Method and device for extracting pond water surface from remote sensing images

Abstract

The present invention relates to the field of remote sensing and computer vision technology. The present invention discloses a method and device for extracting the water surface of a pond in a remote sensing image, which can improve the accuracy of pond detection. The method for extracting the water surface of a pond in a remote sensing image comprises dividing the remote sensing image into a plurality of image blocks with overlapping areas; inputting all the image blocks into a deep learning model for reasoning and processing to obtain a vector graphic corresponding to each of the image blocks; setting an edge threshold; marking edge water areas on each of the vector graphics according to the edge threshold, and removing the edge water areas to obtain a graphic to be fused corresponding to each of the vector graphics; fusing all the graphics to be fused to obtain a final graphic to extract the pond.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing and computer vision technology, and more specifically, to a method and device for extracting pond water surface in remote sensing images. Background Art

[0002] In the field of remote sensing, object interpretation, as a key task to identify and classify surface objects from remote sensing data, provides basic information for sub-fields such as urban planning, environmental monitoring and resource management. With the vigorous development of computer vision technology, the field of object interpretation has increasingly benefited from advanced image processing and machine learning technologies. In terms of water area extraction, deep learning models have achieved remarkable achievements in high accuracy and automation. However, deep learning models usually need to split remote sensing images into multiple small area blocks of fixed size for inference and prediction. For the marginal waters in the area blocks, due to the corresponding identical water body characteristics and the lack of object information, it is difficult for the model to distinguish whether it is a pond or not. Summary of the invention

[0003] The purpose of the embodiment of the present application is to provide a method and device for extracting the water surface of a pond in a remote sensing image, which can improve the accuracy of pond detection. The embodiment of the present application is mainly achieved through the following technical solutions:

[0004] A first aspect of an embodiment of the present application provides a method for extracting a pond water surface in a remote sensing image, comprising:

[0005] Divide the remote sensing image into multiple image blocks with overlapping areas;

[0006] Inputting all the image blocks into a deep learning model for inference and processing to obtain a vector graphic corresponding to each of the image blocks;

[0007] Set edge threshold;

[0008] Marking edge water areas on each of the vector graphics according to the edge threshold, and removing the edge water areas to obtain a to-be-fused graphic corresponding to each of the vector graphics;

[0009] All the graphics to be fused are fused to obtain a final graphic to extract the pond.

[0010] According to one embodiment of the present application, the step of dividing the remote sensing image into a plurality of image blocks with overlapping areas includes:

[0011] Set a fixed size window;

[0012] Set the overlap;

[0013] Setting the sliding stride of the window on the remote sensing image;

[0014] The window slides on the remote sensing image along a predetermined path according to the sliding stride and the overlap, and an image block corresponding to each position of the window on the remote sensing image is obtained.

[0015] According to one embodiment of the present application, the degree of overlap is set to 0.5.

[0016] According to one embodiment of the present application, the predetermined path is:

[0017] The window starts to slide horizontally from the upper left corner of the remote sensing image until the window slides to the upper right corner of the remote sensing image, then the window slides to the lower left according to the overlap to the left side of the remote sensing image, then the window slides horizontally until the window slides from the left side of the remote sensing image to the right side of the remote sensing image, then the window slides to the lower left according to the overlap to the left side of the remote sensing image, and so on, until the window slides to the lower right corner of the remote sensing image.

[0018] According to one embodiment of the present application, all the image blocks are input into a deep learning model for reasoning and processing, and the step of obtaining a vector graphic corresponding to each image block includes:

[0019] Inputting all the image blocks into a deep learning model for inference to obtain a mask corresponding to each image block;

[0020] Vectorization is performed on all the masks to obtain a vector map corresponding to each image block.

[0021] According to an embodiment of the present application, the step of performing vector processing on all the masks to obtain a vector map corresponding to each of the image blocks includes:

[0022] The contour tracking algorithm of OpenCV2 is used to calculate the pixel position of the corner points on each mask corresponding to the mask area in the remote sensing image;

[0023] According to the pixel position corresponding to each mask, each mask is converted into a corresponding vector map.

[0024] A second aspect of an embodiment of the present application provides a device for extracting a pond water surface in a remote sensing image, comprising:

[0025] An image block division module is used to divide the remote sensing image into a plurality of image blocks with overlapping areas;

[0026] A vector graphics acquisition module, used for inputting all the image blocks into a deep learning model for inference and processing, and obtaining a vector graphic corresponding to each image block;

[0027] An edge threshold setting module, used to set the edge threshold;

[0028] A module for obtaining graphics to be fused, used for marking edge water areas on each of the vector graphics according to the edge threshold, and removing the edge water areas to obtain graphics to be fused corresponding to each of the vector graphics;

[0029] The fusion module is used to fuse all the graphics to be fused to obtain a final graphic to extract the pond.

[0030] A third aspect of an embodiment of the present application 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 computer program, the steps of a method for extracting pond water surface in a remote sensing image as described in any one of the above are implemented.

[0031] A fourth aspect of an embodiment of the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for extracting pond water surface in a remote sensing image as described in any one of the above.

[0032] The beneficial effects of the embodiments of the present application include:

[0033] The method for extracting the water surface of a pond in a remote sensing image provided by an embodiment of the present application is to divide the remote sensing image into a plurality of image blocks with overlapping areas, input all the image blocks into a deep learning model for reasoning and processing, obtain a vector graphic corresponding to each of the image blocks, then set an edge threshold, mark the edge water area on each of the vector graphics according to the edge threshold, and remove the edge water area to obtain a graphic to be fused corresponding to each of the vector graphics; all the graphics to be fused are fused to obtain a final graphic. Thus, the present application can obtain a more complete recognition result (i.e., a graphic to be fused) based on image blocks with overlapping areas, thereby improving the completeness of the extraction of ordinary water areas. After fusing all the graphics to be fused, the present application can filter out repeated recognition results and increase the clarity of the final graphic, thereby improving the accuracy of pond detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1A flowchart of a method for extracting pond water surface in a remote sensing image of the present invention in some embodiments;

[0036] Figure 2 The vector graphics corresponding to the four image blocks in the present invention;

[0037] Figure 3 A schematic diagram of the marginal water areas marked in the present invention;

[0038] Figure 4 A combined graph of all graphics to be fused in the present invention;

[0039] Figure 5 It is a schematic diagram of each two adjacent graphics to be fused before fusion in the present invention;

[0040] Figure 6 It is a schematic diagram of each two adjacent graphics to be fused after fusion in the present invention;

[0041] Figure 7 It is a schematic diagram of all graphics to be fused in the present invention after fusion;

[0042] Figure 8 It is a schematic diagram of window division in the present invention;

[0043] Fig. 9 for Figure 8 The image block corresponding to label 1;

[0044] Fig.10 for Figure 8 The image block corresponding to label 2;

[0045] Fig.11 for Figure 8 The image block corresponding to number 3;

[0046] Fig.12 for Figure 8 The image block corresponding to number 4;

[0047] Fig.13 is a schematic diagram of a mask in some embodiments of the present invention;

[0048] Fig.14 Schematic diagrams of masks in other embodiments of the present invention;

[0049] Fig.15 It is a schematic diagram of the mask vectorized and superimposed on the remote sensing image in the present invention;

[0050] Fig.16 It is a schematic diagram of the accuracy verification area of ​​the present invention;

[0051] Fig.17It is a structural block diagram of the device for extracting pond water surface in remote sensing images in some embodiments of the present invention. DETAILED DESCRIPTION

[0052] 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.

[0053] In the prior art, there are mainly two methods for extracting water areas, namely, automated extraction using traditional algorithms and deep learning models.

[0054] Traditional algorithms usually use basic image processing techniques to identify water features, such as thresholding, texture analysis, morphological operations, and edge detection. More specifically, the thresholding method divides water and non-water bodies based on the grayscale value of pixels; texture analysis locates water areas by detecting texture differences in images; morphological operations use mathematical morphological principles to perform operations such as corrosion and dilation; and edge detection focuses on finding edge features of water areas in images.

[0055] Deep learning models build convolutional neural networks (CNN) and other model structures, use a large number of training samples of labeled water areas, and perform a series of complex training processes such as feature learning and back propagation to obtain training weights for identifying and extracting water areas. These models abstract the input image layer by layer through structures such as convolutional layers and pooling layers, so that they can learn features at different levels, which helps the model better understand the information of land objects in the image. Using the back propagation algorithm, the model adjusts the internal weights to minimize the difference between the predicted output and the true label, gradually improving the accuracy of water area extraction. With the help of model reasoning and post-processing algorithms, it can be applied to multi-source images to extract water areas that are closer to visual interpretation.

[0056] Although both of the above two methods can extract water areas, there are significant deficiencies in the extraction of subdivided water areas. For example, in the task of riverbank monitoring, it is necessary to distinguish between rivers, ponds or reservoirs. However, due to reasons such as data resolution, terrain changes and water body characteristics, even visual interpretation cannot accurately distinguish ponds and rivers in a limited image field of view. Therefore, this application proposes a method for extracting pond water surfaces in remote sensing images to solve the above-mentioned defects.

[0057] The specific implementation of the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0058] <Method for extracting pond water surface from remote sensing images>

[0059] like Figure 1 FIG. 1 is a flowchart of a method for extracting pond water surface from remote sensing images provided in an embodiment of the present application. Figure 1 In the method, the method for extracting the pond water surface in the remote sensing image includes:

[0060] S1. Divide the remote sensing image into multiple image blocks with overlapping areas.

[0061] It should be understood that, in the embodiment of the present application, every two adjacent image blocks have an overlapping area, and the overlapping area can be set by those skilled in the art according to actual needs.

[0062] The size of each of the image blocks is the same.

[0063] The present application uses multiple image blocks for collaborative processing, so that the recognition results of adjacent image blocks can be contour compensated, making the recognition results more complete, thereby avoiding incomplete recognition due to missing ground object information.

[0064] S2. Input all the image blocks into a deep learning model for reasoning and processing to obtain a vector graphic corresponding to each image block.

[0065] S3. Set the edge threshold.

[0066] The edge threshold is ten meters. In other implementations, the value of the edge threshold can be set by those skilled in the art according to actual needs.

[0067] The purpose of using the edge threshold in the embodiment of the present application is to suppress the edge water area of ​​each image block and eliminate the phenomenon of misidentification of pond results caused by incomplete water areas in a limited window area.

[0068] S4. Marking edge water areas on each of the vector graphics according to the edge threshold, and removing the edge water areas to obtain a to-be-fused graphic corresponding to each of the vector graphics.

[0069] In the embodiment of the present application, an area extending from the boundary of the image block to the center of the image block and having an extension distance less than the edge threshold is marked as the edge area, and the recognition result involving the edge area is marked as edge water area.

[0070] For example, see Figure 2 and Figure 3 As shown, Figure 2 There are four vector graphics in Figure 3 The dotted line part is the marginal water area that needs to be removed (suppressed). Figure 3 After the edge water area in the figure, each vector graphic corresponding to the to-be-fused graphic is as follows Figure 4 shown.

[0071] This step can effectively avoid the misdetection caused by the loss of ground object information in the edge area of ​​the image block. For non-river target patches, complete recognition results can be obtained in adjacent overlapping windows (i.e., adjacent image blocks), avoiding possible missed detection problems.

[0072] S5. All the graphics to be fused are fused to obtain a final graphic to extract the pond.

[0073] For example, if there is a pothole in the middle area of ​​the first image block, and there is also the same pothole in the non-edge area of ​​the second image block, then the same pothole will be repeatedly identified in the remote sensing image, resulting in overlapping prediction results (see Figure 5 Therefore, it is necessary to fuse all the graphics to be fused and eliminate the repeated parts to obtain an accurate recognition result (i.e., the final graphics). Figure 5 The image blocks in can be referenced after fusion. Figure 6 As shown. The final graphics after fusion of this application can be referred to Figure 7 shown.

[0074] The present application can obtain more complete recognition results (i.e., graphics to be fused) based on image blocks with overlapping areas, thus improving the completeness of common water area extraction. After fusing all graphics to be fused, the present application can filter out repeated recognition results, increase the clarity of the final graphics, and thus improve the accuracy of pond detection.

[0075] The method for extracting pond water surface in remote sensing images proposed in this application is further explained below through some implementation methods.

[0076] Specifically, in this embodiment, step S1 includes:

[0077] S11. Set a window of fixed size.

[0078] In the embodiment of the present application, the size of the window is 640*640 px. In other implementations, the size of the window can be set by those skilled in the art according to actual needs.

[0079] S12. Set the degree of overlap.

[0080] The overlap degree is set to 0.5, so that the overlap area of ​​two adjacent image blocks occupies 50% of the area of ​​a single image block. In other embodiments, the overlap degree can be set by those skilled in the art according to actual needs.

[0081] In an embodiment of the present application, the setting of the overlapping area of ​​each image block should be able to include the maximum size of the extracted object. For example, the largest pond water surface is approximately 400*400 pixels in size, so the sliding window with an overlap degree of 0.5 needs to be set to 800*800 or more.

[0082] S13, setting the sliding step of the window on the remote sensing image.

[0083] The value of the sliding stride is calculated by the size of the window * the overlap. It should be understood that the "*" symbol mentioned herein is understood as a multiplication symbol in mathematics. Therefore, in the embodiment of the present application, the sliding stride is 320px. In other embodiments, the sliding stride can be set by those skilled in the art according to actual needs.

[0084] S14, the window slides on the remote sensing image along a predetermined path according to the sliding step and the overlap, and obtains an image block corresponding to each position of the window on the remote sensing image.

[0085] Furthermore, the predetermined path is that the window slides horizontally from the upper left corner of the remote sensing image until the window slides to the upper right corner of the remote sensing image, then the window slides to the lower left to the left side of the remote sensing image according to the overlap, then the window slides horizontally until the window slides from the left side of the remote sensing image to the right side of the remote sensing image, then the window slides to the lower left to the left side of the remote sensing image according to the overlap, and so on, until the window slides to the lower right corner of the remote sensing image.

[0086] For example, Figure 8 , 9 , 10, 11 and 12, in Figure 8 In the figure, the sliding positions of the windows are in the boxes corresponding to numbers 1, 2, 3 and 4 respectively. Figure 8 The image block corresponding to the window at position 1 is Fig. 9 As shown; Figure 8 The image block corresponding to the window at position 2 is Fig.10 As shown; Figure 8 The image block corresponding to the window at position 3 is Fig.11 As shown; Figure 8 The image block corresponding to the window at position 4 is Fig.12 shown.

[0087] In this embodiment, the step S2 includes:

[0088] S21. Input all the image blocks into a deep learning model for inference to obtain a mask corresponding to each image block.

[0089] The embodiment of the present application uses a post-processing algorithm of a deep learning model to infer the mask. The mask area on the mask represents the recognition result of the deep learning model on the target in the image block. The mask can be understood as a binary image, such as Fig.13 and Fig.14 As shown. Fig.13 In FIG. 1 , the parts indicated by the labels A, B and C (i.e., the gray parts) are the mask areas, and the mask areas are the identification results of the pits. Fig.14 In the figure, the white area is the mask area.

[0090] S22, vectorize all the masks to obtain a vector graph corresponding to each image block. The vector graph can refer to Fig.15 shown.

[0091] Furthermore, the step S22 includes:

[0092] S221, using the contour tracking algorithm of OpenCV2, calculate the pixel position of the corner point on each mask corresponding to the mask area in the remote sensing image.

[0093] S222. Convert each mask into a corresponding vector map according to the pixel position corresponding to each mask.

[0094] In order to test the effect of this application, the inventors used the extraction method based on the sliding window without overlap, the extraction method based on the overlapping window and the extraction method of the pond water surface in the remote sensing image proposed in this application to compare the accuracy of the results of the pond extraction model optimization for the same batch of remote sensing images. The specific comparison results are shown in Table 1 below. The accuracy verification area used in this experiment is a river in Guangdong Province and its surrounding areas (for details, please refer to Fig.16 As shown), it covers an area of ​​20.5088 square kilometers and has a total of 557 ponds.

[0095] Table 1

[0096]

[0097] It should be understood that in Table 1, recall is a measure of the proportion of positive examples recognized by the model to all actual positive examples; precision is a measure of the proportion of positive examples correctly recognized by the model to all positive examples classified as positive; F1 is an indicator that comprehensively considers the model's ability to recognize positive examples and filter negative examples, and can better reflect the overall performance of the model; IOU represents the overlap between the predicted area and the actual target area, and evaluates the spatial positioning accuracy of the model.

[0098] From Table 1, we can draw the following conclusions:

[0099] 1) Comparing the results of No. 2 and No. 1, using a sliding window with overlapping can significantly improve the recall rate of the model. However, it is worth noting that since the overlapping window will identify more results, the precision rate may drop slightly.

[0100] 2) Comparing the various indicator data of No. 1, No. 2 and No. 3, it can be seen that although No. 3 method slightly reduces the recall rate, it also greatly improves the precision rate, and its F1 index and IOU index are improved. This comprehensive strategy will slightly reduce the recall rate when the model recognition accuracy is low, but as the model recognition accuracy improves, the difference in the recall rate between the two will gradually decrease, and this method can greatly optimize the precision rate and improve the spatial positioning ability to achieve a more comprehensive performance improvement.

[0101] In some embodiments, the method for extracting the pond water surface in the remote sensing image may replace step S2 with the step of "inputting all the image blocks into the deep learning model for inference to obtain a mask corresponding to each image block"; and replace step S4 with the step of "marking edge spots on each mask according to the edge threshold, and removing the edge spots to obtain a to-be-fused graphic corresponding to each mask". This embodiment does not need to be converted into vectorized data and can be used in application scenarios that only require raster data results.

[0102] <Device for extracting pond water surface from remote sensing images>

[0103] like Fig.17 , which is a structural block diagram of a device for extracting a pond water surface in a remote sensing image provided by an embodiment of the present application. The device 100 for extracting a pond water surface in a remote sensing image includes:

[0104] An image block division module 101 is used to divide the remote sensing image into a plurality of image blocks with overlapping areas;

[0105] A vector graphics acquisition module 102, configured to input all the image blocks into a deep learning model for inference and processing, and obtain a vector graphics corresponding to each image block;

[0106] An edge threshold setting module 103, used to set an edge threshold;

[0107] The to-be-fused graphics acquisition module 104 is used to mark the edge water area on each of the vector graphics according to the edge threshold, and remove the edge water area to obtain the to-be-fused graphics corresponding to each of the vector graphics;

[0108] The fusion module 105 is used to fuse all the graphics to be fused to obtain a final graphic to extract the pond.

[0109] In some implementations, the image block division module 101 includes:

[0110] Window setting unit, used to set a window of fixed size;

[0111] An overlap setting unit, used for setting the overlap;

[0112] A sliding step setting unit, used to set the sliding step of the window on the remote sensing image;

[0113] The image block acquisition unit is used to slide the window on the remote sensing image along a predetermined path according to the sliding step and the overlap degree, and acquire the image block corresponding to each position of the window on the remote sensing image.

[0114] In some implementations, the vector graphics obtaining module 102 includes:

[0115] A mask acquisition unit, used for inputting all the image blocks into a deep learning model for inference to obtain a mask corresponding to each of the image blocks;

[0116] The vector graphics acquisition unit is used to perform vectorization processing on all the masks to obtain a vector graphics corresponding to each of the image blocks.

[0117] <Electronic equipment>

[0118] A third aspect of an embodiment of the present application 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 computer program, the steps of a method for extracting pond water surface in a remote sensing image as described in any one of the above are implemented.

[0119] <Non-transitory computer-readable storage medium>

[0120] A fourth aspect of an embodiment of the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for extracting pond water surface in a remote sensing image as described in any one of the above.

[0121] It should also be noted that, unless otherwise defined, all technical and scientific terms used in the specification of this application have the same meaning as those commonly understood by those skilled in the art in the technical field of this application. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in the specification of this application includes any and all combinations of one or more related listed items.

[0122] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0123] It should be noted that, in this document, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatus.

[0124] The technical features of the above embodiments can be combined without changing the basic principles of the present invention. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0125] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the scope of patent protection of the present application shall be subject to the attached claims.

Claims

1. A method for extracting pond water surface in remote sensing images, characterized in that: include: Divide the remote sensing image into multiple image blocks with overlapping areas; Inputting all the image blocks into a deep learning model for inference and processing to obtain a vector graphic corresponding to each of the image blocks; Set edge threshold; Marking edge water areas on each of the vector graphics according to the edge threshold, and removing the edge water areas to obtain a to-be-fused graphic corresponding to each of the vector graphics; wherein an area extending from the boundary of the image block to the center of the image block and having an extension distance less than the edge threshold is marked as an edge area, and a recognition result involving the edge area is marked as an edge water area; All the graphics to be fused are fused to eliminate duplicate parts and obtain a final graphic to extract the ponds.

2. The method for extracting pond water surface in remote sensing images according to claim 1, characterized in that: The step of dividing the remote sensing image into a plurality of image blocks with overlapping areas comprises: Set a fixed size window; Set the overlap; Setting the sliding stride of the window on the remote sensing image; The window slides on the remote sensing image along a predetermined path according to the sliding stride and the overlap, and an image block corresponding to each position of the window on the remote sensing image is obtained.

3. The method for extracting pond water surface in remote sensing images according to claim 2, characterized in that: The overlap is set to 0.

5.

4. The method for extracting pond water surface in remote sensing images according to claim 2, characterized in that: The predetermined path is: The window starts to slide horizontally from the upper left corner of the remote sensing image until the window slides to the upper right corner of the remote sensing image, then the window slides to the lower left according to the overlap to the left side of the remote sensing image, then the window slides horizontally until the window slides from the left side of the remote sensing image to the right side of the remote sensing image, then the window slides to the lower left according to the overlap to the left side of the remote sensing image, and so on, until the window slides to the lower right corner of the remote sensing image.

5. The method for extracting pond water surface in remote sensing images according to claim 1, characterized in that: Inputting all the image blocks into a deep learning model for inference and processing, and obtaining a vector graphic corresponding to each image block comprises: Inputting all the image blocks into a deep learning model for inference to obtain a mask corresponding to each image block; Vectorization is performed on all the masks to obtain a vector map corresponding to each image block.

6. The method for extracting pond water surface in remote sensing images according to claim 5, characterized in that: The step of performing vector processing on all the masks to obtain a vector map corresponding to each image block comprises: The contour tracking algorithm of OpenCV2 is used to calculate the pixel position of the corner points on each mask corresponding to the mask area in the remote sensing image; According to the pixel position corresponding to each mask, each mask is converted into a corresponding vector map.

7. A device for extracting pond water surface from remote sensing images, characterized in that: include: An image block division module is used to divide the remote sensing image into a plurality of image blocks with overlapping areas; A vector graphics acquisition module, used for inputting all the image blocks into a deep learning model for inference and processing, and obtaining a vector graphic corresponding to each image block; An edge threshold setting module, used to set the edge threshold; A module for obtaining graphics to be fused is used to mark edge water areas on each of the vector graphics according to the edge threshold, and remove the edge water areas to obtain graphics to be fused corresponding to each of the vector graphics; wherein an area extending from the boundary of the image block to the center of the image block and having an extension distance less than the edge threshold is marked as an edge area, and a recognition result involving an edge area is marked as an edge water area; The fusion module is used to fuse all the graphics to be fused, eliminate duplicate parts, obtain the final graphics, and extract the ponds.

8. An electronic device, characterized in that The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for extracting the water surface of a pond in a remote sensing image as described in any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for extracting pond water surface in remote sensing images as described in any one of claims 1 to 6 above are implemented.

Images