Construction land change dynamic detection system based on remote sensing image

Abstract

This invention relates to the field of image feature analysis technology, specifically to a dynamic detection system for changes in construction land based on remote sensing imagery. The system acquires and registers temporal multispectral images of the target area; extracts the spectral, texture, and shape features of each building patch in parallel to construct a feature sequence; constructs a construction stage template feature sequence, calculates the similarity score between each patch and each construction stage, and obtains the construction stage evaluation result for each patch at each acquisition time based on the similarity score, generating a construction stage time-series map and a feature change trend map for each patch. This invention effectively solves the problem that traditional methods cannot obtain the dynamic evolution law of the construction process during dynamic detection of changes in construction land, thereby improving the diversity of dynamic detection.

Description

Technical Field

[0001] This invention relates to the field of image feature analysis technology, specifically to a dynamic detection system for changes in construction land based on remote sensing images. Background Technology

[0002] With the accelerating pace of global urbanization, effective supervision of construction land has become a core task for urban planning, land resource management, and law enforcement. Remote sensing technology, with its advantages of macroscopic, objective, and periodic observation, has become a primary technical means for dynamic monitoring of construction land.

[0003] Currently, traditional monitoring methods for construction land based on remote sensing images, such as image difference method, ratio method, principal component analysis method, change vector analysis method, and change detection model based on deep learning, all identify areas where the surface state has changed significantly by comparing remote sensing images from two or more different time periods. However, these methods are limited by not distinguishing the specific construction stage of the construction land, and therefore cannot obtain the dynamic evolution pattern of the construction process. As a result, the detection results lack engineering semantics and are difficult to use directly for progress assessment and refined management. Summary of the Invention

[0004] To address the technical problem that traditional methods for detecting construction land cannot capture the dynamic evolution patterns of the construction process, resulting in a lack of diverse detection data, the present invention aims to provide a dynamic detection system for construction land changes based on remote sensing imagery. The specific technical solution adopted is as follows: This invention proposes a dynamic detection system for changes in construction land based on remote sensing imagery, the system comprising: Data acquisition and preprocessing module 101: used to acquire multispectral images of the target area at continuous acquisition times within a preset acquisition period, where each construction site is represented as a patch in the multispectral image; Feature extraction module 102: used to obtain the spectral features of each patch based on the surface reflectance characteristics of each patch in different bands; to obtain the texture features of each patch based on the grayscale features of each patch in the near-infrared band; to obtain the shape features of the construction area in each patch based on the number of pixels belonging to the construction area in each patch; and to construct a feature sequence for each patch based on the spectral features, texture features, and shape features. Template library management module 103: For each construction stage, a template feature sequence is constructed from the sample feature sequences of the already labeled construction stages, so that the sum of the squares of the similarity distances from the template feature sequence to each sample feature sequence is minimized; The temporal matching analysis module 104 is used to calculate the similarity score between each patch and each construction stage based on the similarity distance relationship between the feature sequence of each patch and the feature sequence of each template in the multispectral image, and to obtain the construction stage evaluation result of each patch at each acquisition time based on the similarity score. Result Integration and Visualization Module 105: Based on the construction phase evaluation results of each patch at each acquisition time, it obtains the construction phase transition point and constructs the construction phase time sequence diagram and feature change trend diagram for each patch.

[0005] Furthermore, the system also includes an image registration module, which is used for: Using the multispectral image with the least cloud cover as a benchmark, a registration method based on SIFT feature points is adopted to spatially register other multispectral images by extracting control points.

[0006] Furthermore, the method for obtaining the spectral features of each patch includes: Based on the surface reflectance of each pixel in different bands, the normalized vegetation index (NDI) and normalized building index (NDI) of each pixel are calculated. For each multispectral image, the average of the NDI of all pixels within each patch is calculated as the NDI feature value of that patch; the average of the NDI of all pixels within each patch is calculated as the NDI feature value of that patch; and the NDI feature values ​​are used as the spectral features of each patch.

[0007] Furthermore, the method for obtaining the texture features of each patch includes: A gray-level co-occurrence matrix is ​​constructed based on the near-infrared band image of each patch. The entropy and homogeneity of each patch are calculated based on the probability of different gray levels co-occurring under different spatial relationships. The entropy and homogeneity are used as the texture features of each patch.

[0008] Furthermore, the method for obtaining the shape features of each patch includes: Based on the normalized building index of all pixels, a global threshold is set, and pixels with a normalized building index greater than the global threshold are identified as building region pixels. For each patch, the number of pixels belonging to the building region is counted, and the area of ​​the building region in each patch is calculated using the image spatial resolution. Edge detection is performed on the building region to extract the contour boundary, the convex hull of the building region contour is calculated, and a bounding rectangle with each side of the convex hull parallel is constructed. All bounding rectangles are traversed, and the aspect ratio of the bounding rectangle with the smallest area is recorded. The area of ​​the building region and the aspect ratio are used as the shape features of each patch.

[0009] Furthermore, the method for obtaining the similarity distance relationship includes: For each feature sequence obtained at each acquisition time for each patch, they are arranged in chronological order to construct a temporal feature sequence. The sliding window method is then used to obtain the similarity distance relationship between the feature sequence of each window and the template feature sequence.

[0010] Furthermore, the similarity distance calculation method includes: The similarity distance between two feature sequences with different time lengths is calculated using the dynamic time warping algorithm.

[0011] Furthermore, the similarity score calculation method includes: The similarity distance is negatively correlated and normalized to obtain the similarity score.

[0012] Furthermore, the construction phase assessment method includes: For each real-time data acquisition moment, multiple sliding windows will cover the construction phase evaluation process. Each sliding window has a construction phase selection result. Using a majority voting mechanism, the construction phase selection result with the most votes will be selected. If there is a tie, the construction phase that is the same as the previous data acquisition moment will be selected as the construction phase for the real-time data acquisition moment.

[0013] Furthermore, the method for obtaining the construction phase transition point includes: Based on the aforementioned construction phase assessment results, the first data collection moment when a change occurs during the phase is extracted as the transition point of the construction phase.

[0014] The present invention has the following beneficial effects: The system enhances its ability to represent and distinguish the complex dynamics of construction activities by constructing a feature sequence that integrates spectrum, texture, and morphology, and through multi-dimensional information complementarity. Furthermore, by calculating the similarity between the template feature sequence of each construction stage and the feature sequences of each patch at various periods, the specific construction stage is determined. This represents a leap from traditional detection methods that focus on changes and inconsistencies to dynamic identification of continuous construction processes at specific stages. Ultimately, the system outputs a construction stage time series map and a feature change trend map, effectively solving the problem that traditional methods cannot capture the dynamic evolution patterns of the construction process during dynamic detection of changes in construction land, thereby improving the diversity of dynamic detection. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a structural diagram of a dynamic detection system for changes in construction land based on remote sensing imagery, provided as an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a dynamic detection system for changes in construction land based on remote sensing imagery proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific scheme for a dynamic detection system for changes in construction land based on remote sensing imagery provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a system structure block diagram of a dynamic detection system for changes in construction land based on remote sensing imagery, according to an embodiment of the present invention. The system includes: Data acquisition and preprocessing module 101: Used to acquire multispectral images of the target area at continuous acquisition times within a preset acquisition period, with each construction site in the multispectral image being a map patch.

[0021] In order to achieve dynamic monitoring of changes in construction land, each target area contains multiple construction land areas. When conducting dynamic monitoring, it is necessary to divide each construction land area into separate regions and analyze the dynamic change process of each construction land area separately, thereby achieving dynamic monitoring.

[0022] Because construction land is phased throughout the construction process, the duration of each phase often varies depending on the building type. For example, the initial construction period differs significantly between shopping malls and residential buildings, and the same construction period often lasts for several months. During construction, the vegetation and buildings on the construction land change dramatically, exhibiting distinct characteristics in visible light, near-infrared, and short-wave infrared images. Considering these factors, this embodiment of the invention uses a Landsat sensor to acquire multispectral images of the target area in visible light, near-infrared, and short-wave infrared bands. Finally, based on the national land survey standard building land vector map of the target area, a map patch is generated for each construction land in the multispectral image of the target area, and analysis is performed on a map patch basis.

[0023] In one specific implementation of this invention, the data collection period is set to 10 years, and the collection time is the 1st of each month. For ease of subsequent analysis, the collection time will be represented by the month in the following description of the specific implementation method.

[0024] In one specific implementation of this invention, when performing radiometric calibration and atmospheric correction, the radiance value is converted into surface reflectance by combining the sensor calibration coefficient, and atmospheric correction is performed using the 6S radiative transfer model.

[0025] Feature extraction module 102: used to obtain the spectral features of each patch based on the surface reflectance features of each patch in different bands; to obtain the texture features of each patch based on the grayscale features of each patch in the near-infrared band; to obtain the shape features of the construction area in each patch based on the number of pixels belonging to the construction area in each patch; and to construct a feature sequence for each patch based on the obtained spectral features, texture features and shape features.

[0026] Because the surface reflectance characteristics of construction land in different wavelength bands can reflect changes in vegetation and buildings in the construction area; the near-infrared band significantly reflects changes in surface structure, reflecting changes in surface structure throughout the construction process; the shape of buildings in each patch changes continuously with the construction process, and the relationship between pixels within each patch reflects these changes in shape. In this embodiment, spectral features are extracted by analyzing the surface reflectance of all pixels within the patch at different wavelength bands for each patch; texture features are extracted by analyzing the grayscale features of the patch in the near-infrared band; and structural features are extracted by analyzing the pixel relationships within the patch. Finally, a feature sequence is constructed for each patch based on the obtained spectral, texture, and shape features. The feature sequences extracted from each patch at each acquisition time are arranged chronologically to construct a temporal feature sequence for each patch.

[0027] Template library management module 103: For each construction stage, it constructs a template feature sequence from the sample feature sequences of the already labeled construction stages, so as to minimize the sum of the squares of the similarity distances between the template feature sequence and each sample feature sequence.

[0028] During the evaluation of the construction stage of each map patch at each acquisition time by the time-series matching analysis module 104, it is necessary to combine the feature sequence of each map patch at each acquisition time for discrimination. Therefore, it is necessary to extract the template feature sequence that can accurately represent the construction land from the vacant land period, the initial construction period, the main construction period to the completed period from the sample of the already marked construction stage based on the feature sequence extracted by the feature extraction module 102.

[0029] The characteristic sequences corresponding to samples at different construction stages are typically as follows: Open land stage: The surface remains in a natural state, the six-dimensional characteristic values ​​are relatively stable, and the NDBI value is low and fluctuates little; Early construction stage: NDBI begins to rise significantly, the entropy value increases significantly, the dynamic area begins to grow, and homogeneity decreases; Main construction stage: NDBI remains at a high level, the dynamic area continues to expand rapidly, and the morphological characteristics tend to the design value; Completed stage: All characteristic values ​​tend to stabilize, NDBI remains at a stable high level, and the dynamic area reaches its peak and then stabilizes.

[0030] In this embodiment of the invention, over 200 building map patches with complete and clear construction stages are selected as training samples. The samples must cover different geographical regions, climate zones, and major building types (such as residential, industrial, and commercial). Using the feature extraction module 102, feature values ​​from the above training sample map patches are obtained, and each feature value is standardized to construct a feature sequence. For each construction stage, taking the vacant land stage as an example: Given n sample sequences during the empty period ,in It is a given open space period Given a set of sample feature arrays, construct a template feature sequence such that the sum of the squared similarity distances from the template feature sequence to each sample feature array is minimized. Similarly, the template feature sequences for the other three construction stages can be calculated.

[0031] In one specific implementation of this invention, the template feature sequence is used. Assign initial values ​​and perform iterative optimization. Specifically, from the sample sequence... Randomly select a sample sequence Using this sequence as the initial value of the template feature sequence, calculate... to sample sequence The similarity distance of each sample feature array is used for iterative optimization to improve the template feature sequence. The goal is to minimize the sum of squared similarity distances to each sample's feature array. The specific iterative algorithm is a technique well-known to those skilled in the art and will not be elaborated or limited here.

[0032] In one specific implementation of this invention, the Z-score algorithm is used to standardize each extracted feature. Other normalization methods, such as Min-Max normalization and mean normalization, may also be used in other implementations of this invention, and are not limited or elaborated upon here.

[0033] The temporal matching analysis module 104 is used to calculate the similarity score between each patch and each construction stage based on the similarity distance relationship between the feature sequence of each patch and the feature sequence of each template in the multispectral image, and to obtain the construction stage evaluation result of each patch at each acquisition time based on the similarity score.

[0034] In this embodiment of the invention, the feature sequence extracted from each patch for each month is compared with the template feature sequence for each construction stage. By analyzing the similarity score, the higher the similarity score, the greater the similarity between the two. The highest similarity score is determined as the most likely construction stage for each patch for each month.

[0035] Result Integration and Visualization Module 105: Based on the construction phase evaluation results of each patch at each acquisition time, it obtains the construction phase transition point and constructs the construction phase time sequence diagram and feature change trend diagram for each patch.

[0036] Since each land parcel is of a different type, such as residential, industrial, or commercial land, the duration of each construction phase is also different. Based on the construction phase assessment results of each land parcel at each collection time, the construction phase transition point can be obtained. At the same time, by combining the feature sequence at each collection time, the dynamic detection results of construction land changes in the study area can be analyzed.

[0037] In this embodiment of the invention, a construction phase time series diagram is constructed using the construction phase assessment results and construction phase transition points for each map patch containing 120 months of data. The horizontal axis represents 120 consecutive monthly time points, labeled in "year-month" format; the vertical axis represents four discrete construction phases (vacant land phase, initial construction phase, main construction phase, and completion phase), each represented by a different color, with phase transition points marked by vertical dashed lines. Using the feature extraction module 102, a six-dimensional feature sequence for each map patch containing 120 months is obtained, constructing six parallel sub-maps corresponding to the six-dimensional features. These sub-maps share the same horizontal axis (120 consecutive monthly time points, labeled in "year-month" format) and six different feature values ​​on the vertical axis.

[0038] In summary, by constructing a feature sequence that integrates spectrum, texture, and morphology, and through multi-dimensional information complementarity, the system enhances its ability to represent and distinguish the complex dynamics of construction activities. Furthermore, by calculating the similarity between the template feature sequence of each construction stage and the feature sequences of each patch at various periods, the specific construction stage is determined. This represents a leap from traditional detection methods that focus on changes and inconsistencies to dynamic identification of continuous construction processes at specific stages. Finally, the system's output of construction stage time-series diagrams and feature change trend diagrams effectively solves the problem that traditional methods cannot capture the dynamic evolution patterns of the construction process during dynamic detection of changes in construction land, thereby improving the diversity of dynamic detection.

[0039] Preferably, in some possible implementations of the embodiments of the present invention, considering the interference of cloud cover on subsequent feature extraction, the dynamic detection system for changes in construction land based on remote sensing images in the embodiments of the present invention further includes image registration and an image registration module. The image registration module is used to: take the multispectral image with the least cloud cover as a reference, adopt the registration method based on SIFT feature points, and achieve spatial registration of other multispectral images by extracting control points.

[0040] Areas covered by clouds cannot yield effective surface features (such as building edges, road intersections, and field boundaries). Clouds, cloud shadows, and fog can distort or obscure the true spectral and textural features of the land surface. Images with the least cloud cover expose the largest surface area, providing the densest and most comprehensive set of feature points for feature-based registration algorithms like SIFT. The SIFT algorithm used for multispectral registration is a well-known technique in the field and will not be elaborated upon or limited here.

[0041] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the spectral features of each patch includes: Based on the surface reflectance of each pixel in different bands, the normalized vegetation index and normalized building index of each pixel are calculated. For each multispectral image, since there are multiple pixels within a patch, the average normalized vegetation index of all pixels within each patch is calculated as the normalized vegetation index feature value of that patch; the average normalized building index of all pixels within each patch is calculated as the normalized building index feature value of that patch.

[0042] Therefore, the normalized vegetation index (NDI) and normalized building index (NDI) feature values ​​are used as the spectral features of each patch.

[0043] As a concrete example, taking each patch as the basic unit, the formulas for calculating the eigenvalues ​​of all pixels within it—the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Building Index (NDBI)—can be expressed as follows: NIR represents the surface reflectance of the target pixel in the near-infrared band; RED represents the surface reflectance of the target pixel in the red band; NDVI is used to represent changes in vegetation cover. During construction, vegetation removal will cause NDVI to drop sharply, while green restoration after completion may cause it to rise slightly.

[0044] NIR represents the surface reflectance of a target pixel in the near-infrared band; SWIR represents the surface reflectance of a target pixel in the short-wave infrared band; and NDBI is used to represent the surface features of buildings. After construction begins, the exposure of building materials causes a significant increase in NDBI, which is a key indicator for distinguishing construction stages.

[0045] It should be noted that the NDVI and NDBI parameters mentioned above are technical means well known to those skilled in the art, and the specific meanings of the formulas will not be elaborated here.

[0046] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the texture features of each patch includes: A gray-level co-occurrence matrix is ​​constructed based on the near-infrared band image of each patch. The entropy and homogeneity of each patch are calculated based on the probability of different gray levels co-occurring under different spatial relationships. The entropy and homogeneity are used as the texture features of each patch.

[0047] It should be noted that, in one specific implementation of this invention, the specific parameter settings of the gray-level co-occurrence matrix include: quantizing the pixel values ​​of the original image to 16 levels to achieve a balance between computational efficiency and texture detail; and setting the movement step size to 1 pixel. Considering the spatial anisotropy of the building structure, in order to capture the texture pattern of the dominant direction, the horizontal direction is selected as the computation direction in this embodiment; and a 3×3 neighborhood window is taken as the center of each pixel for local texture calculation to balance detail representation and noise suppression.

[0048] Entropy measures the average number of bits of information needed to describe the texture of an image. During the initial construction phase, the surface is usually covered with vegetation or bare soil, resulting in relatively simple and homogeneous textures and low calculated entropy values. In the early and main construction phases, the surface is severely disturbed, filled with machinery, building materials, potholes, etc., causing the image texture to become extremely complex and disordered, the probability distribution of the GLCM matrix to become dispersed, and the entropy value to increase significantly. During the completed phase, as buildings and roads form regular structures and the surface material tends to be uniform, the entropy value drops from its peak and remains at a moderate level.

[0049] Homogeneity measures the uniformity of local areas in an image. During the initial stage of construction, the surface is covered with vegetation or flat land, resulting in relatively uniform texture and small differences in grayscale between pixels, leading to high homogeneity. During the initial and main construction phases, the surface depressions and the accumulation of building materials cause the texture to become rough and uneven, significantly reducing homogeneity. During the completed phase, the building surface has uniform materials and a regular structure, and the texture becomes uniform again, with homogeneity recovering from its low point. However, due to the introduction of texture by windows and shadows, it is usually impossible to return to the level of the initial stage.

[0050] Therefore, entropy and homogeneity are used as texture features for each patch.

[0051] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the shape feature of each patch includes: Based on the normalized building index of all pixels, a global threshold is set, and pixels with a normalized building index greater than the global threshold are identified as building region pixels. For each patch, the number of pixels belonging to the building region is counted, and the area of ​​the building region in each patch is calculated using the image spatial resolution. Edge detection is performed on the building region to extract the contour boundary, the convex hull of the building region contour is calculated, and the circumscribed rectangle with each side of the convex hull is constructed. All circumscribed rectangles are traversed, and the aspect ratio of the circumscribed rectangle with the smallest area is recorded. The area of ​​the building region and the aspect ratio of the smallest circumscribed rectangle are used as the shape features of each patch.

[0052] It should be noted that when the number of pixels belonging to the building area is 0, it indicates that the area is completely empty and there may be no building area inside the patch. Therefore, the area and aspect ratio of the building area are set to 0.

[0053] In one specific implementation of this invention, the optimal threshold T is calculated using Otsu's method for the NDBI values ​​of all pixels within all patches. Pixels with NDBI ≥ T are identified as building areas, and the number of pixels is counted as N. The area of ​​the building area within each patch is then determined. The calculation formula is: Where R represents the spatial resolution of the image. Throughout the construction process, the area of ​​the building area rapidly increases and then tends to stabilize, so area can be used as an important characteristic value for shape features.

[0054] For each feature: the Canny edge detection algorithm is used to extract the outline boundary of the building area, and then the rotating caliper algorithm is used to calculate the convex hull of the building area outline. The outer rectangles with parallel edges of the convex hull are constructed. All outer rectangles are traversed, and the aspect ratio of the outer rectangle with the smallest area is recorded. During the entire construction process, the aspect ratio will show a trend from random dispersion at the beginning to regular fluctuations, and finally gradually converge to the design value. Therefore, the aspect ratio can be used as an important feature value of shape characteristics.

[0055] Based on the above description, the area of ​​the construction region and the aspect ratio of the minimum bounding rectangle are used as the shape features of each patch.

[0056] It should be noted that, in one specific implementation of this invention, spectral features, texture features, and structural features each contain two types of feature values. Therefore, each patch should correspond to six feature values ​​at each acquisition time, and the feature array of each patch at each acquisition time is a six-dimensional array, with each feature value corresponding to one array element. Finally, the feature arrays obtained for each patch at all acquisition times are arranged in chronological order to construct a temporal feature sequence for each patch.

[0057] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the similarity distance relationship includes: For each feature sequence obtained at each acquisition time for each patch, arrange them in chronological order to construct a temporal feature sequence. Then, use the sliding window method to obtain the similarity distance relationship between the feature sequence of each window and the template feature sequence.

[0058] Since each patch yields a feature sequence at each acquisition time, reflecting its features at that time, the feature sequences of the same patch at different acquisition times are arranged chronologically to construct a temporal feature sequence for the patch over the entire preset acquisition period. A sliding window method is then used on this temporal feature sequence to obtain the similarity distance between the feature sequence of each window and the template feature sequence. In a specific implementation of this invention, the window length is set to l = 12 months, the sliding step size is k = 1 month, and the total number of windows is m = 120 - l + 1 = 109. For the feature sequence of the i-th sliding window of each patch... Defined as: in, ; For each patch The characteristic sequence of a month.

[0059] Preferably, in some possible implementations of the embodiments of the present invention, the similarity distance calculation method includes: The similarity distance between two feature sequences with different time lengths is calculated using a dynamic time warping algorithm. When calculating the similarity distance, the lengths of the two feature sequences may differ. For example, in the template library management module, the feature sequences of two samples may have different lengths due to different building types. Similarly, in the time series matching analysis module 104, when using the sliding window method, the length of each window's feature sequence is different from the template's feature sequence length. If traditional Euclidean distance is used to calculate the similarity distance between two feature sequences with different lengths, interpolation or pruning of the feature sequences is necessary, which can easily introduce false data and affect the authenticity of the original feature sequences. The dynamic time warping algorithm, however, allows for non-linear stretching or compression of sequences along the time axis to find the best morphological alignment, fully compatible with the comparison of sequences of unequal lengths. Therefore, in this embodiment of the invention, the dynamic time warping algorithm is used to calculate the similarity distance between two feature sequences with different time lengths, eliminating the influence of time length.

[0060] It should be noted that the dynamic time warping algorithm is a well-known technique in the field, and its specific details will not be elaborated further.

[0061] Preferably, in some possible implementations of the embodiments of the present invention, the similarity score calculation method includes: performing a negative correlation mapping on the similarity distance and normalizing it to obtain the similarity score. The specific formula is expressed as: in, A value between 0 and 1 indicates a high degree of similarity between the two values; the closer the value is to 1, the higher the similarity between the two values. The similarity distance between the two sequences is given; the positive integer 1 in the denominator is used to avoid a denominator of 0.

[0062] Preferably, in some possible implementations of the embodiments of the present invention, the construction phase evaluation method includes: For each real-time data acquisition moment, multiple sliding windows will cover the construction phase evaluation process. Each sliding window has a construction phase selection result. Using a majority voting mechanism, the construction phase selection result with the most votes will be selected. If there is a tie, the construction phase that is the same as the previous data acquisition moment will be selected as the construction phase for the real-time data acquisition moment.

[0063] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the construction phase transition point includes: Based on the construction phase evaluation results of each map patch at different acquisition times, the first acquisition time at which the phase changes is extracted as the construction phase transition point. As the boundary between two construction phases, the construction phase transition point indicates, to some extent, the duration of each map patch in each construction phase, which is beneficial for quantitative analysis.

[0064] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A dynamic detection system for changes in construction land based on remote sensing imagery, characterized in that, The system includes: Data acquisition and preprocessing module: used to acquire multispectral images of the target area at continuous acquisition times within a preset acquisition period, with each construction site in the multispectral image being a single image patch; Feature extraction module: used to obtain the spectral features of each patch based on the surface reflectance characteristics of each patch in different bands; to obtain the texture features of each patch based on the grayscale features of each patch in the near-infrared band; to obtain the shape features of the construction area in each patch based on the number of pixels belonging to the construction area in each patch; and to construct a feature sequence for each patch based on the spectral features, texture features, and shape features. Template library management module: For each construction stage, it constructs a template feature sequence from the sample feature sequences of the already labeled construction stages, so as to minimize the sum of squared similarity distances between the template feature sequence and each sample feature sequence; The temporal matching analysis module is used to calculate the similarity score between each patch and each construction stage based on the similarity distance relationship between the feature sequence of each patch and the feature sequence of each template in the multispectral image, and to obtain the construction stage evaluation result of each patch at each acquisition time based on the similarity score. The results integration and visualization module is used to obtain the construction stage transition points based on the construction stage evaluation results of each patch at each acquisition time, and to construct the construction stage time sequence diagram and feature change trend diagram for each patch.

2. The construction land change dynamic detection system based on remote sensing imagery according to claim 1, characterized in that, The system also includes an image registration module, which is used for: Using the multispectral image with the least cloud cover as a benchmark, a registration method based on SIFT feature points is adopted to spatially register other multispectral images by extracting control points.

3. The construction land change dynamic detection system based on remote sensing imagery according to claim 1, characterized in that, The method for obtaining the spectral features of each patch includes: Based on the surface reflectance of each pixel in different bands, the normalized vegetation index (NDI) and normalized building index (NDI) of each pixel are calculated. For each multispectral image, the average of the NDI of all pixels within each patch is calculated as the NDI feature value of that patch; the average of the NDI of all pixels within each patch is calculated as the NDI feature value of that patch; and the NDI feature values ​​are used as the spectral features of each patch.

4. The construction land change dynamic detection system based on remote sensing imagery according to claim 1, characterized in that, The method for obtaining the texture features of each patch includes: A gray-level co-occurrence matrix is ​​constructed based on the near-infrared band image of each patch. The entropy and homogeneity of each patch are calculated based on the probability of different gray levels co-occurring under different spatial relationships. The entropy and homogeneity are used as the texture features of each patch.

5. A dynamic detection system for changes in construction land based on remote sensing imagery according to claim 3, characterized in that, The method for obtaining the shape features of each patch includes: Based on the normalized building index of all pixels, a global threshold is set, and pixels with a normalized building index greater than the global threshold are identified as building region pixels. For each patch, the number of pixels belonging to the building region is counted, and the area of ​​the building region in each patch is calculated using the image spatial resolution. Edge detection is performed on the building region to extract the contour boundary, the convex hull of the building region contour is calculated, and a bounding rectangle with each side of the convex hull parallel is constructed. All bounding rectangles are traversed, and the aspect ratio of the bounding rectangle with the smallest area is recorded. The area of ​​the building region and the aspect ratio are used as the shape features of each patch.

6. The construction land change dynamic detection system based on remote sensing imagery according to claim 1, characterized in that, The method for obtaining the similarity distance relationship includes: For each feature sequence obtained at each acquisition time for each patch, they are arranged in chronological order to construct a temporal feature sequence. The sliding window method is then used to obtain the similarity distance relationship between the feature sequence of each window and the template feature sequence.

7. A dynamic detection system for changes in construction land based on remote sensing imagery according to claim 6, characterized in that, The similarity distance calculation method includes: The similarity distance between two feature sequences with different time lengths is calculated using the dynamic time warping algorithm.

8. A dynamic detection system for changes in construction land based on remote sensing imagery according to claim 7, characterized in that, The similarity score calculation method includes: The similarity distance is negatively correlated and normalized to obtain the similarity score.

9. A dynamic detection system for changes in construction land based on remote sensing imagery according to claim 6, characterized in that, The construction phase assessment methods include: For each real-time data acquisition moment, multiple sliding windows will cover the construction phase evaluation process. Each sliding window has a construction phase selection result. Using a majority voting mechanism, the construction phase selection result with the most votes will be selected. If there is a tie, the construction phase that is the same as the previous data acquisition moment will be selected as the construction phase for the real-time data acquisition moment.

10. A dynamic detection system for changes in construction land based on remote sensing imagery according to claim 1, characterized in that, The method for obtaining the construction phase transition point includes: Based on the aforementioned construction phase assessment results, the first data collection moment when a change occurs during the phase is extracted as the transition point of the construction phase.

Images