A surface water null pixel reconstruction method based on optimal similarity

Abstract

The application discloses a surface water empty pixel reconstruction method based on optimal similarity, and is characterized in that the method comprises the following steps: obtaining remote sensing image data sets in a to-be-measured region, and performing empty pixel screening to obtain a first target database with empty pixels and a first reference database without empty pixels; obtaining a three-value image, a two-value image and a one-value image in the to-be-measured region, and performing screening on the first target database and the first reference database to obtain a corresponding second target database and a second reference database; performing image operation on the images in the second reference database to obtain optimal similar images of the images in the second reference database; reconstructing the empty pixels in the second target database based on the optimal similar images to obtain reconstructed images; and supplementing the reconstructed images to the second reference database to complete the reconstruction of the surface water empty pixels. The application can reasonably allocate water resources, maintain the ecological balance of water resources, and realize the reasonable utilization of water resources.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of remote sensing and computer science, and particularly relates to a surface water empty pixel reconstruction method based on optimal similarity. BACKGROUND

[0002] Multiple global surface water datasets have been generated based on remote sensing data. Pekel et al. generated a global surface water dataset (GSWD) from 1984 to 2015 based on Landsat data, which achieved a major breakthrough in the application of remote sensing in surface water monitoring. In theory, this dataset can provide a monthly area time series of all surface water bodies during the Landsat monitoring period. However, since GSWD classifies all contaminated pixels (clouds, cloud shadows, terrain shadows, and scan line corrector on Landsat7) as "no data", about one third of the data is lost, resulting in a low data integrity. In order to control the quality of the monitoring results, most existing studies choose to discard the contaminated images. However, this method is not suitable for areas frequently covered by clouds, nor for areas containing SLC errors in Landsat7 images.

[0003] Lakes have obvious topographic feature information. Mullen et al. extracted lake virtual elevation information using the physical mechanism of lake water appearance, solved the problem of water information loss in cloud-contaminated images, and provided a high-quality continuous global monthly lake area dataset. Although a large number of studies have been carried out on the problem of surface water information loss, there are still many problems that have not been solved. First, the method of characterizing lake elevation information by monitoring the water frequency of independent lakes is affected by whether the lake is a single lake basin. With the intensification of climate change and human activities, the dynamic change of lake area in arid regions is enhanced, and its variability has an increasing trend. Therefore, when the lake is a multi-lake basin, the method of using water frequency to represent lake elevation is not universally meaningful and has low robustness; secondly, since rivers do not have obvious elevation constraint information (large elevation difference rivers are common, such as waterfalls. In addition, rivers also commonly exist in variable channels, and the water appearance position is not strictly affected by the surface elevation), the method of characterizing elevation information based on water frequency does not work. In view of this, it is urgent to develop a post-processing method that can be applied to both lakes and rivers to infer the inundation state of contaminated images.

[0004] For a long time, in the remote sensing monitoring research of surface water dynamic change, the existence of a large number of null pixels leads to the discontinuity of surface water information, seriously hinders the efficient development of surface water remote sensing monitoring work, and leads to the low accuracy of surface water data set or the inability to realize the rapid monitoring of small water body dynamic change. Therefore, the research on obtaining more accurate remote sensing monitoring data is not deep enough in the prior art, and the universality of the null value reconstruction method needs to be further improved. Therefore, it is urgent to propose a surface water null pixel reconstruction method based on optimal similarity, which reconstructs the water coverage of null pixels by using non-null pixel information. SUMMARY

[0005] The purpose of the present application is to provide a surface water null pixel reconstruction method based on optimal similarity, which solves the technical problems of low resolution and the inability of the null value reconstruction method to simultaneously satisfy lakes and rivers in the prior art.

[0006] To achieve the above purpose, the present application provides a surface water null pixel reconstruction method based on optimal similarity, comprising the following steps:

[0007] Obtain the remote sensing image data set in the to-be-measured region, and perform null pixel screening to obtain a first target database with null pixels and a first reference database without null pixels;

[0008] Obtain the surface water body information in the to-be-measured region, and based on the surface water body information, use the CFmask algorithm and the decision tree method to obtain a three-value image, a binary image and a one-value image in the to-be-measured region;

[0009] Screen the first target database and the first reference database based on the three-value image, the binary image and the one-value image in the to-be-measured region to obtain corresponding second target database and second reference database;

[0010] Perform image operation on the images in the second reference database to obtain the optimal similar images of the images in the second reference database;

[0011] Reconstruct the null pixels in the second target database based on the optimal similar images to obtain reconstructed images;

[0012] Supplement the reconstructed images to the second reference database to complete the reconstruction of the surface water null pixels.

[0013] Optionally, the process of obtaining the surface water body information in the to-be-measured region comprises: obtaining an optimal water body identification method in the to-be-measured region, and based on the optimal water body identification method, obtaining the surface water body information in the to-be-measured region.

[0014] Optionally, the process of obtaining the three-value image, the binary image and the one-value image in the to-be-tested region comprises: constructing a decision tree model to classify the surface water body information to obtain a surface water pixel category, the surface water pixel category comprising a water pixel, a non-water pixel and a null pixel; and based on the CFmask algorithm and the decision tree method, obtaining the three-value image of the water pixel, the non-water pixel and the null pixel, the binary image of the water pixel and the non-water pixel, and the one-value image of the water pixel or the non-water pixel or the null pixel.

[0015] Optionally, the process of screening the first target database and the first reference database based on the three-value image, the binary image and the one-value image in the to-be-tested region comprises: when the image is a binary image or a one-value image, classifying into the first reference database without null pixels to obtain a corresponding second reference database; and when the image is a three-value image and simultaneously contains null pixels and water pixels, classifying into the first target database with null pixels to obtain a corresponding second target database.

[0016] Optionally, the formula of the image operation on the image in the second reference database is as follows:

[0017] min(A(S dw △S Dw ))

[0018] Wherein, A is the spatial distribution information of the surface water body, S dw is the water body distribution of the to-be-reconstructed image with null pixels, S Dw is all water body distributions of the images in the second reference database without null pixels, △ represents intersection negation, and min(A) is the minimum value of the non-overlapping spatial area of S dw and S Dw .

[0019] Optionally, the process of obtaining the remote sensing image data set in the to-be-tested region comprises: obtaining long-time series multi-source remote sensing monitoring data in the to-be-tested region to construct a remote sensing image data set in the to-be-tested region.

[0020] Optionally, the process of reconstructing the null pixels in the second target database based on the optimal similar image comprises: obtaining the image with null pixels from the second target database, obtaining the similarity of the spatial distribution of water pixels between the image with null pixels and any image without null pixels in the second reference database and sorting to obtain the highest similarity of the null pixel without null pixels, and replacing the null pixel in the image with null pixels with the highest similarity of the null pixel without null pixels.

[0021] The technical effect of the present application is:

[0022] This invention provides a surface water null pixel reconstruction method based on optimal similarity, solving the problems of the large number of null pixels in existing remote sensing data hindering the efficient application of optical remote sensing imagery, and the fact that existing null reconstruction methods do not simultaneously meet the requirements for lake and river null pixel reconstruction. It achieves a systematic and accurate assessment of surface water changes, thereby accurately grasping the current status of surface water resources. Through the analysis of surface water resource changes, it enables intelligent adjustments to current surface water resource utilization methods, thus rationally allocating water resources, reducing unnecessary waste, maintaining the ecological balance of water resources, and achieving the technical effect of rational water resource utilization. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 This is a schematic diagram of the surface water spatial cell reconstruction method based on optimal similarity in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram illustrating the principle of the image similarity evaluation method in an embodiment of the present invention. Detailed Implementation

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0028] Example 1

[0029] like Figure 1 As shown in the figure, this application provides a method for reconstructing surface water spatial value pixels based on optimal similarity, wherein the method includes:

[0030] Collect long-term, multi-source remote sensing monitoring data to obtain a remote sensing dataset for the area to be measured:

[0031] Specifically, with the rapid development of remote sensing technology in recent decades, satellite-based water body monitoring has become the main means of monitoring the dynamic changes of surface water bodies. The wide application of remote sensing technology makes large-scale water resource monitoring possible, especially in remote and less accessible areas. Among all satellite sensors, the Landsat series sensors have the longest satellite monitoring capability and medium spatial resolution (30 meters). The Landsat series of satellites provides important data support for earth observation. Landsat5 TM, Landsat7 ETM+ and Landsat8 OLI Collection1 Tier1 surface reflectance data are initially provided by the United States Geological Survey (USGS) and are widely used for the identification of open surface water bodies. Landsat5 was launched in 1984 and retired in 2012, with a 16-day revisit period and a spatial resolution of 30m. Landsat7 was launched in 1999, but due to the failure of the scan line corrector (SLC), the images collected after June 2003 showed striping, and was officially replaced by Landast9 at the end of 2020. Landsat8 launched in 2013 carries a land imager and a thermal infrared sensor. Among them, OLI includes 9 bands of blue, green, near-infrared, etc. Here, we use bands 2 (green: 0.52-0.60 μm), band 3 (red: 0.63-0.69 μm), band 4 (near-infrared: 0.77-0.90 μm) and band 5 (short-wave infrared: 1.55-1.75 μm) from Landsat5 and 7, and bands 3 (green: 0.53-0.59 μm), band 4 (red: 0.64-0.67 μm), band 5 (near-infrared: 0.85-0.88 μm) and band 6 (short-wave infrared: 1.57-1.65 μm) from Landsat8. Due to problems such as clouds, cloud shadows, and sensor failures, there are a large number of null pixels in the monitoring images, which poses a challenge to the accurate monitoring of the changes in the range of surface water bodies. Surface water resources are a core important factor affecting the changes and development of human-land-ecosystem. Therefore, accurate monitoring of the dynamic changes of surface water bodies can provide in-depth understanding of the specific change characteristics of any lake, river or reservoir, and provide data support for accurately grasping the trend of surface water resource changes, which is of great significance for regional surface water resource management.

[0032] According to the remote sensing data set in the to-be-measured region, a first target database with null pixels and a first reference database without null pixels are obtained:

[0033] Specifically, according to the image has no empty pixel database is set up respectively, is the important premise of the implementation of the empty value reconstruction method. First, according to the optimal water body identification method to obtain the surface water body information, and according to the CFmask algorithm, decision tree method to divide three value image (water, non water and empty value), two value image (water and non water) or one value image (water or non water or empty value), for screening image provides data basis. In the process of screening satellite image, according to whether there is empty pixel in the project, the image is divided into reference database (no empty pixel) and target database (with empty pixel). In order to ensure the quality of database and work efficiency, the reference database and target database need to be further screened in detail. When there is no empty pixel in the image, that is, two value image or one value image, it is classified into the first reference database, and the corresponding second target database is obtained; when there is empty pixel in the image but no water pixel, it is discarded; when there is empty pixel in the image and contains water pixel, it is classified into the first target database, and the second reference database is obtained.

[0034] According to the geographical environment, the optimal water body identification method in the region is screened out:

[0035] Specifically, the geographical environment characteristics of the study area are determined, and the optimal water body identification method is selected, which is an important basis for accurately obtaining the surface water body information. In order to realize the optimal performance of each method and obtain the most accurate water body classification result, the method selects the optimal water body identification method in different regions by carrying out instance verification on multiple example water bodies in the study area with the measured water body information as the reference value. At present, there are many classification accuracy evaluation methods, such as confusion matrix, KAPPA coefficient, root mean square error (RMSE), omission error and error error, etc. The optimal water body identification method in the region is obtained.

[0036] According to the optimal water body identification method in the region, the surface water body information in the region to be measured is obtained;

[0037] The decision tree model is constructed, and the surface water pixel classification in the region to be measured is obtained;

[0038] According to the CFmask algorithm and the decision tree method, the water, non water and empty value three value image in the region to be measured is obtained;

[0039] The calculation unit is determined and the image operation is carried out, and the optimal similar image of the image in the second reference database is obtained:

[0040] Specifically, the basic principle of the image similarity evaluation method is as follows: Based on a second reference database (denoted as D), the image with the highest similarity (without null values) is determined by calculating the similarity of the spatial distribution of water pixels between the null pixel image and any image in database D. The null pixels in the null pixel image are then replaced with pixels at the corresponding positions of the most similar null pixel image. In this embodiment, image similarity is calculated using the intersection negation image operation method to determine the number of pixels in different categories; the fewer the number, the higher the similarity, and vice versa. Figure 2 d i The satellite imagery database D contains images with null values ​​and images without null values, including D1, D2, ..., D2. j Multiple image data, etc. All data in D that are related to d. i The non-nullable pixels corresponding to the nullable pixel positions are masked to obtain D. j And calculate d i With each D j The number of different types of pixels between them is denoted as DP. For example, D1 and d i The DP value between them is 2, indicating that in the reference database, D1 and the null pixel image d i The number of pixels of different categories is 2. Similarly, the image D with the smallest DP is calculated. j ( Figure 2 The best similar image in the dataset is D5), and then D is used. j The pixel category replacement image d i The null value cells are used to obtain the true water cover information of the null value cells.

[0041] Water cover information with nullable pixels is reconstructed from the best similar image to obtain an image without nullable pixels:

[0042] Specifically, firstly, using the optimal water body identification method selected in the early stage and based on Landsat satellite imagery, the surface water area information of the study area is obtained. Then, a ternary image (water, non-water, and null values) is constructed using the CFmask algorithm and decision tree method. Subsequently, all Landsat images of the study area during the monitoring period are screened, and satellite images without null values ​​are collected as a second reference database (denoted as D), while satellite images containing null values ​​are collected as a second target database (denoted as d). Next, a selection window is determined and image processing is performed. Based on the images in the second reference database D, the most similar images are selected, and the null values ​​in the second target database d are reconstructed, thus obtaining high-quality reconstructed images. The reconstructed null value-free images are then added to the second reference database D, further expanding the scope of the second reference database D, thereby improving the accuracy of the null value reconstruction method in this project. The image processing formula is as follows:

[0043] min(A(S dw △SDw ))

[0044] Wherein, A is the spatial distribution information of surface water body, S dw is the water body distribution of the image to be reconstructed with null pixels, S Dw is all the water body distribution of the reference database without null pixels, △ represents the complement of intersection, min(A) means taking the minimum value of S dw and S Dw The minimum value of the non-overlapping spatial area of the water body.

[0045] The reconstructed null pixel image is supplemented into the reference database to improve the abundance of the reference database and further improve the reconstruction accuracy of the null pixels.

[0046] In summary, the surface water null pixel reconstruction method based on optimal similarity provided in the embodiment has the following technical effects:

[0047] The surface water null pixel reconstruction method based on optimal similarity, wherein the method comprises the following steps. One, selecting a water body index, extracting surface water body information based on Landsat satellite data; two, constructing a water, non-water and null value three-value image using the CFmask algorithm and the decision tree method; three, screening all images in the monitoring period, collecting satellite images without null pixels as a reference database, and collecting satellite images with null pixels as a target database; four, determining a calculation unit and performing image operation, according to the images in the reference database, screening the optimal similar images to reconstruct the null pixels of a specific image in the target database, thereby obtaining an image without null pixels; five, supplementing the reconstructed null pixel image into the reference database to improve the abundance of the reference database and further improve the reconstruction accuracy of the null pixels.

[0048] The embodiment provides a surface water null pixel reconstruction method based on optimal similarity, which achieves high-quality reconstruction of optical remote sensing image null pixels, thereby accurately grasping the present situation of surface water resources in the research area, reasonably utilizing water resources, reducing unnecessary waste, maintaining the ecological balance of water resources, and realizing the reasonable utilization of water resources.

[0049] The embodiment uses multi-source long-time remote sensing satellite data, which can coordinate long-time and high-time and space resolution, realize long-time high-quality monitoring of the dynamic change of surface water body, and verify the accuracy based on measured data, fully prove the accuracy of the estimation result. The surface water resource quantity can be accurately estimated, so that the overall situation of surface water resources can be fully grasped, and the sustainable utilization of surface water resources can be realized.

[0050] The above description is only the preferred embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for reconstructing surface water void pixels based on optimal similarity, characterized in that, The method comprises the following steps: obtaining a remote sensing image dataset in a to-be-tested region, and performing null pixel screening to obtain a first target database with null pixels and a first reference database without null pixels; obtaining surface water body information in the to-be-tested region, and based on the surface water body information, using a CFmask algorithm and a decision tree method to obtain a three-value image, a binary image and a one-value image in the to-be-tested region; screening the first target database and the first reference database based on the three-value image, the binary image and the one-value image in the to-be-tested region to obtain corresponding second target database and second reference database; performing image operation on the images in the second reference database to obtain optimal similar images of the images in the second reference database; reconstructing the null pixels in the second target database based on the optimal similar images to obtain reconstructed images; complementing the reconstructed images to the second reference database to complete the reconstruction of the surface water null pixels; The process of obtaining the three-value image, the binary image and the one-value image in the to-be-tested region comprises: constructing a decision tree model to classify the surface water body information to obtain a surface water pixel category, wherein the surface water pixel category comprises water pixels, non-water pixels and null pixels; based on the CFmask algorithm and the decision tree method, obtaining the three-value image of the water pixels, the non-water pixels and the null pixels, the binary image of the water pixels and the non-water pixels, and the one-value image of the water pixels or the non-water pixels or the null pixels; The process of screening the first target database and the first reference database based on the three-value image, the binary image and the one-value image in the to-be-tested region comprises: when the image is a binary image or a one-value image, classifying it into the first reference database without null pixels to obtain the corresponding second reference database; when the image is a three-value image and simultaneously contains null pixels and water pixels, classifying it into the first target database with null pixels to obtain the corresponding second target database; The formula for performing image operation on the images in the second reference database is as follows: min(A(S dw △S Dw )) Wherein, A is the spatial distribution information of surface water body, S dw is the water body distribution of the image to be reconstructed with null pixels Dw is the water body distribution of all images in the second reference database without null pixels, Δ represents the complement of intersection, min(A) is the minimum value of S dw and S Dw is the minimum value of the non-overlapping spatial area of water bodies.

2. The method for reconstructing surface water null pixels based on optimal similarity according to claim 1, characterized in that The process of obtaining the surface water body information in the to-be-tested region comprises: obtaining an optimal water body identification method in the to-be-tested region, and based on the optimal water body identification method, obtaining the surface water body information in the to-be-tested region.

3. The method for reconstructing surface water null pixels based on optimal similarity according to claim 1, characterized in that The process of obtaining the remote sensing image dataset in the to-be-tested region comprises: obtaining long-time-series multi-source remote sensing monitoring data in the to-be-tested region, and constructing a remote sensing image dataset in the to-be-tested region.

4. The method for reconstructing surface water null pixels based on optimal similarity according to claim 1, characterized in that The process of reconstructing the null pixels in the second target database based on the optimal similar image comprises: obtaining the image with null pixels from the second target database, obtaining the similarity of the spatial distribution of water body pixels between the image with null pixels and any null-pixel-free image in the second reference database and sorting, obtaining the highest similarity null-pixel-free image, and replacing the null pixels in the image with null pixels with the highest similarity null-pixel-free image.

Images