An Impervious Surface Mapping Method, System, Device, and Medium

Abstract

The present invention discloses a method, system, device and medium for impervious surface mapping, which relates to the field of geospatial information technology. The method includes: obtaining a target remote sensing image; extracting gray-level co-occurrence matrix features from the target remote sensing image to obtain first texture features; performing Gabor wavelet transform processing on the target remote sensing image to obtain second texture features; performing Gabor wavelet transform processing on the first texture features to obtain third texture features; determining a three-dimensional texture feature space image according to the first texture features, second texture features and third texture features; classifying the three-dimensional texture feature space image by using a clustering algorithm to obtain the distribution results of pervious surfaces and impervious surfaces; and processing the target remote sensing image according to the distribution results to obtain an impervious surface distribution map. The present invention can realize fast, accurate and automatic extraction of impervious surfaces, which has significant significance for studying the degree of urbanization and its development intensity.

Description

Technical Field

[0001] The present invention relates to the field of geospatial information technology, and particularly to a method, system, device and medium for mapping impervious surfaces. Background Art

[0002] With the acceleration of the urbanization process, land use types such as farmland, vegetation, and water areas within cities have gradually been replaced by artificial surfaces. As the most prominent artificial surface feature of urbanization, impervious surfaces have become the most intuitive indicator for studying the degree and development intensity of urbanization. Impervious surfaces refer to the earth's surface that water cannot penetrate, mainly including roads of various materials such as asphalt, concrete, stones, bricks, and artificial building areas such as building roofs, airports, squares, and parking lots. The large-scale expansion of impervious surfaces has affected the material cycle of the urban ecosystem, thus triggering a series of ecological problems.

[0003] In recent decades, remote sensing technology has been widely used to monitor the earth. In addition, the spatial resolution of remote sensing plays an important role in mapping impervious surfaces. Due to the advantages of medium resolution, such as relatively low acquisition cost, fine data structure, wide coverage, and moderate resolution. Therefore, medium spatial resolution images are the most commonly used scale for mapping impervious surfaces.

[0004] In addition to the influence of remote sensing image resolution, different extraction methods will also affect the results of impervious surfaces. In recent years, scholars at home and abroad have proposed a large number of remote sensing extraction methods for impervious surfaces, including linear spectral mixture analysis (LSMA), regression models, object-based classification methods, artificial neural networks (ANN), deep learning, and spectral index methods, etc.

[0005] LSMA can effectively address the impact of mixed pixels on impervious surface cover extraction. However, due to the complexity of land cover types, the selection of endmembers is more subjective. Regression models can predict and regress continuous variables, are simple to implement, and are highly efficient. However, regression models involve a manual classification process, and the accuracy of estimates can be significantly reduced if there is significant noise in the training samples. Object-based classification methods are limited by image illumination variations, noise, and shadows. The most serious problem is the lack of a universal segmentation algorithm, which greatly hinders the generalization of these methods. Deep learning and artificial neural networks have the ability to learn the essential characteristics of remote sensing datasets, handle nonlinear relationships, and integrate other data and features as well as prior knowledge. However, their accuracy depends on the quality of the training data input to the model. Compared to these algorithms, spectral index methods offer several advantages, including low computational cost, high efficiency, and sensitive spectral response. They can identify impervious surfaces without requiring additional parameters or training samples. However, spectral index methods are susceptible to mixed pixels and seasonal transitions. Due to differences in spectral reflectance, the reflectance values of different bands for the same land cover in the same season can vary. Mixed pixels can easily be misclassified, which in turn affects the accuracy of impervious surface mapping. Except for the spectral index method, the above methods all require setting parameters, rules and providing training samples, which makes it difficult to achieve automatic mapping of impervious surfaces.

[0006] However, spectral index methods are based on the reflectance of land cover. Land cover is diverse and constantly changing. Different types of land cover are intertwined and exhibit nearly homogeneous spectral characteristics. Furthermore, the same type of land cover often exhibits heterogeneity in different scenes. In contrast, texture features can reflect the homogeneous visual characteristics in remote sensing data images and can well reflect the spatial distribution of image grayscale. Texture features incorporate the surface information of remote sensing data images and their relationship to the surrounding environment, effectively accounting for the macroscopic and microscopic structures of the image.

[0007] In summary, there is an urgent need to provide an impervious surface mapping method based on texture features to achieve fast, accurate and automatic extraction of impervious surfaces. Summary of the Invention

[0008] The purpose of the present invention is to provide an impervious surface mapping method, system, equipment and medium to achieve fast, accurate and automatic extraction of impervious surfaces.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] An impervious surface mapping method, comprising:

[0011] Acquire target remote sensing images;

[0012] Extract the gray-level co-occurrence matrix features from the target remote sensing image to obtain the first texture feature;

[0013] Perform Gabor wavelet transform on the target remote sensing image to obtain the second texture feature;

[0014] Perform Gabor wavelet transform on the first texture feature to obtain the third texture feature;

[0015] Determine the three-dimensional texture feature space image according to the first texture feature, the second texture feature and the third texture feature;

[0016] Use the clustering algorithm to classify the three-dimensional texture feature space image to obtain the distribution results of the water-permeable surface and the non-water-permeable surface;

[0017] Process the target remote sensing image according to the distribution results to obtain the non-water-permeable surface distribution map.

[0018] Optionally, the first texture feature includes at least one of the contrast texture feature, the dissimilarity texture feature, the homogeneity texture feature, the angular second moment texture feature, the entropy texture feature, the variance texture feature and the correlation texture feature.

[0019] Optionally, determining the three-dimensional texture feature space image according to the first texture feature, the second texture feature and the third texture feature specifically includes:

[0020] Overlay the target remote sensing images processed by the first texture feature, the second texture feature and the third texture feature, and assign the three color channels of red, green and blue to obtain the three-dimensional texture feature space image.

[0021] Optionally, using the clustering algorithm to classify the three-dimensional texture feature space image to obtain the distribution results of the water-permeable surface and the non-water-permeable surface specifically includes:

[0022] Use the clustering algorithm to classify the three-dimensional texture feature space image to obtain a number of water-permeable surface sub-images and a number of non-water-permeable surface sub-images;

[0023] Merge all the water-permeable surface sub-images or all the non-water-permeable surface sub-images to obtain the distribution results of the water-permeable surface and the non-water-permeable surface.

[0024] Optionally, the clustering algorithm is the K-means unsupervised clustering algorithm.

[0025] Optionally, the number of clusters is 4, the change threshold is 5%, and the maximum number of iterations is 25; each cluster is: non-water-permeable surface, forest, water body and bare land, where forest, water body and bare land are all water-permeable surfaces.

[0026] Optionally, the target remote sensing image is a Sentinel-2 multispectral image.

[0027] An impervious surface mapping system, comprising:

[0028] An acquisition module, configured to acquire a target remote sensing image;

[0029] A first feature extraction module, configured to perform gray-level co-occurrence matrix feature extraction on the target remote sensing image to obtain a first texture feature;

[0030] A second feature extraction module, configured to perform Gabor wavelet transform processing on the target remote sensing image to obtain a second texture feature;

[0031] A third feature extraction module, configured to perform Gabor wavelet transform processing on the first texture feature to obtain a third texture feature;

[0032] A three-dimensional texture feature determination module, configured to determine a three-dimensional texture feature space image according to the first texture feature, the second texture feature, and the third texture feature;

[0033] A classification module, configured to classify the three-dimensional texture feature space image by using a clustering algorithm to obtain a distribution result of pervious surfaces and impervious surfaces;

[0034] A mapping module, configured to process the target remote sensing image according to the distribution result to obtain an impervious surface distribution map.

[0035] An electronic device, comprising a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned impervious surface mapping method.

[0036] A computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the above-mentioned impervious surface mapping method is implemented.

[0037] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0038] The impervious surface mapping method provided by the present invention uses a gray-level co-occurrence matrix and Gabor wavelet transform to extract the texture features of different ground objects, combines them, constructs a three-dimensional texture feature space, screens stable texture feature indicators for impervious surfaces, removes the influence of bare land, cultivated land, water bodies, and shadows on the extraction of impervious surfaces, and finally uses a clustering algorithm for classification, realizing the rapid, accurate, and automatic extraction of impervious surfaces. Description of the Drawings

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0040] Figure 1 It is a flowchart of the impervious surface mapping method provided by the present invention;

[0041] Figure 2 It is a module diagram of the impervious surface mapping system provided by the present invention.

[0042] Symbol description:

[0043] Acquisition module - 1, First feature extraction module - 2, Second feature extraction module - 3, Third feature extraction module - 4, Three - dimensional texture feature determination module - 5, Classification module - 6, Mapping module - 7. Specific embodiments

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0045] The purpose of the present invention is to provide an impervious surface mapping method, system, device and medium to achieve fast, accurate and automatic extraction of impervious surfaces.

[0046] To make the above - mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0047] Embodiment 1

[0048] The embodiment of the present invention provides an impervious surface mapping method. As Figure 1 shown, the impervious surface mapping method includes:

[0049] Step S1: Obtain the target remote - sensing image.

[0050] Preferably, the target remote - sensing image is a Sentinel - 2 multispectral image.

[0051] Step S2: Extract the gray - level co - occurrence matrix features from the target remote - sensing image to obtain the first texture feature.

[0052] Among them, the first texture feature includes at least one of a contrast texture feature, a dissimilarity texture feature, a homogeneity texture feature, an angular second moment texture feature, an entropy texture feature, a variance texture feature, and a correlation texture feature. Preferably, the first texture feature is a contrast texture feature.

[0053] Step S3: Perform Gabor wavelet transform processing on the target remote sensing image to obtain a second texture feature.

[0054] Step S4: Perform Gabor wavelet transform processing on the first texture feature to obtain a third texture feature.

[0055] Step S5: Determine a three-dimensional texture feature space image according to the first texture feature, the second texture feature, and the third texture feature.

[0056] Specifically, superimpose the target remote sensing image after processing the first texture feature, the second texture feature, and the third texture feature, and assign it to the red, green, and blue color channels to obtain a three-dimensional texture feature space image.

[0057] Step S6: Use a clustering algorithm to classify the three-dimensional texture feature space image to obtain the distribution results of the water-permeable surface and the water-impermeable surface.

[0058] Specifically, use a clustering algorithm to classify the three-dimensional texture feature space image to obtain a number of water-permeable surface sub-images and a number of water-impermeable surface sub-images; merge all the water-permeable surface sub-images or all the water-impermeable surface sub-images to obtain the distribution results of the water-permeable surface and the water-impermeable surface.

[0059] Preferably, the clustering algorithm is the K-means unsupervised clustering algorithm. As a specific implementation method, the number of clusters is 4, the change threshold is 5%, and the maximum number of iterations is 25; each cluster is: water-impermeable surface (i.e., water-impermeable surface), forest, water body, and bare land, where forest, water body, and bare land are all water-permeable surfaces (i.e., water-permeable surfaces).

[0060] Step S7: Process the target remote sensing image according to the distribution results to obtain a water-impermeable surface distribution map.

[0061] A specific research example is provided below to illustrate the above method in detail.

[0062] Step 1: First, three cities, Beijing, Huainan, and Jinhua, were selected as the study areas, and single-scene images with zero or near-zero cloud cover (<10%) during the period from March to December 2021 were chosen. Using a DELL G15 personal notebook device with a configuration of 16G DDR4 3200MHz memory, 500G solid-state drive, 3.2GHz Intel i7 6-core 12-thread CPU, and NVIDIA GeForce RTX 3060 discrete graphics card, the corresponding single-scene images were downloaded from the Sentinel Scientific Data Center (https: / / scihub.copernicus.eu / ). For the Beijing area, four Sentinel-2 multispectral images were obtained on May 2, August 10, October 24, and December 13 respectively. For the Huainan area and the Jinhua area, one multispectral image was obtained on March 5 and May 6 respectively. In this embodiment, the Level-1C (L1C) product of Sentinel-2 was used, which was generated through radiometric and geometric corrections. The L1C product provides the reflectance at the top of the atmosphere (TOA). The Level-2 (L2) surface reflectance was obtained through the Sen2Cor atmospheric correction processor, which was integrated into the Sentinel Application Platform (SNAP) version 5.0. The Level-2 reflectance product was geometrically corrected to the Universal Transverse Mercator (UTM) projection system. The images of the three study areas were projected to the WGS1984 UTM zone 50. The 2, 3, and 4 bands (blue, green, red) of the Sentienl-2 L2 reflectance product were used to extract image texture features.

[0063] Step 2: Before extracting the texture of the Sentinel-2 multispectral images, four types of sample data, including impervious surfaces, forests, water bodies, and bare land, were collected in the three study areas. Specifically, using a stratified random sampling strategy, 4000 sample data were randomly generated, with 1000 for each type of land cover. The gray-level co-occurrence matrix and two-dimensional Gabor wavelet transform method as well as the texture features of secondary texture processing of 7 texture features from impervious surfaces, forests, water bodies, and bare land were calculated using the 4000 sample data. On the other hand, in order to quantitatively evaluate the accuracy of impervious surface mapping, a random sampling method was also adopted. There were a total of 6000 validation samples, 2000 for each site, and a random or stratified random sampling scheme was used to estimate the classification accuracy. Google Earth images contained the observational data of this period and had a high spatial resolution. Therefore, the time-series Google Earth images were used as an auxiliary data set for visual interpretation, and each validation sample could be labeled as "impervious surface" or "pervious surface". In addition, since the spatial heterogeneity of impervious surfaces is usually higher than that of natural land cover types, when identifying impervious samples, the impervious area in a 10m×10m window should account for more than 50%. Finally, in order to minimize the influence of the subjective knowledge of the interpretation experts, each validation sample would be interpreted by experts five times.

[0064] Step 3: Use the gray-level co-occurrence matrix, two-dimensional Gabor wavelet transform method, and quadratic texture processing method to extract the texture features of the entire Sentinel-2 single-scene image.

[0065] The texture feature extraction of the image is processed on the ENVI 5.3 and PyCharm Community Edition 2020.2.3 platforms. The gray-level co-occurrence matrix refers to the distribution of gray values that co-occur at a certain offset in the image. It tabulates the frequencies of different combinations of pixel gray values (or gray levels) in the image. And it also captures the digital features of the texture using the spatial relationship of similar gray levels.

[0066] To describe the texture attributes of four types of land cover in three study areas using Sentinel-2 images, GLCM (Grey-Level Co-occurrence Matrix) analysis was used. The GLCM values are calculated using the following formula to quantify the texture features:

[0067] p(i,j,d,θ) = {[x,y],(x + D x ,y + D y )|f(x,y) = i; f(x + D x ,y + D y ) = j; x =

[0068] 0,1,2,3…N x ; y = 0,1,2,3…N y}(1)

[0069] where pi,j,d,θ represents the probability of each pixel occurring in the image domain; fx,y represents the pixel value at coordinates x,y; fx+D x ,y+D y represents the pixel value at coordinates x+D x ,y+D y ; i represents the i-th row of the gray-level co-occurrence matrix, j represents the j-th column of the gray-level co-occurrence matrix, and i,j = 0,1,2,...N-1, where N is the number of gray levels of the image (i.e., the target remote sensing image); x,y are the original pixel positions; x,y are the pixel coordinates; N x and N y respectively represent the number of rows and columns of the image; D x and D yrespectively represent the horizontal and vertical offsets, both of which are 1 in this embodiment; d represents the generated step size, which is set to the texture processing window, specifically 3×3 pixel size in this embodiment; θ refers to the generation direction of pixel pairs, specifically 135° in this embodiment. Seven texture features were used in the study, namely contrast, homogeneity, heterogeneity, angular second moment, energy, entropy, variance, and correlation.

[0070] Contrast:

[0071]

[0072] Dissimilarity:

[0073]

[0074] Homogeneity:

[0075]

[0076] Angular second moment:

[0077]

[0078] Entropy:

[0079]

[0080] Variance:

[0081]

[0082] Correlation:

[0083]

[0084] where P i,j is the element in the i-th row and j-th column of the GLCM; N is the 64 gray levels in the specified image under quantization; u i and u j respectively represent the average value of the pixel in the i-th row and j-th column within the 3×3 range, and σ i and σ j are the standard deviations of the pixel in the i-th row and j-th column within the 3×3 range respectively.

[0085] In addition to the extraction of texture features GLCM based on statistical methods, the Gabor wavelet transform method based on multi-scale and time-frequency analysis is also used to distinguish the textures of impervious surfaces and other land covers. The texture of impervious surfaces has strong regularity and directionality, and wavelets can accurately capture multi-scale and multi-directional texture features. The Gabor function forms a completely non-orthogonal basis function, which provides a local frequency domain description and a basis for expanding signals. Set as the parent wavelet, the Gabor wavelet transform can be obtained by the expansion and rotation of the parent function, and can be calculated by the following formula:

[0086] g mn (x,y)=a -m g(x′,y′)a>1,m,n∈Z (9)

[0087] Among them, g mn (x,y) is regarded as a Gaussian function modulated by a complex sine function; the first parameter m = 0, 1, 2, 3..., the second parameter n = 0, 1, 2, 3..., by changing the values of m and n, a set of filters with different directions and scales can be obtained; Z represents the set of integers; x′ represents the abscissa transformation, and y′ represents the ordinate transformation, and y′ = a -m -xsinθ + ycosθ;[[ID=1⑧]] represents the wavelet transform angle, k refers to the number of wavelet direction transformations, in this method k = 6 is set, so that θ can be transformed to 0°, 30°, 60°, 90°, 120° and 150°; a refers to the scaling coefficient; a -m is the scaling scale, and its value is set to 2. By setting the number of directions k and the scaling scale a -m , the texture of the image can be decomposed into multiple sub-bands, thus overcoming to a certain extent the image differences caused by a certain rotation angle.

[0088] Although there are good differences between impervious surfaces and water bodies and forests in GLCM and wavelet texture features, the use of the two texture feature extraction methods cannot completely solve the problem of the mixture of impervious surfaces and bare land. GLCM can better distinguish between impervious surfaces and bare land because the internal arrangement of impervious surfaces is more regular, while the interior of bare land is more chaotic. In addition, the gray level at the edge of the impervious surface is very different from the surrounding environment, while the difference in gray level at the bare edge is smaller than that of the surrounding environment. The Gabor wavelet can perform edge recognition and detect areas where the gray frequency changes sharply. Therefore, two different features of impervious surfaces and bare land are also studied. The Gabor wavelet is used to perform secondary texture analysis on the image processed by GLCM, thereby enhancing the distinction between impervious surfaces.

[0089] Step 4: To determine the separation degree between the impervious surface and other land covers during the extraction process of GLCM features, Gabor wavelet textures, and quadratic texture features, apply M a 、M b 、M c indices as the main separation degree test indicators. M a 、M b 、M c indices are a numerical measure that quantifies the differences based on the means and standard deviations of the GLCM, Gabor wavelet, and quadratic-extracted texture feature values between the impervious surface and forest, impervious surface and water body, and impervious surface and bare land. Its mean and standard deviation are calculated from the texture feature values extracted at 4000 sample points. The formula can be expressed as follows:

[0090]

[0091]

[0092]

[0093] where u1, u2, u3, and u4 represent the average texture feature values of the impervious surface, forest, water body, and bare land, respectively, in this embodiment, and σ1, σ2, σ3, and σ4 represent the standard deviations of the texture feature values of the impervious surface, forest, water body, and bare land, respectively. If the values of M a 、M b 、M c are greater than 1, then these two classes will be well separated according to this feature. If the values of M a 、M b 、M c are less than 1, it indicates a very poor separation degree.

[0094] In addition, GLCM, Gabor wavelet, and quadratic texture extraction can only distinguish impervious surfaces from one type of feature and cannot completely separate other features simultaneously. The combination of the two texture features cannot avoid the mixing of impervious surfaces and other features. The distribution of impervious surfaces in the three feature spaces of GLCM, Gabor wavelet, and the texture features extracted quadratically shows significant differences, reducing the mixing of impervious surfaces with forests, water bodies, and bare land in the combination of one-dimensional and two-dimensional features. Therefore, three texture features, namely contrast (i.e., the first texture feature), Gabor wavelet feature (i.e., the second texture feature), and quadratic texture (Con_Gabor, i.e., the third texture feature), are combined to form a three-dimensional texture feature (3DTF) space. In this embodiment, in the ENVI remote sensing image processing software, the layers processed by these three texture features are superimposed to form an image, which is respectively assigned to the red, green, and blue channels. Using the K-means unsupervised clustering algorithm in the ENVI remote sensing image processing software, the number of clusters must be determined in advance, the K value is set to 4, including impervious surfaces, forests, water bodies, and bare land, the change threshold is set to 5%, the maximum number of iterations is set to 25 times, and then the image is automatically classified. Then, the classified pervious surfaces are merged. Finally, the image only shows the distribution results of impervious surfaces and pervious surfaces and is saved in the tiff or dat format. Verifying the accuracy of mapping impervious surfaces using an automatic method in different regions requires obtaining verification samples of the area or comparing with existing published impervious surface products.

[0095] Step 5: By comparing the mapping results with the verification samples, the ability of the proposed texture automation method to accurately classify impervious surfaces was evaluated. The confusion matrix, also known as the error matrix, is a standard format for representing accuracy evaluation and is represented in matrix form. The specific evaluation indicators include producer's accuracy (PA), user's accuracy (UA), and overall accuracy (OA). Verification samples are added to the classified impervious surface image, and PA, UA, OA, and Kappa values are calculated in the confusion matrix using ground truth rois function in the ENVI remote sensing image processing software. In addition, the F1 score is calculated from the UA and PA values obtained above:

[0096]

[0097] Embodiment 2

[0098] To implement the method corresponding to the above Embodiment 1 to achieve the corresponding functions and technical effects, the following provides an impervious surface mapping system. As Figure 2 shown, the impervious surface mapping system includes:

[0099] An acquisition module 1, configured to acquire a target remote sensing image.

[0100] The first feature extraction module 2 is configured to perform gray-level co-occurrence matrix feature extraction on the target remote sensing image to obtain the first texture feature.

[0101] The second feature extraction module 3 is configured to perform Gabor wavelet transform processing on the target remote sensing image to obtain the second texture feature.

[0102] The third feature extraction module 4 is configured to perform Gabor wavelet transform processing on the first texture feature to obtain the third texture feature.

[0103] The three-dimensional texture feature determination module 5 is configured to determine a three-dimensional texture feature space image according to the first texture feature, the second texture feature, and the third texture feature.

[0104] The classification module 6 is configured to classify the three-dimensional texture feature space image by using a clustering algorithm to obtain the distribution results of the water-permeable surface and the water-impermeable surface.

[0105] The mapping module 7 is configured to process the target remote sensing image according to the distribution results to obtain a water-impermeable surface distribution map.

[0106] Embodiment III

[0107] The embodiment of the present invention further provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor is used to run the computer program so that the electronic device executes the water-impermeable surface mapping method in Embodiment I. The electronic device may be a server.

[0108] In addition, the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the water-impermeable surface mapping method in Embodiment I is implemented.

[0109] In summary, the present invention uses the gray-level co-occurrence matrix and two-dimensional Gabor wavelet transform methods to extract image textures, and analyzes the unique texture features between impervious surfaces and other land covers based on selected ground object samples to construct a three-dimensional texture feature space. In addition, considering the high degree of automation required by this method, the optimal texture features of the impervious surface are then combined with the K-means unsupervised classifier to ultimately achieve the goal of automatic impervious surface mapping. Among them, the gray-level co-occurrence matrix is widely used in computer graphics to analyze image textures. It describes texture features by calculating the number of occurrences of specific gray values at a certain spatial position, and can describe the relevant state of the spatial gray-level distribution of the image, regardless of the analysis object. The two-dimensional Gabor filter constructed by the two-dimensional Gabor function is usually used in the research of texture feature extraction. The Gabor wavelet transform has very good spatial localization and direction selection characteristics, and can capture the edges and local features of the texture image at different frequencies and directions. In addition, since the image texture features are relatively stable and will not change significantly due to seasonal effects, mapping the impervious surface through texture features has strong robustness. Compared with the prior art, the present invention has the following advantages:

[0110] (1) It does not require collecting training samples, improving the classification efficiency.

[0111] Currently, deep learning and artificial neural networks can learn the basic features of remote sensing datasets, handle non-linear relationships, and integrate other data and features as well as prior knowledge. However, the accuracy of impervious surface mapping depends on the quantity and quality of the training data input into the model. In step 4 of this method, by analyzing the separation and clustering of impervious surfaces and several other land covers in the Sentinel-2 image in the 3DTF space composed of Contrast, Gabor wavelet, and quadratic texture (Con_Gabor), an automatic classification is carried out in combination with the K-means unsupervised separator.

[0112] (2) Impervious surface mapping that is not affected by seasons, improving the robustness of the method.

[0113] Currently, the method of using spectra to extract impervious surfaces is easily affected by seasonal heterogeneity. The main reason is that in summer, when the vegetation is lush, the spectral contrast between the impervious surface and the forest is clear. However, in winter, when the vegetation withers, a large amount of exposed bare land background will cause spectral blurring. In step 3 of this method, the gray-level co-occurrence matrix, two-dimensional Gabor wavelet method, and quadratic texture processing method are introduced to extract the texture features of the entire image. Because texture features can reflect the homogeneous visual features in the remote sensing data image and can well reflect the spatial distribution information of the image gray level, and the texture of the impervious surface, including the internal distribution arrangement and edge range, is a stable physical expression and has stability in different seasons.

[0114] (3) It can perform impervious surface mapping across sensors, improving the applicability of the method.

[0115] There are differences in spatial resolution among different satellite sensors. At the medium-resolution scale, the texture automation method is applied to 16-m GF6-WFV and 30-m Landsat-8 OLI to map impervious surfaces. The spatial distributions of the impervious surfaces mapped from the remote sensing images obtained by the three sensors are consistent, and the mapping errors caused by the influence of spatial resolution are within 3%. Currently, in the coarser-resolution images obtained, the problem of mixed spectra dominates in the process of mapping impervious surfaces using spectra, resulting in lower accuracy in local and regional mapping of impervious surfaces. Compared with methods such as spectral indices, the texture automation method has great potential for wider application.

[0116] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and reference can be made to the description in the method part for related parts.

[0117] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for mapping impervious surfaces, characterized in that, Including: Obtain a target remote sensing image; Extract gray-level co-occurrence matrix features from the target remote sensing image to obtain first texture features; Perform Gabor wavelet transform processing on the target remote sensing image to obtain second texture features; Perform Gabor wavelet transform processing on the first texture features to obtain third texture features; Determine a three-dimensional texture feature space image according to the first texture features, the second texture features, and the third texture features; Use a clustering algorithm to classify the three-dimensional texture feature space image to obtain the distribution results of water-permeable surfaces and water-impermeable surfaces; Process the target remote sensing image according to the distribution results to obtain a water-impermeable surface distribution map.

2. The impervious surface mapping method according to claim 1, wherein The first texture features include at least one of contrast texture features, dissimilarity texture features, homogeneity texture features, angular second moment texture features, entropy texture features, variance texture features, and correlation texture features.

3. The impervious surface mapping method according to claim 1, characterized in that Determine a three-dimensional texture feature space image according to the first texture features, the second texture features, and the third texture features, specifically including: Overlay the target remote sensing images processed by the first texture features, the second texture features, and the third texture features, and assign the three color channels of red, green, and blue to obtain a three-dimensional texture feature space image.

4. The impervious surface mapping method according to claim 1, wherein Use a clustering algorithm to classify the three-dimensional texture feature space image to obtain the distribution results of water-permeable surfaces and water-impermeable surfaces, specifically including: Use a clustering algorithm to classify the three-dimensional texture feature space image to obtain several water-permeable surface sub-images and several water-impermeable surface sub-images; Merge all the water-permeable surface sub-images or all the water-impermeable surface sub-images to obtain the distribution results of water-permeable surfaces and water-impermeable surfaces.

5. The impervious surface mapping method according to claim 1, characterized in that The clustering algorithm is the K-means unsupervised clustering algorithm.

6. The impervious surface mapping method according to claim 5, characterized in that, The number of clusters is 4, the change threshold is 5%, and the maximum number of iterations is 25; each cluster is: water-impermeable surface, forest, water body, and bare land, where forest, water body, and bare land are all water-permeable surfaces.

7. The impervious surface mapping method according to claim 1, characterized in that, The target remote sensing image is a Sentinel-2 multispectral image.

8. An impervious surface mapping system, characterized in that, Including: An acquisition module for acquiring a target remote sensing image; A first feature extraction module for extracting gray-level co-occurrence matrix features from the target remote sensing image to obtain first texture features; A second feature extraction module for performing Gabor wavelet transform processing on the target remote sensing image to obtain second texture features; [[ID= ​ ​ ​ 9. An electronic device, characterized in that, It includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the impervious surface mapping method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program which, when executed by a processor, implements the impervious surface mapping method according to any one of claims 1 to 7.

Images