Land utilization change information extraction method and system based on multi-source remote sensing image

Abstract

The invention relates to the technical field of remote sensing and geographic information, and discloses a land utilization change information extraction method and system based on a multi-source remote sensing image. The method comprises the following steps: acquiring optical or synthetic aperture radar remote sensing images of different time phases; geometric correction, radiation normalization, cloud removal repair and format adaptation are carried out to generate a standardized image data set; selecting an adaptation algorithm according to an image type, and fusing spectrum and texture features to generate an initial change mask; and through pattern spot boundary optimization and base map sleeving verification, an SHP format change pattern spot and a precision report conforming to the industrial standard are output. The system comprises a multi-source image input unit, a preprocessing unit, a change detection unit and a post-processing verification unit. According to the method, through full-process automatic processing, the precision, robustness and achievement normalization of land utilization change identification are improved.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing and geographic information technology, specifically relating to a method and system for extracting land use change information based on multi-source remote sensing images. Background Technology

[0002] With the rapid development of remote sensing technology and the widespread acquisition of multi-source satellite data, land use change monitoring has become a key supporting means for natural resource management, ecological environment assessment, and urban planning decisions. Remote sensing imagery, with its advantages of wide coverage, short update cycles, and rich information dimensions, is widely used in large-scale, long-term land cover dynamic analysis. However, extracting land use change information involves complex image processing procedures, requiring the fusion of data from multiple sensors such as optical and synthetic aperture radar (SAR), and precise comparison across different time phases. This places extremely high demands on data preprocessing, change detection algorithms, and post-processing of results.

[0003] Land use change information extraction methods based on multi-source remote sensing images aim to identify areas of changed land cover types by comparing remote sensing images from two or more time periods, and generate change patches with clear boundaries and attribute information. The core of this technology lies in how to achieve highly consistent and robust identification and refined representation of change areas under complex interference conditions such as atmospheric interference, illumination differences, geometric shifts, and cloud cover.

[0004] Existing technologies still have the following problems in practical applications: First, due to differences in sensor type, imaging angle, and time, multi-source remote sensing images have inconsistent radiometric and geometric characteristics. Conventional preprocessing methods are unable to simultaneously eliminate the coupled effects of atmospheric scattering, illumination changes, and geometric distortions, leading to systematic biases in subsequent change detection. Second, single change detection algorithms are difficult to adapt to the heterogeneous characteristics of optical and SAR images. While deep learning models have potential, they often produce a large number of false detections or missed detections in complex terrain scenes due to insufficient training samples or limited generalization ability. Third, initial change masks generally have jagged edges, holes, and isolated noise pixels. The lack of effective post-processing mechanisms and linkage verification with authoritative base map data results in final map features that do not conform to industry mapping standards and are difficult to use directly in operational GIS platforms.

[0005] The aforementioned problems are particularly prominent in large-scale, high-frequency land change surveys, severely restricting the accuracy, efficiency, and practicality of remote sensing change detection results. There is an urgent need for a full-process solution that integrates standardized preprocessing, adaptive detection, and intelligent post-verification. Summary of the Invention

[0006] This invention provides a method and system for extracting land use change information based on multi-source remote sensing images. By constructing an end-to-end remote sensing image processing and analysis workflow, it integrates key technologies such as geometric correction, radiometric normalization, cloud removal and restoration, multimodal change detection, patch boundary optimization, and base map overlay verification, forming a standardized, automated, and high-precision land use change information extraction system. This method and system can effectively eliminate the spatiotemporal inconsistencies of multi-source remote sensing images, accurately identify land use change areas, and generate vector patch results that conform to industry standards, significantly improving the reliability and practicality of change detection.

[0007] This invention provides a method for extracting land use change information based on multi-source remote sensing images, including: Acquire remote sensing image data from at least two different time phases, wherein the remote sensing image data is in TIFF format and includes optical remote sensing images or synthetic aperture radar remote sensing images; Multi-source remote sensing image preprocessing operations are performed on the remote sensing image data. The preprocessing operations include geometric correction, radiometric normalization, cloud removal, and format adaptation to generate a standardized remote sensing image dataset that is spatially aligned, radiometrically consistent, cloud-free, and retains the original spatial reference information. Based on the standardized remote sensing image dataset, a land use change detection operation is performed. The change detection operation includes selecting an appropriate change detection algorithm according to the image type, extracting spectral features and texture features to construct a difference feature set, and generating a binarized initial change mask. The initial change mask is subjected to post-processing and verification operations for change patches. These operations include patch boundary optimization, base map patch overlay verification, and result output to generate high-precision land use change patch files and change detection accuracy reports that meet industry standards.

[0008] Preferably, the multi-source remote sensing image preprocessing operation is performed on the remote sensing image data, including: Ground control points were selected to perform spatial coordinate matching on remote sensing images of different time phases. A polynomial transformation model was used to unify the images to the same geographic coordinate system. The spatial resolution of the corrected images was consistent with that of the original images, and the geometric position error was less than a pixel. Histogram matching is performed on each band of remote sensing images from different time periods to make the radiation distribution of the target image consistent with that of the reference image, or a linear regression model is used to establish the radiation brightness mapping relationship between corresponding bands of two images, and the radiation scale of the target image is adjusted accordingly. The cloud and cloud shadow regions in the image are identified based on a multi-threshold segmentation method. The multi-threshold includes a near-infrared band reflectance threshold, a blue light band reflectance threshold, and a normalized vegetation index threshold. The identified cloud areas are repaired by a spatiotemporal interpolation algorithm. The spatiotemporal interpolation algorithm uses the pixel values ​​of cloudless areas in adjacent time phases at the same geographical location for weighted interpolation. The preprocessed image is output in a standardized TIFF format that retains the original spatial reference information, and its spatial resolution, projection parameters and coordinate system are consistent with the reference image in the input image.

[0009] Preferably, based on the standardized remote sensing image dataset, a land use change detection operation is performed, including: If the input image is an optical remote sensing image, the image difference method or principal component analysis method is used for change detection; if the input image includes synthetic aperture radar remote sensing image, a deep learning model is used for change detection. The image interpolation method includes calculating the difference in normalized vegetation index, brightness difference, or water index; the principal component analysis method includes performing principal component transformation on two images and extracting the difference of the first principal component as the basis for change discrimination; the deep learning model includes a U-Net network structure or a ChangeNet network structure, whose input is the stitched tensor of two standardized remote sensing images, and whose output is the probability value of each pixel belonging to the change category; A differential feature set is constructed by extracting spectral and texture features. The spectral features include gray values ​​of each band, normalized vegetation index, normalized water index, and soil-regulated vegetation index. The texture features include contrast, correlation, energy, and homogeneity calculated based on the gray-level co-occurrence matrix, as well as variance, entropy, and gradient magnitude calculated based on local windows.

[0010] Preferably, generating the initial change mask for binarization includes: Binarization is performed on the differential feature set by setting a fixed threshold or an adaptive threshold. Pixels with a difference value greater than the threshold are marked as changed regions, and the remaining pixels are marked as unchanged regions. Alternatively, the deep learning model can directly output a binary mask, with the pixel value of changed regions being 1 and the pixel value of unchanged regions being 0.

[0011] Preferably, the optimization of the patch boundaries includes: Morphological closing operations are performed on the initial change mask to fill the internal holes. Then, a region growing algorithm is used to perform connectivity analysis on the edge pixels. Combined with adaptive threshold segmentation, the optimal boundary is determined, isolated pixels and jagged edges are eliminated, and the continuous and complete boundary of the patch is ensured.

[0012] Preferably, the base map patch overlay verification includes: Import the land use status base map data in SHP format, perform spatial overlay analysis between the optimized change maps and the base map maps, and calculate the overlap between their boundaries. When the overlap is greater than 85%, replace the corresponding boundary of the change map map with the boundary of the base map map map to complete the map map standardization process.

[0013] Preferably, the result output includes: Generate SHP format land use change patch files, whose attribute table includes change type, change area, change start and end time, and change confidence fields; at the same time, generate a change detection accuracy report, which includes the missed detection rate, false detection rate, overall accuracy, and Kappa coefficient.

[0014] Preferably, in the geometric correction, the ground control points are evenly distributed across the entire image area, the polynomial transformation is of order two or three, and the parameters are solved with minimizing the sum of squared residuals as the objective function.

[0015] Preferably, the stable and invariant region in the radiation normalization includes bare soil or water bodies, used to fit the linear regression model. , For the target image band value, For reference image band values, The slope This is the intercept.

[0016] This invention also provides a land use change information extraction system based on multi-source remote sensing images, comprising: A multi-source remote sensing image input unit is used to receive TIFF format remote sensing image data from at least two different time phases, wherein the remote sensing image data includes optical remote sensing images or synthetic aperture radar remote sensing images. The multi-source remote sensing image preprocessing unit is used to perform geometric correction, radiometric normalization, cloud removal and format adaptation on the remote sensing image data to generate a standardized remote sensing image dataset. The land use change detection unit is used to select an appropriate change detection algorithm based on the standardized remote sensing image dataset and the image type, extract spectral and texture features, and generate an initial change mask. The post-processing and verification unit for changed patches is used to perform patch boundary optimization, base map patch overlay verification, and result output on the initial change mask, generating a high-precision land use change patch file and a change detection accuracy report.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention solves the problem of low detection accuracy of land use change information extraction caused by geometric misalignment, radiation differences and cloud pollution in multi-source remote sensing images in the prior art by constructing an integrated framework for the entire process.

[0018] 2. The geometric correction process ensures strict spatial alignment of images from different time phases, with errors controlled within one pixel, laying the foundation for subsequent pixel-level comparisons; radiometric normalization eliminates radiometric deviations caused by illumination and atmospheric conditions, improving spectral comparability; cloud removal uses a combination of multi-threshold segmentation and spatiotemporal interpolation to achieve high-precision cloud area restoration and ensure data integrity.

[0019] 3. In the change detection stage, the present invention intelligently adapts the optimal algorithm according to the image type, which not only retains the stability of traditional methods on optical images, but also introduces a deep learning model to process the complex scattering characteristics of synthetic aperture radar images, thereby improving detection robustness.

[0020] 4. The difference feature set integrates spectral and textural multidimensional information to enhance sensitivity to subtle changes. The post-processing stage optimizes the map patch boundaries through region growing and adaptive threshold segmentation to eliminate noise interference; the base map overlay verification mechanism introduces authoritative land use status data as constraints to ensure that the output map patches conform to industry standards, and automatic overlay is achieved when the boundary overlap meets the standard, balancing automation and standardization.

[0021] 5. The final output SHP format patch file contains complete attribute information, and the accuracy report quantitatively evaluates the detection performance, which facilitates seamless integration with business systems such as land, planning, and environmental protection, providing highly reliable decision support for land supervision, ecological assessment, and urban planning. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the multi-source remote sensing image adaptive change detection algorithm selection and multi-dimensional feature fusion in this invention; Figure 3 This is a logical flowchart of the multi-source remote sensing image preprocessing stage in this invention. Figure 4 This is a logical flowchart of the initial change mask generation and post-processing verification stages in this invention. Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between land use change patch boundary optimization and base map overlay verification in this invention; Figure 6 This is a schematic diagram of the multi-level interaction relationships and data flow between the various functional units of the system and the external geographic information system and database in this invention. Detailed Implementation

[0023] Please refer to Figures 1 to 6This invention provides a method and system for extracting land use change information based on multi-source remote sensing images. Its core lies in constructing an end-to-end, fully automated processing chain to achieve high-precision identification and standardized patch output of land use change areas in remote sensing images from different time phases. The method strictly follows a pre-defined S-step sequence, sequentially executing four major stages: remote sensing image acquisition, multi-source remote sensing image preprocessing, land use change detection, and post-processing and verification of changed patches. This ensures that every step from raw data input to final output is reproducible, standardized, and highly robust.

[0024] The method begins with step S1: acquiring remote sensing image data from at least two different time periods. The remote sensing image data is in TIFF format and includes either optical remote sensing imagery or synthetic aperture radar (SAR) remote sensing imagery. In this step, the system receives two or more periods of remote sensing imagery through a standard file interface. Each period of imagery carries complete geospatial metadata, including the projected coordinate system, geographic transformation parameters, number of bands, and bit depth information. The imagery time span covers the variation period to be analyzed, with a typical time interval of 6 months to 5 years.

[0025] The sources of input imagery include, but are not limited to, Gaofen series satellites, Landsat series, Sentinel-2 optical imagery, and Sentinel-1 synthetic aperture radar imagery. All imagery must be submitted in uncompressed TIFF format to ensure lossless preservation of the original radiometric information and spatial reference structure. During reception, the system verifies the integrity of the imagery, including checking header identifiers, band dimension consistency, and georeference validity. If any missing or corrupted images are found, a data retransmission mechanism is triggered until a complete and valid input set is obtained.

[0026] The process then proceeds to step S2: Multi-source remote sensing image preprocessing is performed on the remote sensing image data. This preprocessing includes geometric correction, radiometric normalization, cloud removal, and format adaptation to generate a standardized remote sensing image dataset that is spatially aligned, radiometrically consistent, cloud-free, and retains original spatial reference information. This step is fundamental to the entire process, and its quality directly determines the reliability of subsequent change detection.

[0027] Step S2 first performs a geometric correction sub-operation. This sub-operation selects ground control points to perform spatial coordinate matching on remote sensing images from different time phases, and uses a polynomial transformation model to unify all images to the same geographic coordinate system. Ground control points are extracted from high-resolution base maps or historical orthophotos using an automatic feature matching algorithm, with no fewer than 30 matching points evenly distributed across the entire image area. The order of the polynomial transformation is adaptively selected based on the degree of image distortion, typically second or third order, with the objective function being the minimization of the sum of squared residuals. After correction, the spatial positions of all images are strictly aligned, the geometric position error is controlled within pixels, and the spatial resolution remains consistent with the original image, avoiding additional blurring or information loss introduced by resampling.

[0028] Next, a radiometric normalization sub-operation is performed. This sub-operation addresses the inconsistency in radiance caused by differences in solar elevation angle, atmospheric scattering, and sensor response between images from different time phases. The system first designates one phase image as the reference image and the remaining images as target images. For each band, the histogram cumulative distribution function between the target image and the reference image is calculated, and a one-to-one mapping relationship is established. The pixel values ​​of the target image are replaced with pixel values ​​in the reference image that have the same cumulative probability, thereby achieving forced alignment of the radiative distribution.

[0029] When significant linear offset exists between images, the system switches to a linear regression model: sampling corresponding pixel pairs within stable regions (such as bare soil or water bodies) and fitting the target image band values. Compared with reference image band values Relationship , The slope The intercept is used. This linear regression model is applied to the entire image to complete the radiometric scale adjustment. Both methods ensure that the radiometric characteristics of the output image are highly consistent with the reference image, eliminating spurious change signals caused by non-ground features.

[0030] The cloud removal sub-operation is then executed. This sub-operation addresses the common cloud and cloud shadow contamination issues in optical images. The system first identifies cloud areas based on a multi-threshold segmentation method: pixels with near-infrared reflectance greater than 0.3, blue light reflectance greater than 0.2, and normalized vegetation index less than 0.1 are designated as candidate cloud pixels; further filtering is then performed using a brightness temperature threshold (if the thermal infrared band is available). Cloud shadow areas are determined by back-projecting the cloud location against the solar azimuth angle.

[0031] After identification, spatiotemporal interpolation is performed to repair the cloud areas. The repair algorithm uses pixel values ​​from cloudless areas at the same geographical location in adjacent time phases (one phase before and one phase after), weighted by time distance. The weighting function is as follows: , For the current time phase, The nearest time phase is used. If the nearest time phase also contains clouds, the process is repeated until valid data is found. After restoration, cloud area pixels are assigned reasonable estimates, and the overall restoration accuracy is greater than 90%, meaning the root mean square error of the spectrum between the restored area and the real surface is less than a preset threshold.

[0032] Finally, the format adaptation sub-operation is performed. This sub-operation ensures that all preprocessed images are output in a standardized TIFF format, preserving the spatial reference information of the original images, including projection type, geographic transformation matrix, coordinate origin, and cell size. The band order and data type (e.g., 16-bit unsigned integer) of the output images are completely consistent with the reference images, facilitating direct use by subsequent modules. At this point, step S2 is complete, generating a standardized remote sensing image dataset that is spatially strictly aligned, radiometrically consistent, cloud-free, and format-unified.

[0033] Proceed to step S3: Based on the standardized remote sensing image dataset, perform land use change detection operation. The change detection operation includes selecting an appropriate change detection algorithm according to the image type, extracting spectral features and texture features to construct a difference feature set, and generating a binarized initial change mask.

[0034] Step S3 begins with the detection algorithm selecting a sub-unit to determine the type of input image. If all images are optical remote sensing images, conventional image processing methods are used; if any image contains synthetic aperture radar (SAR) remote sensing images, a deep learning model is used. This decision is made automatically based on the sensor type field in the image metadata.

[0035] When using traditional methods, the system offers two options: image interpolation or principal component analysis. Image interpolation calculates the pixel-by-pixel difference between corresponding bands of two images, generating a differenced image. Typical differences include the normalized difference vegetation index (NDVI) difference. Brightness difference Water index difference , .

[0036] , The normalized vegetation indexes for time 1 and time 2 are respectively. It is in the near-infrared band. It is in the red light band. Multispectral images 1, 2... Band. , These are the water indices for time 1 and time 2, respectively. It is in the green light band. It is in the shortwave infrared band.

[0037] Principal component analysis stitches two images together into a multi-band cube, performs principal component transformation, and extracts the first principal component. and Calculate the difference . It concentrates the information with the largest variance between images and is sensitive to changes.

[0038] When a deep learning model is enabled, the system loads a pre-trained U-Net or ChangeNet network. The input is a tensor formed by stitching two standardized images along the band dimension, with a size of [size missing]. , , For the image height and width, , This represents the number of bands in two phases. The network extracts multi-scale contextual features through an encoder-decoder structure, outputting the probability value of each pixel belonging to the change category. .

[0039] Regardless of the algorithm used, the system synchronously executes the differential feature extraction sub-operation. Spectral features include gray values ​​of each band, normalized vegetation index, normalized water index, and soil-regulated vegetation index; texture features are calculated using a sliding window with a window size of 7×7 pixels. Based on the gray-level co-occurrence matrix, contrast, correlation, energy, and homogeneity are calculated; based on local statistics, variance, entropy, and gradient magnitude are calculated. All features constitute a high-dimensional differential feature set. , For the set of real numbers, For feature dimensions.

[0040] The initial change mask generator sub-operation performs binarization based on the above results. For traditional methods, the system... , Adaptive threshold setting for equal difference images T is automatically determined using the Otsu algorithm to maximize the inter-class variance. If the absolute value of the difference is greater than... If the pixel is a variable, it is marked as a changed pixel. For deep learning models, the system sets a probability threshold of 0.5. Pixels with a value greater than 0.5 are marked as changes. The final output is a binary image. , Indicates the area of ​​change. Indicates the region that does not change.

[0041] Proceed to step S4: Perform post-processing and verification operations on the initial change mask, including patch boundary optimization, base map patch overlay verification, and result output, to generate a high-precision land use change patch file and change detection accuracy report that conforms to industry standards.

[0042] Step S4 first executes the patch boundary optimization sub-operation. The system performs initial change masking. Morphological closing operations are performed, with the structuring element being a 3×30 character shape, to fill internal small holes and connect neighboring fragments. A region growing algorithm is then employed: traversing all changing pixels, and based on their 8-neighborhood connectivity, merging adjacent pixels to form connected regions. For the edge pixels of each connected region, the system calculates their local gradient magnitude and combines this with adaptive thresholding to determine the optimal boundary.

[0043] The adaptive threshold is dynamically adjusted based on the difference in spectral mean between the inside and outside of the region, ensuring that the boundary conforms to the real-world feature contours. This process eliminates isolated pseudo-variable pixels and jagged edges, outputting an optimized variation mask. The boundaries of its patches are continuous and complete.

[0044] The subsequent sub-operation of base map patch overlay verification is performed. The system imports land use status base map patch data in SHP format, which includes the latest land classification vector map data released by authoritative departments. Convert to vector polygons and perform spatial overlay analysis with the base map patches. For each changed patch, calculate its boundary overlap degree with the intersecting base map patches. , defined as the ratio of the length of the common boundary to the perimeter of the changing patch. When In this case, the system replaces the corresponding boundaries of the changed patches with the boundaries of the base map patches, completing the standardization process. This operation ensures that the output patches comply with current land survey regulations, improving the administrative usability of the results.

[0045] The final output sub-operation generates land use change patch files in SHP format. Each patch records the following attributes: change type (inferred from spectral characteristics of the preceding and following phases, such as conversion of cultivated land to construction land), change area (square meters), change start and end time (determined by the image acquisition date), and change confidence score (obtained based on deep learning probability or difference amplitude normalization). A change detection accuracy report is also generated, including the false negative rate (the proportion of undetected true changes), false positive rate (the proportion of stable areas incorrectly labeled as changed), overall accuracy (the proportion of correctly classified pixels), and Kappa coefficient (a classification consistency index considering random consistency). All results can be directly imported into mainstream geographic information system software for subsequent analysis and decision-making.

[0046] At the system level, this invention provides a land use change information extraction system based on multi-source remote sensing images. The system is deployed on a computing platform equipped with a central processing unit, graphics processing unit, memory, and storage devices, running a geographic information system software environment. The system comprises four core functional units.

[0047] The multi-source remote sensing image input unit is responsible for receiving TIFF format remote sensing image data from at least two different temporal phases. This unit is equipped with a standard file parser, capable of reading all metadata under the GeoTIFF specification and verifying image integrity. The input interface supports batch uploads and streaming access, adapting to the needs of different data sources.

[0048] The multi-source remote sensing image preprocessing unit includes a geometric correction subunit, a radiometric normalization subunit, a cloud removal subunit, and a format adaptation subunit. The geometric correction subunit integrates an automatic ground control point matching engine and a polynomial transform solver; the radiometric normalization subunit has a built-in histogram matching and linear regression dual-mode correction module; the cloud removal subunit implements multi-threshold cloud identification and spatiotemporal interpolation restoration algorithms; and the format adaptation subunit ensures that the output conforms to the OGC GeoTIFF standard.

[0049] The land use change detection unit includes a detection algorithm selection subunit, a differential feature extraction subunit, and an initial mask generation subunit. The detection algorithm selection subunit automatically routes to traditional methods or deep learning models based on the sensor type field in the image metadata; the differential feature extraction subunit calculates spectral indices and texture statistics in parallel; and the initial mask generation subunit performs threshold segmentation or neural network inference and outputs a binary mask.

[0050] The post-processing and verification unit for changed patches includes a boundary optimization subunit, a base map overlay subunit, and a result output subunit. The boundary optimization subunit implements morphological operations and region growing algorithms; the base map overlay subunit performs spatial overlay analysis and boundary overlap calculation; and the result output subunit generates standardized SHP files and accuracy reports, and interfaces with external land management databases via an API.

[0051] The entire system achieves high cohesion and low coupling through modular design. Data is passed between units via memory sharing or message queues to ensure processing efficiency. The system supports distributed computing, enabling parallel processing of large-scale imagery blocks, significantly shortening the job cycle. The final output meets the comprehensive requirements of land surveys, ecological monitoring, and urban planning for accuracy, timeliness, and standardization of change information.

[0052] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0053] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for extracting land use change information based on multi-source remote sensing images, characterized in that, include: Acquire remote sensing image data from at least two different time phases, wherein the remote sensing image data is in TIFF format and includes optical remote sensing images or synthetic aperture radar remote sensing images; Multi-source remote sensing image preprocessing operations are performed on the remote sensing image data. The preprocessing operations include geometric correction, radiometric normalization, cloud removal, and format adaptation to generate a standardized remote sensing image dataset that is spatially aligned, radiometrically consistent, cloud-free, and retains the original spatial reference information. Based on the standardized remote sensing image dataset, a land use change detection operation is performed. The change detection operation includes selecting an appropriate change detection algorithm according to the image type, extracting spectral features and texture features to construct a difference feature set, and generating a binarized initial change mask. The initial change mask is subjected to post-processing and verification operations for change patches. These operations include patch boundary optimization, base map patch overlay verification, and result output to generate high-precision land use change patch files and change detection accuracy reports that meet industry standards.

2. The method for extracting land use change information based on multi-source remote sensing images according to claim 1, characterized in that, Perform multi-source remote sensing image preprocessing on the remote sensing image data, including: Ground control points were selected to perform spatial coordinate matching on remote sensing images of different time phases. A polynomial transformation model was used to unify the images to the same geographic coordinate system. The spatial resolution of the corrected images was consistent with that of the original images, and the geometric position error was less than a pixel. Histogram matching is performed on each band of remote sensing images from different time periods to make the radiation distribution of the target image consistent with that of the reference image, or a linear regression model is used to establish the radiation brightness mapping relationship between corresponding bands of two images, and the radiation scale of the target image is adjusted accordingly. The cloud and cloud shadow regions in the image are identified based on a multi-threshold segmentation method. The multi-threshold includes a near-infrared band reflectance threshold, a blue light band reflectance threshold, and a normalized vegetation index threshold. The identified cloud areas are repaired by a spatiotemporal interpolation algorithm. The spatiotemporal interpolation algorithm uses the pixel values ​​of cloudless areas in adjacent time phases at the same geographical location for weighted interpolation. The preprocessed image is output in a standardized TIFF format that retains the original spatial reference information, and its spatial resolution, projection parameters and coordinate system are consistent with the reference image in the input image.

3. The method for extracting land use change information based on multi-source remote sensing images according to claim 2, characterized in that, Based on the standardized remote sensing image dataset, a land use change detection operation is performed, including: If the input image is an optical remote sensing image, the image difference method or principal component analysis method is used for change detection; if the input image includes synthetic aperture radar remote sensing image, a deep learning model is used for change detection. The image interpolation method includes calculating the difference in normalized vegetation index, brightness difference, or water index; the principal component analysis method includes performing principal component transformation on two images and extracting the difference of the first principal component as the basis for change discrimination; the deep learning model includes a U-Net network structure or a ChangeNet network structure, whose input is the stitched tensor of two standardized remote sensing images, and whose output is the probability value of each pixel belonging to the change category; A differential feature set is constructed by extracting spectral and texture features. The spectral features include gray values ​​of each band, normalized vegetation index, normalized water index, and soil-regulated vegetation index. The texture features include contrast, correlation, energy, and homogeneity calculated based on the gray-level co-occurrence matrix, as well as variance, entropy, and gradient magnitude calculated based on local windows.

4. The method for extracting land use change information based on multi-source remote sensing images according to claim 3, characterized in that, Generate an initial change mask for binarization, including: Binarization is performed on the differential feature set by setting a fixed threshold or an adaptive threshold. Pixels with a difference value greater than the threshold are marked as changed regions, and the remaining pixels are marked as unchanged regions. Alternatively, the deep learning model can directly output a binary mask, with the pixel value of changed regions being 1 and the pixel value of unchanged regions being 0.

5. The method for extracting land use change information based on multi-source remote sensing images according to claim 4, characterized in that, The optimization of the patch boundaries includes: Morphological closing operations are performed on the initial change mask to fill the internal holes. Then, a region growing algorithm is used to perform connectivity analysis on the edge pixels. Combined with adaptive threshold segmentation, the optimal boundary is determined, isolated pixels and jagged edges are eliminated, and the continuous and complete boundary of the patch is ensured.

6. The method for extracting land use change information based on multi-source remote sensing images according to claim 5, characterized in that, The base map patch overlay verification includes: Import the land use status base map data in SHP format, perform spatial overlay analysis between the optimized change maps and the base map maps, and calculate the overlap between their boundaries. When the overlap is greater than 85%, replace the corresponding boundary of the change map map with the boundary of the base map map map to complete the map map standardization process.

7. The method for extracting land use change information based on multi-source remote sensing images according to claim 6, characterized in that, The output results include: Generate SHP format land use change patch files, whose attribute table includes change type, change area, change start and end time, and change confidence fields; at the same time, generate a change detection accuracy report, which includes the missed detection rate, false detection rate, overall accuracy, and Kappa coefficient.

8. The method for extracting land use change information based on multi-source remote sensing images according to claim 1, characterized in that, In the geometric correction, ground control points are uniformly distributed across the entire image area, and the polynomial transformation is of order two or three. The parameters are solved with the objective function of minimizing the sum of squared residuals.

9. The method for extracting land use change information based on multi-source remote sensing images according to claim 2, characterized in that, The stable and invariant regions in the radiation normalization include bare soil or water bodies, used to fit the linear regression model. , For the target image band value, For reference image band values, The slope This is the intercept.

10. A land use change information extraction system based on multi-source remote sensing images, characterized in that, include: A multi-source remote sensing image input unit is used to receive TIFF format remote sensing image data from at least two different time phases, wherein the remote sensing image data includes optical remote sensing images or synthetic aperture radar remote sensing images. The multi-source remote sensing image preprocessing unit is used to perform geometric correction, radiometric normalization, cloud removal and format adaptation on the remote sensing image data to generate a standardized remote sensing image dataset. The land use change detection unit is used to select an appropriate change detection algorithm based on the standardized remote sensing image dataset and the image type, extract spectral and texture features, and generate an initial change mask. The post-processing and verification unit for changed patches is used to perform patch boundary optimization, base map patch overlay verification, and result output on the initial change mask, generating a high-precision land use change patch file and a change detection accuracy report.

Images