A dynamic remote sensing monitoring method and system based on land use

Abstract

The present invention belongs to the field of electronic digital data processing technology, and discloses a dynamic remote sensing monitoring method and system based on land use. The method includes: obtaining a multi-source heterogeneous time series data set, generating standardized spatiotemporal data through preprocessing; extracting spatiotemporal correlation features based on a dynamic graph convolutional network, and generating dynamic pattern data through multi-dimensional feature fusion; performing classification decision-making and state transition analysis on the dynamic pattern data, identifying state change areas and generating dynamic analysis reports; constructing an interactive visualization engine to map spatiotemporal heat maps and trend maps; combining incremental learning optimization models to output long-term evolution prediction results. The present invention solves the problems of low efficiency and insufficient noise suppression capability of spatiotemporal correlation modeling of multi-source data through dynamic graph convolutional networks and adaptive cross-modal alignment technology, significantly improves the accuracy of dynamic pattern recognition, supports real-time interactive analysis and long-term prediction, and is suitable for accurate decision-making in scenarios such as industrial monitoring and traffic planning.

Description

Technical Field

[0001] The present application relates to the technical field of electronic digital data processing, and in particular to a dynamic remote sensing monitoring method and system based on land use. Background Art

[0002] With the rapid development of the Internet of Things (IoT), sensor networks, and distributed computing technologies, the scale and complexity of multi-source, heterogeneous data are growing exponentially. Real-time collection, fusion, and analysis of dynamic spatiotemporal data have become core requirements in areas such as industrial equipment monitoring, traffic flow analysis, and healthcare management. However, existing technologies face significant bottlenecks in integrating multi-source, heterogeneous data, recognizing dynamic patterns accurately, and scalability, making it difficult to meet the demands of timely, highly concurrent data analysis.

[0003] 1. Data heterogeneity and integration efficiency issues. Existing technologies typically rely on a single data source or simple rules to achieve multimodal data alignment (such as time series matching of sensor data and log data), and lack the ability to dynamically model cross-modal correlations. For example:

[0004] Data noise suppression: Traditional filtering algorithms (such as mean filtering) require preset fixed parameters and cannot be dynamically adjusted according to data distribution, resulting in a significant decrease in signal-to-noise ratio in high-noise scenarios.

[0005] Cross-modal alignment: Data fusion methods based on static coordinate systems (such as rigid registration) are difficult to adapt to the spatiotemporal drift of multi-source data (such as differences in sensor sampling frequency and inconsistent spatial resolution), resulting in the accumulation of alignment errors.

[0006] 2. Dynamic pattern recognition lacks real-time and generalizability. Existing dynamic analysis models (such as the traditional time series forecasting algorithm ARIMA) are typically based on fixed time windows or static feature extraction, and suffer from the following drawbacks:

[0007] Limitations of spatiotemporal correlation modeling: Discrete time series analysis methods (such as sliding window statistics) cannot capture long-range spatiotemporal dependencies, resulting in a high rate of missed detection of complex dynamic patterns (such as the cross-sensor propagation of equipment failures).

[0008] Weak model generalization ability: Models trained for specific scenarios (such as a single industrial device) are difficult to adapt to other fields (such as medical monitoring). Feature engineering and model architecture need to be repeatedly designed, resulting in high development costs. Summary of the Invention

[0009] This application provides a dynamic remote sensing monitoring method and system based on land use to solve the problem of how to achieve cross-modal dynamic spatiotemporal correlation modeling and evolution law analysis based on multi-source heterogeneous time series data (sensor data, spatial coordinate data and time series data), and improve the real-time decision-making accuracy and system scalability in complex scenarios.

[0010] To solve the above problems, this application provides a dynamic remote sensing monitoring method based on land use, including:

[0011] S100: Acquire a multi-source heterogeneous time series dataset, preprocess the dataset, and generate standardized spatiotemporal data;

[0012] S200: extracting spatiotemporal correlation features from the standardized spatiotemporal data based on a dynamic graph convolutional network, performing multi-dimensional feature fusion using a dynamic pattern recognition model, and generating dynamic pattern data of the target object;

[0013] S300: Performing classification decisions on the dynamic pattern data to generate multi-period state analysis results of the target object, including:

[0014] S310: Performing probability distribution modeling on the dynamic pattern data using a classification model to obtain classification results of each spatial position in a continuous time period;

[0015] The probability distribution of the classification result is defined by the following formula: ;

[0016] in, is the spatiotemporal pattern data, It is the inner product after feature mapping, which represents the weight of spatiotemporal features in the classification decision function. is the predicted probability of land use category, is the number of categories of the classification model, is the bias term;

[0017] S320: Perform state transition analysis on the classification results of the continuous time period to identify the state change area of ​​the target object, and calculate the change rate as follows: ;

[0018] in, Indicates time, is the land use type of the same location in different time periods, Indicates the dynamic mode analysis rate of the location;

[0019] And perform classification statistics on dynamic pattern analysis: ;

[0020] in, Indicates the state of mapping different types of changes to spatial positions, is the indicator function;

[0021] S330: Perform classification decisions on the dynamic pattern data to generate multi-period state analysis results and a dynamic analysis report for the target object;

[0022] S400: Build an interactive visualization engine to map the dynamic analysis report into spatiotemporal data visualization and trend graphs;

[0023] S500: Based on incremental learning of historical data and real-time data, it optimizes the dynamic pattern recognition model and generates long-term evolution prediction results of the target object.

[0024] Furthermore, preprocessing of multi-source heterogeneous time series datasets includes:

[0025] S110: Preprocessing of multi-source heterogeneous time series datasets, including denoising, geometric correction, and radiometric correction;

[0026] S120: performing spatial resolution matching and integration on the preprocessed data to obtain integrated data;

[0027] S130: Perform radiation correction and denoising on the integrated data to construct standardized spatiotemporal data in a unified spatiotemporal coordinate system.

[0028] Furthermore, the spatiotemporal correlation features extracted based on the dynamic graph convolutional network include:

[0029] S210: Construct a spatiotemporal graph structure where nodes represent spatial locations, edges represent spatiotemporal correlations, and the adjacency matrix is ​​updated via a dynamic attention mechanism.

[0030] S220: Inputting the spatiotemporal feature vector into the spatiotemporal feature extraction model for inference to obtain a first classification result output by the model;

[0031] S230: using the first classification result, obtaining change pattern data by comparing multiple time periods;

[0032] S240: Constructing spatiotemporal pattern data for dynamic pattern analysis using the change pattern data.

[0033] Furthermore, mapping the dynamic analysis report into spatiotemporal data visualization and trend graphs includes:

[0034] S410: Formatting the analysis results and storing them as a visual data file; the visual data file includes: the spatial location of the change area, classification information of the change type, and the time series of the change;

[0035] S420: Generate a visualization display interface based on the stored visualization data file using the geographic information system platform. The visualization display interface includes: visualization of changing spatiotemporal data and a trend chart.

[0036] Furthermore, generating the long-term evolution prediction results of the target object includes:

[0037] S510: Statistical calculations are performed based on satellite images and multi-source heterogeneous data acquisition terminal data to obtain land data characteristics, including trends and derivative characteristics of land use types over time;

[0038] S520: performing feature fusion on land data features to obtain comprehensive feature data;

[0039] S530: Predicting changes in land use types at future moments using a land use type prediction model based on the comprehensive feature data to obtain a prediction result;

[0040] S540: Generate a comprehensive report on dynamic pattern analysis based on forecast results and socio-economic environment data.

[0041] A dynamic remote sensing monitoring system based on land use, applied to any of the above methods, comprising:

[0042] The data processing module obtains the data from multi-source heterogeneous data acquisition terminals, pre-processes and integrates the data from multi-source heterogeneous data acquisition terminals to obtain image data;

[0043] The land change identification module extracts spatiotemporal features from image data, uses a dynamic pattern analysis model to detect dynamic changes, and obtains spatiotemporal pattern data for dynamic pattern analysis;

[0044] The type change analysis module uses spatiotemporal pattern data to analyze changes in different time periods and obtain analysis results of type changes;

[0045] The result display module displays the classification results through a visual interface;

[0046] The report generation module combines ground data with multi-source heterogeneous data for comprehensive analysis to generate a forecast report for dynamic pattern analysis.

[0047] The beneficial effects are as follows:

[0048] (1) Efficient alignment and noise suppression of multi-source data. Through cross-modal tensor decomposition and adaptive filtering algorithms, the spatiotemporal drift and noise interference problems of traditional methods when integrating multi-source heterogeneous data (such as sensor data and spatial coordinate data) are solved, the data alignment error is reduced, and the input quality of subsequent analysis is significantly improved.

[0049] (2) Improved accuracy in modeling dynamic spatiotemporal correlations. Based on a dynamic graph convolutional network and a spatiotemporal attention mechanism, we break through the local limitations of traditional sliding window statistics and achieve modeling of long-range spatiotemporal dependencies. This improves the accuracy of recognizing complex dynamic patterns (such as the propagation path of industrial equipment faults).

[0050] (3) Optimization of state change detection efficiency. Probabilistic distribution modeling and state transition probability calculation are performed on dynamic pattern data. Compared with the traditional threshold method, the false alarm rate of the change area is reduced and the detection response time is shortened to milliseconds.

[0051] (4) Enhanced real-time interaction and long-term prediction capabilities. Combining incremental learning optimization with an interactive visualization engine, it supports real-time rendering of dynamic analysis reports and online model updates, reduces the monthly error rate of long-term evolution predictions, and allows users to customize filtering rules to enhance decision-making flexibility.

[0052] Of course, any product implementing the present application does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0054] Figure 1 A flow chart of a dynamic remote sensing monitoring method based on land use provided in an embodiment of the present application;

[0055] Figure 2 This is a structural block diagram of a dynamic remote sensing monitoring system based on land use provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0057] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0058] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0059] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0060] Currently, some technologies have not yet been able to provide a comprehensive, real-time, and accurate land use dynamic monitoring system, making it difficult to provide effective decision-making support in land resource management and policy making.

[0061] In view of this, this application provides a new idea. Figure 1 A flow chart of a dynamic remote sensing monitoring method based on land use provided in an embodiment of the present application is as follows: Figure 1 As shown in , the method may include the following steps:

[0062] S100: Acquire a multi-source heterogeneous time series dataset, preprocess the dataset, and generate standardized spatiotemporal data;

[0063] S200: extracting spatiotemporal correlation features from the standardized spatiotemporal data based on a dynamic graph convolutional network, performing multi-dimensional feature fusion using a dynamic pattern recognition model, and generating dynamic pattern data of the target object;

[0064] S300: Performing classification decisions on the dynamic pattern data to generate multi-period state analysis results of the target object, including:

[0065] S400: Build an interactive visualization engine to map the dynamic analysis report into spatiotemporal data visualization and trend graphs;

[0066] S500: Based on incremental learning of historical data and real-time data, the dynamic pattern recognition model is optimized to generate a long-term evolution prediction result of the target object.

[0067] As can be seen from the above process, this application has significantly improved the data quality and consistency by acquiring satellite images and multi-source heterogeneous data acquisition terminal data and preprocessing and integrating them, laying the foundation for subsequent analysis. The use of spatiotemporal feature extraction and deep learning models for dynamic land use change detection can accurately capture complex dynamic pattern analysis trends, overcome the limitations of traditional methods in dynamic monitoring, and achieve high-precision dynamic monitoring. Through refined classification and analysis of dynamic pattern analysis in different time periods, the generated results can not only accurately reveal the laws of change, but also provide data support for resource management and policy making. Further combined with visualization technology, dynamic pattern analysis can be intuitively displayed through visualization of changing spatiotemporal data and trend charts, allowing users to quickly understand the changing areas and their trends, providing convenience for scientific decision-making.

[0068] The following describes in detail each step of the above process and the effects that can be further produced in conjunction with the embodiments. It should be noted that the limitations such as "first" and "second" involved in this disclosure do not have limitations in terms of size, order, and quantity, but are only used to distinguish them in name.

[0069] First, the above step S100, namely "obtaining satellite images and multi-source heterogeneous data acquisition terminal data, preprocessing and integrating the satellite images and the multi-source heterogeneous data acquisition terminal data to obtain remote sensing image data" is described in detail in conjunction with the embodiment.

[0070] Satellite imagery data typically comes from a variety of remote sensing satellite platforms, such as the Landsat and Sentinel series, or high-resolution commercial satellites (e.g., WorldView and QuickBird). For example, you can access public remote sensing data platforms through international or regional remote sensing data centers, select the target region, time range, and band type to download imagery data. Alternatively, you can request imagery acquisition for specific regions and monitoring needs through the satellite mission scheduling system to obtain real-time imagery with high temporal and spatial resolution. You can also purchase high-resolution or specific-band imagery data from commercial remote sensing data providers to meet more sophisticated monitoring needs.

[0071] Multi-source heterogeneous data acquisition terminals can collect long-term environmental data (such as temperature, humidity, and soil moisture) through equipment such as meteorological stations and soil monitoring stations. This data is typically automatically aggregated to a data center through the Internet of Things (IoT) system. Alternatively, multiple sensor nodes (such as spectral sensors and temperature and humidity sensors) can be deployed in a target area to monitor local environmental information in real time and transmit data via wireless communication technologies. Alternatively, drones or vehicles equipped with sensors can be used to collect high-precision ground data in specific areas, compensating for the limited spatial coverage of fixed sensor networks.

[0072] Satellite imagery data has wide coverage and rich spectral information, and can provide regional land use information. However, it may be affected by factors such as atmospheric conditions and sensor performance, resulting in noise or geometric distortion. Multi-source heterogeneous data acquisition terminal data contains local environmental factors and has limited spatial coverage, but it can provide complementary information to satellite imagery. By preprocessing these two types of data, including denoising, geometric correction, and radiometric correction, interfering factors are eliminated and the authenticity and consistency of the data are improved. Subsequently, through spatial resolution matching and integration operations, the multi-source heterogeneous data acquisition terminal data and satellite imagery data are combined into unified remote sensing imagery data, providing accurate and comprehensive basic data for dynamic monitoring and refined analysis of dynamic pattern analysis.

[0073] As an implementable manner, step S100 can be implemented by steps S110-S130:

[0074] S110: Preprocessing the multi-source heterogeneous time series dataset, including: denoising, geometric correction and radiation correction.

[0075] Satellite imagery data includes images at different time points and spatial resolutions, providing spectral characteristics of the target area (e.g., red, green, blue, and near-infrared bands). The spatial resolution of these satellite imagery data is typically 10 to 30 meters, with a monthly or quarterly frequency. Preprocessing of satellite imagery data includes denoising, geometric correction, and radiometric correction.

[0076] Denoising can be achieved through spatial filtering algorithms (such as Gaussian filters or median filters). Gaussian filters use the following formula: (1)

[0077] in, The original satellite image at coordinates The pixel value at is the pixel value of the satellite image after denoising, is the Gaussian kernel function, defined as: (2)

[0078] This formula reduces the impact of noise by weighted averaging the values ​​of surrounding pixels. is the filter window size, is the standard deviation, which controls the strength of denoising.

[0079] For geometric correction, in order to correct the remote sensing image data to the geographic coordinate system, ground control points (GCPs) are used to calibrate the geometric distortion of the satellite image. Assume that the position of any pixel in the satellite image is , and its corresponding geographic coordinates are , solve the spatial transformation model by the least squares method and match the image coordinate system with the geographic coordinate system. The transformation matrix of geometric correction It can be expressed by the following formula: (3)

[0080] in, is the transformation matrix calculated from the ground control points.

[0081] For radiation correction, the image can be corrected using an atmospheric correction model (such as the 6S model) or an inversion method based on ground-based measured data. Assuming that the original radiation value of the image is , which is converted into surface reflectance through the atmospheric correction formula (4)

[0082] in, The atmospheric scattering effect is is the atmospheric transmittance, and its value depends on atmospheric conditions, sensor height, etc.

[0083] Through the above steps, high-quality satellite image data that has undergone denoising, geometric correction, and radiometric correction is obtained.

[0084] S120: Perform spatial resolution matching and integration on the pre-processed data to obtain integrated data.

[0085] Through a multi-source heterogeneous data collection terminal network or collection platform, ground monitoring data is obtained over a certain period of time. This data usually includes environmental factors related to dynamic pattern analysis, such as soil moisture, temperature, and vegetation cover.

[0086] Remote sensing imagery is obtained by fusing data from multi-source heterogeneous data collection terminals with satellite imagery. To ensure spatial consistency, the spatial resolution of the multi-source heterogeneous data collection terminal data and the satellite imagery must be matched. This can be achieved through spatial interpolation or by using ground sampling points that match the imagery resolution. Common interpolation methods include Kriging or Inverse Distance Weighted (IDW).

[0087] (5)

[0088] in, To integrate data, is the data integration function, and They are satellite image data and multi-source heterogeneous data collection terminal data.

[0089] During the data integration process, there may be some inconsistencies or missing values, which need to be filled through data interpolation or other methods (such as multiple interpolation).

[0090] Through this step, the integrated data It will include data from satellite images and multi-source heterogeneous data collection terminals, and be prepared for subsequent radiometric correction and denoising processing.

[0091] S130: Performing radiation correction and denoising on the integrated data to construct standardized spatiotemporal data in a unified spatiotemporal coordinate system.

[0092] After integration of data Perform radiation correction. Radiation correction adjusts the radiation value of the image data by considering factors such as atmospheric influence, sensor characteristics, and imaging angle to obtain standardized data. It can express more realistic ground information.

[0093] The general process of radiation correction can be expressed as: (6)

[0094] in, is the data after radiation correction, is the radiation correction function, For the integrated data, are atmospheric parameters, is the sensor characteristic.

[0095] After radiation correction, the data may still contain noise, especially the noise introduced by factors such as sensor hardware limitations and external environment changes. In order to improve the data quality, the integrated data needs to be Perform denoising. This denoising operation can use some common image processing techniques, such as median filtering, mean filtering, and bilateral filtering. The purpose is to remove high-frequency noise in the data and retain meaningful signal information.

[0096] De-noising can be done using filtering methods, which can be expressed mathematically as follows: (7)

[0097] in, is the denoised data, is the denoising filter function.

[0098] After denoising and radiometric correction, the data quality needs to be assessed to ensure it meets the requirements of subsequent analysis. This can be done by comparing the data with existing ground-truth data or by using error metrics (such as root mean square error (RMSE) and correlation coefficient) to quantify the data's accuracy and reliability.

[0099] The data quality assessment formula is: (8)

[0100] in, is the data quality assessment value, is the number of data points, and The denoised data and ground truth data are The value of the position.

[0101] Through the above steps, we can obtain It is high-quality remote sensing data that has been radiometrically corrected and denoised, and is suitable for subsequent spatiotemporal pattern analysis and dynamic pattern analysis and detection.

[0102] The above step S200, namely "extracting spatiotemporal features from the remote sensing image data, detecting dynamic changes in land use using a dynamic pattern analysis model, and obtaining spatiotemporal pattern data for dynamic pattern analysis" is described in detail below in conjunction with an embodiment.

[0103] This step describes a dynamic change detection method based on remote sensing imagery data. By extracting spatiotemporal features and integrating them with a dynamic pattern analysis model, it accurately captures land use change patterns. Specifically, spatiotemporal feature extraction involves analyzing temporal trends and spatial distribution characteristics from multi-time series data from remote sensing images. First, using time series analysis methods (such as short-time Fourier transform or sliding window methods), temporal features are extracted from images at different time periods to identify periodic, sudden, or long-term trends for dynamic pattern analysis. Second, using spatial analysis techniques (such as gray-level co-occurrence matrix or texture feature extraction), the spatial distribution information of different land parcels in the imagery is captured, including boundary changes and regional differences in land cover types. The extracted spatiotemporal features are then input into a dynamic pattern analysis model for dynamic change detection. This model typically uses deep learning methods (such as convolutional neural networks (CNNs) or spatiotemporal convolutional networks (ST-CNNs)) to learn the spatiotemporal correlations of large amounts of remote sensing imagery data to predict and classify land use change types. For example, to detect the gradual loss of agricultural land to urban development within a city area, the model can identify land use type transitions at specific locations based on the intensity of change over time and generate spatiotemporal data visualizations and trend graphs.

[0104] As an implementable manner, step S200 can be implemented by steps S210-S230:

[0105] S210: Construct a spatiotemporal graph structure where nodes represent spatial locations, edges represent spatiotemporal correlations, and the adjacency matrix is ​​updated through a dynamic attention mechanism.

[0106] By extracting spatiotemporal features from remote sensing image data, we can identify the time series change patterns in the image. First, we select remote sensing image data at multiple time points (such as quarters or years) as the input dataset. The input dataset includes ,in For time In coordinates The corrected remote sensing image data.

[0107] Furthermore, the sliding window method is used to extract temporal features from continuous remote sensing image data. Assuming that the size of the sliding window is , where the starting time of the window is The end time point is , the spatiotemporal change sequence can be extracted from remote sensing images of multiple periods , this sequence can reflect the changing trend of the image in space and time.

[0108] For time series data , using time characteristic analysis methods such as short-time Fourier transform Or wavelet transform, extract the time domain and frequency domain features of the image. The formula is as follows: (9)

[0109] in, is the spatiotemporal spectrum extracted by short-time Fourier transform, is the frequency, is the imaginary unit, is the length of the time series. This spatiotemporal spectrum can effectively capture the changing characteristics of remote sensing images in the time and frequency domains, providing input data for subsequent deep learning modeling.

[0110] While extracting spatiotemporal features, spatial features also need to be extracted from remote sensing images. Specifically, traditional remote sensing image feature extraction methods (such as gray-level co-occurrence matrix, texture features, etc.) are used to extract the spatial information of the image. For example, the gray-level co-occurrence matrix of the image is calculated. , which is calculated as follows: (10)

[0111] in, is the gray-level co-occurrence matrix is the pixel grayscale value, is the pixel pitch, is the angle, is the Dirac function. The extracted spatial features can reflect the spatial distribution patterns of land use types, vegetation changes, etc.

[0112] Combining the temporal and spatial features, the spatiotemporal feature vector is obtained through feature fusion methods (such as weighted average, principal component analysis, etc.). Assume that the spatiotemporal feature vector is , the temporal and spatial features are fused in the following ways: (11)

[0113] in, and is the weight coefficient, which controls the fusion ratio of time domain features and spatial features. The final spatiotemporal feature vector Can be used as input data for deep learning models.

[0114] S220: Input the spatiotemporal feature vector into a spatiotemporal feature extraction model for inference, and obtain a first classification result output by the model.

[0115] Convolutional neural network (CNN) is selected as the spatiotemporal feature extraction model, and spatiotemporal features are extracted by constructing convolutional layer, pooling layer and fully connected layer. Assume that the spatiotemporal feature vector is The input of the model is the stacked data of multiple spatiotemporal features. The convolutional layer of the model extracts spatiotemporal features through the following convolution operation: (12)

[0116] in, is the convolution kernel, is the output of the convolution operation, which represents the local response to the feature in space and time.

[0117] After the convolution operation, the model will be processed by the pooling layer to reduce the dimensionality to reduce the computational complexity. Assume that the output of the pooling layer is , the final classification or regression is performed through the fully connected layer to obtain the prediction results of land use dynamic changes. Specifically, use The function predicts the category: (13)

[0118] in, is the number of land use categories, Predict the probability of a land-use class at a given location and time.

[0119] Based on the output of the spatiotemporal feature extraction model, the spatiotemporal pattern of dynamic pattern analysis is extracted. Assume that the first classification result of the model output is .

[0120] S230: Using the first classification result, obtaining change pattern data by comparing multiple time periods.

[0121] By comparing multiple time periods of the first classification results, the change pattern of land use is identified and the land use change pattern data is obtained. The extraction of spatiotemporal variation pattern data can be performed in the following ways:

[0122] (14)

[0123] in, Indicates time The amount of change in land use category. If it is positive, it means that land use has changed; if it is zero, it means that land use has not changed.

[0124] S240: Using the change pattern data to construct the spatiotemporal pattern data of the dynamic pattern analysis. The spatiotemporal pattern data will include: dynamic pattern analysis results at multiple time points and change trends of spatial positions.

[0125] Based on the above spatiotemporal change pattern data , constructing a spatiotemporal data set of dynamic land use changes. This spatiotemporal data will include dynamic pattern analysis at multiple time points, as well as spatial trends. This data will provide a basis for long-term land use monitoring and trend prediction.

[0126] At the same time, the extracted spatiotemporal pattern data Save in formatted data storage to ensure data visualization and subsequent analysis.

[0127] The above step S300, i.e., "analyzing the changes in land use in different time periods using spatiotemporal pattern data to obtain analysis results of changes in land use types," is described in detail below in conjunction with an embodiment.

[0128] This step analyzes the land-use classification information in the spatiotemporal pattern data and combines it with dynamic land-use changes over different time periods to generate a detailed analysis of changes in regional land-use types. The land-use classification results in the spatiotemporal pattern data are compared by time period to identify changes in land-use types at the same spatial location over different time periods. Specifically, the classification results in the spatiotemporal pattern data can be compared to identify areas where land-use changes have occurred and to identify the type of change (e.g., from forestland to agricultural land). Statistical analysis of the change areas can be performed to assess the intensity of change in different land-use types (e.g., the area and proportion of agricultural land that has decreased in a particular location) and their temporal trends (e.g., annual decreases or rapid shifts).

[0129] As an implementable manner, step S300 can be implemented by steps S310-S330:

[0130] S310: Probability distribution modeling is performed on the dynamic pattern data through a classification model to obtain classification results of each spatial position in a continuous time period.

[0131] The spatiotemporal pattern data are classified and processed to obtain the land use type data including a second classification result of each spatial position in different time periods; and the land use type data of each time period is obtained according to the second classification result.

[0132] Specifically, using spatiotemporal pattern data on land use dynamics ,Combining the spatiotemporal characteristics and historical land use data, a classification model is constructed.

[0133] First, obtain historical land use classification data , and use these data to train the classification model. Preferably, the classification model used is support vector machine , then the training goal of the model is to minimize the following objective function:

[0134] (15)

[0135] in, is the weight of the classifier, is the bias term, is the penalty coefficient, is the slack variable, is the number of samples. The classification model aims to obtain a decision boundary that can classify new spatiotemporal pattern data by learning historical land use data.

[0136] Spatiotemporal pattern data Input into the trained SVM model for classification. The model will output the probability distribution of different land use types based on the learning of historical data. , which is the second classification result. Specifically, for each position and time , the model calculates the predicted probability of the land use type at that location: (16)

[0137] in, is the inner product after feature mapping, represents the weight of spatiotemporal features in the classification decision function, is the predicted probability of land use category, is the number of categories for the classification model.

[0138] Generate land use type data for each time period based on the second classification results output by the classification model , this data contains the land use classification results of each spatial location in different time periods. This data will provide a basis for dynamic pattern analysis. Specifically, the classification results for each time period can be extracted separately according to the time label in the second classification result. The land use types output by the classification model are divided according to the time series to generate the land use type distribution corresponding to each time point. For example, for the classification data of a certain area, the classification results in 2020 show that area A is "agricultural land" and area B is "construction land"; the classification results in 2025 show that area A becomes "construction land" while area B remains unchanged.

[0139] S320: Performing state transition analysis on the classification results of the continuous time periods to identify the state change area of ​​the target object.

[0140] After obtaining the land use type data, we further analyze the changes in land use type. Specifically, we compare the land use types of the same location in different time periods. , identify the changing areas. For example, calculate the time and Dynamic pattern analysis between: (17)

[0141] if If it is not zero, it means that the land use at that location has changed. The type of change is determined by This calculation can identify the spatial distribution and change trends of dynamic pattern analysis.

[0142] Furthermore, dynamic pattern analysis is classified and counted. Assuming different types of change, such as agricultural land to urban land, forest cover change, etc., the number of each type of change is counted and mapped to spatial locations. The statistical formula is as follows:

[0143] (18)

[0144] in, is the indicator function, if Belongs to a specific change type , otherwise it is zero. This statistical result provides data support for the specific type of dynamic pattern analysis.

[0145] S330: Counting the spatiotemporal evolution patterns of the state change region, and generating a dynamic analysis report including change intensity, correlation, and trend prediction.

[0146] Based on the above change statistics, we can further analyze the trends and patterns of dynamic pattern analysis. We can use regression analysis or time series analysis to identify the patterns of dynamic pattern analysis. For example, we can analyze the trends of dynamic pattern analysis in a certain region by fitting a linear regression model:

[0147] (19)

[0148] in, is the predicted land use type, and are the intercept and slope of the regression model, is the time variable. , can determine the growth or decrease trend of dynamic pattern analysis, thus providing a basis for the prediction of future land use.

[0149] The above step S400, namely "displaying the land use classification result through a visual interface", is described in detail below with reference to an embodiment.

[0150] This step presents complex land use classification results and change analysis data to users in an intuitive and interactive format, improving the data's comprehensibility and practicality. By designing a user-friendly interface, this technology integrates classification results with a geographic information system (GIS) to graphically display the spatial distribution and temporal trends of land use. Specifically, the visualization interface can include the following features: Classification result display: Displays land use classification results for each time period on a geospatial map using color coding or a legend; Trend analysis: Displays classification results for different time periods using a timeline control, allowing users to dynamically view the dynamic pattern analysis process for a particular region; Hotspot area annotation: Utilizes spatiotemporal data visualization technology to highlight hotspot areas where dynamic pattern analysis is frequent, such as urban expansion areas or ecologically sensitive areas; Interactive data query: Users can click on an area on the map to view detailed classification information and change records for that area, among other features.

[0151] As an implementable manner, step S400 can be implemented by steps S410-S420:

[0152] S410: Formatting the analysis result and storing it as a visual data file; the visual data file includes: the spatial location of the change area, classification information of the change type, and the time series of the change.

[0153] The results of land use type change analysis The results of the trend analysis are formatted and stored as visual data files. The formatted data includes the spatial location of the change area, the classification information of the change type, and the time series of the change. and trends in dynamic pattern analysis Stored in a standard data format for subsequent visualization and spatial analysis.

[0154] S420: Generate a visualization display interface based on the stored visualization data file using a geographic information system platform, wherein the visualization display interface includes: visualization of changing spatiotemporal data and a trend chart.

[0155] Utilize a GIS (Geographic Information System) platform to visualize stored data. For example, color-code the spatial distribution of dynamic pattern analysis and plot trend charts to provide a direct overview of dynamic pattern analysis. Specifically, use spatiotemporal data visualization or layered display to identify change areas and the spatial locations of change types, generating intuitive visualizations of analysis results.

[0156] Based on these analysis results, a comprehensive report on the dynamic pattern analysis will be generated. This report will include the spatial distribution, types of change, and trends of the dynamic pattern analysis, aiming to support decision-making for land management and planning. The report will include statistical charts, change analysis results, and trend forecasts to help policymakers understand the dynamics of the dynamic pattern analysis.

[0157] Based on the results of land use type change analysis and trend analysis results , designing a user-friendly visualization interface. This interface will display land use classification information for different time periods, as well as the spatial distribution of dynamic pattern analysis. Using map overlay techniques, land use categories are coded and displayed using different colors or symbols, allowing users to intuitively understand the land use type and changes in each plot. This visualization interface supports multiple user needs. For example, policymakers can view global trends, while local managers can view detailed information on local changes.

[0158] In order to ensure efficient display of the visualization interface, the spatiotemporal data and dynamic mode analysis data Formatting is performed and converted into a format suitable for visualization tools to read. Specifically, the data is converted into a format that can be recognized by the Geographic Information System (GIS) so that it can be displayed on the visualization platform.

[0159] (20)

[0160] in, For data in visual format, It is a data conversion function that converts the classification results into a suitable display format.

[0161] Design different levels of user interaction within the visualization interface. Users at different levels can select different geographic regions and time periods for query and analysis based on their needs. For example, local managers can select changes in a specific city or village, while national policymakers can analyze trends across a wider range of dynamic patterns.

[0162] For visualization of changing spatiotemporal data, data can be analyzed based on dynamic patterns , generate visualization of changing spatiotemporal data to show the intensity of dynamic pattern analysis in different regions. The visualization of changing spatiotemporal data is displayed by aggregating the spatial distribution of dynamic pattern analysis and applying color scale mapping. First, based on time, Dynamic pattern analysis within , calculate the change intensity at each position: (twenty one)

[0163] in, Indicates location The intensity of the change within the time interval, is the indicator function, when When it is not zero, it means that the position is at time If dynamic pattern analysis occurs within the time frame, the value is 1; otherwise, it is 0. By aggregating these change intensities, we can obtain a visualization of the spatiotemporal data of changes.

[0164] The calculated change intensity data Input to the visualization platform, and display the spatiotemporal data visualization through color level mapping. Specifically, the intensity of the change Mapped to a specific color range, areas of higher intensity are represented in red, and areas of lower intensity are represented in green or blue. Spatiotemporal data visualization clearly displays hotspots in dynamic pattern analysis, helping users quickly identify areas with the most significant changes. Users can dynamically adjust the temporal and spatial scopes of spatiotemporal data visualization through the interactive features of the visualization interface. For example, users can select different time periods to view spatiotemporal data visualizations at different stages of dynamic pattern analysis, thereby discovering patterns and trends in dynamic pattern analysis.

[0165] For trend charts, trend analysis results can be This visualization is done as a trend chart, allowing users to select different regions for trend analysis. This chart shows the overall direction of the dynamic pattern analysis, for example, whether agricultural land in a particular region is gradually being converted to urban land. Trend charts typically use line or bar graphs to display the dynamic pattern analysis for each region over different time periods, helping users understand land use trends in a particular area.

[0166] Furthermore, users can interactively select different land use types and time periods to dynamically analyze dynamic patterns across different regions. For example, they can analyze agricultural land changes in a specific city or urban expansion trends over a specific time period for a more detailed analysis.

[0167] The above step S500 , namely “combining ground data with remote sensing data for comprehensive analysis to generate a prediction report of dynamic pattern analysis”, is described in detail below with reference to an embodiment.

[0168] Based on the forecast results, combined with socioeconomic and environmental data (such as population density and ecologically sensitive regional divisions), the spatial distribution, type shifts, and potential impacts of dynamic model analysis under different change scenarios are analyzed. For example, the impact of policy adjustments on the reduction of agricultural land or the expansion of urban land, as well as the environmental costs of different development strategies, are analyzed.

[0169] The forecast report is presented through visualization and statistical analysis, including time series trends for dynamic pattern analysis, spatial distribution maps, spatiotemporal data visualization of change intensity, and assessment recommendations for future scenarios. Critical areas (such as urban expansion hotspots) and high-risk areas (such as the potential destruction of ecological reserves) are also highlighted in the report.

[0170] As an implementable manner, step S500 can be implemented by steps S510-S520:

[0171] S510: Statistical calculations are performed based on satellite images and multi-source heterogeneous data acquisition terminal data to obtain land data characteristics, where the land data characteristics include: trends and derivative characteristics of land use types changing over time.

[0172] Assume that the spatial precision is , the original remote sensing data is adjusted by spatial resampling. Specifically, the resampled remote sensing image data can be expressed as: (twenty two)

[0173] Among them, F(t) is the original remote sensing image data, Represents the resampled remote sensing image data.

[0174] Next, we extract features from the raw remote sensing data. In terms of spatial feature extraction, we focus on the spatial distribution of land use types and their changes. Specifically, the change in land use types can be measured by calculating the difference between the current and previous images, expressed as: (twenty three)

[0175] in, Indicates at time Remote sensing data on is the total number of pixels in the image. In this way, the spatial aggregation and change trend of land use types can be quantified.

[0176] In terms of temporal feature extraction, we focus on the trend of land use types changing over time. By weighting the time series data of land use types, we can obtain the corresponding temporal features: (twenty four)

[0177] in, Indicates at time area land use type, is the weight coefficient of the region, is the number of land use types.

[0178] In addition, it is necessary to extract other derived features, such as ecological benefits, resource utilization efficiency, etc. Assuming these features are , and its calculation formula is: (25)

[0179] in, Represents the area of ​​each pixel, These characteristics can provide a more comprehensive understanding of the spatial distribution and temporal changes of land use.

[0180] S520: Perform feature fusion on the land data features to obtain comprehensive feature data.

[0181] These land data features are fused. The specific method is to use weighted average or principal component analysis (PCA) and other techniques to fuse different types of features. The fused comprehensive feature data is expressed as: (26)

[0182] in, It is Class features, is the corresponding weight coefficient, is the total number of features. Through this weighted fusion, the importance of each feature in the comprehensive analysis is fully reflected.

[0183] S530: Predicting changes in land use types at future moments using a land use type prediction model based on the comprehensive feature data to obtain a prediction result.

[0184] Based on the comprehensive feature data, the prediction model is used to predict the future dynamic pattern analysis. First, input the comprehensive feature data and historical dynamic pattern analysis data The latter includes information such as dynamic pattern analysis rate and change type. Based on these data, a prediction model is established to predict the changes in land use types in the future.

[0185] In terms of prediction model selection, the support vector machine regression (SVR) model can be used for prediction. The prediction process is expressed as: (27)

[0186] in, represents the predicted future land use type, It is a support vector regression model. The input of the model is historical land use data and extracted feature data. After training, it can make predictions about future land use changes.

[0187] The model parameters are optimized by cross-validation and other methods to find the best model parameters. The optimization process is as follows: (28)

[0188] in, is the actual dynamic pattern analysis data, is the predicted value, is a hyperparameter of the model.

[0189] Finally, the trained prediction model is used to predict future land use types. The formula is: (29)

[0190] in, is the predicted land use type.

[0191] S540: Generate a comprehensive report on dynamic pattern analysis based on the prediction results and socio-economic environment data.

[0192] Based on land use forecast results and socio-economic and environmental data, a multi-dimensional comprehensive analysis is conducted to ultimately provide decision makers with policy support and decision-making basis for dynamic model analysis.

[0193] This step can be implemented in a variety of ways, such as building a dynamic feedback system between dynamic pattern analysis and the socioeconomic environment. Using system dynamics models (such as Vensim or AnyLogic) to simulate the interactions between land use, the economy, and the environment, the long-term impacts of different policy scenarios can be analyzed. For example, by adjusting parameters (such as economic growth rate and policy intervention intensity), the effects of policies on dynamic pattern analysis can be evaluated.

[0194] As a preferred implementation method, the prediction results and the socio-economic environment data can be input into a comprehensive decision-making model, and a comprehensive report on dynamic pattern analysis can be generated based on the output results of the comprehensive decision-making model; wherein, the comprehensive decision-making model is obtained by pre-training, and the model parameters are optimized according to the optimization objectives during the model training process; the optimization objectives are constructed based on the impact of dynamic pattern analysis on socio-economy and the risks brought by land use types.

[0195] Specifically, first, input the land use prediction results , and relevant data from socio-economic and environmental sources, such as population changes , policy adjustments and ecological environment index .

[0196] In the process of building a comprehensive decision-making model, the multiple impacts of dynamic pattern analysis on social economy, environmental protection and other factors are considered. Assuming that a multi-objective optimization model is used, the dynamic pattern analysis prediction, social economic data and environmental data are comprehensively considered. The model formula is as follows: (30)

[0197] in, It is the impact of dynamic pattern analysis on social economy, is the risk assessment brought by land use type, is the risk weight, is the weight of each factor.

[0198] Finally, based on the comprehensive analysis results, a comprehensive report on the dynamic pattern analysis is generated. This report includes forecasts of future dynamic pattern analysis trends, the impact of policy adjustments on dynamic pattern analysis, and recommendations for ecological protection and resource management. The report can be presented through a visual interface, such as dynamic pattern analysis spatiotemporal data visualization and trend charts, providing policymakers with intuitive decision-making support.

[0199] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0200] Example 2: Figure 2 FIG. 1 shows a structural block diagram of a dynamic remote sensing monitoring system based on land use according to an embodiment of the present invention. Figure 2 As shown, the system may include:

[0201] The data processing module 201 acquires multi-source heterogeneous data acquisition terminal data, pre-processes and integrates the multi-source heterogeneous data acquisition terminal data to obtain image data;

[0202] The land change identification module 202 extracts spatiotemporal features from the image data, performs dynamic change detection using a dynamic pattern analysis model, and obtains spatiotemporal pattern data of the dynamic pattern analysis.

[0203] The type change analysis module 203 uses the spatiotemporal pattern data to analyze the changes in different time periods to obtain analysis results of type changes.

[0204] The result display module 204 displays the classification results through a visual interface.

[0205] The report generation module 205 performs a comprehensive analysis based on the ground data and multi-source heterogeneous data to generate a prediction report of the dynamic pattern analysis.

[0206] As an implementable manner, the data processing module 201 can be configured to: preprocess the data, the preprocessing including: denoising, geometric correction and radiation correction; perform spatial resolution matching and integration on the preprocessed data and the multi-source heterogeneous data acquisition terminal data to obtain integrated data; perform radiation correction and denoising on the integrated data to obtain remote sensing data.

[0207] As an operative approach, the land change identification module 202 may be configured to: identify time series change patterns from the image data using a sliding window method to obtain a spatiotemporal feature vector; input the spatiotemporal feature vector into a spatiotemporal feature extraction model for inference, obtaining a first classification result output by the model; utilize the first classification result to compare multiple time periods to obtain change pattern data; and utilize the change pattern data to construct spatiotemporal pattern data for the dynamic pattern analysis. The spatiotemporal pattern data will include dynamic pattern analysis results at multiple time points and spatial location change trends.

[0208] As an implementable method, the type change analysis module 203 can be configured to: classify the spatiotemporal pattern data to obtain the type data including a second classification result of each spatial position in different time periods; obtain the type data of each time period based on the second classification result; identify the change area by comparing the type data of each spatial position in different time periods; count the change trend of the change area, and generate the analysis result of the type change.

[0209] As an implementable method, the result display module 204 can be configured to: format the analysis results and store them as a visual data file; the visual data file includes: the spatial location of the change area, classification information of the change type, and the time series of the change; use the geographic information system platform to generate a visual display interface based on the stored visual data file, and the visual display interface includes: visualization of changing spatiotemporal data and trend charts.

[0210] As an implementable manner, the report generation module 205 can be configured to: perform statistical calculations based on satellite images and multi-source heterogeneous data acquisition terminal data to obtain land data characteristics, wherein the land data characteristics include: trends and derivative characteristics of land use types changing over time; perform feature fusion on the land data characteristics to obtain comprehensive feature data; based on the comprehensive feature data, use a land use type prediction model to predict changes in land use types at future moments to obtain prediction results; and generate a comprehensive report on dynamic pattern analysis based on the prediction results and socio-economic environment data.

[0211] As an implementable method, when generating a comprehensive report of dynamic pattern analysis based on the prediction results and socio-economic environmental data, the report generation module 205 can be configured as follows: inputting the prediction results and the socio-economic environmental data into a comprehensive decision-making model, and generating a comprehensive report of dynamic pattern analysis based on the output results of the comprehensive decision-making model; wherein, the comprehensive decision-making model is obtained by pre-training, and the model parameters are optimized according to the optimization objectives during the model training process; the optimization objectives are constructed based on the impact of dynamic pattern analysis on socio-economy and the risks brought by land use types.

[0212] The above is a detailed introduction to the technical solutions provided by this application. Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above embodiments is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the contents of this specification should not be understood as limiting this application.

Claims

1. A dynamic remote sensing monitoring method based on land use, characterized in that: The method comprises: S100: Acquire data from satellite images and multi-source heterogeneous data acquisition terminals, pre-process and integrate the data to obtain remote sensing image data; the satellite image data is from Landsat series, Sentinel series, or high-resolution commercial satellites; the multi-source heterogeneous data acquisition terminals collect temperature, humidity, and soil moisture environmental data from meteorological stations and soil monitoring stations; S200: Obtaining spatiotemporal feature vectors based on continuous remote sensing image data, inputting the spatiotemporal feature vectors into a convolutional neural network for inference, and obtaining the predicted probability of land use category at a given location and time point , and obtain the first classification result of land use category ; Using the first classification results, we can obtain land use change pattern data by comparing multiple time periods. : , in, Indicates the change in land use category at time t; if If it is positive, it means that land use has changed; if it is zero, it means that land use has not changed; S300: Analyze the land use change pattern data to generate analysis results of land use type changes, including: S310: Obtain historical land use classification data , and use these data to train the classification model; Input into the trained classification model for classification to obtain the updated land use category prediction probability ; Updated predicted probability of land use categories It is defined by the following formula: ; in, is the inner product after feature mapping, represents the weight of spatiotemporal features in the classification decision function, is the number of categories of the classification model, is the bias term; S320: Perform state transition analysis on the classification results of the continuous time period to identify the state change area of ​​the target object, and calculate the change rate as follows: ; in, Indicates time, is the land use type of the same location in different time periods, Indicates the dynamic mode analysis rate of the location; And perform classification statistics on dynamic pattern analysis: ; in, Indicates the state of mapping different types of changes to spatial locations; is the indicator function; if Belongs to a specific type of change , otherwise zero; S400: Formatting the analysis results and storing them as a visualization data file, wherein the visualization data file includes the spatial location of the change area, classification information of the change type, and the time series of the change; using a geographic information system platform, generating a visualization display interface based on the stored visualization data file, wherein the visualization display interface includes visualization of the change spatiotemporal data and a trend chart; S500: Statistical calculations are performed based on data from satellite images and multi-source heterogeneous data acquisition terminals to obtain land data characteristics; Performing feature fusion on the land data features to obtain comprehensive feature data; Predicting changes in land use types at future moments using a land use type prediction model based on the comprehensive feature data to obtain a prediction result; The prediction results and socio-economic environment data are input into a comprehensive decision-making model, and a comprehensive report on dynamic pattern analysis is generated based on the output results of the comprehensive decision-making model; the comprehensive decision-making model is obtained by pre-training, and the model parameters are optimized according to the optimization objectives during the model training process.

2. The dynamic remote sensing monitoring method based on land use according to claim 1, characterized in that: The process of extracting spatiotemporal correlation features includes: For time series data extracted from remote sensing images over multiple periods , using short-time Fourier transform Or wavelet transform to extract the time domain and frequency domain features of the image ; Gray Level Co-occurrence Matrix Extracting spatial features of images ; Combining time domain features and spatial features, the spatiotemporal feature vector is obtained through feature fusion method. : (11) in, and is the weight coefficient, which is used to control the fusion ratio of time domain frequency domain features and spatial features. is the space-time feature vector.

3. A dynamic remote sensing monitoring system based on land use, applied to the method according to any one of claims 1 to 2, comprising: A data processing module is used to obtain data from satellite images and multi-source heterogeneous data acquisition terminals, pre-process and integrate the data to obtain remote sensing image data; Land change identification module, used to obtain land use change pattern data; A type change analysis module is used to analyze the land use change pattern data and generate an analysis result of the land use type change; A result display module is used to format the analysis results and store them as a visual data file, which includes the spatial location of the change area, classification information of the change type, and the time series of the change; using the geographic information system platform, a visual display interface is generated based on the stored visual data file, and the visual display interface includes a visualization of the change spatiotemporal data and a trend chart; Report generation module, used to generate comprehensive reports of dynamic pattern analysis.

Images