Land utilization change monitoring model based on live-action three dimensions

Abstract

The invention provides a live-action three-dimensional land utilization change monitoring model, and the model comprises a data collection module which is used for obtaining a multi-angle image of a target region; the three-dimensional modeling module is used for generating a multi-temporal high-precision point cloud and texture network model; the point cloud registration module is used for carrying out automatic registration on the multi-temporal high-precision point cloud so as to obtain the surface morphological structure change of the target area; the semantic segmentation module is used for carrying out ground feature semantic segmentation on the texture network model and labeling sub-regions of the target region; and the change monitoring module is used for comparing the surface structure and the land utilization type of the target area at different time points based on the three-dimensional models of the plurality of patrol flight periods and the semantic segmentation result, and generating a visual change layer and a monitoring report. The three-dimensional characteristics of land utilization and the actual situation under the complex terrain condition can be accurately reflected, and the monitoring precision and reliability of the land utilization change are improved.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional scene reconstruction and real-time monitoring technology, and in particular to a land use change monitoring model based on real-world three-dimensional scenes. Background Technology

[0002] Against the backdrop of rapid technological advancements and increasing demands for intelligent applications, 3D scene reconstruction and real-time positioning technologies have become hot topics in research and industry. With breakthroughs in machine learning, artificial intelligence, and computer vision, more accurate and efficient scene understanding capabilities are crucial for various automated systems. Traditional 3D scene reconstruction methods often rely on complex hardware systems and limited sensor data processing capabilities, which restricts their application scope and processing efficiency.

[0003] In the field of land resource management, accurate and real-time monitoring of land use is crucial for rational land planning, ensuring sustainable resource utilization, and protecting the ecological environment. Currently, monitoring of land use in the canal area mainly relies on a combination of remote sensing data interpretation, Geographic Information System (GIS) analysis, and limited ground inspections. Medium- to high-resolution remote sensing imagery can provide a relatively intuitive view of changes in land use patterns along the canal, such as the expansion of riverbanks, reduction of green space, and increased building density. However, existing monitoring systems primarily focus on the two-dimensional planar distribution of land use, and are relatively inadequate in reflecting land use intensity, spatial utilization efficiency, and the degree of three-dimensional development. Especially in densely developed urban sections along the canal shoreline, monitoring results often fail to reveal the actual height of buildings, the layers of three-dimensional development, and the specific details of the overlapping use of ground space, leading to deviations in land use status assessments from reality. Some areas suffer from poor data timeliness and long update cycles, making it difficult to support dynamic monitoring needs in a rapidly changing environment. Furthermore, the canal basin spans multiple administrative divisions, with inconsistent data source standards and a lack of information sharing mechanisms, further weakening the systematic and coherent nature of monitoring work.

[0004] Overall, the current monitoring of land use status in canals is still in a development stage that mainly relies on "static identification" and lacks "dynamic perception" capabilities. Its spatial accuracy, three-dimensional identification capabilities, and timeliness response level all need to be improved. Summary of the Invention

[0005] The purpose of this application is to provide a land use change monitoring model based on real-scene 3D, which solves the problems of insufficient spatial accuracy, inadequate stereo recognition capability and timeliness response level of traditional land use monitoring models in the prior art when monitoring land use.

[0006] To achieve the above objectives:

[0007] In a first aspect, embodiments of the present invention provide a land use change monitoring model based on real-scene 3D, the model comprising a data acquisition module, a 3D modeling module, a point cloud registration module, a semantic segmentation module, and a change monitoring module; wherein;

[0008] The data acquisition module is used to periodically patrol the target area using a drone equipped with a tilting camera in order to obtain multi-angle images of the target area.

[0009] The 3D modeling module is used to generate multi-temporal high-precision point cloud and texture network models based on the multi-angle images;

[0010] The point cloud registration module is used to automatically register the multi-temporal high-precision point cloud to obtain the changes in the surface morphology and structure of the target area.

[0011] The semantic segmentation module is used to perform semantic segmentation of ground features on the texture network model and to label sub-regions of the target region.

[0012] The change monitoring module is used to compare the surface structure and land use type of the target area at different time points based on the three-dimensional model and semantic segmentation results of multiple flight cycles, and generate a visual change layer and monitoring report.

[0013] In one embodiment, after the data acquisition module periodically patrols the target area using a drone equipped with a tilting camera to acquire multi-angle images of the target area, the method further includes:

[0014] Based on the images acquired by the UAV, a time-series dataset of the target area is established to continuously monitor the state changes of the target area at different time points.

[0015] In one embodiment, when automatically registering the multi-temporal high-precision point cloud using the point cloud registration module to obtain the changes in the landform structure of the target area, the process specifically includes:

[0016] Automatic registration of the multi-temporal high-precision point cloud is performed using a coarse registration method based on morphologically invariant points; and / or

[0017] The multi-temporal high-precision point cloud is automatically registered using a multi-temporal point cloud fine registration method based on morphologically invariant points.

[0018] In one embodiment, the multi-temporal point cloud coarse registration method based on morphologically invariant points automatically registers the multi-temporal high-precision point cloud, including:

[0019] Generate a random spatial plane for the reference point cloud, and determine four coplanar points that satisfy the preset point cloud registration conditions based on the random spatial plane;

[0020] Based on the four coplanar points, determine the four pairs of congruent points between the reference point cloud and the point cloud to be registered;

[0021] Extract the local features of the congruent four-point pair in the target space neighborhood; the local features include the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered;

[0022] The similarity between the local features of the reference point cloud and the local features of the point cloud to be registered is determined, and the coarse registration point cloud result of the reference point cloud is determined based on the similarity.

[0023] In one embodiment, determining the similarity between the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered, and determining the coarse point cloud to be registered based on the similarity, specifically includes:

[0024] The similarity between the local features of the reference point cloud and the local features of the point cloud to be registered is determined based on a preset Euclidean distance comparison function; the local features of the reference point cloud include multiple reference feature vectors, and the local features of the point cloud to be registered include multiple feature vectors to be registered.

[0025] When the similarity is less than a preset similarity threshold, the corresponding reference feature vector is determined in the neighborhood local features of the point cloud to be registered, and the set of corresponding feature vectors to be registered is used as the point cloud result of the coarse registration.

[0026] In one embodiment, the multi-temporal point cloud fine registration method based on morphologically invariant points automatically registers the multi-temporal high-precision point cloud, including:

[0027] Based on coarse registration of morphologically invariant points, the morphologically invariant regions of the changing scene are determined;

[0028] The morphologically invariant region is registered using the iterative nearest point method, and the distance between the first and second point sets of the morphologically invariant region is optimized based on a preset optimization formula to obtain a finely registered point cloud result.

[0029] In one embodiment, the expression for the optimization formula is:

[0030]

[0031] R i and T i These represent the morphologically invariant regions Q to be registered. TIR Registered to the reference morphology invariant region PTIR The optimal rotation and translation parameters, #k represents P TIR and Q TIR The smaller value among the points, Δ i This is the cumulative sum of the minimum Euclidean distances between point sets calculated using the optimal rotation and translation in the i-th algorithm iteration.

[0032] In one embodiment, the step of performing geographic feature semantic segmentation on the texture network model using a semantic segmentation module and labeling sub-regions of the target region specifically includes:

[0033] Using the texture network model and the labels associated with the texture network model as input, a pre-defined fully convolutional neural network is used to concatenate the texture network model and the labels associated with the texture network model to obtain an annotated texture network model, and the annotated texture network model is output to complete sub-region segmentation and annotation.

[0034] In one embodiment, the change monitoring module compares the surface structure and land use type of the target area at different time points based on the three-dimensional model and semantic segmentation results of multiple flight cycles, and generates a visualized change layer and monitoring report, specifically including:

[0035] The three-dimensional model and semantic segmentation results of the target area obtained from the multiple patrol cycles are compared in time to extract information on changes in landform and land cover types in the target area.

[0036] Based on the information on changes in landform and land cover types, determine the conversion status of land use types in the target area;

[0037] Based on the transformation results, a corresponding visualization layer and detection report are generated.

[0038] Secondly, the present invention provides a land use change monitoring system based on real-scene 3D, comprising:

[0039] A processor is configured with the land use monitoring model described in any one of the first aspects; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to control the land use monitoring model to complete land use monitoring.

[0040] This invention provides a land use change monitoring model based on real-scene 3D. The model includes a data acquisition module, a 3D modeling module, a point cloud registration module, a semantic segmentation module, and a change monitoring module. The data acquisition module uses a drone equipped with an oblique camera to periodically patrol the target area to acquire multi-angle images of the target area. The 3D modeling module generates multi-temporal high-precision point clouds and texture network models based on the multi-angle images. The point cloud registration module automatically registers the multi-temporal high-precision point clouds to obtain changes in the landform structure of the target area. The semantic segmentation module performs semantic segmentation of the texture network model and labels sub-regions of the target area. The change monitoring module compares the land structure and land use types of the target area at different time points based on the 3D model and semantic segmentation results from multiple patrol cycles, and generates a visualized change layer and a monitoring report. This application introduces high-precision 3D mapping, UAV patrol, and semantic segmentation technologies to improve the model's spatial accuracy, stereo recognition capability, and timeliness response level for land use. It can more accurately reflect the intensity of land use, spatial utilization efficiency, and degree of three-dimensional development. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 The module intent of the land use change monitoring model based on real-scene 3D provided in the embodiments of the present invention. Detailed Implementation

[0043] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0044] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0045] See Figure 1 This is a schematic diagram of the module composition of a real-scene 3D land use change monitoring model provided in an embodiment of this application. The real-scene 3D land use change monitoring model includes: a data acquisition module, a 3D modeling module, a point cloud registration module, a semantic segmentation module, and a change monitoring module. The data acquisition module is used to periodically patrol the target area using a drone equipped with an oblique camera to acquire multi-angle images of the target area. The 3D modeling module is used to generate multi-temporal high-precision point clouds and texture network models based on the multi-angle images. The point cloud registration module is used to automatically register the multi-temporal high-precision point clouds to obtain changes in the landform structure of the target area. The semantic segmentation module is used to perform semantic segmentation of the texture network model and label sub-regions of the target area. The change monitoring module is used to compare the landform structure and land use type of the target area at different time points based on the 3D model and semantic segmentation results from multiple patrol cycles, and generate a visualized change layer and a monitoring report.

[0046] In this embodiment, multi-temporal 3D point cloud and texture models are used to not only identify land cover types but also extract their height, volume, and spatial information, such as the dynamics of three-dimensional development including new buildings, demolition projects, and terrain changes. Furthermore, through multi-temporal comparison, subtle changes in land use can be accurately captured, including reductions in green space, water body filling, and structural movement, providing dynamic monitoring capabilities. In addition, outputting change layers, data reports, and heat maps can visually present change trends and key problem areas, assisting in urban management, river protection, and the investigation and handling of illegal construction.

[0047] For example, taking the Xijin Reservoir section of the Pinglu Canal in Nanning as a typical application area, the land use change monitoring model based on real-scene 3D provided by the present invention is used to conduct multi-temporal monitoring of the area to verify the application effect of the present invention in actual land use supervision.

[0048] This embodiment selects a target area of ​​approximately 4.8 kilometers, covering typical waterfront greenbelts, historical district riverside buildings, bridge structures, and port facilities. The system first conducted two rounds of patrol flights over the area using a drone equipped with an oblique photography camera in June 2024 and January 2025, acquiring multi-angle, high-resolution image data. Subsequently, the 3D modeling module generates point cloud and texture network models for the corresponding time periods, and the point cloud registration module compares and analyzes terrain morphology changes; the semantic segmentation module completes the fine identification and spatial annotation of landform categories such as buildings, water bodies, green spaces, and transportation facilities.

[0049] Based on the multi-temporal data output by the above modules, the change monitoring module detects and quantifies the changes in surface structure and land use type in the target area between two temporal phases, identifying the following main changes:

[0050] ① The water area decreased from 162,400 square meters to 158,250 square meters, a reduction of approximately 4,150 square meters, which is preliminarily determined to be due to illegal filling of the river.

[0051] ② The waterfront green space decreased from 48,300 square meters to 43,950 square meters, a reduction of 4,350 square meters. The changed area is shown in the image as hard paving and building coverage.

[0052] ③ Seven new structures were added, including four steel-structured simple warehouses and three informal dock platforms, covering a total area of ​​approximately 1,280 square meters;

[0053] ④ The number of sections with a bridge clearance height of less than 4.5 meters has increased from 3 to 5, of which 2 sections have experienced a decrease in clearance due to the construction of new temporary structures under the bridge.

[0054] The monitoring system outputs corresponding change layers and reports, including: a highlighted map of the change area, a heat map of the change in airspace, statistical tables of the change area of ​​various land uses, and a semantic type conversion matrix and spatial overlay analysis layer.

[0055] After being reviewed by the local natural resources authorities, the above results were successfully used to support the demolition of illegal buildings, ecological restoration and planning rectification, significantly improving the dynamic monitoring capacity and management efficiency of land use along the canal.

[0056] In one embodiment, after the data acquisition module periodically patrols the target area using a drone equipped with a tilting camera to acquire multi-angle images of the target area, the method further includes: establishing a time-series dataset of the target area based on the images acquired by the drone, so as to continuously monitor the state changes of the target area at different time points based on the time-series dataset.

[0057] Understandably, drones equipped with oblique photogrammetry systems can periodically patrol a target area along a fixed route over multiple time periods, collecting high-resolution images from different perspectives and at different times. By archiving these images chronologically, a time-series remote sensing image dataset containing multi-dimensional attributes such as space, time, and perspective can be formed. Using the images collected at each time point, corresponding real-world 3D models (point clouds, textured meshes, etc.) can be generated, not only reconstructing the terrain but also capturing vertical information such as building height, structural outlines, and bridge clearance, providing a foundation for subsequent structural change identification. Spatially aligning 3D models or images from different periods and combining them with semantic segmentation and change detection algorithms, changes in the morphology, location, or attributes of land features can be automatically identified, thereby analyzing the dynamic process of land use.

[0058] In one specific implementation, the data acquisition module uses a drone equipped with an oblique photography camera to conduct multiple patrol flights over the target river section in June 2024, December 2024, and March 2025 to acquire multi-angle, high-resolution remote sensing images of the target area.

[0059] Based on the aforementioned imagery, the 3D modeling module constructs a real-world 3D model for each temporal phase, including high-precision point cloud data and a texture network model. The point cloud registration module performs spatial registration and alignment of the multi-temporal 3D models, ensuring that the data from each temporal phase can be compared and analyzed within the same spatial reference frame. Subsequently, the semantic segmentation module identifies and labels features in each temporal model, classifying typical elements such as water bodies, green spaces, transportation facilities, and buildings.

[0060] The change monitoring module performs differential analysis on data from different time phases, constructs a three-dimensional time-series dataset of the target area, and identifies land use changes. Specific monitoring results include:

[0061] ① The water body area decreased by approximately 4,150 square meters between June 2024 and March 2025, which was identified as unplanned river reclamation.

[0062] ② The area of ​​waterfront green space decreased by 4,350 square meters, and some areas were converted into land with hard paving or structures;

[0063] ③ Seven new structures were discovered, covering a total area of ​​approximately 1,280 square meters;

[0064] ④ Two temporary structures were added under the bridge, which reduced the clearance height and posed a safety hazard to shipping.

[0065] The monitoring results mentioned above are automatically generated by the system into change layers, change type annotation maps, area change statistical reports, and text-based monitoring reports, which can be used to support practical management work such as land law enforcement evidence collection, ecological space protection, and urban space remediation.

[0066] This embodiment verifies the model's ability to accurately identify land use changes in the canal area under continuous, multi-temporal conditions. It has significant advantages, especially in detecting three-dimensional development of waterfront spaces, shoreline erosion, and the addition of illegal buildings. It can effectively solve the problem that traditional two-dimensional remote sensing methods cannot reflect three-dimensional changes on the land surface.

[0067] In one embodiment, when the multi-temporal high-precision point cloud is automatically registered using the point cloud registration module to obtain the changes in the landform structure of the target area, the process specifically includes: automatically registering the multi-temporal high-precision point cloud using a coarse registration method based on morphologically invariant points; and / or automatically registering the multi-temporal high-precision point cloud using a fine registration method based on morphologically invariant points.

[0068] Specifically, the multi-temporal point cloud coarse registration method based on morphologically invariant points automatically registers the multi-temporal high-precision point cloud, including: generating a random spatial plane of the reference point cloud, and determining four coplanar points that satisfy preset point cloud registration conditions based on the random spatial plane; determining congruent four-point pairs between the reference point cloud and the point cloud to be registered based on the congruent four-point pairs; extracting local features of the congruent four-point pairs in the target spatial neighborhood; the local features include the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered; determining the similarity between the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered, and determining the point cloud result of the coarse registration of the reference point cloud based on the similarity.

[0069] Specifically, this embodiment uses affine invariance constraints to find four identical coplanar point pairs between the reference point cloud P and the point cloud Q to be registered:

[0070] First, using the 4PCS algorithm, three spatial points are randomly selected in the reference point cloud, where the distance between any two of these three points is greater than or equal to a preset minimum point interval L (usually 0.2 to 0.3 times the maximum coverage length of the reference point cloud). The random spatial plane formed by these three points is denoted as (p...). 1i ,p 2i ,p 3i ).

[0071] Then follow the formula: (O is a random space plane (p)1i ,p 2i ,p 3i Given any point in the array, where a, b, and c are constants and the sum of a, b, and c is 1, calculate the coordinates p of the fourth point that satisfies these conditions. 4i And form a set of four points B. i =(p 1i ,p 2i ,p 3i ,p 4i ), where the coordinates of the fourth point are p 4i With three other points in space, point p 1i p 2i p 3i The distances between them are all greater than or equal to the preset minimum point interval L. Then, find all four coplanar points B = {B1, B2, ..., B...} in the reference point cloud that satisfy the conditions. n}, where n is the total number of coplanar points that satisfy the condition.

[0072] Next, according to the constraint functions of distance invariant constraint and angle invariant constraint, find the set of approximately congruent four points in the point cloud to be registered. Repeat this step until the matching of all coplanar four points in B is completed, find the initial set of matching four point pairs S = {q1, q2, q3, q4}, and then select the combination in S with the largest point cloud coverage as the congruent four point pair J = {p1, p2, p3, p4}.

[0073] The expression for the constraint function is:

[0074]

[0075] Where e1 is the tolerance for the constant side length constraint, and its value ranges from 0.001 to 0.05; e2 is the tolerance for the angle constraint, and its value ranges from 0.01 to 0.1.

[0076] Further, determining the similarity between the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered, and determining the coarse point cloud to be registered based on the similarity, specifically includes: determining the similarity between the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered based on a preset Euclidean distance comparison function; the neighborhood local features of the reference point cloud include multiple reference feature vectors, and the neighborhood local features of the point cloud to be registered include multiple feature vectors to be registered; when the similarity is less than a preset similarity threshold, determining the feature vector to be registered corresponding to each of the reference feature vectors in the neighborhood local features of the point cloud to be registered, and using the set of corresponding feature vectors to be registered as the point cloud result of the coarse registration.

[0077] In this embodiment, a congruent four-point pair radius R-radius spatial neighborhood feature is constructed, and a point feature extraction algorithm is used to calculate the local features of all point clouds in the neighborhood to describe the distribution attribute information of the point clouds in the neighborhood; the feature similarity of the point clouds in the neighborhood is calculated, and the point clouds are registered based on the similarity and the distribution attribute information of the point clouds in the neighborhood.

[0078] Specifically, to construct the spatial neighborhood features of congruent four-point pairs with radius R, the first step is to search for the set of points J inside the neighborhood of the congruent four-point pairs with radius R. R ={P1 R P2 R P3 R P4 R} and S R ={q1 R ,q2 R ,q3 R ,q4 R}; Secondly, calculate the local features of all points in the neighborhood, that is, construct p∈J using a point feature extraction algorithm. R and q∈S R The point feature descriptor is used to describe the neighborhood distribution attribute information of a point and obtain the neighborhood local features J of the reference point cloud. RT ={P1 RT P2 RT P3 RT P4 RT} and the neighborhood local features S of the point cloud to be registered RT ={q1 RT ,q2 RT ,q3 RT ,q4 RT Next, spatial point feature matching and optimization are performed. First, feature similarity is calculated to complete the initial matching. Then, J is calculated using the Euclidean distance comparison function. RT and S RT The similarity of point cloud features is calculated. A smaller similarity value indicates a closer Euclidean distance between the two feature vectors, resulting in higher feature similarity. The ratio of the minimum to the second minimum similarity value is calculated and defined as the Euclidean distance ratio *r*. When *r* is less than 0.6, the point features are considered successfully matched. This process is iterated through J... RT All eigenvectors within S RT Find the initial matching point in the middle, and denote the matching result as E = {(e P 1, e Q 1), (e) P 2, e Q 2), ..., (e P tn e Q tn)}, where tn represents the total number of initial feature matches; considering the existence of mismatches, a random sampling consistency strategy is used to optimize the initial matching results and improve the feature matching accuracy. First, 3 pairs of points are randomly selected from tn point pairs as a sample, and the rigid transformation matrix T between the three points is calculated. Then, it is determined whether the remaining tn-3 point pairs are correct matches under the transformation matrix T: Points satisfying the trust region constraint are denoted as interior points; otherwise, they are denoted as exterior points. All interior points form the consensus set for this sampling. Then, the upper limit of the sampling count is updated based on the number of interior points. The point selection calculation and interior point judgment process are repeated until the number of iterations reaches the upper limit of the sampling count. At this point, the feature matching optimization process is complete. The interior point with the most iterations in the consensus set is taken as the feature matching optimization point pair, i.e., the morphologically invariant matching point pair C = {(C P 1 C Q 1 ), (C P 2 C Q 2 ), ..., (C P tN C Q tN ,)}, where tN is the total number of feature matches, and let J be the morphologically invariant points in the neighborhood of R for the four congruent points. TIP ={P1 TIP P2 TIP P3 TIP P4 TIP} and S TIP ={q1 TIP ,q2 TIP ,q3 TIP ,q4 TIP}

[0079] The expression for the Euclidean distance comparison function is:

[0080]

[0081] Where D represents the similarity. Let i be the i-th reference feature vector of the local features in the neighborhood of the reference point cloud. Let be the i-th feature vector to be registered, representing the local features of the neighborhood of the point cloud to be registered.

[0082] The functional expression for the trust region constraint is:

[0083]

[0084] in, For feature matching point pairs, δ represents the transformation constraint threshold.

[0085] Furthermore, the multi-temporal point cloud fine registration method based on morphologically invariant points automatically registers the multi-temporal high-precision point cloud, including:

[0086] Based on coarse registration of morphologically invariant points, the morphologically invariant regions of the changing scene are determined;

[0087] The morphologically invariant region is registered using the iterative nearest point method, and the distance between the first and second point sets of the morphologically invariant region is optimized based on a preset optimization formula to obtain a finely registered point cloud result.

[0088] Specifically, the expression for the optimization formula is:

[0089]

[0090] R i and T i These represent the morphologically invariant regions Q to be registered. TIR Registered to the reference morphology invariant region P TIR The optimal rotation and translation parameters, #k represents P TIR and Q TIR The smaller value among the points, Δ i This is the cumulative sum of the minimum Euclidean distances between point sets calculated using the optimal rotation and translation in the i-th algorithm iteration.

[0091] In one embodiment, the step of performing semantic segmentation of the texture network model using a semantic segmentation module and labeling the sub-regions of the target region specifically includes: taking the texture network model and the labels associated with the texture network model as input, processing the texture network model and the labels associated with the texture network model in series through a preset fully convolutional neural network to obtain a labeled texture network model, and outputting the labeled texture network model to complete the sub-region segmentation and labeling.

[0092] First, a texture network model of the target region is used as the basic input, and the label information associated with this model is obtained. This texture network model can be generated from remote sensing images, aerial photographs, or other geospatial information acquisition methods. The label information may include prior land cover category information or annotation results obtained through training. Then, the texture network model and its corresponding label information are fused together, specifically by concatenating them and feeding them as input into a pre-defined fully convolutional neural network (FCN). This FCN network structure may include multiple convolutional layers, activation layers, pooling layers, and upsampling modules to extract texture features and achieve the fusion and parsing of spatial semantic information. Under the processing of the neural network, an annotated texture network model is output, thus completing the land cover semantic segmentation operation. This annotation model contains the semantic recognition results of each sub-region in the target region. Each sub-region is assigned a semantic label indicating its land cover category, such as road, water body, building, vegetation, etc. Finally, the annotated texture network model is used as the output, achieving effective segmentation and annotation of sub-regions in the target region.

[0093] By combining texture network models with label information and inputting them into the neural network, the model's accuracy in recognizing feature boundaries and categories is effectively improved, making it particularly suitable for processing target areas with complex textures, such as canals. Furthermore, the output labeled texture network model exhibits clear sub-region boundaries and complete label information, providing accurate foundational data support for subsequent geographic information analysis, spatial modeling, or intelligent decision-making.

[0094] In one embodiment, the change monitoring module compares the surface structure and land use type of the target area at different time points based on the 3D model and semantic segmentation results of multiple flight cycles, and generates a visual change layer and monitoring report. Specifically, this includes: performing a time-series comparison of the 3D model and semantic segmentation results of the target area obtained in the multiple flight cycles to extract information on changes in surface morphology and land cover types in the target area; determining the conversion status of land use types in the target area based on the information on changes in surface morphology and land cover types; and generating a corresponding visual layer and detection report based on the conversion status.

[0095] Here, the system first acquires 3D models of the target area and their corresponding semantic segmentation results from multiple flight cycles using a change monitoring module. These flight cycles can be periodic tasks performed regularly by UAVs or other flight platforms to acquire geospatial information. Next, the change monitoring module compares and analyzes these multiple temporal 3D models and semantic segmentation results. Specifically, this includes: spatially aligning the 3D models from different periods using a registration algorithm; extracting information on changes in the target area's surface structure at different times, such as elevation changes, morphological variations, and geomorphic evolution; and comparing the semantic segmentation results to identify changes in land cover types at different points in time, such as reduced vegetation cover, building expansion, and water body drying. Based on this, the change monitoring module further determines the land use type conversion status of the target area based on both surface morphological and land cover type change information. This determination can be achieved through rule-based reasoning, statistical analysis, or machine learning models to identify type conversion events such as "farmland to construction land," "green space to hardened ground," and "water area reduction." Finally, the system generates corresponding visualization layers based on the conversion results. These layers visually display the location, type, and degree of change in the target area at different points in time. At the same time, it generates monitoring reports that include change statistics, trend analysis, spatial distribution, etc., providing users with quantitative and structured analytical basis.

[0096] In this way, it is possible to achieve time-series comparison and dynamic evolution analysis, accurately identify changes in land use types, and provide intuitive visualization results.

[0097] This application introduces high-precision 3D mapping, UAV patrol, and semantic segmentation technologies to improve the model's spatial accuracy, stereo recognition capability, and timeliness response level for land use. It can more accurately reflect the intensity of land use, spatial utilization efficiency, and degree of three-dimensional development.

[0098] Based on the same concept as the foregoing embodiments, this embodiment of the invention provides a land use change monitoring system based on real-scene 3D, comprising: a processor configured with a land use monitoring model as described in the first aspect; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the instructions to control the land use monitoring model to complete land use monitoring.

[0099] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0100] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A land use change monitoring model based on real-world 3D, characterized in that, include: The module includes a data acquisition module, a 3D modeling module, a point cloud registration module, a semantic segmentation module, and a change monitoring module; among them, The data acquisition module is used to periodically patrol the target area using a drone equipped with a tilting camera in order to obtain multi-angle images of the target area. The 3D modeling module is used to generate multi-temporal high-precision point cloud and texture network models based on the multi-angle images; The point cloud registration module is used to automatically register the multi-temporal high-precision point cloud to obtain the changes in the surface morphology and structure of the target area. The semantic segmentation module is used to perform semantic segmentation of ground features on the texture network model and to label sub-regions of the target region. The change monitoring module is used to compare the surface structure and land use type of the target area at different time points based on the three-dimensional model and semantic segmentation results of multiple flight cycles, and generate a visual change layer and monitoring report.

2. The model according to claim 1, characterized in that, After the data acquisition module periodically patrols the target area using a drone equipped with a tilting camera to acquire multi-angle images of the target area, it further includes: Based on the images acquired by the UAV, a time-series dataset of the target area is established to continuously monitor the state changes of the target area at different time points.

3. The model according to claim 1, characterized in that, When automatically registering the multi-temporal high-precision point cloud using the point cloud registration module to obtain the changes in the landform structure of the target area, the specific steps include: Automatic registration of the multi-temporal high-precision point cloud is performed using a coarse registration method based on morphologically invariant points; and / or The multi-temporal high-precision point cloud is automatically registered using a multi-temporal point cloud fine registration method based on morphologically invariant points.

4. The model according to claim 3, characterized in that, The multi-temporal point cloud coarse registration method based on morphologically invariant points automatically registers the multi-temporal high-precision point cloud, including: Generate a random spatial plane for the reference point cloud, and determine four coplanar points that satisfy the preset point cloud registration conditions based on the random spatial plane; Based on the four coplanar points, determine the four pairs of congruent points between the reference point cloud and the point cloud to be registered; Extract the local features of the congruent four-point pair in the target space neighborhood; the local features include the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered; The similarity between the local features of the reference point cloud and the local features of the point cloud to be registered is determined, and the coarse registration point cloud result of the reference point cloud is determined based on the similarity.

5. The model according to claim 4, characterized in that, The step of determining the similarity between the neighborhood local features of the reference point cloud and the neighborhood local features of the point cloud to be registered, and determining the coarse point cloud to be registered based on the similarity, specifically includes: The similarity between the local features of the reference point cloud and the local features of the point cloud to be registered is determined based on a preset Euclidean distance comparison function; the local features of the reference point cloud include multiple reference feature vectors, and the local features of the point cloud to be registered include multiple feature vectors to be registered. When the similarity is less than a preset similarity threshold, the corresponding reference feature vector is determined in the neighborhood local features of the point cloud to be registered, and the set of corresponding feature vectors to be registered is used as the point cloud result of the coarse registration.

6. The model according to claim 3, characterized in that, The multi-temporal point cloud fine registration method based on morphologically invariant points automatically registers the multi-temporal high-precision point cloud, including: Based on coarse registration of morphologically invariant points, the morphologically invariant regions of the changing scene are determined; The morphologically invariant region is registered using the iterative nearest point method, and the distance between the first and second point sets of the morphologically invariant region is optimized based on a preset optimization formula to obtain a finely registered point cloud result.

7. The model according to claim 6, characterized in that, The expression for the optimization formula is: R i and T i These represent the morphologically invariant regions Q to be registered. TIR Registered to the reference morphology invariant region P TIR The optimal rotation and translation parameters, #k represents P TIR and Q TIR The smaller value among the points, Δ i This is the cumulative sum of the minimum Euclidean distances between point sets calculated using the optimal rotation and translation in the i-th algorithm iteration.

8. The model according to claim 1, characterized in that, The step of performing semantic segmentation of the texture network model using a semantic segmentation module and labeling the sub-regions of the target region specifically includes: Using the texture network model and the labels associated with the texture network model as input, a pre-defined fully convolutional neural network is used to concatenate the texture network model and the labels associated with the texture network model to obtain an annotated texture network model, and the annotated texture network model is output to complete sub-region segmentation and annotation.

9. The model according to any one of claims 1-8, characterized in that, The change monitoring module, based on 3D models and semantic segmentation results from multiple flight cycles, compares the surface structure and land use types of the target area at different time points, and generates a visualized change layer and monitoring report, specifically including: The three-dimensional model and semantic segmentation results of the target area obtained from the multiple patrol cycles are compared in time to extract information on changes in landform and land cover types in the target area. Based on the information on changes in landform and land cover types, determine the conversion status of land use types in the target area; Based on the transformation results, a corresponding visualization layer and detection report are generated.

10. A land use change monitoring system based on real-scene 3D, characterized in that, include: The processor is configured with the land use monitoring model as described in any one of claims 1-9; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to control the land use monitoring model to complete land use monitoring.

Images