Remote sensing mapping method, device and equipment for rural settlements and storage medium

Abstract

The application discloses a rural settlement remote sensing mapping method, device, equipment and storage medium, wherein the method comprises the following steps: acquiring image data sets of a to-be-measured area, and calculating characteristic wave bands of a first random forest classifier; inputting the characteristic wave bands into the first random forest classifier which is pre-trained, so as to output a pixel-level land cover result map; performing land cover type merging, linear impervious surface removal and raster vectorization processing on the land cover result map, so as to obtain a vectorized land cover vector object map; calculating spatial semantic features of the land cover objects, and acquiring an objectized land use result map based on the spatial semantic features; performing land use type re-labeling and rasterization processing on the land use result map, so as to obtain a rasterized land use result map; and performing category merging on the rasterized land use result map, so as to obtain a rural settlement distribution map of the target area. The application is beneficial to improving the accuracy of rural settlement mapping.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing, and in particular to a rural settlement remote sensing mapping method, device, equipment and storage medium. BACKGROUND

[0002] Rural settlements are one of the ways of land use, and are settlements where residents mainly engage in agricultural activities. They usually have rural landscapes such as farmhouses, livestock sheds, warehouses, courtyards, roads, canals, and green spaces beside houses. Large-scale mapping of rural settlements can help understand the situation of land use and provide data support for government departments to develop rural development policies. In the field of remote sensing application research, with the increasing number of satellite data images, the development and production of land cover and land use methods and products at large spatial scales and long time series have gradually been met. In these land cover products, pixels marked as impervious surface or built-up area can be considered as human settlements including rural settlements. The remote sensing images used by these methods or land cover products include coarse spatial resolution images (such as MODIS, VIIRS), medium spatial resolution images (such as Landsat, Sentinel-2), and high-resolution commercial satellite images (such as Worldview). Land cover products based on coarse spatial resolution images, such as the MCD12Q1 land cover product based on MODIS satellites, are limited by their spatial resolution and the size of rural settlements, and they are almost unable to accurately map rural settlements. High spatial resolution remote sensing satellite data has limited spatial coverage, making it difficult to achieve large-area rural settlement mapping. Therefore, medium spatial resolution images are a suitable choice for mapping rural settlements.

[0003] However, the existing mapping method of rural settlements is to use integrated multi-source satellite data to map rural settlements. For example, some "location indicators" are used as auxiliary data to improve the separability of artificial buildings in the feature dimension of Landsat images, so as to detect rural settlements. These indicators include the distance to water bodies, roads, and straw burning fire points, etc. generated by humans, because they all represent human activities near rural settlements. However, this method has some limitations. First, location indicators such as water body distribution and road network are not necessarily applicable to all regions around the world, especially in relatively poor areas where the government has no ability to invest in the construction of geographic spatial databases. In addition, other data sets, such as population census statistics, are also quite time-consuming, so they cannot be updated in a timely manner. Second, the historical data archives of some satellites such as Sentinel-1, 2 and VIIRS are much less than Landsat, so they cannot be combined with Landsat to track the dynamic changes of rural settlements at the scale of decades. Therefore, the existing method is difficult to achieve accurate mapping of rural settlements. SUMMARY

[0004] Embodiments of the present application aim to provide a remote sensing mapping method, device and equipment for rural settlements and a storage medium, so as to improve the accuracy of mapping of rural settlements.

[0005] To solve the above technical problems, the embodiments of the present application provide a remote sensing mapping method for rural settlements, comprising:

[0006] acquiring an image data set of a to-be-tested region, and calculating a feature band of a first random forest classifier based on the image data set, wherein the image data set is a Landsat surface reflectance image data set of the to-be-tested region at a preset period;

[0007] inputting the feature band into the pre-trained first random forest classifier to output a pixel-level land cover result map of the to-be-tested region;

[0008] performing land cover type merging, linear impervious surface removal and raster vectorization processing on the land cover result map to obtain a vectorized land cover object vector map, wherein the land cover object vector map comprises all land cover objects of the to-be-tested region;

[0009] calculating spatial semantic features based on the land cover objects, and inputting the spatial semantic features into a pre-trained second random forest classifier to output an objectified land use result map, wherein the land use result map comprises a land use type of each land cover object;

[0010] performing land use type relabeling and rasterization processing on the land use result map to obtain a rasterized land use result;

[0011] performing category merging on the rasterized land use result to obtain a rural settlement distribution map of a target region.

[0012] To solve the above technical problems, the embodiments of the present application provide a remote sensing mapping device for rural settlements, comprising:

[0013] a feature band calculation unit configured to acquire an image data set of a to-be-tested region, and calculate a feature band of a first random forest classifier based on the image data set, wherein the image data set is a Landsat surface reflectance image data set of the to-be-tested region at a preset period;

[0014] a land cover result map generation unit configured to input the feature band into a pre-trained first random forest classifier to output a pixel-level land cover result map of the to-be-tested region;

[0015] The land cover result map processing unit is configured to perform land cover type merging, linear impervious surface removal and raster vectorization processing on the land cover result map, to obtain a vectorized land cover object vector map of the to-be-detected region.

[0016] The land use result map generation unit is configured to calculate spatial semantic features based on the land cover objects, and input the spatial semantic features into a pre-trained second random forest classifier, to output an objectized land use result map, wherein the land use result map includes a land use type of each land cover object.

[0017] The land use result map processing unit is configured to perform land use type relabeling and rasterization processing on the land use result map, to obtain a rasterized land use result.

[0018] The target settlement distribution map generation unit is configured to perform category merging on the rasterized land use result, to obtain a rural settlement distribution map of the target region.

[0019] To solve the above technical problems, one technical solution of the present application is to provide a computer device, comprising one or more processors, and a memory configured to store one or more programs, so that the one or more processors implement the remote sensing mapping method of the rural settlement according to any one of the above.

[0020] To solve the above technical problems, one technical solution of the present application is a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the remote sensing mapping method of the rural settlement according to any one of the above.

[0021] The embodiments of the present application provide a remote sensing mapping method, device and equipment of a rural settlement and a storage medium. The embodiments of the present application realize the conversion from land cover to land use by generating a land cover result map and a land use result map, so as to extract rural residential areas with complete spatial forms, so that the form of the rural settlement is preserved, and the pixel result of the impervious surface is not limited, thereby facilitating the improvement of the accuracy of the rural settlement mapping. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the schemes in the present application, the drawings needed in the description of the embodiments of the present application will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Figure 1 is an implementation flowchart of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0024] Figure 2 is an arbitrary Landsat surface reflectance image of the to-be-tested area in 2020 provided by an embodiment of the present application;

[0025] Figure 3 is an implementation flowchart of a sub-process of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0026] Figure 4 is still another implementation flowchart of the sub-process of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0027] Figure 5 is still another implementation flowchart of the sub-process of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0028] Figure 6 is still another implementation flowchart of the sub-process of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0029] Figure 7 is a comparison chart of the effects before and after processing linear impervious surface removal provided by an embodiment of the present application;

[0030] Figure 8 is still another implementation flowchart of the sub-process of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0031] Figure 9 is still another implementation flowchart of the sub-process of the remote sensing mapping method for rural settlements provided by an embodiment of the present application;

[0032] Figure 10 is a land use result chart provided by an embodiment of the present application;

[0033] Figure 11 is a schematic diagram of a remote sensing mapping device for rural settlements provided by an embodiment of the present application;

[0034] Figure 12 is a schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] 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 application belongs; the terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application. Unless otherwise defined, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application. The terms "comprises", "comprising", "includes", "including", "has", "having" and the like are used in the detailed description and in the claims to mean that the stated features, steps, characteristics, and the like are included, but not excluded. The terms "first", "second", and the like, used in the description and in the claims, are used for distinguishing between similar objects and are not necessarily used to describe a particular sequential or chronological order.

[0036] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all referring to a common set of embodiments. It is expressly understood that any of the embodiments described herein can be incorporated in to another embodiment.

[0037] In order to make the technical personnel in the art better understand the scheme of the application, the technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings.

[0038] It should be noted that the remote sensing mapping method of rural settlements provided by the embodiments of the application is generally executed by a server, and accordingly, the remote sensing mapping device of rural settlements is generally configured in the server.

[0039] Please refer to Figures 1-2 , Figure 1 An embodiment of the remote sensing mapping method of rural settlements is shown. Figure 2 is any one Landsat surface reflectance image of the to-be-tested region in 2020 provided by the embodiments of the application.

[0040] It should be noted that the method of the application is not limited to the order of the flow shown, and the method includes the following steps: Figure 1

[0041] S1. Obtain an image data set of the to-be-tested region, and calculate the feature bands of the first random forest classifier based on the image data set.

[0042] The image data set is a Landsat surface reflectance image data set of the to-be-tested region in a preset period.

[0043] ​Specifically, the to-be-tested region refers to a target region for which rural settlement mapping is required. In a specific embodiment, if rural settlement mapping is required for the study region in 2020, all Landsat surface reflectance images from July 1, 2019 to June 30, 2021 are collected.

[0044] Referring to Figure 3 , Figure 3 An embodiment of step S1 is shown and described as follows:

[0045] S11. Obtain an image dataset of the to-be-tested region.

[0046] S12. Calculate spectral indices corresponding to the image dataset based on spectral bands and a preset formula.

[0047] S13. Extract score data of near-infrared bands of a preset frequency from the spectral indices and spectral bands of the image dataset, calculate spatial standard deviation based on the score data of the near-infrared bands, and obtain characteristic bands.

[0048] Specifically, 33 features need to be input into the subsequent first random forest classifier, which includes temporal, spatial, and spectral information of the ground pixels. First, the Landsat surface reflectance image includes six spectral bands (blue, green, red, near-infrared, far-infrared 1, and far-infrared 2). Four spectral indices including EVI, NDWI, NDMI, and NBR are calculated based on the spectral bands (calculated by the following formulas 1-4). For all spectral bands and spectral indices (a total of 10) of all collected Landsat reflectance images, the scores of the 20th, 50th, and 80th percentiles from low to high cumulative frequency are taken, that is, score data is obtained. Using the three score data of the near-infrared band, the spatial standard deviation of each pixel in the spatial 3x3 pixel window centered on the pixel is calculated (calculated by the above formula 5). Finally, the image of all 33 training features, i.e., 33 characteristic bands, is obtained.

[0049] Further, the preset formula includes:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] wherein Blue, Green, Red, NIR, Swirl and Swirl2 represent the reflectance values of the blue, green, red, near-infrared and two short-wave infrared bands in the Landsat reflectance data, and G, c1, c2 and L are 2.5, 1, 6 and 7.5, respectively.x i represents the pixel value of the i-th pixel in each spatial 3x3 pixel window, represents the mean value of all pixel values in the corresponding pixel window.

[0056] S2. Inputting the characteristic bands into the first random forest classifier trained in advance to output a pixel-level land cover result map of the region to be measured.

[0057] In the embodiments of the present application, the 33 characteristic bands obtained by the above steps are inputted into the first random forest classifier trained in advance to output a pixel-level land cover result map of the region to be measured.

[0058] Please refer to Figure 4 , Figure 4 shows a specific implementation before step S2, which is described as follows:

[0059] S21. Obtaining a high spatial resolution image of the region to be measured.

[0060] S22. Selecting a preset number of sample points from the high spatial resolution image to obtain sample data.

[0061] S23. Extracting classification features corresponding to the characteristic bands from the sample data according to a preset mode.

[0062] S24. Training the first random forest classifier based on the classification features to obtain the first random forest classifier trained in advance.

[0063] Specifically, before implementing step S2, the embodiments of the present application further include a process of training the first random forest classifier.

[0064] In the embodiment of the present application, a high spatial resolution image of the to-be-measured region is acquired, which can be a Google Earth image, and then sample data is generated based on different land cover type sample points collected from the high spatial resolution image. The sample data can also be generated from other ways, such as field collection of sample points. The sample data includes multiple land cover sample types, for example, including forest, grassland, shrub, water body, wetland, farmland, impervious surface, and bare soil. Then, according to a preset mode, classification features corresponding to feature bands are extracted from the sample data. In a specific embodiment, 350 sample points are selected in the to-be-measured region of 30km x 30km and around for each type of land cover category. Then, in GEE (Google Earth Engine), ee.Image.reduceRegions() is used to extract 33 classification features involved in step S1 for all samples. Finally, the first random forest classifier is trained based on the classification features, and a pre-trained first random forest classifier is obtained. The parameter setting of the first random forest classifier is that the number of trees is 200, and the number of node split features is the square root (rounded) of the number of features.

[0065] It should be noted that steps S1-S2 (including steps S21-S24) of the embodiment of the present application are completed on the Google Earth Engine (GEE) platform. Subsequent steps S3-S6 are to migrate the land cover result executed on GEE to a personal computer (or other computer device) for execution of the embodiment of the present application.

[0066] S3. The land cover result map is subjected to land cover type merging, linear impervious surface removal, and raster vectorization processing to obtain a vectorized land cover object vector map.

[0067] The land cover object vector map includes all land cover objects of the to-be-measured region.

[0068] Referring to Figure 5 , Figure 7 , Figure 5 a specific implementation of step S3 is shown, Figure 7 is a comparison chart of the front and back effects of processing linear impervious surface removal provided by the embodiment of the present application, which is described in detail as follows:

[0069] S31. The land cover type of the current pixel in the land cover result is identified, and the land cover type of the land cover result map is merged and relabeled based on the land cover type to obtain a first land cover result map.

[0070] Referring to Figure 6 , Figure 6 a specific implementation of step S31 is shown, which is described in detail as follows:

[0071] S311. Identify the current pixel in the land cover result and, based on the pixel value of the current pixel, identify the initial land cover type.

[0072] The initial land cover types include water bodies, forests, grasslands, wetlands, farmland, shrublands, impermeable surfaces, and bare soil.

[0073] S312. To combine forests, shrublands and grasslands into natural vegetation, and to combine water bodies and wetlands into water bodies.

[0074] S313. Re-mark natural vegetation, farmland, impermeable surfaces, bare soil, and water bodies in sequence to obtain the first land cover result map.

[0075] Specifically, land cover type merging refers to combining forests, shrubs, and grasslands into one category, labeled as natural vegetation. Water bodies and wetlands are merged into one category, labeled as water bodies. For example, in this application, the current pixel in the land cover result, from 0 to 7, represents water bodies, forests, grasslands, wetlands, farmland, shrubs, impermeable surfaces, and bare soil, respectively. By merging forests, shrubs, and grasslands into natural vegetation, and merging water bodies and wetlands into water bodies, a final classification result with five land cover types is obtained: natural vegetation, farmland, impermeable surfaces, bare soil, and water bodies. In the new classification result, these five cover types are labeled as 1, 2, 3, 4, and 5, respectively.

[0076] S32. Identify the pixels corresponding to roads in the first land cover result map, and use the mode principle to merge the pixels corresponding to roads into the categories adjacent to roads, so as to remove the linear impermeable surface pixels in the first land cover result map and obtain the second land cover result map.

[0077] Specifically, linear impermeable surface removal refers to removing roads from the first surface cover results because they can affect the detection results in rural settlements. For example... Figure 7 As shown, Figure 7 (a) presents a rural landscape, in Figure 7 A road can be observed at the center of (a). Figure 7 (b) is the obtained land cover map, which corresponds to Figure 7(a) The position indicated by the grey line, which contains 5x5 Landsat pixels. The center and the outermost impervious pixels are used to determine whether the center pixel belongs to a road. From the center pixel and each of the five boundary pixels, five line segments can be formed, and five angles can be formed from the two-by-two adjacent line segments. If two or more of the five angles are greater than 90°, the center pixel is considered to be part of a road. Once all the pixels belonging to the road are identified, they are merged into the category of the land cover type with the highest frequency of non-impervious pixels in the 3x3 window centered on them. Figure 7 (c) and (d) show the land cover map before and after this post-processing, respectively, successfully removing linear impervious pixels.

[0078] S33. Vectorize the land cover in the second land cover result map to make the pixels of the same land cover type that are spatially adjacent to each other into a land cover object, to obtain a vectorized land cover object, and obtain the land cover object vector map after all the vectorized land cover objects are generated.

[0079] Specifically, raster vectorization refers to raster vectorization of the second land cover result map into different land cover object polygons, referred to as land cover objects. The vectorization principle is that the pixels of the same land cover type are merged into a land cover object if they are in contact with each other in any one of the eight spatial directions. The embodiment of the application performs vectorization operations by calling the ogr library in the gda l package of python.

[0080] S4. Calculate the spatial semantic features based on the land cover objects, and input the spatial semantic features into the second random forest classifier trained in advance to output an objectified land use result map.

[0081] The land use result map includes the land use type of the land cover object.

[0082] Specifically, the embodiment of the application includes a land cover stage and a land use stage. The above steps S1-S3 are the land cover stage, and steps S4-S6 are the land use stage. The land cover stage adopts an automatic sampling strategy, so that the samples are representative in rural areas; in the land use stage, the semantic information between different adjacent land cover objects is used as a training feature to successfully detect the patches of rural settlements.

[0083] Please refer to Figure 8 , Figure 8 An embodiment of step S4 is shown, which is described in detail as follows:

[0084] S41. Take the land cover object as the target object, and take the land cover objects adjacent to the target object and having common boundaries as the neighbor objects.

[0085] The attributes of the target object include the area of the object, the land cover type, the length and width of the minimum circumscribed rectangle of the object, and the length-width ratio.

[0086] S42. Group all the neighbor objects according to the land cover types of the neighbor objects to obtain neighbor object groups.

[0087] S43. Calculate the topological relationship between the neighbor object groups and the target object, wherein the topological relationship includes the number of neighbor objects in each neighbor object group, the total area, the relative area with respect to the target object, the normalized common contact boundary with respect to the target object, and the area-weighted distance from the target object.

[0088] The area-weighted distance is calculated by first calculating the geographic center of the neighbor object group with the area of all objects in the group as the weight, and then calculating the distance from the geographic center to the center of the target object.

[0089] S44. Take the attributes of the target object and the topological relationship between the neighbor objects and the target object as the spatial semantic features.

[0090] S45. Input the spatial semantic features into the pre-trained second random forest classifier to output a land use result map.

[0091] Specifically, in the land use classification stage, the random forest classifier is still used, but the spatial semantic features are used, that is, the second random forest classifier is retrained using the attributes of the land cover object (target object) itself and the topological relationship between the land cover object and the adjacent land cover objects. There are 30 spatial semantic features. First, the attributes of the target object itself include the area of the object, the land cover type, and the length and width of the minimum circumscribed rectangle of the object, and the length-width ratio. Second, the topological relationship between the target object and the adjacent objects can be divided into 5 groups according to the different land cover types of the adjacent objects. The relationship between each group of neighbor objects and the target object includes the number of neighbor objects, the total area, the relative area with respect to the target object, the normalized common contact boundary with respect to the target object (the ratio of the common edge length to the perimeter of the target object), and the geographic distance (also known as the area-weighted distance) between the center of the neighbor object and the center of the target object. The geographic distance between the center of the neighbor object and the center of the target object is first calculated by calculating the geographic center of all neighbor objects with the area as the weight (formula (6) below), and then the distance from the geographic center to the center of the target object is calculated. Finally, each land cover object contains 30 spatial semantic features for random forest classifier training.

[0092] Further, formula (6) is:

[0093]

[0094] wherein w i represents the area of the neighbor object with the center coordinate (x i , y i ) in the geographical space, and the number of neighbors of this land cover type is n. The geographical center of all the neighbor objects of the target land cover type is (x, y).

[0095] Finally, the spatial semantic features are input into the pre-trained second random forest classifier to output a land use result map.

[0096] Referring to Figure 9 , Figure 9 a specific implementation before step S45 is shown, which is described as follows:

[0097] S451. Obtain a high-resolution remote sensing image map of the region to be measured.

[0098] S452. Compare the high-resolution remote sensing image map with the target land cover result map to randomly select a sample library for constructing a land use classification.

[0099] S453. Train the second random forest classifier according to the sample library to obtain the pre-trained second random forest classifier.

[0100] Specifically, corresponding to the five land cover types, the embodiments of the present application have five land use types, which are natural vegetation land, agricultural land, residential land, bare land and water body land. The difference between the land use type and the land cover type is that the impervious surface category is replaced by the residential land category. In the embodiments of the present application, by obtaining a high-resolution remote sensing image map of the region to be measured, and then comparing the high-resolution remote sensing image map with the target land cover result map, a sample library for constructing a land use classification is randomly selected.

[0101] In a specific embodiment, by comparing the different surface cover objects obtained in step S3 with the high-resolution remote sensing image map, it is determined whether these objects belong to the five land use types. Therefore, the objects from the land cover type to the land use type include a total of 25 conversion types. For example, natural vegetation can be attributed to residential land if it belongs to a part of the residential area and is adjacent to the impervious surface object, while scattered objects containing only a few impervious surfaces (up to 5 in the embodiment of the present application) are considered to belong to the land use type that wraps the object. Finally, the land use classification sample library is selected for the 25 conversion types. Finally, the second random forest classifier is trained according to the land use classification sample library to obtain the pre-trained second random forest classifier.

[0102] S5. Re-labeling and rasterizing the land use result map to obtain a rasterized land use result map.

[0103] Specifically, for the target object that has a class conversion from land cover to land use (such as from A to B), the conversion is determined by the adjacent land cover object of the same class (i.e., the B class). However, the adjacent land cover object (i.e., the B class) may also be converted during the land use classification process (such as from B to C), and if the conversion occurs, the target object is relabeled as the land use type of the adjacent land cover object (i.e., the C class). After the rasterization of the land cover object, a land use result map containing five land use types is generated.

[0104] S6. Obtain the rural residential area distribution map of the target area by merging the rasterized land use result map.

[0105] Please refer to Figure 10 , Figure 10 is the land use result map provided by the embodiment of the present application.

[0106] Specifically, by merging the rasterized land cover object map, i.e., merging other non-residential area categories (natural vegetation land, agricultural land, bare land, water body) into a background category, a binary image of the residential area distribution map and the background pixel is obtained, i.e., the rural residential area distribution map of the target area is obtained. In the distribution map, the residential area pixel value is 1, and the background pixel value is 0.

[0107] In the embodiment of the present application, the image data set of the to-be-tested region is acquired, and the feature band of the first random forest classifier is calculated based on the image data set, wherein the image data set is a Landsat ground surface reflectivity image data set of the to-be-tested region at a preset period; the feature band is input into the first random forest classifier trained in advance, so as to output a pixel-level land cover result map of the to-be-tested region; the land cover result map is subjected to land cover type merging, linear impervious surface removal and raster vectorization processing, so as to obtain a vectorized land cover object vector map, wherein the land cover object vector map includes all land cover objects of the to-be-tested region; the spatial semantic feature is calculated based on the land cover object, and the spatial semantic feature is input into the second random forest classifier trained in advance, so as to output an objectified land use result map, wherein the land use result map includes the land use type of each land cover object; the land use result map is subjected to land use type relabeling and land cover object rasterization processing, so as to obtain a rasterized land use result map; and the rasterized land use result map is subjected to merging, so as to obtain a target region rural settlement distribution map. The embodiment of the present application realizes the conversion from land cover to land use by generating the land cover result map and the land use result map, so as to extract the rural residential area with complete spatial form, and the form of the rural settlement is retained, rather than being limited to obtaining the impervious surface pixel result, thereby being beneficial to improving the accuracy of rural settlement mapping.

[0108] Further, the embodiment of the present application is designed as a hierarchical structure including two stages, so as to realize the conversion from land cover to land use, and finally extract the rural residential area with complete spatial form. Stage 1 is pixel-based land cover mapping, and stage 2 realizes the conversion from the land cover concept to the land use concept by exploring the semantic information between different land cover objects in the rural residential area. Compared with the prior art, the form of the rural settlement is retained, rather than being limited to the impervious surface pixel result obtained in the prior method, and the obtained result is scientific and reasonable.

[0109] Further, the embodiment of the present application adopts an automatic sample generation strategy for land cover classification, which can make the impervious surface sample more representative in rural areas. Meanwhile, in order to solve the problem of the increase of misclassification cases and the appearance of pepper and salt noise caused by the sample representativeness, stage 2 of the hierarchical structure designed in the embodiment of the present application can effectively solve the pepper and salt noise in the land cover classification map by using the semantic information between different ground cover patches, and finally obtain the settlement distribution result.

[0110] Please refer to Figure 11 , as to the above Figure 1The implementation of the method shown in this application provides an embodiment of a remote sensing mapping device for rural settlements, which is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0111] like Figure 11 As shown, the remote sensing mapping device for rural settlements in this embodiment includes: a feature band calculation unit 71, a land cover result map generation unit 72, a land cover result map processing unit 73, a land use result map generation unit 74, a land use result map processing unit 75, and a target area settlement distribution map generation unit 76, wherein:

[0112] The feature band calculation unit 71 is used to acquire the image dataset of the area to be tested and calculate the feature bands of the first random forest classifier based on the image dataset. The image dataset is the Landsat surface reflectance image dataset of the area to be tested during a preset period.

[0113] The land cover result map generation unit 72 is used to input the feature bands into the pre-trained first random forest classifier to output a pixel-level land cover result map of the area to be tested;

[0114] The land cover result map processing unit 73 is used to perform land cover type merging, linear impermeable surface removal and raster vectorization processing on the land cover result map to obtain a vectorized land cover object vector map, wherein the land cover object vector map includes all land cover objects in the area to be measured;

[0115] The land use result map generation unit 74 is used to calculate spatial semantic features based on land cover objects and input the spatial semantic features into a pre-trained second random forest classifier to output an object-oriented land use result map, wherein the land use result map includes the land use type of each land cover object;

[0116] The land use result map processing unit 75 is used to re-label the land use types and rasterize the land use result map to obtain rasterized land use results.

[0117] The target settlement distribution map generation unit 76 is used to obtain the rural settlement distribution map of the target area by merging the categories of the rasterized land use results.

[0118] Furthermore, the characteristic band calculation unit 71 includes:

[0119] The image dataset acquisition unit is used to acquire the image dataset of the area to be tested.

[0120] The spectral index calculation unit is configured to calculate a spectral index corresponding to the image data set based on the spectral band and a preset formula.

[0121] The score data extraction unit is configured to extract score data of a near-infrared band of a preset frequency from the spectral index and the spectral band of the image data set, calculate a spatial standard deviation based on the score data of the near-infrared band, and obtain the characteristic band.

[0122] Further, the land cover result map generation unit 72 further comprises:

[0123] The high spatial resolution image acquisition unit is configured to acquire a high spatial resolution image of the region to be measured.

[0124] The sample point selection unit is configured to select a preset number of sample points from the high spatial resolution image to obtain sample data.

[0125] The classification feature extraction unit is configured to extract classification features corresponding to the characteristic band from the sample data in a preset manner.

[0126] The random forest classifier training unit is configured to train the first random forest classifier based on the classification features to obtain the pre-trained first random forest classifier.

[0127] Further, the land cover result map processing unit 73 comprises:

[0128] The first land cover result map generation unit is configured to identify a land cover type represented by a pixel in the land cover result, and merge and re-label the land cover types in the land cover result map to obtain a first land cover result map.

[0129] The second land cover result map generation unit is configured to identify a pixel corresponding to a road in the first land cover result map, and merge the pixel corresponding to the road into a category adjacent to the road according to a principle of mode to remove a linear impervious surface pixel in the first land cover result map to obtain a second land cover result map.

[0130] The raster vectorization processing unit is configured to perform raster vectorization processing on the land cover in the second land cover result map to merge pixels of the same land cover type that are spatially adjacent into one land cover object to obtain a vectorized land cover object, and obtain a land cover object vector map after all the vectorized land cover objects are generated.

[0131] Further, the first land cover result map generation unit comprises:

[0132] An initial land cover type identifying unit is configured to identify a current pixel in the land cover result and identify an initial land cover type based on a pixel value of the current pixel, wherein the initial land cover type includes water body, forest, grassland, wetland, farmland, shrub, impervious surface and bare soil;

[0133] An initial land cover type merging unit is configured to merge the forest, the shrub and the grassland into natural vegetation, and merge the water body and the wetland into water body;

[0134] An initial land cover type marking unit is configured to sequentially re-mark the natural vegetation, the farmland, the impervious surface, the bare soil and the water body to obtain a first land cover result map.

[0135] Further, the land use result map generating unit 74 comprises:

[0136] A target object attribute unit is configured to take the land cover object as a target object, and take a land cover object having a common boundary with the target object as a neighbor object, wherein an attribute of the target object includes an area of the object, a land cover type, a length and a width of a minimum circumscribed rectangle of the object and a length-width ratio;

[0137] A target grouping generating unit is configured to group all the neighbor objects according to the land cover type of the neighbor objects to obtain a neighbor object grouping;

[0138] A topological relationship calculating unit is configured to calculate a topological relationship between the neighbor object grouping and the target object, wherein the topological relationship includes a number of neighbor objects in each neighbor object grouping, a total area, a relative area with respect to the target object, a normalized common contact boundary with respect to the target object and an area-weighted distance with respect to the target object;

[0139] A spatial semantic feature generating unit is configured to take the attribute of the target object and the topological relationship between the neighbor objects and the target object as spatial semantic features;

[0140] A land use result map output unit is configured to input the spatial semantic features into a second random forest classifier trained in advance to output a land use result map.

[0141] Further, the land use result map generating unit 74 further comprises:

[0142] A high-resolution remote sensing image map obtaining unit is configured to obtain a high-resolution remote sensing image map of a region to be measured;

[0143] A land use classification sample library constructing unit is configured to compare the high-resolution remote sensing image map with the target land cover result map to randomly obtain a sample library for constructing a land use classification;

[0144] The second random forest classifier training unit is configured to train a second random forest classifier according to the sample library, and obtain a pre-trained second random forest classifier.

[0145] To solve the above technical problems, the embodiment of the present application further provides a computer device. For details, please refer to Figure 12 , Figure 12 The basic structure block diagram of the computer device of the embodiment is shown in FIG. 1.

[0146] The computer device 8 includes a memory 81, a processor 82, and a network interface 83 which are connected to each other through a system bus. It should be noted that only the computer device 8 with three components, i.e., the memory 81, the processor 82, and the network interface 83 is shown in the figure, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be alternatively implemented. Among them, those skilled in the art can understand that the computer device herein is a device capable of automatically performing numerical calculation and / or information processing according to pre-set or stored instructions, and the hardware thereof includes but is not limited to a microprocessor, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), an embedded device, etc.

[0147] The computer device can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The computer device can perform human-computer interaction with a user through a keyboard, a mouse, a remote controller, a touchpad, a voice control device, and the like.

[0148] The memory 81 includes at least one type of readable storage medium, including a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 81 can be an internal storage unit of the computer device 8, such as a hard disk or a memory of the computer device 8. In other embodiments, the memory 81 can also be an external storage device of the computer device 8, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 8. Of course, the memory 81 can also include both the internal storage unit and the external storage device of the computer device 8. In the present embodiment, the memory 81 is generally used to store an operating system and various application software installed on the computer device 8, such as the program code of the remote sensing mapping method for rural settlements, etc. In addition, the memory 81 can also be used to temporarily store various data that have been output or will be output.

[0149] The processor 82 can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 82 is generally used to control the overall operation of the computer device 8. In the present embodiment, the processor 82 is used to run the program code or process data stored in the memory 81, such as running the program code of the remote sensing mapping method for rural settlements to implement various embodiments of the remote sensing mapping method for rural settlements.

[0150] The network interface 83 can include a wireless network interface or a wired network interface, and is generally used to establish a communication connection between the computer device 8 and other electronic devices.

[0151] The present application also provides another implementation, i.e., to provide a computer readable storage medium storing a computer program, which can be executed by at least one processor to make the at least one processor execute the steps of the remote sensing mapping method for rural settlements as described above.

[0152] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the method of each embodiment of the present application.

[0153] Obviously, the above-described embodiments are only some of the embodiments of the present application, not all the embodiments, and the drawings show the preferred embodiments of the present application, but do not limit the patent scope of the present application. The present application can be implemented in many different forms, and conversely, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing specific embodiments, or make equivalent replacements to some technical features. Any equivalent structure made by using the content of the specification and drawings, directly or indirectly applied to other related technical fields, is also within the scope of the patent protection of the present application.

Claims

1. A remote sensing mapping method for rural settlements, characterized in that, include: Obtain an image dataset of the area to be tested, and calculate the feature bands of the first random forest classifier based on the image dataset, wherein the image dataset is a Landsat surface reflectance image dataset of the area to be tested during a preset period. The feature bands are input into a pre-trained first random forest classifier to output a pixel-level land cover map of the area to be tested. The land cover result map is processed by merging land cover types, removing linear impermeable surfaces, and raster vectorization to obtain a vectorized land cover object vector map, wherein the land cover object vector map includes all land cover objects in the area to be measured; Spatial semantic features are calculated based on the land cover objects, and the spatial semantic features are input into a pre-trained second random forest classifier to output an object-oriented land use result map, wherein the land use result map includes the land use type of each land cover object; The land use result map is re-labeled with land use types and rasterized to obtain rasterized land use results; By categorizing and merging the rasterized land use results, a distribution map of rural settlements in the target area is obtained.

2. The remote sensing mapping method for rural settlements according to claim 1, characterized in that, The process of acquiring the image dataset of the region to be tested and calculating the feature bands of the first random forest classifier based on the image dataset includes: Obtain the image dataset of the region to be tested; Based on the spectral bands and preset formulas, the spectral index corresponding to the image dataset is calculated; The near-infrared band score data of a preset frequency is extracted from the spectral index and spectral band of the image dataset, and the spatial standard deviation is calculated based on the near-infrared band score data to obtain the feature band.

3. The remote sensing mapping method for rural settlements according to claim 1, characterized in that, Before inputting the feature bands into a pre-trained first random forest classifier to output a pixel-level land cover map of the area to be tested, the method further includes: Acquire a high spatial resolution image of the region to be tested; A predetermined number of sample points are selected from the high spatial resolution image to obtain sample data; According to a preset method, the classification features corresponding to the feature bands are extracted from the sample data; The first random forest classifier is trained based on the classification features to obtain the pre-trained first random forest classifier.

4. The remote sensing mapping method for rural settlements according to claim 1, characterized in that, The process of merging land cover types, removing linear impermeable surfaces, and rasterizing the land cover result map to obtain a vectorized land cover object vector map includes: Identify the current pixel in the land cover result, and merge and re-label the land cover type based on the current pixel to obtain the first land cover result map; Identify the pixels corresponding to roads in the first land cover result map, and use the mode principle to merge the pixels corresponding to roads into the categories adjacent to roads, so as to remove the linear impermeable surface pixels in the first land cover result map and obtain the second land cover result map. By performing raster vectorization on the land cover in the second land cover result map, pixels of the same type of spatially adjacent land cover are merged into one land cover object, thus obtaining the vectorized land cover object. After all the vectorized land cover objects are generated, the land cover object vector map is obtained.

5. The remote sensing mapping method for rural settlements according to claim 4, characterized in that, The land cover result map is merged and re-labeled by land cover type to obtain a first land cover result map, including: Identify the current pixel in the land cover result, and based on the pixel value of the current pixel, identify the initial land cover type, wherein the initial land cover type includes water body, forest, grassland, wetland, farmland, shrubland, impermeable surface and bare soil; The forest, the shrubland, and the grassland are combined into natural vegetation, and the water body and the wetland are combined into a water body; The natural vegetation, farmland, impermeable surfaces, bare soil, and water bodies are re-marked in sequence to obtain the first land cover result map.

6. The remote sensing mapping method for rural settlements according to any one of claims 1 to 5, characterized in that, The step of calculating spatial semantic features based on the land cover object and inputting the spatial semantic features into a pre-trained second random forest classifier to output an object-oriented land use result map includes: The land cover object is taken as the target object, and the land cover objects that share the same boundary with the target object are taken as neighbor objects. The attributes of the target object include the object's area, land cover type, length and width of the object's minimum bounding rectangle, and aspect ratio. Based on the land cover type of the neighbor objects, all the neighbor objects are grouped to obtain neighbor object groups; Calculate the topological relationship between the neighbor object group and the target object, wherein the topological relationship includes the number of neighbor objects in each neighbor object group, the total area, the relative area to the target object, the normalized common contact boundary to the target object, and the area-weighted distance to the target object; The attributes of the target object and the topological relationship between the neighboring objects and the target object are used as the spatial semantic features; The spatial semantic features are input into the pre-trained second random forest classifier to output the land use result map.

7. The remote sensing mapping method for rural settlements according to claim 6, characterized in that, Before inputting the spatial semantic features into the pre-trained second random forest classifier to output the land use result map, the method further includes: Obtain a high-resolution remote sensing image map of the area to be tested; The high-resolution remote sensing image map is compared with the target land cover result map to randomly obtain a sample library for constructing land use classification; The second random forest classifier is trained based on the sample library to obtain the pre-trained second random forest classifier.

8. A remote sensing mapping device for rural settlements, characterized in that, include: The feature band calculation unit is used to acquire the image dataset of the area to be tested and calculate the feature bands of the first random forest classifier based on the image dataset, wherein the image dataset is the Landsat surface reflectance image dataset of the area to be tested during a preset period. The land cover result map generation unit is used to input the feature bands into a pre-trained first random forest classifier to output a pixel-level land cover result map of the area to be tested; The land cover result map processing unit is used to perform land cover type merging, linear impermeable surface removal, and raster vectorization processing on the land cover result map to obtain a vectorized land cover object vector map, wherein the land cover object vector map includes all land cover objects in the area to be measured; A land use result map generation unit is used to calculate spatial semantic features based on the land cover object and input the spatial semantic features into a pre-trained second random forest classifier to output an object-oriented land use result map, wherein the land use result map includes the land use type of each land cover object; The land use result map processing unit is used to re-label the land use types and rasterize the land use result map to obtain rasterized land use results. The target settlement distribution map generation unit is used to obtain a rural settlement distribution map of the target area by merging the categories of the rasterized land use results.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the remote sensing mapping method for rural settlements as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the remote sensing mapping method for rural settlements as described in any one of claims 1 to 7.

Images