Cultivated land-forest land dynamic monitoring and layout optimization system and method fusing multi-source data

Abstract

The invention relates to the technical field of land resource management, and discloses a cultivated land-forest land dynamic monitoring and layout optimization system and method fusing multi-source data, and the system comprises a data layer, a fusion layer, a monitoring layer, an optimization layer and an application layer which are sequentially linked to form full-flow closed-loop management and control. The data layer is used for collecting multi-source data and completing standardized preprocessing; the fusion layer adopts a two-stage fusion strategy to eliminate data heterogeneity so as to generate a cultivated land-forest land comprehensive data set; the monitoring layer realizes cultivated land-forest land dynamic change real-time identification, quantitative measurement and relevance establishment based on a deep learning model; the optimization layer is used for constructing a multi-objective quantitative optimization model and establishing a dynamic updating mechanism to output a layout optimization scheme; and the application layer is used for providing visual display, decision support and early warning push functions. The problems that an existing algorithm is limited in scene adaptability and unsmooth in connection with actual application requirements are solved, monitoring accuracy and real-time performance are improved, and layout optimization is more universal and scientific.

Description

Technical Field

[0001] This invention relates to the field of land resource management technology, specifically to a system and method for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data. Background Technology

[0002] The dynamic balance between arable land and forest land is crucial to food security and ecological stability, and is the core of national land spatial planning. With the advancement of the precision management strategy, related monitoring and layout optimization technologies have evolved from traditional on-site surveys and experience-based decision-making to an intelligent system combining remote sensing, GIS, and algorithmic models, becoming a key support for natural resource supervision and ecological protection and restoration. Current technological development focuses on two main directions: first, upgrading dynamic monitoring of arable land quality to multi-source data fusion and visualization, integrating remote sensing, soil, and meteorological data to achieve quality quantification and dynamic display; second, optimizing the spatial layout of arable land focusing on map patch adaptability, partially extending to cross-land-category management of arable and forest land. However, existing technologies still have significant limitations and cannot meet the needs of collaborative management throughout the entire process.

[0003] For example, patent CN121073304A proposes a dynamic accounting system for the exchange of cultivated land and forest land resources. Although it achieves cross-land category management, its core limitation is that it only focuses on the accounting of resource exchange indicators and the correction of anomalies. It fails to achieve a full-time dynamic monitoring system, cannot accurately capture real-time changes and long-term evolution trends of cultivated land and forest land, and does not form a closed loop throughout the entire process, making it difficult to support collaborative development decisions.

[0004] Current technologies for monitoring and optimizing the layout of cultivated forests face numerous technical bottlenecks. Existing technologies often focus on single land types or are limited to cross-land type indicator calculations, failing to reflect the overall ecosystem. Furthermore, there is a general disconnect between monitoring and optimization; optimization schemes are mostly static outputs, lacking dynamic adjustment mechanisms based on real-time changes, making them difficult to adapt to external environmental changes. Moreover, most technologies involve simple overlaying of multi-source data, failing to adequately cover data from the cultivated forest transition zone, resulting in insufficient identification accuracy. Layout optimization also tends to focus on single objectives, lacking quantitative models for multi-objective collaboration, and some still rely on expert subjective judgment and experience, making the optimization schemes lack scientific rigor and universality. In addition, existing technologies are often designed for specific regions, failing to consider the heterogeneity of complex terrain, resulting in poor integration with practical application needs. Summary of the Invention

[0005] The present invention aims to provide a system and method for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data, in order to solve the problems that existing technologies cannot meet the needs of precise control of national land space, and are prone to causing imbalance in cultivated land and forest protection, unrealistic optimization schemes, low decision-making efficiency, and difficulty in promotion in complex areas.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a dynamic monitoring and layout optimization system for cultivated land and forest land integrating multi-source data, comprising a data layer, a fusion layer, a monitoring layer, an optimization layer, and an application layer that are linked sequentially to form a closed-loop control of the entire process; the data layer is used to collect multi-source data and complete standardized preprocessing; the fusion layer uses a two-level fusion strategy to eliminate data heterogeneity to generate a comprehensive cultivated land-forest land dataset; the monitoring layer uses a deep learning model to realize real-time identification, quantitative calculation, and correlation establishment of dynamic changes in cultivated land and forest land; the optimization layer is used to construct a multi-objective quantitative optimization model and establish a dynamic update mechanism to output layout optimization schemes; the application layer is used to provide visualization, decision support, and early warning push functions.

[0007] Meanwhile, this solution also provides a method for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data, applied to the aforementioned system for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data, including the following steps: S1. Collect multi-source data, construct a multi-source data system covering relevant dimensions of cultivated land and forest land, and complete the standardization process; S2 employs a two-level fusion strategy to eliminate data heterogeneity and generate a comprehensive farmland-forest land dataset. S3, based on a deep learning model, enables real-time identification, quantitative calculation, and correlation establishment of dynamic changes between cultivated land and forest land; S4, construct a multi-objective quantitative optimization model and combine it with a dynamic update mechanism to output a layout optimization scheme; S5 generates land parcel-level implementation paths and provides decision support through visualization.

[0008] The principles and advantages of this scheme are: This scheme integrates multi-source data, eliminates format and spatial heterogeneity through standardized preprocessing, and then employs a two-level fusion strategy. It utilizes an improved CNN-attention mechanism to extract deep-level correlation features and dynamically assigns weights. Combining DS evidence theory with multi-model decision-making results, it generates a high-precision integrated farmland and forestry dataset. Based on this dataset, it introduces differential image enhancement features into an improved CNN-LSTM time-series model to achieve real-time identification of dynamic changes in farmland and forestry. Simultaneously, it constructs coupling coordination and competition intensity models to quantify the ecological coupling and resource competition relationship between the two. Finally, with coordinated development as the core, it uses an improved NSGA-Ⅲ algorithm to construct a multi-objective quantitative optimization model encompassing ecological, economic, and social objectives. Combined with dynamic monitoring results, it establishes a real-time update mechanism for the scheme, outputting plot-level optimization implementation paths adapted to different scenarios.

[0009] Therefore, the advantage of this solution is: First, the multi-source data fusion strategy overcomes the limitations of a single data source. Through deep feature fusion and dynamic weight allocation, it significantly improves the accuracy of cultivated land classification and transition zone identification, enhancing data support capabilities in complex scenarios. Second, the full-process dynamic monitoring system achieves integration from change identification to correlation quantification, shortening the monitoring update cycle to 1-3 days. It can accurately capture short-term mutations and provide real-time data support for collaborative management. Third, the multi-objective dynamic optimization model takes into account ecological protection, economic development, and social needs. Combined with a closed-loop update mechanism, it ensures the scientific nature and dynamic adaptability of the optimization scheme, while improving the practicality of the technology through plot-level implementation path design. Fourth, by integrating the coupling coordination degree and competition intensity model throughout the entire process, it strengthens the characterization of cultivated land correlation and achieves an upgrade from isolated monitoring to collaborative management. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the structure of the farmland-forest land dynamic monitoring and layout optimization system that integrates multi-source data according to the present invention; Figure 2 This is a flowchart illustrating the method for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data according to the present invention. Detailed Implementation

[0011] The following detailed description illustrates the specific implementation method: The system and method for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data in this embodiment achieves precise and coordinated management and control of cultivated land and forest land by constructing a closed-loop technical link of "multi-source data fusion - dynamic collaborative monitoring - multi-objective dynamic optimization". It solves the core problems of existing technologies, such as low degree of single data integration, insufficient monitoring timeliness and accuracy, lack of cultivation-forest correlation mining, lack of quantitative support and dynamic adaptability for layout optimization, and limited applicability of algorithms to various scenarios.

[0012] Option 1 Provides a system for dynamic monitoring and layout optimization of cultivated land and forest land that integrates multi-source data, as shown in the attached figure. Figure 1 As shown, it includes a data layer, a fusion layer, a monitoring layer, an optimization layer, and an application layer that work together to form a closed-loop control system for the entire process.

[0013] The data layer is used to collect multi-source data, build a multi-source data collection system, and complete standardized preprocessing to achieve comprehensive coverage and standardized preprocessing of data related to cultivated land and forest land.

[0014] In this embodiment, the multi-source data includes six major categories: remote sensing data, topographic data, soil data, meteorological and hydrological data, human and social data, and ecological data, ensuring the comprehensiveness and relevance of the data. Among them, the remote sensing data includes at least high-resolution optical remote sensing data (such as Gaofen-1 / 2 / 6, spatial resolution 0.8-2m, temporal resolution 1-5 days), synthetic aperture radar (SAR) data (such as Sentinel-1, spatial resolution 5-20m, all-weather penetration), hyperspectral remote sensing data (such as Gaofen-5, to obtain vegetation spectral characteristics), and nighttime light remote sensing data (such as NPP-VIIRS, reflecting the intensity of human activities).

[0015] Topographic data includes digital elevation model (DEM) data (such as ASTER GDEM, 30m resolution), from which parameters such as elevation, slope, aspect, and topographic relief are extracted.

[0016] Soil data includes soil type maps, soil fertility data (organic matter content, nitrogen, phosphorus and potassium content), and soil moisture data (such as SMAP satellite data and field sampling data).

[0017] Meteorological and hydrological data include precipitation, temperature, sunshine duration, wind speed (such as meteorological station observation data and ERA5 reanalysis data), river system distribution, and groundwater depth data.

[0018] Humanities and social data include land use planning data, farmland ownership confirmation data, population density data, agricultural production input data (fertilizer usage, number of agricultural machines), forestry management data, etc.

[0019] Ecological data includes vegetation cover (NDVI), net primary productivity (NPP), biodiversity distribution data, and ecosystem service value assessment data.

[0020] Data quality is improved by standardizing the format of various types of data (e.g., unifying to GeoTIFF format), spatial registration (using WGS84 coordinate system), and resolution (resampling to 10m); outlier detection (based on the 3σ principle), noise removal (median filtering for optical remote sensing and Lee filtering for SAR data), and missing value imputation (using Kriging interpolation).

[0021] The fusion layer employs a two-level fusion strategy to eliminate data heterogeneity in order to generate a comprehensive farmland-forest land dataset.

[0022] In this embodiment, the two-level fusion strategy is feature-level fusion and decision-level fusion, which eliminates the heterogeneity of multi-source data and strengthens the correlation features between cultivated land and forest land.

[0023] Feature-level fusion is based on an improved CNN-attention mechanism model, which extracts and fuses deep features from various data sources. In this embodiment, firstly, features are extracted from remote sensing data, terrain data, and soil data, such as extracting texture features from optical remote sensing, scattering features from SAR data, and terrain features from DEM data. Then, the weights of each feature are dynamically assigned through an attention mechanism, such as assigning high weights to key texture and terrain features for identifying farmland-forest boundaries, generating a fused feature matrix. The weight allocation calculation formula is as follows: (1) In the formula, The weights of the features of the i-th type of data source; The importance score for the i-th type of feature is learned through training samples by an attention mechanism; n is the number of data source types. This formula can adaptively highlight key features for farmland-forest monitoring and suppress noisy features.

[0024] Decision-level fusion, based on DS evidence theory, involves multi-model decision fusion of the feature-level fused data to obtain classification results. In this embodiment, three basic models—Random Forest (RF), Support Vector Machine (SVM), and Deep Learning (CNN-LSTM)—are selected to perform preliminary classification of cultivated land-forest land types, obtaining the decision results (evidence) of each model. Then, the multi-model decision results are fused using DS evidence synthesis rules to improve classification accuracy. In this embodiment, the synthesis rule formula is as follows: when hour, (2) when hour, (3) In the formula, Assign basic probabilities to category A (cultivated land / forest land / transition zone / other) after merging; The conflict coefficient (measures the degree of conflict in multi-model decision-making). , , The three basic models, Random Forest, Support Vector Machine, and CNN-LSTM, are assigned basic probability values ​​for class A, respectively. , , These are the decision results from the three models. This fusion strategy can effectively reduce the misclassification rate of a single model, especially improving the classification accuracy of the farmland-forest transition zone.

[0025] The monitoring layer uses a deep learning model to achieve real-time identification, quantitative calculation, and correlation establishment of dynamic changes between cultivated land and forest land.

[0026] In this embodiment, a dynamic monitoring process of "real-time identification - change quantification - correlation mining" is constructed to achieve full-time, high-precision monitoring of changes in cultivated land and forest land. Among them, the real-time identification of dynamic changes is based on fused multi-source data and adopts an improved CNN-LSTM time-series prediction model to achieve real-time identification of changes in cultivated land and forest land.

[0027] In this embodiment, the LSTM module captures the temporal dependencies of the data (such as seasonal changes in cultivated land and annual growth changes in forest land), while the CNN module captures spatial features (such as changes in boundary morphology). The model introduces a difference image as input, which is the difference between the current time-phase data and the previous time-phase data, enhancing the feature response of the changed areas. The model outputs a change mask that marks the changed areas and the types of changes. That is, the model output is a change mask, which marks the changed areas (e.g., 1 indicates change, 0 indicates no change) and the types of changes, such as cultivated land to forest land, forest land to cultivated land, cultivated land degradation, forest land degradation, and other changes.

[0028] For the identified areas of change, indicators such as the area of ​​change, the rate of change, and the degree of change are quantitatively calculated. In this embodiment, the quantitative calculation of change can be expressed as follows: (4) (5) (6) In the formula, The total area of ​​a certain variation type; k is the number of pixels of that variation type; The area of ​​a single pixel (based on a 10m resolution). =100m 2 ); Rate of change (unit: m) 2 / sky); The interval between the two monitoring phases is the number of days; D represents the degree of change (percentage). The core functional indicators of the changed area (such as the NDVI value of cultivated land and the NPP value of forest land). This represents the core functional indicators before the change. The larger the D value, the more significant the functional change.

[0029] Then, the correlation between cultivated land and forest land is explored. In this embodiment, the correlation is established through a coupling coordination degree model and a competition intensity model. Specifically, the coupling coordination degree model quantifies the ecological coupling relationship between the two, while the competition intensity model quantifies the land resource competition relationship between them.

[0030] Specifically, the coupling coordination degree model is constructed by first calculating the coupling degree C, which takes a value between 0 and 1. A value closer to 1 indicates a stronger coupling relationship, and is represented as follows: (7) Next, calculate the coupling coordination degree D, which takes a value of 0-1. A value of 0-0.3 indicates misalignment, 0.3-0.7 indicates coordination, and 0.7-1 indicates excellent coordination. Therefore, it is represented as... ;in, (8) In the formula, As a comprehensive development index; The Farmland System Development Index is a weighted average of indicators such as farmland yield, soil fertility, and irrigation security rate. The forest ecosystem development index is obtained by weighting indicators such as vegetation cover, NPP, and ecosystem service value. , These are the weighting coefficients, Based on the analytic hierarchy process (AHP), such as ecological priority areas are determined. Major agricultural production areas .

[0031] The competition intensity model measures competition intensity; a higher value indicates more intense competition. In this embodiment, the formula for calculating competition intensity is: (9) In the formula, This refers to the area of ​​farmland converted to forest land; This refers to the area of ​​forest land converted to arable land; It is the sum of the total area of ​​cultivated land and forest land.

[0032] The optimization layer is used to construct a multi-objective quantitative optimization model and establish a dynamic update mechanism to output a layout optimization scheme. By constructing a layout optimization system with multi-objective quantitative optimization and dynamic adaptation adjustment, the system takes the coordinated development of arable land and forest land as its core, while taking into account the three major objectives of ecology, economy, and society.

[0033] In this embodiment, the multi-objective quantitative optimization model is constructed using an improved NSGA-III (non-dominated sorting genetic) algorithm, transforming each objective function and constraint condition into a fitness function of the model. Its optimization objectives include the total value of ecosystem services. Maximize, gross economic output Maximizing and social welfare index The ecological goals include values ​​such as water conservation, soil retention, carbon sequestration, and biodiversity protection. The economic goal is to maximize the sum of agricultural output from arable land and economic output from forest land. The social goals are to ensure food security (with per capita arable land possession not less than 0.1 hectares) and promote employment (with stable employment in forestry and agriculture).

[0034] The weights of each objective are determined using the entropy weight method; therefore, the formula for calculating the optimization objective is: (10) In the formula, F is the value of the comprehensive objective function; Total value of ecosystem services; Gross economic output; The social welfare index (composed of food security rate and employment stability rate); , , The target weights are determined based on the entropy weight method to avoid subjective bias.

[0035] In this embodiment, the constraints of the multi-objective quantitative optimization model include constraints on arable land retention, forest coverage, land use change costs, and per capita arable land. Specifically, the ecological objective constraint is that the forest coverage rate is not lower than the regional planning standard (e.g., 23%), the economic objective constraint is that the arable land retention is not lower than the permanent basic farmland protection standard, and the social objective constraint is that the land use change costs (e.g., the costs of arable land consolidation and afforestation) are minimized.

[0036] Based on dynamic monitoring results, a real-time update mechanism for the optimization scheme is constructed. In this embodiment, the dynamic update mechanism triggers a recalculation of the multi-objective quantitative optimization model and outputs an updated layout optimization scheme when the monitored changes in cultivated land and forest land exceed a threshold, such as when the coupling coordination degree drops to an imbalance level, the competition intensity exceeds a preset threshold (e.g., 5%), or when changes occur in the external environment (e.g., policy adjustments, climate change). In this embodiment, the adjustment logic is as follows: priority is given to ensuring the core constraints (cultivated land area and forest coverage), and then the new objective priorities are adapted through weight adjustments (e.g., increasing the carbon sink weight of forest land under climate change).

[0037] The application layer provides visualization, decision support, and early warning push functions. It transforms the spatial layout results (raster-scale land use planning) output by the optimized model into plot-scale implementation plans. This includes defining the spatial boundaries between arable land protection red lines and forest ecological red lines; developing management plans for arable land-forest transition zones (such as creating farmland shelterbelts and delineating areas for converting sloping arable land back to forest); planning land consolidation projects (such as arable land quality improvement and forest land degradation restoration); and developing supporting policies (such as ecological compensation and agricultural subsidies) to ensure the implementation of the plans.

[0038] In this embodiment, by integrating six categories of multi-source data and employing a two-level fusion algorithm, data heterogeneity is eliminated, the correlation between cultivated land and forest land is strengthened, and data support capabilities are enhanced. Based on a CNN-LSTM time series model and differential image input, the accuracy of transition zone identification is improved, changes are captured quantified, and the timeliness and accuracy of dynamic monitoring are enhanced. A coupling coordination degree and competition intensity model is constructed to quantify the ecological coupling and resource competition relationship between the two, reflecting the integrity of the ecosystem.

[0039] Option 2 A method for dynamic monitoring and layout optimization of cultivated land and forest land by integrating multi-source data is provided, which is applied to the aforementioned system for dynamic monitoring and layout optimization of cultivated land and forest land by integrating multi-source data, as shown in the attached figure. Figure 2 As shown, it includes the following steps: S1: Collect multi-source data, construct a multi-source data system covering relevant dimensions of cultivated land and forest land, and complete standardization processing.

[0040] In this embodiment, remote sensing data is acquired using a "multi-satellite collaborative acquisition" mode. High-resolution optical remote sensing data is obtained through the Gaofen-1 / 2 / 6 satellite image download platform, selecting images with a spatial resolution of 0.8-2m and a temporal resolution of 1-5 days (prioritizing cloudless periods) to extract the morphology of the farmland-forest boundary and details of land cover. Synthetic aperture radar (SAR) data is acquired from the Sentinel-1 data center, using C-band images with a spatial resolution of 5-20m. Interferometry is used to eliminate cloud and rain interference for monitoring areas with cloudy days or high vegetation cover. Hyperspectral remote sensing data is taken from the Gaofen-5 satellite, using 260 spectral bands to capture the spectral characteristics of vegetation and distinguish between crop types in farmland and vegetation types in forestland. Nighttime light remote sensing data uses the NPP-VIIRS product, downloaded from the NOAA website, to reflect the intensity of human activities and forestry management around farmland.

[0041] The topographic data was downloaded from the USGS website as ASTER GDEM digital elevation model data (30m resolution). The topography was analyzed using ArcGIS software to extract core parameters such as elevation, slope (calculated using a 3×3 window), aspect (calculated clockwise with due north as 0°), and topographic relief (statistically calculated using a 1km×1km window) to determine the topographic suitability of cultivated land to forest land.

[0042] Soil data were obtained from the Ministry of Natural Resources Soil Survey Database, using a 1:50,000 scale. Soil fertility data combined field sampling (collecting soil samples from the 0-20cm topsoil layer using a 200m×200m grid) with laboratory testing (measuring organic matter, nitrogen, phosphorus, and potassium content), while also integrating long-term data from county-level soil fertility monitoring stations. Soil moisture data fused SMAP satellite remote sensing inversion data (spatial resolution 9km) with field sensor monitoring data (soil moisture sensors were deployed according to plots, collecting data hourly), improving the spatial and temporal accuracy of the data.

[0043] Meteorological and hydrological data, including precipitation, temperature, sunshine duration, and wind speed, were downloaded from the Meteorological Science Data Center. Daily-scale data from all meteorological stations in the region were selected, and 10m resolution raster data were generated through spatial interpolation. ERA5 reanalysis data (spatial resolution 0.25°) was also added to address the shortage of meteorological stations in remote areas. River system distribution data were obtained from the 1:100,000 river system map in the National Hydrological Database, and groundwater depth data were generated through interpolation of monthly-scale observation data from regional groundwater monitoring wells.

[0044] The land use planning data and cultivated land ownership confirmation data in the humanities and social data are obtained from the natural resources department and are plot-level data in vector format; the population density data are selected from the WorldPop global population dataset (spatial resolution 100m) and calibrated in combination with county statistical yearbook data; the agricultural production input data (fertilizer usage and number of agricultural machines) are obtained from the agricultural statistical database, and are allocated to plots after being statistically analyzed at the township level; the forestry management data are provided by the forestry department, including afforestation area, management personnel, logging records, etc.

[0045] In the ecological data, vegetation cover (NDVI) was calculated based on Landsat 8 satellite imagery using the formula NDVI = (NIR - R) / (NIR + R), where NIR is the near-infrared band and R is the red band. Net primary productivity (NPP) was calculated using the CASA model, with input data including NDVI, meteorological data, and solar radiation data. Biodiversity distribution data were obtained from the Ministry of Ecology and Environment's Biodiversity Survey Database, and ecosystem service value assessment data were calculated using the equivalent factor method proposed by Xie Gaodi et al., with adjustments made to the equivalent factor based on regional realities.

[0046] The collected multi-source data were then preprocessed, including format standardization, converting all raster data to GeoTIFF format and vector data to SHP format. Spatial registration was performed on the standardized data, aligning various data types with high-resolution optical remote sensing imagery, controlling the registration error within one pixel. Resolution was also standardized to ensure consistent spatial scale. Finally, outlier detection and processing were performed, along with noise removal and missing value imputation.

[0047] S2 employs a two-level fusion strategy to eliminate data heterogeneity and generate a comprehensive dataset of cultivated land and forest land.

[0048] In this embodiment, the heterogeneity of multi-source data is eliminated through step-by-step processing, thereby strengthening the correlation between cultivated land and forest land. Feature-level fusion involves extracting core features from each type of preprocessed data. Specifically, texture features (using Gabor filters) and spectral features are extracted from optical remote sensing data through the convolutional layers of a CNN; backscattering coefficient features are extracted from SAR data; topographic factor features are extracted from DEM data; and fertility feature vectors are extracted from soil data.

[0049] The extracted features are standardized, normalizing each feature to the [0,1] interval to avoid the impact of magnitude differences on the fusion effect. The extracted features are then input into the attention mechanism module, which learns feature importance scores through training samples. Then, the weights of each feature are calculated using the weight allocation formula (1). Assign weights to various features. Weighted summation generates a unified fusion feature matrix, highlighting key features for monitoring cultivated land and forest land (such as boundary texture and topographic relief) while suppressing noise features.

[0050] To build the foundational model for training, decision-level fusion was used. Three models were selected: Random Forest (RF), Support Vector Machine (SVM), and Deep Learning (CNN-LSTM). The fusion feature matrix was used as input, and the labels of four land types from the field survey were used as output. The models were trained and the parameters were optimized (RF optimized the number and depth of decision trees, SVM optimized the kernel function parameters, and CNN-LSTM optimized the number of iterations and the learning rate).

[0051] The data to be classified is input into three trained models to obtain the basic probability assignment of each pixel to the class of each model. , , The initial decision result is obtained. Then, the degree of conflict in the multi-model decision is calculated using formulas (2) and (3). ,when When there is little conflict, the multi-model decision results are directly fused using the DS evidence synthesis rule, and the fused basic probability assignment is calculated. .

[0052] Select The largest category is used as the final classification result for that pixel, generating a high-precision integrated farmland-forest dataset, which in particular improves the classification accuracy of the transition zone.

[0053] S3, based on a deep learning model, enables real-time identification, quantitative calculation, and correlation establishment of dynamic changes between cultivated land and forest land.

[0054] In this embodiment, real-time dynamic change recognition is achieved based on an improved CNN-LSTM temporal prediction model. A combined cultivated land and forest land dataset with consecutive time phases is selected, and difference images between adjacent time phases are calculated to enhance the feature differences of changed areas. The CNN part employs 4 convolutional layers and 2 pooling layers to extract spatial change features from the difference images. The LSTM part uses 2 hidden layers to capture temporal dependencies. Using the difference images and previous time-phase data as input, and the change labels and change type labels from field surveys as output, the model is trained using a cross-entropy loss function and iteratively optimized until the loss function converges.

[0055] The newly acquired temporal data is input into the trained model, which outputs a change mask to accurately mark the location of the changed area, the type of change, and the confidence level of the change, enabling real-time monitoring over a period of 1-3 days.

[0056] Based on the change masking results, formulas (4), (5), and (6) are used to quantify the change indicators, construct a dual-model system of coupling coordination degree model and competition intensity model, quantify the ecological and resource relationship between the two, and establish the correlation between cultivated land and forest land.

[0057] S4, construct a multi-objective quantization optimization model and combine it with a dynamic update mechanism to output a layout optimization scheme.

[0058] In this embodiment, the multi-objective optimization model is constructed using an improved non-dominated sorting genetic algorithm (NSGA-III). The three objectives are weighted and summed to obtain a comprehensive objective function, which is then applied based on set constraints such as farmland retention limits. Forest coverage rate constraints ; variable cost constraints per capita arable land constraints .in, To optimize the cultivated land area; The standard for arable land retention; To optimize the forest area; This represents the area corresponding to the forest coverage rate standard. Total cost of land use change; The upper limit of the budget; This refers to the per capita arable land area. This is the minimum guarantee standard.

[0059] The population size was set to 200, the number of iterations to 100, the crossover probability to 0.8, and the mutation probability to 0.05. The Pareto optimal solution was obtained by using the NSGA-Ⅲ algorithm, and the solution with the best overall benefit was selected as the initial layout optimization scheme (grid size, 10m×10m).

[0060] A real-time update mechanism is constructed based on dynamic monitoring results, and a preset threshold for coupling coordination imbalance is established. ), competition intensity warning threshold ( When detected , When the external environment changes (policy adjustments, climate change, major project construction), the model is updated. At the same time, its adjustment strategy prioritizes ensuring core constraints such as arable land area and forest coverage, then adjusts the target weights according to the new scenario, re-runs the NSGA-Ⅲ algorithm, and outputs the updated optimization scheme.

[0061] S5 generates land parcel-level implementation paths and provides decision support through visualization.

[0062] The optimization results at the raster scale are transformed into plot-level implementation plans. Combining farmland title confirmation data and forest protection red line data, the land use attributes and spatial boundaries of each plot are clarified, generating plot-level optimization maps. Simultaneously, effective decision-making suggestions or supporting data are provided, such as developing differentiated management plans for farmland-forest transition zones (50-100m wide), including slope considerations. The 25° transition zone is designated as a land conversion area for returning farmland to forest, and farmland shelterbelts (row spacing 20m, plant spacing 3m) are established in the transition zone of the plain area.

[0063] For degraded arable land, improvement and upgrading projects will be planned; for degraded forest land, restoration projects will be planned. Supporting policies will be formulated, and an ecological compensation mechanism will be established to ensure the smooth implementation of the plans.

[0064] In this embodiment, the deep fusion and comprehensive coverage of multi-source data solves the problems of single data, low fusion degree, and insufficient accuracy in identifying the transition zone between cultivated and forested areas in traditional technologies. Real-time dynamic monitoring and correlation quantification models compensate for the shortcomings of existing technologies, such as poor monitoring timeliness and insufficient mining of the correlation between cultivated and forested areas. Through multi-objective quantification optimization and dynamic update mechanisms, the challenges of single-objective, experience-driven, and statically fixed traditional layout optimization are overcome. Furthermore, the technical system, verified through pilot projects in different terrain regions, effectively solves the problems of limited adaptability of existing algorithms to various scenarios and poor connection with actual application needs. Ultimately, it achieves a synergistic balance between cultivated and forested protection and utilization, providing reliable technical support for precise management and control of national land space.

[0065] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A dynamic monitoring and layout optimization system for cultivated land and forest land integrating multi-source data, characterized by: The system comprises a data layer, a fusion layer, a monitoring layer, an optimization layer, and an application layer, which work in sequence to form a closed-loop management system for the entire process. The data layer is used to collect multi-source data and perform standardized preprocessing. The fusion layer uses a two-level fusion strategy to eliminate data heterogeneity and generate a comprehensive dataset of cultivated land and forest land. The monitoring layer uses a deep learning model to achieve real-time identification, quantitative calculation, and correlation establishment of dynamic changes in cultivated land and forest land. The optimization layer is used to construct a multi-objective quantitative optimization model and establish a dynamic update mechanism to output layout optimization schemes. The application layer provides visualization, decision support, and early warning push functions.

2. The system for dynamic monitoring and layout optimization of cultivated land and forest land integrating multi-source data as described in claim 1, characterized in that: The two-level fusion strategy consists of feature-level fusion and decision-level fusion. The feature-level fusion involves extracting deep features from each data source and dynamically assigning weights based on an improved CNN-attention mechanism model. The weight assignment formula is as follows: ； In the formula, The weights of the features of the i-th type of data source; Score the importance of the i-th type of feature; n represents the number of data source types; The decision-level fusion is based on the DS evidence theory and integrates the classification results of multiple models.

3. The system for dynamic monitoring and layout optimization of cultivated land and forest land integrating multi-source data according to claim 1, characterized in that: The real-time dynamic change recognition adopts an improved CNN-LSTM time series prediction model. The model introduces a difference image as input, which is the difference between the current time phase data and the previous time phase data. The model outputs a change mask that marks the change area and the change type.

4. The system for dynamic monitoring and layout optimization of cultivated land and forest land integrating multi-source data according to claim 1, characterized in that: The correlation is established by coupling a coordination degree model and a competition intensity model; wherein, the competition intensity is calculated using the following formula: ； In the formula, This refers to the area of ​​farmland converted to forest land; This refers to the area of ​​forest land converted to arable land; It is the sum of the total area of ​​cultivated land and forest land.

5. The system for dynamic monitoring and layout optimization of cultivated land and forest land integrating multi-source data according to claim 4, characterized in that: The coupling coordination degree model is constructed by first calculating the coupling degree C, which is expressed as follows: ； Next, calculate the coupling coordination degree D, which is expressed as: ； ； In the formula, As a comprehensive development index; The farmland system development index; Forest system development index; , These are the weighting coefficients.

6. The system for dynamic monitoring and layout optimization of cultivated land and forest land integrating multi-source data according to claim 1, characterized in that: The multi-objective quantitative optimization model is constructed using an improved NSGA-III algorithm, and the optimization objectives include the total value of ecosystem services. Maximize, gross economic output Maximizing and social welfare index To maximize the objective, the weights of each objective are determined using the entropy weighting method; therefore, the formula for calculating the optimization objective is: ； In the formula, , , These are the target weights.

7. The system for dynamic monitoring and layout optimization of cultivated land and forest land integrating multi-source data according to claim 6, characterized in that: The constraints of the multi-objective quantitative optimization model include constraints on arable land retention, forest coverage, land use change costs, and per capita arable land.

8. A method for dynamic monitoring and layout optimization of cultivated land and forest land by integrating multi-source data, characterized in that, The system for dynamic monitoring and layout optimization of cultivated land and forest land, which integrates multi-source data as described in any one of claims 1-7, includes the following steps: S1. Collect multi-source data, construct a multi-source data system covering relevant dimensions of cultivated land and forest land, and complete the standardization process; S2 employs a two-level fusion strategy to eliminate data heterogeneity and generate a comprehensive farmland-forest land dataset. S3, based on a deep learning model, enables real-time identification, quantitative calculation, and correlation establishment of dynamic changes between cultivated land and forest land; S4, construct a multi-objective quantitative optimization model and combine it with a dynamic update mechanism to output a layout optimization scheme; S5 generates land parcel-level implementation paths and provides decision support through visualization.

9. The method for dynamic monitoring and layout optimization of cultivated land and forest land by integrating multi-source data according to claim 8, characterized in that: The multi-source data includes remote sensing data, topographic data, soil data, meteorological and hydrological data, human and social data, and ecological data. Among them, the remote sensing data includes at least high-resolution optical remote sensing data, synthetic aperture radar data, and hyperspectral remote sensing data.

10. The method for dynamic monitoring and layout optimization of cultivated land and forest land by integrating multi-source data according to claim 8, characterized in that, The dynamic update mechanism triggers the multi-objective quantitative optimization model to recalculate and output an updated layout optimization scheme when the coordination degree of cultivated land-forest land coupling drops to an imbalance level, the competition intensity exceeds a preset threshold, or the external environment changes.

Images