A weak information ground object remote sensing extraction method

Abstract

The application discloses a weak information ground object remote sensing extraction method, which can quickly select beneficial information related to weak information ground object identification in an electronic map; a feature selection and preparation method is used to convert the information into data content and format which can participate in weak information ground object remote sensing extraction; an object-oriented segmentation technology with a small scale threshold is used to accurately generate object monomers which can match the edges of weak information ground objects; and an image processing technology process is used to effectively extract weak information ground objects from remote sensing images with the help of the above information. The application uses a small size threshold segmentation parameter combination to generate object monomers which can match the edges of weak information ground objects, so that the identification precision of the boundaries of weak information ground objects can be significantly improved; and the application uses a spatial registration and image processing technology method process to complete the rapid extraction of weak information ground objects, greatly improves work efficiency, and meets user requirements.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of remote sensing images, and particularly relates to a weak information ground object remote sensing extraction method. BACKGROUND

[0002] Remote sensing images can provide rich and timely information for monitoring and mastering the changes of the earth's surface, and can reflect the current features and historical changes of land objects such as farmland, forest land, water surface, land, mineral resources, and ocean. The popularization of remote sensing technology has enabled China to enter a new era of stereoscopic, multi-level, multi-angle, all-around, and all-weather observation of the earth, greatly improving the observation ability of the earth and the monitoring and management level, and providing sufficient data support and technical support for the construction of China's digital society and the sustainable development of green economy.

[0003] Weak information remote sensing recognition technology has always restricted the accuracy of remote sensing interpretation and the effect of remote sensing application. In the vast desert areas in the western part of China, the spectral information of some road, residential land, new energy land (hydroelectric power station, photovoltaic, wind power, etc.), factory, and mining site is extremely weak and difficult to detect. This often causes misclassification and missed classification of some important land types, and in the Gobi or desert area, the road interpretation plot is often discontinuous, which greatly affects the accuracy and usability of the remote sensing interpretation result. In addition, for some easily mixed land types (such as commercial, cultural, and educational land in construction land), the phenomenon of the same spectrum of different objects also easily causes misclassification and missed classification. Compared with the actual amount of information required for complete recognition of these land types, this kind of remote sensing interpretation problem can also be regarded as a weak information remote sensing recognition problem.

[0004] At present, the main methods for weak information ground object recognition include machine learning and supervised classification methods. The former mainly relies on sample labeling and learning model construction, and the latter mainly establishes an expert knowledge base of weak information ground objects according to spectral, geometric, and texture features. The above methods are relatively effective for some land types whose features are significantly different from other land types (such as regularly arranged photovoltaic power plants), but the recognition effect of small ground objects in the desert background and easily mixed land objects in the city is still not good (different backgrounds, materials, sizes, and land use types cause the same spectrum of different objects). And often the classifier needs to have a detailed understanding of the spatial distribution of the key ground objects in the classification area in order to focus on the recognition of the main distribution area. Essentially, this kind of method relies on sufficient expert knowledge to meet the information required for accurate interpretation in order to complete the recognition of weak information land types, which greatly improves the experience requirement for the interpreter to some extent. SUMMARY

[0005] In view of the above problems in the prior art, the weak information ground object remote sensing extraction method provided by the application solves the problems of complicated sample marking, low interpretation accuracy and strict experience requirement for interpreters of the prior extraction method.

[0006] In order to achieve the above-mentioned purposes, the application adopts the technical scheme of a weak information ground object remote sensing extraction method, comprising the following steps:

[0007] S1, collecting available ground object information in an open source map;

[0008] S2, performing data conversion on the classified ground object information to construct a weak information planar vector element set;

[0009] S3, generating a thematic information enhanced raster data set based on the weak information planar vector element set;

[0010] S4, collecting remote sensing images corresponding to the available ground object information, and performing feature segmentation to obtain an object single body polygon set;

[0011] S5, performing weak information ground object remote sensing extraction on the object single body polygon set based on each thematic layer in the raster data set.

[0012] Further, the available ground object information in the step S1 is ground element data in the open source map with a spatial size less than 1 hectare and an information accuracy greater than 80% when performing random type inspection.

[0013] Further, the step S2 specifically comprises:

[0014] S21, performing spatial consistency processing on the available ground object information;

[0015] S22, classifying the point-like vector data and the line-like vector data with spatial consistency, and respectively using point-to-surface and line-to-surface vector conversion methods to construct a weak information planar vector element set.

[0016] Further, the step S21 specifically comprises:

[0017] Performing a GIS command of projection conversion on the available ground object information, and adopting a transverse Mercator projection mode to perform projection processing, so that the central meridian position range of the processed vector data is within ±3° of the central longitude, to ensure the spatial consistency of the available ground object information;

[0018] In the step S22, for the point-like vector data, after classification, the point-to-surface vector conversion method is used to generate a corresponding two-dimensional planar vector element layer, and a corresponding category digital code is assigned in the attribute table thereof;

[0019] For linear vector data, after classification, the buffer analysis of different lengths is carried out to form planar vector data, and the corresponding category digital code is assigned in the attribute table;

[0020] For planar vector data in the existing land feature information, the corresponding category digital code is assigned in the attribute table.

[0021] Further, the step S3 is specifically:

[0022] The raster conversion is performed on each vector data in the weak information planar vector element layer by using the nearest neighbor pixel method, and then the converted two-dimensional raster data is superimposed on the three-dimensional space according to the XY plane coordinate values of each layer, and the raster data set of thematic information enhancement is generated by using the spatial correlation thereof;

[0023] In the raster conversion, the attribute value of the vector data is converted into the raster cell value.

[0024] Further, the step S4 is specifically:

[0025] S41, setting multi-scale segmentation parameters, including scale, color and smoothness;

[0026] S42, configuring and combining the multi-scale segmentation parameters, using the optimal segmentation parameters to perform small-area multi-scale segmentation on the remote sensing image, and then obtaining an object monomer patch set reflecting the edge features of the weak information land features in the remote sensing image;

[0027] When the land feature types in the segmentation unit do not exceed 2, and the error between the edge of the segmentation unit and the patch boundary in the remote sensing image does not exceed 0.3 pixel size, the corresponding multi-scale segmentation parameter combination is the optimal segmentation parameter.

[0028] Further, the step S5 is specifically:

[0029] Taking the patch in the object monomer patch set as the minimum processing unit, using the object-oriented image analysis method, using each thematic layer in the raster data set, the spatial mode filtering is performed on the weak information land feature remote sensing extraction.

[0030] Further, the formula for performing the spatial mode filtering is:

[0031]

[0032]

[0033] In the formula, M0 is the mode value calculated by the filtering unit, L is the lower limit of the group where the mode is located, U is the upper limit of the group where the mode is located, f b is the difference between the number of groups where the mode is located and the number of adjacent groups of the lower limit, and f aThe difference between the number of the mode group and the number of the adjacent group of the upper limit is the difference of the mode group, and i is the difference of the mode group.

[0034] The beneficial effects of the present application are:

[0035] 1) The present application makes full use of the open source information provided by important places and ground objects on Internet maps and navigation maps, extracts the position and attribute features related to weak information ground objects through data cleaning and feature selection, and makes a thematic information enhanced raster data set, which can greatly enrich the information amount and meet the needs of remote sensing extraction of weak information ground objects.

[0036] 2) The present application uses a smaller size threshold parameter combination to generate object monomers that can match the edges of weak information ground objects, which can significantly improve the recognition accuracy of the boundaries of weak information ground objects.

[0037] 3) The present application uses a method process of spatial registration and image processing technology to complete the rapid extraction of weak information ground objects, which greatly improves the work efficiency and meets the needs of users. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 A weak information ground object remote sensing extraction method process is provided for the present application.

[0039] Figure 2 A weak information ground object remote sensing extraction schematic diagram is provided for the present application. DETAILED DESCRIPTION

[0040] The specific embodiments of the present application are described below to facilitate understanding of the present application by those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all inventions utilizing the concept of the present application are within the scope of protection.

[0041] As shown in Figure 1 A weak information ground object remote sensing extraction method, comprising the following steps:

[0042] S1, collecting available ground object information in open source maps;

[0043] S2, performing data conversion on the collected ground object information classified by categories to construct a weak information planar vector feature set;

[0044] S3, generating a thematic information enhanced raster data set based on the weak information planar vector feature set;

[0045] S4, collecting remote sensing images corresponding to the available ground object information, and performing feature segmentation to obtain an object monomer patch set;

[0046] S5, weak information remote sensing extraction is performed on the object single body spot set based on each thematic layer in the grid data set.

[0047] The available ground feature information in step S1 of the embodiment of the application is ground feature element data with a spatial size less than 1 hectare in an open source map and an information accuracy greater than 80% when a random type test is performed.

[0048] Specifically, open source data of a mobile phone commonly used Internet map and navigation map is used to perform spatial random inspection on the element types and attribute contents of various maps, the number of random points is at least greater than 200, and the effectiveness, accuracy, completeness, consistency and timeliness of the relevant data are checked one by one, the information accuracy of roads, traffic auxiliary facilities, construction land, new energy land (hydroelectric power station, photovoltaic, wind power, etc.), factories, mining sites and other land types or ground features with a spatial size less than 1 hectare in the random test is evaluated, and the vector data in the element layer with a ground feature information accuracy greater than 80% is used as available ground feature information for weak information ground feature remote sensing extraction.

[0049] Step S2 of the embodiment of the application is specifically:

[0050] S21, spatial consistency processing is performed on the available ground feature information;

[0051] S22, the point-like vector data and line-like vector data with spatial consistency are classified, and a point-to-surface and line-to-surface vector conversion method is used to construct a weak information surface-like vector element set.

[0052] Step S21 of the embodiment is specifically:

[0053] A GIS command of projection conversion is performed on the available ground feature information, a transverse Mercator projection mode is used for projection processing, the central meridian position range of the processed vector data is within ±3° of the central meridian, and the spatial consistency of the available ground feature information is ensured; the available ground feature information includes important open source vector data such as place names, locations, roads and traffic facilities in an electronic map.

[0054] In step S22 of the embodiment, after the point-like vector data is classified, a point-to-surface vector conversion method is used to generate a corresponding two-dimensional surface-like vector element layer, and a corresponding category digital code is assigned in the attribute table; the point-like vector data includes place names, locations and traffic auxiliary facilities, and is classified and summarized according to main types such as industry, traffic, commerce, culture, education, health, residence and urban green land.

[0055] For linear vector data, after classification, different length buffer analysis is carried out to form planar vector data, and corresponding category digital code is assigned in the attribute table; wherein, according to the function or administrative level of the road, the road data is classified as expressway, national road, provincial road, and 50m, 30m, 20m, 10m buffer analysis is carried out to form planar vector data.

[0056] For the existing planar vector data in the available ground object information, corresponding category digital code is assigned in the attribute table; the existing planar vector data includes residential area and industrial area range, etc.

[0057] The step S3 of the embodiment of the application is specifically:

[0058] The vector data in the weak information planar vector element layer is subjected to grid conversion by using the nearest neighbor pixel method, and then the converted two-dimensional grid data is superimposed on the three-dimensional space according to the XY plane coordinate values of each layer, and the grid data set of the thematic information enhancement is generated by using the correlation in space;

[0059] In the grid conversion, the attribute value of the vector data is converted into the grid cell value.

[0060] Specifically, the vector-to-grid method provided by the open source GIS software can be used, wherein the input element refers to the vector file to be converted, the output element refers to the grid file generated by conversion, the interpolation method used for conversion is selected as the nearest neighbor pixel method, the grid size set for interpolation processing is consistent with the remote sensing image used for interpretation; the number of thematic layers of the formed grid data set is consistent with the number of two-dimensional grid data superimposed on the three-dimensional space.

[0061] The step S4 of the embodiment of the application is specifically:

[0062] S41, a multi-scale segmentation parameter is set, including a scale factor, a color factor and a smoothness factor;

[0063] The value range of the scale factor is (10, 100) with a step of 10; the value range of the color factor is (0, 1) with a step of 0.1; the value range of the smoothness factor is (0, 1) with a step of 0.1.

[0064] S42, the multi-scale segmentation parameters are configured and combined, the remote sensing image is subjected to small area multi-scale segmentation by using the optimal segmentation parameter, and then the object monomer patch set reflecting the edge features of the weak information ground objects in the remote sensing image is obtained.

[0065] When the number of the ground object types in the segmentation unit is not more than 2 and the error between the edge of the segmentation unit and the boundary of the map spot in the remote sensing image is not more than 0.3 pixel size, the corresponding multi-scale segmentation parameter combination is the optimal segmentation parameter; the map spot is a fine object such as a road or a wind power station in the remote sensing image.

[0066] The step S5 of the embodiment of the application is specifically:

[0067] Taking the map spots in the object monomer spot set as the minimum processing unit, using the object-oriented image analysis method, using each thematic layer in the raster data set, the spatial mode filtering is performed on the weak information ground object remote sensing extraction.

[0068] The formula for performing the spatial mode filtering is:

[0069]

[0070]

[0071] In the formula, M0 is the mode value calculated by the filtering unit, L is the lower limit of the group where the mode is located, U is the upper limit of the group where the mode is located, f b is the difference between the number of the group where the mode is located and the number of the adjacent group of the lower limit, f a is the difference between the number of the group where the mode is located and the number of the adjacent group of the upper limit, and i is the distance of the group.

[0072] In the embodiment, according to the mode filtering method, the thematic layers of the raster data set are used to sequentially complete the interpretation of the weak information ground classes such as roads, urban land use types, new energy land (hydroelectric power station, photovoltaic, wind power, etc.), factories, and mining sites, gradually improve the identification accuracy of the weak information ground objects, and finally obtain high-quality remote sensing interpretation results to meet the user business service needs.

[0073] In the description of the application, it should be understood that the orientations or positional relationships indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the application. In addition, the terms "first", "second", "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implying the number of the technical features indicated. Therefore, the features limited by "first", "second", "third" can explicitly or implicitly include one or more features.

Claims

1. A weak information ground object remote sensing extraction method, characterized in that, The method comprises the following steps: S1, collecting available ground object information in an open source map; S2, performing data conversion on the classified ground object information to construct a weak information planar vector element set; S3, generating a thematic information enhanced raster data set based on the weak information planar vector element set; S4, collecting remote sensing images corresponding to the available ground object information, and performing feature segmentation on the remote sensing images to obtain an object monomer polygon set; S5, performing weak information ground object remote sensing extraction on the object monomer polygon set based on each thematic layer in the raster data set; The step S2 is specifically: S21, performing spatial consistency processing on the available ground object information; S22, classifying the point vector data and the line vector data with spatial consistency, and respectively using a point-to-surface vector conversion method and a line-to-surface vector conversion method to construct the weak information planar vector element set; The step S21 is specifically: A GIS command of performing projection conversion is executed on the available ground object information, and a transverse Mercator projection mode is used for projection processing, so that the central meridian position range of the processed vector data is within ±3° of the central meridian, so as to ensure the spatial consistency of the available ground object information; In the step S22, for the point vector data, after classification, the point-to-surface vector conversion method is used to generate a corresponding two-dimensional planar vector element layer, and a corresponding category digital code is assigned in the attribute table; For the line vector data, after classification, a buffer analysis is performed according to different lengths to form planar vector data, and a corresponding category digital code is assigned in the attribute table; For the existing planar vector data in the available ground object information, a corresponding category digital code is assigned in the attribute table; The step S3 is specifically: Each vector data in the weak information planar vector element layer is converted into a raster by using a nearest neighbor pixel method, and then the converted two-dimensional raster data is superimposed in a three-dimensional space according to the XY plane coordinate values of each layer, and a thematic information enhanced raster data set is generated by using the spatial correlation thereof; In the raster conversion, the attribute value of the vector data is converted into a raster cell value; The step S5 is specifically: Taking a polygon in the object monomer polygon set as a minimum processing unit, an object-oriented image analysis method is used to perform spatial mode filtering on each thematic layer in the raster data set, and weak information ground object remote sensing extraction is performed. 2.The weak information ground object remote sensing extraction method according to claim 1, characterized in that, The available ground object information in the step S1 is ground element data in the open source map with a spatial size less than 1 hectare and an information accuracy greater than 80% when a random type test is performed. 3.The weak information ground object remote sensing extraction method according to claim 1, characterized in that, The step S4 is specifically: S41, setting multi-scale segmentation parameters, including scale, color and smoothness; S42, configuring and combining the multi-scale segmentation parameters, using optimal segmentation parameters to perform small-area multi-scale segmentation on the remote sensing images, and then obtaining an object monomer polygon set reflecting the edge features of weak information ground objects in the remote sensing images; When the number of ground object types in a segmentation unit is not more than 2 and the error between the edge of the segmentation unit and the polygon boundary in the remote sensing image is not more than 0.3 pixel size, the corresponding multi-scale segmentation parameter combination is the optimal segmentation parameter. 4.The weak information ground object remote sensing extraction method according to claim 1, characterized in that, The formula for performing spatial mode filtering is: wherein is the mode value calculated for the filter unit, is the lower limit of the group where the mode is located, is the upper limit of the group where the mode is located, is the difference between the number of the group where the mode is located and the number of the adjacent group below the lower limit of the group, is the difference between the number of the group where the mode is located and the number of the adjacent group above the upper limit of the group, is the gap where the mode group is located.

Images