Remote sensing information sample selection method and system

Abstract

This invention provides a method and system for selecting remote sensing information samples. The method includes: acquiring high spatial resolution finely corrected satellite imagery, high temporal resolution vegetation index cube data, and a sampling area map; acquiring multiple image regions in the sampling area map, the image regions including a first image region and a second image region; acquiring reference sample points and identification labels corresponding to the reference sample points; classifying the second image region based on a classification model, wherein the classification model is obtained after training based on the reference sample points and the identification labels corresponding to the reference sample points, and the identification labels are predetermined according to the sample type. The remote sensing information sample selection method and system provided by this invention can improve the accuracy, reliability, and efficiency of sample selection.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a remote sensing information sample selection method and system. Background Art

[0002] At present, in the field of remote sensing earth observation intelligent identification, due to the extreme complexity and regional heterogeneity of the earth's surface, there is no general surface cover type classification sample library. Before conducting intelligent identification research, practitioners and researchers of remote sensing earth observation intelligent identification need to first build a sufficient and reliable sample library for specific research goals. The general practice is to use the high recognition characteristics of sub-meter high spatial resolution images for the surface, and professionals with rich remote sensing interpretation experience to carry out sample selection, which takes a long time.

[0003] In addition, in the process of manual sample selection, samples are selected based on the high recognition characteristics of the surface using sub-meter high spatial resolution images. For landforms such as farmland and grassland that are difficult to identify using single-period remote sensing images or seasonal random remote sensing images, there is a great risk to the accuracy of sample selection. Summary of the invention

[0004] The present invention provides a remote sensing information sample selection method, which is used to solve the defects of long time consumption and low accuracy of sample selection in the prior art, realizes rapid selection of remote sensing information samples, and improves the accuracy of sample selection.

[0005] In a first aspect, the present invention provides a remote sensing information sample selection method, comprising: obtaining a high spatial resolution accurately corrected satellite image, a high temporal resolution vegetation index cube data and a sampling area map; the high spatial resolution accurately corrected satellite image, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in spatial position relationship; obtaining multiple image areas in the sampling area map, the image areas including a first image area and a second image area; obtaining a reference sample point and an identification label corresponding to the reference sample point, the reference sample point and the identification label corresponding to the reference sample point are determined based on user input, the reference sample point is determined in the first image area based on the high spatial resolution accurately corrected satellite image, the high temporal resolution vegetation index cube data and one or more preset sample types, each of the sample types corresponds to at least one reference sample point; classifying the second image area based on a classification model, the classification model is obtained after training based on the reference sample point and the identification label corresponding to the reference sample point, and the identification label is predetermined according to the sample type.

[0006] Optionally, the obtaining of reference sample points and identification labels corresponding to the reference sample points includes: obtaining single-period reference sample points input by a user based on the high spatial resolution precisely corrected satellite image, and identification labels corresponding to the single-period reference sample points; and / or obtaining multi-period reference sample points input by a user based on a vegetation time series change curve, and identification labels corresponding to the multi-period reference sample points; wherein the vegetation time series change curve is obtained based on the high temporal resolution vegetation index cube data; the single-period reference sample points are reference sample points corresponding to a sample type that is independent of time changes; and the multi-period reference sample points are reference sample points corresponding to a sample type that is related to time changes.

[0007] Optionally, classifying the second image area based on the classification model includes: inputting the second image area, the high spatial resolution precisely corrected satellite image corresponding to the second image area, and the high temporal resolution vegetation index cube data corresponding to the second image area into the classification model, and outputting the classification results of each second image area; wherein the classification results include classification areas with obvious attribution and classification areas without obvious attribution.

[0008] Optionally, the method further includes: if the classification result includes the area with no obvious attribution classification, obtaining additional reference samples and identification labels corresponding to the additional reference samples; retraining the classification model based on the additional reference samples and the identification labels corresponding to the additional reference samples; using the retrained classification model to classify image areas without sample selection; repeating the above steps until all image areas are traversed; wherein the additional reference samples and the identification labels corresponding to the additional reference samples are determined in the area with no obvious attribution classification based on user input.

[0009] Optionally, the acquiring the multiple image regions in the sampling region map includes: using a preset window to traverse the sampling region map to obtain the multiple image regions.

[0010] Optionally, the method of acquiring high spatial resolution and precisely corrected satellite images includes: acquiring high spatial resolution satellite images, the high spatial resolution satellite images including multispectral images and panchromatic images; performing geometric precision correction on the multispectral images and the panchromatic images to obtain geometric precision corrected multispectral images and geometric precision corrected panchromatic images; performing multispectral and panchromatic fusion on the geometric precision corrected multispectral images and the geometric precision corrected panchromatic images to obtain the high spatial resolution and precisely corrected satellite images.

[0011] Optionally, obtaining high temporal resolution vegetation index cube data includes: obtaining a high temporal resolution satellite image set; performing geometric correction and atmospheric correction on each satellite image in the high temporal resolution satellite image set to obtain a high temporal resolution reflectance data product; performing vegetation index calculation on the high temporal resolution reflectance data product to obtain a high temporal resolution vegetation index data product; and performing time series data superposition on the high temporal resolution vegetation index data product to obtain the high temporal resolution vegetation index cube data.

[0012] In a second aspect, the present invention further provides a remote sensing information sample selection system, comprising: a first acquisition module, used to acquire high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and a sampling area map; the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in spatial position relationship; a second acquisition module, used to acquire multiple image areas in the sampling area map, the image areas include a first image area and a second image area; a third acquisition module, used to acquire reference sample points and identification labels corresponding to the reference sample points, the reference sample points and the identification labels corresponding to the reference sample points are determined based on user input, the reference sample points are determined in the first image area based on the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and one or more preset sample types, each of the sample types corresponds to at least one reference sample point; a model classification module, used to classify the second image area based on a classification model, the classification model is obtained after training based on the reference sample points and the identification labels corresponding to the reference sample points, and the identification labels are predetermined according to the sample types.

[0013] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the remote sensing information sample selection methods described above are implemented.

[0014] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the remote sensing information sample selection methods described above.

[0015] In a fifth aspect, the present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned remote sensing information sample selection methods.

[0016] The remote sensing information sample selection method and system provided by the present invention can well provide surface sample information and seasonal variation characteristics by fusing high spatial resolution precision-corrected satellite images and high temporal resolution vegetation index cube data. In addition, the present invention classifies the areas to be selected samples through a classification model, thereby improving the accuracy, reliability and efficiency of sample selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of the remote sensing information sample selection method provided by an embodiment of the present invention;

[0019] Figure 2 It is a schematic diagram of a process for obtaining a high spatial resolution accurately corrected satellite image provided by an embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of the process of high temporal resolution vegetation index cube data provided by an embodiment of the present invention;

[0021] Figure 4 This is the second flow chart of the remote sensing information sample selection method provided by the embodiment of the present invention;

[0022] Figure 5 is a structural diagram of a remote sensing information sample selection system provided by an embodiment of the present invention;

[0023] Figure 6 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] A large amount of sample data is a necessary input condition for artificial intelligence image recognition in various fields. Sample annotation is a task that requires a lot of time and long-term accumulation. With the rapid development of artificial intelligence in recent years, a large number of sample libraries have been accumulated for the industry and academia to develop and improve various artificial intelligence algorithms in major application fields such as medicine, autonomous driving, and face recognition. However, in the field of intelligent recognition of remote sensing earth observation, due to the extreme complexity and regional heterogeneity of the earth's surface, there is currently no general surface cover type classification sample library. Practitioners and researchers of intelligent recognition of remote sensing earth observation need to first build a sufficient and reliable sample library for specific research goals before conducting intelligent recognition research. The general practice is to use the high recognition characteristics of sub-meter high spatial resolution images for the surface, and professionals with rich experience in remote sensing interpretation to carry out sample selection, which takes a long time.

[0026] In the existing technology, samples are selected based on the high recognition characteristics of the surface using sub-meter high spatial resolution images. Although this can well identify specific artificial targets such as houses, roads and airports, there is a great risk of accuracy in sample selection for objects such as farmland and grassland that are difficult to identify based on single-period remote sensing images or seasonal random remote sensing images. Vegetation such as farmland and grassland needs to be distinguished based on seasonal variation characteristics, which is information that sub-meter high spatial resolution remote sensing images cannot provide.

[0027] On the other hand, even if professionals with rich experience in remote sensing interpretation carry out sample selection, there are still difficulties in ensuring comprehensive coverage of various types of complex surfaces to be characterized; especially when the area of ​​the sample to be selected is large, browsing remote sensing images greatly increases the difficulty of comprehensive characterization of surface types by the selected samples, and then a large number of surface elements with very low recognition probability are generated when the artificial intelligence model is constructed, which limits the accuracy of the artificial intelligence model.

[0028] The embodiments of the present invention propose a remote sensing information sample selection method, which can provide surface sample information and seasonal change characteristics by fusing high spatial resolution precision-corrected satellite images and high temporal resolution vegetation index cube data. In addition, the embodiments of the present invention classify the areas for selecting samples through a classification model, thereby improving the accuracy, reliability and efficiency of sample selection.

[0029] Combine the following Figure 1-Figure 4 The remote sensing information sample selection method provided by the present invention is described.

[0030] refer to Figure 1 , Figure 1 This is one of the flow charts of the remote sensing information sample selection method provided by the embodiment of the present invention. The remote sensing information sample selection method provided by the embodiment of the present invention includes:

[0031] Step 110, obtaining a high spatial resolution accurately corrected satellite image, high temporal resolution vegetation index cube data and a sampling area map; the high spatial resolution accurately corrected satellite image, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship;

[0032] Step 120, obtaining a plurality of image regions in the sampling region map, wherein the image regions include a first image region and a second image region;

[0033] Step 130, obtaining reference sample points and identification labels corresponding to the reference sample points, wherein the reference sample points and the identification labels corresponding to the reference sample points are determined based on user input, and the reference sample points are determined in the first image region based on the high spatial resolution precisely corrected satellite image, the high temporal resolution vegetation index cube data, and one or more preset sample types, and each of the sample types corresponds to at least one reference sample point;

[0034] Step 140, classifying the second image region based on a classification model, wherein the classification model is obtained after training based on the reference samples and identification labels corresponding to the reference samples, and the identification labels are predetermined according to the sample type.

[0035] Specifically, spatial resolution refers to the minimum size of a single object or two adjacent objects that can be recognized by an optical sensor. The higher the spatial resolution, the richer the object morphological information contained in the remote sensing image, and the smaller the target that can be recognized. The high spatial resolution in this embodiment can refer to sub-meter high spatial resolution.

[0036] Temporal resolution refers to the minimum time interval between two adjacent remote sensing observations in the same area. The smaller the time interval, the higher the temporal resolution. The high temporal resolution in this embodiment may refer to a monthly high temporal resolution of ten meters.

[0037] The embodiment of the present invention solves the problem that high spatial and high temporal resolution remote sensing information is difficult to fuse and visualize by using high spatial resolution precisely corrected satellite images, high temporal resolution vegetation index cube data and sampling area maps that correspond to each other in spatial position relationship. By using the comprehensive display technology of high spatial resolution and high temporal resolution image information, it is achieved that multi-source and multi-feature data simultaneously support the judgment of the real situation of the sample points during the selection of sample points. Under the same visualization interface, the accuracy of sample selection is improved through the support of multi-source information fusion, and the time consumption and uncertainty of manual identification of multi-layer images caused by window switching are avoided. In addition, the embodiment of the present invention classifies the area to be selected for the sample through a classification model, thereby improving the accuracy, reliability and efficiency of sample selection.

[0038] Next, possible implementations of the above steps in specific embodiments are further described.

[0039] Step 110, obtaining high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area map; the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship.

[0040] Specifically, the high spatial resolution, precisely corrected satellite image is a spatial resolution satellite image that has been precisely corrected; the high temporal resolution vegetation index cube data refers to the vegetation index data product obtained based on a collection of high temporal resolution satellite images; the sampling area map includes the geographical location information of the area to be sampled, which can be a location map of the area to be sampled, a boundary contour map of the area to be sampled, or a high spatial resolution, precisely corrected satellite image. The sampling area map is used to display the location information of the sampling area, so that the user can intuitively and preliminarily select the sampling location, and then determine the sampling location by combining the high spatial resolution, precisely corrected satellite image and the high temporal resolution vegetation index cube data.

[0041] The situation in which the high spatial resolution precisely corrected satellite image, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship means that any position point in the sampling area map has corresponding data in the high spatial resolution precisely corrected satellite image and the high temporal resolution vegetation index cube data, and the corresponding data is determined based on the spatial relationship.

[0042] In one embodiment, there is a point A on the sampling area map, with longitude and latitude of 116.46° east longitude and 39.92° north latitude. The image data of the longitude and latitude of 116.46° east longitude and 39.92° north latitude in the high spatial resolution accurately corrected satellite image can be directly obtained through the spatial mapping relationship, and the vegetation index data of the longitude and latitude of 116.46° east longitude and 39.92° north latitude in the high temporal resolution vegetation index cube data can also be directly obtained through the spatial mapping relationship. It can be understood that the high spatial resolution accurately corrected satellite image corresponding to point A obtained based on the spatial relationship is an image area centered on point A; the size of the image area can be set by the user according to needs.

[0043] Optionally, by establishing a spatial index, high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area maps are spatially superimposed to obtain high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area maps that correspond to each other in spatial position relationship. Since high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area maps all contain location information, a spatial database can be established based on the three, and the corresponding spatial elements in the three can be quickly located according to the spatial index, thereby improving the speed and efficiency of spatial operations.

[0044] Optionally, high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area maps are spatially superimposed and stored locally or in the cloud for backup.

[0045] In one embodiment, the display interface has a main window and a sub-window. The main window initially displays a sampling area map. When a position in the sampling area map window is clicked, the main window displays a high spatial resolution, precisely corrected satellite image corresponding to the clicked position, and the sub-window displays vegetation index cube data corresponding to the position.

[0046] refer to Figure 2 , Figure 2 The present invention provides a flowchart of obtaining a high spatial resolution accurately corrected satellite image. The process of obtaining a high spatial resolution accurately corrected satellite image includes:

[0047] Step 210, obtaining high spatial resolution satellite images, where the high spatial resolution satellite images include multispectral images and panchromatic images;

[0048] Specifically, obtain sub-meter-level high spatial resolution satellite images, such as Gaofen-2 high spatial resolution images. The Gaofen-2 (GF-2) satellite is equipped with a high-resolution 1-meter panchromatic camera and a 4-meter multispectral camera. The Gaofen-2 high spatial resolution images include Gaofen-2 panchromatic images and Gaofen-2 multispectral images. Optionally, the Gaofen-2 high spatial resolution satellite images can be replaced by other satellite image data sources with sub-meter spatial resolution, such as QuickBird, WordView, and Pleiades satellite images.

[0049] Step 220, performing geometric precision correction on the multispectral image and the panchromatic image to obtain a geometric precision corrected multispectral image and a geometric precision corrected panchromatic image;

[0050] Specifically, geometric precision correction refers to the use of ground control points and geometric correction mathematical models to correct errors caused by non-systematic factors, that is, to eliminate geometric deformation in the image and produce an image that meets the requirements of map projection or graphic expression.

[0051] Step 230 performs multispectral and panchromatic fusion on the geometrically precisely corrected multispectral image and the geometrically precisely corrected panchromatic image to obtain a high spatial resolution precisely corrected satellite image.

[0052] Specifically, ENVI can be used to perform multispectral and panchromatic fusion on geometrically precisely corrected multispectral images and geometrically precisely corrected panchromatic images. The resulting high-spatial-resolution precisely corrected satellite images can not only inherit the multispectral band information, but also further improve the spatial resolution.

[0053] When the multispectral spatial resolution has met the sub-meter standard, the multispectral and panchromatic fusion steps can be omitted.

[0054] refer to Figure 3 , Figure 3 The process of obtaining high temporal resolution vegetation index cube data provided by an embodiment of the present invention is as follows:

[0055] Step 310, obtaining a high temporal resolution satellite image set;

[0056] Specifically, the high temporal resolution satellite image collection can be a 10-meter-level high temporal resolution satellite image, such as the European Space Agency Sentinel Series A / B high temporal resolution satellite image series. Optionally, the 6-day return Sentinel Series 2A / 2B data (spatial resolution 10 / 20 meters) formed by two satellites used to construct high temporal resolution data can also be replaced by the 16-day return U.S. Landsat OLI data (spatial resolution 30 meters). The high temporal resolution satellite image collection can be monthly data used to characterize seasonal change characteristics within a year.

[0057] Step 320, performing geometric precision correction and atmospheric correction on each satellite image in the high temporal resolution satellite image set to obtain a high temporal resolution reflectivity data product;

[0058] Specifically, in the geometric correction stage, the geometrically corrected Sentinel Series 2A / B or US Landsat OLI data in Google Earth Engine (GEE) can be directly called. When the above form is used, the high temporal resolution satellite image can be directly used in the form of geometrically corrected images in the visualization interface of GEE; the reflectivity data can be directly called in the form of Top of atmosphere reflectance (TOA) data products in GEE to obtain real high temporal resolution reflectivity data products.

[0059] Step 330, calculating the vegetation index for the high temporal resolution reflectance data product to obtain a high temporal resolution vegetation index data product;

[0060] Specifically, vegetation indices include Normalized Vegetation Index (NDVI), Ratio Vegetation Index (RVI), Difference Vegetation Index (DVI), Soil Brightness Adjusted Vegetation Index (SAVI), Soil Adjusted Vegetation Index (MSAVI) and Enhanced Vegetation Index (EVI), etc. Optionally, the NDVI index is used in this implementation. NDVI is recognized as one of the most effective parameters for characterizing vegetation changes, which can reflect changes in vegetation greenness. NDVI can reflect the background effects of plant canopies, such as soil, wet ground, snow, dead leaves and coarseness, and is related to vegetation cover.

[0061] High temporal resolution vegetation index data products are product data sets established for high temporal resolution reflectance data products based on vegetation index. They can be made into raster data of various scales. The data sets may include data name, data type, data format, resolution, coverage, coordinate system and vegetation index, etc.

[0062] Step 340 , performing time series data superposition on the high temporal resolution vegetation index data product to obtain high temporal resolution vegetation index cube data.

[0063] Specifically, high temporal resolution vegetation index cube data refers to a product data set obtained by arranging high temporal resolution vegetation index data products in chronological order. The data set may include data name, data type, data format, resolution, coverage, coordinate system, vegetation index and time series, etc.

[0064] Step 120: Acquire multiple image regions in the sampling region map, where the image regions include a first image region and a second image region.

[0065] Optionally, multiple image regions in the sampling region map are obtained, including:

[0066] The sampling region map is traversed using a preset window to obtain multiple image regions.

[0067] Specifically, the size of the traversal window is determined and input according to the area size and the total number of pre-selected samples and their distribution estimates; in actual applications, the preset window traverses the sampling area map displayed in the main window, and divides the image into multiple image areas during the traversal process. During the traversal process, the preset window can enter the next image area of ​​the same size according to the preset order or the order input by the user.

[0068] In one embodiment, a high spatial resolution and precisely corrected satellite image corresponding to the area indicated by the sampling area map is displayed in the main window. The sampling area map is arranged as a bottom layer below the high spatial resolution and precisely corrected satellite image, and the preset window is traversed as a top layer. In this embodiment, while the preset window is traversing the sampling area map, the main window can display the high spatial resolution and precisely corrected satellite image of the corresponding position, so as to facilitate the user to select the sampling point.

[0069] In one embodiment, when the sampling area map is a sampling area contour map, a mixed image of the sampling area contour map and a high spatial resolution precisely corrected satellite image corresponding to the area indicated by the sampling area contour map can be displayed in the main window, that is, the sampling area contour lines are added to the high spatial resolution precisely corrected satellite image, and the preset window can traverse the mixed image.

[0070] Step 130, obtaining reference sample points and identification labels corresponding to the reference sample points, wherein the reference sample points and the identification labels corresponding to the reference sample points are determined based on user input, and the reference sample points are determined in the first image area based on the high spatial resolution precisely corrected satellite image, the high temporal resolution vegetation index cube data and one or more preset sample types, and each of the sample types corresponds to at least one of the reference sample points.

[0071] Optionally, obtaining a reference sample point and an identification tag corresponding to the reference sample point includes:

[0072] Obtaining single-period reference sample points input by a user based on high spatial resolution accurately corrected satellite images, and identification labels corresponding to the single-period reference sample points; and / or

[0073] Acquire multiple reference sample points input by the user based on the vegetation time series change curve, and identification labels corresponding to the multiple reference sample points;

[0074] Among them, the vegetation time series change curve is obtained based on the high temporal resolution vegetation index cube data; the single-period reference sample points are the reference sample points corresponding to the sample type that is independent of temporal changes; the multi-period reference sample points are the reference sample points corresponding to the sample type that is related to temporal changes.

[0075] Specifically, click on a location in the sampling area map displayed in the main window, and the user visually interprets the high spatial resolution and precisely corrected satellite image corresponding to the location displayed in the main window, selects a single-period reference sample point that is independent of time changes, and marks the corresponding identification label, such as a house, road, or airport. After selecting a sample in the current preset window, the user can enter the next preset window area for sample selection, and the preset window area where the reference sample point is selected is the first image area. The high spatial resolution and precisely corrected satellite image displayed in the main window can intuitively reflect the actual surface conditions. Through the high spatial resolution and precisely corrected satellite image, specific artificial targets such as houses, roads, and airports can be well identified, thereby improving the accuracy of sample selection for single-period reference samples.

[0076] The secondary window displays the vegetation index cube data corresponding to the selected position, which can be a high-temporal resolution image value time series curve obtained based on the high-temporal resolution vegetation index cube data. When the temporal resolution is monthly data within a year and the clicked position is vegetation, the displayed curve is a vegetation index curve, such as a seasonal curve of ground feature growth. The horizontal axis of the vegetation index curve is time and the vertical axis is the NDVI value. The user visually interprets the high-spatial resolution precision-corrected satellite image displayed in the main window, preliminarily selects candidate sample points, and determines the selected candidate sample points as multi-period reference sample points related to time changes according to the vegetation index curve displayed in the secondary window, and marks the corresponding identification labels, such as farmland or grassland, woodland, etc. After selecting samples in the current preset window area, the user can enter the next preset window area for sample selection, and the preset window area where the reference sample points are selected is the first image area. The vegetation index curve can reflect the differences in vegetation coverage and growth conditions, reflect the seasonal change characteristics of ground features, and can improve the accuracy of selecting samples of ground features such as farmland, grassland or woodland that change over time.

[0077] Due to the extreme complexity and regional heterogeneity of the earth's surface, there is currently no universal surface cover type classification sample library. In order to ensure the classification accuracy of subsequent classification models, at least one reference sample point should be selected for each sample type to be classified.

[0078] Step 140, classifying the second image region based on a classification model, wherein the classification model is obtained after training based on the reference samples and identification labels corresponding to the reference samples, and the identification labels are predetermined according to the sample type.

[0079] Classifying the second image region based on the classification model includes:

[0080] superimposed image data corresponding to the first and second image regions and outputting classification results of each second image region;

[0081] The classification results include areas with obvious classification affiliation and areas with no obvious classification affiliation.

[0082] Specifically, the classification model is used to classify the second image area and the superimposed image data corresponding to the second image area. The classification model can be a supervised classification model, an unsupervised classification model, or a classification model based on a neural network. The classification model is trained using reference samples and identification labels corresponding to the reference samples. The second image area and the superimposed image data corresponding to the second image area are input into the classification model, and the classification results of each second image area are output. The second image area refers to an area where no reference sample has been selected.

[0083] The classification results are represented by the probability of belonging to each category, such as the probability of the point to be classified being grassland is 90%, and the probability of being a house is 8%. The classification results include areas with obvious attribution and areas without obvious attribution. Areas with obvious attribution indicate that the ground feature samples in the area are well representative, while areas with no obvious attribution indicate that the ground feature samples in the area are poorly representative and need to increase samples to improve sample representativeness.

[0084] The probability difference threshold can be preset to determine the type of classification area. The area with a probability difference greater than the preset probability difference threshold is a clearly classified area, and the area with a probability difference less than or equal to the preset probability difference threshold is a non-clearly classified area. The probability difference is the difference between the maximum probability and the second largest probability in the classification result. For example, if the preset probability difference threshold is 5%, the classification result of point A is: 96% grass probability, 4% road probability, 1% airport probability, and the probability difference is 92%, which is greater than the preset probability difference threshold, and point A is judged to be a clearly classified area; the classification result of point B is: 46% grass probability, 44% road probability, 41% airport probability, and the probability difference is 2%, which is less than the preset probability difference threshold, and point B is judged to be a non-clearly classified area; the classification result of point C is: 50% grass probability, 45% road probability, 41% airport probability, and the probability difference is 5%, which is equal to the preset probability difference threshold, and point C is judged to be a non-clearly classified area. By highlighting the non-clearly classified area, a key area to be selected is provided for further manual point selection.

[0085] In one embodiment, the remote sensing information sample selection method further includes:

[0086] Step 410, if the classification result includes an area without obvious attribution classification, obtain additional reference sample points and identification labels corresponding to the additional reference sample points;

[0087] The additional reference points and the identification labels corresponding to the additional reference points are determined in the region without obvious attribution classification based on user input.

[0088] Specifically, if the classification result includes an area with no obvious attribution classification, the user shall visually interpret the high spatial resolution and accurately corrected satellite image of the area with no obvious attribution classification, select the single-period reference sample points that are not related to seasonal characteristics among the additional reference sample points, and the corresponding identification labels. The user shall visually interpret the high spatial resolution and accurately corrected satellite image of the area with no obvious attribution classification, preliminarily select the candidate sample points, and determine the selected candidate sample points as multi-period reference sample points related to time changes and the corresponding identification labels according to the corresponding vegetation index curve graph displayed in the sub-window.

[0089] Step 420, retraining the classification model based on the additional reference samples and the identification labels corresponding to the additional reference samples;

[0090] Step 430, using the retrained classification model to classify the image area where no sample selection is performed;

[0091] Step 440, repeat the above steps until all image areas are traversed;

[0092] If the retrained classification model is used to classify the image area without sample selection, if the classification result includes an area with no obvious classification, the additional reference samples and the identification labels corresponding to the additional reference samples are obtained again, the additional reference samples and the identification labels corresponding to the additional reference samples are used to retrain the classification model, and the trained classification model is used to classify the remaining image area without sample selection.

[0093] In one embodiment, if the number of selected samples is insufficient after traversing all image areas, the remote sensing information sample selection method provided by the embodiment of the present invention may be repeatedly used until the number of selected samples reaches the required number.

[0094] On the one hand, the embodiment of the present invention obtains superimposed image data by superimposing high spatial resolution precision-corrected satellite images and high temporal resolution vegetation index cube data, thereby solving the problem that high spatial and high temporal resolution remote sensing information is difficult to integrate and visualize. The comprehensive display technology of high spatial resolution and high temporal resolution image information realizes that multi-source and multi-feature data simultaneously support the judgment of the real situation of the sample points during the selection of sample points. There is no need to repeatedly switch windows, call images separately, or repeatedly switch multiple layers of images of different seasons for manual judgment. The main window and sub-window with spatial position links are used to display single-period high-recognition images and multi-period seasonal phase change curves representing vegetation growth characteristics respectively. Under the same visualization interface, under the support of multi-source information fusion, the selection of all types of samples is completed at one time and accurately, including types such as buildings that require high-recognition image recognition and types such as farmland, woodland and grassland that require seasonal characteristics to identify. On the one hand, the accuracy of sample selection is improved, and on the other hand, the time consumption and uncertainty of window switching and manual identification of multiple layers of images are saved.

[0095] On the other hand, the embodiment of the present invention realizes a human-computer interactive sample selection method in which the computer uses manually selected samples to perform calculations and then provides feedback to humans on the distribution of points of the type to be determined by the optimization algorithm classification. The human-computer interactive semi-intelligent sample selection method solves the problem of blindly one-way selection by visual interpretation experts who do not know where the algorithm needs to add sample points in order to improve the reliability of classification when performing sample selection in the remaining area after completing sample selection in the local area. In the point selection process after selecting at least one sample point for each type to be classified, the algorithm achieves the purpose of giving periodic feedback to visual interpretation point selection experts by pre-performing a pre-classification and highlighting pixels without obvious classification, and guides point selection experts to pay attention to the area of ​​points to be selected that needs to rely on expert experience for point selection. By following the method of optimizing the point selection based on the need for algorithm optimization in the point selection process, it can ensure that the selected sample points fully cover various types and various forms of complex and diverse expressions of similar objects, thereby improving the accuracy of the artificial intelligence model.

[0096] It should be added that reference Figure 4 , Figure 4 This is the second flow chart of the remote sensing information sample selection method provided by the embodiment of the present invention. The embodiment of the present invention provides a remote sensing information sample selection method, including:

[0097] The steps for comprehensive display of high spatial resolution and high temporal resolution image information are as follows: first, the GF-2 high spatial resolution images and the ESA Sentinel Series A / B high temporal resolution (usually monthly data within a year used to characterize seasonal change characteristics) images are processed into a series of precisely geometrically corrected images with a unified spatial position reference, which are stored locally or in the cloud for backup; the display interface has a main and a secondary window, the main window displays high spatial resolution remote sensing images, and when the mouse clicks on a specific position in the main window, the secondary window displays the numerical time series curve of the high temporal resolution image at that position; when the temporal resolution is monthly data within a year and the clicked position is vegetation, the displayed curve is the growth season curve of the ground object.

[0098] The fixed-size window traverses the area in sequence to select evenly distributed samples. First, determine and enter the size of the traversal window based on the area size and the total number of pre-selected samples and their distribution estimates. After the sample selection is completed within the current window range, click the "NEXT" button to replace the traditional method of manually dragging to enter the next area to be selected, and enter the next window of the same area size to start selecting samples.

[0099] The classification algorithm calculates and highlights the prompt display steps for the low-value area of ​​classification probability. First, at least one or more sample points are selected for each type to be classified. In the subsequent full-process manual selection of sample points, a pre-classification is performed in advance according to the selected sample points. The classification results are represented by the probability of belonging to each category. By highlighting the pixels with equal probability of belonging to each category and no obvious belonging to a certain type, the key candidate area prompts are provided for further manual point selection. The above steps can be repeated during the point selection process to ensure that the samples selected in the entire domain can truly and effectively help the automatic classification algorithm make a high-level type attribution decision.

[0100] The remote sensing information sample selection system provided by the present invention is described below. The remote sensing information sample selection system described below and the remote sensing information sample selection method described above can be referenced to each other.

[0101] refer to Figure 5 , Figure 5 Schematic diagram of the structure of a remote sensing information sample selection system provided by an embodiment of the present invention. The embodiment of the present invention provides a remote sensing information sample selection system, including: a first acquisition module 510, a second acquisition module 520, a third acquisition module 530 and a model classification module 540.

[0102] The first acquisition module 510 is used to acquire a high spatial resolution accurately corrected satellite image, a high temporal resolution vegetation index cube data and a sampling area map; the high spatial resolution accurately corrected satellite image, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship;

[0103] A second acquisition module 520, configured to acquire a plurality of image regions in the sampling region map, wherein the image regions include a first image region and a second image region;

[0104] A third acquisition module 530 is used to acquire reference sample points and identification tags corresponding to the reference sample points, wherein the reference sample points and the identification tags corresponding to the reference sample points are determined based on user input, and the reference sample points are determined in the first image region based on the high spatial resolution accurately corrected satellite image, the high temporal resolution vegetation index cube data, and one or more preset sample types, and each of the sample types corresponds to at least one reference sample point;

[0105] The model classification module 540 is used to classify the second image area based on a classification model, wherein the classification model is obtained after training based on the reference samples and the identification labels corresponding to the reference samples, and the identification labels are predetermined according to the sample type.

[0106] It should be noted here that the above-mentioned device provided in the embodiment of the present invention can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.

[0107] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the remote sensing information sample selection method, which includes:

[0108] Acquire high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area map; the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship;

[0109] Acquire a plurality of image regions in the sampling region map, wherein the image regions include a first image region and a second image region;

[0110] Acquire a reference sample point and an identification tag corresponding to the reference sample point, wherein the reference sample point and the identification tag corresponding to the reference sample point are determined based on a user input, and the reference sample point is determined in the first image region based on the high spatial resolution precisely corrected satellite image, the high temporal resolution vegetation index cube data, and one or more preset sample types, and each of the sample types corresponds to at least one reference sample point;

[0111] The second image region is classified based on a classification model, where the classification model is obtained after training based on the reference sample points and identification labels corresponding to the reference sample points, and the identification labels are predetermined according to the sample type.

[0112] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the remote sensing information sample selection method provided by the above methods, which includes: obtaining high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and a sampling area map; spatially superimposing the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map to obtain superimposed image data; the superimposed image data includes high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and A sampling area map; obtaining multiple image areas in the sampling area map, the image areas including a first image area and a second image area; obtaining reference sample points and identification labels corresponding to the reference sample points, the reference sample points and the identification labels corresponding to the reference sample points are determined based on user input, the reference sample points are determined in the first image area based on one or more preset sample types and the superimposed image data, each of the sample types corresponds to at least one reference sample point; classifying the second image area based on a classification model, the classification model is obtained after training based on the reference sample points and the identification labels corresponding to the reference sample points, and the identification labels are predetermined according to the sample types.

[0114] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the remote sensing information sample selection method provided by the above-mentioned methods, the method comprising: obtaining high spatial resolution precisely corrected satellite images, high temporal resolution vegetation index cube data and a sampling area map; spatially superimposing the high spatial resolution precisely corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map to obtain superimposed image data; the superimposed image data comprises high spatial resolution precisely corrected satellite images, high temporal resolution vegetation index cube data and a sampling area map; obtaining the sampling area The multiple image areas in the figure include a first image area and a second image area; taking reference samples and identification labels corresponding to the reference samples, wherein the reference samples and the identification labels corresponding to the reference samples are determined based on user input, and the reference samples are determined in the first image area based on one or more preset sample types and the superimposed image data, and each sample type corresponds to at least one reference sample; classifying the second image area based on a classification model, wherein the classification model is obtained after training based on the reference samples and the identification labels corresponding to the reference samples, and the identification labels are predetermined according to the sample types.

[0115] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0116] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote sensing information sample selection method, characterized in that: include: Acquire high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and sampling area map; the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship; Acquire a plurality of image regions in the sampling region map, wherein the image regions include a first image region and a second image region; Acquire a reference sample point and an identification tag corresponding to the reference sample point, wherein the reference sample point and the identification tag corresponding to the reference sample point are determined based on a user input, and the reference sample point is determined in the first image region based on the high spatial resolution precisely corrected satellite image, the high temporal resolution vegetation index cube data, and one or more preset sample types, and each of the sample types corresponds to at least one reference sample point; classifying the second image region based on a classification model, wherein the classification model is obtained after training based on the reference sample points and identification labels corresponding to the reference sample points, and the identification labels are predetermined according to the sample type; The obtaining of reference sample points and identification labels corresponding to the reference sample points includes: Acquire a single-period reference sample point input by a user based on the high spatial resolution accurately corrected satellite image, and an identification label corresponding to the single-period reference sample point; and / or Acquire multiple reference sample points input by a user based on a vegetation time series change curve, and identification tags corresponding to the multiple reference sample points; The vegetation time series change curve is obtained based on the high time resolution vegetation index cube data; the single-period reference sample point is a reference sample point corresponding to a sample type that is not related to time changes; the multi-period reference sample point is a reference sample point corresponding to a sample type that is related to time changes; The classifying the second image region based on the classification model includes: Inputting the second image area, the high spatial resolution accurately corrected satellite image corresponding to the second image area, and the high temporal resolution vegetation index cube data corresponding to the second image area into the classification model, and outputting the classification results of each second image area; The classification results include areas with obvious attribution and areas without obvious attribution; The method further comprises: If the classification result includes the region without obvious attribution, obtaining additional reference sample points and identification labels corresponding to the additional reference sample points; retraining the classification model based on the additional reference samples and identification labels corresponding to the additional reference samples; Using the retrained classification model to classify the image area where no sample selection is performed; Repeat the above steps until all image areas are traversed; The additional reference points and the identification labels corresponding to the additional reference points are determined in the region without obvious attribution classification based on user input.

2. The remote sensing information sample selection method according to claim 1, characterized in that: The acquiring of a plurality of image regions in the sampling region map comprises: The sampling region map is traversed using a preset window to obtain a plurality of image regions.

3. The remote sensing information sample selection method according to claim 1, characterized in that: The obtaining of high spatial resolution accurately corrected satellite images comprises: Acquire high spatial resolution satellite images, wherein the high spatial resolution satellite images include multispectral images and panchromatic images; Performing geometric precision correction on the multispectral image and the panchromatic image to obtain a geometric precision corrected multispectral image and a geometric precision corrected panchromatic image; The geometrically precisely corrected multispectral image and the geometrically precisely corrected panchromatic image are subjected to multispectral and panchromatic fusion to obtain the high spatial resolution precisely corrected satellite image.

4. The remote sensing information sample selection method according to claim 1, characterized in that: The step of obtaining high temporal resolution vegetation index cube data comprises: Acquire a collection of high temporal resolution satellite images; Performing geometric precision correction and atmospheric correction on each satellite image in the high temporal resolution satellite image set to obtain a high temporal resolution reflectivity data product; Calculating the vegetation index on the high temporal resolution reflectance data product to obtain a high temporal resolution vegetation index data product; The high temporal resolution vegetation index data product is subjected to time series data superposition to obtain the high temporal resolution vegetation index cube data.

5. A remote sensing information sample selection system, characterized in that: include: A first acquisition module is used to acquire high spatial resolution accurately corrected satellite images, high temporal resolution vegetation index cube data and a sampling area map; the high spatial resolution accurately corrected satellite images, the high temporal resolution vegetation index cube data and the sampling area map correspond to each other in terms of spatial position relationship; A second acquisition module, used to acquire a plurality of image regions in the sampling region map, wherein the image regions include a first image region and a second image region; a third acquisition module, configured to acquire reference sample points and identification tags corresponding to the reference sample points, wherein the reference sample points and the identification tags corresponding to the reference sample points are determined based on user input, and the reference sample points are determined in the first image region based on the high spatial resolution precisely corrected satellite image, the high temporal resolution vegetation index cube data, and one or more preset sample types, and each of the sample types corresponds to at least one reference sample point; The obtaining of reference sample points and identification labels corresponding to the reference sample points includes: Acquire a single-period reference sample point input by a user based on the high spatial resolution accurately corrected satellite image, and an identification label corresponding to the single-period reference sample point; and / or Acquire multiple reference sample points input by a user based on a vegetation time series change curve, and identification tags corresponding to the multiple reference sample points; The vegetation time series change curve is obtained based on the high time resolution vegetation index cube data; the single-period reference sample point is a reference sample point corresponding to a sample type that is not related to time changes; the multi-period reference sample point is a reference sample point corresponding to a sample type that is related to time changes; a model classification module, configured to classify the second image region based on a classification model, wherein the classification model is obtained after training based on the reference sample points and identification labels corresponding to the reference sample points, and the identification labels are predetermined according to the sample type; The classifying the second image region based on the classification model includes: Inputting the second image area, the high spatial resolution accurately corrected satellite image corresponding to the second image area, and the high temporal resolution vegetation index cube data corresponding to the second image area into the classification model, and outputting the classification results of each second image area; The classification results include areas with obvious attribution and areas without obvious attribution; If the classification result includes the region without obvious attribution, obtaining additional reference sample points and identification labels corresponding to the additional reference sample points; retraining the classification model based on the additional reference samples and identification labels corresponding to the additional reference samples; Using the retrained classification model to classify the image area where no sample selection is performed; Repeat the above steps until all image areas are traversed; The additional reference points and the identification labels corresponding to the additional reference points are determined in the region without obvious attribution classification based on user input.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the remote sensing information sample selection method according to any one of claims 1 to 4 are implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the remote sensing information sample selection method according to any one of claims 1 to 4 are implemented.

Images