Dynamic monitoring method and system of land cover remote sensing based on adaptive irregular unit grid

Abstract

The present invention discloses a method and system for dynamic monitoring of land cover using remote sensing based on an adaptive irregular unit grid. The method comprises the following steps: S1, collecting and preprocessing remote sensing image data to construct a standardized data set; S2, extracting feature boundaries based on the standardized data set to construct an initial irregular grid; S3, calculating a change sensitivity index to refine the adaptive irregular unit grid; S4, extracting sample feature vectors within the adaptive irregular unit grid; S5, executing a hierarchical and swarming locust optimization algorithm to obtain an optimal spatial weight distribution of sample points; S6, performing weighted fitting on remote sensing time series data to model the evolution trend of land cover attributes; S7, extracting and outputting land cover change trends to generate monitoring and prediction results; and S8, collecting the latest annotated data and adaptively updating the model and weights. The present invention can achieve high-precision, intelligent, and dynamic remote sensing monitoring of land cover changes and trend prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method and system for dynamic monitoring of land cover by remote sensing based on an adaptive irregular unit grid. Background Art

[0002] In the field of dynamic remote sensing monitoring of land cover, with the continuous improvement of remote sensing image resolution and acquisition frequency, traditional regular gridding and static sample collection methods face increasing limitations in capturing surface change processes and conducting detailed feature analysis. Existing technologies often use regular gridding to spatially partition remote sensing images, then extract features and perform change detection based on grid cells. While this regular grid approach is simple in structure and easy to implement, in practice, it often fails to fully conform to complex surface boundaries and struggles to effectively address surface spatial heterogeneity and multi-scale variation. Particularly in areas with complex feature distribution and variable spatial structure, regular grids often result in a large amount of invalid data processing and blurred boundaries, compromising monitoring accuracy and model generalization. Furthermore, conventional remote sensing change detection and trend analysis are often based on traditional statistical methods or single machine learning models, lacking the capabilities for dynamic weight optimization, spatial feature adaptation, and in-depth mining of temporal information, making it difficult to accurately model the evolutionary trends of land cover attributes within a region.

[0003] On the other hand, existing land cover monitoring methods typically use uniform or empirical weights to process sample points during feature extraction and modeling. These methods lack intelligent mechanisms for dynamically adjusting weight distribution based on spatial neighborhoods, attribute correlations, and temporal statistical characteristics. This results in insufficient response to change-sensitive areas and boundary transition regions, impacting the model's sensitivity and robustness. Furthermore, current optimization algorithms for spatial weighting or parameter optimization are mostly general-purpose methods like particle swarm optimization and genetic algorithms. These methods suffer from slow convergence and a tendency to fall into local optimality. Furthermore, they lack mechanisms such as hierarchical population collaboration, local memory, and diversified search, making them difficult to adapt to the global optimization requirements of complex, high-dimensional, multimodal remote sensing data.

[0004] In summary, the existing dynamic monitoring methods of surface cover using remote sensing have prominent problems such as insufficient spatial resolution, weak boundary fitting ability, unreasonable sample weight distribution, and insufficient intelligence of the optimization algorithm, making it difficult to achieve refined, intelligent, and dynamic monitoring of complex surface changes.

[0005] Therefore, how to provide a method and system for dynamic monitoring of land cover by remote sensing based on an adaptive irregular unit grid is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] One purpose of the present invention is to propose a method and system for dynamic monitoring of surface cover by remote sensing based on an adaptive irregular unit grid. The present invention makes full use of the idea of ​​combining swarm intelligence optimization algorithm with adaptive grid modeling, and describes in detail the entire process of adaptive spatial division of remote sensing images, dynamic weight optimization of sample features, and modeling of the evolution trend of surface cover attributes over time. It has the advantages of high spatial resolution, strong fit with the boundaries of objects, high accuracy in dynamic change detection, and strong model intelligent optimization capability.

[0007] The method for dynamic monitoring of land cover by remote sensing based on an adaptive irregular unit grid according to an embodiment of the present invention includes the following steps:

[0008] S1. Collect remote sensing image data of the target monitoring area for multiple time periods, preprocess the remote sensing image data, and construct a standardized remote sensing image dataset;

[0009] S2. Based on the standardized remote sensing image dataset, the image segmentation method is used to extract the boundary information of the ground objects, and the initial grid structure is constructed according to the spatial structure characteristics and boundary complexity of the ground objects to generate irregular unit grids;

[0010] S3. Based on the irregular unit grid, calculate the change sensitivity index of each grid unit. When the sensitivity is higher than the set threshold, perform the local grid refinement step to form an adaptive irregular unit grid;

[0011] S4, based on the adaptive irregular unit grid, the remote sensing image pixels in each unit grid are used as candidate sample points, and the spatial position and ground feature attributes are extracted to construct the sample feature vector;

[0012] S5. Using the sample feature vector as input, perform position update, population behavior simulation, and solution space search in the locust optimization algorithm to obtain the optimal spatial weight distribution of sample points within each irregular unit grid;

[0013] S6. Based on the optimal spatial weight distribution of sample points, perform weighted fitting on the remote sensing time series data in each adaptive irregular unit grid, model the evolution trend of surface cover attributes over time, and obtain weighted regression results that reflect the dynamic changes of the surface;

[0014] S7. Based on the weighted regression results, extract the change direction, change amplitude and period characteristics of each irregular unit grid, and generate the surface cover prediction image of the target area in the current phase and the future phase;

[0015] S8. Collect the latest phase of remote sensing imagery with real annotation data, compare and analyze it with the surface cover prediction image, and update the population parameters of the locust optimization algorithm based on the error feedback results.

[0016] Optionally, the remote sensing image data specifically includes multi-temporal multispectral images, high-resolution visible light images and vegetation index products.

[0017] Optionally, the preprocessing of the remote sensing image data specifically includes radiation correction, geometric correction, band fusion and unified coordinate projection processing.

[0018] Optionally, the S2 specifically includes:

[0019] S21. Based on the constructed standardized remote sensing image dataset, a multi-scale object-oriented image segmentation method is used to segment the remote sensing image, extract the boundary vector information of the object, and project the boundary vector information of the object into a unified coordinate system;

[0020] S22. Constructing a spatial adjacency relationship graph G=(V, E) based on the feature boundary vector information, where V represents the set of feature objects after segmentation and E represents the adjacency relationship between the objects;

[0021] S23. Based on the spatial adjacency graph, the Voronoi diagram construction method is used to preliminarily divide the monitoring area, and the seed point set is set as , where each seed point The geometric center coordinates corresponding to a feature object, n is the total number of feature objects;

[0022] S24, perform Delaunay triangulation operation, generate a triangular mesh structure on the seed point set, and obtain an initial mesh structure set , where each triangle unit Corresponding to an irregular cell grid candidate area, m is the total number of triangular cells;

[0023] S25. The texture complexity of the ground object is obtained by calculating the texture contrast of the grayscale co-occurrence matrix of the remote sensing image in the triangular unit, analyzing the tangent change rate of the triangular unit boundary point to obtain the boundary curvature, and statistically analyzing the difference between the properties of the triangular unit and its adjacent grid units to obtain the spatial heterogeneity index. Based on the texture complexity of the ground object in each triangular unit, the texture of the ground object is obtained by calculating the texture contrast of the grayscale co-occurrence matrix of the remote sensing image in the triangular unit. , boundary curvature and spatial heterogeneity index , construct the boundary complexity function:

[0024] ;

[0025] in, , , is the weighting coefficient;

[0026] S26, based on the boundary complexity function The values ​​of are used to sort the initial grid cells and set the complexity screening threshold ,when When the corresponding unit is included in the refinement and optimization process, the construction of the initial irregular unit grid is completed.

[0027] Optionally, the S3 specifically includes:

[0028] S31, for the generated irregular unit grid set , in each unit grid Based on the standardized remote sensing image dataset, the corresponding multi-temporal spectral data, texture data and boundary morphological parameters are extracted to construct a set of change sensitivity indicators. ;

[0029] S32, the change sensitivity index set It includes three sensitivity quantitative indicators:

[0030] Regional spectral variability :Indicates unit The degree of time variation dispersion of the mean value of the inner main band;

[0031] Image texture variation index : Indicates the changing trend of texture contrast in multi-temporal gray-level co-occurrence matrix;

[0032] Boundary complexity function value : generated by the boundary complexity function;

[0033] ;

[0034] in, 、 、 are the normalized spectral variability, texture change index and boundary complexity value, 、 and is the weight coefficient;

[0035] S34. Setting sensitivity score threshold set , when the unit grid Rating value When , it is determined that the k-th layer local grid refinement processing needs to be performed;

[0036] S35, perform layered refinement processing, in the original unit grid Generate k-layer nested triangulation internally, insert additional seed points using Delaunay triangulation method, and construct refined grid subsets , where q is the number of subgrid units, ;

[0037] S36. Summarize all refined grid cells to generate an updated irregular cell grid set , as the adaptive irregular unit grid.

[0038] Optionally, the S4 specifically includes:

[0039] S41, Generative Adaptive Irregular Cell Grid Set , in each irregular unit grid Extract the current grid and spatial neighborhood All temporal remote sensing image pixel data within form a set of candidate sample points , k represents the total number of adaptive irregular unit grids, represents the number of sample points in the unit grid and spatial neighborhood;

[0040] S42, sample point set Each sample point , get the spatial position coordinates , extract multi-dimensional feature attribute vector , and historical multi-temporal statistical characteristics ,in is the mean, is the standard deviation, is the maximum difference, d is the feature dimension;

[0041] S43. Constructing the space-attribute-time series joint feature vector for each sample point :

[0042] ;

[0043] in, Represents the spatial ordinate of the jth sample point in the i-th unit grid, , ,..., Represents the multidimensional feature attribute characteristics of the jth sample point in the i-th unit grid, and d is the feature dimension;

[0044] S44, the joint feature vector of all sample points Composition of sample feature vector set ;

[0045] S45. Output the sample feature vector set of all irregular unit grids , used for weight optimization processing of locust optimization algorithm.

[0046] Optionally, the S5 specifically includes:

[0047] S51, based on the output sample feature vector set , the sample feature matrix of each irregular unit grid is used as the input of the optimization algorithm;

[0048] S52, initialize the hierarchical population structure, divide the entire population into M sub-populations, each sub-population is independently distributed in the global search space, and initializes the individual position, speed and search parameters respectively;

[0049] S53. Within each subpopulation, the positions of the individuals in the population are updated according to the social interaction force, gravity force and wind force, and the current optimal fitness position of each individual is recorded and set as the local optimal position. ;

[0050] S54, when each individual updates its position, it adds a local learning and memory strategy to the local optimal position With weight coefficient Introduced into the update formula, the new individual position is :

[0051] ;

[0052] in is the individual's current location, is the memory impact factor, is the social interaction force vector of the jth individual in the i-th subpopulation, is the gravitational force vector acting on the j-th individual in the i-th subpopulation, is the wind force vector acting on the jth individual in the i-th subpopulation;

[0053] S55. For each individual in each subpopulation, regularly check the historical optimal fitness, continuously If the fitness of the generation does not improve, the local memory backtracking strategy is triggered, and the individual jumps back to the vicinity of the local historical optimal position and conducts a neighborhood re-search;

[0054] S56, introduce individual behavior diversification mechanism, and classify some individuals according to the mutation probability Perform Gaussian perturbation or Levy jump, the new position of the individual is :

[0055] ;

[0056] in is the disturbance parameter, is a Gaussian distribution;

[0057] S57. Set up an information exchange mechanism between subpopulations. At a specified iteration cycle or when the global optimum has not improved for a long time, exchange the elite individuals of each subpopulation, introduce the optimal solution of the higher-level subpopulation into the lower-level subpopulation, or perform random exchange of individual positions.

[0058] S58, all individuals after each round of iteration according to the target fitness function of the sample feature vector Perform fitness evaluation with the goal of minimizing weighted regression error and controlling the smoothness of spatial weight distribution;

[0059] S59, loop through steps S53 to S58 until the maximum number of iterations or the fitness convergence condition is reached, and output the optimal spatial weight distribution of sample points for each irregular unit grid. ;

[0060] S510, the optimal sample point spatial weight distribution set of all irregular unit grids , for use in weighted regression model building and dynamic monitoring and analysis of surface cover.

[0061] Optionally, the S6 specifically includes:

[0062] S61. Based on the obtained optimal sample point spatial weight distribution, the sample feature vectors in each adaptive irregular unit grid are matched with the corresponding weights to form a weighted sample feature set;

[0063] S62. In each adaptive irregular unit grid, select corresponding remote sensing time series data as input data for regression modeling, covering land cover attribute observations of all historical phases;

[0064] S63, matching the weighted sample feature set with the remote sensing time series data, assigning corresponding weights to the land cover attribute values ​​of each historical phase, and constructing a weighted data sequence;

[0065] S64, for each adaptive irregular cell grid, performing a regression modeling step based on the weighted data sequence to fit the evolution trend of the surface cover attributes over time;

[0066] S65. Extract the changing trend of the land cover attributes of each adaptive irregular unit grid according to the results of the weighted regression modeling, including characteristics such as the changing direction, changing amplitude, and changing period;

[0067] S66. Output the weighted regression results of each adaptive irregular unit grid as a basis for trend analysis reflecting dynamic changes in the surface.

[0068] The surface cover remote sensing dynamic monitoring system based on an adaptive irregular unit grid according to an embodiment of the present invention includes the following modules:

[0069] Data acquisition module, used to collect remote sensing image data of the target monitoring area for multiple time periods;

[0070] The data preprocessing module is used to perform radiation correction, geometric correction, band fusion and unified coordinate projection processing on remote sensing image data to generate a standardized remote sensing image dataset;

[0071] The grid construction module is used to extract the boundary information of the ground objects based on the standardized remote sensing image dataset, construct the initial irregular unit grid and perform adaptive refinement to generate an adaptive irregular unit grid set;

[0072] The feature extraction module is used to extract the spatial location, multi-dimensional feature attributes and time series statistical features of sample points in the adaptive irregular unit grid and construct a set of sample feature vectors;

[0073] The weight optimization module is used to execute the hierarchical population locust optimization algorithm based on the sample feature vector set to obtain the optimal spatial weight distribution of sample points in each irregular unit grid;

[0074] The weighted regression modeling module is used to perform weighted fitting based on the spatial weight distribution of the optimal sample points and remote sensing time series data, model the evolution trend of surface cover attributes over time, and generate weighted regression results;

[0075] The change analysis and prediction module is used to extract the characteristics of land cover change trends based on weighted regression results and output land cover prediction and dynamic monitoring results;

[0076] The result feedback and update module is used to collect the real annotation data of remote sensing images in the latest phase, and continuously update and adaptively optimize the weight optimization algorithm based on the error feedback mechanism.

[0077] The beneficial effects of the present invention are:

[0078] The present invention deeply integrates the locust optimization algorithm with the local weighted regression method, and combines it with the spatial modeling strategy of adaptive irregular unit grids to achieve the refinement and intelligence of the dynamic monitoring process of remote sensing image surface cover. Compared with the regular grids and traditional weight distribution methods adopted in the existing technology, the present invention can automatically generate adaptive irregular unit grids that are more in line with the actual boundary of the ground objects according to the spatial heterogeneity and change sensitivity of the ground objects, thereby improving the accuracy of boundary detection and change analysis. By introducing the locust optimization algorithm with innovative mechanisms such as hierarchical populations, local learning and behavioral diversity, the global optimization and dynamic adjustment of the spatial weights of sample points are achieved, effectively enhancing the model's adaptability to complex change areas, spatial transition zones and multi-scale structures. Based on the optimal weight distribution, the weighted regression method is used to perform trend modeling on the time series remote sensing data of each grid unit, so that the evolution law of surface cover attributes can be accurately reflected and predicted.

[0079] This invention significantly improves the sensitivity and accuracy of detecting dynamic changes in surface cover, reduces invalid data processing, and improves the efficiency of spatial information utilization, adapting to the actual monitoring needs of different types of land features and regional diversity. Furthermore, its innovative optimization and modeling framework ensures the efficiency and robustness of the monitoring model, promoting the intelligent and automated development of remote sensing surface monitoring technology, and possesses promising application prospects and promotional value. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0081] Figure 1 This is a flow chart of the method for dynamic monitoring of land cover by remote sensing based on an adaptive irregular unit grid proposed by the present invention;

[0082] Figure 2 This is a structural diagram of the surface cover remote sensing dynamic monitoring system based on the adaptive irregular unit grid proposed by the present invention. DETAILED DESCRIPTION

[0083] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0084] refer to Figure 1 , a method for dynamic monitoring of land cover by remote sensing based on adaptive irregular cell grids, comprising the following steps:

[0085] S1. Collect remote sensing image data of the target monitoring area for multiple time periods, preprocess the remote sensing image data, and construct a standardized remote sensing image dataset;

[0086] S2. Based on the standardized remote sensing image dataset, the image segmentation method is used to extract the boundary information of the ground objects, and the initial grid structure is constructed according to the spatial structure characteristics and boundary complexity of the ground objects to generate irregular unit grids;

[0087] S3. Based on the irregular unit grid, calculate the change sensitivity index of each grid unit. When the sensitivity is higher than the set threshold, perform the local grid refinement step to form an adaptive irregular unit grid;

[0088] S4, based on the adaptive irregular unit grid, the remote sensing image pixels in each unit grid are used as candidate sample points, and the spatial position and ground feature attributes are extracted to construct the sample feature vector;

[0089] S5. Using the sample feature vector as input, perform position update, population behavior simulation, and solution space search in the locust optimization algorithm to obtain the optimal spatial weight distribution of sample points within each irregular unit grid;

[0090] S6. Based on the optimal spatial weight distribution of sample points, perform weighted fitting on the remote sensing time series data in each adaptive irregular unit grid, model the evolution trend of surface cover attributes over time, and obtain weighted regression results that reflect the dynamic changes of the surface;

[0091] S7. Based on the weighted regression results, extract the change direction, change amplitude and period characteristics of each irregular unit grid, and generate the surface cover prediction image of the target area in the current phase and the future phase;

[0092] S8. Collect the latest phase of remote sensing imagery with real annotation data, compare and analyze it with the surface cover prediction image, and update the population parameters of the locust optimization algorithm based on the error feedback results.

[0093] In this embodiment, the remote sensing image data specifically includes multi-temporal multispectral images, high-resolution visible light images and vegetation index products.

[0094] In this embodiment, the pre-processing of the remote sensing image data specifically includes radiation correction, geometric correction, band fusion and unified coordinate projection processing.

[0095] In this embodiment, S2 specifically includes:

[0096] S21. Based on the constructed standardized remote sensing image dataset, a multi-scale object-oriented image segmentation method is used to segment the remote sensing image, extract the boundary vector information of the object, and project the boundary vector information of the object into a unified coordinate system;

[0097] S22. Constructing a spatial adjacency relationship graph G=(V, E) based on the feature boundary vector information, where V represents the set of feature objects after segmentation and E represents the adjacency relationship between the objects;

[0098] S23. Based on the spatial adjacency graph, the Voronoi diagram construction method is used to preliminarily divide the monitoring area, and the seed point set is set as , where each seed point The geometric center coordinates corresponding to a feature object, n is the total number of feature objects;

[0099] S24, perform Delaunay triangulation operation, generate a triangular mesh structure on the seed point set, and obtain an initial mesh structure set , where each triangle unit Corresponding to an irregular cell grid candidate area, m is the total number of triangular cells;

[0100] S25. The texture complexity of the ground object is obtained by calculating the texture contrast of the grayscale co-occurrence matrix of the remote sensing image in the triangular unit, analyzing the tangent change rate of the triangular unit boundary point to obtain the boundary curvature, and statistically analyzing the difference between the properties of the triangular unit and its adjacent grid units to obtain the spatial heterogeneity index. Based on the texture complexity of the ground object in each triangular unit, the texture of the ground object is obtained by calculating the texture contrast of the grayscale co-occurrence matrix of the remote sensing image in the triangular unit. , boundary curvature and spatial heterogeneity index , construct the boundary complexity function:

[0101] ;

[0102] in, , , is the weighting coefficient;

[0103] S26, based on the boundary complexity function The values ​​of are used to sort the initial grid cells and set the complexity screening threshold ,when When the corresponding unit is included in the refinement and optimization process, the construction of the initial irregular unit grid is completed.

[0104] In this embodiment, S3 specifically includes:

[0105] S31, for the generated irregular unit grid set , in each unit grid Based on the standardized remote sensing image dataset, the corresponding multi-temporal spectral data, texture data and boundary morphological parameters are extracted to construct a set of change sensitivity indicators. ;

[0106] S32, the change sensitivity index set It includes three sensitivity quantitative indicators:

[0107] Regional spectral variability :Indicates unit The degree of time variation dispersion of the mean value of the inner main band;

[0108] Image texture variation index : Indicates the changing trend of texture contrast in multi-temporal gray-level co-occurrence matrix;

[0109] Boundary complexity function value : generated by the boundary complexity function;

[0110] ;

[0111] in, 、 、 are the normalized spectral variability, texture change index and boundary complexity value, 、 and is the weight coefficient;

[0112] S34. Setting sensitivity score threshold set , when the unit grid Rating value When , it is determined that the k-th layer local grid refinement processing needs to be performed;

[0113] S35, perform layered refinement processing, in the original unit grid Generate k-layer nested triangulation internally, insert additional seed points using Delaunay triangulation method, and construct refined grid subsets , where q is the number of subgrid units, ;

[0114] S36. Summarize all refined grid cells to generate an updated irregular cell grid set , as the adaptive irregular unit grid.

[0115] In this embodiment, the S4 specifically includes:

[0116] S41, Generative Adaptive Irregular Cell Grid Set , in each irregular unit grid Extract the current grid and spatial neighborhood All temporal remote sensing image pixel data within form a set of candidate sample points , k represents the total number of adaptive irregular unit grids, represents the number of sample points in the unit grid and spatial neighborhood;

[0117] S42, sample point set Each sample point , get the spatial position coordinates , extract multi-dimensional feature attribute vector , and historical multi-temporal statistical characteristics ,in is the mean, is the standard deviation, is the maximum difference, d is the feature dimension;

[0118] S43. Constructing the space-attribute-time series joint feature vector for each sample point :

[0119] ;

[0120] in, Represents the spatial ordinate of the jth sample point in the i-th unit grid, , ,..., Represents the multidimensional feature attribute characteristics of the jth sample point in the i-th unit grid, and d is the feature dimension;

[0121] S44, the joint feature vector of all sample points Composition of sample feature vector set ;

[0122] S45. Output the sample feature vector set of all irregular unit grids , used for weight optimization processing of locust optimization algorithm.

[0123] In this embodiment, the S5 specifically includes:

[0124] S51, based on the output sample feature vector set , the sample feature matrix of each irregular unit grid is used as the input of the optimization algorithm;

[0125] S52, initialize the hierarchical population structure, divide the entire population into M sub-populations, each sub-population is independently distributed in the global search space, and initializes the individual position, speed and search parameters respectively;

[0126] S53. Within each subpopulation, the positions of the individuals in the population are updated according to the social interaction force, gravity force and wind force, and the current optimal fitness position of each individual is recorded and set as the local optimal position. ;

[0127] S54, when each individual updates its position, it adds a local learning and memory strategy to the local optimal position With weight coefficient Introduced into the update formula, the new individual position is :

[0128] ;

[0129] in is the individual's current location, is the memory impact factor, is the social interaction force vector of the jth individual in the i-th subpopulation, is the gravitational force vector acting on the j-th individual in the i-th subpopulation, is the wind force vector acting on the jth individual in the i-th subpopulation;

[0130] S55. For each individual in each subpopulation, regularly check the historical optimal fitness, continuously If the fitness of the generation does not improve, the local memory backtracking strategy is triggered, and the individual jumps back to the vicinity of the local historical optimal position and conducts a neighborhood re-search;

[0131] S56, introduce individual behavior diversification mechanism, and classify some individuals according to the mutation probability Perform Gaussian perturbation or Levy jump, the new position of the individual is :

[0132] ;

[0133] in is the disturbance parameter, is a Gaussian distribution;

[0134] S57. Set up an information exchange mechanism between subpopulations. At a specified iteration cycle or when the global optimum has not improved for a long time, exchange the elite individuals of each subpopulation, introduce the optimal solution of the higher-level subpopulation into the lower-level subpopulation, or perform random exchange of individual positions.

[0135] S58, all individuals after each round of iteration according to the target fitness function of the sample feature vector Perform fitness evaluation with the goal of minimizing weighted regression error and controlling the smoothness of spatial weight distribution;

[0136] S59, loop through steps S53 to S58 until the maximum number of iterations or the fitness convergence condition is reached, and output the optimal spatial weight distribution of sample points for each irregular unit grid. ;

[0137] S510, the optimal sample point spatial weight distribution set of all irregular unit grids , for use in weighted regression model building and dynamic monitoring and analysis of surface cover.

[0138] In this embodiment, S6 specifically includes:

[0139] S61. Based on the obtained optimal sample point spatial weight distribution, the sample feature vectors in each adaptive irregular unit grid are matched with the corresponding weights to form a weighted sample feature set;

[0140] S62. In each adaptive irregular unit grid, select corresponding remote sensing time series data as input data for regression modeling, covering land cover attribute observations of all historical phases;

[0141] S63, matching the weighted sample feature set with the remote sensing time series data, assigning corresponding weights to the land cover attribute values ​​of each historical phase, and constructing a weighted data sequence;

[0142] S64, for each adaptive irregular cell grid, performing a regression modeling step based on the weighted data sequence to fit the evolution trend of the surface cover attributes over time;

[0143] S65. Extract the changing trend of the land cover attributes of each adaptive irregular unit grid according to the results of the weighted regression modeling, including characteristics such as the changing direction, changing amplitude, and changing period;

[0144] S66. Output the weighted regression results of each adaptive irregular unit grid as a basis for trend analysis reflecting dynamic changes in the surface.

[0145] In this embodiment, the S7 specifically includes:

[0146] S71. Based on the weighted regression results, analyze the time series data of land cover attributes of each irregular unit grid;

[0147] S72. Within each irregular unit grid, determine the direction of change of the land cover attribute based on the regression trend curve, including increase, decrease, or stability;

[0148] S73, extracting the change amplitude of the surface cover attribute of each irregular unit grid, and quantifying the specific change amount of the attribute during the analysis period;

[0149] S74. Identify the periodic characteristics of the changes in the surface cover attributes of each irregular unit grid and determine the periodic or stage-specific regularity of the attribute fluctuations;

[0150] S75. Generate surface cover prediction images of the target area in the current and future phases based on the change direction, change amplitude, and periodic characteristics.

[0151] In this embodiment, S8 specifically includes:

[0152] S81. Collect the latest remote sensing imagery of the target area, with real annotation data, as the observation benchmark for land cover changes;

[0153] S82. Match the latest phase’s real annotated data with the surface cover prediction image and conduct comparative analysis;

[0154] S83. Identify the prediction error and spatial distribution characteristics within each irregular unit grid based on the comparative analysis results;

[0155] S84. Adjusting and optimizing the population parameters in the locust optimization algorithm based on the prediction error feedback information;

[0156] S85. Use the optimized parameters in the iterative process to achieve adaptive updating and accuracy improvement of the remote sensing dynamic monitoring model.

[0157] refer to Figure 2 , a dynamic monitoring system for land cover remote sensing based on adaptive irregular cell grids, including the following modules:

[0158] Data acquisition module, used to collect remote sensing image data of the target monitoring area for multiple time periods;

[0159] The data preprocessing module is used to perform radiation correction, geometric correction, band fusion and unified coordinate projection processing on remote sensing image data to generate a standardized remote sensing image dataset;

[0160] The grid construction module is used to extract the boundary information of the ground objects based on the standardized remote sensing image dataset, construct the initial irregular unit grid and perform adaptive refinement to generate an adaptive irregular unit grid set;

[0161] The feature extraction module is used to extract the spatial location, multi-dimensional feature attributes and time series statistical features of sample points in the adaptive irregular unit grid and construct a set of sample feature vectors;

[0162] The weight optimization module is used to execute the hierarchical population locust optimization algorithm based on the sample feature vector set to obtain the optimal spatial weight distribution of sample points in each irregular unit grid;

[0163] The weighted regression modeling module is used to perform weighted fitting based on the spatial weight distribution of the optimal sample points and remote sensing time series data, model the evolution trend of surface cover attributes over time, and generate weighted regression results;

[0164] The change analysis and prediction module is used to extract the characteristics of land cover change trends based on weighted regression results and output land cover prediction and dynamic monitoring results;

[0165] The result feedback and update module is used to collect the real annotation data of remote sensing images in the latest phase, and continuously update and adaptively optimize the weight optimization algorithm based on the error feedback mechanism.

[0166] Example 1:

[0167] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a coastal city. Between 2018 and 2022, urbanization in the region has been rapidly advancing, the area of ​​wetlands has been continuously decreasing, and the pressure on ecological protection has increased. Traditional remote sensing dynamic monitoring methods generally use regular grids and uniform weights, but in the wetland-city interface area, problems such as blurred boundaries, untimely response to changes, and large detection errors often occur. In order to improve monitoring accuracy and dynamic analysis capabilities, this embodiment fully adopts the adaptive irregular unit grid surface cover remote sensing dynamic monitoring method and system based on the fusion of locust optimization algorithm and local weighted regression proposed by the present invention.

[0168] In practice, the system first acquired 20 quarterly Sentinel-2 multispectral remote sensing imagery data from 2018 to 2022, with a spatial resolution of 10 meters. All images were uniformly processed through radiometric correction, geometric correction, band fusion, and projection processing, and standardized as the basis for analysis. An intelligent segmentation algorithm, combined with the complexity of the terrain, was used to establish an adaptive irregular cell grid, generating over 2,200 spatial cells. Automatic refinement was particularly effective in urban fringes and wetland areas, ensuring that boundaries more closely aligned with the actual terrain contours.

[0169] During the sample feature extraction and weight optimization phase, the system utilizes a stratified and clustered locust optimization algorithm to dynamically adjust the weight distribution of each sample point, combining spatial location, NDVI, building index, and temporal characteristics. This enhances response to areas of dramatic landform change and heterogeneity. A weighted regression model then fits the changing trends of surface attributes within each cell, enabling continuous, high-precision monitoring of urban expansion and wetland loss.

[0170] When evaluating the effects, measured data were used to compare the performance of the method of the present invention and the traditional regular grid method. The results showed that over a five-year period, the average annual error in wetland area detection using the method of the present invention was only 1.12%, and the average annual error in urban land detection was 1.19%. The accuracy of wetland boundary detection was increased to 93.7%, and the accuracy of urban land boundary detection was increased to 92.0%. The errors of traditional methods were 2.23% and 2.78%, respectively, and the detection accuracy was lower than 90%. In addition, the average image processing time for each period of the present invention is 4.1 minutes, which is about 30% longer than that of traditional methods. These real data show that the present invention not only greatly improves the monitoring accuracy of complex dynamic areas, but also effectively shortens the analysis time, providing more reliable technical support for urban expansion assessment, wetland protection and ecological decision-making.

[0171] Table 1 Comparative effect data of the method of the present invention and the traditional method

[0172] ;

[0173] By analyzing Table 1, it can be clearly seen that the present invention has significant advantages in the accuracy and effect of monitoring dynamic changes in the surface. First of all, judging from the results of wetland area monitoring, the measured wetland area decreased year by year from 2018 to 2022. The wetland area detection values ​​of the method of the present invention in each year are always highly close to the measured values, and the error each year is much smaller than that of the traditional method. For example, the measured wetland area in 2022 was 359.5 square kilometers, the detection value of the present invention was 358.2 square kilometers, and the traditional method was 351.4 square kilometers. The error of the traditional method is significantly greater than that of the present invention.

[0174] Secondly, in terms of wetland boundary detection accuracy, the method of the present invention has continued to improve over the past five years, rising from 92.8% to 94.6%, consistently outperforming traditional methods, indicating a stronger ability to identify complex land boundaries and dynamically changing areas. The monitoring of urban land also reflects the advantages of the present invention. As the measured area of ​​urban land continues to grow, the detection value of the present invention is highly consistent with the measured data. The detection value in 2022 was 249.2 square kilometers, while the actual measurement was 246.7 square kilometers, and the accuracy of urban boundary detection remained above 90%, and increased year by year to 92.9%. Traditional methods have greater errors in urban land detection, and the boundary detection accuracy is always lower than that of the present invention.

[0175] Comprehensive tabular data shows that the method of the present invention not only improves the spatial resolution and boundary discrimination capabilities of dynamic monitoring, but also more accurately reflects the changing trends of surface cover. Compared with the traditional regular grid method, the present invention significantly reduces the error in wetland and urban land area detection and improves the accuracy of boundary detection. Its advantages are particularly prominent in areas with drastic changes and complex distribution of land features. This fully demonstrates the effectiveness and application value of the adaptive irregular unit grid method based on the fusion of the locust optimization algorithm and local weighted regression in the field of remote sensing dynamic monitoring.

[0176] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring of land cover by remote sensing based on an adaptive irregular cell grid, characterized in that: The steps include: S1. Collect remote sensing image data of the target monitoring area for multiple time periods, preprocess the remote sensing image data, and construct a standardized remote sensing image dataset; S2. Based on the standardized remote sensing image dataset, the image segmentation method is used to extract the boundary information of the ground objects, and the initial grid structure is constructed according to the spatial structure characteristics and boundary complexity of the ground objects to generate irregular unit grids; S3. Based on the irregular unit grid, calculate the change sensitivity index of each grid unit. When the sensitivity is higher than the set threshold, perform the local grid refinement step to form an adaptive irregular unit grid; S4, based on the adaptive irregular unit grid, the remote sensing image pixels in each unit grid are used as candidate sample points, and the spatial position and ground feature attributes are extracted to construct the sample feature vector; S5. Using the sample feature vector as input, perform position update, population behavior simulation, and solution space search in the locust algorithm to obtain the optimal spatial weight distribution of sample points within each irregular unit grid; S6. Based on the optimal spatial weight distribution of sample points, perform weighted fitting on the remote sensing time series data in each adaptive irregular unit grid, model the evolution trend of surface cover attributes over time, and obtain weighted regression results that reflect the dynamic changes of the surface; S7. Based on the weighted regression results, extract the change direction, change amplitude and period characteristics of each irregular unit grid, and generate the surface cover prediction image of the target area in the current phase and the future phase; S8. Collect the latest phase of remote sensing imagery with real annotation data, compare and analyze it with the surface cover prediction image, and update the locust algorithm population parameters based on the error feedback results; The S2 specifically includes: S24, perform Delaunay triangulation operation, generate a triangular mesh structure on the seed point set, and obtain an initial mesh structure set , where each triangle unit Corresponding to an irregular cell grid candidate area; S25, based on each triangle unit Complexity of terrain texture , boundary curvature and spatial heterogeneity index , construct the boundary complexity function: ； in, is the weighting coefficient; S26, based on the boundary complexity function The values ​​of are used to sort the initial grid cells and set the complexity screening threshold ,when , the corresponding unit is included in the refinement process to complete the construction of the initial irregular unit grid; The S3 specifically includes: S31, for the generated irregular unit grid set , in each unit grid Based on the standardized remote sensing image dataset, the corresponding multi-temporal spectral data, texture data and boundary morphological parameters are extracted to construct a set of change sensitivity indicators. ; S32, the change sensitivity index set It includes three sensitivity quantitative indicators: Regional spectral variability :Indicates unit The degree of time variation dispersion of the mean value of the inner main band; Image texture variation index : Indicates the changing trend of texture contrast in multi-temporal gray-level co-occurrence matrix; Boundary complexity function value : generated by the boundary complexity function; S33. Normalize the three sensitivity quantitative indicators and construct a comprehensive sensitivity scoring function: ； in, 、 、 are the normalized spectral variability, texture change index and boundary complexity value, 、 and is the weight coefficient; S34. Setting sensitivity score threshold set , when the unit grid Rating value When the Layer local grid refinement processing; S35, perform layered refinement processing, in the original unit grid Internal Generation Layer nested triangulation, using Delaunay triangulation method to insert additional seed points and construct refined grid subsets ,in is the number of sub-grid units, ; S36. Summarize all refined grid cells to generate an updated irregular cell grid set , as the adaptive irregular unit grid.

2. The method for dynamic monitoring of land cover by remote sensing based on adaptive irregular unit grid according to claim 1, characterized in that: The remote sensing image data specifically includes multi-temporal multispectral images, high-resolution visible light images and vegetation index products.

3. The method for dynamic monitoring of land cover by remote sensing based on adaptive irregular unit grid according to claim 1, characterized in that: The pre-processing of remote sensing image data specifically includes radiation correction, geometric correction, band fusion and unified coordinate projection processing.

4. The method for dynamic monitoring of land cover by remote sensing based on adaptive irregular unit grid according to claim 1, characterized in that: The S2 specifically also includes: S21. Based on the constructed standardized remote sensing image dataset, a multi-scale object-oriented image segmentation method is used to segment the remote sensing image, extract the boundary vector information of the object, and project the boundary vector information of the object into a unified coordinate system; S22: Constructing a spatial adjacency relationship graph based on the feature boundary vector information ,in, Represents the set of ground objects after segmentation, Indicates the adjacency relationship between objects; S23. Based on the spatial adjacency graph, the Voronoi diagram construction method is used to preliminarily divide the monitoring area, and the seed point set is set as , where each seed point Corresponding to the geometric center coordinates of a ground object, , n is the total number of ground objects.

5. The method for dynamic monitoring of land cover by remote sensing based on adaptive irregular unit grid according to claim 1, characterized in that: The S4 specifically includes: S41, Generative Adaptive Irregular Cell Grid Set , in each irregular unit grid Extract the current grid and spatial neighborhood All temporal remote sensing image pixel data within form a set of candidate sample points , represents the total number of adaptive irregular cell grids, represents the number of sample points in the unit grid and spatial neighborhood; S42, sample point set Each sample point , get the spatial position coordinates , extract multi-dimensional feature attribute vector , and historical multi-temporal statistical characteristics ,in is the mean, is the standard deviation, is the maximum difference; S43. Construct the space-attribute-time series joint feature vector for each sample point : ； in, represents the spatial ordinate of the jth sample point in the i-th unit grid, Represents the multidimensional feature attribute characteristics of the jth sample point in the i-th unit grid, is the feature dimension; S44, the joint feature vector of all sample points Composition of sample feature vector set ; S45. Output the sample feature vector set of all irregular unit grids , used for locust algorithm weight optimization processing.

6. The method for dynamic monitoring of land cover by remote sensing based on adaptive irregular unit grid according to claim 1, characterized in that: The S5 specifically includes: S51, based on the output sample feature vector set , the sample feature matrix of each irregular unit grid is used as the input of the locust algorithm; S52, initialize the hierarchical population structure and divide the entire population into subpopulations, each subpopulation is independently distributed in the global search space, and each subpopulation initializes its individual position, velocity and search parameters; S53. Within each subpopulation, the positions of the individuals in the population are updated according to the social interaction force, gravity force and wind force, and the current optimal fitness position of each individual is recorded and set as the local optimal position. ; S54, when each individual updates its position, it adds a local learning and memory strategy to the local optimal position With weight coefficient Introduced into the update formula, the new individual position is : ； in is the individual's current location, is the memory impact factor, is the social interaction force vector of the jth individual in the i-th subpopulation, is the gravitational force vector acting on the j-th individual in the i-th subpopulation, is the wind force vector acting on the jth individual in the i-th subpopulation; S55. For each individual in each subpopulation, regularly check the historical optimal fitness, continuously If the fitness of the generation does not improve, the local memory backtracking strategy is triggered, and the individual jumps back to the vicinity of the local historical optimal position and conducts a neighborhood re-search; S56, introduce individual behavior diversification mechanism, and classify some individuals according to the mutation probability Perform Gaussian perturbation or Levy jump, the new position of the individual is : ； in is the disturbance parameter, is a Gaussian distribution; S57. Set up an information exchange mechanism between subpopulations. At a specified iteration cycle or when the global optimum has not improved for a long time, exchange the elite individuals of each subpopulation, introduce the optimal solution of the higher-level subpopulation into the lower-level subpopulation, or perform random exchange of individual positions. S58, all individuals after each round of iteration according to the target fitness function of the sample feature vector Perform fitness evaluation with the goal of minimizing weighted regression error and controlling the smoothness of spatial weight distribution; S59, loop through steps S53 to S58 until the maximum number of iterations or the fitness convergence condition is reached, and output the optimal spatial weight distribution of sample points for each irregular unit grid. ; S510, the optimal sample point spatial weight distribution set of all irregular unit grids , for use in weighted regression model building and surface cover dynamic monitoring and analysis.

7. The method for dynamic monitoring of land cover by remote sensing based on adaptive irregular unit grid according to claim 1, characterized in that: The S6 specifically includes: S61. Based on the obtained optimal sample point spatial weight distribution, the sample feature vectors in each adaptive irregular unit grid are matched with the corresponding weights to form a weighted sample feature set; S62. In each adaptive irregular unit grid, select corresponding remote sensing time series data as input data for regression modeling, covering land cover attribute observations of all historical phases; S63, matching the weighted sample feature set with the remote sensing time series data, assigning corresponding weights to the land cover attribute values ​​of each historical phase, and constructing a weighted data sequence; S64, for each adaptive irregular cell grid, performing a regression modeling step based on the weighted data sequence to fit the evolution trend of the surface cover attributes over time; S65. Extracting the change trend of the surface cover attributes of each adaptive irregular unit grid according to the results of the weighted regression modeling, including the change direction, change amplitude, and change period; S66. Output the weighted regression results of each adaptive irregular unit grid as a basis for trend analysis reflecting dynamic changes in the surface.

8. A system for dynamic monitoring of land cover by remote sensing based on an adaptive irregular unit grid, applying the method for dynamic monitoring of land cover by remote sensing based on an adaptive irregular unit grid according to any one of claims 1 to 7, characterized in that: Includes the following modules: Data acquisition module, used to collect remote sensing image data of the target monitoring area for multiple time periods; The data preprocessing module is used to perform radiation correction, geometric correction, band fusion and unified coordinate projection processing on remote sensing image data to generate a standardized remote sensing image dataset; The grid construction module is used to extract the boundary information of the ground objects based on the standardized remote sensing image dataset, construct the initial irregular unit grid and perform adaptive refinement to generate an adaptive irregular unit grid set; The feature extraction module is used to extract the spatial location, multi-dimensional feature attributes and time series statistical features of sample points in the adaptive irregular unit grid and construct a set of sample feature vectors; The weight optimization module is used to execute the hierarchical population locust algorithm based on the sample feature vector set to obtain the optimal spatial weight distribution of sample points in each irregular unit grid; The weighted regression modeling module is used to perform weighted fitting based on the spatial weight distribution of the optimal sample points and remote sensing time series data, model the evolution trend of surface cover attributes over time, and generate weighted regression results; The change analysis and prediction module is used to extract the characteristics of land cover change trends based on weighted regression results and output land cover prediction and dynamic monitoring results; The result feedback and update module is used to collect the real annotation data of remote sensing images in the latest phase, and continuously update and adaptively optimize the weight optimization algorithm based on the error feedback mechanism.

Images