summarizes them to form a potential energy residual matrix. This residual matrix reflects the functional loss that has not yet been restored in the regional arable land system, rather than a single area numerical gap. During the supplementary analysis phase, the optimization solution module forms a backup resource node library from the unoccupied nodes and imports preset policy and business constraints; the system transforms requirements such as replacing paddy fields with paddy fields, prioritizing land for compensation, cross-regional compensation requiring conversion, and ensuring that ecologically sensitive area boundaries are not damaged after compensation into directed impedances; the lower the impedance, the more feasible the land is in terms of policy, more continuous in terms of ecology, and easier to form an effective alternative in terms of engineering. The system performs constrained optimization along the low-impedance path, no longer using a global search method without constraint guidance, but instead searching for supplementary node combinations in reverse around the location where the potential energy residual needs to be repaired; finally, the decision output module generates a spatial layout scheme containing the coordinates of the compensated plots and outputs a potential energy recovery trajectory map; this trajectory map can show which irrigation branches are restored, which downstream fields have reduced losses, and which fragmented areas are reconnected after the supplementary nodes are gradually activated; Anomaly and boundary condition handling mechanism: If the boundary to be occupied intersects with multiple patches only at small corners, or if the overlap area between the boundary to be occupied and multiple patches is lower than a preset normal threshold, the system first marks the intersection relationship below this scale as pending review; if the overlap area is not lower than the preset normal threshold, it is normally determined as a subset of occupied nodes to avoid misjudgment caused by vector boundary precision error. The methods for determining the minimum effective overlap scale include: obtaining the average boundary registration error between historical vector data and high-precision remote sensing images of the project area, and using a preset multiple of the average boundary registration error as the minimum effective overlap scale; or, the business end can directly specify the minimum effective overlap scale based on the average area of ​​cultivated land parcels in the area, according to a preset tolerance ratio. If there is a time inconsistency in the input data, such as remote sensing images from the current quarter while soil samples are from the previous year, the system will prioritize retaining the data with higher timeliness and use the data with lower timeliness as auxiliary attributes in the coding to prevent old data from dominating the current judgment. If local areas lack elevation or irrigation / drainage data, the system establishes edge relationships based solely on existing spatial adjacencies, while attaching a high uncertainty metric to the relevant edges and prompting manual verification in the output scheme. If the backup resource node library is insufficient to fully compensate for the residual matrix, the system outputs a partial recovery scheme and a warning that the gap is not completely closed, rather than forcibly providing supplementary results that do not conform to reality. In the above-mentioned high-speed expansion scenario, the proposed occupation vector boundary passes through the core paddy field area of ​​a certain irrigation district, directly occupying 6 cultivated plots, of which 2 plots are located at key control positions of the branch canal; after the event is triggered, the system calculates that several paddy field nodes downstream that were not directly occupied also show significant potential energy decay, because the irrigation and drainage connection and contiguous cultivation conditions are destroyed. The system selects a set of supplementary nodes that are more suitable for restoring the paddy field network from the western remediation reserve land of the same irrigation district, the potential reclamation land of neighboring villages, and the reserve plots within the county boundary, while not selecting scattered plots of similar size but located in dryland hilly areas; the output results also provide the coordinates of the compensation plots, the contiguous structure, and the visualization trajectory of the restoration process. To avoid multiple names for the same technical object throughout the document, this specification makes the following unified conventions in this embodiment: First, spatial adjacency relationship is used as a general term, which includes both geometric adjacency between map features and computable adjacency formed by irrigation and drainage connection, road accessibility, and topographic and hydrological relationships; in subsequent embodiments, the edge set is established based on hydrological flow direction data and spatial distance data, which is a refined extraction of spatial adjacency relationship, rather than the introduction of different concepts; Second, the descriptions of highway red line project land occupation boundaries, etc., appearing in the business scenario all refer to the proposed occupation vector boundary input by the business end; Third, supplementary node combination is the node-level result obtained by optimization solution, and the actual land plot corresponding to its mapping is uniformly referred to as compensation land plot in the output stage. The two correspond to the node layer and the land plot layer, respectively, and do not represent two different objects; The purpose of this step is to transform the issue of farmland occupation and compensation from a static area offset to a topological restoration process based on the linkage between quality, ecology and space, so as to achieve a true characterization of the impact of occupation and interpretable decision support for the replenishment layout.

[0020] Furthermore, the multi-source spatiotemporal data includes: remote sensing image data, elevation matrix data, and soil sampling point sequences; the multi-dimensional attribute indicators include: quality indicators, ecological indicators, and spatial indicators.

[0021] This embodiment provides an organization mechanism for multi-source spatiotemporal data and multi-dimensional attribute indicators. Specifically, continuing the aforementioned scenario of high-speed expansion and compensation review, under the premise that the basic closed loop can complete the compensation analysis, if only a single layer or single source data is relied upon, it is easy to misjudge plots that only meet the compensation conditions in appearance as actually recoverable plots. Therefore, remote sensing image data, elevation matrix data and soil sampling point sequences are further introduced, and the plot attributes are subdivided into quality indicators, ecological indicators and spatial indicators. The specific processing steps are as follows: Remote sensing image data mainly reflects the surface condition and temporal changes of cultivated land, such as crop cover, surface moisture content, signs of abandonment, and boundaries of contiguous cultivated areas, which is suitable for describing whether the plot is currently maintaining stable cultivation; Elevation matrix data is mainly used to reflect topographic relief, slope runoff, and low-lying water accumulation conditions, which is suitable for determining whether the plot can naturally support the irrigation and drainage system, and whether it is prone to soil erosion or flood pressure. Soil sampling point sequences reflect relatively stable background attributes such as soil layer thickness, organic matter, pH, and topsoil structure, which are suitable for identifying the long-term productivity base of a plot. The system maps these data from different sources and with different update frequencies onto the same cultivated land patch, so that each plot not only has apparent cover characteristics, but also integrates a comprehensive assessment of topographic suitability and soil carrying capacity. In terms of attribute organization, quality indicators may include arable land grade, soil fertility, topsoil integrity, and irrigation guarantee capacity, which reflect the production capacity base; ecological indicators may include water system connectivity, habitat buffer capacity, ecological sensitivity, and farmland protective barrier stability, which reflect the role of the plot in the regional ecological network; spatial indicators may include contiguousness, boundary regularity, connection strength to existing cultivated areas, and accessibility of roads and canals, which reflect whether the plot is suitable for forming an effective supplement in spatial layout. Through this classification, the system avoids treating plots with soil fertility indicators that meet the standards but spatial contiguousness below the threshold as the same as plots with area indicators that meet the standards but whose irrigation and drainage networks are not connected. If only remote sensing images are used, the condition of the plot may be misjudged during seasonal fallow or short-term waterlogging; if only elevation data is used, low-lying paddy fields may be confused with perennial waterlogged land; if only soil points are used, it is difficult to describe the actual contiguous pattern between patches; therefore, multi-source combination is actually to make up for the observational limitations of a single data source. Anomaly and boundary condition handling mechanism: If the remote sensing image of a certain area is insufficient in effective information due to cloud cover, the system can call previous and subsequent temporal images to supplement it; if it still cannot be supplemented, the weight of the surface condition related indicators of the area will be reduced and marked as pending on-site verification; if the soil sampling points are too sparse, the system will prioritize inheriting the existing farmland quality assessment results, and then supplement them with the interpolation results of similar surrounding landform units, but will indicate in the decision output that they are estimated values; if the elevation matrix resolution is too coarse and cannot accurately depict the micro-topography at the field ridge level, the system will control the hydrological connectivity analysis to the scale of ditches or areas to avoid drawing overly fragmented conclusions; In the eastern section of the irrigation area traversed by the high-speed expansion, remote sensing images showed that a candidate plot had stable planting traces for three consecutive years. However, the elevation matrix indicated that the plot was located in the transition zone between the plateau and the slope, with significant downcutting. Soil samples also showed that the topsoil thickness was below the preset standard. Based on this, the system determined that although the plot appeared to have signs of utilization, it was difficult to provide equivalent compensation for the occupied high-quality paddy fields. Conversely, another remediation reserve area located near the downstream canal system showed that its surface flatness met the preset threshold, its elevation conditions were conducive to water storage and irrigation, and its soil samples also had a high level of organic matter. Therefore, it was given higher substitution potential in subsequent analysis. The purpose of this step is to avoid geometric inconsistencies or biased judgments caused by a single data source by cross-verifying multi-source data and splitting multi-dimensional indicators, thereby achieving a more accurate representation of the functional status of arable land in accordance with the actual mechanisms of farmland.

[0022] Furthermore, the data mapping module includes: a centroid extraction unit, used to extract the centroids of spatial polygons from multi-source spatiotemporal data as a set of nodes; a topology connection unit, connected to the centroid extraction unit, used to extract hydrological flow direction data based on elevation matrix data, extract spatial distance data based on the set of nodes, and establish an edge set based on the hydrological flow direction data and spatial distance data; and a tensor encoding unit, connected to the topology connection unit, used to encode quality indicators, ecological indicators, and spatial indicators into high-dimensional tensors, and attach the high-dimensional tensors to the set of nodes to generate a spatiotemporal potential energy topology map.

[0023] This embodiment provides a refined implementation mechanism for a data mapping module. Specifically, in the aforementioned main scenario, if the polygons are stored as ordinary objects without establishing a unified organization method for nodes, edges, and attributes, the subsequent cascaded damage analysis and supplementary reverse inference will be difficult to implement. Therefore, a centroid extraction unit, a topology connection unit, and a tensor encoding unit are further introduced. The specific processing steps are as follows: The centroid extraction unit performs geometric standardization on cultivated land patches, unifies the coordinate system, corrects overlapping gaps, and performs centroid validity verification on narrow or concave patches. Normally, one centroid is extracted from each patch to serve as a node representative. However, if a patch spans significantly different irrigation control areas, terrain units, or administrative units, the system can split it according to preset rules to ensure that a node corresponds to a single agricultural function state as much as possible. The centroid is not for simplifying the graph, but to provide stable index anchors for subsequent topology solving. The topological connection unit establishes an edge set based on this. The establishment of edges is not based solely on a single matching connection according to the spatial distance threshold, but also takes into account the hydrological flow direction derived from the elevation matrix and the spatial distance between nodes. For farmland, two plots that are close in distance but have opposite uphill and downhill relationships may have weak actual irrigation and drainage connections. Conversely, plots that are spatially distant from the preset nearest neighbor threshold but are on the same branch canal control chain are given higher ecological and production connection weights by the system. Therefore, the system prioritizes connecting nodes with hydrological connection relationships and uses spatial distance as a supplementary condition to suppress unreasonable cross-connections. For example, we can set three map nodes A, B, and C. If the elevation and canal system indicate that the surface runoff and irrigation direction of A flows from B to C, then we can establish directed edges A→B and B→C. If A and C are close in a straight line but are cut off by embankments or roads, then we can not directly connect them. The edge set formed in this way is closer to the actual physical connectivity of farmland. The tensor encoding unit is responsible for organizing quality, ecological, and spatial indicators into high-dimensional tensors and attaching them to nodes. Here, a tensor can be understood as a hierarchical attribute container, such as a node containing multiple components like arable land grade, irrigation guarantee, ecological sensitivity, and contiguousness. The system encodes quality, ecological, and spatial indicators in a fixed order to ensure comparability between different nodes. If a node is superior in quality but isolated spatially, its tensor components exhibit differentiated characteristics. The system utilizes these differences to subsequently identify plots that can be supplemented but are not suitable for supplementation, and plots that are suitable for supplementation and should be prioritized for supplementation. Without the above-mentioned detailed processing, if the attributes of the map features are recorded in ordinary tables, it will be difficult to characterize the physical process of cascading decay of downstream related nodes after the key topology control nodes are occupied in subsequent analysis. After introducing these three units, the data mapping stage has unified the physical attributes, topological connections and computable representations of the map features. Anomaly and boundary condition handling mechanism: If the tortuosity of a certain patch boundary exceeds the set threshold and the geometric centroid falls outside the patch, the system uses an internal representative point to replace it, so as to avoid the node from deviating from the real plot; if there are local abnormal pits or unclear flow directions in flat areas in the elevation matrix, the system can introduce artificial canal system vectors to assist in correction, giving priority to respecting the actual direction of existing irrigation and drainage projects; if some indicators are temporarily missing, the tensor encoding unit can retain the empty space and add a missing mark, so as to reduce its participation in the subsequent solution, rather than forcibly filling unreliable values; In the irrigation area north of the expressway expansion project, after extracting the centroids of 12 adjacent cultivated land patches, the system determined that 4 patches were within the control range of the same distribution canal, while the other 2 patches, although geometrically close, were separated by rural roads and flood control ditches. Based on this, the topology connection unit only established edge relationships that were actually accessible for irrigation and drainage. The tensor encoding unit encoded the soil fertility level, irrigation accessibility, whether it was adjacent to the ecological isolation zone, and the degree of connection with existing high-standard farmland of each node into a unified structure, forming the spatiotemporal potential energy topology map of the area. The purpose of this step is to transform the scattered spatial data into a topological expression that reflects the real connections of farmland and is easy to process in subsequent calculations, thereby achieving a structured model of the functional state of the arable land system.

[0024] Furthermore, the attenuation calculation module includes: a diffusion equation solving unit, used to construct the diffusion equation on the spatiotemporal potential energy topology using the graph Laplacian operator, and obtain the tensor cascade attenuation value by solving the diffusion equation; and a state determination unit, connected to the diffusion equation solving unit, configured to: if the tensor cascade attenuation value is greater than a preset attenuation threshold, mark the corresponding node as a severely damaged node, and include the tensor cascade attenuation value of the severely damaged node in the potential energy residual matrix; otherwise, mark the corresponding node as a slightly damaged node, and apply a preset spatial smoothing filter to filter the tensor cascade attenuation value of the slightly damaged node before including it in the potential energy residual matrix.

[0025] This embodiment provides a refined implementation mechanism for the attenuation calculation module. Specifically, in the aforementioned structure, if the occupied patch itself is only regarded as the loss area, the secondary impact caused by irrigation and drainage blockage, fragmentation, and ecological boundary rupture is easily underestimated. Therefore, a diffusion equation solving unit and a state determination unit are further introduced to perform hierarchical propagation analysis on the impact caused by the occupation event. The specific processing procedure is as follows: The diffusion equation solving unit propagates the occupancy disturbance on the topological graph based on the graph Laplacian operator; its engineering meaning is that after a key patch is occupied, the impact will not remain within the boundary of the patch, but will be transmitted outward along hydrological, agricultural, and ecological connections; for irrigated farmland, this transmission often manifests as a gradual weakening of irrigation guarantee capacity, obstruction of mechanized contiguous operations, increased fragmentation, or a decline in downstream ecological buffering function; the system identifies which nodes are only affected by local disturbances and which nodes have already experienced substantial functional degradation through diffusion solving on the graph; For ease of understanding, a simplified model derivation can be made: If the occupied node is M, it has hierarchical connections with downstream N, O, and even further R; after the system is solved, it is calculated that the irrigation and drainage connectivity and spatial integrity of N decrease simultaneously, O mainly shows an increase in spatial fragmentation, while R only shows slight fluctuations; here, the calculation method is not emphasized, but the attenuation value represents the degree to which the original function of the plot is weakened. The state determination unit classifies relevant nodes into severely damaged and slightly damaged based on a preset attenuation threshold. Severely damaged nodes usually mean that even if the plot is not directly requisitioned, its agricultural function has been reduced to below the preset stability threshold, such as losing stable irrigation conditions or becoming an isolated small plot after being cut off by roads. The process of determining the preset attenuation threshold includes: extracting multi-source spatiotemporal data before and after historical farmland occupation events in this region, calculating the tensor concatenation attenuation value sequence of historical associated nodes before and after occupation, using a clustering algorithm to divide the sequence into two categories: high attenuation and low attenuation, and extracting the lower bound of the high attenuation category as the preset attenuation threshold; slightly damaged nodes are mostly characterized by local impact, short-term recovery, or edge effects caused by neighboring disturbances. The reason for performing spatial smoothing filtering on slightly damaged nodes is that slight attenuation is often mixed with boundary noise, temporal differences and local data anomalies; for example, short-term wet spots in remote sensing images and small misalignments of vector boundaries can cause individual nodes to show measurement deviations; spatial smoothing allows the system to determine whether it belongs to the part of the real damage based more on the overall surrounding pattern, thereby avoiding the inclusion of occasional anomalies as stable losses in the residual matrix. Specifically, a preset spatial smoothing filter is applied to filter the tensor cascade attenuation value of the slightly damaged node, including: obtaining all first-order adjacent nodes of the target slightly damaged node in the spatiotemporal potential energy topology graph, extracting the tensor cascade attenuation value of each first-order adjacent node, using the reciprocal of the spatial distance from each first-order adjacent node to the target slightly damaged node as the weight, performing a weighted average calculation on the above tensor cascade attenuation values, and using the calculated result as the filtered tensor cascade attenuation value of the target slightly damaged node; Furthermore, considering that the edge set in the aforementioned topological graph may contain both geometric adjacency and directed connections such as irrigation and drainage directions, the diffusion equation solving unit does not require all edges to degenerate into completely symmetric relationships when constructing the graph Laplace operator. For edges with a clear flow direction, the system retains their directional weights to express the strength of the influence from upstream to downstream and from the main canal to the branch canal. For edges that only reflect spatial adjacency without significant directionality, the system can use symmetric weights to participate in propagation. That is, the diffusion solution relies on weighted topological perturbation propagation rather than simple geometric diffusion, so that when key topological control nodes fail, downstream nodes are more likely to show higher attenuation values ​​than laterally adjacent nodes, avoiding the distorted judgment that the closest node will necessarily suffer the most damage. Furthermore, the tensor cascade decay values ​​can be propagated separately according to tensor components before being aggregated; in other words, quality indicators, ecological indicators, and spatial indicators do not need to share the exact same diffusion intensity; for example, the quality component and spatial connectivity component corresponding to irrigation and drainage blockage can propagate rapidly along the canal system link, while the ecological buffer component decays more slowly in the adjacent area; the system retains the component source mark when aggregating, so that the subsequent residual matrix can distinguish between production capacity gaps, ecological gaps, and contiguous gaps; in this way, the residual matrix is ​​not only a total loss table, but also reflects which type of functional decay dominates the loss; Without the stratification of heavy and light damage, the system may exhibit two biases: first, it may not be sensitive enough to critical blocking nodes, leading to an underestimation of the actual gap; second, it may overreact to slight edge fluctuations, causing the residual matrix to be amplified by noise. Through the above design, the residual matrix is ​​closer to the actual damage map of the farmland system. Anomaly and boundary condition handling mechanism: If a certain area has historical problems such as abandonment, disaster damage or disrepair of canal system before the occupation event occurs, the system first reads the baseline status before the event to avoid attributing all the original low-quality status to the current project occupation; if the attenuation distribution shows a large area of ​​isolated abnormal data points that are inconsistent with the actual topology, the system marks them as suspected data noise and prioritizes their inclusion in the slightly damaged path processing. If the attenuation level of a node is near the threshold, the system can make a secondary judgment by combining the continuity of the nearest neighbor and the historical monitoring results to prevent the boundary node from frequently oscillating between severe and mild conditions. If there is a ring irrigation and drainage connection or a two-way road in the local subgraph, the system sets an upper limit on the number of propagation rounds in the subgraph and combines the visited mark to suppress repeated amplification, so as to avoid the attenuation value being unreasonably accumulated due to the local closed loop structure. After the highway expansion line crosses the intersection of the irrigation canal and the branch road, the system identified two directly occupied map patches, and five downstream paddy field nodes showed varying degrees of functional degradation. Among them, two nodes adjacent to the intersection area were marked as severely damaged due to the loss of stable water supply and accessibility. The remaining three nodes, although affected by the road at the edges, could still maintain cultivation by bypassing the canal system, so they were marked as slightly damaged and included in the residual matrix after spatial smoothing. The residual matrix generated in this way not only reflects the serious consequences of the failure of the key nodes, but also avoids exaggerating the edge noise into a total loss. The purpose of this step is to differentiate and quantify the direct and cascading losses caused by occupancy events, thereby achieving potential gap identification that more closely approximates the actual degradation process of farmland.

[0026] Furthermore, the preset constraints include: cross-regional compensation conversion coefficients and attribute balance constraints; the optimization solution module includes: an impedance mapping unit, used to map the cross-regional compensation conversion coefficients and attribute balance constraints to the directed impedance of the spatiotemporal potential energy topology; a residual input unit, used to receive the potential energy residual matrix; and a gradient descent unit, connected to the impedance mapping unit and the residual input unit, used to perform gradient descent operations along the low-impedance path of the spatiotemporal potential energy topology under the constraint of the directed impedance, in order to determine the supplementary node combination.

[0027] This embodiment provides a refined implementation mechanism for optimizing the solution module. Specifically, the aforementioned residual matrix can already describe where and what is damaged. However, if subsequent supplementary selections do not consider cross-regional policies and attribute balance, the system may pick out a batch of plots with sufficient area but which do not comply with regulations or are difficult to form effective alternatives. Therefore, impedance mapping unit, residual input unit and gradient descent unit are further introduced. The specific processing steps are as follows: The residual input unit receives the potential energy residual matrix and identifies the type of function that most needs to be supplemented. For example, in this main scenario, if the occupied part is the core paddy field of the irrigation area, the residual is not only reflected in the area loss, but also in the lack of paddy field attributes, irrigation and drainage continuity and contiguous cultivation conditions. The impedance mapping unit converts various constraints into directed impedances in the topology graph. The so-called cross-regional compensation conversion factor means that if the supplementary land is located outside the administrative boundary, outside the irrigation area, or in a different agricultural functional area, its actual substitution effect and policy acceptance may be lower than that of supplementation in the same region, and therefore it will be given higher resistance. The so-called attribute balance constraint means that the key attributes of the occupied land must be compensated by similar attributes. For example, the occupied paddy fields should be compensated by land that can be stably restored to paddy fields, rather than simply replaced by sloping dry land. This impedance is not an abstract analogy, but a computational expression of real constraints. For example, the first, second, and third candidate nodes can be respectively associated with the ecologically sensitive buffer zone of general dry land in neighboring counties with the same irrigation area's remediation potential. Even if the three are similar in area, the first candidate node has a lower impedance due to its compatibility with the irrigation and drainage system of the damaged area and its high degree of attribute matching; the second candidate node has a higher impedance due to the cross-regional conversion requirement; and the third candidate node, although located nearby, will also be given a higher impedance due to its strong ecological constraints. The gradient descent unit gradually determines the supplementary node combination along the low impedance path based on this, which is essentially prioritizing the repair of the plots that can best close the potential energy gap and have the smaller constraints. Furthermore, the directionality of the directed impedance is reflected in two aspects: one is the alternative direction of the candidate supplement node pointing to the damaged residual location, with different directions representing different compensation links; the other is that the compensation efficiency of the same candidate node for different residual components is not consistent. For example, a certain plot has a strong ability to repair gaps in paddy field attributes, but has limited improvement on spatial continuity near ecologically sensitive boundaries. Therefore, its impedance pointing to the quality residual and the impedance pointing to the ecological residual can be set differently. After this processing, the system does not simply assign a single score to each candidate plot, but clarifies which type of gap and which area the plot is more suitable for repairing. Furthermore, to avoid processing bias caused by directly handling purely discrete selection or non-selection states in subsequent gradient updates, the gradient descent unit can first assign a range-based activation intensity to each candidate node before entering the formal iteration. This activation intensity is used to characterize the priority weight of the candidate node entering the final supplementary combination and is configured as a continuous real number variable with a value range between 0 and 1. The range-based activation intensity is configured as a continuous real number variable with a value range between 0 and 1, where 0 indicates that the node is not activated and 1 indicates that the node is fully activated and substantially included in the supplementary combination. After the iterations stabilize, the activation intensity is transcribed into the final supplementary node combination according to the preset threshold or sorting truncation rule. Therefore, the aforementioned gradient descent along the low impedance path can be understood as: first finding the optimal repair direction in the continuous priority space, and then falling back to the set of discrete plots that are executable in terms of business. This process maintains the gradient descent framework in the embodiment, while avoiding the misunderstanding of the optimization process as directly calculating the gradient of the discrete combination without differentiability. Without impedance mapping, the system is prone to degenerate into a coarse search rule that prioritizes area or availability, failing to reflect policy rigidity and attribute equivalence. By rewriting policy requirements as directed impedance, the system's search direction will naturally avoid high-risk, high-loss, and low-matching paths. Anomaly and boundary condition handling mechanism: If the cross-regional compensation rules differ in different cities, the system supports loading the corresponding rule version according to the project approval level; if the rules are not fully configured, cross-regional candidate plots will be upgraded to a cautious state by default, and will only be considered as alternatives, and will not be directly included in the priority sequence; if some backup node attributes have a high degree of matching but poor engineering implementation conditions, such as major irrigation and drainage renovation needs or unclear ownership issues, accessibility restrictions can be superimposed on the impedance. If all backup resources are in a high impedance state, the system outputs the optimal approximate solution under high constraints and prompts for manual verification, instead of producing a seemingly complete but difficult-to-execute result; if the activation intensity of multiple candidate nodes is close and they work together on the same damaged area, the system can add a combined verification before output to remove redundant nodes that repeatedly repair the same residual but do not contribute enough to other residuals, thus preventing local overcompensation and overall imbalance. In the high-speed expansion project, the two reserve lands available for remediation within the county are located in the same irrigation district and the hilly dryland area, respectively. There is also a large reserve area of ​​newly added arable land in a neighboring county. After receiving the residual matrix, the system determined that the most significant gaps in this project stemmed from the high-quality paddy field function and the contiguous canal system structure. After impedance mapping, the reserve land in the same irrigation district was assigned a lower impedance due to its better match with the balance requirements. Although the reserve area in the neighboring county had sufficient area, its impedance was significantly higher due to the cross-regional conversion and insufficient paddy field attributes. Ultimately, the gradient descent unit prioritized the reserve land in the same irrigation district and only considered cross-regional supplementary land plots when there were still unclosed residuals. The purpose of this step is to integrate policy legitimacy, attribute matching, and spatial recoverability into the supplementary search process, thereby achieving a constraint-based solution that moves from supplementable to correct supplementation.

[0028] Furthermore, the gradient descent unit is used to: construct a potential energy residual minimization objective function; the potential energy residual minimization objective function includes a compensation term for the potential energy residual matrix and a penalty term for the directed impedance; assign continuous activation intensities to each backup resource node in the backup resource node library; and calculate the gradient value of the potential energy residual minimization objective function with respect to the activation intensities of each backup resource node. The activation intensity of each backup resource node is updated based on the gradient value to determine the supplementary node combination until the objective function of minimizing the potential residual converges.

[0029] This embodiment provides a refined implementation mechanism for gradient descent units. Specifically, in the aforementioned structure, although candidate plots can be screened through low-impedance paths, without a clear target convergence mechanism, the system may repeatedly oscillate between multiple similar candidate schemes, or only focus on reducing impedance while ignoring the true repair of residuals. Therefore, objective function construction, gradient calculation, activation state update, and convergence control are further introduced. The specific processing procedure is as follows: The gradient descent unit constructs an objective function to minimize the potential energy residual; this objective function does not simply pursue the maximization of the supplementary area, but considers two directions at the same time: one is the compensation term, that is, whether the supplementary nodes can truly make up for the quality, ecological and spatial gaps reflected in the residual matrix after they are activated; the other is the penalty term, that is, the policy losses, cross-regional conversion burden, attribute imbalance risk and implementation resistance that need to be paid to activate these nodes. The practical significance of the objective function is to find an executable balance between the repair effect and the constraint cost; it includes a compensation term and a penalty term, where the penalty term is used to quantify the policy offset burden, attribute imbalance risk and project implementation damping. During the solution process, the gradient value can be understood as a directional indication of whether prioritizing the activation of a certain node would be more beneficial or detrimental to the overall recovery. The system gradually adjusts the activation state of the supplementary node combinations based on this direction. For example, candidate nodes can be denoted as U1, U2, U3, and U4, all of which are initially inactive. In the first round of evaluation, U1, once activated, can significantly repair the downstream paddy field irrigation and drainage connectivity residuals, and has low impedance, so it is prioritized to enter the activation state. Although U2 has a large area, its repair of key residuals is limited, so it is not activated for the time being. U3 can make up for attribute balance, but its cross-regional conversion burden is high, so it is considered as a subsequent candidate. U4 is located at the edge of the ecological buffer zone, and although it can locally make up for the area, its penalty term is large, so its priority is lowered. As the activation state is iteratively updated, the system continuously evaluates whether the remaining residuals are still worth paying the price for some high-impedance nodes, until the overall goal tends to stabilize, that is, the convergence state is reached. Furthermore, to make the relationship between gradient calculation and node combination selection clearer, the activation state can be represented as the continuous activation intensity of the candidate node during the iteration. The higher the value, the closer the node is to being formally included in the supplementary combination. In the output stage, the nodes that meet the threshold condition are then transcribed into discrete final combinations. Correspondingly, the objective function for minimizing the potential energy residual can be written as a tradeoff coefficient including a penalty term. expression The system solves for the function value. To evaluate the set of activation strengths of candidate nodes currently participating in the iteration. To comprehensively quantify the activation intensity set Unrepaired potential energy residual And the combined penalties resulting from cross-regional conversion, attribute imbalance, and implementation resistance. ; The potential energy residual R(A) is calculated as follows: the Frobenius norm of the difference between the potential energy residual matrix and the potential energy compensation provided by all candidate nodes under the current activation intensity set A is calculated; wherein, the potential energy compensation provided by all candidate nodes is obtained by superimposing the activation intensity of each candidate node and its corresponding tensor component after scalar multiplication; the comprehensive penalty C(A) is calculated as follows: the cross-regional compensation conversion coefficient, attribute mismatch index and implementation impedance index of each candidate node are linearly weighted and summed to generate a node-level penalty vector, and then the penalty vector is multiplied by the activation intensity set A. Therefore, the gradient value reflects a slight increase in the activation intensity of a candidate node, and the overall goal is to decrease or increase, thus giving the update direction a clear business meaning. Furthermore, the compensation term does not require a one-to-one correspondence between candidate nodes and residual matrix items. Instead, it allows multiple nodes to collaboratively activate and close a damaged link. For example, if one remediation area mainly restores paddy field properties and another adjacent area mainly restores contiguous structure and canal connection, then the sum of the two is needed to fully compensate for the potential energy gap in a certain area. The gradient descent unit recalculates the overall residual after each update, rather than only considering the independent contribution of a single node. Therefore, it can avoid the problem of single-point effectiveness leading to combined imbalance or single-point generality leading to combined optimality being misjudged. The reason for introducing convergence control is that farmland replenishment should not fall into endless, fragmented adjustments due to local optima; once the system determines that the residual improvement brought by the newly activated node is too small, or the improvement is not enough to offset the new penalty, the solution can be terminated; this termination does not mean that all gaps have been absolutely eliminated, but rather that the most reasonable balance has been reached under the current policy, resource and engineering conditions. Furthermore, convergence can be determined by combining two conditions: first, the decrease in the objective function between two adjacent rounds is less than a preset threshold, indicating that the optimization amount of the objective function for continued searching has significantly decreased; second, the core residual components have reached the acceptable range for business operations, indicating that the key gaps have been basically closed; when the first condition is met but the second condition is not met, the system can output the optimal approximate solution and simultaneously prompt the type of residual that has not been fully recovered, such as a small number of ecological connectivity gaps or cross-regional balance pressures that have not been resolved, so that approvers know the reason for the algorithm's cessation and the remaining problems. Anomaly and boundary condition handling mechanism: If the activation state switches frequently during the iteration process, it indicates that the substitution relationship between candidate nodes is very close. The system can introduce stability constraints and prioritize retaining the scheme that is more consistent with the existing contiguous structure. If the target cannot converge for a long time, it usually indicates that the constraints are too strict or the backup resources are insufficient. The system will output a non-convergence prompt and suggest relaxing the local constraints or supplementing the backup resource investigation. If a certain round of updates leads to a significant imbalance in some key attributes, such as the residual of paddy fields being incorrectly offset by dryland nodes, the system immediately rolls back the update and readjusts the influence strength of the attribute balance constraints. If a candidate node scores highly in continuous activation intensity, but its corresponding patch has real-world obstacles in terms of ownership, remediation cycle, or project timing, the system can downgrade it to a candidate node before discretization and mapping to ensure that the final output result is still feasible. In the supplementary simulation of the high-speed expansion project, the system initially considered three remediation sites within the county and two reserve sites in neighboring counties. After the iteration began, the first phase prioritized activating the two remediation sites connected to the original irrigation area by the same water system, which rapidly reduced the residual of the main paddy fields. The second phase assessment determined that if one reserve site in a neighboring county were to be activated, although the area would be sufficient, the improvement on the core residual would be limited after cross-regional conversion, and the penalty would increase significantly. Therefore, this site was not included in the final combination. Finally, after four rounds of updates, the system reached a stable state and output a set of supplementary node combinations that took into account both contiguousness and policy feasibility. The purpose of this step is to stabilize and form a combination of supplementary land parcels through a gradual convergence approach, thereby achieving an optimal choice that can both repair the potential energy gap and control policy and implementation costs.

[0030] Furthermore, the decision output module includes: a coordinate generation unit, used to extract the coordinates of the compensation plots from the supplementary node combination and generate a spatial layout scheme; and a trajectory reconstruction unit, connected to the coordinate generation unit, used to record the tensor recovery time series data of each node in the spatiotemporal potential energy topology map during the solution process of the gradient descent algorithm, and to render the potential energy recovery trajectory map based on the tensor recovery time series data.

[0031] This embodiment provides a refined implementation mechanism for the decision output module. Specifically, after the aforementioned solution is completed, if the system only outputs a list of recommended selected plots, it will be difficult for the approvers to understand why the supplementary plan is valid and how the restoration process occurs. It is also not conducive to subsequent collaboration with farmland protection, agriculture and rural affairs and water conservancy departments. Therefore, a coordinate generation unit and a trajectory reconstruction unit are further introduced. The specific processing procedure is as follows: The coordinate generation unit extracts the corresponding map patch boundary, center coordinates, administrative region, land type, and adjacent infrastructure information from the determined supplementary node combination to form a spatial layout plan that can be directly incorporated into the land and space business system. This plan is not just point coordinates, but can also include the area scope, the connection path with the original damaged area, and suggestions for the proposed remediation methods, such as restoring paddy fields, implementing land leveling, dredging irrigation canals, and connecting field roads. In this way, the approval personnel can directly see where each compensation plot is located and how it is connected with the existing farming pattern. The trajectory reconstruction unit records the temporal recovery information during the solution process; its core function is not to show the internal details of the algorithm, but to transform the reason why these locations were selected into evidence of how these locations gradually repaired the damaged system. For example, during several rounds of node activation, the system can record that the main irrigation and drainage connections are restored first, the conditions for contiguous farming are restored, and the local edge ecological buffers are improved. Based on this, the trajectory reconstruction unit generates a potential energy recovery trajectory map, which allows approvers to see intuitively how the paddy field properties of downstream damaged nodes recover once a certain reserve area is activated; and how the originally fragmented layout gradually transforms into a concentrated and contiguous area after another remediation area is included. The value of this output method lies in its interpretability; traditional supplementary solutions often provide results in the form of plot number + area, but it is difficult to prove that they can indeed replace the function performed by the occupied plot; through the trajectory map, business personnel can verify that the supplement is not a simple translation of the area, but rather a repair around the link with the most serious residual. Furthermore, to maintain terminology consistency, the output stage uniformly uses "compensated land parcel" to refer to the actual land parcel after the supplementary node combination is mapped, while the supplementary node combination is retained as the node-level name in the optimization solution stage; that is, the coordinate generation unit is responsible for completing the mapping from node-level results to land parcel-level results, the trajectory reconstruction unit still records the recovered time-series data of nodes in the spatiotemporal potential energy topology map, while the approval display interface presents it as the spatial unit of the compensated land parcel; through this correspondence, the problem of unclear object hierarchy caused by mixing supplementary land parcels and compensated land parcels can be avoided; Anomaly and boundary condition handling mechanism: If the final solution contains multiple adjacent nodes but the boundaries still need to be integrated, the coordinate generation unit can automatically perform adjacent plot fusion and retain the original node mapping relationship for easy subsequent filing; if some nodes in the restored trajectory have large temporal fluctuations and unstable interpretation, the trajectory reconstruction unit can perform phased merging of the display layer, retaining only the key restored nodes to avoid affecting the review and understanding due to overly detailed intermediate changes; if there are multiple alternative results with similar scores in the supplementary solution, the system can output the main solution and the alternative solution in parallel, with restored trajectories attached to each, to support joint review and consultation; In the highway expansion project, the system ultimately output the boundary coordinates of three compensation plots. Two of these plots are located in the remediation area west of the original irrigation district, and one is located in the planned high-standard farmland connection area south of the county. The trajectory map shows that after the first compensation plot was activated, the downstream paddy field nodes that were damaged due to the canal being cut off resumed irrigation. After the second compensation plot was activated, the farmland areas that were originally separated by the road re-formed continuous patches. After the third compensation plot was added, the overall quality and spatial balance tended to stabilize. Based on this, the approval personnel were able to understand the restoration role undertaken by each compensation plot. The purpose of this step is to transform the optimization results into business outputs that can be mapped, reviewed, and traced, thereby enabling the spatial implementation of the compensation plan and the interpretable demonstration of the recovery process.

[0032] Furthermore, the system also includes: a tensor computation acceleration module; the tensor computation acceleration module contains a graphics processor or a neural network processor; the data mapping module, event triggering module, decay calculation module, and optimization solution module are all communicatively connected to the tensor computation acceleration module; the tensor computation acceleration module is used to provide hardware acceleration support for the construction of the spatiotemporal potential energy topology graph, the calculation of tensor cascade decay values, and the solution of supplementary node combinations.

[0033] This embodiment provides a tensor computation acceleration mechanism. Specifically, in the aforementioned main scenario, if the project only involves a small number of townships, a general-purpose processor can complete the analysis. However, in provincial review or cross-city linkage projects, the base map size and the number of candidate nodes will increase rapidly. If ordinary serial computation is still used, it will easily lead to insufficient processing time and make it difficult to meet the response speed requirements of scheme comparison and consultation review. Therefore, a tensor computation acceleration module is introduced. The specific processing steps are as follows: The tensor computation acceleration module can be composed of a graphics processor or a neural network processor, which communicates with modules such as data mapping, event triggering, decay calculation and optimization solution; For this system, the purpose of acceleration is not simply to increase the single operation speed, but because the spatiotemporal potential energy topology graph contains a large number of node attributes, edge relationships and cross-round iteration states, especially in the decay propagation and supplementary combination search stages, there is a high frequency of parallel update requirements. Graphics processors are suitable for parallel computation of large-scale nodes and edges in batch processing; neural network processors are suitable for performing tensor attribute encoding and batch state updates; through hardware acceleration, the system can quickly return new analysis results when the proposed occupancy boundary, policy parameters or supplementary constraints are adjusted multiple times on the business side, supporting a working mode of reviewing and comparing simultaneously. For example, in a simplified sandbox, if the candidate node library contains U1 to U10000, and each round of solution requires evaluating the repair effect of these nodes on multiple residual components, the ordinary serial approach requires processing each node sequentially; the acceleration module can load these nodes in batches in parallel and complete the state update; another example is that once the occupancy boundary is finely adjusted, the affected adjacent links also need to be recalculated. Hardware acceleration can make this recalculation limited to the relevant subgraph and completed quickly, without waiting for the entire base graph to be refreshed. Introducing an acceleration module can also reduce the risk of system interaction blockage in complex scenarios. For natural resources authorities, a joint review meeting often requires comparing multiple route options or various supplementary combinations at the same time. If each response takes too long, it will weaken the system's usability. Anomaly and boundary condition handling mechanism: If the acceleration hardware is unavailable, the driver is faulty, or the video memory is insufficient, the system automatically switches to CPU compatibility mode and prioritizes completing local analyses with smaller scopes; at the same time, it prompts the user that the current operation is in a degraded state; if the area to be processed exceeds the single video memory capacity, the system can slice it into administrative grids or watershed units and calculate it in segments, and keep the node identifiers consistent when summarizing the results to avoid loss of cross-regional communication due to slicing; if the communication between the acceleration module and the business module is interrupted, the system retains the most recent stable graph structure snapshot to prevent the results from being unreproducible due to interruption in the task; When reviewing alternative routes for highway expansion projects at the provincial level, business personnel revised the proposed red line three times to compare whether it would avoid a certain high-grade paddy field. With the support of the acceleration module, the system can quickly reconstruct the local potential energy map, recalculate the propagation of damage in the surrounding area, and generate new supplementary combinations, enabling the consultation personnel to complete multiple rounds of scheme screening within the same working period. If some acceleration nodes are offline for maintenance on the same day, the system will automatically switch to degraded mode, only recalculating the changed county sub-maps and retaining the original global base map results. The purpose of this step is to provide stable parallel computing support for large-scale farmland topology analysis, thereby enabling interactive solution and ensuring business continuity in highly complex scenarios.

[0034] Furthermore, the system also includes: a graph database module; the graph database module is communicatively connected to the data mapping module and is used to persistently store the spatiotemporal potential energy topology graph; the graph database module is communicatively connected to the event triggering module and is used to respond to the query requests of the event triggering module and return the topological adjacency data of the occupied node subset.

[0035] This embodiment provides a graph database persistence and query mechanism. Specifically, after the aforementioned system can complete accelerated calculation, if the topology graph is temporarily reconstructed from the original layer for each project review, redundant calculation overhead will be generated, and it will not be conducive to the review of cross-batch projects, the tracking of historical schemes, and local queries. Therefore, a graph database module is further introduced to persistently store the spatiotemporal potential energy topology graph and return topological adjacency data to the event triggering module. The specific processing procedure is as follows: The graph database module stores the constructed spatiotemporal potential topology graph in a node-edge-attribute structure; each node can correspond to a cultivated land patch unit, saving its geometric identifier, administrative region, quality attribute, ecological attribute, spatial attribute, and latest temporal label; each edge records information such as hydrological connections, spatial proximity relationships, and policy resistance directions between nodes; compared with ordinary relational tables, graph databases are more suitable for storing and querying multi-hop adjacency links, especially suitable for business needs in this system that require quickly finding the first-order, second-order, or even more distant connectivity influence range after occupying a node; When the event triggering module receives a vector boundary to be occupied, it can first quickly locate the intersecting nodes based on the spatial index, and then initiate an adjacency query to the graph database to request the return of the upstream and downstream nodes, parallel nodes, and related edge attributes of these nodes. The practical value of doing this is that the system does not need to re-traverse the entire base graph every time it is triggered, but can directly extract the local subgraph from the stored topology for analysis. For example, if the occupied nodes are K1 and K2, the graph database can immediately return the directly connected nodes L1, L2, and L3, as well as the adjacent nodes and edge attributes of the further affected nodes M1 and M2, for the subsequent decay calculation module to continue to use. Persistent storage also benefits project lifecycle management; once the same area is subject to multiple occupancy reviews in different years and different projects, the system can compare the topology changes before and after, identify the cumulative damaged areas and the recovered areas, and avoid duplicate occupancy, duplicate counting or mutual offsetting between different projects. Furthermore, to ensure consistency between module relationships and terminology, the topological adjacency data returned by the graph database module uniformly refers to the query results of node sets, edge sets, and their edge attributes on the local subgraph, where edge attributes may include hydrological flow direction, spatial distance weight, uncertainty marker, and directional impedance-related fields. The event triggering module only reads the subset of occupied nodes that intersect with the boundary of the currently intended occupied vector and their outward adjacent links, without changing the basic topology definition in the graph database; thus, the graph database module undertakes the functions of persistence and query reuse, while the attenuation calculation module and the optimization solution module continue to complete the cascading propagation and supplementary solution on this local subgraph, and the responsibilities before and after remain consistent. Anomaly and boundary condition handling mechanism: If some node versions in the graph database are older, while the original base map has been updated, the system will include a version stamp when returning the query results; when the version inconsistency reaches a preset level, the event triggering module will first trigger a local reconstruction before performing formal analysis; if the adjacency query result is empty, the system will not directly determine that the node has no impact, but will first check whether it is caused by incomplete edge relationships, unmerged data slices, or failed boundary correction; if necessary, the basic topology connectivity mode based on the original spatial adjacency will be called; if the database write fails, the system will at least retain the most recently successfully constructed topology snapshot and mark the current task as temporary to prevent historical link breakage; During the review phase of the high-speed expansion project, the staff compared the proposed occupied vector boundaries with the initial review and the revised ones. The system directly retrieved the existing spatiotemporal potential energy topology map of the irrigation area from the graph database, and quickly returned the occupied nodes and their upstream and downstream adjacency relationships after the new proposed occupied vector boundaries were entered. Since the channel connections, land parcel attributes and policy resistance of the area had been persistently saved in the previous review, it was not necessary to rebuild the entire county map from scratch this time. Only the information of the two recently renovated land parcels needed to be updated locally to regenerate the attenuation analysis and compensation layout results. The purpose of this step is to reduce the computational resource consumption of repeated map building and improve event response efficiency by persistent storage of the topology map and local query reuse, thereby achieving continuity and traceability of farmland occupation and compensation analysis under multiple projects and multiple time phases.

[0036] 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A farmland resource analysis system based on multi-source spatiotemporal data fusion, characterized in that, The system includes: The data mapping module is used to receive multi-source spatiotemporal data, extract spatial adjacency relationships and multi-dimensional attribute indicators from the multi-source spatiotemporal data; delineate spatial polygons from the multi-source spatiotemporal data, extract the centroids of the spatial polygons to generate a set of nodes, establish an edge set between the set of nodes based on the spatial adjacency relationships, and convert the multi-dimensional attribute indicators into tensors and map them to the set of nodes to construct a spatiotemporal potential energy topology graph. The event triggering module, which is communicatively connected to the data mapping module, is used to receive the proposed occupancy vector boundary, locate the occupied node subset that has spatial intersection with the proposed occupancy vector boundary in the spatiotemporal potential energy topology map, and set the tensor value of the occupied node subset to zero. The decay calculation module is communicatively connected to the event triggering module and is used to calculate the tensor cascade decay value of the occupied node subset to the surrounding connected nodes based on the graph Laplacian operator, and generate the potential energy residual matrix based on the tensor cascade decay value. The optimization solution module is connected to the attenuation calculation module to obtain preset constraints, convert the constraints into directional impedance, and use a constrained gradient descent algorithm in the backup resource node library composed of unoccupied nodes in the spatiotemporal potential energy topology graph to solve for the supplementary node combination that makes up for the potential energy residual matrix, with the goal of minimizing the potential energy residual matrix. The decision output module is communicatively connected to the optimization solution module and is used to generate a spatial layout scheme containing the coordinates of the compensated plots based on the supplementary node combination, and output the potential energy recovery trajectory map.

2. The farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The multi-source spatiotemporal data includes: remote sensing image data, elevation matrix data, and soil sampling point sequences; The multidimensional attribute indicators include: quality indicators, ecological indicators, and spatial indicators.

3. The farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 2, characterized in that, The data mapping module includes: The centroid extraction unit is used to extract the centroids of spatial polygons from the multi-source spatiotemporal data as the node set. A topology connection unit, connected to the centroid extraction unit, is used to extract hydrological flow direction data based on the elevation matrix data, extract spatial distance data based on the node set, and establish the edge set based on the hydrological flow direction data and the spatial distance data. The tensor encoding unit, connected to the topology connection unit, is used to encode the quality indicators, ecological indicators, and spatial indicators into high-dimensional tensors, and to attach the high-dimensional tensors to the node set to generate the spatiotemporal potential energy topology graph.

4. The farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 3, characterized in that, The attenuation calculation module includes: The diffusion equation solving unit is used to construct the diffusion equation on the spatiotemporal potential energy topology using the graph Laplacian operator, and to obtain the tensor cascade decay value by solving the diffusion equation. The state determination unit, connected to the diffusion equation solving unit, is configured to: if the tensor cascade decay value is greater than a preset decay threshold, mark the corresponding node as a severely damaged node and include the tensor cascade decay value of the severely damaged node in the potential energy residual matrix; otherwise, mark the corresponding node as a slightly damaged node, and apply a preset spatial smoothing filter to filter the tensor cascade decay value of the slightly damaged node before including it in the potential energy residual matrix.

5. A farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The preset constraints include: cross-regional compensation conversion factor and attribute balance constraint. The optimization solution module includes: Impedance mapping unit, used to map the cross-regional compensation conversion factor and the attribute balance constraint condition to the directed impedance of the spatiotemporal potential energy topology graph; The residual input unit is used to receive the potential energy residual matrix; The gradient descent unit, connected to the impedance mapping unit and the residual input unit, is used to perform a gradient descent operation along the low-impedance path of the spatiotemporal potential energy topology graph under the constraint of the directed impedance, in order to determine the supplementary node combination.

6. A farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 5, characterized in that, The gradient descent unit is used for: Construct an objective function that minimizes the potential energy residual; The objective function for minimizing the potential energy residual includes a compensation term for the potential energy residual matrix and a penalty term for the directed impedance; Assign continuous activation intensities to each backup resource node in the backup resource node library; calculate the gradient value of the potential residual minimization objective function with respect to the activation intensities of each backup resource node; The activation intensity of each backup resource node is updated based on the gradient value to determine the supplementary node combination until the objective function of minimizing the potential residual converges.

7. A farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The decision output module includes: A coordinate generation unit is used to extract the coordinates of the compensation plot from the supplementary node combination and generate the spatial layout scheme. The trajectory reconstruction unit, connected to the coordinate generation unit, is used to record the tensor recovery time series data of each node in the spatiotemporal potential energy topology graph during the solution process of the gradient descent algorithm, and to render the potential energy recovery trajectory graph based on the tensor recovery time series data.

8. A farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 1, characterized in that, The system also includes: a tensor computation acceleration module; The tensor computation acceleration module includes a graphics processor or a neural network processor. The data mapping module, the event triggering module, the decay calculation module, and the optimization solution module are all communicatively connected to the tensor calculation acceleration module. The tensor computation acceleration module provides hardware acceleration support for the construction of the spatiotemporal potential energy topology graph, the calculation of the tensor cascade decay value, and the solution of the supplementary node combination.

9. A farmland resource analysis system based on multi-source spatiotemporal data fusion according to claim 8, characterized in that, The system also includes: a graph database module; The graph database module is communicatively connected to the data mapping module and is used to persistently store the spatiotemporal potential energy topology graph; The graph database module is communicatively connected to the event triggering module and is used to respond to the query request of the event triggering module and return the topological adjacency data of the occupied node subset.

Images