Multi-temporal remote sensing change detection optimization method for territorial surveying and mapping

Abstract

The invention relates to a multi-temporal remote sensing change detection optimization method for territorial surveying and mapping. The method comprises the following steps: acquiring a multi-temporal remote sensing image in the same region and dividing the multi-temporal remote sensing image into pattern spots or regular grid operation units; selecting a ground object which can be repeatedly recognized in the multi-temporal image and meets a preset stability condition as an anchor point, performing anchor-point-based consistency re-check on each temporal image, and enabling the temporal which does not pass the re-check not to be included in an available temporal set of the operation unit; determining reference period candidate time phases in the available time phase set, evaluating and sorting according to a preset reliability judgment rule, and selecting the first sorted time phase as a reference period; when the candidate time phases do not meet the preset reliability condition, selecting the candidate time phase with the maximum passing number of the anchor points as a temporary reference, if the candidate time phases are parallel, selecting the candidate time phase with the maximum passing rate of the anchor points, and marking the candidate time phase with low reliability; and comparing the reference period with other available time phases except the reference period, extracting a change region, and outputting the change region, a reference period identifier and a reliability mark.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing information processing and land surveying technology, specifically to a multi-temporal remote sensing change detection and optimization method for land surveying. Background Technology

[0002] Existing technologies, such as the land use dynamic change monitoring method and system based on multimodal remote sensing data disclosed in patent document CN120356113A, reveal a series of significant shortcomings and drawbacks in current multi-temporal remote sensing change detection optimization methods used for land surveying in engineering practice. Firstly, existing technologies generally use land use type or sub-region as the analysis unit, relying on a consistent baseline database and difference index for change determination. Their core assumption is that the same land use type possesses relatively fixed statistical characteristics under stable conditions. However, in actual land surveying scenarios, significant spatial heterogeneity and structural differences often exist within the same land category. For example, construction land in urban-rural fringe areas may include both long-term stable roads and buildings, as well as frequently changing temporary construction land and construction areas. This internal difference leads to insufficient representativeness of the baseline characteristic mean and standard deviation, thus reducing the difference index's ability to distinguish between real changes and data quality fluctuations, and easily resulting in strong threshold dependence and poor regional adaptability.

[0003] Secondly, existing solutions emphasize multimodal data fusion and feature correlation analysis, but their consistency judgments remain largely at the feature level or statistical level, lacking direct constraints on the geometric consistency and spatial stability between images. This is especially true in high-resolution land surveying applications, where slight registration errors, shadow changes, or local occlusion can cause linked shifts in multimodal features, leading to misjudgments as abnormal areas. This is particularly evident when monitoring areas with high-frequency changes such as urban renewal and road construction. Thirdly, existing technologies primarily distinguish between data quality anomalies and sudden event anomalies by the number of differences in the correlation matrix. However, this method relies on empirically set feature dimensions and threshold N1, and once anomaly diagnosis begins, it often directly marks areas at the sub-region scale, lacking the ability to spatially trace the source of anomalies. It is difficult to determine whether anomalies are caused by a few unstable targets, local occlusion, or seasonal differences in imaging conditions, thus easily introducing too many areas requiring manual verification into land surveying results and reducing the efficiency of automated processing. Meanwhile, while existing methods propose data restoration mechanisms, the restoration process still relies on statistical regression of historical multimodal characteristics, failing to fully consider the geometric constraints of stable features within the same region. The reliability of the restoration results in terms of spatial location and boundary morphology is difficult to quantify and assess, which poses a significant limitation in land surveys and change surveys where high accuracy of changed boundaries is required. Furthermore, these methods typically assume a one-way comparison between real-time data and the baseline database in multi-temporal processing, lacking a bidirectional consistency verification and arbitration mechanism between multiple temporal phases. When an image from a particular phase is affected by extreme imaging conditions, it is easily marked as an anomaly or a high-priority change, resulting in decreased temporal utilization and information waste. Summary of the Invention

[0004] The purpose of this invention is to provide an optimized method for multi-temporal remote sensing change detection in land surveying, thereby addressing some of the shortcomings and deficiencies pointed out in the background art.

[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: a multi-temporal remote sensing change detection and optimization method for land surveying, comprising: acquiring multi-temporal remote sensing images of the same area and dividing them into patches or regular grid operation units;

[0006] Select ground objects that can be repeatedly identified in multi-temporal images and meet preset stability conditions as anchor points, and perform consistency verification based on anchor points for each temporal image. Temporal images that fail the verification are not included in the available temporal set of this operation unit.

[0007] Within the set of available time phases, candidate reference time phases are determined, evaluated and sorted according to preset reliability judgment rules, and the first-ranked one is selected as the reference time phase. When none of the candidate time phases meet the preset reliability conditions, the candidate time phase with the maximum number of anchor point passes is selected as the temporary reference. If they are tied, the one with the maximum anchor point pass rate is selected and marked as low reliability. The change area is extracted by comparing the reference time phase with the other available time phases excluding the reference time phase, and the change area, reference time phase identifier, and reliability marker are output.

[0008] Furthermore, the selection of anchor points includes extracting anchor point candidate objects from each temporal image, retaining only candidate objects that can be repeatedly located in no less than a preset number of temporal images; within the same work unit, no less than two candidate objects constitute a candidate anchor point group, and when the mutual distance ratio or included angle relationship of the anchor point group in different temporal images meets a preset tolerance, it is determined as an anchor point, otherwise it is discarded.

[0009] Furthermore, the consistency review includes, for the reference period candidate time phase and any other candidate time phase, performing anchor point consistency determination on the other candidate time phase based on the reference period candidate time phase, and performing anchor point consistency determination on the reference period candidate time phase based on the other candidate time phase; when the two determination results are inconsistent or only one determination passes, the other candidate time phase is determined as an object to be arbitrated, and after arbitration, it is decided to remove it or mark it as low confidence for subsequent sorting.

[0010] Furthermore, when the work unit is marked as a low-reliability reference, the candidate time phase ranked first is selected as a temporary reference, and the candidate time phase ranked second and passing the consistency review is selected as an auxiliary reference; the candidate change regions are obtained by comparing the temporary reference and the auxiliary reference with the remaining available time phases, respectively. Let the same candidate change region be region A and region B under the two references, and calculate their spatial overlap. for:

[0011]

[0012] in, This represents the intersection of region A and region B. Let A represent the union of regions A and B. This represents the area or number of pixels corresponding to the intersection or union. Indicates the spatial overlap; when If the value is not lower than the preset threshold, the output is a confirmed change; otherwise, the output is a change to be verified, accompanied by a low reliability reference mark.

[0013] Furthermore, the anchor point consistency determination includes: locating the same anchor point in the candidate reference phase and the other candidate phases and obtaining its relative displacement; when the relative displacements of multiple anchor points in the same work unit meet the preset consistency conditions, the determination is passed; anchor points whose relative displacements deviate from the majority of anchor points are recorded as abnormal anchor points and do not participate in the determination.

[0014] Furthermore, when the two determinations are inconsistent or only one determination passes, a third candidate time phase, different from the candidate time phase of the reference period and the other candidate time phases, is selected from the candidate time phase set of the reference period as the arbitration time phase. Anchor point consistency determination is then performed on the candidate time phase of the reference period and the other candidate time phases based on the arbitration time phase. Candidate time phases that fail the arbitration determination are eliminated, and candidate time phases that pass the arbitration determination but have one-way inconsistencies are marked as low confidence for subsequent sorting.

[0015] Furthermore, the preset consistency condition is determined by multi-anchor point voting: within the same work unit, the dominant direction and dominant amplitude range of the relative displacement of each anchor point are statistically analyzed. When the number of anchor points falling into the dominant amplitude range is not less than a preset proportion and the anchor points are dispersed within the work unit, the anchor point consistency determination is passed.

[0016] Furthermore, the determination of abnormal anchor points is carried out in a hierarchical manner: when the relative displacement of an anchor point deviates from the dominant amplitude range but its neighboring anchor points meet the consistency condition, the anchor point is marked as a temporary abnormal anchor point and is removed only in this determination; when the anchor point is marked as a temporary abnormal anchor point in at least a preset number of time-to-time comparisons, it is determined as a permanent abnormal anchor point and removed from subsequent consistency reviews.

[0017] Furthermore, the neighboring anchor points are determined according to the spatial neighborhood within the work unit, and the determination of temporary abnormal anchor points includes occlusion screening: when the local area of ​​the anchor point shows signs of cloud, shadow or occlusion in the candidate phase of the reference period or other candidate phases, the anchor point is marked as a temporary abnormal anchor point; when the occlusion signs disappear and the relative displacement of the anchor point returns to the dominant amplitude range, the temporary abnormality mark is removed and the anchor point resumes participation in the consistency determination.

[0018] Furthermore, the permanent anomalous anchor point is determined only when the cross-temporal constraint is met: when the anchor point is marked as a temporary anomalous anchor point in at least a preset number of time-phase comparisons under different seasons or different imaging conditions, it is determined as a permanent anomalous anchor point.

[0019] The beneficial effects of this invention are as follows: The multi-temporal remote sensing change detection optimization method for land surveying proposed in this invention introduces stable, repeatable features across time phases as anchor points and uses operational units as the basic processing objects. It performs bidirectional consistency verification and arbitration judgment on multi-temporal remote sensing images, effectively avoiding misjudgments caused by differences in imaging conditions, local occlusion, or geometric micro-offsets. Through step-by-step screening of temporal availability and adaptive selection of reference periods, change detection is established on a reliable reference basis, significantly improving the stability and consistency of multi-temporal remote sensing change analysis in complex land scenarios and reducing the interference of non-realistic changes on surveying results.

[0020] Furthermore, this invention achieves dynamic self-purification of the anchor point set through multi-anchor point voting, hierarchical processing of abnormal anchor points, and cross-temporal constraint mechanisms, effectively suppressing the impact of local anomalies and long-term instability factors on change determination. In low-reliability reference scenarios, it introduces dual-reference cross-validation and spatial overlap determination to further enhance the credibility and interpretability of change results. Therefore, this invention not only improves the accuracy of multi-temporal change detection in land surveying, but also allows the output results to directly serve change data entry, key area verification, and subsequent surveying and mapping workflows. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the multi-temporal remote sensing change detection process based on anchor point consistency verification according to the present invention.

[0022] Figure 2 This is a logic diagram for determining the consistency and confirming the changes of multi-temporal anchor points in this invention.

[0023] Figure 3 This is a functional relationship framework diagram of anchor point consistency determination and abnormal evolution in this invention.

[0024] Figure 4 This is a schematic diagram of anchor point candidate object extraction and geometric stability verification in work unit G142 in Embodiment 1 of the present invention.

[0025] Figure 5 This is a schematic diagram of the reference period candidate phase sorting and consistency verification based on the anchor point passage situation in Embodiment 1 of the present invention.

[0026] Figure 6 This is a schematic diagram of dual-reference change detection and spatial overlap determination under low-reliability reference conditions in Embodiment 1 of the present invention.

[0027] Figure 7 This is a schematic diagram of the statistical analysis of the relative displacement amplitude direction of anchor points and the consistency determination of multi-anchor point voting in Embodiment 2 of the present invention.

[0028] Figure 8This is a schematic diagram of the cross-temporal evolution process of the abnormal anchor point hierarchical processing in Embodiment 2 of the present invention.

[0029] Figure 9 This is a schematic diagram of the permanent anomaly anchor point determination based on spatial neighborhood and occlusion screening in Embodiment 2 of the present invention. Detailed Implementation

[0030] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] Combined with appendix Figure 1 This invention relates to an optimized method for multi-temporal remote sensing change detection in land surveying, comprising the acquisition of multi-temporal remote sensing images and the construction of operational units. Multi-temporal remote sensing images are remote sensing image data covering the same land surveying area but acquired at different times. The images can originate from the same remote sensing sensor or different remote sensing sensors, and the acquisition time interval can be set according to the needs of the land surveying task. To ensure the comparability of subsequent change detection, after image acquisition, the multi-temporal remote sensing images are first subjected to spatial range unification processing, so that each temporal image corresponds to the same surveying area under the same geographic coordinate reference.

[0032] After image acquisition, the land surveying area is divided into multiple operational units, which serve as the basic processing objects for change detection and assessment. Operational units can be divided into patch form, corresponding to existing land use units, administrative units, or natural feature boundaries, based on the needs of land surveying operations; or into regular grid form, where regular grids have uniform size and regular shape in space to ensure consistent processing scale across operational units. Through these methods, each operational unit maintains a clear and fixed spatial correspondence in remote sensing images acquired at different times.

[0033] By mapping multi-temporal remote sensing images onto patch or regular grid work units, unified organization of multi-temporal remote sensing data at the work unit level is achieved. This provides a stable data foundation for subsequent anchor point selection, consistency verification, reference period determination, and change area extraction based on work units, thereby avoiding the instability and error accumulation problems caused by directly performing change detection at the whole image level.

[0034] After acquiring multi-temporal remote sensing images and dividing the work units, the ground features within each work unit are screened to determine anchor points for consistency verification. Anchor points are ground features that can be stably identified in all multi-temporal remote sensing images. These objects exhibit stable location, minimal morphological changes, and are not easily affected by seasonal variations or temporary occlusion across different imaging times. Therefore, ground features are extracted from each temporal image, and their spatial location, morphological features, and identifiability are compared. Only ground features that can be repeatedly located in at least a preset number of temporal images are retained as candidate anchor points.

[0035] Based on the candidate anchor points, the candidates are further screened according to preset stability conditions. These conditions include that the spatial displacement of the candidate object in multi-temporal images is within an allowable range, and that there is no significant occlusion or structural damage in its surrounding area. Candidate objects that meet the preset stability conditions are identified as anchor points and used for subsequent consistency verification processing.

[0036] After determining the anchor points, a consistency check is performed on each temporal remote sensing image within each work unit, based on the anchor points. The consistency check compares the spatial relationships and relative changes of the same anchor point in images from different time periods to determine the comparability between the temporal images. If a temporal image shows a significant offset, missing elements, or unstable identification at the anchor point location, that temporal image is deemed to have failed the consistency check. Temporal images that fail the consistency check are not included in the available temporal image set of the corresponding work unit, thus avoiding the introduction of geometrically inconsistent or unreliable temporal data into the subsequent change detection process.

[0037] After determining the available time phase set for each work unit, reference period screening is performed on each time phase image in the available time phase set. The time phase images in the available time phase set are identified as candidate reference periods, and each candidate reference period is evaluated according to a preset reliability judgment rule. The reliability judgment rule reflects the comprehensive reliability of the candidate time phases in terms of imaging quality and consistency, and its evaluation process incorporates at least the anchor point consistency verification results and image quality stability. Based on the evaluation results, the candidate reference periods are ranked, and the candidate time phase ranked first is selected as the reference period for that work unit, used as the benchmark for subsequent change detection.

[0038] When none of the candidate reference periods meet the preset reliability conditions, a temporary reference strategy is adopted to avoid interrupting the change detection process due to a lack of high-quality references. In this case, the candidate period with the largest number of anchor points passing is selected as the temporary reference, where the number of anchor points passing represents the number of anchor points that pass the consistency review. When multiple candidate periods have the same number of anchor points passing, and all of them are the maximum value, their anchor point pass rates are further compared. The anchor point pass rate is the ratio of the number of anchor points that pass the review to the total number of anchor points participating in the review, and the candidate period with the largest anchor point pass rate is selected as the temporary reference. For the temporary reference period selected in the above manner, the corresponding work unit is marked as a low-reliability reference to indicate that the reliability of this reference period is lower than that of the normal reference period.

[0039] After determining the reference period or provisional reference period, the reference period is compared with other available temporal images (excluding the reference period) to extract the changed areas within the corresponding work unit. The extracted changed areas are output along with the reference period information used, and simultaneously, the reference period identifier and reference period reliability marker for the work unit are also output. Through this method, the change detection results not only include spatial change information but also clearly reflect the reference period on which the change determination is based and its reliability, thus providing a reliable basis for subsequent data entry into the land surveying database and manual verification.

[0040] Combined with appendix Figure 2 Anchor point selection is conducted independently within each operational unit to ensure that the anchor points accurately reflect the spatial stability of that operational unit across multi-temporal remote sensing images. Specifically, ground features within the operational unit are extracted from remote sensing images of each temporal phase to obtain a set of candidate objects for anchor point selection. Subsequently, the candidate objects are repeatedly located in images acquired at different times, retaining only those that can be stably identified and located in images of at least a preset number of temporal phases, thereby eliminating ground features that appear only in a few temporal phases or are easily affected by occlusion.

[0041] After completing the initial screening, at least two candidate objects are selected from the candidate objects within the same work unit to form a candidate anchor point group. For each candidate anchor point group, the spatial relationships between the candidate objects are calculated in remote sensing images acquired at different times, and the changes in their distance ratios or angle relationships at different time phases are compared. When the changes in the distance ratios or angle relationships of the candidate anchor point group in multi-temporal images are all within a preset tolerance range, the candidate anchor point group is determined to have good geometric stability in time, and the candidate objects within the candidate anchor point group are identified as anchor points. If the spatial relationships of the candidate anchor point group at any time phase exceed the preset tolerance, the candidate object combination is considered to lack sufficient stability, and the corresponding candidate object is removed from the anchor point candidate set, thereby ensuring that the finally selected anchor points have a reliable consistency basis in multi-temporal remote sensing images.

[0042] After anchor point selection, a consistency review is performed on the reference period candidate phases and other candidate phases within the work unit to determine whether the images of different time phases are reliably comparable. The consistency review adopts a two-way judgment method. First, using the reference period candidate phase as a benchmark, an anchor point consistency judgment is performed on any other candidate phase. By comparing the spatial position relationship and relative change of each anchor point in the two phase images, it is determined whether the other candidate phases are consistent with the reference period candidate phase. Then, using the other candidate phases as a benchmark, the same anchor point consistency judgment is performed on the reference period candidate phase, thus forming a two-way consistency evaluation result.

[0043] The aforementioned two-way judgment effectively avoids the biases caused by one-way judgment. When the two anchor point consistency judgments are consistent and both pass, the candidate reference period is considered to have good geometric consistency and comparability with other candidate reference periods. When the two judgments are inconsistent or only one judgment passes, the current judgment result is considered to have uncertainty. For candidate reference periods with uncertainty, they are identified as objects to be arbitrated and do not directly participate in subsequent reference period ranking or change detection processing, but instead enter the arbitration stage.

[0044] During the arbitration phase, additional criteria are introduced based on pre-defined arbitration rules to further assess the candidates for arbitration. After arbitration, a final decision is made on the candidate time phases based on the arbitration results. If the arbitration results indicate that the candidate time phase lacks reliable consistency, it is removed from the available time phase set of the corresponding work unit. If the arbitration results indicate that the candidate time phase still has comparability to a certain extent but has unstable factors, it is marked as low reliability and allowed to participate in subsequent ranking, but its ranking priority is reduced.

[0045] When the results of two anchor point consistency determinations performed on a candidate reference period and other candidate periods are inconsistent, or when only one determination passes, the current consistency review result is considered disputed and requires further determination through an arbitration mechanism. In this case, a third candidate period, which is different from both the candidate reference period and other candidate periods, is selected from the set of candidate reference periods as the consistency judgment benchmark for arbitration. This third candidate period has passed basic quality checks during the preliminary screening process and meets the conditions for use in consistency analysis.

[0046] After determining the arbitration period, using it as a benchmark, anchor point consistency determinations are performed on the candidate reference period and other candidate periods. The determination process is consistent with the aforementioned consistency review process, used to assess the geometric consistency and comparability between each candidate period and the arbitration period. By comparing the two sets of arbitration determinations, it is further confirmed which candidate reference period is more consistent with the other candidate periods in the overall timeline.

[0047] Based on the arbitration decision, the candidate time phases are processed in the final stage. When a candidate time phase fails to pass the consistency judgment with the arbitration time phase, it is considered that the candidate time phase has obvious problems in terms of geometric consistency or image stability, and it is removed from the available time phase set of the corresponding work unit. When a candidate time phase passes the arbitration judgment but still fails the one-way consistency judgment, it is considered that the candidate time phase has a certain degree of comparability but insufficient stability, and it is marked as low confidence. Its priority is reduced in the subsequent reference period sorting and change detection process, thereby realizing the hierarchical management and rational utilization of images of different time phases.

[0048] When a work unit is marked as a low-reliability reference during the reference period screening process, a dual-reference judgment mechanism is introduced to enhance the credibility of the change detection results and avoid distortion due to insufficient reliability of a single reference period. Specifically, in the ranking results of the candidate time phases of the reference period, the candidate time phase ranked first is selected as a temporary reference, while the candidate time phase ranked second and having passed the consistency review is selected from the remaining candidate time phases as an auxiliary reference, so as to ensure that both references have a basic temporal consistency basis.

[0049] After determining the temporary and auxiliary references, each reference is used as a benchmark and compared with other available time phases (excluding the corresponding reference) to obtain two sets of candidate change regions within the same work unit. For the same change target extracted under both reference conditions, let region A be the change region obtained with the temporary reference as the benchmark, and region B be the change region obtained with the auxiliary reference as the benchmark. To measure the consistency of the change results under the two references, the spatial overlap between region A and region B is calculated. The calculation method is as follows:

[0050]

[0051] in, This represents the spatial intersection of region A and region B. The symbol represents the spatial union of regions A and B. This indicates the area size or the number of pixels contained in the corresponding spatial region. It is used to characterize the degree of spatial consistency of candidate change regions under two reference conditions.

[0052] After completing the spatial overlap calculation, Compare with a preset threshold. When spatial overlap When the value is not lower than a preset threshold, the changed region is considered to have high consistency and stability under different reference conditions, and is output as a confirmed changed region; when the spatial overlap is... If the change is below the preset threshold, it is considered that the change result is greatly affected by unreliable factors in the reference period. It is then output as a change area to be verified, and a low-reliability reference mark is attached to remind the subsequent review or manual verification process of land surveying results to pay close attention to it.

[0053] Combined with appendix Figure 3 Anchor point consistency determination is used to assess the degree of spatial geometric consistency between a candidate reference time phase and other candidate time phases. Specifically, during consistency determination, the same anchor point is first located in both the candidate reference time phase and other candidate time phases. By matching the spatial positions of the anchor point in the images of the two time phases, the relative displacement information of the anchor point between the two time phases is obtained. The relative displacement reflects the spatial offset of the anchor point in the images at different times and is used to characterize the geometric consistency between the images.

[0054] After obtaining the relative displacements of multiple anchor points, a comprehensive analysis of the relative displacements of each anchor point is performed within the same working unit. When the relative displacements of multiple anchor points are consistent in direction and amplitude and meet the preset consistency conditions, the candidate reference period is considered to have good geometric consistency with other candidate time phases, and the anchor point consistency judgment result is passed. The preset consistency conditions are used to limit the allowable range of relative displacement fluctuations to exclude abnormal offsets caused by changes in imaging conditions or local interference.

[0055] During the consistency analysis, anchor points whose relative displacements significantly deviate from the characteristics of most anchor points are identified as anomalous anchor points. Anomalous anchor points may be affected by local occlusion, identification errors, or changes in ground features, and their relative displacements are not representative; therefore, they are not included in the overall consistency assessment process.

[0056] To improve the stability and anti-interference capability of anchor point consistency determination, the preset consistency conditions are determined using a multi-anchor point voting method. Within the same work unit, the relative displacement of each anchor point participating in the consistency determination is calculated between the candidate reference time phase and other candidate time phases, and the directional and amplitude characteristics of the relative displacement are statistically analyzed. By summarizing the distribution of the relative displacements of each anchor point, the direction of the most frequently occurring relative displacement is determined as the dominant direction, and the dominant amplitude interval corresponding to this dominant direction is determined to characterize the overall displacement characteristics of the work unit under the current time phase comparison.

[0057] After determining the dominant direction and dominant amplitude range, the number of anchor points falling within the dominant amplitude range is counted and compared with the total number of anchor points participating in the judgment. When the number of anchor points falling within the dominant amplitude range is not less than a preset proportion, and the anchor points are dispersed within the working unit space rather than concentrated in a local area, it is considered that most anchor points have given consistent judgment results for the current time-relative comparison, thus the anchor point consistency judgment is deemed passed. By introducing the dispersed distribution constraint, the dominant influence of a few local abnormal anchor points on the overall judgment result is effectively avoided.

[0058] In the aforementioned consistency determination process, anchor points deviating from the dominant amplitude range are further subjected to abnormal anchor point classification processing. When the relative displacement of an anchor point deviates from the dominant amplitude range, but other spatially adjacent anchor points still meet the consistency conditions, it is considered that the anchor point may be affected by local noise or short-term interference, and it is marked as a temporary abnormal anchor point, which is only removed in the consistency determination of the current time-to-time comparison. If, in subsequent time-to-time comparisons, the anchor point is marked as a temporary abnormal anchor point in no less than a preset number of comparisons, it is considered that the anchor point does not have long-term stability, and it is determined as a permanent abnormal anchor point, and it is completely removed from the subsequent consistency review process.

[0059] To further improve the accuracy of abnormal anchor point identification, a neighboring anchor point and occlusion screening mechanism is introduced when handling temporary abnormal anchor points. Neighboring anchor points are determined based on the spatial neighborhood relationship within the work unit. That is, taking the anchor point to be identified as the center, other anchor points adjacent to its position within a preset spatial range are selected as neighboring anchor points to assist in determining the cause of the abnormal relative displacement of the anchor point.

[0060] When determining temporary abnormal anchor points, in addition to considering whether the relative displacement of the anchor point deviates from the dominant amplitude range, occlusion screening is also performed on the local area where the anchor point is located. Occlusion screening analyzes the image features of the area surrounding the anchor point in the candidate reference phase or other candidate phases to determine whether there are signs of clouds, shadows, or other occlusions. When occlusion is detected in the local area of ​​the anchor point, it is considered that the current abnormal relative displacement of the anchor point may be caused by changes in imaging conditions rather than true geometric inconsistency. The anchor point is then marked as a temporary abnormal anchor point and removed during the current consistency determination process.

[0061] In subsequent time-relative comparisons, objects marked as temporary anomalous anchor points are continuously tracked. When the occlusion indication disappears in subsequent candidate reference phases or other candidate phases, and the relative displacement of the anchor point falls back into the dominant amplitude range in the new time-relative comparison, the anomalous state of the anchor point is considered resolved, its temporary anomalous anchor point label is removed, and it is reinstated to participate in anchor point consistency determination. Through the above dynamic processing method based on spatial neighborhood and occlusion screening, false anomalies caused by temporary occlusion and real unstable anchor points are effectively distinguished.

[0062] To avoid erroneously rejecting anchors with long-term stability due to short-term environmental changes or occasional imaging anomalies, a cross-temporal constraint mechanism is introduced to determine permanent anomalous anchors. This cross-temporal constraint ensures that the determination of anchor anomalies has sufficient temporal and conditional coverage, thereby improving the reliability of anomalous anchor identification.

[0063] Specifically, for ground features marked as temporary anomalous anchor points, continuous tracking and analysis are conducted during subsequent multi-temporal consistency verification. Only when an anchor point is marked as a temporary anomalous anchor point in at least a preset number of temporal comparisons will it proceed to the permanent anomalous anchor point determination process. The temporal comparisons cover remote sensing images acquired in different seasons, or remote sensing images acquired under imaging conditions with differences in imaging angle, lighting conditions, or sensor status, to ensure that the anomalous behavior of the anchor point is not caused by a single seasonal feature or a single imaging condition.

[0064] If an anchor point consistently exhibits an abnormal state in multiple time-phase comparisons that satisfy the aforementioned cross-temporal constraints, it is considered to lack long-term stability and is identified as a permanently abnormal anchor point. It will not be considered a valid anchor point in subsequent consistency reviews and change detection processes. This cross-temporal constraint mechanism effectively avoids misclassifying anchor points affected by short-term disturbances as permanently abnormal.

[0065] Example 1:

[0066] This embodiment uses a suburban area as the experimental zone. The experimental zone covers approximately 12.6 square kilometers and includes typical land types such as cultivated land, construction land, water bodies, and forest land, making it representative for land surveying and change monitoring. Four phases of multi-temporal remote sensing images covering the same area were used as input data: March 15, 2024; June 20, 2024; September 18, 2024; and December 5, 2024. The spatial resolution of each image was 0.5 meters. After a unified projection coordinate transformation of the images, a regular grid was used to divide the area into operational units, with each grid having a side length of 200 meters. This resulted in 315 operational units within the experimental zone. Figure 4 The pattern is consistent with the regular grid.

[0067] This embodiment uses work unit numbered G142 as an example for illustration. Figure 4 The spatial location of work unit G142 within the overall grid is marked with a thick outline. This work unit covers a section of urban side road and two paved plazas, with sparse vegetation and bare land in the surrounding area, and a relatively high proportion of artificial structures, providing conditions for constructing stable anchor points.

[0068] First, anchor point candidate objects were extracted from the G142 work unit in each remote sensing image. Candidate objects were prioritized for hard targets that were structurally stable and reproducibly identifiable across multiple temporal phases, including road intersections, curb turning points, bridge and culvert entrances / exits, corners of regular plazas, and corners of large buildings. After automatic extraction and manual verification, 38 candidate objects were obtained from the March 15, 2024 image; 41 candidate objects were obtained from the June 20, 2024 image; 39 candidate objects were obtained from the September 18, 2024 image; and 36 candidate objects were obtained from the December 5, 2024 image. Figure 4 The spatial distribution of candidate objects in one period is illustrated in scatter plot form.

[0069] In this embodiment, the preset time phase number is set to 3, meaning that candidate objects must be repeatedly located in at least 3 out of 4 image periods before proceeding to the next step. After performing cross-time matching positioning on 38 to 41 candidate objects, a total of 24 candidate objects met the repeatable positioning condition, including 7 road intersections, 9 square corners, and 8 building corners. The remaining candidate objects were eliminated. The eliminated objects were mainly due to unclear corners caused by vegetation obstruction or changes in structural form caused by temporary construction enclosures. Taking one building corner candidate object as an example, this corner point was clearly visible in the March and June imagery, but was covered by temporary scaffolding in the September and December imagery. The repeatable positioning only met the requirement of 2 periods, which did not meet the preset time phase number requirement, and therefore it was eliminated.

[0070] After completing the repeatable location screening, candidate anchor point groups are formed within the same work unit, consisting of no fewer than two candidate objects, to test the cross-temporal stability of geometric relationships. In this embodiment, each group contains two or three candidate objects, preferentially combined as road intersections and plaza corners or building corners to improve geometric constraint strength. Taking the two-point anchor point group consisting of candidate objects P1 and P2 as an example, P1 is a road intersection, and P2 is the northeast corner of a plaza. The spatial distance between the two points in each period of imagery is calculated: 118.40 meters in March, 118.46 meters in June, 118.39 meters in September, and 118.45 meters in December. Using March as the baseline, the distance ratios are calculated as follows: 1.0005 in June, 0.9999 in September, and 1.0004 in December. The preset tolerance is set to 0.002, meaning that if the absolute value of the distance ratio deviating from 1 does not exceed 0.002, the requirement is met. The deviations of the ratios of the three periods of the anchor point group were 0.0005, 0.0001 and 0.0004, respectively, all of which were less than the preset tolerance. Therefore, the anchor point group was deemed to have passed, and P1 and P2 were identified as candidates for stable anchor points.

[0071] Taking the three-point anchor group P3, P4, and P5 as an example, P3 is the southwest corner of the building, P4 is the northwest corner of the building, and P5 is the southwest corner of the plaza. The angles formed by P3P4 and P3P5 are calculated respectively. The angles in March are 62.15 degrees, in June 62.22 degrees, in September 62.18 degrees, and in December 62.16 degrees. The preset angle tolerance is 0.50 degrees, meaning that an angle variation not exceeding 0.50 degrees is considered to meet the stability requirements. The changes in the angles of this group relative to March in the three periods are 0.07 degrees, 0.03 degrees, and 0.01 degrees, respectively, all within the tolerance range, thus passing the judgment. Therefore, P3, P4, and P5 are included in the anchor group as stable anchor points. By combining the distance ratio and angle relationship as two criteria, the long-term geometric stability of the anchor points can be quantitatively verified without relying on complex models.

[0072] To verify the effectiveness of the removal mechanism, this embodiment presents a candidate anchor point group that failed the tolerance judgment. Candidate objects Q1 and Q2 are selected to form a two-point anchor point group. Q1 is a corner point of a small temporary parking lot, and Q2 is the boundary inflection point of an adjacent bare land. The distance between the two points is 96.10 meters in March, 96.08 meters in June, 97.62 meters in September, and 97.58 meters in December. Based on March, the distance ratio in September is 1.0158, and in December it is 1.0153, a deviation significantly greater than 0.002. Image verification revealed that the bare land was leveled and hardened into a new parking area starting in September, resulting in a real displacement of the boundary inflection point, which is a geometric relationship disruption caused by changes in land features. According to the rules of this invention, this anchor point group lacks stability and should be removed to avoid mistaking real-change targets as anchor points, thus affecting subsequent consistency verification.

[0073] After completing the above steps, the G142 work unit finally determined a set of 18 anchor points, including 6 road intersections, 7 plaza corners, and 5 building corners. This set of anchor points all met the requirement of at least three repeated positioning phases and passed the preset tolerance test for distance ratios or angle relationships. Figure 4 The final anchor points are identified by different shapes. When this set of anchor points is used for subsequent consistency verification of multi-temporal images, stable geometric relationships can be used as a reference to effectively filter out temporal phases that are obstructed, affected by construction, or have large local registration errors.

[0074] After anchor point screening, candidate reference phases were determined for the available phase sets within work unit G142. Based on preliminary image quality assessment and anchor point pass statistics, the number of anchor points passed for the images from March 15, 2024, and June 20, 2024, within this work unit was 18 each, with a pass rate of 1.00. The image from September 18, 2024, had 15 anchor points passed due to construction obstructions in some areas, with a pass rate of 0.83. The image from December 5, 2024, was affected by a low solar altitude angle, resulting in heavy local shadows; the number of anchor points passed was 16, with a pass rate of 0.89. According to the preset reliability judgment rules, after ranking the four phases of images, the March image ranked first, the June image ranked second, and the September and December images ranked lower. Therefore, the March 15, 2024 image was tentatively selected as the candidate reference phase.

[0075] During the consistency review phase, using the image from March 15, 2024 as the baseline, anchor point consistency was assessed sequentially for the June, September, and December imagery. Taking the March-to-June assessment as an example, the relative displacement of each of the 18 anchor points was calculated. The results showed that the relative displacement amplitudes of 16 anchor points were concentrated in the range of 0.12 meters to 0.28 meters, indicating good directional consistency. The remaining two anchor points showed deviations due to local shading, but these deviations did not exceed the preset consistency criteria, and the overall assessment passed. Subsequently, using the June imagery as the baseline, a reverse consistency assessment was performed on the March imagery. The resulting relative displacement distribution was highly consistent with the previous assessment, and the assessment also passed. Figure 5 The corresponding position in the middle shows that both the forward and reverse directions have passed.

[0076] Next, using the March imagery as a baseline, a consistency assessment was performed on the September imagery. The results showed that 5 out of 18 anchor points had relative displacement amplitudes exceeding 0.80 meters, significantly deviating from the dominant amplitude range, and therefore failed the assessment. However, when a reverse assessment was performed on the March imagery using the September imagery as a baseline, only 2 anchor points were identified as abnormal, with the remaining anchor points concentrated within a reasonable range, and the reverse assessment passed. Due to the inconsistency between the two assessments, the September imagery was designated as the subject of arbitration.

[0077] The processing of December imagery was similar. When using March as the baseline, four anchor points in the December imagery were deemed to have excessive relative displacement due to road shadow elongation, and thus failed the assessment. However, when the March imagery was assessed in reverse using December as the baseline, only one anchor point showed an anomaly, while the remaining anchor points met the consistency criteria. The two assessment results were inconsistent, therefore the December imagery was also identified as subject to arbitration.

[0078] For the two images to be arbitrated in September and December, a third candidate image, different from the March image and its corresponding image to be arbitrated, was selected from the candidate image set of the reference period as the arbitration image. Based on the aforementioned ranking results, the image of June 20, 2024, was selected as the arbitration image. First, using the June image as a benchmark, anchor point consistency was assessed between the March and September images. The results showed that all 18 anchor points in the March image met the consistency criteria, while 4 anchor points in the September image significantly deviated from the dominant amplitude range, and the arbitration decision failed. Subsequently, using the June image as a benchmark, consistency was assessed between the March and December images. The March image passed the consistency assessment, while only 2 anchor points in the December image deviated from the dominant amplitude range, and the overall consistency criteria were met. Figure 5 The results of the positive decision, negative decision, and arbitration decision are summarized and displayed in a light-colored matrix.

[0079] Based on the arbitration results, a final decision was made regarding the September imagery, removing it from the available timeframe set of work unit G142. For the December imagery, while the arbitration decision was successful, a one-way inconsistency was found during the bidirectional consistency review with the March imagery. Therefore, the December imagery was marked as a low-confidence timeframe, but it was retained for subsequent sorting and change detection; however, its weight was reduced in the reference period selection and change determination. Ultimately, work unit G142 formed a stable set of available timeframes, with the March imagery serving as the primary reference period, the June imagery as a high-confidence available timeframe, the December imagery as a low-confidence available timeframe, and the September imagery being removed. This result is highly consistent with the conclusions of manual image interpretation.

[0080] Continuing with work unit G142, the image dated March 15, 2024 was selected as a temporary reference, the image dated June 20, 2024 as an auxiliary reference, the image dated December 5, 2024 as a low-confidence available phase for change detection, and the image dated September 18, 2024 was no longer used for subsequent processing.

[0081] During the change detection phase, the March imagery was first used as a temporary reference and compared with the June and December images. Candidate change areas were extracted within the G142 work unit through a combined assessment of spectral and structural changes. In the comparison between the March and December images, a suspected change area was identified in the southeast corner of the work unit, preliminarily determined to be a transition from bare land to hardened surface. This area was designated as Region A in the detection results using the March imagery as a reference, with a spatial range corresponding to 1250 pixels.

[0082] Subsequently, using the June imagery as an auxiliary reference, a comparative analysis was conducted with the December imagery. Corresponding candidate change regions were detected at the same locations. This region was designated as Region B in the detection results using the June imagery as a reference, with a spatial range corresponding to 1320 pixels. Overlay analysis revealed subtle differences between Region A and Region B at their spatial boundaries, primarily caused by shadow effects and differences in segmentation accuracy.

[0083] To quantitatively assess the consistency of changes under two reference conditions, the spatial overlap between region A and region B is calculated using the following formula:

[0084]

[0085] in, This represents the number of pixels in the intersection of region A and region B. This represents the number of pixels in the union of region A and region B. Statistically, the intersection of regions A and B has 1100 pixels, and the union has 1470 pixels. Substituting these values ​​into the formula yields:

[0086]

[0087] In this embodiment, the preset spatial overlap threshold is 0.65. When the calculated (\Omega) value is 0.748, which is higher than the preset threshold, it is determined that the candidate change region has high consistency under the temporary reference and auxiliary reference conditions, and can be identified as a confirmed change region. Figure 6 The spatial overlap distribution and threshold determination results of each candidate change region are displayed in scatter plot form.

[0088] During the change results output phase, the changed area was output as a confirmed change area, along with reference period information and low-reliability reference markers, to highlight the importance of boundary accuracy in subsequent land surveying and mapping result verification. Manual review revealed that ground hardening construction was completed in this area between October and November 2024, consistent with the remote sensing change detection results.

[0089] Meanwhile, a small-scale suspected change area was detected north of the G142 work unit. This area formed region A (260 pixels) in the comparison between March and December, and region B (190 pixels) in the comparison between June and December. The two regions intersect with 95 pixels and have a union of 355 pixels. The spatial overlap is calculated as follows:

[0090]

[0091] The result was significantly lower than the preset threshold of 0.65, therefore the area of ​​change was identified as an area to be verified, and a low-reliability reference marker was output. Subsequent manual inspection confirmed that the change in this area was mainly caused by the difference between the extension of winter shadows and the surface moisture reflection, and did not belong to the actual land cover change.

[0092] Example 2:

[0093] After completing the anchor point set construction and determining the candidate reference phases, this embodiment continues to use work unit G142 as an example to illustrate the anchor point consistency determination process, in order to verify the feasibility and effectiveness of this determination mechanism in multi-temporal remote sensing change detection. Anchor point consistency determination is based on the spatial stability of anchor points in different temporal images. Through relative displacement statistics and a multi-anchor point voting mechanism, it avoids individual abnormal anchor points from interfering with the overall determination result. The statistical results and determination process are as follows: Figure 7 As shown.

[0094] In the specific implementation process, the image from March 15, 2024, was used as a candidate reference phase, and the image from June 20, 2024, was selected as another candidate phase to be determined. First, the 18 anchor points within the G142 work unit were spatially located one by one in the two image periods, obtaining the coordinate positions of each anchor point in the two image periods. Then, using the anchor point coordinates in the reference phase image as a benchmark, the relative planar displacement amplitude and direction of the corresponding anchor point in the June image were calculated. For example... Figure 7 The polar coordinate statistics show that the relative displacement amplitude of 16 out of the 18 anchor points is concentrated in the range of 0.12 meters to 0.28 meters, and the displacement direction is consistent, mainly concentrated in the northeast direction. The directional deviation is controlled within 15 degrees, reflecting that the overall registration relationship between the two images in this operation unit is relatively stable.

[0095] At the same time, such as Figure 7 As indicated by the cross marks, two other anchor points showed relative displacement amplitudes of 0.46 meters and 0.51 meters respectively, significantly higher than the other anchor points. Image verification confirmed that these two anchor points were located at the edge of the road shadow, and their local areas were significantly affected by changes in lighting, leading to decreased anchor point positioning accuracy, and not actual spatial displacement of the ground features.

[0096] To avoid misleading the overall consistency determination by individual anchor points, this embodiment does not directly rely on the maximum or minimum displacement values ​​for determination. Instead, it introduces a multi-anchor-point voting mechanism to determine preset consistency conditions. Specifically, within the same work unit, the dominant direction and dominant amplitude range of the relative displacement of 18 anchor points are statistically analyzed, and anchor points falling within the dominant range and meeting the direction conditions are included in the set of valid voting anchor points. Figure 7The statistical results show that there are 16 anchor points that meet the dominant amplitude range of 0.12 meters to 0.28 meters and the directional deviation does not exceed 15 degrees, accounting for 0.89% of the total number of anchor points, which is higher than the preset proportion threshold of 0.70.

[0097] In this example, the 16 valid anchor points are dispersed across the work unit's spatial area, located at various locations such as road intersections, plaza corners, and building corners, and are not concentrated in any particular local area. Therefore, according to the preset rules, the anchor point consistency determination for work unit G142 between the images from March 15, 2024, and June 20, 2024, is passed. For the two anchor points whose relative displacement deviates from the dominant amplitude range, they are marked as abnormal anchor points and will not participate in the voting calculation in this consistency determination, but they will not be immediately removed from the anchor point set to retain the possibility of re-participating in subsequent temporal phase determinations.

[0098] In the process of determining the consistency between images from March 15, 2024 and June 20, 2024, such as Figure 8 As shown, among the 18 anchor points, two anchor points have relative displacement amplitudes of 0.46 meters and 0.51 meters, respectively, which significantly deviate from the dominant amplitude range of 0.12 meters to 0.28 meters. However, the relative displacements of their neighboring anchor points all meet the consistency condition. Spatial neighborhood analysis reveals that there are 3 to 4 neighboring anchor points around these two anchor points, and their relative displacement amplitudes are concentrated within the dominant range with good directional consistency. This indicates that the anomaly is more likely caused by shadows or local reflection differences, rather than actual geometric changes.

[0099] According to the hierarchical processing rules of the present invention, the above two anchor points are marked as temporary abnormal anchor points and removed from the voting statistics only in this consensus determination. The remaining 16 anchor points continue to participate in the consensus voting and complete the overall determination.

[0100] Subsequently, in the consistency assessment of the images from March 15, 2024 and December 5, 2024, relative displacement calculations were performed again for the same batch of anchor points. For example... Figure 8 As shown in the mid-anchor point evolution curve, among the two temporary abnormal anchor points, the relative displacement amplitude of one anchor point dropped to 0.29 meters, approaching the dominant amplitude range and showing a recovery trend. Therefore, the abnormality marker was removed, and it resumed participation in the consistency assessment. The relative displacement amplitude of the other anchor point, however, remained at 0.78 meters, with a significant deviation in direction. Neighborhood verification revealed that this anchor point is located at the boundary between the road shadow and the building facade, and its local image morphology is continuously changing due to the influence of a low solar altitude angle.

[0101] In the subsequent consistency assessment of images from June 20, 2024, and December 5, 2024, the anchor point that still exhibited anomalies underwent further relative displacement analysis. Its relative displacement amplitude was 0.81 meters, still significantly deviating from the dominant amplitude range. Therefore, this anchor point was marked as a temporary abnormal anchor point in three consecutive simultaneous relative comparisons, reaching the preset threshold of three occurrences. According to the hierarchical processing rules of this invention, this anchor point was determined as a permanent abnormal anchor point and removed from all subsequent consistency reviews and voting statistics, no longer serving as a reference for geometric stability.

[0102] Based on the above-mentioned abnormal anchor point classification results, this embodiment further combines... Figure 9 This paper provides examples illustrating the spatial neighborhood determination method of adjacent anchor points, the occlusion screening mechanism, and the process of determining permanent abnormal anchor points under cross-temporal constraints.

[0103] In the specific implementation process, the spatial neighborhood of adjacent anchor points is first determined. For example... Figure 9 As shown, a spatial neighborhood with a radius of 25 meters is defined within the work unit, centered on each anchor point. Other anchor points falling within this neighborhood are considered neighboring anchor points. Within work unit G142, the number of neighboring anchor points for a single anchor point is typically 3 to 6, a range that effectively reflects the stability of the local geometry surrounding the anchor point. This neighborhood scale matches the road width and plaza scale within the work unit, effectively covering local structural features without introducing cross-regional interference.

[0104] In the comparison of images from March 15, 2024 and June 20, 2024, further occlusion screening was performed on the aforementioned anchor points that had been marked as temporary anomalies. For example... Figure 9 As shown, by checking the grayscale consistency and texture integrity of the local image area surrounding the anchor point, it was found that the anchor point had significant shadow coverage in a local area of ​​the June image, with its grayscale value decreasing by approximately 18 grayscale levels compared to the reference period image, resulting in reduced edge contrast and decreased anchor point positioning accuracy. Meanwhile, the four neighboring anchor points did not exhibit similar shadows or occlusion, and their relative displacements all fell within the dominant amplitude range. Therefore, according to the rules of this invention, this anchor point was marked as a temporary abnormal anchor point in this determination and removed from this consistency vote, but this does not affect its eligibility for re-determination in subsequent time phases.

[0105] In a subsequent comparison of images from March 15, 2024, and September 18, 2024, construction barriers appeared in the area where the anchor point was located, obscuring the identifiable structural outline in some local images. The relative displacement amplitude of the anchor point reached 0.92 meters, significantly deviating from the dominant amplitude range. Meanwhile, three neighboring anchor points remained stable, while one, affected by the same obstruction, was simultaneously marked as a temporary anomaly. Due to the clear obstruction and the fact that the neighboring anchor points generally met the consistency criteria, this anchor point will continue to be marked as a temporary anomaly.

[0106] In a comparison of images from March 15, 2024, and December 5, 2024, the anchor point was screened for occlusion again. The results showed that the construction barriers in the area had been removed, the local shadow area had significantly decreased, the anchor point's edge structure had reappeared, and its relative displacement amplitude had dropped to 0.27 meters, a significant decrease and approaching the upper limit of the dominant amplitude range, meeting the trend condition for resuming participation in the consistency determination. Furthermore, its relative displacement direction was consistent with that of the five neighboring anchor points. Under these conditions, the occlusion indication was determined to have disappeared. According to the rules of this invention, the temporary anomaly marker on the anchor point was removed, and its participation in the consistency determination and multi-anchor point voting was restored.

[0107] On the other hand, for another anchor point that had been repeatedly marked as a temporary anomaly in the previous stage of Example 2, a cross-temporal constraint was introduced to determine the permanent anomaly anchor point. This anchor point was marked as a temporary anomaly due to shadowing in the image comparison between March 15, 2024 and June 20, 2024; it was again marked as a temporary anomaly due to construction obstruction in the image comparison between March 15, 2024 and September 18, 2024; and it was again marked as a temporary anomaly due to long shadows caused by low solar altitude angles in the image comparison between June 20, 2024 and December 5, 2024. These three time-phase comparisons covered different seasons such as spring, autumn, and winter, and also covered different imaging conditions such as high and low solar altitude angles, satisfying the preset cross-temporal constraint conditions.

[0108] Since this anchor point was marked as a temporary anomalous anchor point in at least three time-relative comparisons under different seasons or imaging conditions, and its relative displacement deviated from the dominant amplitude range for a long period, it was determined that it lacked long-term geometric stability. According to the rules of this invention, this anchor point was ultimately determined as a permanent anomalous anchor point and was completely removed from all subsequent consistency reviews, anchor point voting, and reference period determination processes, and would no longer participate in any geometric consistency analysis.

[0109] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An optimized method for multi-temporal remote sensing change detection in land surveying, characterized in that... include: Acquire multi-temporal remote sensing images of the same area and divide them into patches or regular grid operation units; Select ground objects that can be repeatedly identified in multi-temporal images and meet preset stability conditions as anchor points, and perform consistency verification based on anchor points for each temporal image. Temporal images that fail the verification are not included in the available temporal set of this operation unit. Within the set of available time phases, candidate reference time phases are determined, evaluated and sorted according to preset reliability judgment rules, and the first-ranked one is selected as the reference time phase. When none of the candidate time phases meet the preset reliability conditions, the candidate time phase with the maximum number of anchor point passes is selected as the temporary reference. If they are tied, the one with the maximum anchor point pass rate is selected and marked as low reliability. The change area is extracted by comparing the reference time phase with the other available time phases excluding the reference time phase, and the change area, reference time phase identifier, and reliability marker are output.

2. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 1, characterized in that... The selection of anchor points includes extracting anchor point candidate objects from each temporal image, retaining only candidate objects that can be repeatedly located in no less than a preset number of temporal images; within the same work unit, no less than two candidate objects constitute a candidate anchor point group, and when the mutual distance ratio or included angle relationship of the anchor point group in different temporal images meets the preset tolerance, it is determined as an anchor point, otherwise it is discarded.

3. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 1, characterized in that... The consistency review includes, for the reference period candidate time phase and any other candidate time phase, performing anchor point consistency determination on the other candidate time phase based on the reference period candidate time phase, and performing anchor point consistency determination on the reference period candidate time phase based on the other candidate time phase; when the two determination results are inconsistent or only one determination passes, the other candidate time phase is determined as an object to be arbitrated, and after arbitration, it is decided to remove it or mark it as low confidence for subsequent sorting.

4. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 1, characterized in that... When the work unit is marked as a low reliability reference, the candidate time phase ranked first is selected as a temporary reference, and the candidate time phase ranked second and passing the consistency review is selected as an auxiliary reference; the candidate change areas are obtained by comparing them with the remaining available time phases respectively. When the same change area is extracted under both references and the spatial overlap is not lower than the preset threshold, the output is a confirmed change; otherwise, the output is a change to be checked and a low reliability reference mark is attached.

5. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 3, characterized in that... The anchor point consistency determination includes: locating the same anchor point in the candidate reference phase and the other candidate phases and obtaining its relative displacement; when the relative displacements of multiple anchor points in the same work unit meet the preset consistency conditions, the determination is passed; anchor points whose relative displacements deviate from the majority of anchor points are recorded as abnormal anchor points and do not participate in the determination.

6. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 3, characterized in that... When the two determination results are inconsistent or only one determination passes, a third candidate time phase that is different from the candidate time phase of the reference period and the other candidate time phases is selected from the candidate time phase set of the reference period as the arbitration time phase, and the anchor point consistency determination is performed on the candidate time phase of the reference period and the other candidate time phases based on the arbitration time phase. Candidate phases that fail the arbitration decision are removed, while candidate phases that pass the arbitration decision but have one-way inconsistencies are marked as low confidence for subsequent ranking.

7. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 5, characterized in that... The preset consistency condition is determined by multi-anchor point voting: within the same work unit, the dominant direction and dominant amplitude range of the relative displacement of each anchor point are statistically analyzed. When the number of anchor points falling into the dominant amplitude range is not less than a preset proportion and the anchor points are dispersed within the work unit, the anchor point consistency determination is passed.

8. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 5, characterized in that... The determination of abnormal anchor points is carried out in a hierarchical manner: when the relative displacement of an anchor point deviates from the dominant amplitude range but its neighboring anchor points meet the consistency condition, the anchor point is marked as a temporary abnormal anchor point and is removed only in this determination; when the anchor point is marked as a temporary abnormal anchor point in at least a preset number of time-to-time comparisons, it is determined as a permanent abnormal anchor point and removed from subsequent consistency reviews.

9. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 8, characterized in that... The neighboring anchor points are determined according to the spatial neighborhood within the work unit, and the determination of temporary abnormal anchor points includes occlusion screening: when the local area of ​​the anchor point shows signs of cloud, shadow or occlusion in the candidate phase of the reference period or other candidate phases, the anchor point is marked as a temporary abnormal anchor point. When the occlusion signs disappear and the relative displacement of the anchor point returns to the dominant amplitude range, the temporary anomaly marker is removed and participation in the consistency determination is resumed.

10. The multi-temporal remote sensing change detection optimization method for land surveying according to claim 8, characterized in that... The permanent anomalous anchor point is determined only when the cross-temporal constraint is met: when the anchor point is marked as a temporary anomalous anchor point in at least a preset number of time-phase comparisons under different seasons or different imaging conditions, it is determined as a permanent anomalous anchor point.

Images