Land resource dynamic monitoring and early warning method and system based on multi-source remote sensing data fusion

Abstract

The invention relates to the technical field of land resource monitoring, in particular to a land resource dynamic monitoring and early warning method and system based on multi-source remote sensing data fusion, and the method comprises the steps: employing an unmanned plane to periodically collect optical images, SAR echoes and LiDAR point clouds, constructing a ground three-dimensional digital model, and carrying out the land parcel division; performing fusion to form a multi-dimensional feature vector, establishing an LSTM land parcel feature evolution model, and predicting a change rate interval of each feature in a current period based on a historical sequence; constructing a time sequence difference change detection algorithm, calculating a land parcel change rate, and screening potential abnormal land parcels by taking a prediction interval as an anomaly judgment threshold value; a double-branch convolutional neural network is adopted to identify crop states, growth stages and construction violation behaviors, abnormity is judged and determined, and confidence is given; spatial clustering is carried out on determined abnormal land parcels, accurate boundaries are obtained in combination with a three-dimensional model, multi-level early warning information is generated, and the decision-making efficiency and response speed of land resource monitoring are improved.

Description

Technical Field

[0001] This application relates to the field of land resource monitoring technology, specifically to a method and system for dynamic monitoring and early warning of land resources based on multi-source remote sensing data fusion. Background Technology

[0002] Traditional land resource monitoring relies primarily on manual inspections and satellite remote sensing image interpretation, which suffers from drawbacks such as long monitoring cycles, poor timeliness, and limited coverage. Existing remote sensing monitoring technologies often employ a single data source, such as using only optical imagery for land cover classification or relying solely on radar data for change detection. This single-data-source monitoring method faces numerous limitations in practical applications: optical imagery is susceptible to cloud and fog conditions, making all-weather monitoring impossible; while synthetic aperture radar can penetrate clouds, image interpretation is difficult, hindering accurate identification of land cover types; and while lidar can acquire precise three-dimensional information, it lacks spectral characteristics, making it unable to effectively distinguish between different land use types.

[0003] Existing methods often rely on fixed thresholds to identify abnormal changes, which are ill-suited to the diverse land use patterns across different regions and seasons. This can easily lead to misjudging normal agricultural activities as abnormal changes or overlooking genuine illegal land use. Furthermore, current land resource monitoring suffers from deficiencies in anomaly identification and early warning response. Monitoring results often only provide location information for the changed area, failing to accurately determine the specific type and severity of the change, and lacking refined classification capabilities. Early warning information generation is also rather crude, unable to provide tiered warnings based on the type, scale, and confidence level of the anomaly, thus impacting the efficiency and speed of decision-making in land resource monitoring.

[0004] In view of this, this application proposes a method and system for dynamic monitoring and early warning of land resources based on the fusion of multi-source remote sensing data. Summary of the Invention

[0005] To achieve the above objectives, this application provides a method and system for dynamic monitoring and early warning of land resources based on multi-source remote sensing data fusion, the specific technical solution of which is as follows:

[0006] A dynamic monitoring and early warning method for land resources based on multi-source remote sensing data fusion includes:

[0007] By periodically collecting ground optical images, SAR echoes and LiDAR point cloud data by UAVs, a ground 3D digital model is generated based on the LiDAR point cloud data and the land parcels are divided. A spatiotemporal registration algorithm is constructed to map the ground optical images and SAR echoes acquired in the same period to the corresponding land parcels, thereby generating a ground 3D digital model.

[0008] For the divided plots, texture feature parameters of optical images are extracted, backscattering coefficient and polarization features of SAR echoes are extracted, and elevation statistics and surface roughness of LiDAR point clouds are extracted. Multidimensional feature vectors of plots are generated through feature fusion algorithms.

[0009] Construct a land parcel feature evolution model based on a long short-term memory network, infer the current period's land parcel change pattern based on historical periodic multidimensional feature vectors of land parcels, and generate a range of land parcel feature change rates;

[0010] A change detection algorithm based on time-series difference is established. The change rate of land parcels in the current period is calculated based on the multi-dimensional feature vector of land parcels. The range of land parcel feature change rate is used as the judgment threshold range of potential abnormal land parcels. Potential abnormal land parcels are marked according to the judgment threshold range of potential abnormal land parcels.

[0011] A land parcel status identification model based on a dual-branch convolutional neural network is constructed. The optical images and SAR echoes of potential anomalous land parcels are input into the land parcel status identification model to identify the planting status and growth stage of crops on the ground and illegal construction behavior on the ground. The model determines whether potential anomalous land parcels are confirmed anomalous land parcels and outputs the classification results and confidence scores of confirmed anomalous land parcels.

[0012] Spatial clustering analysis is performed on identified anomalous land parcels to form ground anomaly regions. The geographical boundaries of these regions are then determined using a 3D digital model of the ground. Finally, multi-level early warning information is generated based on the classification results and confidence scores of the identified anomalous land parcels.

[0013] Preferably, an integrated optical camera, synthetic aperture radar, and lidar scanner are used on a drone to collect optical images, SAR echoes, and LiDAR point cloud data of the monitored area.

[0014] Based on the collected LiDAR point cloud data, after noise removal and point cloud thinning preprocessing, and after separating ground points from non-ground points, an irregular triangular mesh interpolation method is used to generate a ground 3D digital model.

[0015] A watershed algorithm based on slope and curvature, combined with edge detection results from optical imagery, is used to divide and encode the 3D digital model of the ground.

[0016] A spatiotemporal registration algorithm is constructed to perform geometric correction and registration of optical images and SAR echoes and map them to the corresponding land parcels.

[0017] Preferably, the optical image is converted into a grayscale image, and the grayscale co-occurrence matrix method is used to calculate and extract energy, contrast, correlation and entropy at a set distance and multiple directions, and the feature values ​​in multiple directions are averaged to obtain feature parameters.

[0018] Radiometric calibration and speckle noise suppression were performed on the SAR echoes. The backscattering coefficient was calculated, and a scattering matrix was constructed from the SAR echo data. The polarization decomposition method was used to extract the components characterizing surface scattering, dihedral scattering, and volume scattering. At the same time, the polarization entropy and the average scattering angle were calculated.

[0019] Preferably, for the point cloud data within each plot, calculate the elevation statistics parameters, including the average elevation, elevation standard deviation, maximum elevation, and minimum elevation;

[0020] Surface roughness is calculated by dividing the point cloud within the plot into a grid and using a local plane fitting method.

[0021] After extracting the feature parameters from each data source, a feature fusion algorithm is used to normalize all the extracted feature parameters. The normalized optical texture features, SAR polarization features, and LiDAR elevation features are then spliced ​​together in a predetermined order to generate a multi-dimensional feature vector of the land parcel.

[0022] Preferably, a land parcel feature evolution model is constructed based on a long short-term memory network;

[0023] The multidimensional feature vectors of land parcels in historical cycles are organized into a time series, and the feature change rate series between adjacent weeks are calculated and standardized as input to the land parcel feature evolution model.

[0024] The architecture of a Long Short-Term Memory (LSTM) network is designed, including an input layer, a fully connected layer, and an output layer. The input layer includes an LSTM layer with a forget gate, an input gate, and an output gate. The output layer is used to output the predicted value of the feature change rate for the next cycle and an uncertainty estimate for quantifying the prediction confidence.

[0025] Preferably, based on the predicted value of the feature change rate and the uncertainty estimate of the plot feature evolution model, a change rate confidence interval is constructed for each dimension feature of the plot, forming a multidimensional feature change rate interval for the plot.

[0026] Preferably, the multidimensional feature vectors of land parcels for the current and historical monitoring periods are obtained, the feature difference vector is calculated, and the land parcel change rate for the current period is calculated.

[0027] The weights are determined based on the coefficients of variation of each feature dimension in historical data, and the overall change rate of the land parcel is calculated using weighted Euclidean distance.

[0028] Calculate the overall relative change rate of the land parcel, as well as the positive and negative change rates that characterize the degree of increase and decrease in the feature value.

[0029] Preferably, a dynamic judgment threshold range is set based on the range of land parcel feature change rate. When the calculated land parcel feature change rate exceeds the dynamic judgment threshold range, the abnormality of the land parcel feature deviating from the prediction range is calculated and summarized into the comprehensive change rate abnormality of the land parcel.

[0030] A multi-criteria anomaly detection mechanism is constructed. When a plot of land meets any of the conditions of the anomaly detection mechanism, the plot is marked as a potential anomaly plot.

[0031] The anomaly detection mechanism includes: the existence of a preset number of feature dimensions whose rate of change exceeds the prediction interval; the anomaly degree of the comprehensive rate of change of the land parcel being greater than a preset threshold; and the comprehensive rate of change of the land parcel exceeding the comprehensive rate of change range derived from the boundaries of each dimension of the prediction interval.

[0032] Preferably, a land parcel status recognition model based on a dual-branch convolutional neural network is constructed to identify whether a potential abnormal land parcel is a confirmed abnormal land parcel;

[0033] The dual-branch convolutional neural network processes optical images and SAR data separately. The optical image branch adopts a residual connection structure and inserts a multi-scale feature extraction layer composed of multiple parallel dilated convolutions to capture features of different receptive fields.

[0034] The SAR data branch adds polarization feature coding to the input of the optical image branch network structure.

[0035] Preferably, an attention-based feature fusion mechanism is constructed to adaptively calculate the weights of the optical and SAR features and perform weighted fusion.

[0036] A multi-task learning structure is designed at the end of a dual-branch convolutional neural network, and parallel classification heads are set up to simultaneously identify crop planting status, growth stage and illegal construction behavior on the ground.

[0037] Based on the prediction results of the land parcel status identification model for the identification task, a comprehensive confidence score is calculated. Combined with the preset threshold for illegal construction and the confidence threshold for crop anomalies, the system determines whether a potential abnormal land parcel is a confirmed abnormal land parcel and outputs the classification results.

[0038] Preferably, spatial clustering analysis is performed on the identified anomalous land parcels, and a density-based clustering algorithm is used to merge spatially adjacent land parcels with the same anomalous category into continuous ground anomalous areas.

[0039] Using a 3D digital model of the ground, the geographic boundaries of the anomaly region are identified and the geometric properties of the anomaly region are calculated through the convex hull algorithm and the Alpha-shape algorithm.

[0040] Preferably, a multi-level early warning mechanism is constructed, which calculates a risk score based on the severity of the anomaly type, the confidence level of identification, and the area of ​​the region to determine the early warning level, and generates a structured early warning report containing information on the location and type of the anomaly.

[0041] The land resource dynamic monitoring and early warning system based on multi-source remote sensing data fusion is used to implement the land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion, including: a data acquisition module, a feature extraction module, a feature change rate prediction module, a potential abnormal land parcel identification module, an abnormal land parcel identification and classification module, and a hierarchical early warning module.

[0042] The data acquisition module periodically collects ground optical images, SAR echoes, and LiDAR point cloud data through UAVs. Based on the LiDAR point cloud data, it generates a ground three-dimensional digital model and divides the land into plots. It constructs a spatiotemporal registration algorithm to map the ground optical images and SAR echoes acquired in the same period to the corresponding plots, thereby generating a ground three-dimensional digital model.

[0043] The feature extraction module extracts texture feature parameters from optical images, backscattering coefficients and polarization features from SAR echoes, and elevation statistics and surface roughness from LiDAR point clouds for the divided land parcels, and generates multidimensional feature vectors for the land parcels through feature fusion algorithms.

[0044] The feature change rate prediction module constructs a land parcel feature evolution model based on a long short-term memory network, infers the current period's land parcel change pattern based on historical periodic multi-dimensional feature vectors of land parcels, and generates a land parcel feature change rate interval.

[0045] The potential abnormal land parcel identification module establishes a change detection algorithm based on time-series difference, calculates the land parcel change rate of the current period based on the multi-dimensional feature vector of the land parcel, uses the range of land parcel feature change rate as the judgment threshold range of potential abnormal land parcels, and marks potential abnormal land parcels according to the judgment threshold range of potential abnormal land parcels.

[0046] The abnormal land parcel identification and classification module constructs a land parcel status identification model based on a dual-branch convolutional neural network. It inputs the optical images and SAR echoes of potential abnormal land parcels into the land parcel status identification model to identify the planting status, growth stage of ground crops, and illegal construction behavior on the ground. It determines whether potential abnormal land parcels are confirmed abnormal land parcels and outputs the classification results and confidence scores of confirmed abnormal land parcels.

[0047] The hierarchical early warning module performs spatial clustering analysis on identified abnormal land parcels to form ground anomaly areas, marks the geographical boundaries of the ground anomaly areas based on the ground three-dimensional digital model, and generates multi-level early warning information based on the classification results and confidence scores of the identified abnormal land parcels.

[0048] The beneficial effects of this application are as follows: This application achieves comprehensive and three-dimensional acquisition of ground information through the collaborative acquisition and spatiotemporal registration of multi-source remote sensing data. The three-dimensional digital model generated from LiDAR point clouds provides a high-precision terrain foundation, offering a unified geographic reference framework for subsequent precise land parcel delineation and spatial matching of multi-source data.

[0049] This application constructs a multi-dimensional description system for land parcels by extracting complementary feature parameters from different remote sensing data sources. The multi-dimensional feature vectors fully utilize the respective advantages of optical, microwave, and lidar technologies to achieve a comprehensive representation of land cover type, physical structure, and geometric morphology.

[0050] This application uses a long short-term memory network to establish a land parcel feature evolution model, which can effectively capture the temporal dependence and long-term change trend of land parcel features; by generating a range of land parcel feature change rates, it provides a dynamic benchmark for anomaly detection.

[0051] This application presents a change detection algorithm based on temporal difference, which can accurately quantify the actual degree of change in land parcels and achieve adaptive screening of abnormal land parcels by combining dynamic threshold ranges. It effectively reduces the false alarm rate caused by seasonal changes and normal agricultural activities, significantly improving the accuracy of land resource monitoring.

[0052] The dual-branch convolutional neural network constructed in this application fully utilizes the complementary advantages of optical and SAR data to achieve accurate identification of land parcel status in all weather and all time. The model can accurately determine the authenticity of abnormal land parcels, output specific anomaly types and confidence scores, and improve the accuracy of illegal land use identification.

[0053] This application integrates discrete anomalous plots into continuous anomalous areas through spatial clustering analysis, improving the practicality of monitoring results; combined with the precise geographic location of the three-dimensional digital model and the multi-level early warning mechanism, it provides regulatory authorities with accurate spatial location information and differentiated response strategies. Attached Figure Description

[0054] Figure 1 The flowchart of the land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion provided in this application;

[0055] Figure 2 A flowchart of multidimensional feature extraction and fusion of multi-remote sensing data provided in this application;

[0056] Figure 3 A flowchart for modeling and predicting the evolution of land parcel features provided for this application;

[0057] Figure 4 The flowchart for change detection and potential anomaly identification provided for this application;

[0058] Figure 5The flowchart for land parcel status identification and anomaly detection provided in this application;

[0059] Figure 6 The structural diagram of the land resource dynamic monitoring and early warning system based on multi-source remote sensing data fusion provided in this application. Detailed Implementation

[0060] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0061] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0063] Example 1

[0064] Reference Figures 1 to 4 This is the first embodiment of the present application, such as Figure 1 As shown, a method for dynamic monitoring and early warning of land resources based on multi-source remote sensing data fusion is provided.

[0065] Step 1: Periodically collect ground optical images, SAR echoes and LiDAR point cloud data using UAVs. Generate a ground 3D digital model based on the LiDAR point cloud data and divide the land into plots. Construct a spatiotemporal registration algorithm to map the ground optical images and SAR echoes acquired in the same period to the corresponding plots, thereby generating a ground 3D digital model.

[0066] Ground data is collected using a multi-sensor platform mounted on a drone. The drone flies at a fixed altitude and speed along a preset route, with its onboard optical camera, synthetic aperture radar (SAR), and lidar scanner working synchronously to collect optical images, SAR echoes, and LiDAR point cloud data of the monitored area. The optical camera acquires high-resolution images in the visible and near-infrared bands, the SAR sensor emits microwave signals and receives SAR echo data reflected from the ground, and the lidar scanner emits laser pulses and records the time and intensity information of the returned signals to form LiDAR point cloud data. By integrating a high-precision GNSS receiver and inertial measurement unit (IMU) on the drone, the position and attitude parameters of the flight platform are recorded in real time, providing accurate geographic reference information for subsequent data processing.

[0067] A 3D digital model of the ground is constructed based on the acquired LiDAR point cloud data. The raw point cloud data undergoes preprocessing, including noise point removal and point cloud thinning. Statistical filtering algorithms are used to remove outlier noise points, and then a voxel mesh downsampling method is used to thin the point cloud, dividing the 3D space into sections with side lengths of... The cubic voxels are used, and all points within each voxel are replaced with its centroid, which reduces data redundancy while preserving terrain features.

[0068] Ground points are separated from non-ground points using an improved progressive morphological filtering algorithm. Based on the separated ground points, a continuous digital elevation model is generated using the triangular network of irregularities (TIN) interpolation method with an interpolation resolution of 0.5 meters, forming a complete three-dimensional digital ground model.

[0069] The generated 3D digital model of the ground is divided into plots, and a watershed algorithm based on slope and curvature is used to identify terrain feature lines and calculate the slope of each grid point. and curvature , where z is the elevation value, and x and y are the plane coordinates. or The time markers are used as terrain feature points. and These are the slope threshold and curvature threshold, respectively. For example, The value is 15°. The value is set to 0.1. Feature points are connected to form the plot boundary lines. Boundary optimization is performed by combining the edge detection results from the optical imagery. The Canny edge detection operator is used to extract image edges, which are then fused with terrain feature lines to generate accurate plot boundaries. The divided plots are coded, assigned unique identifiers, and their spatial extent, center coordinates, and area attributes are recorded.

[0070] A spatiotemporal registration algorithm for multi-source data is constructed to achieve accurate mapping of optical imagery and SAR echo data to land parcels. Geometric correction is performed on the sensors, and based on the UAV's position and attitude information acquired by the differential GPS / INS integrated navigation system onboard the UAV, a transformation relationship from the sensor coordinate system to the geographic coordinate system is established. For optical imagery data, the sensor coordinates of the UAV-acquired image data are mapped to geographic coordinates. For SAR echo data, a range-Doppler model is used for geometric positioning, establishing a mapping relationship between slant range image coordinates and ground coordinates.

[0071] Based on geometric correction, a feature-point-based fine registration method is employed to improve registration accuracy. SIFT feature points are extracted from optical and SAR images, and corresponding point pairs are obtained through feature descriptor matching. The RANSAC algorithm is used to eliminate mismatched points, and the affine transformation parameters between images are estimated. The registered optical images and SAR echo data are cropped and resampled according to the plot boundaries, with each plot corresponding to a set of multi-source remote sensing data slices. A plot data index table is established to record the storage location and metadata information of each plot in different time phases and different sensor data, including acquisition time, sensor parameters, and auxiliary information such as weather conditions.

[0072] This step utilizes multi-source remote sensing technology to monitor the ground in multiple aspects and efficiently manages the collected multi-source remote sensing data, providing a standardized data foundation for subsequent feature extraction and change detection, thereby improving the automation level and processing efficiency of land resource monitoring.

[0073] Step 2: For the divided plots, extract texture feature parameters from the optical imagery, extract backscattering coefficients and polarization features from the SAR echoes, and extract elevation statistics and surface roughness from the LiDAR point clouds. Then, generate multi-dimensional feature vectors for each plot using a feature fusion algorithm. See also... Figure 2 This is a flowchart of the multidimensional feature extraction and fusion process for multi-remote sensing data in this step.

[0074] For each plot of land defined in step 1, multi-dimensional feature parameters are extracted from optical imagery, SAR echo data, and LiDAR point cloud data. For texture feature extraction from optical imagery, the RGB image corresponding to each plot is first converted to grayscale imagery, and the gray-level co-occurrence matrix (GLCM) method is used to calculate texture features. When calculating the GLCM, spatial relationship parameters between pixel pairs are set, including distance d and direction θ, where direction θ takes four values: 0°, 45°, 90°, and 135°, and distance d is determined based on the image resolution. For example, when the image resolution is 0.1 m / pixel, d takes values ​​of 1, 3, and 5 pixels. A normalized GLCM is then constructed. , where i and j represent the gray values ​​of pixel pairs, and texture feature parameters are calculated based on this matrix.

[0075] Four main texture features—energy, contrast, correlation, and entropy—are extracted from the gray-level co-occurrence matrix of optical images. Energy features... The formula reflects the coarseness of image texture and the uniformity of grayscale distribution. Where L is the number of gray levels. Contrast characteristics characterize the sharpness of the texture, and are calculated using the formula: Contrast Characteristics Correlation characteristics The formula describes the degree of linear correlation between pixel gray levels. ,in , These are the mean values ​​in the row and column directions, respectively. , This represents the corresponding standard deviation. Entropy feature. The formula reflects the complexity of the texture. The final value of each texture feature is obtained by averaging the feature values ​​in the four directions. This multi-directional averaging process improves the rotation invariance of the features.

[0076] For feature extraction from SAR echo data, radiometric calibration and speckle noise suppression are performed first. A Lee filter is used to filter the SAR image, with a filter window size of 7×7 pixels to effectively suppress speckle noise while preserving edge information. The backscattering coefficients for each polarization channel are calculated, and the digital numerical values ​​(DN) are converted to backscattering coefficients. The conversion formula is: Where K is a calibration constant determined by sensor parameters, and DN is the original integer value stored for each pixel after sensor imaging and analog-to-digital conversion of the remote sensing image, used to represent the relative magnitude of the electromagnetic signal strength received by that pixel. For SAR data, a scattering matrix S is constructed, and polarization feature parameters are extracted.

[0077] Based on the scattering matrix S, the polarization decomposition characteristics are calculated. The Pauli decomposition method is used to decompose the scattering matrix into three basic scattering mechanism components: single scattering component. Characterizing surface scattering properties, secondary scattering component Reflecting dihedral scattering characteristics, volume scattering component Exhibits bulk scattering characteristics, and These are the diagonal elements of the scattering matrix S of a fully polarimetric SAR. Representing the horizontal polarization of the transmitter and the horizontal polarization of the receiver, the resulting complex scattering coefficients are: , This represents the complex scattering coefficients obtained from both the transmit and receive vertical polarizations. The polarization entropy is then calculated. and average scattering angle The formula for calculating polarization entropy is: ,in The probability is the normalized eigenvalue of the coherence matrix. The gray level index parameter for pixel pairs, and the average scattering angle. , The scattering angles corresponding to each feature vector; polarization features can effectively characterize the scattering mechanism and physical properties of ground objects, and improve the accuracy of ground object classification.

[0078] For LiDAR point cloud data, elevation statistics and surface roughness features of each site are extracted. Elevation statistics parameters, including average elevation, are calculated for the point cloud data within each site. Standard deviation of elevation Maximum elevation and minimum elevation Where N is the number of point clouds within the plot. Let k be the elevation value of the k-th point. Calculate the elevation range. and elevation variation coefficient These parameters reflect the vertical structural characteristics of the land parcel.

[0079] Surface roughness is calculated using a local plane fitting method. The point cloud within the plot is divided into grids, for example, with a grid size of 1 meter × 1 meter. Within each grid, a local plane is fitted using the least squares method; the distance from each point to the fitted plane is then calculated. Surface roughness is defined as Where M is the number of grid cells, This represents the number of points within the m-th grid. The terrain slope is also calculated simultaneously. and terrain relief , For the first The average elevation of each grid cell, with X and Y corresponding to the actual distance axes on the ground in the row and column directions of the grid cell, respectively.

[0080] After extracting features from various data sources, a feature fusion algorithm is used to generate multidimensional feature vectors for the land parcels. First, all features are normalized using a min-max normalization method to map feature values ​​to the [0,1] interval. Then, optical texture features, SAR polarization features, and LiDAR elevation features are sequentially concatenated to form a feature vector of dimension D. , It is the Dth normalized feature component.

[0081] This step fully leverages the complementary advantages of different sensor data through multi-source data feature extraction and fusion methods. Optical images provide spectral and texture information of the ground, SAR data reflects the microwave scattering characteristics of the ground, and LiDAR data accurately describes the three-dimensional structure of the terrain. The fused multi-dimensional feature vector comprehensively characterizes the integrated attributes of the land parcel, providing a rich and reliable feature foundation for subsequent change detection and anomaly identification, which can significantly improve the accuracy and reliability of land resource monitoring.

[0082] Step 3: Construct a land parcel feature evolution model based on a Long Short-Term Memory (LSTM) network. Based on the multi-dimensional feature vectors of land parcels from historical cycles, infer the current cycle's land parcel change patterns and generate intervals for land parcel feature change rates. (See also...) Figure 3 This is a flowchart for modeling and predicting the evolution of land parcel features in this step.

[0083] A land parcel feature evolution model based on a Long Short-Term Memory (LSTM) network is constructed to learn and predict the temporal variation patterns of multidimensional feature vectors of land parcels. Historical time-series data is preprocessed to organize the multidimensional feature vectors of each land parcel into a time series across different monitoring periods. Let the feature vector of land parcel a in the t-th monitoring period be... Construct a historical feature sequence for this plot of land. Where T is the total number of historical monitoring periods. Calculate the vector of historical characteristic change rates between adjacent weeks. ,in To monitor the periodic intervals, a rate of change series was constructed. .

[0084] To eliminate the influence of different feature units and numerical ranges, the rate of change data is standardized, and the mean of the rate of change for each feature dimension is calculated. and standard deviation ,in The rate of change of the b-th dimension of feature of plot a in the t-th period is represented by the standardized rate of change. .

[0085] The architecture of the LSTM network is designed, comprising an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives the sequence of feature change rates within a sliding time window; the LSTM layer contains multiple LSTM units, each containing three gating mechanisms: a forget gate, an input gate, and an output gate. The forget gate determines which information from the previous cell state needs to be forgotten, the input gate controls the degree of reception of current input information, the cell state is updated by combining the outputs of the forget gate and the input gate, and the output gate controls the output from the cell state to the hidden state.

[0086] A fully connected layer is connected after the LSTM layer for feature mapping. The fully connected layer uses the ReLU activation function. Increase non-linear expressive power. The output layer produces two outputs, including: a predicted value of the feature rate of change. and uncertainty estimation The predicted rate of change represents the expected rate of change vector for the land parcel in the next period, while the uncertainty estimate represents the variance of the prediction, used to construct the confidence interval for the rate of change. The output layer uses a linear activation function with the same dimension as the input feature vector.

[0087] The land parcel feature evolution model is trained using historical data to construct a training set and employs the mean squared error loss function. Optimize prediction accuracy, among which The number of training samples. The true rate of change is used, while introducing an uncertainty loss term. This allows the model to learn about the uncertainty of predictions. Total loss function. ,in The balancing coefficient is used. The Adam optimizer is used for model training, and an early stopping strategy is employed to prevent overfitting.

[0088] After training, the feature change rate interval is generated for each plot using the plot feature evolution model. For plot a, the feature change rate sequence of its most recent w periods is input, and the model outputs the predicted change rate for the next period. and uncertainty estimation .

[0089] Based on the prediction results, a confidence interval for the rate of change of the features is constructed. For the b-th dimension feature, the interval for its rate of change is: ,in, The confidence coefficient is determined based on the required confidence level. This indicates that the land parcel feature evolution model applies to land parcel a in the next period t+1. Point prediction values ​​of dimensional features This represents the variance estimate of the predicted values ​​given by the land parcel feature evolution model. For example, at a 95% confidence level... The value is 1.96. Combining the change rate intervals of each dimension forms the multidimensional characteristic change rate interval of the land parcel. ,in and Let be the lower and upper bounds of the rate of change of the b-th dimension feature, respectively. The number of feature dimensions.

[0090] This step constructs a land parcel feature evolution model based on an LSTM network, enabling effective learning and prediction of the temporal patterns of multidimensional feature change rates for land parcels. The constructed model fully utilizes the temporal dependence information of change rates contained in historical monitoring data, capturing both long-term trends and short-term fluctuation patterns of land parcel feature changes. The generated feature change rate intervals not only provide the expected range of normal change rates for land parcels but also quantify the uncertainty of predictions, providing a scientifically sound benchmark for subsequent anomaly detection. Compared to traditional statistical methods, this deep learning-based temporal modeling method exhibits stronger nonlinear fitting capabilities and adaptability, handling complex multidimensional feature change rate evolution patterns and significantly improving the intelligence level and early warning accuracy of land resource dynamic monitoring.

[0091] Step 4: Establish a change detection algorithm based on temporal difference. Calculate the land parcel change rate for the current period based on the multi-dimensional feature vector of the land parcels. Use the range of land parcel feature change rates as the threshold range for identifying potentially anomalous land parcels. Mark potentially anomalous land parcels according to this threshold range. (See also...) Figure 4 This is a flowchart for change detection and potential anomaly identification in this step.

[0092] In step 4, a change detection algorithm based on temporal difference is established to identify potential abnormal plots by calculating the degree of change of the multidimensional feature vector of the plot during adjacent monitoring weeks.

[0093] Obtain the multidimensional feature vectors of the land parcels in the current monitoring period t and the previous monitoring period t-1. For land parcel a, let its feature vector in the current period be... The eigenvector of the previous period is ,in Indicates land parcel In the current monitoring cycle The 3D eigenvalues Indicates land parcel In the previous monitoring cycle The 3D eigenvalues Let be the dimension of the feature vector. Calculate the feature difference vector between two cycles. ,in This represents the change in the b-th dimension feature. Indicates land parcel The The change in dimensional features between two adjacent monitoring weeks is used to calculate the land parcel change rate for the current period. ,in For the monitoring period time interval, Let be the rate of change of the b-th dimension feature.

[0094] To comprehensively assess the overall degree of change in the land parcels, a weighted Euclidean distance is used to calculate the overall change rate. First, the weight coefficients for each feature dimension are determined. The weights are then determined based on the coefficients of variation of the features in historical data, and the historical coefficient of variation for the b-th feature is calculated. ,in The standard deviation of this feature in historical data. Historical mean; feature weights Defined as Features with smaller coefficients of variation receive greater weight because these features are relatively stable under normal circumstances, and their changes are more likely to reflect anomalies. The overall change rate of plot a. The calculation formula is This rate of change comprehensively reflects the intensity of changes in multiple characteristic dimensions of the land parcel.

[0095] Based on the calculation of the rate of change, the concept of relative rate of change is introduced to eliminate the influence of the characteristic dimensions. For each characteristic dimension, the relative rate of change is calculated. ,in To prevent division by zero of small constants. For example, The value can be 0.001.

[0096] Overall relative change rate of the land parcel for: ;

[0097] Simultaneously calculate the directional change index and define the positive change rate. :

[0098] ;

[0099] and negative rate of change ;

[0100] positive rate of change and negative rate of change Separating the characteristic values ​​to represent the degree of increase and decrease helps identify change patterns in different types of land parcels.

[0101] The dynamic threshold range for anomaly detection is constructed using the parcel feature change rate interval generated in step 3. For the b-th dimension feature of parcel a, its predicted change rate interval is obtained from step 3. ,in This is the lower bound of the rate of change. This is the upper bound of the rate of change. Define the anomaly degree of the characteristic rate of change. ,when hour, ;when hour, ;when hour, .in To prevent division by zero of small constants. For example, The value can be 0.01. Anomaly degree of the comprehensive rate of change of the land parcel. This reflects the extent to which the land parcel change rate deviates from the expected range.

[0102] A multi-criteria fusion anomaly detection mechanism is constructed. Based on the rate of change range provided in step 3, a comprehensive anomaly detection criterion is set. When parcel a meets any of the following conditions, parcel a is marked as a potential anomaly parcel:

[0103] (1) There exists at least The rate of change of each feature dimension exceeds the prediction interval, i.e. ,in (2) Overall rate of change anomaly ,in (3) Overall rate of change The comprehensive interval derived beyond all characteristic rate of change intervals ,in , .

[0104] This step establishes a change detection algorithm based on temporal difference, achieving accurate quantification and anomaly identification of dynamic changes in land parcels. The algorithm fully utilizes the feature change rate interval generated in step 3 as a dynamic threshold, enabling the anomaly judgment criteria to adapt to historical change patterns and seasonal norms of land parcels. The multi-criteria fusion strategy comprehensively considers different situations such as single-dimensional anomalies, multi-dimensional anomalies, and combined anomalies, improving the comprehensiveness of detection. Compared with traditional fixed threshold methods, the dynamic threshold method based on prediction intervals in this step has higher adaptability and a lower false alarm rate, providing reliable candidate targets for subsequent refined identification of land parcel status, effectively improving the accuracy and timeliness of land resource monitoring and early warning.

[0105] Step 5: Construct a land parcel status recognition model based on a dual-branch convolutional neural network. Input optical images and SAR echoes of potential anomalous land parcels into the model to identify the planting status and growth stage of crops on the ground, as well as any illegal construction activities. Determine whether a potential anomalous land parcel is a confirmed anomalous land parcel, and output the classification result and confidence score for confirmed anomalous land parcels. See also Figure 5 This is a flowchart for the land parcel status identification and anomaly detection in this step.

[0106] A land parcel status recognition model based on a dual-branch convolutional neural network is constructed to perform refined identification and classification of potential anomalous land parcels marked in step 4. Optical imagery and SAR echo data of potential anomalous land parcels are preprocessed, and the image regions corresponding to the land parcels are cropped and adjusted to a uniform size.

[0107] Furthermore, for optical images, the RGB three-channel data are extracted and normalized, mapping pixel values ​​from [0,255] to the [0,1] interval. The normalization formula is as follows: ,in These are the original pixel values. These are the normalized values. For SAR echo data, the complex form of the echo signal is converted into an amplitude image. The calculation formula is: Where I is the in-phase component and Q is the quadrature component, and the amplitude image is... Logarithmic transformation of the amplitude value , To prevent infinitely large small constants from occurring in logarithmic operations. For example, The value is 1e-10.

[0108] A dual-branch convolutional neural network architecture is designed, which uses a dual-branch structure convolutional neural network to process optical images and SAR data respectively. The dual-branch structure convolutional neural network includes an optical image branch and a SAR data branch.

[0109] The optical image branch of the dual-branch convolutional neural network comprises multiple convolutional modules, each consisting of a convolutional layer, a normalization layer, and an activation layer. For example, the first convolutional module uses 64 7×7 kernels with a stride of 2 to extract low-level features. Subsequent convolutional modules employ a residual connection structure, containing two 3×3 convolutional layers with kernel numbers of 128, 256, and 512 respectively. The formula for calculating residual connections is... ,in For the input feature map, For convolution operations, For convolution weights, This is used to output feature maps. The residual structure employed effectively alleviates the gradient vanishing problem in deep networks and improves feature learning capabilities.

[0110] Multi-scale feature extraction layers are inserted between convolutional modules, employing dilated convolution to capture features from different receptive fields. Three parallel dilated convolution branches are configured, for example, with varying dilation rates. , , They are respectively , , The convolution kernel size is 3×3. The output feature maps of the three branches are concatenated to form a multi-scale feature representation. ,in Indicates the void ratio The convolutional output is used. Feature fusion is performed through 1×1 convolution, which reduces the channel dimension while maintaining the expressive power of the features.

[0111] The SAR data branch of the dual-branch convolutional neural network adopts a network structure similar to that of the optical image branch, but is optimized for the characteristics of SAR images. A polarization feature encoding module is added before the first convolutional layer to encode SAR data with different polarization modes into feature tensors. For fully polarimetric SAR data, a covariance matrix is ​​constructed. ,in For the first The scattering vector of each pixel. for Conjugate transpose, superscript This indicates a conjugate transpose. This represents the total number of pixels. Eigenvalue decomposition is performed on the covariance matrix. Where U is the eigenvector matrix, It is a matrix The conjugate transpose of . This is the eigenvalue diagonal matrix. Using eigenvalues ​​and eigenvectors as input channels enhances the network's ability to utilize polarization information.

[0112] A feature fusion mechanism is constructed to adaptively fuse features from the optical image branch and the SAR data branch. An attention mechanism is used to calculate the weights of the features from the two branches:

[0113] Optical feature weights :

[0114] ;

[0115] SAR feature weights :

[0116] ;

[0117] in, and These are the feature vectors of the two branches, and These are learnable weight parameters.

[0118] Fusion features The calculation formula is The constructed adaptive fusion mechanism can dynamically adjust the contribution of each data source according to the characteristics of different plots.

[0119] A multi-task learning structure is designed in the classification layer of the network to simultaneously identify crop planting status, growth stage, and illegal construction behavior. Three parallel classification heads are set up, each containing two fully connected layers and a softmax output layer.

[0120] Crop planting status classification head output Probability distribution of crop types ,in , For the first The logit value of crop-like crops, For the first The logit value of crop-like crops.

[0121] Growth stage classification head output The probability of each growth stage , The first classification of the head for the growth stage The predicted probabilities for each category. The illegal construction detection head outputs binary classification results. This represents the probability of a normal construction being an illegal construction. and The two categories of probabilities for the illegal construction detection head represent the predicted probabilities of normal or illegal construction, respectively. For example, Set as 10 major crops, It is set to 5 growth stages.

[0122] The land parcel status recognition model training employs a multi-task joint loss function, with a total loss of [missing information]. ,in , , The cross-entropy loss for the three tasks are respectively. For L2 regularization terms, , , , The loss weights are used. A stochastic gradient descent optimizer with momentum is used to train the network.

[0123] For each potential anomalous block, its optical imagery and SAR data are input to obtain prediction results from three tasks. A comprehensive confidence score is then calculated. , For the first The information entropy of the prediction results for each potential anomalous land parcel reflects the uncertainty of the prediction. For the first One potentially abnormal plot of land. When the probability of illegal construction... At that time, it was directly identified as an abnormal land parcel, among which This is the threshold for determining illegal construction. For crop-related anomalies, when the predicted crop type does not match historical records or the growth stage deviates abnormally from the normal range, and the confidence score... The land parcel was identified as an anomaly, among which This is the confidence threshold.

[0124] This step constructs a land parcel status recognition model based on a dual-branch convolutional neural network, achieving accurate identification and fine-grained classification of potentially anomalous land parcels. The model fully utilizes the spectral texture information of optical imagery and the backscattering characteristics of SAR data, capturing land cover features at different levels through multi-scale feature extraction. An adaptive feature fusion mechanism effectively integrates the complementary advantages of multi-source information. The constructed multi-task learning framework can simultaneously complete multiple recognition tasks, improving the model's efficiency and practicality. The land parcel status recognition model constructed in this step can accurately distinguish between normal changes and genuine anomalies, significantly reducing the false alarm rate and providing reliable technical support for accurate monitoring and timely early warning of land resources.

[0125] Step 6: Perform spatial clustering analysis on the identified anomalous plots to form ground anomaly areas. Determine the precise geographical boundaries of the ground anomaly areas based on the ground 3D digital model. Generate multi-level early warning information based on the classification results and confidence scores of the identified anomalous plots.

[0126] Spatial clustering analysis was performed on the identified anomalous plots in step 5. The density-based DBSCAN algorithm was used to merge spatially adjacent anomalous plots of the same category into contiguous anomalous regions. A spatial neighborhood radius was set. and minimum number of plots Two parameters are used to calculate the anomaly of any given plot of land. Set of plots within the neighborhood ,in Indicates land parcel and Euclidean distance between the centroids Let represent the radius distance of the spatial neighborhood. When At that time, A core plot is marked, and all plots within its neighborhood that share the same anomaly category are grouped into the same cluster. By recursively expanding the neighborhood of the core plot, spatially connected anomaly regions are formed. For example, for detecting illegal construction on farmland, the following settings are configured: It is 50 meters. The area is divided into three plots to ensure that the merged region has a meaningful spatial scale. The density-based clustering method used in this step can identify anomalous regions of arbitrary shapes, avoiding the limitations imposed by traditional methods on region shape.

[0127] The geographic boundaries of anomalous areas are accurately identified using a 3D digital ground model. For each clustered anomalous area, the set of boundary points for all contained land parcels is extracted. The outer contour of the anomalous area is calculated using the convex hull algorithm, and then the boundary shape is optimized using the Alpha-shape algorithm to generate boundary lines that more closely resemble the actual contours of ground features. The 2D boundary lines are projected onto the 3D digital model, and the elevation information of the boundary points is obtained through interpolation to construct a 3D boundary model of the anomalous area.

[0128] Calculate the geometric properties of the abnormal region, including its area. ,perimeter and shape complexity index ,in These are the coordinates of the boundary points. The number of boundary points. Indicates the first [unclear] on the boundary of the abnormal region The index of each point. These geometric attributes are stored together with the geographic coordinates of the anomaly area to form a complete spatial information description of the anomaly area.

[0129] Based on the classification results and confidence scores of abnormal land parcels, a multi-level early warning mechanism is constructed, establishing a three-level early warning system: Level 1 warning corresponds to high-risk anomalies, Level 2 warning corresponds to medium-risk anomalies, and Level 3 warning corresponds to low-risk anomalies. The determination of the warning level comprehensively considers the severity of the anomaly type and the confidence level of identification, defining a risk score. ,in Score the severity of the abnormality type. The average confidence score is given. The normalized area of ​​the outlier region. , , These are the corresponding weighting coefficients. For example, illegal construction... It can be set to 0.9, for those not planted according to regulations. It can be set to 0.6, when A Level 1 alert is triggered. For each alert level, a structured alert report is generated, containing the location, type, area, confidence level, and recommended handling measures for the anomaly.

[0130] Multi-level early warning information is prioritized, with higher-level warnings pushed to regulatory authorities first. An early warning information data package is constructed, containing vector boundary data of the abnormal area, keyframe image data, anomaly attribute tables, and early warning report documents. Anomaly areas are overlaid on a base map using a geographic information system platform, with different colors indicating different warning levels; for example, red represents a Level 1 warning, orange a Level 2 warning, and yellow a Level 3 warning. A time-series change comparison chart is generated to show the evolution of the abnormal area over different monitoring periods. Simultaneously, an early warning feedback mechanism is established to record the handling results and actual verification status of each warning, for subsequent model optimization and threshold adjustment.

[0131] This step integrates discrete anomalous plots into meaningful anomalous areas through spatial clustering and boundary calibration, improving the practicality of monitoring results. The construction of the 3D boundary model fully utilizes the elevation information from LiDAR data, achieving a three-dimensional representation of the anomalous areas. The multi-level early warning mechanism established in this step provides differentiated management based on risk levels, ensuring the rational allocation of limited regulatory resources. The early warning process achieves a complete closed loop from data collection and intelligent analysis to decision support, providing scientific, efficient, and precise technical support for the dynamic monitoring of land resources, and significantly improving the informatization and intelligentization level of land resource management.

[0132] Example 2

[0133] Reference Figure 6 This is the second embodiment of the present application, which provides a dynamic monitoring and early warning system for land resources based on multi-source remote sensing data fusion.

[0134] The system includes: a data acquisition module, a feature extraction module, a feature change rate prediction module, a potential abnormal land parcel identification module, an abnormal land parcel identification and classification module, and a graded early warning module.

[0135] The data acquisition module periodically collects ground optical images, SAR echoes, and LiDAR point cloud data using UAVs. Based on the LiDAR point cloud data, it generates a ground 3D digital model and divides the land into plots. It constructs a spatiotemporal registration algorithm to map the ground optical images and SAR echoes acquired in the same period to the corresponding plots, thereby generating a ground 3D digital model.

[0136] The feature extraction module extracts texture feature parameters from optical images, backscattering coefficients and polarization features from SAR echoes, and elevation statistics and surface roughness from LiDAR point clouds for the divided land parcels, and generates multi-dimensional feature vectors for the land parcels through feature fusion algorithms.

[0137] The feature change rate prediction module constructs a land parcel feature evolution model based on a long short-term memory network, infers the current period's land parcel change pattern based on historical periodic multi-dimensional feature vectors of land parcels, and generates a land parcel feature change rate interval.

[0138] The potential abnormal land parcel identification module establishes a change detection algorithm based on time-series difference, calculates the land parcel change rate of the current period based on the multi-dimensional feature vector of the land parcel, uses the range of land parcel feature change rate as the judgment threshold range of potential abnormal land parcels, and marks potential abnormal land parcels according to the judgment threshold range of potential abnormal land parcels.

[0139] The abnormal land parcel identification and classification module constructs a land parcel status identification model based on a dual-branch convolutional neural network. It inputs the optical images and SAR echoes of potential abnormal land parcels into the land parcel status identification model to identify the planting status, growth stage of ground crops, and illegal construction behavior on the ground. It determines whether potential abnormal land parcels are confirmed abnormal land parcels and outputs the classification results and confidence scores of confirmed abnormal land parcels.

[0140] The hierarchical early warning module performs spatial clustering analysis on identified abnormal land parcels to form ground anomaly areas, marks the precise geographical boundaries of the ground anomaly areas based on the ground three-dimensional digital model, and generates multi-level early warning information based on the classification results and confidence scores of the identified abnormal land parcels.

[0141] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0142] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of this application without departing from the spirit and scope of protection of the claims. All of these variations are within the protection scope of this application.

Claims

1. A land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion, characterized in that, The method comprises the following steps: Periodically collecting optical images, SAR echoes and LiDAR point cloud data of the ground by a UAV, generating a three-dimensional digital model of the ground based on the LiDAR point cloud data and dividing the ground into plots, constructing a space-time registration algorithm to map the optical images and SAR echoes of the same period to the corresponding plots to generate a three-dimensional digital model of the ground; For the divided plots, extracting texture feature parameters of the optical images, extracting backscattering coefficients and polarization features of the SAR echoes, and extracting elevation statistics and surface roughness of the LiDAR point cloud, and generating a multi-dimensional feature vector of the plot through a feature fusion algorithm; Constructing a plot feature evolution model based on a long short-term memory network, deducing the plot change rule of the current period based on the multi-dimensional feature vector of the plot of the historical period, and generating a plot feature change rate interval; Establishing a change detection algorithm based on time series difference, calculating the plot change rate of the current period according to the multi-dimensional feature vector of the plot, taking the plot feature change rate interval as a judgment threshold range of the potential abnormal plot, and marking the potential abnormal plot according to the judgment threshold range of the potential abnormal plot; Constructing a plot state recognition model based on a double-branch convolutional neural network, inputting the optical images and SAR echoes of the potential abnormal plot into the plot state recognition model, recognizing the ground crop planting state, growth stage and ground illegal construction behavior, and determining whether the potential abnormal plot is a determined abnormal plot, and simultaneously outputting the classification result and confidence score of the determined abnormal plot; Performing spatial clustering analysis on the determined abnormal plot to form a ground abnormal area, marking the geographical boundary of the ground abnormal area according to the three-dimensional digital model of the ground, and generating multi-level early warning information according to the classification result and confidence score of the determined abnormal plot.

2. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 1, characterized in that, An optical camera, a synthetic aperture radar and a laser radar scanner are integrated on a UAV to collect optical images, SAR echoes and LiDAR point cloud data of a monitoring area; Based on the collected LiDAR point cloud data, after noise removal and point cloud thinning preprocessing, and separation of ground points and non-ground points, an irregular triangle network interpolation method is used to generate a three-dimensional digital model of the ground; A watershed algorithm based on slope and curvature is used to divide and encode the plots of the three-dimensional digital model of the ground by fusing the edge detection results of the optical images; A space-time registration algorithm is constructed to geometrically correct and register the optical images and SAR echoes and map them to the corresponding plots.

3. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 2, characterized in that, The optical images are converted into gray images, and the gray level co-occurrence matrix method is used to calculate and extract energy, contrast, correlation and entropy in a set distance and multiple directions, and the feature values in multiple directions are averaged to obtain feature parameters; The SAR echoes are radiometrically calibrated and speckle noise suppressed, the backscattering coefficients are calculated, and a scattering matrix is constructed for the SAR echo data, and a polarization decomposition method is used to extract components representing surface scattering, dihedral scattering and volume scattering characteristics, and polarization entropy and average scattering angle are calculated.

4. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 3, characterized in that, For the point cloud data in each plot, the elevation statistical parameters are calculated, including mean elevation, elevation standard deviation, maximum elevation and minimum elevation; The surface roughness is calculated by grid division of point cloud in the plot and local plane fitting method; After extracting the characteristic parameters of each data source, a feature fusion algorithm is used to normalize all extracted characteristic parameters, and the normalized optical texture features, SAR polarization features and LiDAR elevation features are spliced in a predetermined order to generate a plot multi-dimensional feature vector.

5. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 4, characterized in that, A plot feature evolution model is constructed based on a long short-term memory network; The plot multi-dimensional feature vectors of historical periods are organized into a time series, the feature change rate sequence between adjacent periods is calculated and standardized to serve as the input of the plot feature evolution model; The architecture of the long short-term memory network is designed, including an input layer, a fully connected layer and an output layer; the input layer includes an LSTM layer of forget gate, input gate and output gate, and the output layer is used to output the feature change rate prediction value of the next period and the uncertainty estimation for quantifying the prediction confidence.

6. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 5, characterized in that, Based on the feature change rate prediction value and the uncertainty estimation predicted by the plot feature evolution model, a change rate confidence interval is constructed for each dimension of the plot, forming a multi-dimensional feature change rate interval of the plot.

7. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 6, characterized in that, The plot multi-dimensional feature vectors of the current and historical monitoring periods are obtained, the feature difference vector is calculated, and the plot change rate of the current period is calculated; Based on the coefficient of variation of each feature dimension in the historical data, the weight is determined, and the comprehensive change rate of the plot is calculated based on the weighted Euclidean distance. The comprehensive relative change rate of the plot is calculated, and the positive change rate and the negative change rate representing the increase and decrease of the feature value are calculated.

8. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 7, characterized in that, Based on the plot feature change rate interval, a dynamic decision threshold range is set, when the calculated plot feature change rate exceeds the dynamic decision threshold range, the abnormality degree of the plot feature deviating from the predicted interval is calculated, and is summarized as the comprehensive change rate abnormality degree of the plot; A multi-criteria fusion abnormality judgment mechanism is constructed, when the plot meets any abnormality judgment mechanism condition, the plot is marked as a potential abnormal plot: The abnormality judgment mechanism includes: there are a preset number of feature dimensions whose change rate exceeds the predicted interval, the comprehensive change rate abnormality degree of the plot is greater than a preset threshold, and the comprehensive change rate of the plot exceeds the comprehensive change rate interval derived from the boundaries of each dimension of the predicted interval.

9. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 1, characterized in that, A plot state recognition model based on a double-branch convolutional neural network is constructed to identify whether a potential abnormal plot is a confirmed abnormal plot; The double-branch convolutional neural network processes optical images and SAR data respectively, the optical image branch adopts a residual connection structure, and a multi-scale feature extraction layer composed of multiple parallel hole convolutions is inserted to capture features of different receptive fields; The SAR data branch adds polarization feature encoding at the input end of the optical image branch network structure.

10. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 9, characterized in that, A feature fusion mechanism based on an attention mechanism is constructed to adaptively calculate the weights of the optical and SAR two branch features and perform weighted fusion; A multi-task learning structure is designed at the end of the double-branch convolutional neural network, and parallel classification heads are set to simultaneously recognize crop planting states, growth stages and ground illegal construction behaviors; According to the prediction result of the land parcel state recognition model on the recognition task, a comprehensive confidence score is calculated, and combined with a preset illegal building judgment threshold and a crop anomaly confidence threshold, whether the potential abnormal land parcel is a determined abnormal land parcel is determined, and a classification result is output.

11. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 10, characterized in that, Spatial clustering analysis is performed on the determined abnormal land parcel, and a density-based clustering algorithm is used to merge spatially adjacent land parcels with the same abnormal category into a continuous ground anomaly area. The ground three-dimensional digital model is used to demarcate the geographical boundary of the abnormal area by the convex hull algorithm and the Alpha-shape algorithm, and the geometric properties of the ground anomaly area are calculated.

12. The land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to claim 11, characterized in that, A multi-level early warning mechanism is constructed, the risk score is calculated according to the severity of the abnormal type, the recognition confidence and the area of the region to determine the warning level, and a structured warning report containing the abnormal position and type information is generated.

13. A land resource dynamic monitoring and early warning system based on multi-source remote sensing data fusion, which is used to realize the land resource dynamic monitoring and early warning method based on multi-source remote sensing data fusion according to any one of claims 1-12, characterized in that, It includes: A data acquisition module, a feature extraction module, a feature change rate prediction module, a potential abnormal land parcel recognition module, an abnormal land parcel recognition classification module, and a hierarchical warning module; The data acquisition module periodically collects optical images, SAR echoes and LiDAR point cloud data on the ground by a UAV, generates a ground three-dimensional digital model based on LiDAR point cloud data and performs land parcel division, and constructs a space-time registration algorithm to map the ground optical images and SAR echoes obtained in the same period to the corresponding land parcel, and generate a ground three-dimensional digital model; The feature extraction module extracts the texture feature parameters of the optical image, extracts the backscattering coefficient and polarization feature of the SAR echo, and extracts the elevation statistics and surface roughness of the LiDAR point cloud for the divided land parcel, and generates a land parcel multi-dimensional feature vector through a feature fusion algorithm; The feature change rate prediction module constructs a land parcel feature evolution model based on a long short-term memory network, deduces the current period land parcel change rule based on the historical period land parcel multi-dimensional feature vector, and generates a land parcel feature change rate interval; The potential abnormal land parcel recognition module establishes a change detection algorithm based on time series difference, calculates the current period land parcel change rate according to the land parcel multi-dimensional feature vector, takes the land parcel feature change rate interval as the determination threshold range of the potential abnormal land parcel, and marks the potential abnormal land parcel according to the determination threshold range of the potential abnormal land parcel; The abnormal land parcel recognition classification module constructs a land parcel state recognition model based on a double-branch convolutional neural network, inputs the optical image and SAR echo of the potential abnormal land parcel into the land parcel state recognition model, recognizes the ground crop planting state, growth stage and ground illegal building behavior, determines whether the potential abnormal land parcel is a determined abnormal land parcel, and outputs the classification result and confidence score of the determined abnormal land parcel; The hierarchical warning module performs spatial clustering analysis on the determined abnormal land parcel to form a ground anomaly area, demarcates the geographical boundary of the ground anomaly area according to the ground three-dimensional digital model, and generates multi-level warning information according to the classification result and confidence score of the determined abnormal land parcel.

Images