A real scene three-dimensional remote sensing target recognition method and system fusing spectral, height and texture features

Abstract

This invention relates to the field of remote sensing image processing technology, and discloses a method and system for real-scene 3D remote sensing target recognition that integrates spectral, elevation, and texture features. The invention first acquires registered remote sensing data of the area to be identified, constructs candidate target patches, maps spectral, elevation, and texture information to the same candidate target patch, extracts corresponding features, and determines the initial target type. It then calls target type verification rules to perform confidence verification, and generates target recognition results based on the confidence verification results and feature conflict handling conditions. This invention can solve the problem of lack of typified processing when multiple feature recognition tendencies are inconsistent.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a method and system for real-scene three-dimensional remote sensing target recognition that integrates spectral, elevation and texture features. Background Technology

[0002] Remote sensing target identification is a common technique in natural resource surveys and surface target monitoring. Existing methods typically utilize multispectral remote sensing images to extract spectral features of land features and combine them with texture features from orthophotos to identify surface targets such as buildings, cultivated land, forest land, and bare land. In the area of ​​cultivated land change detection, Chinese invention patent CN111598045B discloses a remote sensing method for cultivated land change detection based on object atlases and mixed spectra. This method constructs object atlases from remote sensing images from multiple years of contemporaneous data and combines mixed spectral analysis, spectral features, geometric features, and texture features to detect cultivated land changes. This type of method can improve the ability to identify land feature changes by utilizing image objects and multiple image features.

[0003] With the application of real-scene 3D data, digital surface models, and oblique photogrammetry results, remote sensing target recognition has begun to incorporate 3D spatial information. Chinese invention patent CN110135354B discloses a change detection method based on a real-scene 3D model. This method utilizes the color and geometric information of the real-scene 3D model for object-oriented segmentation and identifies changed land cover types based on the changed areas. Chinese invention patent CN115641412B discloses a method for generating a 3D semantic map based on hyperspectral data. This method collects hyperspectral data, image texture data, and elevation distribution data, and forms a 3D semantic map based on the land cover classification results. These technologies demonstrate that spectral features, texture features, and elevation features can already be used for remote sensing target recognition and 3D semantic representation.

[0004] However, in the process of real-scene 3D remote sensing target recognition, after candidate target patches are formed, spectral features, elevation features, and texture features may correspond to different recognition tendencies. Existing technologies mostly focus on multi-source data fusion, object segmentation, or target classification, lacking targeted processing for how to confirm, downgrade, or mark as pending verification according to target type when candidate target patches have conflicting verification results based on different features. This can easily affect the stable output of target recognition results. Therefore, it is necessary to provide a remote sensing target recognition technology that can perform credibility verification of spectral features, elevation features, and texture features at the object level of candidate target patches and output target recognition results according to target type constraints. Summary of the Invention

[0005] The purpose of this invention is to provide a real-scene 3D remote sensing target recognition method and system that integrates spectral, elevation and texture features. This method can solve the problem of lack of typified processing when the spectral features, elevation features and texture features of candidate target patches are inconsistent after their formation. It can also realize the confirmation, downgrading and output of target recognition results as labels to be verified.

[0006] To achieve the above objectives, the present invention provides the following solution: A real-scene 3D remote sensing target recognition method that integrates spectral, elevation, and texture features includes: Acquire registration remote sensing data for the region to be identified. The registration remote sensing data includes multispectral remote sensing images, three-dimensional geometric data, and texture image data that are under the same spatial reference. Candidate target patches are constructed based on the registered remote sensing data, and the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data are mapped to the same candidate target patch object. Based on the spectral, elevation, and texture information mapped to the same candidate target patch, the spectral, elevation, and texture features of the candidate target patch are extracted, and the initial target type of the candidate target patch is determined. Invoke the target type verification rules corresponding to the initial target type; the target type verification rules include dominant features, auxiliary features, and feature conflict handling conditions. The credibility of spectral features, elevation features, and texture features is verified according to the target type verification rules to obtain the credibility verification results. Based on the credibility verification results and feature conflict handling conditions, the initial target type is confirmed, downgraded, or marked as pending verification to generate the target recognition results.

[0007] Preferably, candidate target patch objects are constructed based on registered remote sensing data, and the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data are mapped to the same candidate target patch object, including: Based on the registered remote sensing data, the region to be identified is segmented into candidate target patches to obtain the target patch objects; Based on the spatial extent of the candidate target patch, the data range corresponding to the candidate target patch in the multispectral remote sensing image, 3D geometric data, and texture image data is determined respectively. Spectral information is obtained from the data range corresponding to the multispectral remote sensing image, elevation information is obtained from the data range corresponding to the three-dimensional geometric data, and texture information is obtained from the data range corresponding to the texture image data. The spectral information, elevation information, and texture information are then mapped to the same candidate target patch object.

[0008] Preferably, based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, the spectral features, elevation features, and texture features of the candidate target patch object are extracted, including: Based on spectral information, extract the band reflectance characteristics, interspectral difference characteristics, and band combination characteristics of candidate target patches; Extract the object elevation features, neighborhood relative elevation features, and three-dimensional surface undulation features of candidate target patches based on elevation information; Based on texture information, extract the boundary texture features, internal texture uniformity features, and directional texture continuity features of candidate target patches.

[0009] Preferably, determining the initial target type of the candidate target patch object includes: The spectral features, elevation features, and texture features are matched with preset target type feature conditions to obtain spectral matching results, elevation matching results, and texture matching results, respectively. Based on the combination of spectral matching results, elevation matching results, and texture matching results, the initial target type corresponding to the candidate target patch object is determined; Establish the correspondence between candidate target patch objects and initial target types.

[0010] Preferably, the target type verification rule corresponding to the initial target type is invoked, including: The dominant features are determined based on the initial target type. The dominant features are the features that are given priority in confidence verification among spectral features, elevation features, and texture features. Auxiliary features are determined based on the initial target type. These auxiliary features are spectral features, elevation features, and texture features that work in conjunction with the dominant features to participate in the credibility verification. Based on the target type support relationship between the dominant feature and the auxiliary feature, the feature conflict handling conditions are determined. The feature conflict handling conditions include confirmation conditions, downgrade conditions, and conditions to be verified.

[0011] Preferably, the credibility of spectral features, elevation features, and texture features is verified according to the target type verification rules to obtain the credibility verification results, including: Based on the target type verification rules, determine the feature verification conditions corresponding to spectral features, elevation features, and texture features respectively; The spectral features, elevation features, and texture features are matched with their corresponding feature verification conditions to obtain the spectral verification results, elevation verification results, and texture verification results, respectively. Based on the verification results corresponding to the dominant features and the verification results corresponding to the auxiliary features, the credibility verification results of the candidate target patch objects are generated. When the support relationships between the spectral verification results, elevation verification results, and texture verification results for the initial target type are inconsistent, it is determined that there is a feature conflict in the credibility verification results.

[0012] Preferably, based on the credibility verification results and feature conflict handling conditions, the initial target type is confirmed, downgraded, or marked as pending verification, including: When both the verification results corresponding to the dominant feature and the verification results corresponding to the auxiliary feature support the initial target type, the initial target type is confirmed as the confirmation condition is met. If the verification result corresponding to the dominant feature supports the initial target type, but the verification result corresponding to the auxiliary feature does not support the initial target type and does not support different target types, the degradation condition is determined to be met, and the initial target type is downgraded. If the verification result corresponding to the dominant feature does not support the initial target type, or if the verification result corresponding to the dominant feature and the verification result corresponding to the auxiliary feature support different target types, the initial target type is marked as pending verification.

[0013] Preferably, the dominant and auxiliary features are determined according to the initial target type, including: When the initial target type is a suspected constructed target, elevation features and texture features are determined as the dominant features, and spectral features are determined as the auxiliary features; When the initial target type is a suspected farmland disturbance target, spectral features and texture features are determined as the dominant features, and elevation features are determined as the auxiliary features; When the initial target type is a suspected forest disturbance target, spectral features and elevation features are determined as the dominant features, and texture features are determined as the auxiliary features.

[0014] Preferably, generating target recognition results includes: Based on the initial target type and the results of confirming, downgrading or marking the initial target type as pending verification, determine the target type and review status of the candidate target patch objects; Based on the spatial extent of the candidate target patch in the registered remote sensing data, determine the spatial location and three-dimensional outline extent of the candidate target patch; Target recognition results are generated based on target type, verification status, spatial location, and three-dimensional contour range.

[0015] A real-scene 3D remote sensing target recognition system that integrates spectral, elevation, and texture features includes: The data acquisition module is used to acquire the registration remote sensing data of the area to be identified. The registration remote sensing data includes multispectral remote sensing images, three-dimensional geometric data and texture image data under the same spatial reference. The patch construction module is used to construct candidate target patch objects based on the registered remote sensing data, and to map the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data to the same candidate target patch object. The feature determination module is used to extract the spectral features, elevation features, and texture features of candidate target patches based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, and to determine the initial target type of the candidate target patch object; The rule invocation module is used to invoke the target type verification rule corresponding to the initial target type. The target type verification rule includes the dominant feature, auxiliary feature, and feature conflict handling conditions. The target generation module is used to verify the credibility of spectral features, elevation features, and texture features according to the target type verification rules, obtain the credibility verification results, and confirm, downgrade, or mark the initial target type as pending verification based on the credibility verification results and feature conflict handling conditions, thereby generating target recognition results.

[0016] The present invention discloses the following beneficial effects: (1) This invention acquires registered remote sensing data of the area to be identified and limits the registered remote sensing data to include multispectral remote sensing images, three-dimensional geometric data and texture image data under the same spatial reference, so that spectral information, elevation information and texture information participate in subsequent processing under the same spatial reference. This can avoid the problem that the corresponding range of the same surface target is inconsistent in different data sources when multiple source remote sensing data participate in identification independently, and provide a unified data basis for multi-feature judgment on candidate target patches.

[0017] (2) This invention constructs candidate target patch objects based on registered remote sensing data and maps the spectral information corresponding to multispectral remote sensing images, the elevation information corresponding to three-dimensional geometric data, and the texture information corresponding to texture image data to the same candidate target patch object. This transforms the judgment unit for target identification from a single pixel or a single image region to a candidate target patch object with a spatial range. Thus, at least three types of judgment criteria—spectral information, elevation information, and texture information—are formed within a candidate target patch object, which can reduce misjudgments that occur when determining the target type based solely on a single spectral or texture representation.

[0018] (3) This invention extracts spectral features, elevation features, and texture features based on the spectral information, elevation information, and texture information mapped to the same candidate target patch, and determines the initial target type of the candidate target patch. This ensures that the initial target type is not formed based on a single source feature, but is determined jointly by the three types of features on the same candidate target patch. Therefore, when candidate target patches have similar spectral performance, local elevation anomalies, or texture changes, the initial target type can be constrained by the three types of features simultaneously, reducing the impact of single feature bias on the target recognition result.

[0019] (4) This invention calls the target type verification rules corresponding to the initial target type and limits the target type verification rules to include dominant features, auxiliary features, and feature conflict handling conditions, so that different initial target types can be verified for credibility according to their corresponding dominant and auxiliary features. As a result, spectral features, elevation features, and texture features are no longer simply used in parallel, but are judged in a primary and secondary relationship according to the initial target type, which solves the problem of lack of targeted processing when the recognition tendencies of multiple types of features are inconsistent after the formation of candidate target patches.

[0020] (5) This invention verifies the credibility of spectral features, elevation features, and texture features according to the target type verification rules, obtains the credibility verification result, and confirms, downgrades, or marks the initial target type as pending verification based on the credibility verification result and feature conflict handling conditions, thus generating the target identification result. Therefore, the target identification result no longer only retains the initial classification conclusion, but forms three processing results—confirmation, downgrading, and pending verification mark—when there is a conflict among the three types of features, so that suspected target patches can enter different result states according to the degree of feature consistency, reducing the situation where feature conflict patches are directly used as the target determination result. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a method provided in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the mapping between multi-source remote sensing data and candidate target patches provided in an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram illustrating feature extraction of candidate target patches provided in an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram illustrating the generation of target type feature conditions provided in an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram illustrating the target type verification rules and target recognition result generation provided in an embodiment of the present invention.

[0027] Figure 6 This is a schematic diagram of the target recognition result display interface provided in an embodiment of the present invention.

[0028] Figure 7 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this embodiment provides a real-scene 3D remote sensing target recognition method that integrates spectral, elevation, and texture features, including: Step 100: Acquire the registration remote sensing data of the area to be identified; the registration remote sensing data includes multispectral remote sensing images, three-dimensional geometric data and texture image data under the same spatial reference; Step 200: Construct candidate target patch objects based on the registered remote sensing data, and map the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data to the same candidate target patch object; Step 300: Based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, extract the spectral features, elevation features, and texture features of the candidate target patch object, and determine the initial target type of the candidate target patch object; Step 400: Invoke the target type verification rules corresponding to the initial target type; the target type verification rules include dominant features, auxiliary features, and feature conflict handling conditions; Step 500: Perform credibility verification on spectral features, elevation features, and texture features according to the target type verification rules to obtain credibility verification results. Based on the credibility verification results and feature conflict handling conditions, confirm, downgrade, or mark the initial target type as pending verification to generate target recognition results.

[0032] The method provided by this invention can be applied to scenarios such as remote sensing interpretation of natural resources, surface target surveys, construction change identification, farmland use status identification, and forest disturbance identification. It is particularly suitable for scenarios where the same candidate target patch exhibits differences in spectral, elevation, and texture representation. For example, the area to be identified can be an urban construction edge area, a farmland protection area, a forest boundary area, or an area with frequent construction activities. The above scenarios are merely examples and are not intended to limit the scope of protection of this invention.

[0033] The executing entity in this embodiment can be a remote sensing interpretation platform, a natural resource monitoring platform, an image processing server, or an electronic device with remote sensing data processing capabilities. The executing entity can receive registered remote sensing data from external data sources or call upon existing remote sensing data results for target identification. This embodiment does not limit the acquisition equipment for multispectral remote sensing images, 3D geometric data, and texture image data, as long as the three types of data can be under the same spatial reference and support the construction, feature extraction, and target type verification of candidate target patches.

[0034] For step 100, the area to be identified is the surface space where remote sensing target identification is required. The area to be identified can be determined according to administrative boundaries, the scope of regulatory tasks, the coverage area of ​​remote sensing images, or a manually designated area. Registered remote sensing data refers to a set of data that can correspond to each other under the same spatial reference. Under the same spatial reference, multispectral remote sensing images, 3D geometric data, and texture image data can establish a correspondence through coordinate positions, spatial indexes, or object ranges; it does not require that the three types of data have completely identical data structures, spatial resolutions, or acquisition times.

[0035] Multispectral remote sensing imagery is used to provide the reflection response of ground features within the area to be identified in multiple spectral bands. Three-dimensional geometric data is used to provide the three-dimensional geometric and elevation representations of surface objects within the area to be identified. 3D geometric data can be geometric data from digital surface models, digital elevation models, laser point clouds, oblique photogrammetry 3D models, or 3D geometric results formed from the above data. Texture imagery data is used to provide image representations of ground feature boundaries, internal textures, and directional textures within the area to be identified. Texture imagery data can be orthophotos, 3D model texture images, or visible light texture images corresponding to the area to be identified.

[0036] This embodiment uses registered remote sensing data as input. It does not forcibly convert multispectral remote sensing imagery, 3D geometric data, and texture imagery into the same data format, nor does it simply overlay the three types of data into a single image. In actual processing, multispectral remote sensing imagery, elevation-based 3D results, and texture imagery data usually come from different sources, and their data accuracy and representation methods may also differ. This embodiment requires that the three types of data can establish a correspondence with the same candidate target patch object under the same spatial reference, and subsequent feature extraction is completed around the same candidate target patch object.

[0037] Regarding step 200, this embodiment constructs candidate target patch objects based on the registered remote sensing data. Candidate target patch objects are suspected target areas with continuous spatial extent within the area to be identified. Candidate target patch objects can be obtained through object-oriented segmentation, or they can originate from existing monitoring patches, change detection patches, manually drawn patches, or other spatial objects that can represent suspected target areas. Candidate target patch objects are used to carry subsequent spectral, elevation, and texture information and should not be interpreted as single pixels or isolated classification results without spatial extent.

[0038] Object-oriented segmentation can be performed based on image boundaries, texture differences, spectral differences, 3D surface morphology, or existing spatial patch boundaries. Object-oriented segmentation is used to form spatial units that correspond to surface targets and is not limited to using a specific segmentation algorithm. When existing monitoring patches or change detection patches meet the spatial continuity requirements, these existing patches can be used as candidate target patches, or the existing patches can be merged with the object-oriented segmentation results to form candidate target patches.

[0039] After obtaining the candidate target patches, this embodiment determines the data range corresponding to the candidate target patches in multispectral remote sensing imagery, 3D geometric data, and texture imagery data based on the spatial extent of the candidate target patches. For multispectral remote sensing imagery, pixels or statistical data of each band falling within the spatial extent of the candidate target patches can be acquired as spectral information. For 3D geometric data, 3D patches, elevation points, point cloud sets, or elevation statistical data falling within the spatial extent of the candidate target patches can be acquired as elevation information. For texture imagery data, texture pixels, boundary textures, or local texture regions falling within the spatial extent of the candidate target patches can be acquired as texture information.

[0040] This embodiment maps spectral information from multispectral remote sensing images, elevation information from 3D geometric data, and texture information from texture image data to the same candidate target patch object. The mapping can be based on the spatial boundaries, center position, 3D projection range, or spatial indexing relationship of the candidate target patch object. After mapping, each candidate target patch object corresponds to a set of spectral information, a set of elevation information, and a set of texture information. When multiple source data are independently identified and then merged, inconsistencies in object boundaries can easily occur; this embodiment completes the information mapping at the patch object level, which helps maintain consistency in target judgment units.

[0041] Figure 2 This is a schematic diagram illustrating the mapping between multi-source remote sensing data and candidate target patches provided in an embodiment of the present invention. Figure 2 As shown, multispectral remote sensing imagery, 3D geometric data, and texture imagery data are inputs from different sources, all pointing to the same spatial reference. This same spatial reference indicates that the three types of data can establish a correspondence under a unified coordinate position, spatial index, or object range. Three information streams are derived from the same spatial reference to the candidate target patch object: spectral information, elevation information, and texture information. This means that the spectral information corresponding to the multispectral remote sensing imagery, the elevation information corresponding to the 3D geometric data, and the texture information corresponding to the texture imagery data are all mapped to the same candidate target patch object. The candidate target patch object is represented by an irregular patch outline, with internal contour lines used to illustrate spatial partitions or local morphological differences within the patch object.

[0042] Regarding step 300, this embodiment extracts the spectral features, elevation features, and texture features of candidate target patches based on their spectral, elevation, and texture information mapped to the same candidate target patch object. All spectral, elevation, and texture features are calculated using the same candidate target patch object. Data from adjacent candidate target patch objects are not directly incorporated into the current candidate target patch object, but can be used to form neighborhood references.

[0043] Let the first The candidate target image objects are: . The corresponding effective spectral pixel set is , The number of effective spectral pixels is . The corresponding set of valid elevation sampling points is , The number of valid elevation sampling points is Effective spectral pixels and effective elevation sampling points refer to those falling within... Within the spatial range, and not marked as invalid data units by clouds, shadows, obvious noise, or data holes.

[0044] For spectral characteristics, let The Middle Each effective spectral pixel in the band The reflectance is as follows ,but In the band Average reflection characteristics It can be represented as: In the formula, Indicates the first The candidate target patch is in the band Average reflection characteristics under, express Number of effective spectral pixels in the middle express The Middle Each effective spectral pixel in the band The reflectivity of the sample.

[0045] If a multispectral remote sensing image includes both red and near-infrared bands, then Vegetation spectral characteristics It can be represented as: In the formula, express The average reflectance characteristics in the near-infrared band, express The average reflection characteristics of the red light band. To prevent positive numbers with a denominator of zero. A positive number much smaller than the band reflection characteristic can be used, for example... . Used to characterize the vegetation spectral state of candidate target patches.

[0046] If a multispectral remote sensing image includes both shortwave infrared and near-infrared bands, then Architectural spectral characteristics It can be represented as: In the formula, express The average reflection characteristics of the shortwave infrared band. This is used to characterize the spectral representation of built-up or hardened surfaces of candidate target patches. If the multispectral remote sensing image does not include the shortwave infrared band, this can be omitted. And adopt the band reflection characteristics, inter-spectral difference characteristics or other band combination characteristics that can be obtained to form the band.

[0047] In a natural resource supervision task, candidate target map objects Average reflection characteristics of the red light band The average reflectance characteristic of the near-infrared band is 0.22. The average reflectance characteristic of the shortwave infrared band is 0.31. The value is 0.50. Substituting this into the above formula, The calculated result is approximately 0.17; The calculated result is approximately 0.23. The calculation result indicates... The vegetation exhibits weak spectral characteristics, while there are also some spectral characteristics of constructed or hardened surfaces. The above values ​​are used to illustrate the calculation process of spectral characteristics and are not used as fixed judgment thresholds.

[0048] For elevation characteristics, let... The Middle The elevation values ​​of each valid elevation sampling point are: ,but object elevation characteristics It can be represented as: In the formula, Indicates the first The elevation characteristics of the candidate target patch objects. express Number of valid elevation sampling points in the middle express The Middle The elevation values ​​of each effective elevation sampling point.

[0049] set up The neighborhood buffer range is , The median elevation of the effective elevation sampling points is ,but Neighborhood relative elevation characteristics It can be represented as: In the formula, Indicates the first The height variation of each candidate target patch object relative to its neighborhood buffer range. It can be by The outer boundary expands outward to form, and Not included The relative elevation features of the neighborhood are used to reduce the impact of overall terrain undulation on the determination of target height.

[0050] Three-dimensional surface undulation features It can be represented as: In the formula, express Internal elevation values ​​around the object's elevation characteristics The degree of dispersion. The larger the value, the more pronounced the undulations of the three-dimensional surface inside the candidate target patch.

[0051] Continue with For example, The elevation values ​​of the five valid elevation sampling points are 117.30m, 117.80m, 118.20m, 118.60m, and 118.70m, respectively. According to the object elevation characteristic formula... The calculated result is 118.12m. If... neighborhood buffer range median elevation If it is 114.90m, then The calculated result is 3.22m. Based on the formula for the characteristics of three-dimensional surface undulations, Approximately 0.52m. The above elevation results show that... Compared to the surrounding surface, there are significant changes in elevation, while the internal elevation undulations can be represented by three-dimensional geometric data.

[0052] For texture features, let The corresponding texture image gray-level co-occurrence matrix is , and This indicates the grayscale level. For the normalized gray-level co-occurrence matrix, satisfying . Uniform internal texture features It can be represented as: In the formula, Indicates the first The internal texture uniformity of the candidate target patch object. Due to It has been normalized. The value of is usually between 0 and 1. The larger, the more The internal texture becomes more uniform.

[0053] set up The area is , The perimeter is ,but compact boundary features It can be represented as: In the formula, Indicates the first The boundary compactness of the candidate target patch objects express area, express The perimeter. The value of is usually no greater than 1. The larger the value, the more compact or regular the boundaries of the candidate target patch are. It is not used alone to determine the target type, but rather participates in subsequent verification together with spectral features, elevation features, and other texture features.

[0054] In the same processing task, The normalized gray-level co-occurrence matrix is ​​used to calculate the uniformity of internal texture features. The value is 0.74, indicating a compact boundary feature. The value is 0.66. This result indicates... The internal texture is relatively stable, and the boundaries are relatively compact. Combined with the aforementioned... , , and , It simultaneously possesses three types of feature expressions: spectral, elevation, and texture, so subsequent target type determination does not rely on a single spectral indicator.

[0055] Through the above calculations, candidate target patches can be formed into spectral feature groups, elevation feature groups, and texture feature groups. The spectral feature group can include... , Band reflection characteristics and inter-spectral differences. Elevation characteristic groups can include... , and Texture feature sets can include , In addition, there are directional texture continuity features. All three types of features use the same candidate target patch object as the calculation object, and the feature data will not be mixed between different spatial objects.

[0056] Figure 3 This is a schematic diagram illustrating feature extraction of candidate target patches provided in an embodiment of the present invention. Figure 3 As shown, candidate target patch objects As a single spatial object, the candidate target patch object yields spectral feature groups, elevation feature groups, and texture feature groups respectively; the spectral feature group shows... and The elevation feature group shows , and The texture feature group shows and The three arrows represent the feature formation relationship between the same candidate target patch and the three feature groups, while the three rectangles on the right represent the aggregation results of different feature groups.

[0057] This embodiment determines the initial target type of candidate target patches based on spectral features, elevation features, and texture features. Specifically, spectral features, elevation features, and texture features can be matched with preset target type feature conditions to obtain spectral matching results, elevation matching results, and texture matching results, respectively. The preset target type feature conditions can be feature ranges, feature combination relationships, or feature judgment rules corresponding to different target types. Subsequently, based on the combination relationship of the spectral matching results, elevation matching results, and texture matching results, the initial target type corresponding to the candidate target patches is determined, and a correspondence between the candidate target patches and the initial target type is established.

[0058] Target type characteristic conditions can be generated from historically confirmed sample map features. Historically confirmed sample map features refer to map features whose target type has been confirmed through manual interpretation, on-site verification, operational data, or existing high-precision interpretation results. For any target type... and any feature In this embodiment, the target type in the sample patches has been confirmed through statistical analysis. Corresponding features The distribution of the target type is determined, and target type feature conditions are generated based on the quantiles. Let the target type be... The corresponding historical confirmed sample patch set is as follows ,feature exist The 10th quantile is The 90th quantile is Then the target type Corresponding features The target type feature condition can be expressed as: In the formula, Indicates the first The candidate target patch objects have the following features The feature values ​​are determined by this method. This approach utilizes the distribution range of historically confirmed sample patches to form feature conditions, eliminating the need to manually assign fixed weights to each feature. When the number of historically confirmed samples for a particular target type is small, it can be supplemented using historical samples from the same area, adjacent areas, or a small number of manually verified samples. The corresponding target type feature conditions are updated as the number of samples increases.

[0059] This embodiment uses the 10th and 90th quantiles to form the feature interval. The 10th and 90th quantiles can exclude a small number of outliers in historical samples while retaining the characteristic distribution range of most confirmed samples. This interval is derived from the sample distribution and does not require manually setting multiple sets of weight parameters.

[0060] Figure 4 This is a schematic diagram illustrating the generation of target type feature conditions provided in an embodiment of the present invention. For example... Figure 4 As shown, the historically confirmed sample patch set As a sample source, the distribution curve within the dashed box represents the target type. Lower features The sample distribution; the two vertical markers below the curve correspond to... and The two rectangles on the right represent the 10th and 90th quantiles, respectively; the rightmost rectangle represents the interval... Forming target type The feature conditions are shown in the diagram. The arrows indicate the data transfer direction from the sample set, feature distribution, quantile extraction to the generation of feature conditions.

[0061] Table 1 shows examples of target type feature conditions generated from historically confirmed sample patches. The values ​​in Table 1 are used to illustrate how target type feature conditions are formed and are not considered as fixed thresholds in this invention.

[0062] Table 1 Examples of Target Type Feature Conditions

[0063] Referring to Table 1, of It is 0.17. It is 0.23. It is 3.22m. It is 0.52m. It is 0.74. The value is 0.66. All of the above eigenvalues ​​fall within the range of characteristic conditions corresponding to the suspected constructed target. This can generate initial target types that are suspected to be constructed targets. These initial target types are formed based on multiple features; a single spectral anomaly will not directly translate into a confirmed result.

[0064] When determining the initial target type, this embodiment can calculate candidate target patch objects separately. Support values ​​for feature groups of different target types. Let the candidate target patch object be... The set of candidate target types is , express Any target type in the dataset. Let the feature group... Includes several features , It can be a spectral feature group, an elevation feature group, or a texture feature group. If eigenvalues Satisfy target type The corresponding target type feature condition is denoted as If not satisfied, then record as Feature groups For target type Support value It can be represented as: In the formula, Representing feature groups China The number of valid features. Invalid features are not counted. .like A value of 0 indicates that the feature set cannot be used for... Provide valid supporting results.

[0065] For candidate target patch objects and feature groups ,like Not less than Then the feature group is determined. Supported target types ;like Below Then the feature group is determined. Target type not supported When the same feature group has support values ​​for multiple target types, the target type with the highest support value can be selected as the supported target type for that feature group; if the highest support value is lower than... If the target type is not specified, then the feature group does not support any target type. Spectral matching results, elevation matching results, and texture matching results can all be obtained from the valid feature matching results in their respective feature groups.

[0066] This embodiment determines the initial target type of candidate target patches based on the combination of support values ​​for different target types from spectral feature groups, elevation feature groups, and texture feature groups. For example, the target type with the highest combined support value from the three feature groups can be determined as the initial target type, or the target type can be determined as the initial target type when at least two feature groups support the same target type. If the support values ​​of the three feature groups for different target types are close, or no target type receives sufficient support, the candidate target patch can be marked as a target to be verified, or no initial target type can be generated.

[0067] Table 2 shows an example of feature calculation and support relationships for candidate target patches. The data in Table 2 is only used to illustrate how spectral features, elevation features, and texture features participate in target type determination and is not used as a fixed threshold in this invention.

[0068] Table 2. Examples of feature calculation and support relationships for candidate target patches.

[0069] In Table 2, the spectral, elevation, and texture features of O1 all support the possibility of a built-up target; therefore, the initial target type of O1 is a suspected built-up target. The spectral and texture features of O2 support the possibility of farmland disturbance, but the elevation feature does not support either, nor does it support different target types. Therefore, O2 can still be classified as a suspected farmland disturbance target as an initial target type, and will be downgraded in subsequent reviews. The spectral, elevation, and texture features of O3 all support the possibility of forest disturbance; therefore, the initial target type of O3 is a suspected forest disturbance target. The spectral features of O4 support the possibility of a built-up target, but the elevation and texture features do not. Therefore, O4 can be classified as a suspected built-up target as an initial target type, but it should not be directly used as a confirmation result and will be placed in a pending verification state in subsequent reviews. Thus, the initial target type represents the preliminary identification result of the candidate target patch before the credibility verification. The credibility verification result is used to further distinguish between the confirmed, downgraded, and pending verification states.

[0070] The initial target type represents the preliminary identification result of candidate target patches before confidence verification. Initial target types can include suspected construction targets, suspected farmland disturbance targets, and suspected forest land disturbance targets. These target types are used to represent suspected target categories in remote sensing interpretation and do not directly represent administrative conclusions such as illegal construction, farmland encroachment, or forest land destruction. In practical applications, the initial target type can be adjusted according to the regulatory task. Different initial target types all form preliminary identification results through spectral features, elevation features, and texture features.

[0071] For step 400, this embodiment invokes the target type verification rules corresponding to the initial target type. The target type verification rules include dominant features, auxiliary features, and feature conflict handling conditions. Dominant features are the features in the initial target type that prioritize participation in the credibility verification. Auxiliary features are those that cooperate with the dominant features in the credibility verification. Feature conflict handling conditions are used to confirm, downgrade, or mark the initial target type as pending verification when the recognition tendencies corresponding to spectral features, elevation features, and texture features are inconsistent.

[0072] This embodiment determines the dominant and auxiliary features based on the initial target type. If the initial target type is a suspected constructed target, elevation and texture features can be determined as dominant features, and spectral features as auxiliary features. Suspected constructed targets are usually identified based on relative height changes, boundary regularity, and internal texture, as a single spectral feature can easily be confused with bare land, hardened ground, or areas with material accumulation. If the initial target type is a suspected farmland disturbance target, spectral and texture features can be determined as dominant features, and elevation features as auxiliary features. Suspected farmland disturbance targets usually require a combination of ground cover status and changes in cultivated texture for judgment, with elevation changes serving as a supplementary criterion. If the initial target type is a suspected forest disturbance target, spectral and elevation features can be determined as dominant features, and texture features as auxiliary features. Suspected forest disturbance targets usually require a combination of vegetation spectral status and canopy height changes for judgment, with texture changes helping to identify areas with broken or exposed canopies.

[0073] The configuration of the dominant and auxiliary features described above is used to illustrate that different initial target types can have different verification focuses, and does not mean that this embodiment can only identify the above three types of targets. In other application scenarios, the dominant and auxiliary features can be reconfigured according to the physical and image performance of the target type. For example, when identifying other surface targets, the dominant features can be determined based on the stable performance of the target in spectrum, elevation, and texture, and other features can be used as auxiliary features.

[0074] Feature conflict handling conditions can include confirmation conditions, degradation conditions, and pending verification conditions. Confirmation conditions confirm the initial target type when both the dominant and auxiliary features support it. Degradation conditions downgrade the initial target type when the dominant feature supports it but the auxiliary features do not adequately support it. Pending verification conditions mark the initial target type as pending verification when the dominant feature does not support it, or when the dominant and auxiliary features support different target types. Feature conflict handling conditions can distinguish between candidate target patches with consistent features and candidate target patches with conflicting features.

[0075] Figure 5 This diagram illustrates the target type verification rules and target recognition result generation provided in an embodiment of the present invention. Figure 5 As shown, after the initial target type τ0 is input into the target type verification rules, the dominant feature verification result and the auxiliary feature verification result are formed respectively. The dominant feature verification result and the auxiliary feature verification result are then combined into the feature conflict handling conditions. The feature conflict handling conditions include confirmation conditions, downgrade conditions, and pending verification conditions, which correspond to confirmation markers, downgrade markers, and pending verification markers, respectively. The confirmation markers, downgrade markers, and pending verification markers are further summarized to form the target recognition result. Figure 5The arrows in the text indicate the data flow of the target recognition result after the initial target type has passed through the target type verification rules, feature conflict handling conditions, and verification status marking.

[0076] Table 3 shows examples of target type verification rules. The dominant and auxiliary features in Table 3 are used to illustrate the verification focus for different initial target types and do not imply that the present invention can only be implemented according to the target types listed in Table 3.

[0077] Table 3 Examples of Target Type Review Rules

[0078] In Table 3, the confirmation condition, downgrade condition, and pending verification condition correspond to different review statuses. The confirmation condition is met when both the dominant feature and the auxiliary feature support the initial target type; the downgrade condition is met when the dominant feature supports the initial target type, but the auxiliary feature does not support the initial target type and does not support different target types; the pending verification condition is met when the dominant feature does not support the initial target type, or when the dominant feature and the auxiliary feature support different target types respectively.

[0079] Referring to Table 3, for suspected construction targets, if a candidate target patch only approximates a hardened surface in terms of spectral features, but its elevation and texture features do not support the suspicion of construction, then the candidate target patch will not be directly confirmed as a suspected construction target. For suspected farmland disturbance targets, if the spectral and texture features support the suspicion of farmland disturbance, but the elevation features do not support the suspicion of farmland disturbance and do not support other target types, then the candidate target patch can be downgraded. For suspected forest land disturbance targets, if both the spectral and elevation features support the suspicion of forest land disturbance, and the texture features also support the suspicion of forest land disturbance, then the candidate target patch can be confirmed.

[0080] Regarding step 500, this embodiment performs credibility verification on spectral features, elevation features, and texture features according to target type verification rules to obtain credibility verification results. Specifically, feature verification conditions corresponding to spectral features, elevation features, and texture features are determined according to target type verification rules. Then, spectral features, elevation features, and texture features are matched with their corresponding feature verification conditions to obtain spectral verification results, elevation verification results, and texture verification results. The spectral verification result indicates whether the spectral features support the initial target type. The elevation verification result indicates whether the elevation features support the initial target type. The texture verification result indicates whether the texture features support the initial target type.

[0081] set up The initial target type is For each feature group This embodiment is based on Determine feature groups Does it support initial target type? .like Not less than Then the feature group Support initial target type .like Below However, there are differences. target type ,and Not less than Then the feature group Supports different target types. Below And it does not exist. Not less than Different target types Then the feature group Initial target type is not supported, nor are different target types.

[0082] When a dominant feature comprises multiple feature groups, such as when the dominant features of a suspected constructed target include elevation and texture features, this embodiment can require that all dominant feature groups support the initial target type before determining that the verification result corresponding to the dominant feature supports the initial target type. If any dominant feature group supports a different target type, or if the target types supported by multiple dominant feature groups are inconsistent, then the verification result corresponding to the dominant feature is determined to be conflicting. The verification results corresponding to auxiliary features can be determined in the same manner.

[0083] This embodiment generates a credibility verification result for candidate target patches based on the verification results corresponding to the dominant feature and the auxiliary feature. If the support relationships of the spectral verification result, elevation verification result, and texture verification result for the initial target type are consistent, then there is no feature conflict in the credibility verification result. If the support relationships of the spectral verification result, elevation verification result, and texture verification result for the initial target type are inconsistent, then there is a feature conflict in the credibility verification result. For example, if the spectral verification result supports the initial target type, but the elevation verification result and texture verification result do not, it can be determined that there is a difference in recognition tendency among the three types of features. Furthermore, if the verification result corresponding to the dominant feature points to one target type, and the verification result corresponding to the auxiliary feature points to another target type, it can be determined that there is a feature conflict in the credibility verification result.

[0084] If there are no feature conflicts in the credibility verification results, and both the verification results corresponding to the dominant feature and the verification results corresponding to the auxiliary features support the initial target type, this embodiment determines that the confirmation conditions are met and confirms the initial target type. The confirmed initial target type can be used as the target type of candidate target patch objects. This processing method is suitable for situations where spectral features, elevation features, and texture features have a consistent support relationship for the initial target type.

[0085] When the verification result corresponding to the dominant feature supports the initial target type, but the verification result corresponding to the auxiliary feature does not support the initial target type and does not support different target types, this embodiment determines that the downgrade condition is met and downgrades the initial target type. Downgrade indicates that the candidate target patch object still retains the recognition direction of the initial target type, but the verification status is lower than the confirmation status. The downgraded candidate target patch object can enter the target recognition result as a low-confidence target or a weakly supported target. The downgrade process retains target clues while not using weakly supported results as confirmation results.

[0086] When the verification results corresponding to the dominant feature do not support the initial target type, or when the verification results corresponding to the dominant feature and the verification results corresponding to the auxiliary feature support different target types, this embodiment determines that the conditions for pending verification are met and marks the initial target type as pending verification. The pending verification mark indicates that there are significant conflicts between multiple features of the candidate target patch, requiring further confirmation through manual interpretation, on-site verification, or other business data. The pending verification mark can distinguish candidate target patches with strong feature conflicts from confirmed targets, reducing the likelihood of conflicting patches being directly used as the determined target result.

[0087] Table 4 shows examples of target recognition results. The verification status in Table 4 is determined based on the confidence verification results and feature conflict handling conditions.

[0088] Table 4 Examples of target recognition results

[0089] Based on Tables 1 to 4, both the elevation and texture features of O1 support a suspected built-up target, and the spectral features also support a suspected built-up target; therefore, O1 enters the confirmed state. The spectral and texture features of O2 support a suspected farmland disturbance target, but the elevation feature does not support a suspected farmland disturbance target and does not support different target types; therefore, O2 enters the downgraded state. The spectral and elevation features of O3 support a suspected forest land disturbance target, and the texture feature also supports a suspected forest land disturbance target; therefore, O3 enters the confirmed state. The spectral features of O4 support a suspected built-up target, but the elevation and texture features, as the dominant features, do not support a suspected built-up target; therefore, O4 enters the pending verification state. Thus, the target identification results include target type and verification status, with verification status including confirmed, downgraded, and pending verification.

[0090] This embodiment generates target identification results based on the initial target type and the results of confirming, downgrading, or marking the initial target type as pending verification. The target identification results may include target type, verification status, spatial location, and three-dimensional contour range. The target type is determined based on the initial target type and the corresponding confirmation, downgrading, or pending verification marking results. The verification status indicates whether the target type is in a confirmed, downgraded, or pending verification state. The spatial location is determined based on the spatial range of the candidate target patch in the registered remote sensing data. The three-dimensional contour range is determined based on the spatial range of the candidate target patch in the three-dimensional geometric data. The target identification results can be used for remote sensing interpretation and display, target screening, generation of monitoring patches, or assignment of review tasks.

[0091] Figure 6 This is a schematic diagram of the target recognition result display interface provided in an embodiment of the present invention. Figure 6 As shown, the interface includes a layer list, a 2D view, a target recognition result table, attribute details, and statistical information. The layer list displays data layers such as multispectral imagery, DEM elevation, texture imagery, candidate patches, and recognition results; the 2D view displays the spatial location and outline of candidate target patches O1 to O9; the target recognition result table displays the patch ID, target type, verification status, dominant feature verification result, auxiliary feature verification result, and confidence level for each candidate target patch; the attribute details display the NDVI, NDBI, H, ΔH, σH, T, C, and corresponding target type condition intervals for the selected candidate target patches; and the statistical information displays the quantity distribution of different target types and different verification statuses. Figure 6 The interface content is used to illustrate how the target recognition results are organized, and does not require the target recognition results to be displayed in the same interface format.

[0092] Target identification results are not directly equivalent to administrative determination results. Target identification results represent remote sensing identification results obtained based on registered remote sensing data, candidate target patches, spectral features, elevation features, texture features, and target type verification rules. For matters requiring confirmation through legal procedures, target identification results can be used as supplementary judgment information and further processed in conjunction with planning approval data, land use control data, forest boundary data, on-site verification data, or other business data. The verification status in the target identification results can distinguish between confirmed targets, low-confidence targets, and targets awaiting verification.

[0093] like Figure 7 As shown in the figure, this embodiment of the invention also provides a real-scene 3D remote sensing target recognition system that integrates spectral, elevation, and texture features. The system includes a data acquisition module, a patch construction module, a feature determination module, a rule invocation module, and a target generation module.

[0094] The data acquisition module is used to acquire the registration remote sensing data of the area to be identified. The registration remote sensing data includes multispectral remote sensing images, three-dimensional geometric data and texture image data under the same spatial reference.

[0095] The patch construction module is used to construct candidate target patch objects based on the registered remote sensing data, and to map the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data to the same candidate target patch object.

[0096] The feature determination module is used to extract the spectral features, elevation features, and texture features of candidate target patches based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, and to determine the initial target type of the candidate target patch object.

[0097] The rule invocation module is used to invoke the target type verification rules corresponding to the initial target type. The target type verification rules include dominant features, auxiliary features, and feature conflict handling conditions.

[0098] The target generation module is used to verify the credibility of spectral features, elevation features and texture features according to the target type verification rules, obtain the credibility verification results, and confirm, downgrade or mark the initial target type as pending verification based on the credibility verification results and feature conflict handling conditions, and generate target recognition results.

[0099] The data acquisition module, patch construction module, feature determination module, rule invocation module, and target generation module can be integrated into the same electronic device or distributed across multiple electronic devices. Data can be transferred between the modules via a local bus, network interface, or data service interface. This module division is only for illustrating the functional implementation of the invention and does not imply that each module must exist as an independent hardware component. In actual implementation, a module can be implemented by multiple processing units, or multiple modules can be implemented by the same processing unit.

[0100] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0101] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for real-scene 3D remote sensing target recognition that integrates spectral, elevation, and texture features, characterized in that, include: Acquire registration remote sensing data for the region to be identified, including multispectral remote sensing images, three-dimensional geometric data, and texture image data under the same spatial reference; Candidate target patch objects are constructed based on the registered remote sensing data, and the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data are mapped to the same candidate target patch object. Based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, the spectral features, elevation features, and texture features of the candidate target patch object are extracted, and the initial target type of the candidate target patch object is determined. Invoke the target type verification rule corresponding to the initial target type; the target type verification rule includes dominant features, auxiliary features, and feature conflict handling conditions; The credibility of the spectral features, elevation features, and texture features is verified according to the target type verification rules to obtain the credibility verification result. Based on the credibility verification result and the feature conflict handling conditions, the initial target type is confirmed, downgraded, or marked as pending verification to generate the target recognition result.

2. The real-scene 3D remote sensing target recognition method that integrates spectral, elevation, and texture features according to claim 1, characterized in that, Candidate target patch objects are constructed based on the registered remote sensing data, and the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data are mapped to the same candidate target patch object, including: The region to be identified is segmented into objects based on the registered remote sensing data to obtain the candidate target patch objects; Based on the spatial range of the candidate target patch, the data ranges corresponding to the candidate target patch in the multispectral remote sensing image, the three-dimensional geometric data, and the texture image data are determined respectively. The spectral information is obtained from the data range corresponding to the multispectral remote sensing image, the elevation information is obtained from the data range corresponding to the three-dimensional geometric data, and the texture information is obtained from the data range corresponding to the texture image data. The spectral information, the elevation information, and the texture information are then mapped to the same candidate target patch object.

3. The real-scene 3D remote sensing target recognition method that integrates spectral, elevation, and texture features according to claim 1, characterized in that, Based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, the spectral features, elevation features, and texture features of the candidate target patch object are extracted, including: Based on the spectral information, the band reflectance characteristics, interspectral difference characteristics, and band combination characteristics of the candidate target patch are extracted; Based on the elevation information, extract the object elevation features, neighborhood relative elevation features, and three-dimensional surface undulation features of the candidate target patch; Based on the texture information, extract the boundary texture features, internal texture uniformity features, and directional texture continuity features of the candidate target patch object.

4. The real-scene 3D remote sensing target recognition method that integrates spectral, elevation, and texture features according to claim 1, characterized in that, Determining the initial target type of the candidate target patch object includes: The spectral features, elevation features, and texture features are matched with preset target type feature conditions to obtain spectral matching results, elevation matching results, and texture matching results, respectively. Based on the combination relationship between the spectral matching results, the elevation matching results, and the texture matching results, the initial target type corresponding to the candidate target patch object is determined; Establish the correspondence between the candidate target patch objects and the initial target type.

5. The real-scene 3D remote sensing target recognition method that integrates spectral, elevation, and texture features according to claim 1, characterized in that, Invoking the target type review rule corresponding to the initial target type includes: The dominant feature is determined based on the initial target type, and the dominant feature is the feature that preferentially participates in the credibility verification among the spectral feature, the elevation feature, and the texture feature. The auxiliary feature is determined based on the initial target type. The auxiliary feature is a feature among the spectral feature, the elevation feature, and the texture feature that works in conjunction with the dominant feature to participate in the credibility verification. Based on the target type support relationship between the dominant feature and the auxiliary feature, the feature conflict handling conditions are determined, including confirmation conditions, downgrade conditions, and conditions to be verified.

6. The real-scene three-dimensional remote sensing target recognition method that integrates spectral, elevation, and texture features according to claim 5, characterized in that, The credibility of the spectral features, elevation features, and texture features is verified according to the target type verification rules to obtain credibility verification results, including: Based on the target type verification rules, determine the feature verification conditions corresponding to the spectral features, the elevation features, and the texture features, respectively; The spectral features, elevation features, and texture features are matched with the corresponding feature verification conditions to obtain spectral verification results, elevation verification results, and texture verification results; Based on the verification results corresponding to the dominant features and the verification results corresponding to the auxiliary features, a credibility verification result for the candidate target patch object is generated; When the support relationships between the spectral verification results, the elevation verification results, and the texture verification results for the initial target type are inconsistent, it is determined that there is a feature conflict in the credibility verification results.

7. The real-scene 3D remote sensing target recognition method fusing spectral, elevation, and texture features according to claim 6, characterized in that, Based on the credibility verification results and the feature conflict handling conditions, the initial target type is confirmed, downgraded, or marked for pending verification, including: When both the verification results corresponding to the dominant feature and the verification results corresponding to the auxiliary feature support the initial target type, it is determined that the confirmation condition is met, and the initial target type is confirmed. If the verification result corresponding to the dominant feature supports the initial target type, and the verification result corresponding to the auxiliary feature does not support the initial target type and does not support different target types, then the degradation condition is determined to be met, and the initial target type is downgraded. When the verification result corresponding to the dominant feature does not support the initial target type, or when the verification result corresponding to the dominant feature and the verification result corresponding to the auxiliary feature support different target types, it is determined that the condition to be verified is met, and the initial target type is marked as to be verified.

8. The method for real-scene 3D remote sensing target recognition by fusing spectral, elevation, and texture features according to claim 5, characterized in that, The dominant feature and the auxiliary feature are determined according to the initial target type, including: When the initial target type is a suspected constructed target, the elevation feature and the texture feature are determined as the dominant feature, and the spectral feature is determined as the auxiliary feature; When the initial target type is a suspected farmland disturbance target, the spectral features and the texture features are determined as the dominant features, and the elevation features are determined as the auxiliary features; When the initial target type is a suspected forest disturbance target, the spectral features and the elevation features are determined as the dominant features, and the texture features are determined as the auxiliary features.

9. The method for real-scene 3D remote sensing target recognition by fusing spectral, elevation, and texture features according to claim 1, characterized in that, Generate target recognition results, including: Based on the initial target type and the results of confirming, downgrading or marking the initial target type as pending verification, the target type and review status of the candidate target patch are determined. Based on the spatial range of the candidate target patch in the registered remote sensing data, determine the spatial location and three-dimensional contour range of the candidate target patch; The target recognition result is generated based on the target type, the verification status, the spatial location, and the three-dimensional contour range.

10. A real-scene 3D remote sensing target recognition system that integrates spectral, elevation, and texture features, characterized in that, include: The data acquisition module is used to acquire the registration remote sensing data of the area to be identified. The registration remote sensing data includes multispectral remote sensing images, three-dimensional geometric data and texture image data under the same spatial reference. The patch construction module is used to construct candidate target patch objects based on the registered remote sensing data, and to map the spectral information corresponding to the multispectral remote sensing image, the elevation information corresponding to the three-dimensional geometric data, and the texture information corresponding to the texture image data to the same candidate target patch object. The feature determination module is used to extract the spectral features, elevation features, and texture features of the candidate target patch object based on the spectral information, elevation information, and texture information mapped to the same candidate target patch object, and to determine the initial target type of the candidate target patch object; The rule invocation module is used to invoke the target type verification rule corresponding to the initial target type. The target type verification rule includes dominant features, auxiliary features, and feature conflict handling conditions. The target generation module is used to perform credibility verification on the spectral features, elevation features and texture features according to the target type verification rules, obtain credibility verification results, and confirm, downgrade or mark the initial target type as pending verification according to the credibility verification results and the feature conflict handling conditions, thereby generating target recognition results.

Images