A high spatial resolution remote sensing image semantic change detection method and system

Abstract

The application discloses a kind of high spatial resolution remote sensing image semantic change detection method, system, it is related to remote sensing image analysis technical field, obtains remote sensing image and is preprocessed to obtain time series dataset;Grid division is carried out to target area;Extract the structural semantic features of each grid and overall area;Change index is extracted by constructing time series curve;Material density parameter is calculated;The semantic change detection is carried out by comprehensively structural features, time series curve and material parameter.The application realizes the accurate identification to local reconstruction+facade renovation and other mixed reconstruction scene by the multi-level detection framework of grid unit division and overall area analysis.Chequered grid division method is used, building area is decomposed into minimum functional unit, effectively separates the complex situation of local semantic great change and overall semantic unchanged, fundamentally solves the misjudgment problem caused by overall feature analysis of traditional method.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image analysis technology, specifically to a method and system for detecting semantic changes in remote sensing images with high spatial resolution. Background Technology

[0002] With the rapid development of remote sensing technology, high spatial resolution remote sensing imagery is playing an increasingly important role in the field of land surface change detection. This technological advancement has demonstrated significant advantages in improving change detection accuracy, enabling refined urban management, and promoting the construction of smart cities.

[0003] In the process of urban renewal and renovation, old residential area renovation projects often include different types of engineering activities such as facade renovation and partial demolition and reconstruction. These engineering activities exhibit complex spectral and textural features in remote sensing imagery, posing a significant challenge to the accurate identification of semantic changes.

[0004] In the remote sensing image change detection technology system, building semantic change recognition occupies a core position. The key to achieving accurate recognition lies in the comprehensive analysis of building structural features, material properties, and temporal change patterns. Currently, most existing building change detection methods are based on dual-temporal image comparison analysis, identifying changed areas by extracting overall feature differences. Undeniably, these methods can achieve good results in conventional change detection scenarios, providing some technical support for urban management. However, they also have significant limitations:

[0005] Existing methods struggle to effectively distinguish between facade renovation and demolition / reconstruction, two semantically fundamentally different types of change. Because these two types share highly similar image texture features, existing techniques relying solely on overall feature analysis are prone to misjudgment, misidentifying local reconstruction as overall renovation or vice versa, severely impacting the accuracy and practicality of change detection results. This limitation is particularly pronounced in complex scenarios such as the renovation of old residential communities, where a mixture of local structural reconstruction and overall facade renovation often exists. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for detecting semantic changes in remote sensing images with high spatial resolution, so as to overcome the shortcomings of the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting semantic changes in high spatial resolution remote sensing images, comprising:

[0008] Acquire remote sensing images of the target area, preprocess the images, and obtain a time-series image dataset;

[0009] Based on the aforementioned time-series image dataset, the target region is divided into grid cells;

[0010] Structural semantic features are extracted for each grid cell and the entire region.

[0011] Based on the aforementioned structural semantic features, a time-series change process curve is constructed, and change indicators are extracted;

[0012] Calculate the material density parameter based on the time-series change process curve;

[0013] Based on the structural semantic features, temporal change process curves, and material density parameters, semantic change detection is performed on the target region.

[0014] In a preferred embodiment, the remote sensing image includes high spatial resolution optical remote sensing image and synthetic aperture radar image, and the preprocessing includes registration, radiometric correction and cropping of the optical remote sensing image and the synthetic aperture radar image.

[0015] In a preferred embodiment, the grid cell division is based on the target area in the pre-reconstruction image.

[0016] In a preferred embodiment, the structural semantic features include main structural integrity, layer consistency, and spatial layout matching degree, wherein:

[0017] The integrity of the main structure is based on extracting the edge contour of the target area from optical images, calculating the contour continuity index, and analyzing the echo intensity distribution characteristics based on synthetic aperture radar images. The continuous proportion of the main structure is calculated comprehensively, and when the continuous proportion is ≥90%, it is determined that the main structure is unchanged.

[0018] The layer consistency is determined by analyzing the number of layers in the target area through the pixel stacking features in the vertical direction of the image, verifying the layer calculation results based on the relationship between shadow length and solar altitude angle, and comparing the layer difference between the corresponding areas before and after the transformation. When the layer difference is ≤0, it is determined that the number of layers has not changed.

[0019] The spatial layout matching degree is calculated by extracting feature points from the images before and after the renovation, and using a feature matching algorithm to calculate the matching degree of key functional areas. When the matching score is ≥85, it is determined that the spatial layout has not changed.

[0020] In a preferred embodiment, the change index includes change stage, duration and change rate, wherein the change stage is divided into no change stage, surface modification stage, main body deconstruction stage, main body reconstruction stage and stable stage. When only the no change stage, surface modification stage and stable stage are present, it is marked as a renovation candidate. When the no change stage, main body deconstruction stage, main body reconstruction stage and stable stage are present, it is marked as a reconstruction candidate.

[0021] In a preferred embodiment, the calculation of the material density parameter includes fusing features from optical images and synthetic aperture radar images. Specifically, it includes extracting the material spectral features from the optical images, analyzing the echo intensity features from the synthetic aperture radar images, and fusing the optical and radar features to calculate the material density parameter. When the difference between the material density parameter and the pre-modification value is ≤15%, it is determined that the material has not been substantially replaced. When the difference between the material density parameter and the pre-modification value is ≥40%, it is determined that the material has been substantially replaced.

[0022] In a preferred embodiment, the semantic change detection includes determining the semantic type of each grid cell individually and deriving an overall semantic determination result based on the determination results of all grid cells, specifically including:

[0023] The semantic type is determined for each grid cell individually. The determination results of all grid cells are statistically analyzed. When the proportion of grid cells determined to have undergone a major semantic change is ≥20% and they are continuously distributed, the overall determination is local reconstruction. When the proportion of grid cells determined to have undergone a major semantic change is <20% or they are discretely distributed, the overall determination is overall renovation.

[0024] The present invention also provides a data acquisition module for acquiring remote sensing images of a target area;

[0025] The preprocessing module is used to preprocess the images to obtain a time-series image dataset;

[0026] The grid division module is used to divide the target area into grid cells based on the time-series image dataset;

[0027] The feature extraction module is used to extract structural semantic features for each grid cell and the entire region respectively;

[0028] The time-series analysis module is used to construct a time-series change process curve based on the structural semantic features and extract change indicators;

[0029] The material analysis module is used to calculate the material density parameters based on the time-series change process curve;

[0030] The semantic detection module is used to detect semantic changes in the target region based on the structural semantic features, the temporal change process curve, and the material density parameters.

[0031] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0032] A multi-layered detection architecture combining grid cell division and overall region analysis enables accurate identification of mixed renovation scenarios such as partial reconstruction and facade renovation. Employing a checkerboard grid division method, the building area is decomposed into the smallest functional units, effectively separating the complex situation of drastic local semantic changes from those of unchanged overall semantics, fundamentally solving the misjudgment problem caused by the overall feature analysis of traditional methods.

[0033] By integrating multimodal remote sensing data and comprehensively utilizing multi-dimensional features such as structural integrity, layer consistency, and spatial layout matching, a comprehensive semantic feature system was constructed. Combining temporal change process modeling with material density analysis, a multi-verification mechanism was established, significantly improving the accuracy of distinguishing between facade renovation and demolition / reconstruction.

[0034] By constructing time-series change process curves and extracting quantitative indicators such as change stages, duration, and rate of change, the characteristics of different stages in the renovation process can be accurately identified. This method is particularly suitable for long-term engineering projects such as the renovation of old residential communities, and can effectively track different types of evolution processes, such as surface modification and main structure reconstruction.

[0035] The independent judgment mechanism based on grid cells can accurately locate the specific location and scope of semantic changes. By setting reasonable threshold standards, it can capture meaningful local reconstructions while avoiding misjudging discrete, minor changes as significant alterations, thus providing a reliable basis for urban planning and management decisions. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0037] Figure 1 This is a flowchart of the method of the present invention.

[0038] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1, please refer toFigure 1 As shown in this embodiment, a method for detecting semantic changes in high spatial resolution remote sensing images includes:

[0041] S1: Acquire remote sensing images of the target area, preprocess the images, and obtain a time-series image dataset;

[0042] S2: Based on the time-series image dataset, the target area is divided into grid cells;

[0043] S3: Extract structural semantic features for each grid cell and the entire region;

[0044] S4: Based on the aforementioned structural semantic features, construct a time-series change process curve and extract change indicators;

[0045] S5: Calculate the material density parameter based on the time-series change process curve;

[0046] S6: Based on the structural semantic features, temporal change process curves, and material density parameters, perform semantic change detection on the target region.

[0047] With the rapid development of remote sensing technology, high spatial resolution remote sensing imagery is playing an increasingly important role in the field of land surface change detection. This technological advancement has demonstrated significant advantages in improving change detection accuracy, enabling refined urban management, and promoting the construction of smart cities.

[0048] In the process of urban renewal and renovation, old residential area renovation projects often include different types of engineering activities such as facade renovation and partial demolition and reconstruction. These engineering activities exhibit complex spectral and textural features in remote sensing imagery, posing a significant challenge to the accurate identification of semantic changes.

[0049] In the remote sensing image change detection technology system, building semantic change recognition occupies a core position. The key to achieving accurate recognition lies in the comprehensive analysis of building structural features, material properties, and temporal change patterns. Currently, most existing building change detection methods are based on dual-temporal image comparison analysis, identifying changed areas by extracting overall feature differences. Undeniably, these methods can achieve good results in conventional change detection scenarios, providing some technical support for urban management. However, they also have significant limitations:

[0050] Existing methods struggle to effectively distinguish between facade renovation and demolition / reconstruction, two semantically fundamentally different types of change. Because these two types share highly similar image texture features, existing techniques relying solely on overall feature analysis are prone to misjudgment, misidentifying local reconstruction as overall renovation or vice versa, severely impacting the accuracy and practicality of change detection results. This limitation is particularly pronounced in complex scenarios such as the renovation of old residential communities, where a mixture of local structural reconstruction and overall facade renovation often exists.

[0051] As described in S1-S6 above, a multi-layered detection architecture combining grid cell division and overall region analysis enables accurate identification of mixed renovation scenarios such as partial reconstruction and facade renovation. A checkerboard grid division method is used to decompose the building area into the smallest functional units, effectively separating the complex situation of significant local semantic changes from overall semantic invariance, fundamentally solving the misjudgment problem caused by the overall feature analysis of traditional methods. By integrating multimodal remote sensing data and comprehensively utilizing multi-dimensional features such as structural integrity, layer consistency, and spatial layout matching, a comprehensive semantic feature system is constructed. Combining temporal change process modeling and material density analysis, a multi-verification mechanism is formed, significantly improving the accuracy of distinguishing between facade renovation and demolition / reconstruction.

[0052] By constructing a time-series change process curve and extracting quantitative indicators such as change stages, duration, and rate of change, the characteristics of different stages in the renovation process can be accurately identified. This method is particularly suitable for long-term engineering projects such as the renovation of old residential areas, and can effectively track different types of evolution processes such as surface modification and main structure reconstruction. Based on an independent judgment mechanism using grid cells, the specific location and scope of semantic changes can be accurately located. By setting reasonable threshold standards, meaningful local reconstructions can be captured while avoiding misjudging discrete, minor changes as significant renovations, providing a reliable decision-making basis for urban planning and management.

[0053] In one embodiment, step S1 includes:

[0054] S11. Acquire multi-temporal, high spatial resolution optical remote sensing images and concurrent synthetic aperture radar (SAR) images of the target area. Optical remote sensing images provide rich texture and spectral information, while SAR images have penetration and sensitivity to structural changes. The combination of the two provides a multi-source data foundation for subsequent analysis. The data acquisition time span covers the early, middle, and late stages of the transformation, with the time interval between two adjacent images not exceeding 30 days to ensure temporal continuity.

[0055] S12. Perform precise spatiotemporal registration on the acquired optical remote sensing images and SAR images. Employ feature point matching and geometric correction algorithms to unify multi-source images to the same geographic coordinate system and spatial resolution, eliminating displacement errors caused by sensor differences and imaging conditions, and ensuring pixel-level alignment between images from different time phases;

[0056] S13. Perform radiometric correction on the registered images, including atmospheric correction and sensor radiometric calibration. For optical images, use dark target subtraction or radiative transfer model to eliminate the effects of atmospheric scattering and absorption; for SAR images, eliminate radiometric distortion caused by antenna mode and topographic effects through calibration processing, thereby obtaining the true surface reflectance or backscattering coefficient and improving data comparability.

[0057] S14. Based on the boundary range of the target area, crop the calibrated image to extract the region of interest. The cropping operation is based on geographic vector boundaries or manually defined to ensure that all temporal images cover the same spatial range, ultimately generating a spatiotemporally aligned time-series image dataset, providing standardized input for subsequent grid division and feature extraction.

[0058] As described in S11-S14 above, the registration and radiometric correction of multi-source images effectively eliminate geometric and radiometric distortions, ensuring the consistency of temporal data and laying a reliable foundation for subsequent accurate analysis. Combining the rich detail of optical images with the structural sensitivity of SAR images enhances the ability to distinguish between facade renovation and demolition reconstruction, particularly improving detection robustness in complex scenarios. Strict time interval control and spatiotemporal alignment processing enable the modeling of temporal changes to accurately capture gradual changes, avoiding misjudgments caused by data inconsistencies. The standardization and algorithmization of sub-steps automate the preprocessing workflow, improving engineering application efficiency and ensuring the repeatability of results, making it suitable for large-scale urban monitoring scenarios.

[0059] In one embodiment, step S2 includes:

[0060] S21. Use the pre-renovation imagery as the baseline data for grid division to ensure that all subsequent analyses are based on a unified initial state. The reason for choosing pre-renovation imagery is that the building structure remains in its original state at this time, providing a reliable reference baseline for subsequent change detection.

[0061] S22. Using deep learning segmentation algorithms or traditional edge detection algorithms, accurately extract building outlines from pre-renovation images. Optimize the outline boundaries through morphological operations to eliminate noise interference and obtain accurate building area boundaries.

[0062] S23. Determine the grid cell size based on the minimum functional zone size of the building. By analyzing the internal structural characteristics of the building, identify the minimum functional unit (such as a single room, balcony, etc.), and set the grid cell size to 1 / 2 of the minimum functional zone size to ensure that each grid cell can independently represent local semantic features.

[0063] S24. Based on the building outline and the calculated unit size, generate a regular mesh covering the entire building area. For irregular building outlines, use a boundary adaptive adjustment algorithm to ensure that the mesh units completely cover the target area while avoiding including too much background information.

[0064] S25. Perform quality inspection on the generated mesh, including evaluating indicators such as mesh cell integrity and boundary alignment. Adjust and optimize non-compliant meshes through manual verification or automated testing to ensure the accuracy and applicability of the mesh generation.

[0065] Among them, the grid cell size is usually 2-5 meters, which can be adjusted according to the building type and resolution; the grid shape is preferably a square grid, and a rectangular grid can be used in special scenarios; the boundary processing is to use bilinear interpolation to process boundary pixels to ensure the accuracy of feature extraction.

[0066] As described in S21-S25 above, grid division based on pre-renovation images ensures spatial consistency across all temporal analyses, laying the foundation for accurate temporal comparison. The grid size design based on minimum functional partitions captures local architectural details while avoiding feature fragmentation caused by excessive subdivision, achieving an optimal balance between computational efficiency and feature integrity. The adaptive grid generation algorithm effectively handles various building shapes and sizes, particularly suitable for the irregular building layouts in older residential areas. Scientifically sound grid division provides a robust spatial framework for subsequent feature extraction and change detection, ensuring that each grid cell can perform independent semantic analysis while maintaining overall coherence. A grid verification mechanism guarantees the quality of the division, and combined with clear technical parameter standards, this method possesses good operability and repeatability, making it suitable for engineering-scale application.

[0067] In one embodiment, the structural semantic features in step S3 include the integrity of the main structure, the consistency of the number of layers, and the spatial layout matching degree, as detailed below:

[0068] S31. The process for extracting the integrity of the main structure includes:

[0069] S311. Optical Image Edge Contour Extraction: An improved Canny edge detection algorithm is used, combined with prior knowledge of the building structure, to extract the main building contour. Multi-scale edge enhancement techniques are employed to preserve structural edges and filter out texture noise.

[0070] S312. Contour Continuity Quantitative Analysis: Based on the extracted edge contours, calculate the following continuity indices:

[0071] Contour break point density: the number of edge break points per unit length;

[0072] Contour curvature consistency: the range of curvature variation between adjacent edge segments;

[0073] Regularity of corner point distribution: Spatial distribution pattern of structural corner points.

[0074] S313, SAR echo intensity characteristic analysis; using the backscattering characteristics of synthetic aperture radar imagery: extract the σ0 value distribution characteristics of the building area; analyze the spatial variation coefficient of echo intensity; calculate the contribution ratio of structure scattering and volume scattering.

[0075] S314. Calculation of continuous proportion; Establishment of multi-source feature fusion model:

[0076] Continuity percentage = w1 × contour continuity score + w2 × echo stability score + w3 × scattering consistency score;

[0077] The weighting coefficients were determined through optimization based on a large amount of experimental data.

[0078] S32. The layer consistency analysis process includes:

[0079] S321, Pixel stacking feature extraction; Based on high-resolution optical imagery: Using object-oriented image analysis methods to identify building floor division features; Extracting vertical texture periodicity through morphological profile analysis; Predicting the number of floors using deep learning models (such as ResNet);

[0080] S322, Shadow geometry verification; Establish an accurate shadow analysis model: Calculate the solar altitude angle and azimuth angle based on the imaging time; Theoretically estimate the building height based on the building outline and solar geometry; Compare the matching degree between the measured shadow length and the theoretical calculation value;

[0081] S323, Multi-temporal layer number comparison; Constructing a layer number change detection framework: Calculate the layer number independently for each grid cell; Establish the mapping relationship between the layer number before and after the modification; Calculate the statistical significance of the layer number difference;

[0082] S33, Spatial Layout Matching Degree Calculation Process;

[0083] S331, Multi-scale feature point detection; Adaptive feature extraction strategy is adopted: Detect overall building outline feature points at a large scale; Extract detailed feature points such as doors and windows at a fine scale; Combine multiple feature descriptors such as SIFT and ORB;

[0084] S332, Feature Matching and Optimization; Implementing Strict Matching Quality Control: Using the RANSAC algorithm to remove mismatched points; Verifying matching consistency based on geometric constraints; Calculating the spatial distribution uniformity of matching points;

[0085] S333, Functional Area Matching Scoring; Establishing a Comprehensive Scoring System:

[0086] Match score = Base matching rate × Spatial consistency coefficient × Feature stability factor;

[0087] The weights of each component are determined through machine learning optimization.

[0088] S34. Technical parameters and judgment criteria;

[0089] S341. Determination of the integrity of the main structure:

[0090] Continuous proportion ≥ 90%: Structure unchanged (confidence level > 95%);

[0091] 70% ≤ continuous percentage < 90%: further verification is needed;

[0092] A continuous percentage of less than 70% indicates a change in the structure;

[0093] S342, Layer Consistency Determination:

[0094] The difference in the number of floors is ≤0: the number of floors remains unchanged;

[0095] Layer difference > 0: Requires verification in conjunction with other features;

[0096] Shadow verification difference >10%: Trigger manual review;

[0097] S343. Spatial layout matching degree determination:

[0098] Match score ≥ 85: Layout unchanged;

[0099] Match score 60 ≤ Match score < 85: Some variations;

[0100] Match score < 60: Major layout changes;

[0101] As described in S31-S34 above, a complementary verification mechanism is formed by combining the detail advantages of optical imagery with the structural sensitivity of SAR imagery, significantly improving the reliability of feature extraction. From macroscopic structure to microscopic layout, a complete semantic feature description system for buildings is constructed, providing sufficient basis for accurately distinguishing between renovation and reconstruction. The judgment threshold optimized based on a large amount of experimental data ensures both detection sensitivity and control of false alarm rate, demonstrating good stability in practical applications. Through multi-feature cross-validation and confidence assessment, a complete quality control chain is formed to ensure the credibility of feature extraction results. The clear processing flow and judgment criteria make this method highly operable and repeatable, suitable for large-scale engineering applications.

[0102] In one embodiment, the time-series change process modeling process in step S4 includes:

[0103] S41, Intelligent identification sub-step during the change phase;

[0104] S411. A stage identification algorithm based on Hidden Markov Model (HMM) is adopted:

[0105] A five-state Markov chain is constructed, corresponding to five stages of change; the state transition probability is calculated based on multi-temporal feature sequences; and the Viterbi algorithm is used to solve for the optimal state sequence.

[0106] S412. Specific Stage Characteristics Definition:

[0107] No change phase: The change rate of all structural semantic features is <5%;

[0108] Surface finishing stage: Texture feature change rate > 30%, structural feature change rate < 10%;

[0109] Main body deconstruction stage: Structural integrity index decline rate > 15% / ten-day period;

[0110] Main reconstruction phase: Structural integrity index increases at a rate >10% / ten-day period, number of layers may change;

[0111] Stable phase: All characteristic change rates remain <3% for more than 30 days;

[0112] S42, Sub-step for precise quantification of duration;

[0113] S421. Establish a time series analysis model:

[0114] The sliding window method is used to detect the start and end times of each stage; the CUSUM algorithm is used to identify inflection points of change; and the cumulative number of days and working day adjustment coefficient for each stage are calculated.

[0115] S422, Key Duration Threshold Setting:

[0116] Surface finishing stage: Normal range 15-60 days;

[0117] Main body deconstruction + reconstruction phase: Normal range 90-180 days;

[0118] Stage anomaly warning: Automatic review is triggered when the threshold range is exceeded;

[0119] S43, Sub-step for dynamic calculation of rate of change;

[0120] S431. Construct a multi-dimensional rate of change index system:

[0121] Short-term rate of change: based on a 7-day sliding window;

[0122] Medium-term trend: Based on 30-day trend analysis;

[0123] Long-term trend: based on full-cycle regression analysis;

[0124] S432, Specific Calculation Model:

[0125] Surface texture change rate = Δtexture feature value / Δt;

[0126] Rate of change of the main outline = Δoutline integrity / Δt;

[0127] Overall change intensity = w1 × texture change + w2 × outline change + w3 × layer change;

[0128] S433, Phase Determination and Candidate Labeling Mechanism;

[0129] The logic for determining renovation candidates:

[0130] Phase sequence ∈ {no change → surface modification → stable}; surface modification duration ≤ 60 days; maximum texture change rate ≥ 30%; maximum contour change rate ≤ 10%; marked as renovation candidate;

[0131] Logic for reconstructing candidates:

[0132] The phase sequence ∈ {no change → main body deconstruction → main body reconstruction → stable}; the duration of (deconstruction + reconstruction) is ≥90 days; the maximum contour change rate is ≥50%; there is evidence of layer changes or structural reorganization; it is marked as a reconstruction candidate.

[0133] S434. Abnormal handling and quality control;

[0134] Phase jump detection: Automatically marks abnormal phase transitions and requires manual verification;

[0135] Data missing handling: Time series interpolation and smoothing techniques are used to fill data gaps;

[0136] Confidence assessment: Provide a confidence score (0-100%) for each decision;

[0137] As described in S41-S43 above, this method, through a five-stage detailed division, comprehensively describes the continuous change process from surface modification to structural reconstruction, effectively capturing the full-cycle characteristics of the renovation project. Combining quantitative indicators across three dimensions—stage, duration, and rate—a comprehensive change assessment system is formed, providing ample basis for semantic judgment. The judgment threshold, optimized based on extensive experimental data, is adaptable to renovation projects of different scales and types, exhibiting good generalization ability. A robust anomaly detection and handling mechanism ensures reliable analysis results even with incomplete or noisy data. Clear judgment logic and quality control processes make this method highly operable and reliable in practical engineering applications, providing strong technical support for urban renewal management.

[0138] In one embodiment, step S5 includes:

[0139] Material density parameter calculation process

[0140] S51, Sub-step for extracting spectral features of optical image materials

[0141] Establish a multispectral feature analysis system:

[0142] (1) Extraction of spectral reflectance features;

[0143] Continuous spectra in the range of 400-2500 nm were obtained using hyperspectral imaging technology; characteristic spectral curves of building surface materials were extracted; and key parameters such as spectral absorption depth and reflection peak position were calculated.

[0144] (2) Quantitative analysis of texture features;

[0145] Contrast, correlation, energy, and homogeneity are calculated based on the gray-level co-occurrence matrix (GLCM); multi-scale and multi-directional texture features are extracted using Gabor filter banks; and local binary mode (LBP) features are calculated to describe the microstructure of the material surface.

[0146] (3) Material type identification;

[0147] Material recognition model = f(spectral features, texture features, spatial context features);

[0148] Supports identification of common building materials such as paint, tiles, stone, metal, and glass;

[0149] S52, SAR image echo intensity characteristic analysis sub-step;

[0150] Constructing a radar feature extraction framework:

[0151] (1) Backscattering characteristics analysis;

[0152] Extract the time series of the σ0 backscattering coefficient; analyze the scattering characteristics of the HH, HV, VH, and VV polarization channels; calculate the eigenvalue decomposition parameters of the scattering matrix;

[0153] (2) Measurement of interferometric coherence;

[0154] Calculate coherence coefficients based on multi-temporal InSAR data; analyze the spatial distribution patterns of decoherence and complex coherence; extract scattering stability indices related to material density.

[0155] (3) Penetration characteristics assessment;

[0156] Analyze the penetration capability of L-band SAR through building materials; assess the contribution of the internal structure of the material to the echo signal; calculate the ratio of volume scattering to surface scattering.

[0157] S53, Multi-source feature fusion and compaction calculation sub-step;

[0158] Establish a feature fusion and parameter calculation model:

[0159] (1) Feature standardization processing;

[0160] Z-score standardization was used to eliminate the influence of dimensions; principal component analysis (PCA) was applied to reduce dimensionality and remove redundancy; and a feature importance weight allocation mechanism was established.

[0161] (2) Compactness parameter calculation model;

[0162] Material density parameter = α × spectral stability + β × texture uniformity + γ × scattering uniformity + δ × penetration response;

[0163] Where: α+β+γ+δ=1, the weights are adaptively adjusted based on the material type;

[0164] (3) Analysis of temporal changes;

[0165] Calculate the time series of compactness parameters throughout the entire renovation cycle; analyze the trends and abrupt change points of parameter changes; identify characteristic patterns related to construction activities;

[0166] (4) Judgment criteria and quality control;

[0167] Material replacement determination system:

[0168] The material was not substantially changed;

[0169] Judgment criteria: The difference in density parameters is ≤15%;

[0170] Typical characteristics: stable spectral features, consistent texture patterns, and continuous scattering properties;

[0171] Applicable scenarios: facade cleaning, repair of the same material, and color replacement;

[0172] The materials have been partially replaced.

[0173] Judgment criteria: 15% < difference in compactness parameter < 40%;

[0174] Typical characteristics: Local features change, but the overall pattern remains unchanged;

[0175] Processing strategy: Further analysis should be conducted in conjunction with spatial distribution;

[0176] The material has been substantially replaced.

[0177] Judgment criteria: The difference in density parameters is ≥40%;

[0178] Typical characteristics: changes in spectral curve morphology, reorganization of texture structure, and alteration of scattering mechanism;

[0179] Applicable scenarios: Change of material type, replacement of structural materials;

[0180] Quality control mechanism:

[0181] Confidence assessment;

[0182] Confidence level = Feature consistency × Temporal stability × Spatial coherence;

[0183] Manual review is triggered when the confidence level is <0.7;

[0184] Handling abnormal situations;

[0185] Weather impact correction: Remove outlier data caused by rain or snow.

[0186] Seasonal effect compensation: Establishing a baseline model for seasonal changes;

[0187] Sensor differential calibration: unifying data standards for different sensors;

[0188] As described in S51-S53 above, optical imaging provides rich information about the material surface, while SAR imaging reveals the internal structural characteristics of the material. The combination of these two forms a complete material characterization system. Based on rigorous mathematical models and physical mechanisms, quantitative assessment of material changes is achieved, significantly reducing subjective judgment errors. Through full-cycle time-series analysis, the dynamic process of material changes can be captured, providing data support for engineering progress assessment. A robust quality control mechanism and anomaly handling strategy ensure reliable detection performance even under complex environmental conditions. Clear judgment criteria and a comprehensive processing flow make this method highly operable and reliable in practical engineering applications, providing a powerful technical means for identifying and replacing building materials.

[0189] In one embodiment, the mesh cell-level semantic determination process in step S6 is as follows:

[0190] S61, Multi-evidence fusion determination sub-step;

[0191] Establish a fusion judgment model based on DS evidence theory:

[0192] Semantic type determination = F(structural evidence, temporal evidence, material evidence);

[0193] in:

[0194] Structural evidence = {integrity of main structure, consistency of number of floors, and matching degree of spatial layout};

[0195] Temporal evidence = {Sequence of change phases, duration pattern, and rate of change characteristics};

[0196] Material evidence = {Changes in material density, stability of spectral characteristics, consistency of scattering properties};

[0197] Specific judgment rules:

[0198] (1) Judgment condition for semantic unchanged (renovation):

[0199] Structural evidence: The main structure has a continuous proportion of ≥85%, the difference in the number of floors is ≤0, and the layout matching score is ≥80.

[0200] Temporal evidence: Only the sequence of no change → surface modification → stable phase appears;

[0201] Material evidence: Difference in density parameters ≤15%;

[0202] (2) Criteria for determining semantic drastic change (reconstruction):

[0203] Structural evidence: The main structure has a continuous proportion of less than 70%, or the difference in the number of floors is greater than 0, or the layout matching score is less than 60.

[0204] Temporal evidence: Entity remains intact without change → Subject deconstruction → Subject reconstruction → Stable sequence;

[0205] Material evidence: Difference in density parameters ≥ 40%;

[0206] S62, Determine the confidence level calculation sub-step; Construct a multi-dimensional confidence level assessment system:

[0207] Unit confidence level = w1 × structural confidence level + w2 × time series confidence level + w3 × material confidence level;

[0208] Structural confidence = f(feature consistency, measurement accuracy, noise level);

[0209] Time series confidence = g(stage clarity, data completeness, trend significance);

[0210] Material confidence level = h(characteristic stability, sensor consistency, environmental adaptability);

[0211] Confidence level classification:

[0212] High confidence level: ≥0.8, directly use the automatic judgment result;

[0213] Medium confidence level: 0.6-0.8, validated in conjunction with spatial context;

[0214] Low confidence level: <0.6, triggering manual review mechanism;

[0215] Overall semantic integration and analysis process;

[0216] S63, Spatial Distribution Pattern Analysis Sub-step; employing advanced spatial statistical analysis techniques:

[0217] (1) Connectivity analysis;

[0218] Identifying continuously distributed regions based on the 8-neighborhood principle;

[0219] Calculate the morphological parameters of connected regions, such as area, perimeter, and compactness.

[0220] Analyze spatial clustering and distribution patterns;

[0221] (2) Spatial autocorrelation analysis;

[0222] Calculate the global Moran's I index to assess the overall spatial clustering.

[0223] Hotspot areas were identified using Getis-OrdGi statistics;

[0224] Establish a quantitative descriptive model for spatial heterogeneity;

[0225] S64, Overall Semantic Determination Sub-step

[0226] Establish a graded judgment system:

[0227] (1) Criteria for determining local reconstruction:

[0228] The proportion of semantically drastically changed units is ≥20%; the area of ​​the largest connected region is ≥50% of the total drastically changed area; the spatial clustering index is ≥0.6;

[0229] The assessment is for partial reconstruction plus overall facade renovation;

[0230] (2) Criteria for overall renovation judgment:

[0231] The proportion of semantically drastically changed units is <20%; the proportion of semantically drastically changed units is ≥20%, but the area of ​​the largest connected region is <30% of the total drastically changed area; the spatial clustering index is <0.4;

[0232] It was determined to be a complete facade renovation;

[0233] S65, Result Optimization and Verification Sub-step;

[0234] Implementing multi-level quality control specifically includes:

[0235] (1) Spatial consistency optimization;

[0236] (2) The semantic label distribution is optimized using a Conditional Random Field (CRF) model;

[0237] (3) Smoothing anomaly determination results based on spatial neighborhood relationships;

[0238] Maintain semantic coherence in the boundary region;

[0239] Multi-scale verification mechanism;

[0240] (1) Grid cell level: random sampling and manual verification;

[0241] (2) Architectural level: Comparison with field survey data;

[0242] (3) Regional level: Verify with urban planning archives;

[0243] The abnormal situation handling mechanism specifically includes:

[0244] Handling of conflict of evidence;

[0245] (1) When the results of different evidence sources are inconsistent, the conflict resolution algorithm is activated;

[0246] (2) Make weighted decisions based on the reliability of evidence and historical accuracy;

[0247] (3) Introduce an expert knowledge base for arbitration judgment;

[0248] Boundary case handling;

[0249] (1) Establish a fuzzy judgment mechanism for edge cases that are close to the threshold;

[0250] (2) Provide probabilistic outputs instead of just binary decisions;

[0251] (3) Retain the uncertainty measure for subsequent analysis;

[0252] As described in S61-S65 above, the progressive analysis framework from grid cells to the overall building ensures both the sensitivity of local change detection and the accuracy of overall judgment. Spatial distribution pattern analysis effectively identifies meaningful areas of continuous change, avoiding misjudgments based on simple proportional thresholds. The collaborative judgment mechanism using multi-source evidence significantly improves the reliability of semantic type recognition and reduces the risk of insufficient evidence from a single source. From confidence assessment to spatial optimization and multi-scale verification, a complete quality assurance chain is formed. Clear judgment criteria, a robust anomaly handling mechanism, and an adjustable parameter system enable this method to demonstrate good adaptability and stability in practical applications. It not only provides binary judgment results but also outputs rich information such as confidence levels and spatial distribution characteristics, providing comprehensive decision support for urban planning and management.

[0253] In one implementation, the basis for determining the key threshold was verified through multiple sets of comparative experiments. All experimental samples came from real-world scenarios of old residential area renovation. The experimental data were based on a unified high spatial resolution optical remote sensing image (0.5-meter level) and SAR image dataset. The preprocessing procedure was consistent with the aforementioned embodiments of the present invention.

[0254] Verification experiment to ensure that the integrity of the main structure is continuous with a percentage of ≥90%;

[0255] Experimental objective:

[0256] Determine the critical continuous ratio that distinguishes between unchanged and changed main building structures, balance the missed detection rate of structural changes with the false judgment rate of unchanged structures, and adapt to the accurate detection needs of mixed renovation scenarios.

[0257] Experimental samples:

[0258] We selected 120 old residential community renovation cases and categorized them into three types based on renovation type: pure renovation scenarios (40 cases, only facade modification, main structure unchanged), pure reconstruction scenarios (40 cases, main structure completely replaced), and mixed scenarios of partial reconstruction and overall renovation (40 cases, partial structure replacement + overall facade modification). All samples include complete optical and SAR images before and after the renovation, and the actual structural change labels have been obtained through on-site surveys (unchanged structure = 1, structural change = 0).

[0259] Experimental steps:

[0260] According to the aforementioned method of the present invention, the continuous proportion of the main structure of 120 samples is calculated respectively (by fusing the continuous index of the edge contour of optical image and the echo intensity distribution characteristics of SAR image).

[0261] Four candidate thresholds were set: 70%, 80%, 90%, and 95%. Each threshold was used as the judgment standard to calculate the accuracy rate of determining unchanged structure, the accuracy rate of determining structural change, the overall accuracy rate, and the misjudgment rate at each threshold.

[0262] By comparing the detection performance at different thresholds, the optimal threshold is selected.

[0263] Table 1: Comparison of experimental performance under different thresholds for judging the integrity of the main structure

[0264]

[0265] Experimental conclusion:

[0266] When the threshold is set to 90%, the overall accuracy reaches its highest level (96%), with a missed detection rate of only 4% for structural changes and a false positive rate of less than 8% for unchanged structures. This achieves the best balance between accurately identifying structural changes and avoiding misjudgments of renovation. This threshold can effectively distinguish between pure renovation (unchanged main structure) and pure reconstruction / partial reconstruction (changes in the main structure), which meets the core requirement of prioritizing the accuracy of structural change detection in urban renewal supervision. Therefore, the threshold for determining the integrity of the main structure is determined to be a continuous percentage ≥ 90%.

[0267] Verification experiment on spatial layout matching score ≥ 85 points;

[0268] Experimental objective:

[0269] Determine the critical score that distinguishes between unchanged and changed building spatial layout, accurately identify layout changes such as door and window relocation and unit type adjustment caused by partial reconstruction, and avoid misjudgment in pure renovation scenarios.

[0270] Experimental samples:

[0271] Eighty building samples were selected, including 40 pure renovation scenarios (spatial layout unchanged, only facade modification) and 40 partial reconstruction scenarios (spatial layout changes, such as adjustment of door and window positions, and reconstruction of apartment layouts). Key feature points (door and window corners, apartment layout inflection points) were extracted from the images before and after the renovation of all samples, and the actual layout change labels were obtained through on-site surveying.

[0272] Experimental steps:

[0273] According to the aforementioned method of the present invention, the SIFT feature extraction algorithm and the FLANN matching algorithm are used to calculate the spatial layout matching score (out of 100 points) for each sample.

[0274] Five candidate thresholds were set: 60, 70, 80, 85, and 90 points. The accuracy rates for determining layout unchanged, layout changed, and overall accuracy were calculated for each threshold.

[0275] The optimal threshold is selected to balance the detection accuracy of both types of scenarios.

[0276] Experimental results:

[0277] Table 2: Comparison of experimental performance under different judgment thresholds for spatial layout matching degree

[0278]

[0279] Experimental conclusion:

[0280] The highest overall accuracy rate (92%) is achieved at a score of 85. Among them, the accuracy rate for judging layout changes reaches 95%, which can accurately capture spatial layout adjustments caused by local reconstruction. The misjudgment rate for unchanged layout is only 11%, which can effectively avoid misjudging pure renovation scenarios as layout changes. This meets the need to accurately distinguish between local structural changes and surface modifications in mixed renovation scenarios. Therefore, the judgment threshold for spatial layout matching is determined to be a matching score ≥ 85.

[0281] Verification experiment with ≥20% semantically drastically changed grid cells;

[0282] Experimental objective:

[0283] Determine the critical mesh percentage that distinguishes between partial reconstruction + overall renovation and overall renovation + minor repairs, so as to avoid misjudging minor repairs as effective partial reconstruction, while ensuring that substantial partial reconstruction is not missed.

[0284] Experimental samples:

[0285] One hundred mixed renovation cases were selected, including 50 cases of partial reconstruction + overall renovation (replacement of local continuous area structures and overall facade modification), and 50 cases of overall renovation + minor repairs (only minor changes such as wall repairs and small balcony modifications, without substantial structural replacement). All samples were divided into grid units according to the aforementioned method of this invention (unit size is 1 / 2 of the smallest functional area of ​​the building).

[0286] Experimental steps;

[0287] According to the method of the foregoing embodiments of the present invention, the number of grid cells with semantically drastic changes in each sample is determined, and the proportion of them to the total number of grid cells is calculated.

[0288] Four candidate thresholds of 10%, 15%, 20%, and 25% were set, and the overall semantic judgment accuracy and misjudgment rate under each threshold were statistically analyzed in conjunction with the results of field investigation (construction logs, planning approval documents).

[0289] The optimal threshold can be selected to accurately distinguish between the two types of scenarios.

[0290] Experimental results:

[0291] Table 3: Comparison of experimental performance under different percentage thresholds of semantically drastic grid cell changes

[0292]

[0293] Experimental conclusion:

[0294] At the 20% threshold, the overall renovation accuracy rate reaches 96%, the probability of misjudging sporadic repairs as partial reconstruction is only 4%, and the accuracy rate of partial reconstruction remains at 92%, effectively distinguishing between substantial partial reconstruction and sporadic repairs. Considering the need for accurate statistical analysis of renovation results in urban renewal supervision, this threshold ensures that only structural changes accounting for more than 20% continuously are recognized as effective partial reconstruction. Therefore, the threshold for determining semantically significant grid units is determined to be ≥20% and continuously distributed.

[0295] Verification experiments on temporal feature thresholds (surface modification ≤ 60 days, deconstruction + reconstruction ≥ 90 days);

[0296] Experimental basis:

[0297] The fundamental differences in building renovation processes are significant: renovation mainly involves surface treatment (such as paint renovation and tile replacement) and has a short construction period; reconstruction requires a full process of demolition, foundation construction, main structure construction and topping out, and has a long construction period. The construction periods of the two types of scenarios do not overlap.

[0298] Experimental samples:

[0299] Complete construction logs for 60 renovation projects were collected, including 30 pure renovation projects and 30 pure reconstruction projects. All projects correspond to complete time-series remote sensing images (one period every 15 days).

[0300] Experimental steps:

[0301] Extract the change stages of each project based on time-series images (no change, surface modification, main body deconstruction, main body reconstruction, stable);

[0302] The duration of the surface finishing phase in pure renovation projects and the duration of the main structure deconstruction + main structure reconstruction phases in pure reconstruction projects are statistically analyzed.

[0303] Analyze the project duration distribution patterns in the two scenarios to determine the critical threshold.

[0304] Experimental results:

[0305] Table 4: Statistics on the distribution of construction period for key stages of different renovation types

[0306]

[0307] Experimental conclusion:

[0308] For pure renovation projects, the surface finishing phase lasts ≤60 days, while for pure reconstruction projects, the main structure deconstruction + main structure reconstruction phases last ≥90 days. The project durations for these two scenarios do not overlap. Therefore, a surface finishing phase of ≤60 days is defined as a renovation characteristic, and a main structure deconstruction + main structure reconstruction phase of ≥90 days is defined as a reconstruction characteristic, ensuring the accuracy of the timeline phase division.

[0309] Verification experiment on the difference in material density parameters (≤15% / ≥40%)

[0310] Experimental objective:

[0311] Distinguish between the critical difference between materials that have not been substantially replaced (refurbished) and materials that have been substantially replaced (reconstructed) to avoid misjudgment caused by changes in surface coating.

[0312] Experimental samples:

[0313] Seventy building material samples were selected, of which 35 were renovation samples (repairs with the same material or surface coating updates, without substantial material replacement) and 35 were reconstruction samples (material type replacement, such as brick → reinforced concrete).

[0314] Experimental steps:

[0315] According to the aforementioned method of the present invention, the material spectral characteristics of optical images and the echo intensity characteristics of SAR images are extracted, and the difference in material compactness parameters before and after the modification is calculated.

[0316] Five candidate thresholds were set: 10%, 15%, 20%, 30%, and 40%. The accuracy of material replacement determination under each threshold was statistically analyzed.

[0317] Filter for the optimal dual threshold.

[0318] Experimental results:

[0319] Table 5: Comparison of experimental performance under different threshold schemes for material density parameters

[0320]

[0321] Experimental conclusion:

[0322] The overall accuracy rate is highest (95.5%) when the difference is set to ≤15% (material not actually changed) and ≥40% (material actually changed). This can avoid misjudging slight parameter fluctuations in the same material renovation as material change, and can accurately identify material type replacement caused by reconstruction. Therefore, this dual threshold was determined.

[0323] Example 2, please refer to Figure 2 As shown in this embodiment, a high spatial resolution remote sensing image semantic change detection system includes:

[0324] (1) Data acquisition module, used to acquire remote sensing images of the target area;

[0325] The data acquisition module includes a multi-source data acquisition submodule and a data quality control submodule;

[0326] The multi-source data acquisition submodule includes:

[0327] Optical image acquisition unit: Supports the acquisition of high-resolution optical remote sensing data ranging from 0.5 to 2 meters;

[0328] SAR image acquisition unit: supports synchronous acquisition of multi-band synthetic aperture radar data;

[0329] Metadata Management Unit: Unifies the management of metadata such as time, location, and sensor parameters;

[0330] Data quality control submodule:

[0331] Cloud cover detection and filtering unit: automatically identifies and removes images with high cloud cover;

[0332] Imaging quality assessment unit: assesses data quality based on indicators such as signal-to-noise ratio and contrast ratio;

[0333] Time series integrity check unit: Ensures the continuity and integrity of time series data;

[0334] (2) A preprocessing module, used to preprocess the images to obtain a time-series image dataset;

[0335] The preprocessing module includes a spatiotemporal registration submodule and a radiation processing submodule;

[0336] The spatiotemporal registration submodule includes:

[0337] Automatic registration unit: Achieves accurate registration of multi-source data based on feature point matching;

[0338] Geometric correction unit: Eliminates geometric distortions caused by terrain and sensor attitude;

[0339] Resolution unification unit: Resamples multi-source data to a uniform spatial resolution;

[0340] The radiation processing submodule includes:

[0341] Atmospheric correction unit: Uses a radiative transfer model for accurate atmospheric correction;

[0342] Radiometric calibration unit: realizes the conversion of sensor digital quantization values ​​to surface reflectance;

[0343] Noise suppression unit: Eliminates image noise based on filtering algorithms;

[0344] (3) A grid division module, used to divide the target area into grid units based on the time-series image dataset;

[0345] The mesh generation module includes a building outline extraction submodule and an adaptive mesh generation submodule;

[0346] The building outline extraction submodule includes:

[0347] Deep learning segmentation unit: Accurate extraction of building outlines based on U-Net network;

[0348] Edge optimization unit: Employs morphological operations to optimize the quality of the contour boundary;

[0349] Contour verification unit: Ensures the accuracy of contour extraction through manual sampling;

[0350] The adaptive mesh generation submodule includes:

[0351] Unit size calculation unit: Automatically calculates the optimal grid size based on building functional zoning;

[0352] Mesh generation unit: Generates a regular mesh system covering the building area;

[0353] Boundary processing unit: Optimizes the mesh generation effect for irregular boundaries;

[0354] (4) Feature extraction module, used to extract structural semantic features for each grid cell and the overall region respectively;

[0355] The feature extraction module includes a structural feature analysis submodule and a multi-source feature fusion submodule;

[0356] The structural feature analysis submodule includes:

[0357] Outline continuity analysis unit: calculates the continuous proportion of the building edge outline;

[0358] Layer number change detection unit: identifies layer number changes based on pixel stacking and shadow analysis;

[0359] Spatial layout matching unit: evaluates layout consistency through feature point matching;

[0360] The multi-source feature fusion submodule includes:

[0361] Feature standardization unit: unifies the data range and distribution of different features;

[0362] Feature selection unit: Screens key features based on importance assessment;

[0363] Feature weighting unit: Assigning appropriate weight coefficients to different features;

[0364] (5) A time series analysis module, used to construct a time series change process curve based on the structural semantic features and extract change indicators;

[0365] The time series analysis module includes a change process modeling submodule and a change index quantification submodule;

[0366] The change process modeling submodule includes:

[0367] Stage identification unit: Identifying change stages based on hidden Markov models;

[0368] Time series curve construction unit: generates complete time series curves of the change process;

[0369] Inflection point detection unit: Identifies key time points in the process of change;

[0370] The change index quantification submodule includes:

[0371] Duration calculation unit: accurately quantifies the duration of each stage;

[0372] Rate of change analysis unit: calculates the intensity of change per unit time;

[0373] Trend prediction unit: Predicts changing trends based on historical data;

[0374] (6) Material analysis module, used to calculate material density parameters based on the time-series change process curve;

[0375] The material analysis module includes a multimodal feature extraction submodule and a compactness calculation submodule;

[0376] The multimodal feature extraction submodule includes:

[0377] Spectral Feature Analysis Unit: Extracts the material spectral features of optical images;

[0378] Radar Feature Analysis Unit: Analyzes the echo intensity characteristics of SAR images;

[0379] Texture feature calculation unit: quantifies the texture properties of material surfaces;

[0380] The compactness calculation submodule includes:

[0381] Feature fusion unit: Enables deep fusion of optical and radar features;

[0382] Parameter calculation unit: Calculates material density parameters based on fused features;

[0383] Change assessment unit: assesses the temporal variation pattern of material density;

[0384] (7) Semantic detection module, used to detect semantic changes in the target area based on the structural semantic features, temporal change process curve and material density parameters.

[0385] The semantic detection module includes a unit-level judgment submodule and an overall semantic integration submodule;

[0386] The unit-level decision submodule includes:

[0387] Evidence fusion unit: fusing multi-source evidence based on DS evidence theory;

[0388] Semantic classification unit: Implements semantic type determination at the grid unit level;

[0389] Confidence assessment unit: Provides a confidence score for each decision;

[0390] Overall semantic integration submodule:

[0391] Spatial distribution analysis unit: Analyzes the spatial distribution patterns of semantic changes;

[0392] Overall Judgment Unit: The overall semantic judgment is derived based on the grid statistical results;

[0393] Result optimization unit: Improves judgment accuracy through spatial consistency optimization;

[0394] System workflow:

[0395] Data input stage: The system receives multi-temporal high-resolution remote sensing image data; automatically performs data quality checks and format standardization;

[0396] Preprocessing stage: Perform spatiotemporal registration and radiometric correction; generate standardized time-series image datasets;

[0397] Grid generation phase: Extract building outlines based on pre-renovation images; generate an adaptive grid system;

[0398] Feature analysis phase: Structural feature extraction and temporal analysis are performed in parallel; material property analysis is performed simultaneously.

[0399] Semantic determination stage: Semantic determination at the grid unit level is performed based on multi-source evidence; overall semantic change conclusions are drawn through spatial integration;

[0400] Results output stage: Generates a semantic change detection report; outputs a spatial distribution map of the changed areas; provides judgment criteria and confidence assessment;

[0401] As mentioned above, the entire process from data input to result output is automated, significantly improving work efficiency and reducing manual intervention. A dedicated data fusion module fully leverages the complementary advantages of optical and SAR data, enhancing detection reliability. An intelligent judgment mechanism based on evidence theory provides a scientifically reliable basis for semantic change detection. A robust quality control system and a user-friendly interface give the system significant engineering application value. Modular design facilitates functional expansion and algorithm upgrades, adapting to future technological development needs.

[0402] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting semantic changes in high spatial resolution remote sensing images, characterized in that, include: Remote sensing images of the target area are acquired, and the images are preprocessed to obtain a time-series image dataset. The remote sensing images include images of the renovation of the facades of old residential areas and images of demolition and reconstruction. Based on the aforementioned time-series image dataset, the target region is divided into grid cells; Structural semantic features are extracted for each grid cell and the entire region. The structural semantic features include the integrity of the main structure, the consistency of the number of layers, and the matching degree of the spatial layout, wherein: The integrity of the main structure is based on extracting the edge contour of the target area from optical images, calculating the contour continuity index, and analyzing the echo intensity distribution characteristics based on synthetic aperture radar images. The continuous proportion of the main structure is calculated comprehensively, and when the continuous proportion is ≥90%, it is determined that the main structure is unchanged. The layer consistency is determined by analyzing the number of layers in the target area through the pixel stacking features in the vertical direction of the image, verifying the layer calculation results based on the relationship between shadow length and solar altitude angle, and comparing the layer difference between the corresponding areas before and after the transformation. When the layer difference is ≤0, it is determined that the number of layers has not changed. The spatial layout matching degree is calculated by extracting feature points from the images before and after the renovation, and using a feature matching algorithm to calculate the matching degree of key functional areas. When the matching score is ≥85 points, it is determined that the spatial layout has not changed. Based on the aforementioned structural semantic features, a time-series change process curve is constructed, and change indicators are extracted; Calculate the material density parameter based on the time-series change process curve; The calculation of the material density parameter includes fusing features from optical images and synthetic aperture radar images. Specifically, it includes extracting the material spectral features from the optical images, analyzing the echo intensity features from the synthetic aperture radar images, and fusing optical and radar features to calculate the material density parameter. When the difference between the material density parameter and the pre-modification value is ≤15%, it is determined that the material has not been substantially replaced. When the difference between the material density parameter and the pre-modification value is ≥40%, it is determined that the material has been substantially replaced. Based on the structural semantic features, temporal change process curves, and material density parameters, semantic change detection is performed on the target region.

2. The method for detecting semantic changes in high spatial resolution remote sensing images according to claim 1, characterized in that: The remote sensing images include high spatial resolution optical remote sensing images and synthetic aperture radar images. The preprocessing includes registration, radiometric correction, and cropping of the optical remote sensing images and synthetic aperture radar images.

3. The method for detecting semantic changes in high spatial resolution remote sensing images according to claim 1, characterized in that: The grid cell division is based on the target area in the pre-reconstruction image.

4. The method for detecting semantic changes in high spatial resolution remote sensing images according to claim 1, characterized in that: The change indicators include the change stage, duration, and rate of change. The change stage is divided into a no-change stage, a surface modification stage, a main body deconstruction stage, a main body reconstruction stage, and a stable stage. When only the no-change stage, surface modification stage, and stable stage are present, the candidate is marked as a renovation candidate. When the no-change stage, main body deconstruction stage, main body reconstruction stage, and stable stage are present, the candidate is marked as a reconstruction candidate.

5. The method for detecting semantic changes in high spatial resolution remote sensing images according to claim 1, characterized in that: The semantic change detection includes determining the semantic type of each grid cell individually, and deriving an overall semantic determination result based on the determination results of all grid cells, specifically including: The semantic type is determined for each grid cell individually. The determination results of all grid cells are statistically analyzed. When the proportion of grid cells determined to have undergone a major semantic change is ≥20% and they are continuously distributed, the overall determination is local reconstruction. When the proportion of grid cells determined to have undergone a major semantic change is <20% or they are discretely distributed, the overall determination is overall renovation.

6. A high spatial resolution remote sensing image semantic change detection system, used to implement the high spatial resolution remote sensing image semantic change detection method according to any one of claims 1-5, characterized in that, include: The data acquisition module is used to acquire remote sensing images of the target area; The preprocessing module is used to preprocess the images to obtain a time-series image dataset; The grid division module is used to divide the target area into grid cells based on the time-series image dataset; The feature extraction module is used to extract structural semantic features for each grid cell and the entire region respectively; The time-series analysis module is used to construct a time-series change process curve based on the structural semantic features and extract change indicators; The material analysis module is used to calculate the material density parameters based on the time-series change process curve; The semantic detection module is used to detect semantic changes in the target region based on the structural semantic features, the temporal change process curve, and the material density parameters.

Images