Three-dimensional scene modeling method and system for forestry investigation based on multi-source data fusion

Abstract

The application discloses a forestry investigation three-dimensional scene modeling method and system based on multi-source data fusion, and relates to the technical field of three-dimensional scene modeling.The forestry investigation three-dimensional scene modeling method and system based on multi-source data fusion comprises the following steps: S1, preprocessing orthographic remote sensing image data and spatial modeling constraint data; S2, identifying the dislocation degree according to the multi-factor projection consistency discriminant analysis result and triggering a hierarchical processing strategy; S3, correcting the image projection dislocation through an iterative correction and convergence judgment mechanism, adjusting the local texture mapping relationship, and judging whether the correction process is terminated; and S4, constructing a forestry investigation three-dimensional scene, and executing the output, backtracking and reconstruction decision of the result.The problem that in the prior art, due to the lack of projection consistency constraints of orthographic remote sensing images and terrain elevation data in complex terrain regions such as slopes, the visual dislocation and distortion of image texture and terrain geometric structure in the three-dimensional scene are prone to occur is solved.

Description

Technical Field

[0001] This invention relates to the field of 3D scene modeling technology, specifically to a method and system for 3D scene modeling in forestry surveys based on multi-source data fusion. Background Technology

[0002] In forestry resource surveys, ecological monitoring, and management, three-dimensional spatial representation has gradually become an important technical foundation for improving survey accuracy and decision-making efficiency. With the continuous development of remote sensing, geographic information systems, and 3D visualization technologies, constructing 3D scenes of forest areas using multi-source spatial data to intuitively reflect topographic relief, land cover, and spatial patterns has become a common technical approach in forestry informatization. Existing technologies typically integrate remote sensing imagery, elevation information, and various spatial data to reconstruct and display the 3D morphology and resource distribution of forest areas, meeting the needs of spatial cognition, regional analysis, and planning assistance in forestry surveys. In practical applications, these 3D scenes are widely used for displaying forest resource survey results, planning survey routes, assisting in sample plot layout, topographic analysis, and comprehensive expression of multi-scale spatial information, gradually forming a forestry survey technology system centered on 3D modeling, which has been continuously promoted and applied in related industry practices.

[0003] For example, the invention patent with announcement number CN116778105B discloses a modeling method based on the fusion of multi-precision 3D surveying and mapping data. This method delineates the surveying and mapping area and establishes a reference coordinate system for the covered area. It combines UAV imaging, lidar mapping, and multi-anchor point deployment to acquire multi-source surveying and mapping data, and performs zonal collection and multi-scale modeling of the terrain within the surveying and mapping area. At the same time, it introduces a surveying and mapping condition monitoring and image quality assessment mechanism to perform multi-level screening and reorganization of the acquired surveying and mapping images. It then fuses the planar real-scene map with the 3D terrain model constructed from multiple sets of terrain surveying data. For areas with errors, it improves the model matching degree through re-surveying and iterative correction, and finally outputs a 3D terrain model with acceptable credibility, thereby achieving high-precision 3D reconstruction of the terrain of the surveying and mapping area.

[0004] For example, the invention patent with publication number CN120563751A discloses a geological 3D modeling method and system based on multi-data fusion. This method is based on geological profile mapping results and geophysical exploration data. It divides different density sections by analyzing the proportional characteristics of the spacing between profile nodes, and generates section adjustment coefficients by combining the differences in node elevation and spatial position. It performs node elevation offset and angle correction on high-density and low-density sections respectively. On this basis, it constructs multi-layer profile reconstruction primitives, forms coupling features by extracting the relationship between node line segment lengths and turning angles, and performs difference analysis with adjacent profiles. Finally, it integrates the relationship between node segment lengths, elevation offset information and original coordinate data to generate geological 3D stitching information, realizing geological 3D structure modeling under the conditions of multiple profiles and multiple data sources.

[0005] Existing 3D scene modeling technologies mostly rely on the fusion of multi-source spatial data and 3D reconstruction to achieve basic representation of terrain morphology and land cover. However, under the complex terrain conditions of forestry surveys, there is insufficient constraint on the projection relationship between images and terrain. The modeling process mainly focuses on data accuracy or local correction, lacking a systematic discrimination and control mechanism for the projection consistency of areas such as slopes. This can easily lead to visual misalignment and distortion between image texture and terrain geometry in 3D scenes, affecting the intuitiveness and credibility of forestry survey results.

[0006] Therefore, in order to address the above problems, there is an urgent need for a method and system for three-dimensional scene modeling of forestry surveys based on multi-source data fusion. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for modeling three-dimensional scenes in forestry surveys based on multi-source data fusion. This solves the problem that in existing three-dimensional scene modeling for forestry surveys, the lack of projection consistency constraints between orthophoto remote sensing images and terrain elevation data in complex terrain areas such as slopes easily leads to visual misalignment and distortion of image textures and terrain geometry in the three-dimensional scene.

[0008] Technical solution To achieve the above objectives, this invention employs the following technical solution: a forestry survey 3D scene modeling method based on multi-source data fusion, comprising: S1, acquiring orthorectified remote sensing image data and spatial modeling constraint data, and preprocessing the orthorectified remote sensing image data and spatial modeling constraint data; S2, performing multi-factor projection consistency discrimination analysis on the orthorectified remote sensing image data and spatial modeling constraint data, identifying the degree of misalignment between the image and the terrain within the slope area based on the results of the multi-factor projection consistency discrimination analysis, and triggering a hierarchical processing strategy; S3, correcting the image projection misalignment through an iterative correction and convergence determination mechanism, dynamically adjusting the local texture mapping relationship, and determining whether the correction process has terminated; S4, constructing a forestry survey 3D scene based on the projection correction, conducting consistency delivery analysis by fusing boundary ground-hugging properties, projection consistency, and micro-topography level maintenance, and executing output, retrospective, and reconstruction decisions for the 3D scene results based on the consistency delivery analysis results.

[0009] Furthermore, the specific process of acquiring orthorectified remote sensing image data and spatial modeling constraint data is as follows: Orthorectified remote sensing image data includes: multi-band pixel value data, grayscale pixel value data, image acquisition time data, solar altitude angle data, solar azimuth angle data, sensor observation angle data, and image geographic reference data; Spatial modeling constraint data includes: topographic elevation data, vector land cover data, forest boundary data, and control point data; among which, topographic elevation data includes forest area topographic elevation raster data and topographic elevation sampling interval data; vector land cover data includes land cover zone boundary point set data and land cover category attribute data; forest boundary data includes forest boundary polyline point set data, forest boundary closure marker data, and boundary topology consistency marker data; control point data includes ground control point data, control point acquisition time data, and control point quality marker data.

[0010] Furthermore, the specific preprocessing steps for orthophoto remote sensing image data and spatial modeling constraint data are as follows: For terrain elevation data, a spike and collapse correction is performed using a sliding window midpoint filtering algorithm and a robust regression correction algorithm for local outliers; for terrain elevation data and orthophoto remote sensing image data, resolution consistency and pixel alignment are achieved using a unified grid resampling algorithm and pixel center alignment rules; for vector land cover data and forest boundary data, topological consistency is restored using topological self-checking and self-intersection resolution algorithms; for land cover zone boundary point set data and forest boundary polyline point set data, self-intersection, duplicate points, fracture endpoints, and pseudo-closed loop detection and resolution are performed; and the ground-level stability of the boundary on the slope is improved using a boundary polyline uniform densification algorithm; for ground control point data, quality screening is performed using a three-dimensional coordinate consistency verification algorithm and outlier removal algorithm, unifying it to a spatial reference consistent with the orthophoto remote sensing image data; and for orthophoto remote sensing image data and spatial modeling constraint data, standardization and normalization are performed using distribution standardization and linear normalization algorithms.

[0011] Furthermore, the specific process of performing multi-factor projection consistency discrimination analysis on orthophoto remote sensing image data and spatial modeling constraint data is as follows: For image grayscale pixel data, the Sobel gradient operator is used to obtain a texture gradient grayscale map; for forest terrain elevation raster data, combined with solar altitude angle data and solar azimuth angle data, a slope shadow geometric projection algorithm is used to obtain a slope shadow grayscale map; within a sliding window, normalized mutual information calculation is performed on the texture gradient grayscale map and the slope shadow grayscale map, and the structural information missing value is obtained by subtracting one from the mutual information value; for forest terrain elevation raster data, second-order curvature calculation and flow direction accumulation analysis are used to obtain the terrain linear structure; for image multi-band pixel data, vegetation index enhancement and Canni edge detection are used to obtain the image linear structure, and the average nearest distance is calculated through nearest-nearest-distance matching to obtain the linear structure deviation; using solar altitude angle data, solar azimuth angle data, and forest terrain data... A slope shadow mask is generated from elevation raster data and an image shadow mask is generated from image grayscale pixel data. Viewpoint consistency filtering is performed using sensor observation angle data, and the mask difference ratio is statistically analyzed to obtain the occlusion inconsistency degree. The main slope direction is calculated from the forest terrain elevation raster data, and the main texture direction is calculated from the image grayscale pixel data. The angle between the two directions is normalized to obtain the direction deviation degree. The natural logarithm of the missing structural information degree is calculated to obtain the logarithmic modulation term of the structural information. The exponential value of the linear structural deviation degree is calculated to obtain the linear structural exponential amplification term. The square root of the occlusion inconsistency degree is calculated to obtain the occlusion consistency adjustment term. The reciprocal of the direction deviation degree is calculated to obtain the direction consistency suppression term. The logarithmic modulation term of the structural information, the linear structural exponential amplification term, the occlusion consistency adjustment term, and the direction consistency suppression term are multiplied together to obtain the slope projection misalignment criterion value.

[0012] Furthermore, the specific process of identifying the degree of misalignment between the image and the terrain within the slope area and triggering a hierarchical processing strategy based on the results of multi-factor projection consistency discrimination analysis is as follows: Real-time comparison of the slope projection misalignment criterion value and the slope projection misalignment criterion threshold. The slope projection misalignment criterion threshold includes a primary criterion threshold and a secondary criterion threshold: When the slope projection misalignment criterion value is less than the secondary criterion threshold, orthophoto remote sensing image data, terrain elevation data, vector land cover data, forest boundary data, and ground control point data are created and archived into the modeling database and entered into the 3D scene construction and output module; when the slope projection misalignment criterion value is greater than or equal to the secondary criterion threshold and... When the value is less than the first-level criterion threshold, a misaligned hot zone raster is generated based on the slope projection misalignment criterion value, and a correction zone is divided in combination with the land use zoning boundary point set data. The misaligned hot zone and the corresponding correction zone are used as the correction range. Based on the ground control point data, local texture coordinate remapping is performed on the orthophoto image in the slope area, and linear structure consistency constraints and orientation consistency constraints are introduced. The image then enters the projection consistency constraint iteration and convergence control module. When the slope projection misalignment criterion value is greater than or equal to the first-level criterion threshold, the misaligned hot zone is isolated to participate in the direct texture attachment output. Available control points are selected based on the control point quality label data, and a ground control point supplementary log is generated.

[0013] Furthermore, the specific process of the iterative correction and convergence determination mechanism is as follows: Effective control points are selected based on the control point quality marker data; spatial calculation is performed on the image geographic reference data to map the control point coordinates to the image coordinate system; the difference in the projection positions of the control points in the image before and after correction is calculated to obtain the control point residual degree; the natural logarithmic value of the correction structural information missing degree plus one is calculated to obtain the correction structural information logarithmic modulation term; the exponential value of the correction occlusion inconsistency degree is calculated and added to one to obtain the correction occlusion exponential enhancement term; the sum of the control point residual and the correction direction deviation degree is added to one as the comprehensive suppression factor; the correction structural information logarithmic modulation term and the correction occlusion exponential enhancement term are multiplied as the consistency driving term, and adjusted with the reciprocal of the comprehensive suppression factor to obtain the slope correction convergence value.

[0014] Furthermore, the specific process for correcting image projection misalignment, dynamically adjusting local texture mapping relationships, and determining whether the correction process has terminated is as follows: Real-time comparison of slope correction convergence values ​​and slope correction convergence thresholds. The slope correction convergence thresholds include a first-level convergence threshold and a second-level convergence threshold. When the slope correction convergence value is less than the second-level convergence threshold, the corrected texture attachment result and correction parameters are output. The forest boundary polyline point set data is then checked for closure according to the closure marker data. The ground-attached forest boundary data and vector land cover data are created and archived into the correction database, and then the process proceeds to 3D scene construction. The output module continues to perform local texture coordinate remapping when the slope correction convergence value is greater than or equal to the second-level convergence threshold and less than the first-level convergence threshold. It enhances linear structure extraction in the boundary neighborhood corresponding to the land use partition boundary point set and adjusts the sliding window scale based on the terrain elevation sampling interval data. When the slope correction convergence value is greater than or equal to the first-level convergence threshold, it isolates the misaligned hot zone as a verification zone and generates control point residual logs, boundary topology consistency marker data verification results, and correction parameters. Simultaneously, it outputs the corrected area results (excluding the verification zone) and enters the 3D scene construction and output module.

[0015] Furthermore, based on projection correction, a 3D forestry survey scene is constructed. The specific process for conducting consistency delivery analysis by integrating boundary ground-fitting, projection consistency, and micro-topography preservation is as follows: A 3D land surface is constructed based on topographic elevation data. The consistency-corrected orthophoto remote sensing image is then attached as a texture to the 3D land surface. Forest boundary data and vector land cover data are loaded to construct a 3D forestry survey scene model, which is then used as the 3D scene output. The forest boundary polyline point set data and land cover zone boundary point set data are mapped to the topographic surface using a ground-fitting projection algorithm. Based on pixel center alignment rules, ground-fitted 3D boundary polyline point set data is formed. The corrected texture is then attached... If edge detection is performed on the grayscale pixel data of the image and mapped to the terrain surface, the spatial deviation between the image and the 3D boundary polyline point set data is calculated and normalized to obtain the 3D boundary misalignment rate; the curvature probability distribution of the terrain elevation data before and after correction is calculated and normalized using Jensen-Shannon divergence and exponential decay to obtain the micro-topography fidelity; the difference between the 3D boundary misalignment rate and one is calculated to obtain the boundary consistency preservation term; the reciprocal of the slope projection misalignment criterion value plus one is calculated to obtain the slope projection consistency suppression term; the square root of the micro-topography fidelity plus one is calculated to obtain the micro-topography preservation enhancement term; the boundary consistency preservation term, the slope projection consistency suppression term, and the micro-topography preservation enhancement term are multiplied to obtain the 3D scene consistency delivery value.

[0016] Furthermore, the specific process for executing the output, retrospective, and reconstruction decisions of the 3D scene results based on the consistency delivery analysis results is as follows: Real-time comparison of the 3D scene consistency delivery value and the 3D scene consistency delivery threshold. The 3D scene consistency delivery threshold includes a primary delivery threshold and a secondary delivery threshold: When the 3D scene consistency delivery value is greater than or equal to the primary delivery threshold, output the forestry survey 3D scene result file, slope projection misalignment criterion value distribution report, boundary misalignment report, and micro-topography fidelity report; archive the image acquisition time data, control point acquisition time data, sensor observation angle data, and image geographic benchmark data to the modeling database; when the 3D scene consistency delivery value is greater than or equal to the secondary delivery threshold and less than .... When the threshold is reached, the process backtracks to the projection consistency constraint iteration and convergence control module: for partitions where the 3D boundary misalignment rate is greater than the average, forest boundary closure marker data constraints are introduced to participate in the correction; for partitions where the slope projection misalignment criterion value is greater than the average, control point quality marker data is called to filter control points and adjust the correction partition range; micro-topography fidelity is recalculated and the archive is updated; when the 3D scene consistency delivery value is less than the secondary delivery threshold, a delivery anomaly database is created and the current 3D scene output file, partition-level slope projection misalignment criterion value, 3D boundary misalignment rate and micro-topography fidelity report are archived; a check area list and ground control point supplementation suggestion log are output; and the layer publishing driven by vector land category attribute data is paused in the check area.

[0017] The second aspect of this invention provides a forestry survey 3D scene modeling system based on multi-source data fusion, comprising: an acquisition and preprocessing module for acquiring orthophoto remote sensing image data and spatial modeling constraint data, and preprocessing the orthophoto remote sensing image data and spatial modeling constraint data; a slope projection consistency discrimination module for performing multi-factor projection consistency discrimination analysis on the orthophoto remote sensing image data and spatial modeling constraint data, identifying the degree of misalignment between the image and the terrain in the slope area based on the multi-factor projection consistency discrimination analysis results, and triggering a hierarchical processing strategy; a projection consistency constraint iteration and convergence control module for correcting image projection misalignment through an iterative correction and convergence determination mechanism, dynamically adjusting local texture mapping relationships, and determining whether the correction process has terminated; and a 3D scene construction and output module for constructing a forestry survey 3D scene based on projection correction, conducting consistency delivery analysis by fusing boundary ground-hugging properties, projection consistency, and micro-topography level, and executing output, retrospective, and reconstruction decisions of the 3D scene results based on the consistency delivery analysis results.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) This invention achieves the coordinated expression of image texture and terrain geometry under complex slope conditions by uniformly discriminating and hierarchically controlling the projection relationship of multi-source spatial data, thereby significantly improving the overall spatial consistency of the three-dimensional scene of forestry survey and effectively solving the problem of texture and terrain misalignment in complex areas such as slopes in the existing technology. (2) In the process of three-dimensional modeling, the present invention introduces an iterative correction and convergence determination mechanism, which enables the image projection relationship to be optimized step by step under controlled conditions, thereby realizing the effect of quantifiable and terminateable projection correction process, effectively solving the problem that projection correction in the prior art relies on human experience and is difficult to converge stably; (3) This invention, by incorporating the boundary ground-hugging property and terrain representation into the modeling process, realizes the continuous and stable presentation of forest boundaries along the terrain in the three-dimensional scene, thereby achieving the effect of high consistency between boundary semantics and spatial morphology, effectively solving the problem of boundaries being suspended, broken or semantically distorted on the three-dimensional terrain in the prior art. (4) This invention establishes a consistent delivery analysis mechanism to uniformly evaluate and classify the results of three-dimensional scene, thereby achieving the effect that the modeling results are verifiable and traceable, effectively solving the problem of lack of quality quantification and decision-making basis for three-dimensional scene results in the prior art.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] Figure 1 This is a flowchart of the forestry survey 3D scene modeling method based on multi-source data fusion according to the present invention; Figure 2 This is a structural diagram of the forestry survey 3D scene modeling system based on multi-source data fusion according to the present invention; Figure 3 This is a schematic diagram of the convergence evolution curve of the slope correction convergence value according to the present invention; Figure 4 This is a flowchart of the three-dimensional scene consistency delivery determination process of the present invention. Detailed Implementation

[0021] 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.

[0022] Please see Figures 1-4This invention provides a technical solution: a method for modeling a 3D scene of a forestry survey based on multi-source data fusion, comprising the following steps: S1, collecting orthorectified remote sensing image data and spatial modeling constraint data, and preprocessing the orthorectified remote sensing image data and spatial modeling constraint data; S2, performing multi-factor projection consistency discrimination analysis on the orthorectified remote sensing image data and spatial modeling constraint data, identifying the degree of misalignment between the image and the terrain in the slope area based on the results of the multi-factor projection consistency discrimination analysis, and triggering a hierarchical processing strategy; S3, correcting the image projection misalignment through an iterative correction and convergence determination mechanism, dynamically adjusting the local texture mapping relationship, and determining whether the correction process has terminated; S4, constructing a 3D scene of a forestry survey based on the projection correction, conducting consistency delivery analysis by fusing boundary ground-hugging properties, projection consistency, and micro-topography level, and executing output, retrospective, and reconstruction decisions of the 3D scene results based on the consistency delivery analysis results.

[0023] Specifically, the process of collecting orthorectified remote sensing image data and spatial modeling constraint data is as follows: collecting orthorectified remote sensing image data, which includes: multi-band pixel value data, grayscale pixel value data, image acquisition time data, solar altitude angle data, solar azimuth angle data, sensor observation angle data, and image geographic reference data. By uniformly organizing the multi-temporal and multi-view remote sensing imaging results, the integrity of the image in terms of radiometric characteristics, geometric reference, and time markers is ensured.

[0024] Spatial modeling constraint data is collected, including topographic elevation data, vector land cover data, forest boundary data, and control point data. By uniformly collecting and organizing geometric constraints, semantic constraints, and control constraints, a multi-source constraint set that can simultaneously support spatial correction and semantic consistency analysis is formed.

[0025] The topographic elevation data includes forest area topographic elevation raster data and topographic elevation sampling interval data. By uniformly collecting and organizing geometric, semantic, and control constraints, a multi-source constraint set is formed that can simultaneously support spatial correction and semantic consistency analysis. Vector land cover data includes land cover zoning boundary point sets and land cover category attribute data. Through the collaborative expression of boundary geometry and category attributes, an accurate description of the spatial distribution of land covers and their semantic classification relationships is achieved. Forest boundary data includes forest boundary polyline point sets, forest boundary closure marker data, and boundary topological consistency marker data. Through the collaborative expression of boundary geometry and category attributes, an accurate description of the spatial distribution of land covers and their semantic classification relationships is achieved. Control point data includes ground control point data, control point acquisition time data, and control point quality marker data. Through the comprehensive identification of the timeliness and quality level of control points, a quantifiable basis is provided for geometric correction and accuracy assessment. Remote sensing satellite image grayscale pixel value data is used; airborne lidar is used to collect forest area topographic elevation raster data; and total station data is used to collect ground control point data.

[0026] This implementation plan, through the systematic acquisition and organization of orthorectified remote sensing imagery and spatial modeling constraint data, establishes a multi-source data framework covering image representation, topographic geometry, spatial boundaries, and control information. This framework ensures that various data types possess a unified descriptive basis and constraints at the spatial reference, temporal identification, and quality attribute levels, thereby reducing the risk of inconsistencies in scale, accuracy, and semantics between different data sources. By constructing a dataset with complete geometric support and semantic orientation, image texture, topographic relief, forest boundaries, and control point information can be jointly analyzed and mutually constrained within the same modeling system. This provides a continuous, controllable, and verifiable data support environment for subsequent slope projection consistency determination, stable convergence of the correction process, and the overall credibility of the 3D scene results.

[0027] Specifically, the preprocessing process for orthorectified remote sensing image data and spatial modeling constraint data is as follows: For orthorectified remote sensing image data, a segmented histogram matching algorithm and a locally contrast-limited adaptive histogram equalization algorithm are used to unify the cross-scene brightness distribution, reducing the interference of statistical distribution differences between different images of the same area on subsequent structure determination; for orthorectified remote sensing image data, a phase correlation method and a multi-scale mutual information maximization registration algorithm are used for coarse alignment, ensuring consistency between texture structure extraction and the spatial reference corresponding to the slope structure; for topographic elevation data, a sliding window median filtering algorithm and a robust regression correction algorithm for local outliers are used to correct spikes and collapses, with a large median filtering window. Based on the topographic elevation sampling interval data and local elevation gradient statistics, the loss function form and parameters used in robust regression correction are selected from the elevation residual distribution analysis results, preserving the resolvable morphology of valleys and ridges while suppressing local isolated anomalies. For topographic elevation data and orthorectified remote sensing image data, resolution consistency and pixel alignment are achieved through a unified grid resampling algorithm and pixel center alignment rules. The unified grid resolution is determined based on the minimum common scale of the spatial resolution of the topographic elevation sampling interval data and the orthorectified remote sensing image, avoiding boundary drift introduced by sampling interval differences. For vector land cover data and forest boundary data, topological self-checking and self-intersection resolution algorithms are used. Topological consistency restoration involves detecting and eliminating self-intersections, duplicate points, fracture endpoints, and pseudo-closed loops in the boundary point sets of land use zones and forest boundary polyline point sets. A uniform densification algorithm for boundary polylines is used to improve the ground-hugging stability of the boundaries on slopes. The spacing of the uniform densification is determined based on the topographic elevation sampling interval data and the statistical results of the local curvature of the boundary, ranging from one to three times the topographic elevation sampling interval, reducing the probability of overhangs and fractures during subsequent ground-hugging. Ground control point data undergoes quality screening using a 3D coordinate consistency verification algorithm and an outlier removal algorithm, unifying it to a spatial reference consistent with orthorectified remote sensing data, and outlier removal is performed. The threshold is determined based on the statistical distribution of control point residuals and the analysis results of their confidence intervals. The value range is within the confidence interval covering no less than 95% of the valid control point residual samples, and is used as a stability constraint for subsequent deformation correction. Orthophoto remote sensing image data and spatial modeling constraint data are standardized and normalized using distribution standardization and linear normalization algorithms. First, the distribution standardization algorithm is used to uniformly constrain the statistical differences in mean, variance and distribution shape of different data sources. Then, the linear normalization algorithm is used to map the standardization results to the [0,1] numerical range to ensure the comparability and numerical stability of various criteria in the subsequent projection consistency discrimination and convergence judgment process.

[0028] In this implementation plan, this step involves systematically and uniformly preprocessing imagery, terrain, and spatial constraint data. This ensures that the multi-source data achieves a consistent and comparable state in terms of brightness distribution, spatial location, resolution scale, geometric shape, and topological relationships, mitigating the cumulative impact of differences in data sources, variations in acquisition conditions, and noise anomalies on subsequent analysis results. By establishing a stable spatial reference system and reliable geometric constraints, image texture representation, terrain undulation features, boundary structure morphology, and control point accuracy can synergistically exert their constraining effects within the same analytical framework. This significantly improves the stability of structural feature determination, the controllability of slope projection analysis, and the continuity and overall reliability of results during subsequent correction and 3D scene construction.

[0029] Specifically, the process of performing multi-factor projection consistency discrimination analysis on orthophoto remote sensing image data and spatial modeling constraint data is as follows: For the image grayscale pixel value data, the Sobel gradient operator is used to calculate the texture gradient grayscale map. By highlighting the locations of grayscale abrupt changes, the response intensity of potential structural edges in the image is enhanced, providing a stable texture change representation basis for subsequent structural consistency analysis. For the forest area topographic elevation raster data, combined with solar altitude angle data and solar azimuth angle data, a slope shadow geometric projection algorithm is used to calculate the slope shadow grayscale map. By introducing real illumination geometric relationships to characterize the modulation effect of topographic undulations on shadow distribution, the shadow representation can reflect the actual slope morphology characteristics. Within a sliding window, the texture gradient grayscale map and the slope shadow grayscale map are normalized and cross-referenced. The information calculation algorithm calculates the mutual information value, and subtracting the mutual information from 1 yields the structural information missing degree. This quantifies the weakening of the matching between image texture and terrain structure in the form of information correlation decay, enhancing the measurability of local differences. The sliding window size is determined based on the consistency analysis results between the terrain elevation sampling interval data and the image spatial resolution. The window edges are processed using a mirror extension method. The probability distribution in the mutual information statistics process is obtained through fixed-bin histogram statistics, and the number of bins is determined based on the number of samples within the window and the grayscale dynamic range constraint. A second-order curvature calculation algorithm is used to extract slope curvature features from the forest terrain elevation raster data, and a flow direction accumulation analysis algorithm is combined to extract the linear terrain structure, including ridgelines and valley lines. This is further analyzed using curvature and hydrological data. Morphological joint constraint stability extraction is used to extract the dominant linear skeleton with topographic control significance; vegetation index enhancement algorithm is used to enhance the forest edge area using multi-band pixel value data of the image, and then the Canni edge detection algorithm is used to extract the linear structure of the image. The characteristics of the forest edge transition zone are amplified by vegetation spectral differences, which improves the continuity and recognizability of structural lines in the image; the nearest distance matching algorithm is used to calculate the average nearest distance between the topographic linear structure and the image linear structure to obtain the linear structure deviation, and the overall offset trend of the two types of structural lines at the position level is characterized by spatial proximity relationship, so as to avoid excessive interference of local anomalies on the evaluation results; using solar altitude angle data, solar azimuth angle data and forest area topographic elevation raster data, a slope shadow mask is generated by a slope shadow projection algorithm, which is physically consistent with the image linear structure. The projection relationship is used to constrain the shadow distribution range and improve the geometric rationality of occlusion judgment. The Otsu threshold segmentation algorithm is used to generate image shadow masks for image grayscale pixel data. The boundary between bright and dark areas in the image is determined by adaptive threshold, so that the shadow extraction process has stable adaptability to changes in overall brightness. After screening edge areas for viewing angle consistency by combining sensor observation angle data, the pixel difference ratio of the two types of shadow masks is statistically analyzed to obtain the occlusion inconsistency degree. The occlusion judgment result is corrected by introducing observation geometric conditions to reduce the influence of false inconsistency caused by viewing angle differences. The slope aspect calculation algorithm is used to obtain the main slope direction for forest terrain elevation raster data. The slope orientation is used as the directional constraint benchmark of terrain structure to reflect the dominant extension trend of terrain in space.A gradient orientation histogram statistical algorithm is used to obtain the main texture direction from the grayscale pixel data of the image. By statistically analyzing the concentrated gradient distribution, the dominant texture orientation information in the image is extracted. The angle between two directions is calculated and normalized according to the maximum angle to obtain the orientation deviation. This deviation is used to characterize the overall consistency level between the image texture direction and the terrain slope aspect in the form of scale-independent angle difference.

[0030] The natural logarithm of the missing structural information degree is calculated by adding one to obtain the logarithmic modulation term for structural information. This term is used to enhance the sensitivity of low- and medium-strumental missing areas to the overall criterion while suppressing the explosion of extreme missing values. The exponential value of the linear structural deviation degree, base e, is calculated to obtain the linear structural exponential amplification term. This term enhances the response amplitude of the terrain linear structure and image linear structure in the early stage of spatial misalignment through exponential mapping. The square root of the occlusion inconsistency degree is calculated by adding one to obtain the occlusion consistency adjustment term. This term reduces the excessive dominance of local strong occlusion differences on the overall projection determination in a nonlinear way. The reciprocal of the direction deviation degree is calculated by adding one to obtain the direction consistency suppression term. This term explicitly constrains the negative impact of slope aspect and texture main direction deviation on projection consistency through an inverse proportional suppression mechanism. The logarithmic modulation term for structural information, the exponential amplification term for linear structure, the occlusion consistency adjustment term, and the direction consistency suppression term are multiplied to obtain the slope projection misalignment criterion value. The specific calculation formula is as follows: ; In the formula, This represents the slope projection misalignment criterion value, which is used to comprehensively characterize the overall projection consistency level of image texture relative to the terrain geometry under slope conditions. It represents the degree of missing structural information, reflecting the change in the degree of information coupling between image texture structure and terrain shadow structure within a local spatial window; It represents the deviation of the linear structure, describing the degree of spatial location and continuity of the topographically dominated linear skeleton and the linear features of the forest edge in the image; The occlusion inconsistency is represented by the degree of mismatch between the image shadow and the terrain shadow caused by differences in slope orientation, terrain undulation, and observation geometry; t represents the discrimination calculation index, which is used to identify the calculation status of the slope projection misalignment criterion at the current analysis scale, spatial location, or processing stage.

[0031] In this implementation scheme, a multi-dimensional consistency representation mechanism is constructed by systematically and collaboratively characterizing image texture features, terrain structure features, illumination occlusion relationships, and directional consistency. This mechanism can comprehensively reflect the spatial coupling state between images and terrain under slope conditions, enabling the misalignment, occlusion mismatch, and directional deviation between image texture and terrain geometry to be stably quantified under a unified discrimination framework. By comprehensively mapping local structural correlation, linear skeleton alignment degree, shadow projection rationality, and directional consistency into a single slope projection misalignment criterion value, the overall evaluation and comparable expression of projection consistency level in complex slope scenes are realized. This provides a core quantitative basis with physical meaning, spatial sensitivity, and discrimination stability for subsequent misalignment partitioning identification, hierarchical correction decision-making, and 3D scene consistency control.

[0032] Specifically, the process of identifying the degree of misalignment between the image and the terrain within the slope area and triggering a graded processing strategy based on the results of multi-factor projection consistency discriminant analysis is as follows: Real-time comparison of slope projection misalignment criterion values ​​and slope projection misalignment criterion thresholds. The slope projection misalignment criterion thresholds include a primary criterion threshold and a secondary criterion threshold. These thresholds are obtained through quantile statistical analysis based on the statistical distribution of slope projection misalignment criterion values ​​in historical remote sensing image and terrain matching sample data. The primary criterion threshold corresponds to the high-risk quantile interval of the statistical distribution, with a value range of 0.65–0.85, while the secondary criterion threshold corresponds to the medium-risk quantile interval of the statistical distribution, with a value range of 0.35–0.55. When the slope projection misalignment criterion value is less than the secondary criterion threshold, it is determined that the projection consistency meets the requirements, indicating that the image texture and terrain structure have reached an acceptable spatial matching accuracy in the slope area. The orthophoto remote sensing image data, terrain elevation data, vector land cover data, forest boundary data and ground control point data are created and archived into the modeling database to form a stable data baseline that can be directly used for 3D modeling and survey applications, and then enter the 3D scene construction and output module.

[0033] When the slope projection misalignment criterion value is greater than or equal to the secondary criterion threshold and less than the primary criterion threshold, it is determined that there is a correctable misalignment in the projection consistency. This indicates that the misalignment is mainly concentrated in a local slope area and has the conditions to be repaired through constraint correction. A misalignment hotspot raster is generated based on the slope projection misalignment criterion value, and correction zones are divided by combining the land use zoning boundary point set data. The resolution of the misalignment hotspot raster is determined based on the topographic elevation sampling interval. Spatially continuous misalignment areas are extracted through pixel adjacency connectivity analysis to achieve spatial focusing of the correction range and avoid global indiscriminate adjustment. The misalignment hotspot and its corresponding correction zone are used as the correction... The scope of application is determined by establishing spatial constraints based on ground control point data, providing a stable and traceable geometric reference for local correction. Local texture coordinate remapping is performed on orthophotos within slope areas. During the correction process, linear structural consistency constraints and directional consistency constraints are introduced simultaneously. Through dual constraints of structure and direction, new geometric distortions are avoided during the correction process, gradually reducing the spatial deviation between image texture and terrain structure. Orthophoto data, terrain elevation data, vector land cover data, forest boundary data, and ground control point data are used as inputs into the projection consistency constraint iteration and convergence control module.

[0034] When the slope projection misalignment criterion value is greater than or equal to the first-level criterion threshold, it is determined that the projection consistency is seriously insufficient, indicating that there is a systematic misalignment risk between the current image and the terrain, which has exceeded the reliable range of correction. The misalignment hotspot is isolated from the texture pasting process and its participation in the 3D scene texture pasting output is suspended to prevent high-risk areas from negatively transmitting to the overall semantic and geometric quality of the 3D scene. Based on the control point quality marker data, available control points are screened and ground control point supplementary logs are generated. At the same time, the time consistency verification results of the image acquisition time data and the control point acquisition time data are archived to investigate whether the misalignment is due to external acquisition factors with inconsistent timing.

[0035] In this implementation plan, the projection consistency between images and terrain within the slope area is stably diverted and orderly managed, allowing data with different degrees of misalignment to enter direct modeling, constraint correction, or risk isolation paths respectively. This achieves pre-control and dynamic scheduling of the 3D modeling input quality. By deeply coupling the slope projection misalignment criterion value with the modeling process, it ensures that data meeting the projection consistency standard can be quickly accumulated into a reliable modeling foundation, while also ensuring that data with local misalignment is targeted for repair within a controlled range. At the same time, it implements effective isolation and source management for abnormal areas that exceed the credibility range. Overall, a modeling access and correction closed loop driven by consistency criteria is constructed, providing a systematic guarantee for the geometric stability, semantic reliability, and traceability of results in subsequent 3D scene construction.

[0036] Specifically, the iterative correction and convergence determination mechanism involves the following steps: Valid control points are selected based on the control point quality label data. By introducing quality level constraints and stability discrimination mechanisms, low-reliability control points with the risk of amplified observation noise or temporal inconsistencies are eliminated, ensuring that the control points involved in the calculation meet the correction requirements in terms of geometric accuracy and timeliness. Spatial calculations are performed on the image geographic reference data, mapping the control point coordinates to the image coordinate system. A rigorous geometric model is used to jointly solve the image exterior orientation elements and the relationship between the ground space, ensuring consistency and reversibility of the coordinate mapping process at both the overall scale and local regions. The difference in the projected positions of the control points in the image before and after correction is calculated to obtain the control point residual. By applying unified dimensional constraints and numerical stabilization to the residuals, they can objectively reflect the change in geometric consistency before and after the correction iteration.

[0037] The natural logarithm of the corrected structural information missing degree plus one is calculated to obtain the logarithmic modulation term of the corrected structural information. This logarithmic mapping compresses the numerical amplitude in the high-missing interval while enhancing the sensitivity of the correction process to changes in medium and low structural missing values. The base-e exponent of the corrected occlusion inconsistency is calculated and added to a constant to obtain the correction occlusion exponential enhancement term. The nonlinear amplification characteristic of the exponential function on occlusion differences is utilized to strengthen the constraint effect of occlusion relationships in slope projection consistency correction. The sum of the control point residuals and the correction direction deviation is added to one as a comprehensive suppression factor. The synergistic accumulation of residuals and direction deviations reflects the joint influence of geometric constraints and direction consistency on correction stability. The logarithmic modulation term of the corrected structural information and the correction occlusion exponential enhancement term are multiplied to obtain the consistency driving term, which is adjusted by the reciprocal of the comprehensive suppression factor to obtain the slope correction convergence value. This value characterizes the comprehensive stability of image texture and terrain structure under multiple consistency constraints in the current correction iteration. The specific calculation formula is as follows: ; In the formula, This represents the slope correction convergence value, which reflects the overall stability of image texture and terrain structure under multiple consistency constraints in the i-th correction iteration. Represents the control point residual, used to quantify the level of change in geometric consistency of control points in the image space before and after correction; It indicates the degree of missing structural information after correction, and is used to characterize the degree of improvement in information matching between the texture structure and terrain structure of the corrected image. It represents the degree of inconsistency in the correction of shading, and is used to measure the intensity of the impact of the adjustment of the slope shading relationship on the overall structural consistency during the correction iteration process; represents the deviation of the correction direction, used to describe the degree of residual deviation between the main direction of the image texture and the main direction of the terrain slope after correction; i represents the correction iteration index, used to identify the i-th iteration state in the current slope projection consistency correction process.

[0038] In this embodiment, Table 1 is an iterative feature data table of the slope correction convergence process, which records in detail the changes of key constraint factors and corresponding slope correction convergence values ​​in multiple iterations of slope projection consistency correction, and is used to quantify the consistency evolution state of image texture and terrain structure in the correction process. Specifically: In the first iteration, the missing information degree of the corrected structure was 0.65, the inconsistency of the corrected occlusion was 0.80, the control point residual was 0.45, the deviation of the corrected direction was 0.35, the comprehensive suppression factor was 1.80, and the convergence value of the slope correction was 0.8974; In the second iteration, the missing information degree of the corrected structure was 0.52, the inconsistency of the corrected occlusion was 0.65, the control point residual was 0.40, the deviation of the corrected direction was 0.30, the comprehensive suppression factor was 1.70, and the convergence value of the slope correction was 0.7181; In the third iteration, the missing information degree of the corrected structure was 0.40, the inconsistency of the corrected occlusion was 0.50, the control point residual was 0.32, the deviation of the corrected direction was 0.26, the comprehensive suppression factor was 1.58, and the convergence value of the slope correction was 0.564. 1; In the 4th iteration, the correction structural information missingness was 0.30, the correction occlusion inconsistency was 0.38, the control point residual was 0.25, the correction direction deviation was 0.20, the comprehensive suppression factor was 1.45, and the slope correction convergence value was 0.4455; In the 5th iteration, the correction structural information missingness was 0.22, the correction occlusion inconsistency was 0.28, the control point residual was 0.18, the correction direction deviation was 0.15, the comprehensive suppression factor was 1.33, and the slope correction convergence value was 0.3473; In the 6th iteration, the correction structural information missingness was 0.16, the correction occlusion inconsistency was 0.20, the control point residual was 0.14, the correction direction deviation was 0.12, the comprehensive suppression factor was 1.26, and the slope correction convergence value was 0.2617.

[0039] Table 1 Iterative characteristic data of the slope correction convergence process

[0040] like Figure 3 The diagram shows the convergence evolution curve of the slope correction convergence value. As shown in Table 1, the slope correction convergence value exhibits a continuous decrease and gradually slowing trend with increasing iterations. Specifically, the convergence value decreases significantly in the first to third iterations, indicating that missing structural information and occlusion inconsistencies are rapidly reduced in the early correction stages. The rate of decrease slows considerably in the fourth to sixth iterations, reflecting a gradual shift in the correction process towards fine-tuning subtle consistency issues. Overall, the slope correction convergence value evolution curve clearly reflects the transition from rapid to stable convergence in slope projection correction driven by multi-constraint consistency, serving as an important basis for determining correction termination and assessing the reliability of 3D scene construction.

[0041] In this implementation scheme, by unifying control point geometric constraints, structural consistency constraints, occlusion consistency constraints, and directional consistency constraints into a single convergence criterion framework, a comprehensive quantitative assessment of the stability of the slope projection correction process is achieved. This allows the correction iteration to move beyond relying on a single error index or empirical judgment, instead basing the judgment on the synergistic changes of multi-source consistency information. Consequently, it can sensitively identify the unstable state of structure and occlusion relationships in the early stages of correction, and effectively suppress the excessive influence of local residuals or directional deviations on the overall results in the later stages of correction. This results in a convergence criterion that gradually converges with iteration, exhibits smooth numerical changes, and has clear physical meaning. This mechanism ensures that slope projection consistency correction has a controllable convergence path and a reliable termination criterion under complex terrain conditions, providing crucial support for the stable implementation of texture attachment, boundary grounding, and micro-topography representation in subsequent 3D scene construction.

[0042] Specifically, the process of correcting image projection misalignment, dynamically adjusting local texture mapping relationships, and determining whether the correction process has terminated involves real-time comparison of slope correction convergence values ​​and slope correction convergence thresholds. The slope correction convergence thresholds include a first-level convergence threshold and a second-level convergence threshold. These thresholds are determined through confidence interval analysis based on the statistical distribution of convergence values ​​from historical convergence samples during the correction iteration process. The first-level convergence threshold corresponds to the convergence anomaly interval, with a value range of 0.60–0.80, while the second-level convergence threshold corresponds to the stable convergence interval, with a value range of 0.30–0.50. When the slope correction convergence value is less than the secondary convergence threshold, it is determined to be a correction convergence, indicating that the correction result under multiple constraints has reached a stable state and the residual deviation is within a controllable range. The corrected texture attachment result and correction parameters are output. The forest boundary polyline point set data is checked for closure according to the closed marker data to avoid topological destruction caused by local correction after the boundary is attached to the ground. The forest boundary data and vector land cover data after being attached to the ground are created and archived into the correction database and then entered into the 3D scene construction and output module.

[0043] When the slope correction convergence value is greater than or equal to the second-level convergence threshold and less than the first-level convergence threshold, it is determined that the correction can continue to be optimized. This indicates that the current correction result is not yet completely stable but has room for further improvement. Local texture coordinate remapping is then performed to refine the spatial matching relationship between the image and the terrain in a progressive manner. Linear structure extraction is enhanced in the boundary neighborhood corresponding to the land use zone boundary point set first. Local correction oscillations are suppressed by strengthening key structure constraints to stabilize the decreasing trend of linear structure deviation. The sliding window scale is adjusted according to the terrain elevation sampling interval data. The sliding window scale is determined according to the integer multiple relationship of the terrain elevation sampling interval data. Its value range is jointly constrained by the statistical results of the distance between points of the terrain elevation sampling interval data and the boundary point set of the land use zone. The spatial range covered by the window is not less than the spatial interval between adjacent elevation sampling points and does not exceed the average distance between points of the boundary point set of the land use zone in the local neighborhood. This ensures that the calculation scale of structure and occlusion features is consistent with the terrain resolution to maintain consistent calculation scale.

[0044] When the slope correction convergence value is greater than or equal to the first-level convergence threshold, it is judged as insufficient correction, indicating that automatic correction is difficult to converge effectively under the existing constraints, and there are structural or data source level problems. The misalignment hot zone is isolated as a verification zone to prevent high-risk areas from continuing to participate in the modeling process and amplifying the error. Control point residual logs, boundary topology consistency marker data verification results and correction parameters are generated. At the same time, the results of the corrected areas other than the verification zone are output into the 3D scene construction and output module.

[0045] In this implementation plan, a quantifiable, traceable, and controllable decision-making mechanism for correction termination and optimization is formed by implementing hierarchical judgment and dynamic scheduling of slope correction convergence values. This transforms the slope projection consistency correction process from a single result judgment to process management based on the evolution of convergence states. By introducing differentiated handling paths corresponding to different convergence intervals, reliable output of stable regions, directional optimization of improveable regions, and effective isolation of high-risk regions are achieved. This ensures the continuity of overall 3D modeling while preventing the spread and amplification of local errors in subsequent processes. This mechanism effectively improves the robustness and interpretability of the correction process, keeping the correction results under control in terms of geometric consistency, structural stability, and data availability, providing a clear and reliable quality grading basis for 3D scene construction and output.

[0046] Specifically, the process of constructing a 3D forestry survey scene based on projection correction and conducting consistency delivery analysis by integrating boundary ground-hugging properties, projection consistency, and micro-topography level preservation is as follows: a 3D land surface is constructed based on topographic elevation data; discrete elevation information is transformed into a computable spatial carrier through continuous topographic surface reconstruction; orthophotos after consistency correction are attached as textures to the 3D land surface, so that the image texture is strictly constrained by topographic geometry in space; forest boundary data and vector land cover data are loaded to achieve the collaborative expression of 3D geometric structure and survey semantic information; a 3D forestry survey scene model is constructed; and the current 3D forestry survey scene model is used as the 3D scene result. Forest boundary polyline point sets and land use zone boundary point sets are mapped onto the terrain surface using a ground-fitting projection algorithm. While maintaining the planar coordinates of the polyline point sequence, the 3D coordinates of the polyline points are retrieved and updated in the forest terrain elevation raster data according to pixel center alignment rules, forming a ground-fitted 3D boundary polyline point set. This avoids geometric inconsistencies such as overhangs, ground penetrations, and local flipping of the boundary. Edge detection is performed on the image grayscale pixel values ​​based on the corrected texture attachment results. Using the same mapping rules as the ground-fitting projection algorithm, the data is mapped onto the terrain surface to obtain forest edge candidate point sets. The nearest neighbor spatial distance between the ground-fitted 3D boundary polyline point set and the forest edge candidate point set is calculated in the terrain surface space, and the nearest neighbor spatial distance is normalized along the boundary length dimension to obtain the 3D boundary misalignment rate. The forest terrain elevation raster data and terrain elevation sampling are then used to obtain the final boundary misalignment rate. For the spacing data, a second-order curvature calculation algorithm was used to obtain curvature gratings from the uncorrected and corrected terrain elevation data. The curvature gratings were then binned to obtain curvature histograms. The number of bins in the curvature histogram was determined by the terrain elevation sampling spacing data and the curvature value range, ensuring that the spatial scale of a single bin was not less than the spatial interval between adjacent elevation sampling points. The curvature histograms were normalized according to the total frequency to obtain a curvature probability distribution. The Jensen-Shannon divergence algorithm was used to evaluate the distribution difference between the two curvature probability distributions, obtaining the curvature distribution difference value. Based on the statistical distribution range of the curvature distribution difference value in the corrected samples, an exponential decay normalization mapping relationship was determined. The curvature distribution difference value was converted into a fidelity index using an exponential decay normalization algorithm, represented by a negative exponential function of the difference value. The closer the curvature distribution difference value is to zero, the closer the micro-topography fidelity is to one, thus obtaining the micro-topography fidelity.

[0047] The difference between the 3D boundary misalignment rate and one is calculated to obtain the boundary consistency preservation term, which measures the overall preservation level of forest and land type boundaries after ground contact. The reciprocal of the slope projection misalignment criterion value plus one is calculated to obtain the slope projection consistency suppression term. Before entering the delivery evaluation, the slope projection misalignment criterion value has been limited to a dimensionless interval consistent with the 3D boundary misalignment rate through the mapping of plus one and the reciprocal, thus weakening the negative impact of slope projection misalignment on the delivery evaluation in an inverse proportion. The square root of the micro-topography fidelity plus one is calculated to obtain the micro-topography preservation enhancement term, which highlights the contribution of micro-scale terrain structure preservation to scene credibility through nonlinear amplification. The boundary consistency preservation term, the slope projection consistency suppression term, and the micro-topography preservation enhancement term are multiplied to obtain the 3D scene consistency delivery value. The specific calculation formula is as follows: ; In the formula, This represents the 3D scene consistency delivery value, which comprehensively reflects the overall delivery quality level of the 3D scene in terms of boundary ground consistency, slope projection consistency, and micro-topography preservation capabilities. It represents the three-dimensional boundary misalignment rate, used to describe the degree of spatial offset of the boundary semantics after it is attached to the ground in three dimensions; It represents the slope projection misalignment criterion value, which describes the overall projection consistency of image texture relative to the terrain structure under slope conditions; It represents the fidelity of micro-topography, used to assess the ability of the correction process to preserve the micro-scale undulations of the terrain.

[0048] This implementation plan unifies the consistency-corrected multi-source spatial data onto a 3D surface carrier, achieving stable fusion of geometric structure, image texture, and survey semantic information within the same spatial framework. This enables 3D forestry survey scenes to possess a complete quality expression capability that can be quantitatively assessed. By jointly measuring boundary ground-hugging consistency, slope projection consistency, and micro-topography preservation capability, a comprehensive delivery criterion is constructed that simultaneously reflects the macro-spatial matching status and the micro-topographic morphology preservation. This ensures that 3D scene results no longer rely solely on single visual or geometric indicators but possess an overall credibility judgment basis under multi-dimensional consistency constraints. This process effectively improves the controllability and stability of 3D scenes in terms of structural continuity, semantic accuracy, and topographic realism, providing a unified and reliable quantitative basis for the delivery evaluation, quality grading, and subsequent applications of forestry survey results.

[0049] Specifically, the process of making output, backtracking, and reconstruction decisions for 3D scene results based on the consistency delivery analysis results is as follows: Real-time comparison of 3D scene consistency delivery values ​​and 3D scene consistency delivery thresholds. The 3D scene consistency delivery thresholds include a primary delivery threshold and a secondary delivery threshold. The primary and secondary delivery thresholds are determined through quantile interval analysis based on the statistical distribution of consistency delivery values ​​in the 3D scene quality evaluation samples. The primary delivery threshold corresponds to the quality stability interval, with a value range of 0.60–0.80, and the secondary delivery threshold corresponds to the quality risk interval, with a value range of 0.30–0.50. When the consistency delivery value of the 3D scene is greater than or equal to the first-level delivery threshold, it is determined that the delivery requirements are met. This indicates that the 3D scene has reached a stable and reliable comprehensive state in terms of geometric structure, projection relationship and micro-topography representation. The output includes forestry survey 3D scene result files, slope projection misalignment criterion value distribution report, boundary misalignment report and micro-topography fidelity report. This provides complete data support for survey planning, analysis and decision-making and result application. The image acquisition time data, control point acquisition time data, sensor observation angle data and image geographic benchmark data are archived into the modeling database.

[0050] When the 3D scene consistency delivery value is greater than or equal to the second-level delivery threshold and less than the first-level delivery threshold, it is determined that the delivery can be improved. This indicates that the overall 3D scene has basic usability, but there is still room for optimization in some local areas. The process is traced back to the projection consistency constraint iteration and convergence control module: for partitions where the 3D boundary misalignment rate is greater than the average value, the forest boundary closure marker data constraint is enhanced to participate in the correction. By strengthening the key boundary constraints, the unstable propagation of boundary semantics in the correction process is suppressed. For partitions where the slope projection misalignment criterion value is greater than the average value, the control point quality marker data is preferentially called to filter quality control points and adjust the correction partition range to improve the geometric reliability and convergence efficiency of local correction. The micro-topography fidelity is recalculated and the archive is updated.

[0051] When the 3D scene consistency delivery value is less than the secondary delivery threshold, it is determined that the delivery requirements are not met. This indicates that there are systemic risks in the current 3D scene in terms of boundary grounding, slope projection consistency, or micro-topography preservation. A delivery anomaly database is created and the current 3D scene results, zonal-level slope projection misalignment criterion values, 3D boundary misalignment rates, and micro-topography fidelity reports are archived to the delivery anomaly database. At the same time, a check area list and a log of ground control point supplementation suggestions are output. The publishing of vector land use category attribute data-driven layers is suspended in the check area to prevent erroneous semantics from being used in survey planning. After the check, the projection consistency constraint iteration and convergence control module is entered again for correction.

[0052] like Figure 4The diagram illustrates the 3D scene consistency delivery judgment process. Using the 3D scene consistency delivery value as the core driving force, a hierarchical decision-making and backtracking optimization process with closed-loop feedback capabilities is constructed. By mapping evaluation results to different processing branches, an orderly connection is achieved from risk isolation and local optimization to deliverables: In the judgment phase, different threshold ranges correspond to different handling paths, distinguishing between high-risk states requiring interruption and review, optimizable states requiring backtracking optimization, and stable states that can directly proceed to deliverables; in the execution phase, each branch clearly defines its subsequent actions and data flow, ensuring that abnormal scenarios do not directly enter the deliverables release stage while ensuring that optimizable scenarios can form an iterative improvement loop through backtracking correction; in the results phase, all paths ultimately form traceable records through result archiving and status updates, providing a basis for subsequent modeling optimization, quality assessment, and process auditing. This process design transforms 3D scene delivery from a one-time judgment into a cyclical, backtrackable, and quantifiable dynamic decision-making process, thereby improving overall delivery reliability and engineering controllability.

[0053] This implementation plan introduces a tiered judgment mechanism for the consistency of 3D scene deliverables, enabling systematic control and closed-loop decision-making over the delivery status of 3D scene results. This transforms the delivery process from a single output to a traceable, retrospective, and optimizable quality management workflow. By adopting differentiated processing strategies for different delivery intervals, it effectively distinguishes between 3D scene states requiring review and recalibration, those requiring continuous optimization, and those ready for direct delivery, preventing the spread of quality risks during the results release and application stages. This process ensures that high-risk areas are promptly isolated and corrected while fully releasing the application value of stable areas, improving the reliability, consistency, and engineering controllability of 3D scene results in investigation planning, analysis and decision-making, and subsequent data utilization.

[0054] like Figure 2As shown, the second aspect of this invention provides a forestry survey 3D scene modeling system based on multi-source data fusion, comprising: an acquisition and preprocessing module for acquiring orthophoto remote sensing image data and spatial modeling constraint data, and preprocessing the orthophoto remote sensing image data and spatial modeling constraint data; a slope projection consistency discrimination module for performing multi-factor projection consistency discrimination analysis on the orthophoto remote sensing image data and spatial modeling constraint data, identifying the degree of misalignment between the image and the terrain in the slope area based on the multi-factor projection consistency discrimination analysis results, and triggering a hierarchical processing strategy; a projection consistency constraint iteration and convergence control module for correcting image projection misalignment through an iterative correction and convergence determination mechanism, dynamically adjusting local texture mapping relationships, and determining whether the correction process has terminated; and a 3D scene construction and output module for constructing a forestry survey 3D scene based on projection correction, conducting consistency delivery analysis by fusing boundary ground-hugging properties, projection consistency, and micro-topography level, and executing output, retrospective, and reconstruction decisions of the 3D scene results based on the consistency delivery analysis results.

[0055] This implementation plan achieves closed-loop control over the entire process of orthorectified remote sensing image data and spatial modeling constraint data, from acquisition, analysis, and correction to deliverables. This transforms the construction of 3D forestry survey scenes from traditional static modeling to a dynamic modeling process with discrimination, correction, and decision-making capabilities. The system can identify projection misalignment risks between imagery and terrain in the early stages of modeling, gradually reduce geometric deviations through constraint-driven iterative correction in intermediate stages, and uniformly evaluate and classify boundary grounding, projection consistency, and micro-topography preservation in the deliverables stage. This ensures that the output 3D forestry survey scenes possess a stable and consistent comprehensive quality foundation in terms of spatial geometry, structural representation, and application reliability, providing reliable 3D scene support for subsequent survey planning, results application, and data reuse.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0057] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for three-dimensional scene modeling in forestry surveys based on multi-source data fusion, characterized in that, Includes the following steps: S1, Collect orthorectified remote sensing image data and spatial modeling constraint data, and preprocess the orthorectified remote sensing image data and spatial modeling constraint data; S2, perform multi-factor projection consistency analysis on orthophoto remote sensing image data and spatial modeling constraint data, identify the degree of misalignment between image and terrain in the slope area based on the results of multi-factor projection consistency analysis and trigger a hierarchical processing strategy; S3 corrects image projection misalignment through an iterative correction and convergence determination mechanism, dynamically adjusts local texture mapping relationships, and determines whether the correction process has terminated. S4. Based on projection correction, a 3D forestry survey scene is constructed. Consistency delivery analysis is carried out by integrating boundary ground-fitting properties, projection consistency, and micro-topography level preservation. Based on the consistency delivery analysis results, decisions are made on the output, retrospection, and reconstruction of the 3D scene results. The specific process of performing multi-factor projection consistency discrimination analysis on orthorectified remote sensing image data and spatial modeling constraint data is as follows: The Sobel gradient operator was used to obtain the texture gradient grayscale map from the image grayscale pixel data. A slope shadow grayscale map was obtained from the forest area topographic elevation raster data combined with solar altitude and azimuth data using a slope shadow geometric projection algorithm. Within a sliding window, normalized mutual information was calculated between the texture gradient grayscale map and the slope shadow grayscale map, and the structural information missingness was obtained by subtracting one from the mutual information value. Second-order curvature calculation and flow direction accumulation analysis were used to obtain the linear structure of the terrain from the forest area topographic elevation raster data. The multi-band pixel data from the image was enhanced by vegetation index and subjected to Canney edge detection. The linear structure of the image is acquired, and the average nearest distance is calculated through nearest-nearest-distance matching to obtain the linear structure deviation. Slope shadow masks are generated using solar altitude angle data, solar azimuth angle data, and forest terrain elevation raster data. Image shadow masks are generated using image grayscale pixel value data. Viewpoint consistency screening is performed by combining sensor observation angle data, and the mask difference ratio is statistically analyzed to obtain the occlusion inconsistency. The main slope direction is calculated from the forest terrain elevation raster data, and the main texture direction is calculated from the image grayscale pixel value data. The angle between the two directions is normalized to obtain the direction deviation. Calculate the natural logarithm of the missing structural information degree plus one to obtain the logarithmic modulation term of the structural information; Calculate the exponential value of the deviation of the linear structure to obtain the exponential amplification term of the linear structure; Calculate the square root of the occlusion inconsistency plus one to obtain the occlusion consistency adjustment term; The reciprocal of the direction deviation plus one is calculated to obtain the direction consistency suppression term; the logarithmic modulation term of structural information, the exponential amplification term of linear structure, the shading consistency adjustment term and the direction consistency suppression term are multiplied to obtain the slope projection misalignment criterion value; The specific process of identifying the degree of misalignment between the image and the terrain within the slope area based on the results of multi-factor projection consistency discriminant analysis and triggering a graded processing strategy is as follows: Real-time comparison of slope projection misalignment criterion values ​​and slope projection misalignment criterion thresholds. The slope projection misalignment criterion thresholds include primary criterion thresholds and secondary criterion thresholds: When the slope projection misalignment criterion value is less than the secondary criterion threshold, the orthophoto remote sensing image data, topographic elevation data, vector land cover data, forest boundary data and ground control point data are created and archived into the modeling database and then entered into the 3D scene construction and output module. When the slope projection misalignment criterion value is greater than or equal to the secondary criterion threshold and less than the primary criterion threshold, a misalignment hot zone raster is generated based on the slope projection misalignment criterion value and the correction zone is divided in combination with the land use zoning boundary point set data. The misalignment hot zone and the corresponding correction zone are the correction range. Based on the ground control point data, the local texture coordinate remapping is performed on the orthophoto remote sensing image in the slope area, and linear structure consistency constraints and orientation consistency constraints are introduced. The process then enters the projection consistency constraint iteration and convergence control module. When the slope projection misalignment criterion value is greater than or equal to the first-level criterion threshold, the misalignment hot zone is isolated to participate in the direct texture attachment output, and available control points are selected based on the control point quality marker data and a ground control point supplementary log is generated. The specific process of the iterative correction and convergence determination mechanism is as follows: Valid control points are selected based on the control point quality marker data. Spatial calculations are performed on the image geographic reference data to map the control point coordinates to the image coordinate system. The difference between the projection positions of the control points in the image before and after correction is calculated to obtain the control point residual. Calculate the natural logarithmic value of the missing information of the corrected structure and add one to obtain the logarithmic modulation term of the corrected structure information; calculate the exponential value of the inconsistency of the corrected shading and add it to one to obtain the exponential enhancement term of the corrected shading; sum the control point residuals and the deviation of the correction direction and add one to it as the comprehensive suppression factor; multiply the logarithmic modulation term of the corrected structure information and the exponential enhancement term of the corrected shading as the consistency driving term, and adjust it with the reciprocal of the comprehensive suppression factor to obtain the slope correction convergence value.

2. The method for three-dimensional scene modeling of forestry surveys based on multi-source data fusion according to claim 1, characterized in that: The specific process for acquiring orthorectified remote sensing image data and spatial modeling constraint data is as follows: The orthophoto remote sensing image data is collected, which includes: multi-band pixel value data, grayscale pixel value data, image acquisition time data, solar altitude angle data, solar azimuth angle data, sensor observation angle data, and image geographic reference data. Collect spatial modeling constraint data, which includes: topographic elevation data, vector land cover data, forest boundary data, and control point data; The topographic elevation data includes forest area topographic elevation raster data and topographic elevation sampling interval data; vector land cover data includes land cover zone boundary point set data and land cover category attribute data; forest land boundary data includes forest land boundary polyline point set data, forest land boundary closure marker data and boundary topology consistency marker data; and control point data includes ground control point data, control point acquisition time data and control point quality marker data.

3. The method for three-dimensional scene modeling of forestry surveys based on multi-source data fusion according to claim 1, characterized in that: The specific process for preprocessing orthorectified remote sensing image data and spatial modeling constraint data is as follows: For terrain elevation data, a sliding window midpoint filtering algorithm and a robust regression correction algorithm for local outliers are used to correct spikes and collapses; for terrain elevation data and orthophoto remote sensing image data, a unified grid resampling algorithm and pixel center alignment rules are used to achieve resolution consistency and pixel alignment. Topological consistency is restored for vector land use data and forest boundary data through topological self-checking and self-intersection resolution algorithms. Self-intersection, duplicate points, fracture endpoints and pseudo-closed loops are detected and resolved for land use zone boundary point set data and forest boundary polyline point set data. The ground-hugging stability of the boundary on the slope is improved by the boundary polyline uniform densification algorithm. Ground control point data are screened for quality using a three-dimensional coordinate consistency verification algorithm and an outlier removal algorithm, and unified to a spatial reference consistent with orthophoto remote sensing data. Orthophoto remote sensing data and spatial modeling constraint data are standardized and normalized using distribution standardization and linear normalization algorithms.

4. The method for three-dimensional scene modeling of forestry surveys based on multi-source data fusion according to claim 1, characterized in that: The specific process of correcting image projection misalignment, dynamically adjusting local texture mapping relationships, and determining whether the correction process has terminated is as follows: Real-time comparison of slope correction convergence values ​​and slope correction convergence thresholds, including primary convergence thresholds and secondary convergence thresholds: When the slope correction convergence value is less than the second-level convergence threshold, the corrected texture attachment result and correction parameters are output. The forest boundary polyline point set data is checked for closure according to the closed marker data. The forest boundary data after ground attachment and vector land cover data are created and archived into the correction database and then entered into the 3D scene construction and output module. When the slope correction convergence value is greater than or equal to the second-level convergence threshold and less than the first-level convergence threshold, local texture coordinate remapping continues, linear structure extraction is enhanced in the boundary neighborhood corresponding to the land use partition boundary point set, and the sliding window scale is adjusted according to the topographic elevation sampling interval data. When the slope correction convergence value is greater than or equal to the first-level convergence threshold, the misalignment hot zone is isolated as the verification zone, and the control point residual log, boundary topology consistency marker data verification results and correction parameters are generated. At the same time, the results of the corrected areas other than the verification zone are output, and the process enters the 3D scene construction and output module.

5. The method for three-dimensional scene modeling of forestry surveys based on multi-source data fusion according to claim 1, characterized in that: The specific process of constructing a 3D forestry survey scene based on projection correction and conducting consistency delivery analysis by integrating boundary ground-hugging properties, projection consistency, and micro-topography preservation is as follows: A three-dimensional land surface is constructed based on topographic elevation data. Orthophotos with consistency correction are applied as textures to the three-dimensional land surface. Forest boundary data and vector land cover data are loaded to construct a three-dimensional scene model for forestry surveys. This model is then used as the final three-dimensional scene result. Forest boundary polyline point sets and land cover zone boundary point sets are mapped to the topographic surface using a ground-mounted projection algorithm. Based on pixel center alignment rules, ground-mounted three-dimensional boundary polyline point sets are formed. Edge detection is performed on the image grayscale pixel values ​​based on the corrected texture application results and mapped to the topographic surface. The spatial deviation between the corrected and the three-dimensional boundary polyline point sets is calculated and normalized to obtain the three-dimensional boundary misalignment rate. The curvature probability distribution of the topographic elevation data before and after correction is calculated and normalized using Jensen-Shannon divergence and exponential decay to obtain the micro-topographic fidelity. The difference between the 3D boundary misalignment rate and one is calculated to obtain the boundary consistency preservation term; the reciprocal of the slope projection misalignment criterion value plus one is calculated to obtain the slope projection consistency suppression term; the square root of the micro-topography fidelity plus one is calculated to obtain the micro-topography preservation enhancement term; the boundary consistency preservation term, the slope projection consistency suppression term, and the micro-topography preservation enhancement term are multiplied to obtain the 3D scene consistency delivery value.

6. The method for three-dimensional scene modeling of forestry surveys based on multi-source data fusion according to claim 5, characterized in that: The specific process for making decisions on the output, backtracking, and reconstruction of 3D scene results based on the consistency delivery analysis results is as follows: Real-time comparison of 3D scene consistency delivery values ​​and 3D scene consistency delivery thresholds. The 3D scene consistency delivery thresholds include primary delivery thresholds and secondary delivery thresholds. When the 3D scene consistency delivery value is greater than or equal to the first-level delivery threshold, output the forestry survey 3D scene result file, slope projection misalignment criterion value distribution report, boundary misalignment report and micro-topography fidelity report, and archive the image acquisition time data, control point acquisition time data, sensor observation angle data and image geographic benchmark data into the modeling database; When the 3D scene consistency delivery value is greater than or equal to the second-level delivery threshold and less than the first-level delivery threshold, the process is backtracked to the projection consistency constraint iteration and convergence control module: for partitions where the 3D boundary misalignment rate is greater than the average value, forest boundary closure marker data constraints are introduced to participate in the correction; for partitions where the slope projection misalignment criterion value is greater than the average value, control point quality marker data is called to filter control points and adjust the correction partition range; micro-topography fidelity is recalculated and archive is updated. When the 3D scene consistency delivery value is less than the secondary delivery threshold, a delivery anomaly database is created and the current 3D scene output files, zonal slope projection misalignment criterion values, 3D boundary misalignment rate and micro-topography fidelity report are archived. A check area list and ground control point supplementation suggestion log are output. The publishing of vector land category attribute data-driven layers is suspended in the check area.

7. A forestry survey 3D scene modeling system based on multi-source data fusion, employing the forestry survey 3D scene modeling method based on multi-source data fusion as described in any one of claims 1-6, comprising: The acquisition and preprocessing module is used to acquire orthorectified remote sensing image data and spatial modeling constraint data, and to preprocess the orthorectified remote sensing image data and spatial modeling constraint data. The slope projection consistency discrimination module is used to perform multi-factor projection consistency discrimination analysis on orthophoto remote sensing image data and spatial modeling constraint data. Based on the results of the multi-factor projection consistency discrimination analysis, it identifies the degree of misalignment between the image and the terrain in the slope area and triggers a hierarchical processing strategy. The projection consistency constraint iteration and convergence control module is used to correct image projection misalignment through an iterative correction and convergence determination mechanism, dynamically adjust local texture mapping relationship and determine whether the correction process has terminated. The 3D scene construction and output module is used to construct a 3D scene for forestry surveys based on projection correction. It conducts consistency delivery analysis by integrating boundary ground-fitting properties, projection consistency, and micro-topography level preservation. Based on the consistency delivery analysis results, it executes decisions on the output, retrospection, and reconstruction of the 3D scene results.

Images