Automatic detection method for urban and rural house form change based on multi-source time-series remote sensing images

Abstract

The application relates to the technical field of image data processing, in particular to a kind of urban and rural house form change automatic detection method based on multi-source time sequence remote sensing image, comprising: determining scattering center point according to the pixel decibel value of SAR image;All scattering center points are projected in the same coordinate system to obtain scattering space point set, and the scattering space point set is clustered to obtain each cluster;Determine the target cluster labeled as adhesion house, determine each group in the target cluster according to the distance distribution between the scattering center points in the target cluster;According to grouping, the target cluster is split into each independent house area, the house boundary is corrected to obtain the real plane contour according to the azimuth relationship between the house boundary of independent house area and the illumination direction of SAR sensor;Based on the contour change of the real plane contour of independent house in adjacent monitoring period, determine the house form change detection result.Through the technical scheme of the application, the accuracy of house individual detection and house form change detection is improved.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, specifically to an automatic detection method for changes in the morphology of urban and rural buildings based on multi-source time-series remote sensing images. Background Technology

[0002] Monitoring changes in urban and rural housing morphology relies on high-frequency remote sensing observations. Optical images are susceptible to cloud and rain interference and have poor temporal continuity. In practice, SAR (Synthetic Aperture Radar) imagery data is often needed to complete the observation chain. Urban and rural housing roofs are mostly made of concrete or corrugated steel sheets. When adding floors or expanding, the newly poured roof surface will exhibit strong backscattering in SAR images, creating a significant radiation difference compared to the original roof. However, the characteristics of SAR oblique side-view imaging lead to overlapping of buildings and compression of facades. When buildings are densely distributed, echoes from adjacent buildings are superimposed on the same pixel, and there is a shift between the location of enhanced scattering and the actual boundary of the building. A single, single-angle observation cannot deconstruct this superposition relationship.

[0003] By introducing multi-source SAR time-series observations with different orbits and incident angles, the scattering characteristic sequence of the same building can be obtained from multiple perspectives. The scattering performance of the building from each perspective is determined by its orientation, height, and adjacency relationship, and the pattern of scattering characteristics changing with perspective itself contains individual identification information.

[0004] How to extract individual identification information of independent houses from multi-source time-series SAR observations and accurately locate radiation differences to specific individual houses is a core problem that urgently needs to be solved in the detection of changes in urban and rural housing morphology. Summary of the Invention

[0005] To address the technical problem of extracting individual building identification information from multi-source temporal SAR observations and accurately pinpointing radiation differences to specific buildings, this invention aims to provide an automatic detection method for urban and rural building morphology changes based on multi-source temporal remote sensing imagery. The specific technical solution adopted is as follows: This invention provides an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing imagery, the method comprising: Acquire multi-source time-series SAR intensity image data of the monitoring area, and determine the scattering center point based on the pixel decibel value of each SAR image; By projecting the scattering center points of all SAR images onto the same coordinate system, a set of scattering spatial points is obtained. The set of scattering spatial points is then clustered to obtain clusters that are marked as houses. Identify the target clusters marked as contiguous houses, and determine the grouping of each house marked as an independent house in the target cluster based on the distance distribution between the scattering centers in the target cluster; The target cluster is divided into individual house areas according to the grouping. The house boundaries are corrected according to the azimuth relationship between the house boundaries of the individual house areas and the SAR sensor illumination direction to obtain the true planar outline of the individual houses. Based on the changes in the actual planar outline of detached houses during adjacent monitoring periods, the results of house shape change detection are determined.

[0006] Further, determining the scattering center point based on the pixel decibel value of each SAR image includes: Divide any SAR image into individual pixel sub-blocks and obtain the mean and standard deviation of the decibel values ​​of the pixels in each pixel sub-block. The strong and weak scattering thresholds of the sub-blocks are determined based on the mean and standard deviation of the decibel values. Pixels in the sub-blocks whose decibel values ​​are greater than the strong and weak scattering thresholds are marked as strong scattering candidate points. Non-maximum suppression is applied to strong scattering candidate points to obtain the scattering center points among the strong scattering candidate points.

[0007] Further, the step of performing non-maximum suppression on strong scattering candidate points to obtain the scattering center points among the strong scattering candidate points includes: Determine strong scattering candidate points and their pixel neighborhoods. When the pixel decibel value of a strong scattering candidate point is less than the pixel decibel value of any pixel in its neighborhood, the strong scattering candidate point is suppressed. Each suppressed strong scattering candidate point is marked as a non-scattering center point, thus obtaining the remaining scattering center points among the strong scattering candidate points.

[0008] Furthermore, the process of clustering the scattering spatial points to obtain clusters labeled as houses includes: Based on the ground resolution of SAR images, DBSCAN clustering is performed on the scattering spatial point set to obtain each cluster labeled as a house. The convex hull of the cluster is used as the initial extent of the house, and the geometric center of the convex hull is used as the initial position of the house.

[0009] Further, identifying the target clusters marked as adherent houses includes: Obtain the center point of the ascending orbit scattering from the ascending orbit perspective in the cluster, and determine the first spatial distribution center of the ascending orbit scattering center point; Obtain the center point of the descending orbit scattering from the descending orbit perspective in the cluster, and determine the second spatial distribution center of the descending orbit scattering center point; Determine the planar distance between the first spatial distribution center and the second spatial distribution center; If the planar distance is less than or equal to a preset distance threshold, the cluster is marked as an independent house; If the planar distance is greater than a preset distance threshold, the cluster is marked as an adhered house and the cluster is used as the target cluster.

[0010] Further, determining the grouping of each labeled independent house within the target cluster based on the distance distribution between scattering centers in the target cluster includes: Obtain the point-to-point distance between any two scattering center points in the target cluster and obtain the set of point-to-point distances; A histogram is constructed based on the set of point-to-point distances, and a curve is fitted to the histogram to obtain the distance distribution curve. The initial grouping in the target cluster is determined based on the number of peaks in the distance distribution curve; Identify the boundary ambiguity points in the initial grouping and regroup the boundary ambiguity points to obtain the groups that are marked as independent houses.

[0011] Furthermore, determining the initial grouping in the target cluster based on the number of peaks in the distance distribution curve includes: If there are two peaks in the distance distribution curve, then obtain the valley distance between the two peaks at the valley point. If the point-to-point distance between any two scattering centers in the target cluster is less than the trough distance, then the two scattering centers are determined to belong to the same initial group. If the number of peaks in the distance distribution curve is one or more, hierarchical agglomerative clustering is used to divide the target cluster into initial groups.

[0012] Further, the step of determining the boundary ambiguity points in the initial grouping and regrouping the boundary ambiguity points to obtain each group labeled as an independent house includes: Obtain the first distance between the scattering center point in the initial group and the geometric center point of the initial group; Obtain the second distance between the scattering center point in the initial group and the geometric center point of its adjacent group; Compare the first and second distances to mark some of the scattering center points in the initial grouping as boundary ambiguity points; The first coordinate centroid is obtained from the set of scattering center points corresponding to the target angle in the initial group; The second coordinate centroid is obtained from the set of scattering center points corresponding to the target angle in adjacent groups; Based on the distances of the fuzzy boundary points relative to the first and second coordinate centroids, the fuzzy boundary points are regrouped to obtain groups that are marked as independent houses.

[0013] Further, the step of correcting the house boundaries based on the azimuth relationship between the house boundaries of the detached house area and the SAR sensor illumination direction to obtain the true planar outline of the detached houses includes: Obtain the direction vector corresponding to the SAR sensor illumination direction in the SAR image under the target angle marker; Obtain the minimum bounding rectangle of the convex hull corresponding to the independent house area, use the minimum bounding rectangle as the initial correction benchmark, and determine the house boundary; Starting from the midpoint of the building boundary, a ray along the direction vector is obtained. The boundary type of the building boundary is determined based on the directional relationship of the ray relative to the inside and outside of the building boundary. The boundary type of the house boundary is used to determine the house boundary to be corrected, and the corrected house boundary is used to obtain the true planar outline of the detached house.

[0014] Furthermore, the boundary types of the house boundary include facing edges and back edges; the process of determining the house boundary to be corrected based on the boundary type, and correcting the house boundary to be corrected to obtain the true planar outline of the detached house, includes: If the boundary of a house is determined to be a back edge under all angle markings, then the boundary of the house is determined to be a boundary to be corrected. Obtain the adjacent boundaries of the house boundary to be corrected. If the adjacent boundary is an facing edge, use the adjacent boundary as a reference to correct the house boundary to obtain the true plan outline of the detached house. If the adjacent boundary is a back edge, a new adjacent boundary is obtained. If the new adjacent boundary is a front edge, the house boundary to be corrected is corrected based on the new adjacent boundary to obtain the true plan outline of the detached house.

[0015] The present invention has the following beneficial effects: This invention utilizes the spatial distribution differences and viewpoint-dependent laws of multi-angle SAR scattering centers to effectively extract individual building identification information from overlapping echoes, accurately pinpointing temporal radiation differences to individual buildings. This solves the core problems of traditional single-angle SAR being unable to deconstruct overlapping and overlapping data, multi-source temporal SAR observations being unable to extract individual building identification information from independent buildings, and radiation changes being difficult to accurately attribute to specific buildings. The adhering house separation technology provided by this invention is based on the spatial distribution of multi-angle scattering centers and homogeneous geometric constraints. It can automatically separate adhering houses that have merged due to overlapping without relying on external house base vector reference data, thus realizing the accurate identification of independent houses in dense building areas and significantly improving the accuracy of individual detection of independent houses. This invention also corrects the building boundary by using the orientational relationship between the building boundary and the SAR sensor illumination direction, effectively eliminating geometric stretching distortion caused by SAR overlay, obtaining the true planar outline of the building, ensuring the reliability of subsequent outline change identification, and finally obtaining accurate building shape change detection results. It is suitable for large-scale, high-frequency automated supervision applications of urban and rural housing management departments. Attached Figure Description

[0016] Figure 1 The flowchart shows the steps of an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images, as provided in one embodiment of the present invention. Figure 2 This is a detailed flowchart of step S1 in an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images provided in an embodiment of the present invention. Figure 3 This is a detailed flowchart of step S2 in an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images provided in an embodiment of the present invention. Figure 4 This is a detailed flowchart of step S3 in an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images provided in an embodiment of the present invention. Figure 5 This is a detailed flowchart of step S4 in an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images, provided in an embodiment of the present invention. Detailed Implementation

[0017] The following description, in conjunction with the accompanying drawings, details the specific scheme of an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images provided by the present invention.

[0018] Example 1:

[0019] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of an automatic detection method for urban and rural building morphology changes based on multi-source time-series remote sensing images provided by an embodiment of the present invention.

[0020] The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing images includes the following steps: Step S1: Obtain multi-source time-series SAR intensity image data of the monitoring area, and determine the scattering center point based on the pixel decibel value of each SAR image. In this embodiment, multi-source time-series SAR intensity image data covering the monitoring area are collected. Here, "multi-source" refers to multiple angles, and the angle is denoted as k, called angle identifier k. The angle includes a combination of orbit direction (including ascending and descending orbits), observation azimuth angle (including ascending and descending orbit angles), and incident angle. The image type is an intensity product generated by radiometric calibration of ground distance or slant distance single-view complex data. The time coverage of the collection should include the period before and after the change in building form. The interval between adjacent collection periods is set according to the monitoring frequency (such as monthly or quarterly), but the overall span of the time coverage should not be less than 3 months to accumulate sufficient time-series features to ensure that changes in building form caused by the addition, expansion, or demolition of urban and rural buildings can be captured. The image orbit orientation includes two modes: ascending and descending. The combination of ascending and descending orbits allows the same target to be imaged in two directions with an observation azimuth angle difference of approximately 180°, thereby obtaining complementary side-looking geometric information. The incident angle covers three intervals: 25°–35°, 35°–45°, and 45°–50°. At least two SAR images (i.e., SAR intensity images, hereinafter referred to as "images" in this invention) with different incident angles are acquired in each interval. The temporal sampling density is no less than one scene per month, and the total number of effective image scenes is ensured to be greater than 10. Images with different orbit orientations, observation azimuth angles, and incident angle combinations together constitute a multi-angle, multi-temporal SAR intensity image stack. Each image scene in the stack is accompanied by orbit parameter metadata, which includes at least the orbit orientation, observation azimuth angle, incident angle, slant range sampling interval, and azimuth sampling interval, as parameter inputs for subsequent processing. For the K scenes of SAR images with different orbits and incident angle combinations, radiometric calibration is performed scene by scene. The original DN (Digital Number) values ​​are converted into backscattering coefficients (unit: dB). Lee filtering (window size set to 5×5 pixels) is used to perform adaptive speckle noise suppression to reduce the interference of the inherent granular noise of coherent imaging on the subsequent scattering center localization. The filtered multi-source time-series SAR intensity image data are uniformly stored in a spatiotemporal database, and an index is established according to the angle identifier k, orbital direction, observation azimuth, incident angle and timestamp, which will be used for building identification and change detection in subsequent embodiments.

[0021] Specifically, please refer to Figure 2 Step S1, determining the scattering center point based on the pixel decibel value of each SAR image, includes: Step S11: Divide any SAR image into individual pixel sub-blocks and obtain the mean and standard deviation of the decibel values ​​of the pixels in the pixel sub-blocks. Step S12: Determine the strong scattering threshold of the sub-block based on the mean and standard deviation of the decibel values, and mark the sub-blocks whose decibel values ​​are greater than the strong scattering threshold as strong scattering candidate points. Step S13: Perform non-maximum suppression on the strong scattering candidate points to obtain the scattering center point among the strong scattering candidate points.

[0022] More specifically, step S13 includes: Determine strong scattering candidate points and their pixel neighborhoods. When the pixel decibel value of a strong scattering candidate point is less than the pixel decibel value of any pixel in its neighborhood, the strong scattering candidate point is suppressed. Each suppressed strong scattering candidate point is marked as a non-scattering center point, thus obtaining the remaining scattering center points among the strong scattering candidate points.

[0023] In this embodiment, in order to automatically detect the form of urban and rural houses using multi-angle SAR images, it is necessary to identify and acquire the house areas in the SAR images. Because different incident angles of satellite radar during the acquisition of SAR images can cause backscattering of buildings, resulting in signal overlap between different buildings; Therefore, it is necessary to first identify building targets in SAR images to obtain their initial positions and rough outlines; In SAR images, the corners, ridges, and junctions of walls and ground of buildings exhibit stable strong scattering structures; while non-building targets such as vehicles and trees only appear at certain angles; therefore, by locating strong scattering from multiple angles in SAR images, buildings and non-building areas in the image can be distinguished. For each SAR image, the image is divided into non-overlapping sub-blocks (the size of the sub-blocks is set according to the ground resolution, so that the side length of the sub-block corresponds to about 15m-30m on the ground, so as to ensure that the scattering points of a typical detached house (10m×10m) are counted in the same sub-block); and for each sub-block, the mean decibel value B1 and the standard deviation of the decibel value B2 of the pixels in it are obtained, and the decibel value of each pixel is recorded as the pixel decibel value. A CFAR (Constant False Alarm Rate) threshold strategy is adopted to adaptively obtain the strong and weak scattering threshold coefficient β; specifically, the false alarm rate P = 1 × If the pixel values ​​of a SAR image approximately follow a Gaussian distribution after multi-look processing, then the strong scattering threshold coefficient β = (1-P), where (·) is the inverse function of the standard normal distribution. Substituting it into the calculation, we get β≈3.72; The strong and weak scattering threshold T (used to distinguish between strong and weak scattering points) is set as T=B1+β×B2. For each pixel in a sub-block, if the pixel decibel value is greater than T, it is marked as a strong scattering candidate point; otherwise, if it is less than or equal to T, it is directly marked as a weak scattering point. To ensure that strong scattering structures are identified only at a single point, non-maximum suppression is applied to strong scattering candidate points. For each strong scattering candidate point, its pixel neighborhood (8-neighborhood) is obtained. If its pixel decibel value is less than the pixel decibel value of any pixel in its 8-neighborhood, the strong scattering candidate point is suppressed and not retained. Only the local maximum point in the 8-neighborhood is retained. That is, if the strong scattering candidate point is not a local maximum point in its pixel neighborhood, it can be marked as a non-scattering center point. For the other strong scattering candidate points that are local maximum points in their own pixel neighborhood, they are retained and marked as scattering centers. Finally, the remaining strong scattering candidate points are all scattering centers. This yields the set of scattering centers. Each scattering center point in the set is bound to its source data label, including angle identifier k, orbital direction, observation azimuth, incident angle, and geographic coordinates. The combination of orbital direction, observation azimuth, and incident angle can be determined by the angle identifier k.

[0024] Step S2: Project the scattering center points of all SAR images onto the same coordinate system to obtain a scattering spatial point set, and cluster the scattering spatial point set to obtain each cluster marked as a house. All scattering center points in SAR images from different angles are projected onto the same coordinate system (specifically using image-attached orbital parameters and external reference DEMs (Digital Elevation Models), such as SRTM (Shuttle Radar Topography Mission), and the range-Doppler model is used to geocode each SAR intensity image into WGS-84 (World Geodetic System 1984), and resampled to a uniform spatial resolution (e.g., 2m×2m), which is the existing technology); thus obtaining the multi-angle scattering spatial point set D.

[0025] Specifically, please refer to Figure 3 Step S2, which involves clustering the scattering spatial points into clusters labeled as houses, includes: Step S21: Based on the ground resolution of the SAR image, perform DBSCAN clustering on the scattering spatial point set to obtain each cluster labeled as a house. Step S22: Use the convex hull of the cluster as the initial range of the house, and the geometric center of the convex hull as the initial position of the house.

[0026] In this embodiment, the scattering spatial point set D is clustered using DBSCAN (Density-Based Spatial Clustering of Applications with Noise), resulting in several clusters. Specifically, the neighborhood radius eps and the minimum number of core points (MinPts) are set for this clustering algorithm. The neighborhood radius eps is set to 1.5R to 3R, where R is the ground resolution of the SAR image. The minimum number of core points (MinPts) is set to 3 to 5. The scattering spatial point set D is then clustered using the parameters set above. After processing, several clusters are output. After clustering, each cluster represents and labels a house. The convex hull of each cluster is extracted as the initial boundary of the house, and the geometric center coordinates of the convex hull are used as the initial position of the house. It should be noted that the houses labeled here cannot be simply regarded as independent houses; for ease of subsequent differentiation, they can be called candidate houses.

[0027] Step S3: Determine the target cluster marked as connected houses, and determine the grouping of each house marked as an independent house in the target cluster based on the distance distribution between the scattering center points in the target cluster; Specifically, step S3, determining the target clusters marked as adherent houses, includes: Obtain the center point of the ascending orbit scattering from the ascending orbit perspective in the cluster, and determine the first spatial distribution center of the ascending orbit scattering center point; Obtain the center point of the descending orbit scattering from the descending orbit perspective in the cluster, and determine the second spatial distribution center of the descending orbit scattering center point; Determine the planar distance between the first spatial distribution center and the second spatial distribution center; If the planar distance is less than or equal to a preset distance threshold, the cluster is marked as an independent house; If the planar distance is greater than a preset distance threshold, the cluster is marked as an adhered house and the cluster is used as the target cluster.

[0028] The above embodiment obtained the initial location of each candidate house. Due to the overlapping situation in SAR images, the strong scattering centers of two or more independent houses may be close to each other in space, which makes them easy to be merged into a house label by the clustering algorithm. Therefore, it is necessary to analyze the changes in the house from different angles to determine whether there is any situation where the houses are stuck together. When two houses are arranged adjacent to each other with a small distance between them, the offset of the scattering center of the roof of the back house is enough to make it fall near the scattering center of the front house. This aliasing phenomenon manifests differently under different observation azimuth angles. When the SAR sensor is illuminated from the opposite direction, the aliasing direction is flipped by about 180°. The back house that was originally aliased may become the front house, so its roof is no longer covered by the aliasing. The two sets of scattering centers that were originally aliased may naturally separate in space. Therefore, the adhesion of houses can be judged by observing the changes in the scattering center under multiple angles. For any cluster of candidate houses, obtain the scattering center points from the ascending orbit viewpoint, denoted as ascending orbit scattering center points, and calculate the first spatial distribution center Z1 of these ascending orbit scattering center points; and obtain the descending orbit scattering center points from the descending orbit viewpoint and calculate their second spatial distribution center Z2. Obtain the planar distance DZ between the current house distribution centers Z1 and Z2; When DZ ≤ preset distance threshold (set to 0.5R, where R is ground resolution), the candidate houses corresponding to the cluster are marked as independent houses, and the scattering center points within the cluster all belong to the same house; When DZ > preset distance threshold, the candidate houses corresponding to the cluster are marked as connected houses. The scattering center point in the cluster may come from two or more independent houses; the cluster marked as connected houses is denoted as target cluster C.

[0029] Therefore, the candidate houses obtained in the above embodiments can be divided into two categories: detached houses and attached houses; the attached houses need to be separated in subsequent embodiments.

[0030] Specifically, please refer to Figure 4 Step S3, determining the grouping of each house marked as an independent building in the target cluster based on the distance distribution between scattering centers in the target cluster, includes: Step S31: Obtain the point-to-point distance between any two scattering center points in the target cluster and obtain the set of point-to-point distances; Step S32: Construct a histogram based on the set of point-to-point distances, and perform curve fitting on the histogram to obtain the distance distribution curve; Step S33: Determine the initial grouping in the target cluster based on the number of peaks in the distance distribution curve; More specifically, step S33 includes: If there are two peaks in the distance distribution curve, then obtain the valley distance between the two peaks at the valley point. If the point-to-point distance between any two scattering centers in the target cluster is less than the trough distance, then the two scattering centers are determined to belong to the same initial group. If the number of peaks in the distance distribution curve is one or more, hierarchical agglomerative clustering is used to divide the target cluster into initial groups.

[0031] For the range of houses identified as adhering houses in the above embodiments, it is necessary to separate the adhering houses; in order to perform the separation operation, it is first necessary to determine the actual number of houses contained in the target cluster C; The offset of each mask will cause some scattering centers to be spatially closer, but there will be inherent geometric constraints between the scattering centers from the corners of the same house; therefore, the house can be divided based on geometric constraints.

[0032] Specifically, first determine the distance distribution between the scattering centers in the adhered houses: For any pair of scattering center points formed by two scattering center points in the target cluster C of the house, obtain the point-to-point distance between the scattering center point pairs; and obtain the set of point-to-point distances E for all point-to-point distances; The point-to-point distance set E can reflect the distribution of corner distances in the currently connected houses; A histogram is constructed for all distance values ​​in the point-to-distance set E, and the histogram is fitted with a curve using the least squares method to obtain the distance distribution curve. For situations where the dimensions of different detached houses in urban and rural areas are relatively similar, and the corner spacing within the same detached house is relatively similar but there are significant differences in corner spacing between different detached houses (i.e., obvious separation between inner and outer distances): Adhesive houses will form two peaks on the distance distribution curve; the first peak of the distance distribution curve is located at a smaller distance, representing the smaller distance between the corners of the same house; while the second peak is located at a larger distance, representing the larger distance between the corners of different houses; it should be noted that the smaller distance and the larger distance are relative and are clear characteristics.

[0033] For cases where the dimensions of different independent houses are relatively close or the interior and exterior distances are significantly separated, the adjoining houses will form one or more peaks on the distance distribution curve.

[0034] Furthermore, the number of independent houses among the clustered houses is determined based on the distance distribution curve: First, determine the number of peaks in the distance distribution curve mentioned above; If the number of peaks is 2, obtain the valley point between the two peaks of the distance distribution curve, and obtain the valley distance d' at the valley point. Use the valley distance d' as the basis for determining the number of independent houses in the connected houses. For any pair of scattering center points in the target cluster C, the valley distance d' is used as the connectivity distance threshold. If the distance between any two scattering center points is less than d', an undirected connected edge is established between the two scattering center points. Finally, using the existing connected component extraction algorithm, all scattering center points belonging to the same connected component are divided into the same initial group. Therefore, the target cluster C can be divided into several initial groups, each corresponding to an independent house.

[0035] If the number of peaks is equal to 1 or the number of peaks is greater than 2, hierarchical agglomerative clustering is used for partitioning, and the inter-class distance threshold is set to 1.5R (R is the ground resolution), thereby obtaining several initial groups of the target cluster C; the number of initial groups represents the number of independent houses in the clustered houses.

[0036] Step S34: Determine the boundary ambiguity points in the initial grouping and regroup the boundary ambiguity points to obtain each group marked as an independent house.

[0037] More specifically, step S34 includes: Obtain the first distance between the scattering center point in the initial group and the geometric center point of the initial group; Obtain the second distance between the scattering center point in the initial group and the geometric center point of its adjacent group; Compare the first and second distances to mark some of the scattering center points in the initial grouping as boundary ambiguity points; The first coordinate centroid is obtained from the set of scattering center points corresponding to the target angle in the initial group; The second coordinate centroid is obtained from the set of scattering center points corresponding to the target angle in adjacent groups; Based on the distances of the fuzzy boundary points relative to the first and second coordinate centroids, the fuzzy boundary points are regrouped to obtain groups that are marked as independent houses.

[0038] In this embodiment, for each initial group, the geometric center point is obtained; The above embodiments can only determine the number of individual houses in a clustered house; the house splitting still needs to be accurately performed based on SAR imaging features. For each scattering center point in the target cluster C of the adhered houses, first obtain the first distance between it and the geometric center point of its initial group F1. Furthermore, the second distance between the scattering center point and the geometric center point of its nearest neighboring initial group (adjacent group) F2 is obtained. ; when × > hour( A concentration of 1.2 to 1.5 can achieve good results, with the preferred concentration being... The scattering center point is located near the spatial boundary of the two initial groups F1 and F2, so the ownership of the house is uncertain, and it is marked as a boundary ambiguity point; thus, some scattering center points in the initial group F1 are marked as boundary ambiguity points. For each boundary ambiguity point obtained above, obtain the angle identifier k of its corresponding SAR image (hereinafter referred to as the target angle identifier, used to represent any identical angle identifier). Under the same target angle marker k, the overlapping offset of the same independent house is relatively similar. Therefore, the scattering center points from the corner of the same independent house have similar coordinate distances in the SAR image. Obtain the scattering center points from all SAR images of the target angle identifier k in the initial group F1 to obtain the corresponding set of scattering center points, and obtain the first coordinate centroid X1 of these scattering center points; Obtain the scattering center points from all SAR images of the target angle identifier k in the initial group F2 to obtain the corresponding set of scattering center points, and obtain the second coordinate centroid X2 of these scattering center points; Obtain the distances c1 and c2 between the boundary fuzzy point and the first coordinate centroid X1 and the second coordinate centroid X2, respectively; When c1 > c2, it means that the blurred boundary point is closer to the scattering center point under the same target angle label k in the initial group F2, so it is reassigned to the initial group F2; otherwise, it remains in the initial group F1. After regrouping all the points with ambiguous boundaries, the target cluster C is finally split into several sub-clusters / groups; Each sub-cluster / group represents and is labeled as an independent house. The convex hull of the sub-cluster is obtained as the initial range, and the geometric center point of the convex hull is used as the initial position of the independent house. At this point, each connected house is divided into several independent houses. Combined with the houses identified as independent houses in the previous operations, all independent house areas are obtained.

[0039] It should be noted that the set of scattering center points corresponding to the target angle identifier k in the initial group includes at least the corresponding boundary blur point, so the set of scattering center points is not empty; the set of scattering center points corresponding to the target angle identifier k in the adjacent group may be empty. In this case, it is not necessary to redistribute the boundary blur point, but to keep it in the initial group F1.

[0040] Step S4: The target cluster is divided into individual house areas according to the grouping. The house boundaries are corrected according to the azimuth relationship between the house boundaries of the individual house areas and the SAR sensor illumination direction to obtain the true planar outline of the individual houses. Considering that due to the overlapping effect, the strong echo generated by the wall will be shifted away from the SAR sensor, thus causing the actual range of the house to be stretched in the direction of the overlapping; Therefore, the initial location of the independent house is the actual location of the house, but the initial range differs from the actual house shape. In order to facilitate the automatic detection of changes in urban and rural house shapes, the house boundary needs to be corrected to obtain the true outline of the house. In the initial position and initial range of the independent house obtained in the above embodiment, the house boundary on the side closer to the SAR sensor is determined by the scattering center point of the corner facing the bottom of the wall. These corner scattering center points have no projection error on the ground; therefore, the house boundary on the side facing the wall is more consistent with the real boundary of the house. On the other side, the boundary of the house is formed by the boundary of the roof because the wall itself cannot receive SAR sensor signals. However, since the roof will produce a projection effect in the SAR image, it will be stretched away from the SAR sensor.

[0041] Specifically, please refer to Figure 5 Step S4, which involves correcting the house boundaries based on the azimuth relationship between the house boundaries of the detached house area and the SAR sensor illumination direction to obtain the true planar outline of the detached houses, includes: Step S41: Obtain the direction vector corresponding to the SAR sensor illumination direction in the SAR image under the target angle marker; Step S42: Obtain the minimum bounding rectangle of the convex hull corresponding to the independent house area, use the minimum bounding rectangle as the initial correction benchmark, and determine the house boundary. Step S43: Starting from the midpoint of the house boundary, obtain the ray along the direction vector, and determine the boundary type of the house boundary based on the directional relationship of the ray relative to the inside and outside of the house boundary. For the convex hull of each initial range (independent house area) of each independent house obtained in the above embodiments, its minimum bounding rectangle is obtained, and the minimum bounding rectangle is used as the initial correction reference of the house, and each side is used as the initial house boundary. Under the target angle identifier k, obtain the direction vector G of the SAR sensor illumination direction in the SAR image; For each edge in the initial correction datum of the detached house, determine its condition under the target angle label k; Take the midpoint of each side of the initial correction datum (minimum bounding rectangle), and draw a ray from the midpoint along the direction of the direction vector G. The bounded planar region enclosed by the convex hull of the initial correction datum is defined as the interior of the house, or the interior of the house boundary, which also defines the exterior of the house boundary (the outside of the house). If the ray points outside the boundary of the house, that is, the ray does not intersect with other sides after starting from that side of the smallest bounding rectangle, then that side is the back side, located at the far end of the detached house in the direction of the SAR sensor's illumination. If the ray points inside the boundary of the house, that is, the ray starts from that side of the smallest bounding rectangle, passes through the interior area of ​​the house, and then exits from the opposite side, then that side is the facing side, located near the detached house in the direction of the SAR sensor's illumination.

[0042] Step S44: Determine the house boundary to be corrected based on the boundary type of the house boundary, and correct the house boundary to obtain the true planar outline of the detached house.

[0043] More specifically, the boundary types of the building boundary include facing edges and back edges; step S44 includes: If the boundary of a house is determined to be a back edge under all angle markings, then the boundary of the house is determined to be a boundary to be corrected. Obtain the adjacent boundaries of the house boundary to be corrected. If the adjacent boundary is an facing edge, use the adjacent boundary as a reference to correct the house boundary to obtain the true plan outline of the detached house. If the adjacent boundary is a back edge, a new adjacent boundary is obtained. If the new adjacent boundary is a front edge, the house boundary to be corrected is corrected based on the new adjacent boundary to obtain the true plan outline of the detached house.

[0044] In this embodiment, the boundary conditions of detached houses are analyzed for SAR images under each angle marker k. When the boundary of a detached house is determined to be the facing side under a certain angle marker k, it means that there is a certain angle that ensures that this side is accurately measured. Therefore, this house boundary is used as the reference boundary of the detached house. If a certain boundary of a detached house is not identified as an facing edge under all angle indicators k, that is, it is identified as a back edge (called a always back edge), it means that the detached house is always facing away from the sensor. Therefore, this boundary of the detached house needs to be corrected and is called the boundary to be corrected. For the house boundary to be corrected, first obtain the house boundaries adjacent to it (referred to as "adjacent boundaries"). If an adjacent boundary is identified as an approaching edge, then that adjacent boundary is selected as the reference; based on the coordinates and orientation of that adjacent boundary, the orthogonal parallel constraint is used for deduction. Specifically, by utilizing the geometric characteristics that the facade lines of the same independent house are parallel or perpendicular to each other, and based on the adjacent boundary as the facing side, combined with the a priori standard length-to-width ratio of the building in the monitoring area, the true cut-off point of the house boundary to be corrected on the adjacent boundary is recalculated, and a perpendicular line is drawn from the new true cut-off point to the adjacent boundary to determine the true position of the house boundary to be corrected. Finally, the target endpoint coordinates of the corrected house boundary are output to obtain the corrected house boundary and determine the true planar outline of the independent house.

[0045] If adjacent boundaries are also always facing away, then trace along the building outline to find a new adjacent boundary that is facing away, and use this as a benchmark to perform the above parallel deduction; if the new adjacent boundary is still always facing away, then continue tracing along the building outline until there is a boundary that is facing away. If there is an extreme case where there is no facing away, it indicates that there is an error in the data collection or data processing process, and manual intervention is required.

[0046] Furthermore, the process of correcting the house boundary first yields the corrected rectangular outline of the detached house (the outline formed by the edges of the minimum bounding rectangle is the rectangular outline). This corrected rectangular outline is then spatially superimposed on the initial minimum bounding rectangle corresponding to the detached house (ensuring the maximum overlap). If the area of ​​the corrected rectangular outline is smaller than the area of ​​the initial minimum bounding rectangle corresponding to the detached house (indicating that the original convex hull has been stretched due to the overlay), then the corrected rectangular outline is taken as the true planar outline of the detached house. If the area of ​​the corrected rectangle is greater than or equal to the area of ​​the initial minimum bounding rectangle (indicating slight overlay and stretching, and no need for correction), then the original rectangle outline (initial minimum bounding rectangle) is directly used as the actual plan outline of the house; this rollback logic ensures that the correction operation will not amplify the error due to numerical anomalies.

[0047] Step S5: Based on the changes in the actual planar outline of the detached house during adjacent monitoring periods, determine the detection results of the house morphology change.

[0048] The actual planar outline of each independent house in the SAR image obtained in the above embodiment is then used for automatic change detection of each independent house based on multi-source temporal remote sensing. For the same monitoring area, multiple multi-source remote sensing images from different monitoring periods are stacked; the multi-source remote sensing images include high-resolution optical images, infrared images, and synthetic aperture radar images from different time phases. The above-described methods for separating individual houses and correcting boundaries are applied to the image stacks for each time period to obtain a set of real planar outlines of independent houses for each time period. Spatial overlay analysis of the real planar contour sets of two adjacent time periods is performed to match houses; For successfully matched detached houses, the spectral texture features of optical images and temperature anomaly areas of infrared images can be combined to help confirm whether the detached house has been renovated, expanded or demolished (existing technology). When the rate of change of the outline area of ​​a detached house exceeds a preset change threshold (e.g., 20%), it is determined that the detached house has undergone a change in form, and the change type is recorded as expansion or partial demolition based on the increase or decrease in area. Houses that existed in the previous period but failed to match in the next period are judged to be completely demolished; outlines that were added in the next period and had no corresponding independent houses in the previous period are judged to be newly built. The detected morphological changes of independent houses are analyzed to generate standard vector format house morphological change detection result patch files, which are then used by urban and rural housing management departments.

[0049] This invention utilizes the spatial distribution differences and viewpoint-dependent laws of multi-angle SAR scattering centers to effectively extract individual building identification information from overlapping echoes, accurately pinpointing temporal radiation differences to individual buildings. This solves the core problems of traditional single-angle SAR being unable to deconstruct overlapping and overlapping data, multi-source temporal SAR observations being unable to extract individual building identification information from independent buildings, and radiation changes being difficult to accurately attribute to specific buildings. The adhering house separation technology provided by this invention is based on the spatial distribution of multi-angle scattering centers and homogeneous geometric constraints. It can automatically separate adhering houses that have merged due to overlapping without relying on external house base vector reference data, thus realizing the accurate identification of independent houses in dense building areas and significantly improving the accuracy of individual detection of independent houses. This invention also corrects the building boundary by using the orientational relationship between the building boundary and the SAR sensor illumination direction, effectively eliminating geometric stretching distortion caused by SAR overlay, obtaining the true planar outline of the building, ensuring the reliability of subsequent outline change identification, and finally obtaining accurate building shape change detection results. It is suitable for large-scale, high-frequency automated supervision applications of urban and rural housing management departments.

[0050] Example 2:

[0051] This invention also proposes an automatic detection device for urban and rural building morphology changes based on multi-source time-series remote sensing imagery. The device can be a data processing device such as a computer or server, or a combination of multiple devices.

[0052] Based on the hardware structure of the automatic detection device for urban and rural housing morphology changes based on multi-source time-series remote sensing images described above, various embodiments of the automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing images of the present invention are implemented.

[0053] Furthermore, the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores an automatic detection program for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery. When executed by a processor, the automatic detection program for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery implements the steps of the method corresponding to the above-described automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery.

[0054] The method implemented when the automatic detection program for changes in urban and rural housing morphology based on multi-source time-series remote sensing images is executed can be referred to in various embodiments of the automatic detection method for changes in urban and rural housing morphology based on multi-source time-series remote sensing images of the present invention, and will not be repeated here.

Claims

1. An automatic detection method for urban and rural housing morphological changes based on multi-source time-series remote sensing imagery, characterized in that, The method includes: Acquire multi-source time-series SAR intensity image data of the monitoring area, and determine the scattering center point based on the pixel decibel value of each SAR image; By projecting the scattering center points of all SAR images onto the same coordinate system, a set of scattering spatial points is obtained. The set of scattering spatial points is then clustered to obtain clusters that are marked as houses. Identify the target clusters marked as contiguous houses, and determine the grouping of each house marked as an independent house in the target cluster based on the distance distribution between the scattering centers in the target cluster; The target cluster is divided into individual house areas according to the grouping. The house boundaries are corrected according to the azimuth relationship between the house boundaries of the individual house areas and the SAR sensor illumination direction to obtain the true planar outline of the individual houses. Based on the changes in the actual planar outline of detached houses during adjacent monitoring periods, the results of house shape change detection are determined.

2. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 1, characterized in that, The determination of the scattering center point based on the pixel decibel value of each SAR image includes: Divide any SAR image into individual pixel sub-blocks and obtain the mean and standard deviation of the decibel values ​​of the pixels in each pixel sub-block. The strong and weak scattering thresholds of the sub-blocks are determined based on the mean and standard deviation of the decibel values. Pixels in the sub-blocks whose decibel values ​​are greater than the strong and weak scattering thresholds are marked as strong scattering candidate points. Non-maximum suppression is applied to strong scattering candidate points to obtain the scattering center points among the strong scattering candidate points.

3. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 2, characterized in that, The step of performing non-maximum suppression on strong scattering candidate points to obtain the scattering center points among the strong scattering candidate points includes: Determine strong scattering candidate points and their pixel neighborhoods. When the pixel decibel value of a strong scattering candidate point is less than the pixel decibel value of any pixel in its neighborhood, the strong scattering candidate point is suppressed. Each suppressed strong scattering candidate point is marked as a non-scattering center point, thus obtaining the remaining scattering center points among the strong scattering candidate points.

4. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 1, characterized in that, The process of clustering scattering spatial points to obtain clusters labeled as houses includes: Based on the ground resolution of SAR images, DBSCAN clustering is performed on the scattering spatial point set to obtain each cluster labeled as a house. The convex hull of the cluster is used as the initial extent of the house, and the geometric center of the convex hull is used as the initial position of the house.

5. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 1, characterized in that, The determination of the target clusters marked as adherent houses includes: Obtain the center point of the ascending orbit scattering from the ascending orbit perspective in the cluster, and determine the first spatial distribution center of the ascending orbit scattering center point; Obtain the center point of the descending orbit scattering from the descending orbit perspective in the cluster, and determine the second spatial distribution center of the descending orbit scattering center point; Determine the planar distance between the first spatial distribution center and the second spatial distribution center; If the planar distance is less than or equal to a preset distance threshold, the cluster is marked as an independent house; If the planar distance is greater than a preset distance threshold, the cluster is marked as an adhered house and the cluster is used as the target cluster.

6. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 1, characterized in that, The step of determining the grouping of each labeled independent house in the target cluster based on the distance distribution between scattering centers in the target cluster includes: Obtain the point-to-point distance between any two scattering center points in the target cluster and obtain the set of point-to-point distances; A histogram is constructed based on the set of point-to-point distances, and a curve is fitted to the histogram to obtain the distance distribution curve. The initial grouping in the target cluster is determined based on the number of peaks in the distance distribution curve; Identify the boundary ambiguity points in the initial grouping and regroup the boundary ambiguity points to obtain the groups that are marked as independent houses.

7. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 6, characterized in that, The method of determining the initial grouping in the target cluster based on the number of peaks in the distance distribution curve includes: If there are two peaks in the distance distribution curve, then obtain the valley distance between the two peaks at the valley point. If the point-to-point distance between any two scattering centers in the target cluster is less than the trough distance, then the two scattering centers are determined to belong to the same initial group. If the number of peaks in the distance distribution curve is one or more, hierarchical agglomerative clustering is used to divide the target cluster into initial groups.

8. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 6, characterized in that, The process of determining the boundary ambiguity points in the initial grouping and regrouping the boundary ambiguity points to obtain groups labeled as independent houses includes: Obtain the first distance between the scattering center point in the initial group and the geometric center point of the initial group; Obtain the second distance between the scattering center point in the initial group and the geometric center point of its adjacent group; Compare the first and second distances to mark some of the scattering center points in the initial grouping as boundary ambiguity points; The first coordinate centroid is obtained from the set of scattering center points corresponding to the target angle in the initial group; The second coordinate centroid is obtained from the set of scattering center points corresponding to the target angle in adjacent groups; Based on the distances of the fuzzy boundary points relative to the first and second coordinate centroids, the fuzzy boundary points are regrouped to obtain groups that are marked as independent houses.

9. The automatic detection method for urban and rural housing morphology changes based on multi-source time-series remote sensing imagery according to claim 1, characterized in that, The process of correcting the house boundaries based on the azimuth relationship between the house boundaries of the detached house area and the SAR sensor illumination direction to obtain the true planar outline of the detached houses includes: Obtain the direction vector corresponding to the SAR sensor illumination direction in the SAR image under the target angle marker; Obtain the minimum bounding rectangle of the convex hull corresponding to the independent house area, use the minimum bounding rectangle as the initial correction benchmark, and determine the house boundary; Starting from the midpoint of the building boundary, a ray along the direction vector is obtained. The boundary type of the building boundary is determined based on the directional relationship of the ray relative to the inside and outside of the building boundary. The boundary type of the house boundary is used to determine the house boundary to be corrected, and the corrected house boundary is used to obtain the true planar outline of the detached house.

10. The automatic detection method for urban and rural housing morphology changes based on multi-source temporal remote sensing imagery according to claim 9, characterized in that, The boundary types of house boundaries include facing edges and back edges; the process of determining the house boundary to be corrected based on the boundary type, and correcting the house boundary to be corrected to obtain the true planar outline of the detached house includes: If the boundary of a house is determined to be a back edge under all angle markings, then the boundary of the house is determined to be a boundary to be corrected. Obtain the adjacent boundaries of the house boundary to be corrected. If the adjacent boundary is an facing edge, use the adjacent boundary as a reference to correct the house boundary to obtain the true plan outline of the detached house. If the adjacent boundary is a back edge, then obtain the new adjacent boundary of the adjacent boundary; if the new adjacent boundary is a front edge, then use the new adjacent boundary as a reference to modify the house boundary to be modified and obtain the true plan outline of the detached house.

Images