Feature matching and three-dimensional information collaborative building change grading recognition method and system

Abstract

The application belongs to the field of remote sensing monitoring and building change detection, and discloses a building change grading recognition method and system based on feature matching and three-dimensional information cooperation. The application realizes accurate association of cross-time building instances by constructing an adaptive matching function that fuses global weight and local geometric features. The application generates a three-dimensional point set by Poisson disc sampling, extracts elevation, gradient and other features, and filters out non-building interference. The application outputs results by two-level decision logic, includes changes that do not reach a significant level but have monitoring qualifications into a continuous tracking list, and realizes the capture of small changes in time series accumulation. The application can systematically suppress false changes caused by vegetation, noise and the like, significantly improve detection accuracy and reliability, and is suitable for high-precision progressive change discovery scenarios such as urban renewal and illegal building control.

Description

Technical Field

[0001] This application relates to the field of remote sensing monitoring and building change detection technology, and in particular to a method and system for hierarchical identification of building changes based on feature matching and three-dimensional information collaboration. Background Technology

[0002] With the acceleration of urbanization, automatic building change detection based on multi-temporal remote sensing imagery has become a core technical means for land spatial planning, illegal building control, and urban renewal monitoring.

[0003] However, existing automated detection processes generally suffer from high false change rates and insufficient reliability in their initial outputs, severely limiting their direct application in practical business. Specifically, existing technologies mainly face the following three bottlenecks:

[0004] Firstly, at the temporal correlation level, most methods rely on fixed intersection-union ratio thresholds or simple spatial overlay relationships to match the temporal correspondence of building instances. They cannot adaptively adjust the matching strategy according to the actual geometric scale of the building (such as large-area factory buildings and small-area ancillary facilities), resulting in a high mismatch rate in scenarios with large deformation or irregular edges, making it difficult to establish accurate spatiotemporal correspondence.

[0005] Secondly, in terms of change discrimination, traditional three-dimensional verification methods are often limited to simply comparing whether the elevation difference between two digital surface models exceeds a certain static threshold. They lack in-depth exploration of the essential characteristics of the target's geometric structure and are prone to misjudging elevation fluctuations caused by seasonal vegetation growth, temporary stockpiles, or data noise as real building changes. At the same time, they are also difficult to distinguish between flat building roofs and undulating vegetation canopies.

[0006] Third, at the decision-making mechanism level, existing change identification methods generally adopt a single-period static binary judgment logic (i.e., "change" or "no change"), which lacks the ability to track and accumulate gradual small changes (such as local construction during building addition, roof renovation, etc.), resulting in a large number of real small changes that do not reach the single detection threshold being directly filtered out, causing missed detections.

[0007] In addition, while some change recognition schemes based on deep learning black box models can learn complex features, their decision-making process is not transparent, and their generalization ability and robustness in complex urban scenarios still need to be improved. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this application aims to provide a method and system for hierarchical recognition of building changes based on feature matching and 3D information collaboration. This method, using multi-temporal building outline vectors and corresponding DSM data output from an automated process as input, (1) achieves precise temporal correlation between building targets of different scales and shapes in different time phases by integrating a globally preset weight and an adaptive instance matching scoring function based on local geometric features, overcoming the problem of high mismatch rates in fixed threshold matching scenarios with large / small deformed targets; (2) distinguishes between "smooth, regular building roofs" and "undulating vegetation canopies / disorderly temporary objects / occlusion noise" from a 3D structural dimension by uniformly sampling three-dimensional points on a Poisson disk within the candidate area and combining geometric continuity features such as elevation standard deviation, local slope, and DSM void rate. This systematically filters out non-structural pseudo-changes; and (3) through a two-level logical judgment consisting of a monitoring qualification threshold and a change significance threshold, small changes that do not reach a significant level but have a monitoring basis are included in the continuous monitoring list for cross-temporal cumulative tracking and closed-loop re-judgment, thereby effectively suppressing multi-source pseudo-changes while achieving no omissions and accurate hierarchical output of gradual small building changes (such as partial reconstruction, layer-by-layer addition, etc.).

[0009] To achieve the above objectives, this application adopts the following technical solution:

[0010] This application provides a method for hierarchical identification of building changes based on feature matching and 3D information, including the following steps:

[0011] Acquire multi-temporal building outline vector data and corresponding digital surface model data, wherein the building outline vector data includes geometric boundary information of building instances;

[0012] The adaptive instance matching step is to construct an adaptive matching score function that integrates global preset weights and local geometric features. The matching score function comprehensively considers the contour overlap, contour center distance and shape similarity between any two building instances. Based on the adaptive matching score function, the building instances of the current time phase are associated with the building instances of the previous time phase and the targets in the continuous monitoring list, thereby dividing the successfully matched area, the newly added candidate area and the disappeared candidate area.

[0013] Perform three-dimensional sampling and credibility analysis steps: For the successfully matched area, newly added candidate area and disappeared candidate area, calculate the average elevation difference between different time phases, and perform three-dimensional point sampling within the geometric boundary of each candidate area to generate a uniformly distributed set of sampling points. Based on the set of sampling points, extract multi-dimensional features that characterize the continuity of the geometric structure. Based on the multi-dimensional features, identify and filter out pseudo changes caused by unstructured objects, and retain the change areas with structural features.

[0014] Execute multi-threshold hierarchical decision-making steps: For the retained change area, combine its area change and height change, and use the preset monitoring eligibility threshold and change significance threshold to perform two-level logical judgment to divide the change area into three categories: confirmed change, change that needs to be continuously monitored, and invalid change;

[0015] Update the continuous monitoring list: Store the target information corresponding to the changes that are determined to require continuous monitoring into the continuous monitoring list, and use the continuous monitoring list as the input data source for the adaptive instance matching step in the next time phase processing, so as to realize closed-loop tracking across time phases.

[0016] As a preferred technical solution, in the adaptive instance matching step, the adaptive matching scoring function adopts a two-layer weighting mechanism:

[0017] Define a global basic weight, which is determined by machine learning training based on the priority of three factors: contour overlap, contour center distance, and shape similarity. This global basic weight is used to reflect the general importance of different features in the global scope.

[0018] Define local adaptive parameters, which are dynamically calculated based on the specific geometric attributes of the two building instances to be matched. The geometric attributes include the degree of area similarity between the two instances, the proportional attenuation factor based on the centroid distance, and the degree of similarity in the outline perimeter, which are used to reflect the local geometric characteristics of the current instance pair.

[0019] Based on the product of the global basic weights and the local adaptive parameters, the dynamic weight coefficients corresponding to each feature item in the adaptive matching score function are calculated in a normalized manner, so that the matching score function can adaptively adjust the contribution ratio of each feature item according to building instances of different scales and shapes, thereby improving the accuracy of temporal correlation.

[0020] As a preferred technical solution, the area similarity in the local adaptive parameters is obtained by calculating the ratio of the smaller to the larger area of ​​the two instances. The proportional decay factor in the local adaptive parameters is obtained by normalizing the centroid distance by introducing the square root of the average area of ​​the two instances as a normalization factor. The contour perimeter similarity in the local adaptive parameters is obtained by calculating the ratio of the smaller to the larger perimeter of the two instances. The adaptive matching score function integrates the normalized contour overlap index, the centroid distance index processed by the Gaussian kernel function, and the shape similarity index transformed by the normalized Hausdorf distance through weighted summation, and finally outputs a matching score value between zero and one.

[0021] As a preferred technical solution, in the three-dimensional sampling and credibility analysis step, the three-dimensional point sampling adopts the Poisson disk sampling method. By generating a set of sampling points with uniform spacing inside the geometric polygon of the candidate region, it ensures that the sampling points uniformly cover the candidate region and avoids verification bias caused by uneven sampling density. The multidimensional features characterizing the continuity of the geometric structure include at least: the elevation standard deviation feature for describing the dispersion of the elevation values ​​of the sampling points, the local gradient feature for reflecting the changes in surface slope, and the void ratio feature for statistically analyzing the proportion of invalid elevation data. By jointly analyzing the above multidimensional features, it is possible to distinguish between building changes with a smooth surface and continuous geometric structure and non-structural pseudo-changes with drastic surface undulations, chaotic slopes, or severe shading.

[0022] As a preferred technical solution, the local gradient feature is calculated based on a triangular network constructed from the sampling point set. It is obtained by averaging the slope values ​​of all triangular faces in the triangular network, and is used to quantify the flatness of the candidate region's surface. The void ratio feature is calculated by statistically analyzing the proportion of elevation values ​​that are null or outlier among the sampling points in the candidate region, and is used to identify areas caused by vegetation obstruction or missing data. During the credibility analysis process, if a candidate region simultaneously meets the conditions that its elevation dispersion is below a preset upper limit, its surface slope is below a preset upper limit, and its void ratio is below a preset upper limit, then the region is determined to be a building change region with a real geometric structure; otherwise, it is determined to be a non-structural pseudo-change and is removed.

[0023] As a preferred technical solution, the building change hierarchical recognition method based on feature matching and 3D information synergy described in claim 1 is characterized in that the multi-threshold hierarchical decision-making step includes:

[0024] Define the monitoring eligibility determination logic to determine whether the area change of the changed area reaches the preset area monitoring threshold or the height change reaches the preset height monitoring threshold. If either condition is met, the area is determined to be eligible for monitoring.

[0025] Define the logic for determining the significance of changes, and determine whether the area change of a qualified monitoring area reaches a preset threshold for significant area change, or whether the height change reaches a preset threshold for significant height change. If either condition is met, the change in the area is determined to be significant.

[0026] Based on the Boolean combination of the monitoring eligibility determination logic and the change significance determination logic, the final hierarchical decision result is output.

[0027] As a preferred technical solution, the rule for determining the hierarchical decision-making result is as follows:

[0028] If the changed area is qualified for monitoring and the change is significant, it is determined to be a confirmed change and is directly output as a high-confidence detection result;

[0029] If a changed area is eligible for monitoring but its change does not meet the significance standard, it is determined to be a change that needs to be continuously monitored. The geometric boundary, area and elevation characteristics of the area are stored in the continuous monitoring list so that it can re-participate in adaptive instance matching and verification in subsequent time-phase processing to achieve time-series cumulative judgment of minor changes.

[0030] If the area of ​​change is not eligible for monitoring, it is deemed an invalid change and will be filtered out without further processing.

[0031] As a preferred technical solution, before performing the adaptive instance matching step, a data preprocessing step is also included, which is used to perform topology checks and repairs on the input multi-temporal building outline vector data, and to denoise and fill holes in the digital surface model data; after performing the three-dimensional sampling and credibility analysis step, a spatial topology consistency verification step is also included, which is used to check whether the verified change area has spatial conflicts with features such as roads, water bodies or green spaces. If a conflict occurs, the change area is marked or removed to further improve the geographic logical rationality of the output results.

[0032] As a preferred technical solution, the dynamic maintenance mechanism for the continuous monitoring list includes:

[0033] After each round of time phase processing is completed, the continuous monitoring list is updated, removing targets that have been determined to be confirmed changes or invalid changes, and retaining targets that are still under monitoring.

[0034] When matching targets in the continuous monitoring list with the current time-phase instance, targets in the continuous monitoring list are given a higher matching priority or matching tolerance to accommodate the slow deformation or gradual change of building form that may occur during long-term monitoring, and to prevent mismatch of historical monitoring targets due to minor geometric differences.

[0035] This application also provides a building change hierarchical recognition system that combines feature matching and 3D information, the system comprising:

[0036] The data acquisition module is used to acquire multi-temporal building outline vector data and corresponding digital surface model data;

[0037] The matching module is used to construct an adaptive matching score function that integrates global preset weights and local geometric features. Based on this function, the building instances in the current time phase are associated with the building instances in the previous time phase and the targets in the continuous monitoring list, and the matching successful areas, newly added candidate areas and disappeared candidate areas are output.

[0038] The analysis module is used to calculate the average elevation difference of each candidate region and perform three-dimensional point sampling, extract multi-dimensional features that characterize the continuity of the geometric structure, and filter out non-structural pseudo-changes based on the multi-dimensional features.

[0039] The decision module combines area change and height change, uses monitoring eligibility threshold and change significance threshold to perform two-level logical judgment, and outputs graded results for confirmed changes, changes requiring continuous monitoring, and invalid changes.

[0040] The storage and update module is used to store the continuous monitoring list and update the continuous monitoring list with changes that are determined to require continuous monitoring, so that they can be called in the next phase of processing.

[0041] Compared with the prior art, the beneficial effects of this application are as follows:

[0042] By constructing an adaptive matching scoring function that integrates global preset weights and local geometric features, this application can dynamically adjust the contribution weights of contour overlap, center distance and shape similarity in the final score according to the specific area, centroid distance and perimeter of the building instance to be matched. This breaks through the limitations of fixed threshold matching, improves the temporal correlation accuracy between buildings of different scales and forms in previous and subsequent time phases, and lays a reliable foundation for instance correspondence for subsequent analysis.

[0043] At the three-dimensional perception level, this application abandons the single elevation difference threshold filtering method and instead uses Poisson disk sampling to generate a uniformly distributed three-dimensional point set in the candidate area, and extracts multi-dimensional geometric structure features such as elevation standard deviation, local gradient and void ratio. The candidate area is verified in three dimensions from the three dimensions of surface flatness, slope continuity and data integrity. This can effectively identify and filter out unstructured pseudo changes caused by vegetation, temporary objects and data noise, and significantly reduce the false alarm rate.

[0044] At the multi-threshold hierarchical decision-making level, this application innovatively introduces a two-level judgment logic of monitoring eligibility and change significance. Combined with a dynamically updated continuous monitoring list, it continuously tracks and accumulates judgments across time phases for targets that have monitoring potential but do not reach a significant level (such as minor elevation changes). This avoids directly filtering out minor real changes as noise and prevents the misreporting of minor fluctuations that do not reach engineering significance as formal changes. It resolves the contradiction between suppressing false changes and capturing gradual changes, and achieves accurate hierarchical classification and closed-loop management of change results. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the steps of the hierarchical identification method for building changes based on feature matching and 3D information collaboration according to the present invention.

[0046] Figure 2 This is a schematic diagram of adaptive instance matching;

[0047] Figure 3 A schematic diagram illustrating the reliability analysis of sub-meter level elevation point sampling;

[0048] Figure 4 This is a graph showing the results of a credibility analysis based on sub-meter elevation sampling points.

[0049] Figure 5 This is a schematic diagram of the logic judgment of a multi-threshold hierarchical monitoring system. Detailed Implementation

[0050] To enable those skilled in the art to better understand the present application, the technical solutions in specific embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0051] This application provides a method for hierarchical recognition of building changes through feature matching and 3D information collaboration.

[0052] The premise for implementing the method of this invention is that multi-temporal building outline vector data and corresponding digital surface model (DSM) data have been obtained through existing automated building extraction processes. The building outline vector data contains the geometric boundary information of the building instance.

[0053] In practice, the input data comes from existing automated building extraction processes. The system receives data from the target area at two different time phases, denoted as T1 and T2. This data includes vector data of building instance outlines composed of polygons, and precisely registered digital surface model (DSM) data. The DSM data records elevation information of the ground surface, including building roofs and vegetation canopies. The system initializes with a continuously monitored list, which is empty on the first execution and automatically updated after each processing iteration.

[0054] like Figure 1 As shown, the method for hierarchical recognition of building changes through feature matching and 3D information synergy includes the following steps:

[0055] S1, Perform the adaptive instance matching step.

[0056] An adaptive matching score function is constructed that integrates globally preset weights and local geometric features. This function comprehensively considers the contour overlap, contour center distance, and shape similarity between any two building instances. Based on the adaptive matching score function, building instances in the current time phase are associated with building instances in the previous time phase and targets in the continuous monitoring list. This process delineates successfully matched regions, newly added candidate regions, and disappeared candidate regions.

[0057] like Figure 2As shown, the core of this step is to establish a cross-temporal correspondence between building targets. First, the matching score S is calculated, and its expression is:

[0058] .

[0059] in For intersection, union, and comparison, and The coordinates of the center point of the contour are: The average area of ​​the two contours. For normalized Hausdorff distance. Weights , , A two-layer adaptive mechanism is used for calculation. Unlike fixed weights, the weights in this invention... , , The system uses a two-layer adaptive mechanism for dynamic calculation, which integrates global preset weights and local geometric feature adaptive coefficients.

[0060] Through geometric overlap ( ), centroid proximity ( Boundary similarity The three features are fused to calculate the matching score, establish the temporal correspondence between building instances in different time periods, and achieve accurate initial screening of candidate areas of change.

[0061] Step S1.1: Preset global basic weights.

[0062] First, preset the global base weights. , Global base weights , , In this embodiment, the preferred method is determined by using logistic regression to maximize matching accuracy. , , .

[0063] Local adaptive parameters , , Calculate geometric properties based on examples.

[0064] Area adaptive coefficient: This reflects the degree to which the areas are similar. The value ranges from (0,1), with 1 for equal areas and a smaller value for larger differences.

[0065] Centroid adaptive coefficient: This reflects the decrease in the distance between the centroids. When the distance is 0, it takes the value 1; when the distance approaches infinity, it approaches infinity.

[0066] Perimeter adaptive coefficient: This reflects the similarity of shapes. These are the perimeters of the two instances, with values ​​ranging from (0,1]. The value is 1 when the perimeters are equal, and the value decreases as the difference increases.

[0067] Final weight , , The matching threshold is set to 0.6. Regions with a score greater than or equal to 0.6 are considered successfully matched regions, while regions with a score less than 0.6 are considered newly added or disappeared candidate regions.

[0068] Step S1.2: Eigenvalue calculation.

[0069] right Perform a spatial Boolean operation on the outlines of two building instances to calculate the ratio of the area of ​​their intersection to the area of ​​their union, i.e.:

[0070] .

[0071] The value range is [0,1]. The closer the value is to 1, the higher the overlap between the two contours.

[0072] For example, take building instance M1 in phase T1 (outline area 216.3). ), T2 phase building example M2 (outline area 218.7) The area of ​​the intersection of their outlines is 209.5. The combined area is 225.5. The calculation yields:

[0073] .

[0074] Calculate separately centroid coordinates of the profile ( ), Then calculate the Euclidean distance between the centroids:

[0075] .

[0076] Combine the average area of ​​the two contours After normalization, the final feature values ​​are: The value ranges from (0,1), and the closer the value is to 1, the closer the centroid is.

[0077] For example, the above Center of mass and The Euclidean distance from the centroid is 0.012, and the average area is... =217.5 Calculated .

[0078] Using normalized Hausdorff distance Quantification and Boundary differences in the contours The value range is (0,1], and the closer the value is to 0, the more consistent the boundary shape; the boundary similarity feature value is:

[0079] The closer the value is to 1, the higher the similarity of the contour shapes.

[0080] For example, the above and of The boundary similarity eigenvalues ​​are calculated as follows:

[0081] .

[0082] Step S1.3: Calculation of local adaptive coefficients.

[0083] Calculate the specific values ​​of the local adaptive coefficients based on the geometric properties of the instance pair:

[0084] Area adaptive coefficient:

[0085] .

[0086] Distance adaptive coefficient:

[0087] .

[0088] Perimeter Adaptive Coefficient: The perimeters of the two instances are respectively and ,but:

[0089] .

[0090] Step S1.4: Adaptive weight synthesis and matching score calculation.

[0091] Substituting the global base weights and local adaptive coefficients into the formula, we synthesize the adaptive weights. :

[0092] ;

[0093] ;

[0094] .

[0095] The final matching score is obtained by multiplying the adaptive weights by the feature values ​​and summing the results.

[0096] .

[0097] Step S1.5: Region classification.

[0098] The matching threshold is set to 0.6. Based on the score results, each target will be classified into the following three categories: (1) successfully matched area (there is a corresponding relationship between the two periods); (2) newly added candidate area (newly appeared in period T2); (3) disappeared candidate area (existed in period T1 but not matched in period T2).

[0099] Based on the matching judgment logic, a matching threshold is set for verification in engineering practice. Batch matching and classification were completed for 4527 building instances:

[0100] Successfully matched regions: This is an architectural example where there is a clear spatiotemporal correspondence between the two phases;

[0101] Add new candidate regions: Furthermore, it features a unique silhouette specific to the T2 phase.

[0102] Vanishing candidate regions: Furthermore, it is a contour that exists in period T1 but has no matching target in period T2.

[0103] After classification, all candidate regions are used as input for the next step S2 to achieve preliminary screening of the changing regions.

[0104] S2, perform the three-dimensional sampling and credibility analysis steps.

[0105] like Figure 3 As shown, for successfully matched regions, newly added candidate regions, and disappeared candidate regions, the average elevation difference between different time phases is calculated. Three-dimensional point sampling is performed within the geometric boundaries of each candidate region to generate a uniformly distributed set of sampling points. Multidimensional features representing the continuity of the geometric structure are extracted based on the sampling point set. False changes caused by unstructured objects are identified and filtered out based on these multidimensional features. Regions exhibiting structural features are retained.

[0106] This step aims to verify the authenticity of the changes from a three-dimensional geometric perspective.

[0107] Step S2.1: Calculate the average elevation difference ΔH.

[0108] First, calculate the average elevation difference ΔH using the following formula:

[0109] .

[0110] in, The total number of sampling points within the candidate region. , They are respectively the T1 and T2 periods DSM elevation values ​​of each sampling point A positive value indicates an increase in regional elevation, while a negative value indicates a decrease in elevation.

[0111] For example, taking a candidate region with a successful match, 200 sampling points are selected within it. The sum of the elevation differences of each sampling point is 12.6m. The calculation yields... This indicates that the elevation of the region did not change significantly between the two periods.

[0112] Step S2.2: Poisson disk sampling.

[0113] The sampling method uses the Poisson disk sampling method to generate a set of points with uniform spacing inside the candidate polygon. The sampling point spacing is set to 0.3 meters, which is about 5 times the ground resolution of the image, to ensure that there are no less than 50 sampling points in each area.

[0114] Step S2.3: Extraction of three-dimensional structural continuity features.

[0115] Based on the DSM elevation values ​​of the sample point set, according to Figure 3 (c) The feature extraction rules calculate three core discriminative features: elevation standard deviation, local gradient mean, and DSM hole rate. The definitions, judgment thresholds, and implementation examples of each feature are as follows. The three features are combined to accurately distinguish between structural changes and unstructured pseudo-changes:

[0116] (1) Standard deviation of elevation

[0117] Calculate the standard deviation of the elevation values ​​of all sampling points within the candidate area. The smaller the value, the smoother the surface; the larger the value, the more undulating the surface. Judgment threshold: <0.8m (meets the flatness characteristics of building roofs).

[0118] For example, the standard deviation of the elevation of the sampling points in the building roof area is 0.35m (meets the threshold), while that in the vegetation canopy area is 1.2m (exceeds the threshold and is judged as a pseudo-change).

[0119] (2) Local gradient mean

[0120] Construct a Delaunay triangulation for the sampling point set, calculate the slope value of all triangular faces and take the average value. The smaller the value, the flatter the surface, and the larger the value, the steeper the surface slope. Judgment threshold: <0.1 (consistent with the flat roof characteristics of a building).

[0121] For example, the average slope of flat building roofs is 0.04 (meets the threshold), while that of steep building facades is 0.32 (exceeds the threshold and is judged as a pseudo-change).

[0122] (3) DSM void ratio

[0123] The percentage of invalid DSM elevation data (NoData) in the sampled points within the candidate area is calculated. The lower the value, the higher the data integrity, and the higher the value, the more severe the area occlusion. Judgment threshold: <20% (meets the characteristics of an unobstructed building area).

[0124] For example, the void ratio in unobstructed building areas is 3% (meets the threshold), while the void ratio in tree-covered areas is 28% (exceeds the threshold and is judged as a pseudo-change).

[0125] Step S2.4: False change identification and region screening.

[0126] If a candidate region meets all three thresholds mentioned above, it is determined to be a region of structural change and proceeds to the next step, S3 hierarchical decision-making. If any threshold is not met, it is determined to be a non-structural pseudo-change (caused by vegetation shading, data noise, interference from temporary objects, etc.) and is directly filtered out.

[0127] S3, execute the multi-threshold hierarchical decision-making step.

[0128] For the retained areas of change, the changes in area and height are considered. A two-level logical judgment is performed using preset monitoring eligibility thresholds and change significance thresholds. The areas of change are categorized into three types: confirmed changes, changes requiring continuous monitoring, and invalid changes.

[0129] Step S 3.1: Calculation of characteristic quantities.

[0130] For each area of ​​structural change, the area change is calculated based on the building outline vector data from periods T1 and T2. Average elevation difference calculated based on S2 Determine the height variation and extract the basic feature quantities for grading.

[0131] Step S 3.2: Setting the tiered threshold.

[0132] Based on the "Real Estate Surveying Standard" and the accuracy of DSM data elevation measurement, monitoring eligibility thresholds and change significance thresholds are preset. The threshold settings are tailored to the actual application needs of the project, and the specific values ​​are as follows:

[0133] Monitoring eligibility threshold: Area monitoring threshold High monitoring threshold (Twice the standard error of DSM elevation measurement);

[0134] Significance threshold for change: Area change threshold height change threshold (Minimum standard for building floor height in the "Real Estate Measurement Standard").

[0135] Step S 3.3: Two-level logic determination.

[0136] Based on the aforementioned preset thresholds, strictly in accordance with Figure 5 The logical flow of the multi-threshold hierarchical monitoring system involves executing two levels of judgment sequentially, while also referencing... Figure 4 The visualization results of elevation characteristics in structurally changing areas are used to assist in verification, ensuring that the judgment logic aligns with engineering practice and that the judgment results are accurate and reliable. The specific judgment conditions are as follows:

[0137] (1) Qualification determination for monitoring

[0138] The conditions for determining whether an area meets the basic qualifications for subsequent monitoring and assessment are as follows: If the change in area or the change in height reaches the monitoring threshold, it is determined that the entity is qualified for monitoring.

[0139] (2) Determination of the significance of change

[0140] To determine whether a region is eligible for monitoring, the change in its data reaches the significance level recognized by the engineering project, the following conditions must be met: A change is considered significant if the change in area or the change in height reaches a significant threshold.

[0141] Step S 3.4: Hierarchical decision making.

[0142] Based on the logical combination of the two-level judgment conditions, combined with Figure 5 The decision-making logic classification rules, and in comparison Figure 4 The differences in elevation characteristics among different types of structural change areas were verified, and structural change areas were clearly divided into three categories: confirmed changes, changes requiring continuous monitoring, and invalid changes. The judgment rules and engineering implementation examples for each type are as follows, ensuring that the classification results are highly consistent with the characteristics shown in the attached figures:

[0143] (1) Confirm the changes: (Qualified for monitoring and showing significant changes)

[0144] Judgment rule: Changes that are considered to be real and of high confidence level are directly output as detection results;

[0145] For example, in a certain area of ​​an added-floor building, , ,satisfy ( (True) ( If true, it is determined as a confirmed change.

[0146] (2) Changes that require continuous monitoring: (Qualified for monitoring but the changes are not significant)

[0147] Judgment rules: If there are minor changes in the area but they do not reach the level of engineering significance, the area should be included in the continuous monitoring list for tracking; add the core information such as the outline coordinates, area, average elevation, and characteristic parameters of the area to the continuous monitoring list as the matching input for the next time phase (T3) S1 step;

[0148] For example, a minor adjustment area in a certain building. , ,satisfy ( (True) and ( If the result is false, it is determined to be a change that requires continuous monitoring.

[0149] (3) Invalid changes: (Not qualified for monitoring)

[0150] Judgment rule: Areas with minimal changes and no practical engineering significance are directly filtered out;

[0151] For example, a tiny area, , Not satisfied The condition for determination is that the change is invalid.

[0152] Step S 3.5: Statistics and closed-loop update of classification results.

[0153] After this round of detection is completed, the updated continuous monitoring list is automatically saved and used as one of the matching inputs for step S1 when processing the next time phase (T3) data. Combined with... Figure 5 The closed-loop monitoring logic enables intelligent tracking and progressive judgment of building change targets across time phases, and... Figure 4 The credibility analysis results form a complete closed loop for engineering implementation, ensuring the continuity and integrity of the entire testing process.

[0154] Update the continuous monitoring list. Store the target information corresponding to changes that are determined to require continuous monitoring into the continuous monitoring list. Use the continuous monitoring list as the input data source for the adaptive instance matching step in the next time phase processing to achieve closed-loop tracking across time phases.

[0155] The continuous monitoring list is crucial for capturing gradual changes. For changes deemed necessitating continuous monitoring, the system writes their core information, such as outline coordinates, area, and average elevation, into the list. In the next time phase (T3), during adaptive instance matching, the system matches building instances from the current time phase (T3) with instances from phase T2 and historical targets in this list. This mechanism allows for the cumulative determination of minor changes. For example, if a building's height increases by 0.5 meters in the current monitoring, although it doesn't meet the 2.2-meter addition standard, it's still included in the monitoring list because it's greater than 0.4 meters. If another 0.5-meter change is detected in the next phase, bringing the cumulative change to 1.0 meter, the system will identify the trend in subsequent continuous tracking until it reaches a significant level and is confirmed as an addition, effectively avoiding missed detections of gradual changes.

[0156] This application also provides a hierarchical building change recognition system that combines feature matching and 3D information collaboration. The system includes a data acquisition module, a matching module, an analysis module, a decision-making module, and a storage and update module.

[0157] The data acquisition module is used to acquire multi-temporal building outline vector data and corresponding digital surface model data.

[0158] The matching module is used to construct an adaptive matching score function that integrates globally preset weights and local geometric features. Based on this function, building instances in the current time phase are associated with building instances in the previous time phase and targets in the continuously monitored list. The output includes successfully matched areas, newly added candidate areas, and disappeared candidate areas.

[0159] The analysis module calculates the average elevation difference for each candidate region and performs 3D point sampling. It extracts multidimensional features characterizing the continuity of the geometric structure. Based on these multidimensional features, it filters out non-structural pseudo-changes.

[0160] The decision-making module combines changes in area and height. It performs a two-level logical judgment using monitoring eligibility thresholds and change significance thresholds. It outputs a categorized result for confirmed changes, changes requiring continued monitoring, and invalid changes.

[0161] The storage and update module stores the continuous monitoring list and updates the list with changes deemed to require continuous monitoring, for use in the next time phase processing.

[0162] During implementation, the system may also include preprocessing and post-verification stages. Before executing the adaptive instance matching step, the system performs topology checks and repairs on the input vector data, and denoises and fills holes in the DSM data. After executing the 3D sampling and credibility analysis steps, the system performs spatial topology consistency verification. It checks whether the verified change areas spatially conflict with features such as roads, water bodies, or green spaces. If a conflict occurs, the changed area is marked or removed. For example, small-area change patches appearing within cultivated land, even if they pass 3D verification, will be judged as unreasonable changes and filtered out. In addition, the continuous monitoring list has a dynamic maintenance mechanism. Confirmed or invalid targets are removed after each round of processing. Targets in the list are given higher matching priority or fault tolerance during matching to adapt to the long-term slow deformation of building forms and prevent mismatch of historical monitoring targets.

[0163] It should be noted that the terms "first," "second," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. "A plurality" or "several" indicates at least two. Unless otherwise stated, terms such as "front," "back," "left," "right," "lower," and / or "upper" are for illustrative purposes only and are not limited to a location or spatial orientation. Terms such as "comprising" or "including" indicate that the elements or objects preceding "comprising" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0164] The singular forms “a,” “the,” and “the” used in this application specification and appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0165] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for hierarchical recognition of building changes based on feature matching and 3D information synergy, characterized in that, Includes the following steps: Acquire multi-temporal building outline vector data and corresponding digital surface model data, wherein the building outline vector data includes geometric boundary information of building instances; The adaptive instance matching step is to construct an adaptive matching score function that integrates global preset weights and local geometric features. The matching score function comprehensively considers the contour overlap, contour center distance and shape similarity between any two building instances. Based on the adaptive matching score function, the building instances of the current time phase are associated with the building instances of the previous time phase and the targets in the continuous monitoring list, thereby dividing the successfully matched area, the newly added candidate area and the disappeared candidate area. Perform three-dimensional sampling and credibility analysis steps: For the successfully matched area, newly added candidate area and disappeared candidate area, calculate the average elevation difference between different time phases, and perform three-dimensional point sampling within the geometric boundary of each candidate area to generate a uniformly distributed set of sampling points. Based on the set of sampling points, extract multi-dimensional features that characterize the continuity of the geometric structure. Based on the multi-dimensional features, identify and filter out pseudo changes caused by unstructured objects, and retain the change areas with structural features. Execute multi-threshold hierarchical decision-making steps: For the retained change area, combine its area change and height change, and use the preset monitoring eligibility threshold and change significance threshold to perform two-level logical judgment to divide the change area into three categories: confirmed change, change that needs to be continuously monitored, and invalid change; Update the continuous monitoring list: Store the target information corresponding to the changes that are determined to require continuous monitoring into the continuous monitoring list, and use the continuous monitoring list as the input data source for the adaptive instance matching step in the next time phase processing, so as to realize closed-loop tracking across time phases.

2. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy according to claim 1, characterized in that, In the adaptive instance matching step, the adaptive matching scoring function employs a two-layer weighting mechanism: Define a global basic weight, which is determined by machine learning training based on the priority of three factors: contour overlap, contour center distance, and shape similarity. This global basic weight is used to reflect the general importance of different features in the global scope. Define local adaptive parameters, which are dynamically calculated based on the specific geometric attributes of the two building instances to be matched. The geometric attributes include the degree of area similarity between the two instances, the proportional attenuation factor based on the centroid distance, and the degree of similarity in the outline perimeter, which are used to reflect the local geometric characteristics of the current instance pair. Based on the product of the global basic weights and the local adaptive parameters, the dynamic weight coefficients corresponding to each feature item in the adaptive matching score function are calculated in a normalized manner, so that the matching score function can adaptively adjust the contribution ratio of each feature item according to building instances of different scales and shapes, thereby improving the accuracy of temporal correlation.

3. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy according to claim 2, characterized in that, The area similarity in the local adaptive parameters is obtained by calculating the ratio of the smaller to the larger area of ​​the two instances. The proportional decay factor in the local adaptive parameters is obtained by normalizing the centroid distance by introducing the square root of the average area of ​​the two instances as a normalization factor. The contour perimeter similarity in the local adaptive parameters is obtained by calculating the ratio of the smaller to the larger perimeter of the two instances. The adaptive matching score function fuses the normalized contour overlap index, the centroid distance index processed by the Gaussian kernel function, and the shape similarity index transformed by the normalized Hausdorf distance through a weighted summation method, and finally outputs a matching score value between zero and one.

4. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy according to claim 1, characterized in that, In the three-dimensional sampling and credibility analysis step, the three-dimensional point sampling adopts the Poisson disk sampling method. By generating a set of sampling points with uniform spacing inside the geometric polygon of the candidate region, it ensures that the sampling points uniformly cover the candidate region and avoids verification deviation caused by uneven sampling density. The multidimensional features characterizing the continuity of the geometric structure include at least: the elevation standard deviation feature for describing the dispersion of the elevation values ​​of the sampling points, the local gradient feature for reflecting the changes in surface slope, and the void ratio feature for statistically analyzing the proportion of invalid elevation data; by jointly analyzing the above multidimensional features, the architectural changes with a smooth surface and continuous geometric structure can be distinguished from non-structural pseudo-changes with drastic surface undulations, chaotic slopes, or severe shading.

5. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy according to claim 4, characterized in that, The local gradient feature is calculated based on the triangular network constructed from the sampling point set. It is obtained by averaging the slope values ​​of all triangular faces in the triangular network and is used to quantify the flatness of the candidate region surface. The void ratio feature is calculated by statistically analyzing the proportion of empty or outlier elevation values ​​among the sampling points in the candidate region and is used to identify areas caused by vegetation shading or missing data. During the credibility analysis, if a candidate region simultaneously meets the conditions that its elevation dispersion is lower than the preset upper limit, its surface slope is lower than the preset upper limit, and its void ratio is lower than the preset upper limit, then the region is determined to be a building change region with a real geometric structure; otherwise, it is determined to be a non-structural pseudo-change and is removed.

6. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy according to claim 1, characterized in that, The multi-threshold hierarchical decision-making steps include: Define the monitoring eligibility determination logic to determine whether the area change of the changed area reaches the preset area monitoring threshold or the height change reaches the preset height monitoring threshold. If either condition is met, the area is determined to be eligible for monitoring. Define the logic for determining the significance of changes, and determine whether the area change of a qualified monitoring area reaches a preset threshold for significant area change, or whether the height change reaches a preset threshold for significant height change. If either condition is met, the change in the area is determined to be significant. Based on the Boolean combination of the monitoring eligibility determination logic and the change significance determination logic, the final hierarchical decision result is output.

7. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy as described in claim 6, characterized in that, The rules for determining the results of the hierarchical decision-making process are as follows: If the changed area is qualified for monitoring and the change is significant, it is determined to be a confirmed change and is directly output as a high-confidence detection result; If a changed area is eligible for monitoring but its change does not meet the significance standard, it is determined to be a change that needs to be continuously monitored. The geometric boundary, area and elevation characteristics of the area are stored in the continuous monitoring list so that it can re-participate in adaptive instance matching and verification in subsequent time-phase processing to achieve time-series cumulative judgment of minor changes. If the area of ​​change is not eligible for monitoring, it is deemed an invalid change and will be filtered out without further processing.

8. The method for hierarchical recognition of building changes based on feature matching and 3D information synergy according to claim 1, characterized in that, Before performing the adaptive instance matching step, a data preprocessing step is also included, which is used to perform topology checks and repairs on the input multi-temporal building outline vector data, and to denoise and fill holes in the digital surface model data. After performing the three-dimensional sampling and credibility analysis step, a spatial topology consistency verification step is also included, which is used to check whether the verified change area has spatial conflicts with features such as roads, water bodies or green spaces. If a conflict occurs, the change area is marked or removed to further improve the geographic logic rationality of the output results.

9. The method for hierarchical identification of building changes based on feature matching and 3D information synergy according to claim 1, wherein the dynamic maintenance mechanism of the continuous monitoring list includes: After each round of time phase processing is completed, the continuous monitoring list is updated, removing targets that have been determined to be confirmed changes or invalid changes, and retaining targets that are still under monitoring. When matching targets in the continuous monitoring list with the current time-phase instance, targets in the continuous monitoring list are given a higher matching priority or matching tolerance to accommodate the slow deformation or gradual change of building form that may occur during long-term monitoring, and to prevent mismatch of historical monitoring targets due to minor geometric differences.

10. A hierarchical building change recognition system that combines feature matching and three-dimensional information, characterized in that, include: The data acquisition module is used to acquire multi-temporal building outline vector data and corresponding digital surface model data; The matching module is used to construct an adaptive matching score function that integrates global preset weights and local geometric features. Based on this function, the building instances in the current time phase are associated with the building instances in the previous time phase and the targets in the continuous monitoring list, and the matching successful areas, newly added candidate areas and disappeared candidate areas are output. The analysis module is used to calculate the average elevation difference of each candidate region and perform three-dimensional point sampling, extract multi-dimensional features that characterize the continuity of the geometric structure, and filter out non-structural pseudo-changes based on the multi-dimensional features. The decision module combines area change and height change, uses monitoring eligibility threshold and change significance threshold to perform two-level logical judgment, and outputs graded results for confirmed changes, changes requiring continuous monitoring, and invalid changes. The storage and update module is used to store the continuous monitoring list and update the continuous monitoring list with changes that are determined to require continuous monitoring, so that they can be called in the next phase of processing.

Images