Multi-source remote sensing image accurate water body extraction method and system based on space-time constraint

Abstract

The application discloses a kind of multi-source remote sensing image accurate water body extraction method and system based on space-time constraint, using the characteristics that target area exists rich water period and the period that sun direct point is located in the hemisphere where the area is, first, monthly water body of target area each year rich water period is extracted, a brand-new maximum water area range is made as the space constraint of water body extraction, then different parameters are set for the rich water period and dry period of target area respectively to be used as the time constraint of water body extraction, while removing mountain shadow and building shadow, the integrity of water body extraction result is guaranteed, and it can be applicable to water body extraction in different periods.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water resource remote sensing monitoring, and particularly relates to a multi-source remote sensing image accurate water body extraction method and system based on space-time constraints. BACKGROUND

[0002] Surface water is an important indicator of water resources, and plays an important role in climate regulation, biogeochemical cycling and surface energy balance. Timely and effective monitoring of the dynamic situation of large-scale surface water is helpful for the scientific management and utilization of water resources, and is of great significance to the sustainable development of regional ecosystems and economy. Compared with traditional surface water detection and extraction methods, remote sensing images have the characteristics of strong timeliness and wide coverage, and can quickly obtain large-scale ground object information. With the development and popularization of remote sensing science and technology, high-resolution remote sensing images have become one of the main data for surface water detection and extraction, and can provide more comprehensive, accurate and near real-time water body information. At present, the methods for extracting and detecting surface water from remote sensing images mainly include the following: (1) threshold method: including single-band method and multi-band method, among which NDWI, MNDWI and other water body index methods are more commonly used, which select the band closely related to water body identification, construct different water body index models and give the corresponding threshold value to realize the extraction of water body information; (2) classifier method: including SVM, decision tree and object-oriented method, such as the commonly used random forest classification, which uses multiple bands or features of remote sensing images to construct decision trees, and classifies water bodies by integrating the results of multiple decision trees; (3) other methods: including deep learning, mixed pixel decomposition and other methods, such as convolutional neural network (CNN) which is increasingly used to identify water bodies and other objects from high-resolution satellite images, and has produced the most advanced results. However, many of the above methods usually have a key problem, that is, in some mountainous and urban areas, it is difficult to accurately distinguish water bodies from mountain shadows, building shadows and other noises.

[0003] In order to solve these problems, many methods have been proposed to accurately remove the interference of mountain shadows, building shadows and other noises while extracting water bodies. Liu et al. proposed a new shadow water index (NSWI), which fully utilizes the rich spectral, shape and texture information of high-resolution images, so that water bodies and shadows are effectively distinguished, and the classification accuracy is improved. Tatar Nurollah uses adjacent pixel context information to propose an object-based shadow detection framework, which effectively solves the confusion between water bodies and shadows. Huang et al. proposed a novel pixel-object dual-layer machine learning framework for extracting water and identifying water types from optical high-resolution remote sensing images in urban areas, which achieved good results in water body extraction in complex urban environments.

[0004] However, the above method still has obvious limitations: (1) the mountain shadow or building shadow removal is not clean enough, or the mountain shadow and building shadow interference cannot be accurately removed at the same time; (2) some water body pixels are also damaged while removing the mountain shadow and building shadow, so that the water body extraction result is not complete; (3) for some seasonal features obvious, change more sensitive water body object, cannot be applied to different periods such as wet season and dry season at the same time. Therefore, how to accurately remove the interference of mountain shadow, building shadow and other noises, and can cope with the seasonal changes and differences of different periods of water body area, so as to ensure the accuracy and integrity of water body extraction is the problem to be solved at present. SUMMARY

[0005] In view of the problem that the existing water body extraction method cannot accurately remove the interference of mountain shadow, building shadow and other noises while ensuring the completeness of water body extraction, and is suitable for different periods such as wet season and dry season at the same time, based on the previous related research, the present application proposes a multi-source remote sensing image accurate water body extraction method based on space-time constraint. By using the characteristics that the target area has a wet season and the sun's direct point is located in the hemisphere where the target area is located, the monthly water body of the target area in the wet season of each year is first extracted, a new maximum water area range is made as the spatial constraint of water body extraction, and then different parameters are set for the wet season and dry season of the target area as the time constraint of water body extraction, so that the mountain shadow and building shadow can be removed while ensuring the completeness of water body extraction result, and the water body extraction can be applied to different periods.

[0006] The present application provides a multi-source remote sensing image accurate water body extraction method based on space-time constraint, comprising the following steps:

[0007] Step 1, extract the monthly water body range of the target area in the wet season of the study period of Sentinel-2 remote sensing image, and merge it with the maximum water area range of the target area in the maximum water area range layer of JRC GSW data set, to generate the spatial constraint of water body extraction;

[0008] Step 1.1, use random forest classification method to extract the monthly water body in the wet season of the target area in the study period;

[0009] Step 1.2, remove the mountain shadow, building and building shadow existing in the water body extracted in step 1.1;

[0010] Step 1.3, generate the spatial constraint;

[0011] Step 2, use random forest classification method and modified normalized difference water index (MNDWI) to process Sentinel-2 remote sensing image, and extract the monthly water body in the study period of the target area;

[0012] Step 2.1, remove the cloud and cloud shadow in the Sentinel-2 remote sensing image, and synthesize the monthly median image;

[0013] Step 2.2, combine random forest classification and MNDWI method to extract water body;

[0014] Step 2.3, remove the residual mountain shadow, building and building shadow in the water body extracted in step 2.2;

[0015] Step 3, use Sentinel-1 remote sensing image to supplement the extraction of monthly water body in the area covered by cloud and cloud shadow and data missing in the target area of Sentinel-2 remote sensing image;

[0016] Step 3.1, make Sentinel-1 remote sensing image feature classification sample;

[0017] Step 3.2, filter processing of Sentinel-1 remote sensing image;

[0018] Step 3.3, combine random forest classification and threshold method to extract water body;

[0019] Step 3.4, remove the residual mountain shadow, building and building shadow in the water body extracted in step 3.3;

[0020] Step 4, combine the water body data extracted in step 2 and step 3 to generate the final monthly water body extraction result of the target area.

[0021] Moreover, in step 1.1, a large number of target area Sentinel-2 remote sensing image feature classification samples are manually drawn, including water, land and cloud. The cloud and cloud shadow in the Sentinel-2 remote sensing image of the target area in the wet season are removed, and the median synthesis image of each month is generated. Based on the manually drawn feature classification sample, the monthly water body in the wet season of the target area is automatically extracted by using the random forest classification method.

[0022] Moreover, in step 1.2, the area with an altitude higher than N1 m and a slope greater than N2 degrees is regarded as the mountain shadow area, and the DEM data and the slope data calculated therefrom are used for masking to remove a small amount of mountain shadow in the monthly water body extracted in step 1.1, N1 and N2 are threshold values; the global human settlement layer GHSL in JRC GSW data set is used for masking to remove building shadow and misextracted building noise, and a more accurate and complete monthly water body data of the target area in the wet season during the study period is generated.

[0023] Moreover, in step 1.3, the monthly water body data of the target area in the wet season during the study period, from which the mountain shadow, building and building shadow are removed in step 1.2, is combined with the maximum water area range of the target area in the maximum water area range layer of JRC GSW data set to generate a new maximum water area range, and the spatial constraint of water body extraction is obtained.

[0024] Furthermore, in step 2.1, the cloud and cloud shadow removal processing is performed on the Sentinel-2 remote sensing images in the target area research period, the median composite images of each month are generated, and the areas covered by clouds and cloud shadows and the areas with missing data are marked as Nodata areas.

[0025] Furthermore, in step 2.2, the random forest classification method and the MNDWI method are first used to process the median composite images of each month of Sentinel-2 to extract water bodies, wherein the random forest classification method is based on the artificial drawing of the land cover classification samples in step 1.1 to identify water bodies and identify the cloud-covered areas that are not removed in step 2.1, and mark them as Nodata areas; the segmentation threshold of the MNDWI method is automatically determined by the OTSU algorithm; the water body results obtained by the two methods are marked as W1 and W2, and then the spatial constraints generated in step 1 are used to mask W1 and W2 to retain the water bodies within the spatial constraints to remove mountain shadow and building shadow. Using the SWO (Surface water occurrence) layer in the JRC GSW dataset, the target area is divided into high-frequency water area and other area by setting a threshold, the area with SWO value greater than the set threshold δ in the target area is regarded as high-frequency water area, and the area with SWO value less than or equal to the set threshold δ in the target area is regarded as other area. For the high-frequency water area, the union of W1 and W2 is taken as the final water body result, while in the other area, the intersection of W1 and W2 is taken. In addition, δ is adjusted higher in the dry season and lower in the wet season as a time constraint to adapt to different periods, further improving the water body extraction accuracy. Finally, the water body extraction results of the high-frequency water and other areas are combined to obtain the further water body extraction results of the Sentinel-2 remote sensing images in the target area research period.

[0026] Furthermore, in step 2.3, the area with an elevation higher than N1 m and a slope greater than N2 degrees is regarded as a mountain shadow area, and the DEM data and the slope data calculated therefrom are used to mask the residual small amount of mountain shadow in the monthly water bodies obtained in step 2.2, N1 and N2 are set thresholds; at the same time, the global human settlement layer GHSL in the JRC GSW dataset is used to mask, further removing residual building shadow and mis- extracted building noise, to generate the final water body extraction results of the Sentinel-2 monthly median composite images in the target area research period.

[0027] Furthermore, in step 3.1, a large number of Sentinel-1 remote sensing image land cover classification samples in the target area are manually drawn, including water and land.

[0028] Moreover, in step 3.2, all Sentinel-1 remote sensing images of each month in the target area research period are synthesized into one scene to obtain monthly average synthetic images, and the monthly average synthetic images are filtered to reduce the influence of noise.

[0029] Moreover, in step 3.3, the random forest classification method and the threshold segmentation method are used to process the monthly average synthetic images of Sentinel-1 to extract water bodies, wherein the random forest classification method is based on the ground feature classification samples manually drawn in step 3.1, the threshold segmentation method is based on the VV and VH bands of the Sentinel-1 remote sensing images, and the OTSU algorithm is used to automatically determine the segmentation threshold, and when the values of the VV and VH bands are both less than the threshold value determined by the OTSU algorithm, the water body is determined; the water bodies extracted by the two methods are marked as W3 and W4 respectively, then the Nodata region marked in step 2 is used to mask W3 and W4 to retain the water bodies in the Nodata region, and then the spatial constraint generated in step 1 is used to mask the water bodies in the Nodata region to retain the water bodies within the spatial constraint range to remove the mountain shadow and the building shadow. With the aid of the SWO layer in the JRC GSW data set, the region with a SWO value greater than the threshold value δ in the target area is regarded as a high-frequency water region, and the region with a SWO value less than or equal to the threshold value δ in the target area is regarded as another region, for the high-frequency water region, the union of W3 and W4 is taken as the final water body result, and for the other region, the intersection of W3 and W4 is taken. In addition, the value of δ is adjusted to be higher in the dry season and lower in the wet season, which is used as a time constraint to adapt to different periods, further improving the water body extraction accuracy. Finally, the water body extraction results of the high-frequency water and the other region are combined to obtain the further water body extraction result of the Nodata region in the target area.

[0030] Moreover, in step 3.4, the region with an altitude higher than N1 m and a slope greater than N2 degrees is regarded as a mountain shadow region, and the DEM data and the slope data calculated therefrom are used to mask and remove a small amount of residual mountain shadow in the monthly water body obtained in step 3.3, N1 and N2 are threshold values; at the same time, the global human settlement layer GHSL in the JRC GSW data set is used to mask and further remove the building shadow and the mis-removed building noise, to obtain the final water body extraction result of the Nodata region in the target area research period.

[0031] The application also provides a multi-source remote sensing image accurate water body extraction system based on space-time constraints, which is used to realize the multi-source remote sensing image accurate water body extraction method based on space-time constraints.

[0032] Moreover, comprising a processor and a memory, the memory is used to store program instructions, and the processor is used to call the program instructions in the memory to execute the method for extracting water body from multi-source remote sensing image based on space-time constraint.

[0033] Compared with the prior art, the present application has the following advantages:

[0034] 1) The present application takes advantage of the fact that the target area has a wet season and the sun's direct point is located in the hemisphere where the target area is located, and a new maximum water area range is made as a spatial constraint for water body extraction, which can accurately remove the influence of mountain shadow and building shadow, and can maximize the protection of water body while removing mountain shadow and building shadow.

[0035] 2) Considering the obvious seasonal characteristics of the target area, the water body often shrinks seriously in the dry season, and the present application uses different threshold values in the wet season and the dry season to divide the high-frequency water area and other areas, and takes different water body extraction measures in the two areas as the time constraint for water body extraction, which can improve the robustness of the method and make it applicable to some water body object extraction with obvious seasonal characteristics and sensitive changes. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0037] Figure 1 The overall flowchart of the method for extracting water body from multi-source remote sensing image based on space-time constraint of the present application.

[0038] Figure 2 The flowchart for making space constraint of the present application.

[0039] Figure 3 The flowchart for extracting water body based on Sentinel 2 remote sensing image of the embodiment of the present application.

[0040] Figure 4 The flowchart for extracting water body based on Sentinel 1 remote sensing image of the embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0042] The present application provides a spatio-temporal constraint-based multi-source remote sensing image accurate water body extraction method and system, and the technical solutions of the present application will be further described below in combination with the drawings and embodiments.

[0043] Embodiment 1

[0044] In this embodiment, the Poyang Lake Basin located in the middle and lower reaches of the Yangtze River is selected as the target area. The wet season of this area is from April to October every year, which basically overlaps with the period when the sun is directly above the northern hemisphere (from March 21 to September 23). The data used are Sentinel-2 and Sentinel-1 remote sensing images during 2016-2022, as well as other data such as JRC Global Surface Water (GSW) and Space Shuttle Radar Topography Mission (SRTM). The programming tool is selected as GEE (Google Earth Engine) cloud computing platform. As shown in the figure, the specific implementation process of the spatio-temporal constraint-based multi-source remote sensing image accurate water body extraction method of the present application is as follows: Figure 1

[0045] Step 1, generate spatial constraints for water body extraction.

[0046] Since the wet season of the Poyang Lake Basin basically coincides with the period when the sun is directly above the northern hemisphere, there are almost no mountain shadows and building shadows on the optical remote sensing images during the wet season in this area, and the peak value of water area also usually appears during this period. In this embodiment, the monthly water bodies of the Poyang Lake Basin during April to September in 2016-2022 are extracted from the Sentinel-2 remote sensing images, and are combined with the maximum water area range of the Poyang Lake Basin in the JRC GSW data set maximum water area range layer to generate a new maximum water area range as the spatial constraint for water body extraction (as shown in the figure Figure 2 ). This part includes three parts of wet season monthly water body extraction, shadow removal post-processing and spatial constraint generation.

[0047] Step 1.1, use the random forest classification method to extract the monthly water bodies of the Poyang Lake Basin during the wet season.

[0048] ​First, manually draw enough number of Sentinel-2 remote sensing image ground feature classification samples in Poyang Lake Basin on GEE platform. Table 1 shows the specific information of Sentinel-2 remote sensing image samples, which include water, land and cloud, and the corresponding sample numbers are 693, 459 and 177 respectively. Then use the cloud probability dataset (S2_CLOUD_PROBABILITY) of Sentinel-2 remote sensing image and the Directional Distance Transform function provided in GEE to remove the cloud and cloud shadow of each scene of Sentinel-2 remote sensing image in Poyang Lake Basin from April to September each year, and use the ee.ImageCollection.median() function in GEE to generate the median composite image of each month. Finally, based on the manually drawn ground feature classification samples, the monthly water body in Poyang Lake Basin from April to September each year is automatically extracted by using the random forest classification method.

[0049] Table 1 Sentinel-2 image ground feature classification sample information

[0050]

[0051] Step 1.2, remove the mountain shadow, building and building shadow in the water body extracted in step 1.1.

[0052] There are still a small amount of mountain shadow and building shadow in the monthly water body generated in step 1.1 during the wet season, and there are also a small amount of misextracted building area. In this embodiment, the area with an altitude higher than 500 m and a slope greater than 3 degrees is regarded as a mountain shadow area, and the DEM data from the Space Shuttle Radar Topography Mission (SRTM) and the slope data calculated therefrom are used to mask to remove the small amount of remaining mountain shadow. At the same time, the global human settlement layer (GHSL) in the JRC GSW dataset is used to mask to further remove building shadow and misextracted building noise, and to generate more accurate and complete monthly water body data in Poyang Lake Basin from April to September each year from 2016 to 2022.

[0053] Step 1.3, generate spatial constraints.

[0054] Merge the 42 monthly water bodies in Poyang Lake Basin from April to September each year from 2016 to 2022 produced in step 1.2 with the maximum water area range in the Poyang Lake Basin in the maximum water area range layer (max_extent) in the JRC Global Surface Water Mapping Layers v1.4 dataset to generate a new and more complete maximum water area range data as the spatial constraints for water body extraction to remove the influence of mountain shadow and building shadow.

[0055] Step 2. Water extraction based on Sentinel-2 remote sensing images.

[0056] As Figure 3 shown, the present application combines the random forest classification method and the modified normalized difference water index (MNDWI) to process Sentinel-2 remote sensing images and extract monthly water bodies in the Poyang Lake Basin from 2016 to 2022.

[0057] Step 2.1. Remove clouds and cloud shadows from Sentinel-2 remote sensing images and synthesize monthly median images.

[0058] Using the cloud probability dataset corresponding to Sentinel-2 remote sensing images (S2_CLOUD_PROBABILITY) and the Directional Distance Transform function provided in GEE, remove clouds and cloud shadows from each Sentinel-2 remote sensing image in the Poyang Lake Basin from 2016 to 2022, and use the ee.ImageCollection.median() function in GEE to generate a monthly median synthesis image. At the same time, mark the areas with missing data and cloud and cloud shadow coverage in the Sentinel-2 monthly median synthesis image as Nodata areas, so as to supplement the extraction using Sentinel-1 remote sensing images in the subsequent steps.

[0059] Step 2.2. Extract water bodies by combining random forest classification and MNDWI method.

[0060] The random forest classification method has the advantages of strong anti-overfitting ability, strong ability to process large-scale data, and strong robustness, and the MNDWI index has the advantages of resisting atmospheric interference, vegetation suppression, and easy calculation when extracting water bodies. These two methods have good applicability and high precision, and can be applied to more regions and scenes. The combination of the two methods can integrate the advantages of both and further improve the accuracy and recall rate of water body extraction. First, the random forest classification method and the MNDWI method are used to process the median composite images of Sentinel-2 every month to extract water bodies. The random forest classification method is based on the land cover classification samples drawn by hand in step 1.1 to identify water bodies and identify the cloud-covered areas that are not removed in step 2.1, and also mark them as Nodata areas. The segmentation threshold of the MNDWI method is automatically determined using the OTSU algorithm. The water bodies extracted by the two methods are marked as W1 and W2, and then based on the ee.Image.updateMask() function in GEE, the spatial constraints generated in step 1 are used to mask W1 and W2 to retain the water bodies within the spatial constraints to remove mountain shadows and building shadows. The SWO layer in the JRC Global Surface Water Mapping Layers v1.4 dataset is used to divide the Poyang Lake Basin into high-frequency water regions and other regions by setting a threshold. The regions with SWO greater than the set threshold δ are considered as high-frequency water regions, and the regions with SWO less than or equal to the set threshold δ are considered as other regions. For high-frequency water regions, the union of W1 and W2 is taken as the final water body result, while for other regions, the intersection of W1 and W2 is taken. In addition, considering the obvious seasonal characteristics of the Poyang Lake Basin, water bodies often shrink significantly in the dry season, so the δ value is appropriately adjusted higher in the dry season and lower in the wet season as a time constraint to adapt to different periods and further improve the water body extraction accuracy. Finally, the water body extraction results of high-frequency water regions and other regions are combined to obtain the further water body extraction results of the Poyang Lake Basin Sentinel-2 remote sensing images.

[0061] Step 2.3, remove the residual mountain shadow, building and building shadow in the water body extracted in step 2.2.

[0062] Regions with an altitude higher than 500m and a slope greater than 3 degrees are considered as mountain shadow areas, and the DEM from the Shuttle Radar Topography Mission (SRTM) and the slope data calculated therefrom are used to mask the remaining small amount of mountain shadow in the water body generated in step 2.2. Meanwhile, the global human settlement layer (GHSL) from the JRC GSW dataset is used to further remove building shadows and mis-removed building noise to obtain the final water body extraction results of the Poyang Lake Basin based on Sentinel-2 monthly median composite images.

[0063] Step 3, water body supplement extraction using Sentinel-1 image.

[0064] Sentinel-2 image is an optical remote sensing image, which can show more realistic ground color and texture information compared with SAR image, so it is more applied to water body extraction. Sentinel-1 satellite carries C-band synthetic aperture radar (SAR) and can observe all-weather. Since optical images are affected by weather, i.e. there are cloud and cloud shadow areas, therefore for the areas covered by cloud and cloud shadow in Sentinel-2 image and the areas of data missing, they can be supplemented by Sentinel-1 image of the same period and the same area to ensure the integrity of the water body extraction results. The specific process is shown in Figure 4

[0065] Step 3.1, making Sentinel-1 remote sensing image ground object classification samples.

[0066] A sufficient number of Sentinel-1 remote sensing image ground object classification samples in Poyang Lake Basin were manually drawn on GEE platform. Table 2 shows the specific information of Sentinel-1 remote sensing image samples, which include water and land, and the corresponding sample numbers are 694 and 353 respectively.

[0067] Table 2 Information of Sentinel-1 image ground object classification samples

[0068]

[0069] Step 3.2, Sentinel-1 remote sensing image filtering processing.

[0070] All Sentinel-1 remote sensing images of Poyang Lake Basin from 2016 to 2022 were combined into a monthly average composite image, and the ee.Image.focalMean() function in GEE was used to filter the image to reduce noise.

[0071] Step 3.3, water body extraction combined with random forest classification and threshold method.

[0072] ​Firstly, the random forest classification method and the threshold segmentation method are respectively used to process the monthly average composite images of Sentinel-1 to extract water bodies. The random forest classification method is based on the ground feature classification samples manually drawn in step 3.1, and the threshold segmentation method is based on the VV and VH bands of the Sentinel-1 remote sensing image, and the OTSU algorithm is also used to automatically determine the segmentation threshold. When the values of the VV and VH bands are both less than the threshold value determined by the OTSU algorithm, it is determined that the water body. The water bodies extracted by the two methods are respectively marked as W3 and W4, and then the Nodata region marked in step 2 is used to mask W3 and W4 to retain the water bodies in the Nodata region. Further, the spatial constraint generated in step 1 is used to mask the water bodies in the Nodata region to retain the water bodies within the spatial constraint range to remove the mountain shadow and building shadow. Similarly, the SWO (Surface water occurrence) layer in the JRC Global Surface Water Mapping Layers v1.4 dataset is used to determine the high-frequency water region in the Poyang Lake Basin, and the region with SWO greater than the threshold value δ is regarded as the high-frequency water region, and the region with SWO less than or equal to the threshold value δ is regarded as the other region. For the high-frequency water region, the union of W3 and W4 is taken as the final water body result, and for the other region, the intersection of W3 and W4 is taken. In addition, considering the obvious seasonal characteristics of the Poyang Lake Basin, the water body often shrinks seriously in the dry season, so the δ value is appropriately increased in the dry season and appropriately decreased in the wet season, which is used as a time constraint to be applicable to different periods, and further improve the water body extraction accuracy. Finally, the water body extraction results of the high-frequency water region and the other region are combined to obtain the further water body extraction result of the Sentinel-1 remote sensing image in the Poyang Lake Basin.

[0073] Step 3.4, removing the residual mountain shadow, building and building shadow in the water body extracted in step 3.3.

[0074] The region with an altitude higher than 500m and a slope greater than 3 degrees is regarded as the mountain shadow region, and the DEM from the Space Shuttle Radar Topography Mission (SRTM) and the slope data calculated therefrom are used to mask to remove the small amount of mountain shadow still present in the water body generated in step 3.3. At the same time, the global human settlement layer (GHSL) from JRC is used to mask to further remove the building shadow and the mis- extracted building noise, and the final water body extraction result of the Nodata region of the Sentinel-1 monthly average composite image is obtained.

[0075] Step 4, combining the water body data extracted in steps 2 and 3 to generate the final monthly water body extraction result in the Poyang Lake Basin.

[0076] Example 2

[0077] Based on the same inventive concept, the application also provides a multi-source remote sensing image accurate water body extraction system based on space-time constraints, comprising a processor and a memory, the memory is used for storing program instructions, and the processor is used for calling the program instructions in the memory to execute the multi-source remote sensing image accurate water body extraction method based on space-time constraints.

[0078] In specific implementation, the method provided by the technical scheme of the application can be automatically run by a computer software technology, and the system device of the method, such as a computer readable storage medium storing the corresponding computer program of the technical scheme of the application and a computer device including the running of the corresponding computer program, should also be within the protection scope of the application.

[0079] The above only describes some embodiments of the application and is not used to limit the application. For those skilled in the art, the application can have various changes. Any change, equivalent replacement or improvement within the spirit and principle of the application should be included in the protection scope of the application. The application is suitable for the regions where the periods of abundant water and the periods of the sun directly shining on the hemisphere coincide and the seasonal changes of water bodies are very obvious. The accurate water body extraction of the target region is realized by designing a program through the GEE cloud computing platform.

Claims

1. A method for accurate water body extraction from multi-source remote sensing images based on spatiotemporal constraints, characterized in that, Includes the following steps: Step 1: Extract the monthly water bodies during the high-water season in the target area of ​​the Sentinel-2 remote sensing image during the study period, and merge them with the maximum water area of ​​the target area in the maximum water area layer of the JRC GSW dataset to generate spatial constraints for water body extraction. Step 1.1: Use the random forest classification method to extract the monthly water bodies during the high-water season in the target area during the study period; A large number of Sentinel-2 remote sensing images of the target area were manually drawn, including three categories: water, land, and clouds. Clouds and cloud shadows were removed from the Sentinel-2 remote sensing images of the target area during the high water season, and median composite images for each month were generated. Based on the manually drawn land feature classification samples, the monthly water bodies of the target area during the high water season were automatically extracted using the random forest classification method. Step 1.2: Remove the shadows of mountains, buildings, and structures from the water body extracted in Step 1.1; Step 1.3: Generate spatial constraints; The monthly water body data of the target area during the high-water season of the research period, in step 1.2, which removed the shadows of mountains, buildings, and buildings, is merged with the maximum water body range of the target area in the maximum water body range layer of the JRC GSW dataset to generate a new maximum water body range, which serves as a spatial constraint for water body extraction. Step 2: Process Sentinel-2 remote sensing images using the random forest classification method and the modified normalized difference water index MNDWI to extract monthly water bodies in the target area during the study period; Step 2.1: Remove clouds and cloud shadows from the Sentinel-2 remote sensing image and synthesize a monthly median image; Cloud and cloud shadow removal processing was performed on each Sentinel-2 remote sensing image of the target area during the study period to generate a median composite image for each month, and areas covered by clouds and cloud shadows as well as areas with missing data were marked as Nodata areas; Step 2.2: Extract water bodies by combining random forest classification and the MNDWI method; First, the Random Forest classification method and the MNDWI method were used to process the monthly median composite images from Sentinel-2 to extract water bodies. The Random Forest classification method was based on the manually drawn land cover classification samples in step 1.1 to identify cloud cover areas that were not completely removed in step 2.1 while identifying water bodies, and these areas were also marked as Nodata areas. The segmentation threshold of the MNDWI method was automatically determined using the Otsu's maximum inter-class variance algorithm. The water body results extracted by the two methods were labeled as W1 and W2. Then, the spatial constraints generated in step 1 were used to mask W1 and W2, retaining the water bodies within the spatial constraints and removing mountain shadows and building shadows. Using the SWO layer in the JRCGSW dataset, the target area was divided into high-frequency water areas and other areas by setting a threshold. Areas with SWO values ​​greater than the set threshold were classified as high-frequency water areas. The area is considered a high-frequency water region, and the target area is defined as having a SWO value less than or equal to a set threshold. The region is treated as another region. For high-frequency water regions, the union of W1 and W2 is taken as the final water body result, while for other regions, the intersection of W1 and W2 is taken. In addition, the water level is adjusted during the dry season. Values ​​should be lowered during the high-water season. The value is used as a time constraint to apply to different periods and further improve the accuracy of water body extraction; finally, the water body extraction results of high-frequency water and other areas are merged to obtain the water body extraction results of Sentinel-2 remote sensing images of the target area within the study period. Step 2.3: Remove residual mountain shadows, building shadows, and building shadows from the water body extracted in Step 2.2; Step 3: Using Sentinel-1 remote sensing imagery, monthly water body replenishment and extraction are performed on areas covered by clouds and cloud shadows and areas with missing data in the Sentinel-2 remote sensing imagery of the target area during the research period. Step 4: Combine the water data extracted in Step 2 and Step 3 to generate the final monthly water extraction results for the target area.

2. The method for accurate water body extraction based on spatiotemporal constraints from multi-source remote sensing images as described in claim 1, characterized in that: In step 1.2, the altitude is higher than m and slope greater than The area with a slope of a certain degree is considered the mountain shadow area. A mask is then applied using DEM data and the slope data calculated from it to remove a small amount of mountain shadow from the monthly water bodies extracted in step 1.

1. , The set threshold; We used the global human settlement layer GHSL from the JRCGSW dataset for masking to remove building shadows and misextracted building noise, generating more accurate and complete monthly water body data for the target area during the high-water season.

3. The method for accurate water body extraction based on spatiotemporal constraints from multi-source remote sensing images as described in claim 1, characterized in that: In step 2.3, the altitude will be higher than m and slope greater than The area with a slope of 10 degrees is considered the mountain shadow area. Using DEM data and the slope data calculated from it, a mask is applied to remove the small amount of residual mountain shadow in the monthly water body obtained in step 2.

2. , The threshold was set; at the same time, the global human settlement layer GHSL in the JRCGSW dataset was used for masking to further remove residual building shadows and misextracted building noise, generating the final water extraction results of the Sentinel-2 monthly median composite image of the target area during the study period.

4. The method for accurate water body extraction based on spatiotemporal constraints from multi-source remote sensing images as described in claim 1, characterized in that: Step 3 includes the following steps: Step 3.1: Create land cover classification samples from Sentinel-1 remote sensing images; A large number of land cover classification samples from Sentinel-1 remote sensing images of the target area were manually drawn, including samples of water and land. Step 3.2, Sentinel-1 remote sensing image filtering processing; All Sentinel-1 remote sensing images of the target area for each month during the study period were combined into one scene to obtain the monthly average composite image. The monthly average composite image was then filtered to reduce the impact of noise. Step 3.3: Extract water bodies by combining random forest classification and thresholding. Step 3.4: Remove the residual mountain shadows, building shadows, and building shadows extracted from the water body in Step 3.

3.

5. The method for accurate water body extraction based on spatiotemporal constraints from multi-source remote sensing images as described in claim 4, characterized in that: In step 3.3, the random forest classification method and the threshold segmentation method are used to process the monthly average synthetic image of Sentinel-1 to extract water bodies. The random forest classification method is based on the manually drawn land cover classification samples in step 3.1, and the threshold segmentation method is based on the VV and VH bands of the Sentinel-1 remote sensing image. The OTSU algorithm is used to automatically determine the segmentation threshold. When the values ​​of both the VV and VH bands are less than the threshold determined by the OTSU algorithm for the corresponding band, it is identified as a water body. The water bodies extracted by these two methods are marked as W3 and W4, respectively. Then, the Nodata region marked in step 2 is used to mask W3 and W4 to retain the water bodies in the Nodata region. Then, the spatial constraints generated in step 1 are used to mask the water bodies in the Nodata region to retain the water bodies within the spatial constraints, so as to remove the shadows of mountains and buildings. Using the SWO layer in the JRCGSW dataset, areas with SWO values ​​greater than a threshold are targeted. The area is considered a high-frequency water region, and the SWO value within the target area is less than or equal to the threshold. The region is considered as another region. For the high-frequency water region, the union of W3 and W4 is taken as the final water body result, while for other regions, the intersection of W3 and W4 is taken. In addition, the water level was raised during the dry season. Values ​​should be lowered during the high-water season. The value is used as a time constraint to apply to different periods, further improving the accuracy of water body extraction. Finally, the water extraction results of high-frequency water and other areas are merged to obtain the further water body extraction results of the Sentinel-1 remote sensing image in the Nodata area of ​​the target area.

6. The method for accurate water body extraction based on spatiotemporal constraints from multi-source remote sensing images as described in claim 4, characterized in that: In step 3.4, the altitude will be higher than m and slope greater than The area with a slope of a certain degree is considered the mountain shadow area. A mask is then applied using DEM data and the slope data calculated from it to remove the small amount of residual mountain shadow in the monthly water body obtained in step 3.

3. , The threshold was set; at the same time, the global human settlement layer GHSL in the JRCGSW dataset was used for masking to further remove building shadows and misextracted building noise, so as to obtain the final water extraction results of the monthly average synthetic image of Nodata area Sentinel 1 during the study period.

7. A system for precise water body extraction based on spatiotemporal constraints from multi-source remote sensing images, characterized in that, It includes a processor and a memory, the memory being used to store program instructions, and the processor being used to call the program instructions in the memory to execute the method for accurate water body extraction based on spatiotemporal constraints of multi-source remote sensing images as described in any one of claims 1-6.

Images