Method and device for identifying soil type of ground substrate, equipment and medium

Abstract

The application provides a surface matrix soil type remote sensing identification method and device, equipment and medium, and belongs to the technical field of soil type identification. The method comprises the following steps: determining a bottom plate plot corresponding to each interpretation plot; judging whether an absolute inheritance rule is triggered based on the land class attribute of the bottom plate plot corresponding to each interpretation plot; if the absolute inheritance rule is triggered, the surface matrix attribute of the interpretation plot is directly corrected to obtain a corrected soil type plot set; if the absolute inheritance rule is not triggered, judging whether a collaborative correction rule is triggered based on the land class attribute; if the collaborative correction rule is triggered, the surface matrix attribute of the interpretation plot is corrected based on distance weighted voting to obtain the corrected soil type plot set; and a surface matrix soil type distribution map is generated based on the corrected soil type plot set. The application can improve the efficiency and accuracy of surface matrix soil type remote sensing identification.

Description

Technical Field

[0001] This application belongs to the field of soil type identification technology, and more specifically, it relates to a remote sensing identification method, device, equipment, and medium for surface matrix soil type. Background Technology

[0002] The surface matrix is ​​the basic material carrier that supports vegetation, water bodies, construction land and other surface cover layers. Its soil type and spatial distribution information are core basic data for land resource surveys, ecological environment monitoring, desertification prevention and control and national spatial planning. Conducting large-scale, high-precision identification of surface matrix soil types has important engineering and scientific research value.

[0003] Currently, methods for obtaining surface matrix soil types are mainly divided into two categories: field surveys and remote sensing identification. Traditional field surveys rely on manual field sampling and laboratory testing. While the results are reliable, they suffer from problems such as long investigation cycles, high labor costs, and limited coverage, making it difficult to meet the needs of rapid surveys over large areas. With the development of remote sensing technology, surface information interpretation based on satellite imagery has become the mainstream technical method, which can improve identification efficiency to a certain extent.

[0004] However, existing remote sensing identification methods have the following problems: On the one hand, remote sensing images are easily affected by factors such as soil moisture, surface roughness, and vegetation cover, resulting in prominent phenomena of different spectra for the same object and different objects with the same spectrum. It is difficult to reliably distinguish different soil types by relying solely on spectral information, leading to low accuracy of identification results. On the other hand, remote sensing classification results are mostly based on raster pixels, resulting in blurred boundaries of image patches and poor consistency with the boundaries of real-world features. Furthermore, they fail to effectively integrate with prior geospatial data such as land use, topography, and geology, and the interpretation results lack spatial logical constraints, making it difficult to guarantee overall accuracy.

[0005] Therefore, improving the efficiency and accuracy of remote sensing identification of surface matrix soil types is a technical problem that needs to be solved in this field. Summary of the Invention

[0006] This application provides a method, apparatus, equipment, and medium for remote sensing identification of surface matrix soil types, in order to improve the efficiency and accuracy of remote sensing identification of surface matrix soil types.

[0007] According to one aspect of the embodiments of this application, a remote sensing identification method for surface matrix soil type is provided, comprising: Orthorectified seamless mosaic imagery is obtained based on multi-source remote sensing data of the target area; Image segmentation is performed on the orthophoto seamless mosaic image to obtain multiple image objects. The multidimensional feature vector of each image object is extracted. The multidimensional feature vectors are then input into a pre-trained multi-level rule decision tree model to perform hierarchical discrimination and classification on each image object, thereby obtaining the preliminary surface matrix type corresponding to each image object. Each preliminary surface matrix type is converted into a vector map patch to obtain a preliminary remote sensing interpretation patch vector set; the preliminary remote sensing interpretation patch vector set includes multiple interpretation patches. Acquire land use data from the Third National Land Survey of the target area, which includes multiple base map patches; spatially align the land use data from the Third National Land Survey with orthophoto seamless mosaic imagery. Determine the corresponding intersecting base map features for each interpreted map feature, and determine whether the absolute inheritance rule is triggered based on the land use attributes of the corresponding intersecting base map features. If the absolute inheritance rule is triggered, the surface matrix properties of the interpreted patch will be directly corrected to obtain the corrected soil type patch set; If the absolute inheritance rule is not triggered, the land type attribute is used to determine whether the collaborative correction rule is triggered. If the collaborative correction rule is triggered, the surface matrix attribute of the interpreted patch is corrected based on distance-weighted voting to obtain the corrected soil type patch set. A distribution map of surface matrix soil types is generated based on the corrected soil type patch set.

[0008] According to one aspect of the embodiments of this application, a remote sensing identification device for surface matrix soil type is provided, comprising: The data preprocessing module is used to obtain orthorectified seamless mosaic images based on multi-source remote sensing data of the target area; The preliminary surface matrix type identification module is used to segment the orthophoto seamless mosaic image to obtain multiple image objects and extract the multi-dimensional feature vector of each image object. The multi-dimensional feature vectors are then input into the trained multi-level rule decision tree model to perform hierarchical discrimination and classification on each image object to obtain the preliminary surface matrix type corresponding to each image object. The remote sensing preliminary interpretation patch generation module is used to convert each preliminary surface matrix type into vector patches to obtain a remote sensing preliminary interpretation patch vector set; the remote sensing preliminary interpretation patch vector set includes multiple interpretation patches; The Third National Land Survey acquisition module is used to acquire land use data from the Third National Land Survey of the target area. The land use data includes multiple base map patches; the land use data from the Third National Land Survey is spatially aligned with orthophoto seamless mosaic imagery. The rule triggering module is used to determine the intersecting base plate patches corresponding to each interpreted patch, and to determine whether to trigger the absolute inheritance rule based on the land use attributes of the intersecting base plate patches corresponding to each interpreted patch; The absolute inheritance rule module is used to directly correct the surface matrix properties of the interpreted patch if the absolute inheritance rule is triggered, so as to obtain the corrected soil type patch set. The collaborative correction rule module is used to determine whether to trigger the collaborative correction rule based on the land type attribute if the absolute inheritance rule is not triggered. If the collaborative correction rule is triggered, the surface matrix attribute of the interpreted patch is corrected based on distance-weighted voting to obtain the corrected soil type patch set. The surface matrix soil type distribution map generation module is used to generate a surface matrix soil type distribution map based on the corrected soil type patch set.

[0009] According to one aspect of the embodiments of this application, an electronic device is provided, which includes a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement the above-described remote sensing identification method for surface matrix soil type.

[0010] According to one aspect of the embodiments of this application, the computer program product includes a computer program stored in a computer-readable storage medium. A processor of an electronic device reads the computer program from the computer-readable storage medium and executes the computer program, causing the electronic device to perform the aforementioned remote sensing identification method for surface matrix soil type.

[0011] The technical solutions provided in this application embodiment may have the following beneficial effects: This application employs a collaborative processing model combining multi-source remote sensing data and land use data from the Third National Land Survey. By generating orthorectified seamless mosaic images and spatially aligning them with the land use data from the Third National Land Survey, it ensures that remote sensing information and prior geographical knowledge maintain a unified benchmark, avoiding identification biases caused by data mismatches. This reduces the workload of manual processing and repeated verification, improves the identification efficiency of surface matrix soil types over large areas, and meets the needs of rapid surveys. This application segmentation of the orthorectified seamless mosaic images yields image objects. Multi-dimensional feature vectors are extracted from each image object, and a trained multi-level rule-based decision tree model is used to perform hierarchical classification of each image object. This objectively and stably extracts the essential characteristics of the surface matrix, reducing the subjectivity and uncertainty of manual interpretation. It effectively alleviates the problems of susceptibility to interference and unstable classification in traditional remote sensing identification, improving the accuracy of preliminary surface matrix type judgment. Simultaneously, this application corrects surface matrix attributes by triggering absolute inheritance rules or collaborative correction rules, making the boundaries of the corrected soil type patch set more closely match the boundaries of actual land features, solving the defects of vague classification results and discrepancies with the actual situation. The resulting surface matrix soil type distribution map is more accurate and usable, providing reliable data support for land resource management, ecological monitoring, and other tasks. Therefore, the embodiments of this application can improve the efficiency and accuracy of remote sensing identification of surface matrix soil types. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating the remote sensing identification method for surface matrix soil type provided in this application embodiment; Figure 2 Structural block diagram of the remote sensing identification device for surface matrix soil type provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0015] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0016] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0017] Figure 1 This is a schematic flowchart of the remote sensing identification method for surface matrix soil type provided in an embodiment of this application. The method is executed by an electronic device and may include: S101: An orthorectified seamless mosaic image is obtained based on multi-source remote sensing data of the target area.

[0018] In this embodiment, the target area is the geographic spatial range for identifying surface matrix soil types. This embodiment first determines the geographic boundaries and survey scale of the target area, and then selects multi-source remote sensing data covering the target area accordingly.

[0019] In this embodiment, an orthorectified seamless mosaic image is obtained based on multi-source remote sensing data of the target area, including: Multiple apparent radiance images were obtained by radiometric calibration of multi-source remote sensing data; Atmospheric correction was performed on each apparent radiance image to obtain multiple surface reflectance images; Geometric and orthorectification are performed on surface reflectance images from various locations to obtain multiple orthorectified images; each orthorectified image includes a high-resolution orthorectified image and a medium-resolution orthorectified image; the spatial resolution of the high-resolution orthorectified image is less than or equal to a first spatial resolution threshold; the spatial resolution of the medium-resolution orthorectified image is greater than the first spatial resolution threshold and less than or equal to a second spatial resolution threshold, and the first spatial resolution threshold is less than the second spatial resolution threshold. Image fusion is performed on the panchromatic and multispectral bands of each high-resolution orthophoto to obtain the corresponding high-resolution fused image. Raster registration and resampling were performed on each high-resolution fused image and each medium-resolution orthophoto to obtain multiple registered images with consistent spatial reference. By mosaicking multiple registered images with consistent spatial references, an orthorectified seamless mosaic image is obtained.

[0020] In this embodiment, the multi-source remote sensing data includes multiple raw remote sensing images covering the target area, acquired by different satellite platforms at different times, and may include high-resolution and medium-resolution remote sensing images. The spatial resolution of high-resolution remote sensing images is less than or equal to 2 meters (e.g., GF-2 images), while the spatial resolution of medium-resolution remote sensing images is between 2 meters and 30 meters (e.g., Sentinel-2 MSI images, Landsat 8 OLI images). Each raw remote sensing image contains its own spatial reference information and radiometric calibration parameters, which are recorded in the image header file. Each raw remote sensing image consists of regularly arranged pixels, and each pixel stores a raw digital quantization value. This value is dimensionless and its magnitude reflects the relative intensity of ground surface radiation recorded by the satellite platform, but it does not have absolute physical meaning before calibration.

[0021] Radiometric calibration is performed on multi-source remote sensing data. For each original remote sensing image, the following operations are performed: read the radiometric calibration parameters from the image header file, including the gain coefficients for each band. With offset The original digital quantization value of each pixel in the original remote sensing image (i.e. Substitute into the formula ,in For this pixel in the band The apparent radiance values ​​are calculated for each pixel in each band. The calculated apparent radiance values ​​are then filled row-by-row and column-by-column according to the pixel's spatial position in the original remote sensing image, maintaining the spatial arrangement of the original image. After processing all pixels, the apparent radiance image corresponding to the original remote sensing image is obtained. After scene-by-scene processing, the apparent radiance image corresponding to each scene of the original remote sensing image is obtained.

[0022] Atmospheric correction is performed on each apparent radiance image, eliminating the influence of aerosols and water vapor based on a radiative transfer model to obtain a surface reflectance image. Geometric and orthorectification corrections are then performed on the surface reflectance image, and positional and topographical deviations are corrected using ground control points and a digital elevation model to obtain an orthorectified image. The orthorectified image is divided into high-resolution orthorectified images (spatial resolution ≤ 2m) and medium-resolution orthorectified images (2m < resolution ≤ 30m). Panchromatic and multispectral image fusion is performed on the high-resolution orthorectified image to enhance spatial detail. Raster registration and resampling are performed on the high-resolution fused image and the medium-resolution orthorectified image to ensure all images have a consistent spatial reference and pixel size. Mosaic processing is performed on the spatially reference-consistent registered images, generating a seamless orthorectified mosaic image through tone matching and optimal stitching. The seamless orthorectified mosaic image is a complete image without stitching gaps or tonal differences.

[0023] This embodiment improves the spectral and spatial accuracy of images by standardizing and preprocessing multi-source remote sensing data. The collaborative processing of high-resolution and medium-resolution orthorectified images balances recognition accuracy and coverage. Seamlessly mosaicked orthorectified images achieve complete coverage across the entire region, meeting the needs of large-scale surveys.

[0024] S102: Perform image segmentation on the orthophoto seamless mosaic image to obtain multiple image objects, and extract the multidimensional feature vector of each image object; input each multidimensional feature vector into the trained multi-level rule decision tree model to perform hierarchical discrimination and classification on each image object to obtain the preliminary surface matrix type corresponding to each image object.

[0025] In this embodiment, image segmentation is the process of dividing an orthophoto-seamless mosaic image into regions with uniform internal features. Image objects are spatial units with independent attributes obtained through segmentation. The multi-level rule-based decision tree model is a pre-trained hierarchical classification model used to determine land surface matrix types. The preliminary land surface matrix type is the initial classification result without correction from the third national land survey's land use data.

[0026] For example, this embodiment employs a multi-scale region segmentation method to perform image segmentation on orthorectified seamless mosaic images. With the constraint of minimizing spectral and shape heterogeneity, the orthorectified seamless mosaic images are divided into multiple image objects. For each image object, a multi-dimensional feature vector is extracted, which is used as the discrimination node at each level of the subsequent multi-level rule-based decision tree model. The segmented image objects replace the original pixels as the basic analysis unit, effectively avoiding the fragmentation problem that easily occurs in pixel-level classification, making the classification results more consistent with the actual distribution of land features. Extracting multi-dimensional feature vectors can characterize the differences in different surface matrix soil types from multiple dimensions such as spectral index features, spatial texture features, color space transformation features, and topographic features, improving the separability between different types. The pre-trained multi-level rule-based decision tree model is used for automatic hierarchical discrimination, replacing the traditional method of manually establishing interpretation markers, ensuring the objectivity and consistency of the classification process. The obtained preliminary surface matrix types provide input for subsequent vector patch conversion and land use data correction in the third national land survey, forming a progressively optimized recognition path from coarse to fine, balancing processing efficiency and final accuracy.

[0027] The multidimensional feature vector is a set of multiple indicators that comprehensively characterize the attributes of image objects, including spectral index features, spatial texture features, color space transformation features, and topographic features. As a specific implementation of these features in this embodiment, the multidimensional feature vector includes at least: abundance values ​​obtained through sub-pixel spectral unmixing (belonging to spectral features); slope and topographic humidity indices extracted from the digital elevation model (belonging to topographic features); homogeneity, texture contrast, and information entropy calculated through the gray-level co-occurrence matrix (belonging to spatial texture features); brightness and humidity components obtained through tassel transformation (belonging to color space transformation features); Euclidean distance from the water system calculated through spatial distance analysis (belonging to spatial location features); and bare soil index obtained through band operations (belonging to spectral index features). Each of these features corresponds to a different physical meaning and is used to characterize the differentiated performance of different surface matrix soil types in terms of spectral reflectance, spatial texture, color space, and topography.

[0028] The multi-level rule-based decision tree model employs a hierarchical discriminant structure, determining the surface matrix type of each image object step-by-step from coarse to fine. The specific discriminant process is as follows: Sub-pixel spectral unmixing is performed on each image object. Typical land cover spectra are selected as endmembers to determine the abundance value of pure rock or gravel. Image objects with abundance values ​​greater than a preset abundance threshold are identified as candidate objects of high exposed rock and gravel, thus obtaining a candidate set of high exposed rock and gravel objects. The preset abundance threshold is based on experience, for example, 70%. For each image object in the candidate set of high exposed rock and gravel, the slope and gray-level co-occurrence matrix homogeneity of the image object are determined. The image objects with a slope greater than the slope threshold and a homogeneity less than the homogeneity threshold are identified as exposed bedrock, and the preliminary surface matrix type corresponding to the image object is determined as bedrock type. The slope threshold is obtained based on experience, for example, 15°, and the homogeneity threshold is obtained based on experience, for example, 0.3. For each image object in the candidate set of highly exposed gravel that was not identified as exposed bedrock, the brightness component, information entropy, and Euclidean distance from the water system of each image object were determined. The image objects whose brightness component was greater than the brightness component threshold, whose information entropy was greater than the information entropy threshold, and whose Euclidean distance from the water system was less than the Euclidean distance threshold were identified as gravel matrix, and the preliminary surface matrix type corresponding to each image object was obtained as gravel type. The brightness component threshold was obtained based on experience, for example, the top 10% quantile of the brightness component of the whole area; the information entropy threshold was obtained based on experience, for example, 5.0; and the Euclidean distance threshold from the water system was obtained based on experience, for example, 500 meters. For each image object not included in the candidate set of high exposed rock and gravel, the bare soil index, brightness component, texture contrast, and Euclidean distance from the water system are determined. Image objects with a bare soil index greater than a first bare soil index threshold, a brightness component greater than a brightness component threshold, a texture contrast greater than a texture contrast threshold, and a Euclidean distance from the water system less than a threshold are identified as sandy soil types. The preliminary surface matrix type corresponding to each image object is thus determined to be sandy soil type. The first bare soil index threshold is obtained empirically, for example, 0.4; the brightness component threshold is obtained empirically, for example, 0.6; the texture contrast threshold is obtained empirically, for example, 0.5; and the Euclidean distance from the water system threshold is obtained empirically, for example, 500 meters. For each image object not included in the candidate set of highly exposed gravel and not identified as sandy soil, the moisture component, bare soil index, and topographic moisture index of each image object are determined. Image objects with a moisture component greater than a moisture component threshold, a bare soil index less than a second bare soil index threshold, and a topographic moisture index greater than a topographic moisture index threshold are identified as clay types, thus obtaining the preliminary surface matrix type corresponding to each image object as clay type. The moisture component threshold is obtained based on experience, for example, 0.7; the second bare soil index threshold is obtained based on experience, for example, 0.2; and the topographic moisture index threshold is obtained based on experience, for example, 8.0. For each image object that was not included in the candidate set of high exposed gravel, was not identified as sandy soil, and was not identified as clay, it was identified as loam soil, and the preliminary surface matrix type corresponding to each image object was loam soil.

[0029] In this embodiment, sub-pixel spectral unmixing is a process of decomposing mixed pixels and determining the content of each component. Abundance is the proportion of a certain land feature component in a pixel. The high-exposed rock and gravel candidate object set is a set of image objects with a high content of rock or gravel. Homogeneity is an index reflecting the uniformity of texture within an image object. Luminance component is an index reflecting the intensity of surface reflection. Information entropy is an index reflecting the complexity of texture. Euclidean distance from the water system is the spatial distance from the image object to the nearest water system. Bare soil index is an index reflecting the degree of surface bareness. Texture contrast is an index reflecting the degree of texture difference. Humidity component is an index reflecting the surface humidity status. Topographic humidity index is an index reflecting the regional water catchment conditions.

[0030] This embodiment first uses sub-pixel spectral unmixing to quickly screen areas with highly exposed gravel, achieving stratified differentiation between gravel and soil, thus reducing classification complexity. Then, combining features such as slope, homogeneity, brightness component, humidity component, and topographic humidity index, it classifies the soil in a step-by-step manner, prioritizing gravel over soil, which aligns with the spatial distribution pattern of the surface matrix. After determining the bedrock type, gravel type, sandy soil type, clay type, and loam type in each branch, the results are merged to obtain the preliminary surface matrix type corresponding to each image object. This stratified classification logic reduces misclassification of easily confused types, improves the reliability of the preliminary classification, and provides accurate intermediate results for subsequent vector patch conversion and multi-source data collaborative correction, ensuring the rationality and stability of the overall identification process.

[0031] S103: Convert each preliminary surface matrix type into vector plots to obtain a preliminary remote sensing interpretation plot vector set; the preliminary remote sensing interpretation plot vector set includes multiple interpretation plots.

[0032] In this embodiment, vector patches are geospatial vector units with clearly defined boundaries and attributes. The preliminary interpretation patch vector set is a set of vectors formed by assigning the preliminary surface matrix type corresponding to each image object as attribute information to vector patches whose outlines are defined by the boundaries of that image object. Each vector patch is an interpretation patch. This embodiment converts the raster format image object classification results (i.e., preliminary surface matrix types) into vector format patch data (i.e., the preliminary interpretation patch vector set), which facilitates subsequent spatial operations and attribute association with land use data from the Third National Land Survey in the same format.

[0033] For example, each image object is treated as an independent vector patch, and the preliminary surface matrix type corresponding to the image object is written as an attribute value into the attribute table of the vector patch, thus obtaining each interpreted patch and its corresponding surface matrix attribute. After all image objects have been processed, they are aggregated to form a preliminary remote sensing interpreted patch vector set covering the entire target area. Each vector patch in the preliminary remote sensing interpreted patch vector set is an interpreted patch, and each interpreted patch contains its boundary coordinates and surface matrix attribute.

[0034] S104: Obtain land use data from the Third National Land Survey of the target area, which includes multiple base map patches; spatially align the land use data from the Third National Land Survey with the orthophoto seamless mosaic image.

[0035] In this embodiment, the land use data from the Third National Land Survey is high-precision land use vector data generated during the Third National Land Survey. It includes the boundary locations and land use attributes of each plot. Each plot is a base map patch, and the land use attribute is the soil type recorded in the base map patch of the Third National Land Survey land use data. The geographic coordinate system and projected coordinate system of the Third National Land Survey land use data are completely consistent with the orthophoto seamless mosaic image, that is, spatial alignment is achieved.

[0036] For example, land use data from the third national land survey covering the target area is acquired, and the spatial boundary coordinates and corresponding land use attributes of each parcel in the data are read. The coordinate reference of the land use data from the third national land survey is transformed to the same projection coordinate system as the orthophoto seamless mosaic image, so that the two are accurately superimposed under the same spatial structure. After spatial alignment, each base map patch in the land use data from the third national land survey has a unique geographical location and land use attribute, which is used for constraint correction of the vector set of the preliminary remote sensing interpretation patch in subsequent steps.

[0037] S105: Determine the corresponding intersecting base plate features for each interpreted feature; determine whether the absolute inheritance rule is triggered based on the land use attributes of the corresponding intersecting base plate features for each interpreted feature.

[0038] In this embodiment, determining the intersecting base plate patches corresponding to each interpreted patch includes: Determine the bounding rectangle of each interpreted feature and the bounding rectangle of each base plate feature respectively; Determine whether there is coordinate overlap between the bounding rectangle of each interpreted feature and the bounding rectangle of each base feature; Remove patch pairs that do not have coordinate overlap to obtain a candidate set of intersecting patch pairs; the candidate set of intersecting patch pairs includes multiple patch pairs, each patch pair includes an interpreted patch and a corresponding intersecting base plate patch.

[0039] In this embodiment, the outer rectangle is the smallest rectangular area that encloses the patch. Coordinate overlap means that two rectangles intersect in space. The candidate set of intersecting patch pairs is a combination of patches that have spatial overlap.

[0040] For example, in this embodiment, the bounding rectangle of each interpreted patch and each base plate patch is calculated respectively. It is determined whether there is coordinate overlap between each bounding rectangle pair in the X-axis and Y-axis directions. Patch pairs that do not have coordinate overlap are filtered out, and patch pairs that have coordinate overlap are retained to form a set of candidate intersecting patch pairs.

[0041] This embodiment uses a bounding rectangle for rapid filtering, which avoids performing precise geometric calculations on all patch pairs and effectively improves the efficiency of spatial operations on massive patches.

[0042] Based on the land use attributes of the corresponding intersecting base plots for each interpreted plot, it is determined whether the land use attribute includes water area, hardened road surface, or building land. If it includes any of the above land use attributes, the absolute inheritance rule is triggered. If it does not include the above land use attributes, the absolute inheritance rule is not triggered, and the interpreted plot enters the judgment process of the collaborative correction rule.

[0043] S106: If the absolute inheritance rule is triggered, the surface matrix properties of the interpreted patch will be directly corrected to obtain the corrected soil type patch set.

[0044] In this embodiment, the absolute inheritance rule is a rule that forcibly corrects the surface matrix attributes of interpreted patches based on authoritative land use attributes. The land use attributes of the intersecting base patches corresponding to each interpreted patch are determined one by one. When land use attributes such as "water area," "paved road surface," or "building land" exist, the absolute inheritance rule is triggered, and the surface matrix attributes of the interpreted patch are directly corrected to a non-soil or artificial matrix type (e.g., building land), resulting in a corrected set of soil type patches.

[0045] S107: If the absolute inheritance rule is not triggered, then the land type attribute is used to determine whether the collaborative correction rule is triggered. If the collaborative correction rule is triggered, then the surface matrix attribute of the interpreted patch is corrected based on distance-weighted voting to obtain the corrected soil type patch set.

[0046] In this embodiment, for interpreted patches that do not trigger the absolute inheritance rule, based on the land use attributes and spatial dominance ratio of the corresponding intersecting base patches, it is determined whether there are dry land, forest land, paddy fields, or inland tidal flats among the land use attributes, and whether the spatial dominance ratio is greater than the spatial dominance ratio threshold (this spatial dominance ratio threshold is based on experience, for example, 80%). If so, the collaborative correction rule is triggered, and the surface matrix attributes of the surrounding adjacent interpreted patches that share the boundary with the interpreted patch are introduced. Distance-weighted voting is performed using the boundary length or center point distance between the interpreted patch and each surrounding adjacent interpreted patch as weights. The surface matrix attributes of the interpreted patch are corrected according to the voting results to obtain the corrected soil type patch set. If the above conditions do not exist, the collaborative correction rule is not triggered, and the interpreted patch retains its original surface matrix attributes.

[0047] Specifically, for each pair of intersecting polygons in the candidate set of intersecting polygon pairs, the intersecting polygons between the interpreted polygon and the base polygon are calculated using a polygon precision clipping algorithm, and the area of ​​these intersecting polygons is calculated as the overlap area. For each interpreted polygon, the overlap areas between the interpreted polygon and all intersecting base polygons are summed to obtain the total overlap area of ​​the interpreted polygon; the spatial dominance ratio is obtained by dividing the total overlap area by the total area of ​​the interpreted polygon itself.

[0048] The collaborative correction rule is a voting adjustment rule based on the surface matrix properties of neighboring interpreted map pieces. Distance-weighted voting is a decision-making method that determines weights based on proximity and counts the votes for various surface matrix types. Spatial dominance ratio is the proportion of the total overlap area between the interpreted map piece and the base map piece to the total area of ​​the interpreted map piece itself.

[0049] For example, for each interpreted patch that does not trigger the absolute inheritance rule, the land use attributes of its corresponding intersecting base patches are extracted, and the total overlap area between the interpreted patch and each intersecting base patch is divided by the total area of ​​the interpreted patch itself to obtain the spatial dominance ratio. When the land use attribute is "dry land", "forest land", "paddy field" or "inland tidal flat", and the spatial dominance ratio is greater than 80%, the collaborative correction rule is triggered. The surface matrix attributes of the surrounding adjacent interpreted patches that share the boundary with the interpreted patch are extracted, and the weighted votes for each matrix type are calculated using the boundary length as the weight. The surface matrix attribute with the highest vote is taken as the corrected surface matrix attribute of the interpreted patch, resulting in the corrected soil type patch set.

[0050] Specifically, if the surface matrix attribute of the interpreted map patch is "loam," and the land use attribute of the base map patch is "paddy field" with a spatial dominance greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpreted map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "loamy clay," and the surface matrix attribute of the interpreted map patch is corrected to "loamy clay." If the land use attribute of the base map patch is "dry land" with a spatial dominance greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpreted map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "loam," and the surface matrix attribute of the interpreted map patch is corrected to "loam." If the land use type of the base map patch is "forest" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpretable map patches sharing the same boundary are introduced, and a vote is conducted using boundary length as the weight. The surface matrix attribute with the most votes is "loam," and the surface matrix attribute of the interpretable map patch is corrected to "loam." If the land use type of the base map patch is "inland tidal flat" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpretable map patches sharing the same boundary are introduced, and a vote is conducted using boundary length as the weight. The surface matrix attribute with the most votes is "sandy loam," and the surface matrix attribute of the interpretable map patch is corrected to "sandy loam."

[0051] If the surface matrix attribute of the interpreted map patch is "sandy soil," and the land use attribute of the base map patch is "paddy field" with a spatial dominance greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpreted map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "sandy loam," and the surface matrix attribute of the interpreted map patch is corrected to "sandy loam." If the land use attribute of the base map patch is "dry land" with a spatial dominance greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpreted map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "sandy soil," and the surface matrix attribute of the interpreted map patch is corrected to "sandy soil." If the land use type of the base map patch is "forest" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpretable map patches sharing the same boundary are introduced, and a vote is conducted using boundary length as the weight. The surface matrix attribute with the most votes is "loamy sandy soil," and the surface matrix attribute of the interpretable map patch is corrected to "loamy sandy soil." If the land use type of the base map patch is "inland tidal flat" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpretable map patches sharing the same boundary are introduced, and a vote is conducted using boundary length as the weight. The surface matrix attribute with the most votes is "sandy clay," and the surface matrix attribute of the interpretable map patch is corrected to "sandy clay."

[0052] If the surface matrix attribute of the interpreted map patch is "clay," and the land use attribute of the base map patch is "paddy field" with a spatial dominance greater than 80%, a collaborative correction rule is triggered. The matrix attributes of adjacent interpreted map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "clay," and the surface matrix attribute of the interpreted map patch is corrected to "clay." If the land use attribute of the base map patch is "dry land" with a spatial dominance greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpreted map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "loamy clay," and the surface matrix attribute of the interpreted map patch is corrected to "loamy clay." If the land use type of the base map patch is "Forest Land" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpretable map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "Loamy Clay," and the surface matrix attribute of the interpretable map patch is corrected to "Loamy Clay." If the land use type of the base map patch is "Inland Tidal Flat" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attributes of adjacent interpretable map patches sharing the same boundary are introduced, and a voting system is conducted with boundary length as the weight. The surface matrix attribute with the most votes is "Loamy Clay," and the surface matrix attribute of the interpretable map patch is corrected to "Loamy Clay." If the land use type of the base map patch is "swamp" and its spatial dominance is greater than 80%, a collaborative correction rule is triggered. The surface matrix attribute of the neighboring map patches that share the boundary with the interpreted map patch is introduced, and a vote is conducted with the boundary length as the weight. The surface matrix attribute with the most votes is "silt", and the surface matrix attribute of the interpreted map patch is corrected to "silt".

[0053] S108: Generate a distribution map of surface matrix soil types based on the corrected soil type patch set.

[0054] In this embodiment, the corrected soil type patch set is first subjected to topological checks and attribute normalization to ensure that the patches are free of overlap and gaps, and that the attribute fields are complete and standardized. Then, the surface matrix soil types are read from the corrected soil type patch set, including soil types (such as sandy soil, clay, and loam) and non-soil types (such as construction land), and differentiated rendering styles are set for each to clearly distinguish different types on the map. Subsequently, geographic basic elements are overlaid, and cartographic elements such as map title, legend, scale, and coordinate system are configured to complete the map finishing. Finally, a surface matrix soil type distribution map conforming to spatial data standards is output, which can be used for practical work such as land resource management, ecological monitoring, desertification control, and national spatial planning.

[0055] As can be seen from the above, the embodiments of this application adopt a collaborative processing mode of multi-source remote sensing data and land use data from the Third National Land Survey. By generating orthorectified seamless mosaic images and achieving spatial alignment with the land use data from the Third National Land Survey, it ensures that remote sensing information and geographical prior knowledge maintain a unified benchmark, avoiding identification deviations caused by data mismatch. This reduces the workload of manual processing and repeated verification, improves the identification efficiency of surface matrix soil types over a large area, and meets the needs of rapid surveys. The embodiments of this application obtain image objects by image segmentation of orthorectified seamless mosaic images, extracts multi-dimensional feature vectors for each image object, and uses a pre-trained multi-level rule decision tree model to perform hierarchical discrimination and classification of each image object. This can objectively and stably extract the essential features of the surface matrix, reduce the subjectivity and uncertainty caused by manual interpretation, effectively alleviate the problems of easy interference and unstable classification in traditional remote sensing identification, and improve the accuracy of preliminary surface matrix type judgment. Meanwhile, this application embodiment corrects the surface matrix properties by triggering absolute inheritance rules or collaborative correction rules, making the boundaries of the corrected soil type patch set more closely match the boundaries of actual land features, thus solving the defects of vague classification results and inconsistencies with the actual situation. The final generated surface matrix soil type distribution map is more realistic and usable, providing reliable data support for land resource management, ecological monitoring, and other work. Therefore, this application embodiment can improve the efficiency and accuracy of remote sensing identification of surface matrix soil types.

[0056] In one embodiment of this application, the trained multi-level rule decision tree model is obtained through training in the following manner: The historical measured surface matrix type set and historical orthorectified seamless mosaic image were acquired in the field. The points in the historical measured surface matrix type set were spatially anchored with the historical orthorectified seamless mosaic image to construct multiple regions of interest and determine the measured surface matrix type corresponding to each region of interest, which served as the training sample label set. The historical measured surface matrix type set and the historical orthorectified seamless mosaic image have the same radiometric and spatial reference. For each region of interest, a multi-dimensional feature vector is extracted from the region of interest covered by the historical orthophoto seamless mosaic image as a training sample feature set. The training sample feature set and training sample label set are input into the decision tree model for training to obtain the initial multi-level rule decision tree model; The accuracy evaluation operation is performed multiple times until the preset stopping condition is met, resulting in a trained multi-level rule decision tree model. The steps involved in the accuracy evaluation operation include: Based on the current multi-level rule decision tree model, image segmentation, hierarchical discrimination and classification, vector patch conversion and fusion correction steps are performed on historical orthophoto seamless mosaic images to obtain the corrected historical surface matrix type patch set; The overall accuracy and Kappa coefficient were calculated based on the historical measured surface matrix type set and the corrected historical surface matrix type patch set. If the overall accuracy is less than the preset accuracy threshold, or the Kappa coefficient is less than the preset coefficient threshold, the current multi-level rule decision tree model will be retrained to obtain the updated multi-level rule decision tree model. The preset stopping condition is that the overall accuracy is greater than or equal to the preset accuracy threshold, and the Kappa coefficient is greater than or equal to the preset coefficient threshold.

[0057] In this embodiment, the historical measured surface matrix type set is a collection of surface matrix types collected in the field. The historical orthophoto seamless mosaic image is the previously processed global image data. Spatial anchoring is the operation of precisely matching the measured points with the image; specifically, it involves mapping the GPS coordinates collected in the field to the pixel coordinates of the image through coordinate transformation, so that the measured type and the image pixels correspond precisely in space. The region of interest is the sample selection spatial range (e.g., 3×3 pixels) constructed with the anchor point as the center. The training sample label set is the annotation of the real matrix type corresponding to each region of interest. The training sample feature set is the multidimensional feature data corresponding to each region of interest.

[0058] For example, historical measured surface matrix types and historical orthorectified seamless mosaic images are acquired. Spatial anchoring is performed between the historical measured locations and the historical orthorectified seamless mosaic images to construct multiple regions of interest (ROIs). The measured surface matrix types corresponding to each ROI are used as the training sample label set. For each ROI, multi-dimensional feature vectors are extracted from the ROIs covered by the historical orthorectified seamless mosaic images to form a training sample feature set. The training sample feature set and the training sample label set are input into a decision tree model for training, resulting in the current multi-level rule-based decision tree model.

[0059] In this embodiment, the accuracy evaluation operation is a process for verifying the classification performance of the current multi-level rule decision tree model. Overall accuracy is the degree of consistency between the classification results and the measured data (the proportion of correctly classified samples to the total number of samples). The Kappa coefficient is a consistency evaluation index of the classification results (the degree of consistency after excluding the influence of random consistency). The preset accuracy threshold is the minimum accuracy standard for a qualified model (e.g., 90%). The preset coefficient threshold is the minimum Kappa standard for a qualified model (e.g., 0.85). The preset stopping condition is the condition for determining whether model training is complete, i.e., the overall accuracy is greater than or equal to the preset accuracy threshold and the Kappa coefficient is greater than or equal to the preset coefficient threshold.

[0060] This embodiment utilizes a cyclical accuracy evaluation and iterative retraining mechanism to continuously optimize the multi-level rule-based decision tree model. Overall accuracy and the Kappa coefficient are used as quantization stopping conditions to ensure the model reaches a preset standard before being put into practical application.

[0061] For example, image segmentation is performed on historical orthophoto mosaic images to obtain historical image objects. Historical multidimensional feature vectors are extracted and input into the current multi-level rule decision tree model (initial model) to obtain the historical preliminary land matrix type corresponding to each historical image object. Each historical preliminary land matrix type is converted into vector patches to obtain a set of historical remote sensing preliminary interpreted patch vectors. Fusion correction is performed based on historical land use data from the third national land survey to obtain a corrected set of historical land matrix type patches. Overall accuracy and Kappa coefficient are calculated based on the historical measured land matrix type set and the corrected historical land matrix type patch set. When the overall accuracy is greater than or equal to a preset accuracy threshold (e.g., 90%) and the Kappa coefficient is greater than or equal to a preset coefficient threshold (e.g., 0.85), training stops, and a trained multi-level rule decision tree model is obtained. If the conditions are not met, the current multi-level rule decision tree model is retrained to obtain an updated multi-level rule decision tree model, and the accuracy evaluation process is returned until the preset stopping conditions are met.

[0062] In this embodiment, the historical measured surface matrix type set includes multiple historical measured surface matrix types; Based on the current multi-level rule-based decision tree model, image segmentation, hierarchical classification, vector patch conversion, and fusion correction steps are performed on historical orthophoto seamless mosaic images to obtain a corrected historical surface matrix type patch set, including: Image segmentation is performed on historical orthophoto seamless mosaic images to obtain multiple historical image objects. The historical multidimensional feature vector of each historical image object is extracted. The historical multidimensional feature vectors are input into the current multi-level rule decision tree model to classify each historical image object and obtain the historical preliminary surface matrix type corresponding to each historical image object. Each historical preliminary surface matrix type is converted into a vector map patch, resulting in a preliminary historical remote sensing interpretation patch vector set; Acquire historical land use data from the Third National Land Survey and establish a priori mapping relationship library between the land use attributes of historical land use data and various historical measured land surface matrix types. Based on the prior mapping relation library, the land use attributes of historical land use data from the Third National Land Survey are used to trigger absolute inheritance rules or collaborative correction rules to correct the vector set of historical remote sensing preliminary interpretation patches, resulting in a corrected historical surface matrix type patch set; the historical land use data from the Third National Land Survey are spatially aligned with the vector set of historical remote sensing preliminary interpretation patches.

[0063] In this embodiment, the distribution frequency and confidence level of each surface matrix type under each land use attribute are statistically analyzed to establish a priori mapping relationship database between land use attributes of historical land survey data and each historical measured surface matrix type. The priori mapping relationship database records the surface matrix types and their confidence levels corresponding to different land use attributes (such as paddy fields, dry land, forest land, water areas, paved roads, and building land). If the land use attribute of the base map patch is water area, paved road, or building land, the absolute inheritance rule is triggered, and the surface matrix attribute of the corresponding interpreted map patch is forcibly corrected to a non-soil or artificial matrix type. If the land use attribute is dry land or forest land, and the spatial dominance ratio is greater than a preset threshold (e.g., 80%), the collaborative correction rule is triggered, and the surface matrix attributes of surrounding adjacent interpreted map patches are introduced for distance-weighted voting correction.

[0064] This embodiment establishes a priori mapping relationship database between land use attributes of historical land use data from the Third National Land Survey and various historical measured land surface matrix types. This database provides a basis for subsequent correction processes of historical remote sensing preliminary interpretation vector sets, ensuring the objectivity and reliability of the correction results. Furthermore, the priori mapping relationship database can be continuously updated and improved as historical measured land surface matrix type sets accumulate, thereby continuously enhancing the correction capabilities based on the Third National Land Survey land use data.

[0065] In this embodiment, the overall accuracy and Kappa coefficient are calculated based on the historical measured surface matrix type set and the corrected historical surface matrix type patch set, including: Based on the spatial coordinates of each point in the historical measured land matrix type set, a spatial overlay analysis is performed with the corrected historical land matrix type patch set to obtain the predicted land matrix type for each point. Based on the predicted surface matrix type and the corresponding historical measured surface matrix type at each location, the comparison is performed point by point, the number of samples for each category combination is counted, and a confusion matrix is ​​constructed. Based on the confusion matrix, the overall accuracy is obtained by calculating the ratio of the sum of the diagonal elements to the total number of validation points; and the expected accuracy of random consistency is obtained based on the confusion matrix. The Kappa coefficient is calculated based on the overall accuracy and the random consistent expected accuracy.

[0066] In this embodiment, spatial coordinates represent the geographical location of a point. Spatial overlay analysis is an analysis operation that spatially matches points with map features. Predicted surface matrix type is the output type corresponding to the point. Point-by-point comparison is the operation of comparing the predicted surface matrix type with its corresponding historical measured surface matrix type one by one. Category combination is the pairing result of the predicted surface matrix type and its corresponding historical measured surface matrix type. Confusion matrix is ​​a square matrix table showing the classification correctness and incorrectness. Diagonal elements are samples in the matrix whose predicted surface matrix type and corresponding historical measured surface matrix type are consistent. Total number of validation points is the total number of samples participating in the accuracy test. Overall accuracy is the global accuracy of the classification results. Random consistency expected accuracy is the expected consistency probability under random classification conditions. Kappa coefficient is the classification consistency evaluation index after eliminating the influence of random consistency.

[0067] For example, the spatial coordinates of each point in the historical measured surface matrix type set are extracted. Spatial overlay analysis is performed on these points and the corrected historical surface matrix type patch set to determine the predicted surface matrix type corresponding to each point. The predicted surface matrix type of each point is compared point-by-point with the corresponding historical measured surface matrix type. The number of samples is counted according to the combination of measured and predicted types to construct a complete confusion matrix. The elements on the diagonal of the confusion matrix are summed, and the sum is divided by the total number of validation points to obtain the overall accuracy. The random consensus expected accuracy is calculated based on the row and column sums in the confusion matrix and the total number of samples. The overall accuracy and the random consensus expected accuracy are substituted into the calculation process to obtain the Kappa coefficient. After the calculation is completed, the overall accuracy and the Kappa coefficient are used simultaneously as the criteria for judging the accuracy of the current multi-level rule decision tree model, to determine whether the preset stopping conditions are met or whether the current multi-level rule decision tree model needs to be retrained.

[0068] This embodiment uses a historically measured set of surface matrix types as a benchmark for accuracy evaluation, ensuring the results are accurate and reliable. The confusion matrix visually reflects the misclassification of each category, facilitating the identification of error sources. Overall accuracy is used in conjunction with the Kappa coefficient to comprehensively and objectively evaluate the classification effect. The high-precision evaluation results provide accurate direction for the iterative optimization of the current multi-level rule-based decision tree model, effectively improving the reliability and stability of the current multi-level rule-based decision tree model, and providing accuracy assurance for achieving efficient and accurate identification of surface matrix soil types.

[0069] In this embodiment, the corrected historical surface matrix type atlas includes multiple historically corrected interpretation patches; The current multi-level rule decision tree model is retrained to obtain an updated multi-level rule decision tree model, including: The historical corrected interpretation patches corresponding to the overall accuracy being less than the preset accuracy threshold or the Kappa coefficient being less than the preset coefficient threshold are marked as difficult sample samples, thus obtaining the difficult sample sample set; Increase the sample weight of each difficult example to obtain a weighted sample set; Based on the weighted sample set, the current multi-level rule decision tree model is retrained, and the model parameters are adjusted to increase the classification accuracy of hard examples, resulting in an updated multi-level rule decision tree model.

[0070] In this embodiment, the historically corrected interpretation patch is a historical classification spatial unit that has undergone accuracy verification and attribute correction. Difficult examples are sample data that the multi-level rule decision tree model struggles to distinguish correctly; that is, samples that the current multi-level rule decision tree model misclassifies or has low classification confidence. The difficult example set is the collection of all difficult examples. Sample weights are weighting coefficients reflecting the difficulty or importance of sample classification. The weighted sample set is the training dataset formed by assigning higher weights to difficult examples.

[0071] This embodiment addresses the issue of insufficient accuracy by retraining a multi-level rule-based decision tree model. The core focus is on improving classification ability by concentrating on error-prone, difficult-to-distinguish samples. By assigning higher weights to each difficult sample, the current multi-level rule-based decision tree model pays closer attention to these hard-to-distinguish samples during training, thereby strengthening the learning of differences in easily confused features. Retraining does not start from scratch; instead, it optimizes parameters using a weighted sample set, retaining existing effective features while focusing on improving weaker aspects. Independent validation with a difficult sample set ensures the model improvement is targeted. Iterative iteration continues until a stopping condition is met, forming a closed-loop optimization mechanism that continuously improves the overall accuracy of the current multi-level rule-based decision tree model, guaranteeing stability and accuracy in real-world recognition scenarios.

[0072] For example, this embodiment extracts historical corrected interpretation patches corresponding to cases where the overall accuracy is less than a preset accuracy threshold or the Kappa coefficient is less than a preset coefficient threshold. These patches are uniformly marked as difficult examples and combined to form a difficult example sample set. Each difficult example sample is assigned a sample weight greater than that of regular samples (e.g., the weight value is set to 2 to 10 times). The difficult examples and their weights are weighted to obtain a weighted sample set. Based on the weighted sample set, the current multi-level rule decision tree model is retrained, and the parameters of the current multi-level rule decision tree model are adjusted to increase the classification accuracy of difficult examples, resulting in an updated multi-level rule decision tree model. The updated current multi-level rule decision tree model is independently validated using the difficult example sample set, and spatial overlay analysis and point-by-point comparison are performed to calculate the overall accuracy and Kappa coefficient after validation. If the validation results still do not meet the preset stopping conditions (e.g., overall accuracy < 90% or Kappa coefficient < 0.85), the retraining and validation steps are repeated, and iterative optimization continues until both the overall accuracy and Kappa coefficient reach the preset standards.

[0073] This embodiment focuses on targeted optimization of difficult examples, precisely addressing the weaknesses in current multi-level rule-based decision tree models. The combination of sample weight adjustment and iterative retraining efficiently improves the accuracy of easily confused types, while remaining fully compatible with the underlying principles of multi-level rule-based decision tree models.

[0074] Corresponding to the remote sensing identification method for surface matrix soil type in the above embodiment, Figure 2 This is a structural block diagram of a remote sensing identification device for surface matrix soil type according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The remote sensing identification device 20 for surface matrix soil type includes: a data preprocessing module 21, a preliminary surface matrix type identification module 22, a remote sensing preliminary interpretation patch generation module 23, a land survey acquisition module 24, a rule triggering module 25, an absolute inheritance rule module 26, a collaborative correction rule module 27, and a surface matrix soil type distribution map generation module 28.

[0075] Among them, the data preprocessing module 21 is used to obtain orthorectified seamless mosaic images based on multi-source remote sensing data of the target area; The preliminary surface matrix type identification module 22 is used to perform image segmentation on the orthophoto seamless mosaic image to obtain multiple image objects and extract the multi-dimensional feature vector of each image object; input each multi-dimensional feature vector into the trained multi-level rule decision tree model to perform hierarchical discrimination and classification on each image object to obtain the preliminary surface matrix type corresponding to each image object; The remote sensing preliminary interpretation patch generation module 23 is used to convert each preliminary surface matrix type into vector patches to obtain a remote sensing preliminary interpretation patch vector set; the remote sensing preliminary interpretation patch vector set includes multiple interpretation patches; The land use data acquisition module 24 of the Third National Land Survey is used to acquire land use data of the target area from the Third National Land Survey. The land use data of the Third National Land Survey includes multiple base map patches. The land use data of the Third National Land Survey is spatially aligned with the orthophoto seamless mosaic image. The rule triggering module 25 is used to determine the intersecting base plate patches corresponding to each interpreted patch, and to determine whether to trigger the absolute inheritance rule based on the land use attributes of the intersecting base plate patches corresponding to each interpreted patch. Absolute inheritance rule module 26 is used to directly correct the surface matrix properties of the interpreted patch if the absolute inheritance rule is triggered, so as to obtain the corrected soil type patch set. The collaborative correction rule module 27 is used to determine whether to trigger the collaborative correction rule based on the land type attribute if the absolute inheritance rule is not triggered. If the collaborative correction rule is triggered, the surface matrix attribute of the interpreted patch is corrected based on distance-weighted voting to obtain the corrected soil type patch set. The surface matrix soil type distribution map generation module 28 is used to generate a surface matrix soil type distribution map based on the corrected soil type patch set.

[0076] In one embodiment of this application, the data preprocessing module 21 is specifically used for: Multiple apparent radiance images were obtained by radiometric calibration of multi-source remote sensing data; Atmospheric correction was performed on each apparent radiance image to obtain multiple surface reflectance images; Geometric and orthorectification are performed on surface reflectance images from various locations to obtain multiple orthorectified images; each orthorectified image includes a high-resolution orthorectified image and a medium-resolution orthorectified image; the spatial resolution of the high-resolution orthorectified image is less than or equal to a first spatial resolution threshold; the spatial resolution of the medium-resolution orthorectified image is greater than the first spatial resolution threshold and less than or equal to a second spatial resolution threshold, and the first spatial resolution threshold is less than the second spatial resolution threshold. Image fusion is performed on the panchromatic and multispectral bands of each high-resolution orthophoto to obtain the corresponding high-resolution fused image. Raster registration and resampling were performed on each high-resolution fused image and each medium-resolution orthophoto to obtain multiple registered images with consistent spatial reference. By mosaicking multiple registered images with consistent spatial references, an orthorectified seamless mosaic image is obtained.

[0077] In one embodiment of this application, the rule triggering module 25 is specifically used for: Determine the bounding rectangle of each interpreted feature and the bounding rectangle of each base plate feature respectively; Determine whether there is coordinate overlap between the bounding rectangle of each interpreted feature and the bounding rectangle of each base feature; Remove patch pairs that do not have coordinate overlap to obtain a candidate set of intersecting patch pairs; the candidate set of intersecting patch pairs includes multiple patch pairs, each patch pair includes an interpreted patch and a corresponding intersecting base plate patch.

[0078] In one embodiment of this application, the remote sensing identification device 20 for surface matrix soil type further includes: a model training module; When the model training module obtains a pre-trained multi-level rule decision tree model, it is specifically used for: The historical measured surface matrix type set and historical orthorectified seamless mosaic image were acquired in the field. The points in the historical measured surface matrix type set were spatially anchored with the historical orthorectified seamless mosaic image to construct multiple regions of interest and determine the measured surface matrix type corresponding to each region of interest, which served as the training sample label set. The historical measured surface matrix type set and the historical orthorectified seamless mosaic image have the same radiometric and spatial reference. For each region of interest, a multi-dimensional feature vector is extracted from the region of interest covered by the historical orthophoto seamless mosaic image as a training sample feature set. The training sample feature set and training sample label set are input into the decision tree model for training to obtain the current multi-level rule decision tree model. The accuracy evaluation operation is performed multiple times until the preset stopping condition is met, resulting in a trained multi-level rule decision tree model. The steps involved in the accuracy evaluation operation include: Based on the current multi-level rule decision tree model, image segmentation, hierarchical discrimination and classification, vector patch conversion and fusion correction steps are performed on historical orthophoto seamless mosaic images to obtain the corrected historical surface matrix type patch set; The overall accuracy and Kappa coefficient were calculated based on the historical measured surface matrix type set and the corrected historical surface matrix type patch set. If the overall accuracy is less than the preset accuracy threshold, or the Kappa coefficient is less than the preset coefficient threshold, the current multi-level rule decision tree model will be retrained to obtain the updated multi-level rule decision tree model. The preset stopping condition is that the overall accuracy is greater than or equal to the preset accuracy threshold, and the Kappa coefficient is greater than or equal to the preset coefficient threshold.

[0079] In one embodiment of this application, the historical measured surface matrix type set includes multiple historical measured surface matrix types; the model training module is specifically used for: Image segmentation is performed on historical orthophoto seamless mosaic images to obtain multiple historical image objects. The historical multidimensional feature vector of each historical image object is extracted. The historical multidimensional feature vectors are input into the current multi-level rule decision tree model to classify each historical image object and obtain the historical preliminary surface matrix type corresponding to each historical image object. Each historical preliminary surface matrix type is converted into a vector map patch, resulting in a preliminary historical remote sensing interpretation patch vector set; Acquire historical land use data from the Third National Land Survey and establish a priori mapping relationship library between the land use attributes of historical land use data and various historical measured land surface matrix types. Based on the prior mapping relation library, the land use attributes of historical land use data from the Third National Land Survey are used to trigger absolute inheritance rules or collaborative correction rules to correct the vector set of historical remote sensing preliminary interpretation patches, resulting in a corrected historical surface matrix type patch set; the historical land use data from the Third National Land Survey are spatially aligned with the vector set of historical remote sensing preliminary interpretation patches.

[0080] In one embodiment of this application, the model training module is specifically used for: Based on the spatial coordinates of each point in the historical measured land matrix type set, a spatial overlay analysis is performed with the corrected historical land matrix type patch set to obtain the predicted land matrix type for each point. Based on the predicted surface matrix type and the corresponding historical measured surface matrix type at each location, the comparison is performed point by point, the number of samples for each category combination is counted, and a confusion matrix is ​​constructed. Based on the confusion matrix, the overall accuracy is obtained by calculating the ratio of the sum of the diagonal elements to the total number of validation points; and the expected accuracy of random consistency is obtained based on the confusion matrix. The Kappa coefficient is calculated based on the overall accuracy and the random consistent expected accuracy.

[0081] In one embodiment of this application, the corrected historical surface matrix type patch set includes multiple historically corrected interpretation patches; the model training module is specifically used for: The historical corrected interpretation patches corresponding to the overall accuracy being less than the preset accuracy threshold or the Kappa coefficient being less than the preset coefficient threshold are marked as difficult sample samples, thus obtaining the difficult sample sample set; Increase the sample weight of each difficult example to obtain a weighted sample set; Based on the weighted sample set, the current multi-level rule decision tree model is retrained, and the model parameters are adjusted to increase the classification accuracy of hard examples, resulting in an updated multi-level rule decision tree model.

[0082] It should be noted that the specific limitations of the above-described embodiments of the remote sensing identification device 20 for surface matrix soil type can be found in the limitations of the remote sensing identification method for surface matrix soil type described above, and will not be repeated here. Each module of the above-described device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in the processor of the electronic device in hardware form or independent of the processor, or it can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0083] This application provides an electronic device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the method provided in any optional embodiment of this application.

[0084] In one alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3 The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may further include a transceiver 304, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this electronic device 300 does not constitute a limitation on the embodiments of this application.

[0085] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0086] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0087] The memory 303 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0088] The memory 303 is used to store computer programs that execute the embodiments of this application, and the execution is controlled by the processor 301. The processor 301 is used to execute the computer programs stored in the memory 303 to implement the steps shown in the foregoing method embodiments.

[0089] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one computer program, which is loaded and executed by a processor of a computer device to enable the computer to implement any of the above-described remote sensing identification methods for surface matrix soil types.

[0090] In one possible implementation, the aforementioned computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a solid-state drive (SSD), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc. The random access memory can include resistive random access memory (ReRAM) and dynamic random access memory (DRAM).

[0091] In an exemplary embodiment, a computer program or computer program product is also provided, which includes computer instructions loaded and executed by a processor to enable a computer to implement any of the above-described remote sensing identification methods for surface substrate soil types.

[0092] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0093] In other words, the data collection and processing in this application should strictly comply with the requirements of relevant national laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0094] It should be further noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The implementation methods described in the above exemplary embodiments do not represent all implementation methods consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0095] It should be understood that "multiple" as used in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0096] Furthermore, the step numbers described herein are merely illustrative of one possible execution order between steps. In some other embodiments, the steps may not be executed in the order of their numbers, such as two steps with different numbers being executed simultaneously, or two steps with different numbers being executed in the reverse order of the illustration. This application does not limit this.

[0097] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. Optionally, the program is stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0098] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A remote sensing method for identifying surface matrix soil types, characterized in that, include: Orthorectified seamless mosaic imagery is obtained based on multi-source remote sensing data of the target area; The orthophoto seamless mosaic image is segmented to obtain multiple image objects, and the multidimensional feature vector of each image object is extracted. The multidimensional feature vectors are input into a trained multi-level rule decision tree model to perform hierarchical discrimination and classification on each image object to obtain the preliminary surface matrix type corresponding to each image object. Each of the preliminary surface matrix types is converted into vector plots to obtain a preliminary remote sensing interpretation vector set; the preliminary remote sensing interpretation vector set includes multiple interpretation plots. Acquire land use data from the third national land survey of the target area, the land use data from the third national land survey includes multiple base map patches; the land use data from the third national land survey is spatially aligned with the orthophoto seamless mosaic image. Determine the corresponding intersecting base map features for each interpreted map feature, and determine whether the absolute inheritance rule is triggered based on the land use attributes of the corresponding intersecting base map features. If the absolute inheritance rule is triggered, the surface matrix properties of the interpreted patch will be directly corrected to obtain the corrected soil type patch set; If the absolute inheritance rule is not triggered, then the collaborative correction rule is determined based on the land type attribute. If the collaborative correction rule is triggered, then the surface matrix attribute of the interpreted map patch is corrected based on distance-weighted voting to obtain the corrected soil type map patch set. A distribution map of surface matrix soil types is generated based on the modified soil type patch set.

2. The remote sensing identification method for surface matrix soil type as described in claim 1, characterized in that, The orthorectified seamless mosaic image obtained from the multi-source remote sensing data based on the target area includes: Radiometric calibration of the multi-source remote sensing data yields multiple apparent radiance images; Atmospheric correction was performed on each of the aforementioned apparent radiance images to obtain multiple surface reflectance images; Geometric correction and orthorectification are performed on each of the surface reflectance images to obtain multiple orthorectified images; each orthorectified image includes a high-resolution orthorectified image and a medium-resolution orthorectified image; the spatial resolution of the high-resolution orthorectified image is less than or equal to a first spatial resolution threshold; the spatial resolution of the medium-resolution orthorectified image is greater than the first spatial resolution threshold and less than or equal to a second spatial resolution threshold, wherein the first spatial resolution threshold is less than the second spatial resolution threshold. Image fusion is performed on the panchromatic band and multispectral band of each high-resolution orthophoto to obtain the corresponding high-resolution fused image. Raster registration and resampling are performed on each of the high-resolution fused images and each of the medium-resolution orthophotos to obtain multiple registered images with consistent spatial references. The registered images with consistent spatial references are mosaicked to obtain the orthorectified seamless mosaic image.

3. The remote sensing identification method for surface matrix soil type as described in claim 1, characterized in that, The process of determining the intersecting base plate patches corresponding to each interpreted patch includes: Determine the outer rectangle of each interpreted patch and the outer rectangle of each base plate patch respectively; Determine whether there is coordinate overlap between the outer rectangle of each interpreted patch and the outer rectangle of each base plate patch; By filtering out patch pairs that do not have coordinate overlap, a set of candidate intersecting patch pairs is obtained; the set of candidate intersecting patch pairs includes multiple patch pairs, each patch pair including an interpreted patch and a corresponding intersecting base plate patch.

4. The remote sensing identification method for surface matrix soil type as described in claim 1, characterized in that, The trained multi-level rule decision tree model was obtained through training in the following manner: A set of historical measured surface matrix types and historical orthorectified seamless mosaic images are acquired in the field. The points in the set of historical measured surface matrix types are spatially anchored to the historical orthorectified seamless mosaic images to construct multiple regions of interest. The measured surface matrix types corresponding to each region of interest are determined as a training sample label set. The set of historical measured surface matrix types and the historical orthorectified seamless mosaic images have the same radiometric and spatial reference. For each region of interest, a multi-dimensional feature vector is extracted from the region of interest covered by the historical orthophoto seamless mosaic image as a training sample feature set; The training sample feature set and the training sample label set are input into the decision tree model for training to obtain the current multi-level rule decision tree model. The accuracy evaluation operation is performed multiple times until the preset stopping condition is met, thus obtaining the trained multi-level rule decision tree model. The steps of the accuracy evaluation operation include: Based on the current multi-level rule decision tree model, image segmentation, hierarchical discrimination and classification, vector patch conversion and fusion correction steps are performed on the historical orthophoto seamless mosaic image to obtain the corrected historical surface matrix type patch set; The overall accuracy and Kappa coefficient were calculated based on the historical measured surface matrix type set and the corrected historical surface matrix type patch set. If the overall accuracy is less than a preset accuracy threshold, or the Kappa coefficient is less than a preset coefficient threshold, then the current multi-level rule decision tree model will be retrained to obtain an updated multi-level rule decision tree model. The preset stopping condition is that the overall accuracy is greater than or equal to a preset accuracy threshold, and the Kappa coefficient is greater than or equal to a preset coefficient threshold.

5. The remote sensing identification method for surface matrix soil type as described in claim 4, characterized in that, The set of historical measured surface matrix types includes multiple historical measured surface matrix types; Based on the current multi-level rule decision tree model, the historical orthophoto mosaic image is subjected to image segmentation, hierarchical discrimination and classification, vector patch conversion and fusion correction steps to obtain a corrected historical surface matrix type patch set, including: Image segmentation is performed on the historical orthophoto seamless mosaic image to obtain multiple historical image objects. The historical multidimensional feature vector of each historical image object is extracted. The historical multidimensional feature vectors are input into the current multi-level rule decision tree model to classify each historical image object and obtain the historical preliminary surface matrix type corresponding to each historical image object. Each of the aforementioned preliminary historical surface matrix types is converted into vector plots to obtain a preliminary historical remote sensing interpretation vector set; Acquire historical land use data from the Third National Land Survey, and establish a priori mapping relationship library between the land use attributes of the historical land use data and each of the historical measured land surface matrix types; Based on the prior mapping relationship library, using the land category attributes of the historical land use data from the third national land survey, the absolute inheritance rule or collaborative correction rule is triggered to correct the vector set of the historical remote sensing preliminary interpretation map patches, resulting in the corrected historical surface matrix type map patch set; the historical land use data from the third national land survey is spatially aligned with the vector set of the historical remote sensing preliminary interpretation map patches.

6. The remote sensing identification method for surface matrix soil type as described in claim 4, characterized in that, The calculation of overall accuracy and Kappa coefficient based on the historical measured surface matrix type set and the corrected historical surface matrix type patch set includes: Based on the spatial coordinates of each point in the historical measured surface matrix type set, a spatial overlay analysis is performed with the corrected historical surface matrix type patch set to obtain the predicted surface matrix type of each point. Based on the predicted surface matrix type and the corresponding historical measured surface matrix type of each point, the comparison is performed point by point, the number of samples for each category combination is counted, and a confusion matrix is ​​constructed. Based on the confusion matrix, the ratio of the sum of the diagonal elements to the total number of verification points is calculated to obtain the overall accuracy; and the random consistency expected accuracy is obtained based on the confusion matrix. The Kappa coefficient is calculated based on the overall accuracy and the random consistent expected accuracy.

7. The remote sensing identification method for surface matrix soil type as described in claim 6, characterized in that, The revised historical surface matrix type atlas includes multiple historical revised interpretation patches; The step of retraining the current multi-level rule decision tree model to obtain an updated multi-level rule decision tree model includes: The historical corrected interpretation patches corresponding to the overall accuracy being less than a preset accuracy threshold or the Kappa coefficient being less than a preset coefficient threshold are marked as difficult sample samples, thus obtaining a difficult sample set; Increase the sample weight of each difficult example to obtain a weighted sample set; Based on the weighted sample set, the current multi-level rule decision tree model is retrained, and the model parameters are adjusted to increase the classification accuracy of hard examples, resulting in an updated multi-level rule decision tree model.

8. A remote sensing identification device for surface matrix soil type, characterized in that, include: The data preprocessing module is used to obtain orthorectified seamless mosaic images based on multi-source remote sensing data of the target area; The preliminary surface matrix type identification module is used to perform image segmentation on the orthophoto seamless mosaic image to obtain multiple image objects, and extract the multi-dimensional feature vector of each image object; input each multi-dimensional feature vector into the trained multi-level rule decision tree model to perform hierarchical discrimination and classification on each image object to obtain the preliminary surface matrix type corresponding to each image object; The remote sensing preliminary interpretation patch generation module is used to convert each of the preliminary surface matrix types into vector patches to obtain a remote sensing preliminary interpretation patch vector set; the remote sensing preliminary interpretation patch vector set includes multiple interpretation patches; The land use data acquisition module is used to acquire land use data from the third national land survey of the target area. The land use data from the third national land survey includes multiple base map patches. The land use data from the third national land survey is spatially aligned with the orthophoto seamless mosaic image. The rule triggering module is used to determine the intersecting base plate patches corresponding to each interpreted patch, and to determine whether to trigger the absolute inheritance rule based on the land use attributes of the intersecting base plate patches corresponding to each interpreted patch; The absolute inheritance rule module is used to directly correct the surface matrix properties of the interpreted patch if the absolute inheritance rule is triggered, so as to obtain the corrected soil type patch set. The collaborative correction rule module is used to determine whether to trigger the collaborative correction rule based on the land type attribute if the absolute inheritance rule is not triggered. If the collaborative correction rule is triggered, the surface matrix attribute of the interpreted patch is corrected based on distance-weighted voting to obtain the corrected soil type patch set. The surface matrix soil type distribution map generation module is used to generate a surface matrix soil type distribution map based on the modified soil type patch set.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing a computer program, which is loaded and executed by the processor to implement the remote sensing identification method for surface matrix soil type as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is loaded and executed by a processor to implement the remote sensing identification method for surface matrix soil type as described in any one of claims 1 to 7.

Images