Optimization method and device for remote sensing image object fusion and filtering based on overlapping blocks

Abstract

The present invention discloses an optimization method and device for remote sensing image object fusion and filtering based on overlapping blocks, which can improve the accuracy of the prediction results of remote sensing image maps, reduce the degree of loss of detail information in remote sensing image maps, and save storage memory space. The optimization method for remote sensing image object fusion and filtering based on overlapping blocks includes dividing the remote sensing image into multiple groups of windows; using a machine learning model to predict each group of windows to obtain corresponding prediction results; converting the prediction results into corresponding geometric objects; marking the geometric objects corresponding to the boundaries of two of the windows as truncated objects, checking the topological relationship between all geometric objects corresponding to the other two windows and the truncated objects, if it is determined that the truncated object intersects with at least one geometric object in the other two windows and the type is consistent, it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric object are saved in the final result; the final result is used for surface monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing and computer vision, and more specifically, to an optimization method and device for remote sensing image object fusion and filtering based on overlapping blocks. Background Art

[0002] At present, in the field of remote sensing, machine learning models are used to identify and extract a whole scene of remote sensing images for large-scale surface monitoring. Machine learning models usually use image block processing methods, such as sliding window method and pyramid method. Among them, the sliding window method is to divide the large-scale remote sensing image into multiple windows with overlapping areas, and then use the algorithm to process each window to obtain the prediction result of each window. Finally, by merging the prediction results of all windows, the processing result of the entire large-scale remote sensing image is obtained; the pyramid method uses image pyramid technology to construct remote sensing images into a pyramid structure according to different resolutions, in which the higher-level images are obtained by downsampling the lower-level images, and the images of each layer are predicted to obtain the prediction results, and the prediction results are fused or post-processed to obtain the final large-scale remote sensing image processing results.

[0003] Both of these image block processing methods have defects. Specifically, there are obvious feature differences between the center position and edge position (that is, the overlapping area of ​​the current window and other windows) of each window in the sliding window method. The center position can obtain relatively complete global information, while the edge position lacks sufficient global information, which leads to inconsistent or incoherent prediction results of the edge position. The pyramid method requires the use of remote sensing images of different resolutions, which easily causes the loss of detail information in the remote sensing image; and as the number of pyramid layers increases, the pyramid method requires more remote sensing images of different resolutions, thereby increasing memory consumption. Summary of the invention

[0004] The purpose of the embodiment of the present application is to provide an optimization method and device for remote sensing image object fusion and filtering based on overlapping blocks, which can improve the accuracy of the prediction results of remote sensing image maps, reduce the loss of detail information in remote sensing image maps, and save storage memory space. The embodiment of the present application is mainly achieved through the following technical solutions:

[0005] A first aspect of an embodiment of the present application provides an optimization method for remote sensing image object fusion and filtering based on overlapping blocks, comprising:

[0006] Dividing the remote sensing image into a plurality of windows, wherein each two adjacent windows have an overlapping area, and each four of the windows are grouped into one group;

[0007] Use the machine learning model to predict each group of windows and obtain the prediction results corresponding to each group of windows;

[0008] Convert the prediction results corresponding to each group of windows into corresponding geometric objects, and store the geometric objects;

[0009] Mark the geometric object corresponding to the boundary of two windows in each group of windows as a truncated object, check the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, then it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric object are saved in the final result;

[0010] Mark the non-edge-adjacent geometric objects corresponding to each window as untruncated objects, check the topological relationship between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group, and if the number of intersections between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group is greater than 0, save the untruncated objects corresponding to the current window in the final result;

[0011] The final results are used for surface monitoring.

[0012] According to one embodiment of the present application, the overlapping range of the overlapping area is 50%-70% of the window area.

[0013] According to one embodiment of the present application, the multiple windows are distributed in multiple rows and columns, each group of windows includes a first window, a second window, a third window and a fourth window, the first window and the second window are current row windows, the third window and the fourth window are next row windows, the first window and the third window are current column windows, and the second window and the fourth window are next column windows.

[0014] According to one embodiment of the present application, the machine learning model is an image semantic segmentation model.

[0015] According to an embodiment of the present application, the step of marking a geometric object corresponding to a boundary between two windows in each group of windows as a truncated object, checking the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, and if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, then it is considered that the truncated object is correctly identified, and the step of saving the truncated object and the intersecting geometric object in the final result includes:

[0016] Marking the geometric object corresponding to the lower boundary of the first window and the second window as a first truncated object, checking the topological relationship between all geometric objects corresponding to the third window and the fourth window and the first truncated object, if it is determined that the first truncated object intersects with at least one geometric object in the third window and the fourth window respectively and the types are consistent, it is considered that the first truncated object is correctly identified, and the first truncated object and the intersecting geometric objects are saved in the final result;

[0017] Marking the geometric objects corresponding to the left borders of the second window and the fourth window as second truncated objects, checking the topological relationship between all geometric objects corresponding to the first window and the third window and the second truncated objects, if it is determined that the second truncated object intersects with at least one geometric object in the first window and the third window respectively and the types are consistent, it is considered that the second truncated object is correctly identified, and the second truncated object and the intersecting geometric objects are saved in the final result;

[0018] The first truncated object and the second truncated object both belong to the truncated object.

[0019] A second aspect of the embodiment of the present application provides an optimization device for remote sensing image object fusion and filtering based on overlapping blocks, comprising:

[0020] A window division module is used to divide the remote sensing image into a plurality of windows, wherein each two adjacent windows have an overlapping area, and each four windows are grouped into a group;

[0021] A prediction result acquisition module is used to use a machine learning model to predict each group of windows and obtain the prediction results corresponding to each group of windows;

[0022] A conversion module, used to convert the prediction results corresponding to each group of windows into corresponding geometric objects and store the geometric objects;

[0023] A truncated object identification module is used to mark the geometric object corresponding to the boundary of two windows in each group of windows as a truncated object, check the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, and if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric object are saved in the final result;

[0024] The untruncated object identification module is used to mark the non-edge-adjacent geometric objects corresponding to each window as untruncated objects, check the topological relationship between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group, and if the number of intersections between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group is greater than 0, save the untruncated objects corresponding to the current window in the final result;

[0025] The final results are used for surface monitoring.

[0026] According to one embodiment of the present application, the truncated object recognition module includes:

[0027] a first truncated object identification unit, used for marking the geometric object corresponding to the lower boundary of the first window and the second window as a first truncated object, checking the topological relationship between all geometric objects corresponding to the third window and the fourth window and the first truncated object, and if it is determined that the first truncated object intersects with at least one geometric object in the third window and the fourth window respectively and the types are consistent, it is considered that the first truncated object is correctly identified, and the first truncated object and the intersecting geometric objects are saved in the final result;

[0028] a second truncated object identification unit, used for marking the geometric objects corresponding to the left borders of the second window and the fourth window as second truncated objects, checking the topological relationship between all geometric objects corresponding to the first window and the third window and the second truncated object, and if it is determined that the second truncated object intersects with at least one geometric object in the first window and the third window respectively and the types are consistent, it is considered that the second truncated object is correctly identified, and the second truncated object and the intersecting geometric objects are saved in the final result;

[0029] The first truncated object and the second truncated object both belong to the truncated object.

[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 the optimization method for remote sensing image object fusion and filtering based on overlapping blocks as described 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 any of the above-described optimization methods for overlapping block-based remote sensing image object fusion and filtering.

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

[0033] The optimization method for remote sensing image object fusion and filtering based on overlapping blocks provided in the embodiment of the present application is to divide the remote sensing image into multiple windows, each two adjacent windows have overlapping areas, and each four windows are grouped together, and each group of windows is processed collaboratively, so that both the detail information at the boundary of the window and the detail information not attached to the edge of the window can be identified and processed. Therefore, the present application can deeply consider the context information in each window, make the prediction result more complete, effectively avoid the loss of detail information in the window, and improve the accuracy of the prediction result. In addition, the image division method of the present application only requires a remote sensing image of one resolution, and does not require images of additional sizes (i.e., other resolutions), thereby saving storage memory space. 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 1 A flowchart of an optimization method for remote sensing image object fusion and filtering based on overlapping blocks of the present invention in some embodiments;

[0036] Figure 2 A schematic diagram of the structure of multiple windows in some embodiments of the present invention;

[0037] Figure 3 is a schematic diagram of the structure of the current group window in some embodiments of the present invention;

[0038] Figure 4 It is a schematic diagram of the remote sensing image in some embodiments of the present invention;

[0039] Figure 5 It is a schematic diagram of processing geometric objects at the boundary in the present invention;

[0040] Figure 6 It is a schematic diagram of processing geometric objects at the boundary in the present invention;

[0041] Figure 7 Schematic diagram of remote sensing images in other embodiments of the present invention;

[0042] Figure 8 It is a schematic diagram of processing geometric objects that are not aligned with the edges in the present invention;

[0043] Fig. 9It is a schematic diagram of processing geometric objects that are not aligned with the edges in the present invention;

[0044] Fig.10 It is a schematic diagram of the final result in the present invention;

[0045] Fig.11 A flowchart of another embodiment of the optimization method for remote sensing image object fusion and filtering based on overlapping blocks of the present invention;

[0046] Fig.12 It is a structural block diagram of some embodiments of the optimization device for overlapping block-based remote sensing image object fusion and filtering of the present invention. DETAILED DESCRIPTION

[0047] 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.

[0048] In the field of remote sensing, the size of remote sensing images can vary greatly. Low-resolution images have fewer pixels, and the number of pixels on the long side of their remote sensing images may range from hundreds to thousands. For example, a MODIS satellite image has about 2400x2400 pixels. Medium-resolution images have higher spatial resolution and can provide more detailed information about land objects. The number of pixels on the long side of such images is usually between thousands and ten thousand. For example, Landsat-8 images can usually have 7620x7620 pixels. High-resolution images have higher spatial resolution and can provide more detailed information about land objects. The number of pixels on the long side of such images may range from tens of thousands to hundreds of thousands. For example, the typical image size of Wor l dView-3 is 15000x15000.

[0049] Due to the structure and design of the model and the limitations of computing resources, machine learning models usually have certain requirements or restrictions on the size of the input remote sensing images. Common sizes include 256x256, 512x512, and 1024x1024.

[0050] As the spatial resolution of remote sensing or aerial photograph data is getting higher and higher, and the amount of data is getting larger and larger, how to use limited computer resources to efficiently process big data has become an inevitable problem in the process of writing image processing algorithms. In the prior art, the most common processing method is to divide the image into blocks, and according to the computing resources, only a small number of block data are processed each time until all blocks are processed. However, the scale of remote sensing objects is variable, and the machine learning model has limitations on the size of remote sensing images. It is very common for the same object to be truncated in different processing blocks. Therefore, when processing truncated objects, it is impossible to effectively consider the global context information, which affects the accuracy and completeness of the recognition results of remote sensing objects.

[0051] In view of the problems caused by the above-mentioned block division, the commonly used block division methods for large-scale remote sensing object extraction are the sliding window method and the pyramid method. Both methods have corresponding defects, that is, the sliding window method mentioned above is prone to inconsistent or incoherent prediction results of edge positions; the pyramid method is prone to loss of detail information in remote sensing image maps and increases memory consumption. In view of these defects, this application proposes an optimization method and device for remote sensing image object fusion and filtering based on overlapping blocks. The specific implementation methods of this application will be further described in conjunction with the accompanying drawings.

[0052] <Optimization method of remote sensing image object fusion and filtering based on overlapping blocks>

[0053] like Figure 1 FIG. 1 is a flowchart of an optimization method for fusion and filtering of remote sensing image objects based on overlapping blocks provided in the first aspect of an embodiment of the present application. The optimization method for fusion and filtering of remote sensing image objects based on overlapping blocks includes:

[0054] S101, dividing the remote sensing image into a plurality of windows, wherein every two adjacent windows have an overlapping area, and every four of the windows are grouped into one group.

[0055] It should be understood that the multiple windows are distributed in multiple rows and columns, each group of windows includes a first window, a second window, a third window and a fourth window, the first window and the second window are windows in the current row, the third window and the fourth window are windows in the next row, the first window and the third window are windows in the current column, and the second window and the fourth window are windows in the next column. In other implementations, the number of windows in each group is not limited to four, and can be set by those skilled in the art according to actual needs.

[0056] In the embodiment of the present application, the row and column numbers are used to record the current processing window, for example, Figure 2As shown, the processing window in the first row and the first column is recorded as "1-1"; the processing window in the first row and the second column is recorded as "1-2", which is the first window (current window); the processing window in the first row and the third column is recorded as "1-3", which is the second window; the processing window in the second row and the first column is recorded as "2-1"; the processing window in the second row and the second column is recorded as "2-2", which is the third window; the processing window in the second row and the third column is recorded as "2-3", which is the fourth window. The current group window being processed is referred to in Figure 3 shown.

[0057] The size of the window may be 512 x 512 or 1024 x 1024. Since different models and tasks in the field of computer vision may have different image size requirements, the size of the window may be kept consistent with the image size required by the model prediction.

[0058] The overlapping range of the overlapping area is 50%-70% of the area of ​​each window.

[0059] Compared with the pyramid method, the present application does not need to reduce the high-resolution image to a lower resolution, and can fully consider the processing of edge information within the local window without causing loss of detail information.

[0060] S102: Use the machine learning model to predict each group of windows to obtain prediction results corresponding to each group of windows.

[0061] The machine learning models include but are not limited to image semantic segmentation models such as Swin Transformer and deep l abv3.

[0062] The prediction operation described in this application is an image recognition operation, which is to use the trained model to recognize the remote sensing image to extract objects in the remote sensing image. The prediction operation can recognize objects such as buildings, water bodies, bare soil, docks, and ships.

[0063] The prediction result refers to the objects extracted from the remote sensing image, such as buildings, water bodies, bare soil, docks, ships, etc., which are used to generate geometric objects.

[0064] S103: Convert the prediction results corresponding to each group of windows into corresponding geometric objects, and store the geometric objects.

[0065] The process of converting the prediction result into the corresponding geometric object is to extract the contour of the prediction result in the remote sensing image, obtain the coordinates of the contour and save them as object types such as Polygon, Multi Polygon, etc.

[0066] S104. Mark the geometric objects corresponding to the boundaries of two windows in each group of windows as truncated objects, check the topological relationship between all geometric objects corresponding to the other two windows and the truncated objects, and if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the types are consistent, it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric objects are saved in the final result.

[0067] The type being consistent refers to the same prediction results.

[0068] The schematic diagram of processing the geometric objects corresponding to the boundary can refer to Figure 4 , Figure 5 and Figure 6 As shown, Figure 5 and Figure 6 The number 1 represents soil, the number 4 represents buildings, and the number 5 represents water.

[0069] S105. Mark the non-bordering geometric objects corresponding to each window as untruncated objects, check the topological relationship between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group, and if the number of intersections between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group is greater than 0 times (that is, the geometric object exists in at least two windows), save the untruncated object corresponding to the current window in the final result; the final result is used for surface monitoring.

[0070] The non-edge-adjacent geometric objects refer to geometric objects that are not located at the border of the window.

[0071] The schematic diagram of processing the non-edge-fitting geometric object can be referred to Figure 7 , Figure 8 and Fig. 9 As shown, Figure 8 and Fig. 9 The number 1 represents soil, the number 4 represents buildings, and the number 5 represents water.

[0072] In the embodiment of the present application, the number of intersections can be adjusted according to actual needs. When the model recognition effect is poor, the number of intersections can be appropriately increased to improve the precision of the prediction results.

[0073] The schematic diagram of the final result can be referred to Fig.10 shown.

[0074] The optimization method for remote sensing image object fusion and filtering based on overlapping blocks provided in the embodiment of the present application is to divide the remote sensing image into multiple windows, each two adjacent windows have overlapping areas, and each four windows are grouped together, and each group of windows is processed collaboratively, so that both the detail information at the boundary of the window and the detail information not attached to the edge of the window can be identified and processed. Therefore, the present application can deeply consider the context information in each window, make the prediction result more complete, effectively avoid the loss of detail information in the window, and improve the accuracy of the prediction result. In addition, the image division method of the present application only requires a remote sensing image of one resolution, and does not require images of additional sizes (i.e., other resolutions), thereby saving storage memory space.

[0075] In some possible embodiments, such as Fig.11 As shown, the step S104 includes:

[0076] S1041. Mark the geometric object corresponding to the lower boundary of the first window and the second window as the first truncated object, check the topological relationship between all geometric objects corresponding to the third window and the fourth window and the first truncated object, and if it is determined that the first truncated object intersects with at least one geometric object in the third window and the fourth window respectively and the types are consistent, it is considered that the first truncated object is correctly identified, and the first truncated object and the intersecting geometric objects are saved in the final result.

[0077] S1042. Mark the geometric objects corresponding to the left borders of the second window and the fourth window as the second truncated objects, check the topological relationship between all geometric objects corresponding to the first window and the third window and the second truncated object, and if it is determined that the second truncated object intersects with at least one geometric object in the first window and the third window respectively and the types are consistent, it is considered that the second truncated object is correctly identified, and the second truncated object and the intersecting geometric objects are saved in the final result; the first truncated object and the second truncated object both belong to the truncated object.

[0078] In an embodiment of the present application, the remote sensing image is divided into multiple overlapping windows, and multiple geometric objects are filtered in combination with a non-maximum suppression algorithm (that is, an optimization method for remote sensing image object fusion and filtering based on overlapping blocks), so as to achieve more accurate remote sensing object detection.

[0079] In order to verify that the present application can ensure the integrity of the graphics without reducing the precision of the prediction results, the inventors adopted a block-based method, an overlapping block-based method, and an optimization method of remote sensing image object fusion and filtering based on overlapping blocks of the present application to perform accuracy verification of remote sensing object extraction on the same batch of remote sensing images. The specific accuracy verification results are shown in Table 1.

[0080] Table 1

[0081]

[0082]

[0083] From the data analysis in Table 1, it can be seen that the overlapping block-based method improves the recall rate and image integrity, but reduces the precision rate compared to the block-based method. However, the present application combines the non-maximum suppression algorithm and suppresses the inconsistent or incoherent prediction results in the edge area of ​​the image through multi-window collaborative processing, which can improve the image integrity without reducing the precision rate of the prediction results.

[0084] In Table 1, F1 represents an indicator for comprehensively evaluating the performance of the classification model; IoU represents the overlap, which is the standard for the accuracy of detecting corresponding objects.

[0085] It should be understood that the non-maximum suppression algorithm refers to the optimization method for remote sensing image object fusion and filtering based on overlapping blocks.

[0086] <Optimization device for remote sensing image object fusion and filtering based on overlapping blocks>

[0087] The second aspect of the embodiment of the present application provides an optimization device 200 for remote sensing image object fusion and filtering based on overlapping blocks, such as Fig.12 As shown, the optimization device 200 for remote sensing image object fusion and filtering based on overlapping blocks includes:

[0088] The window division module 201 is used to divide the remote sensing image into a plurality of windows, each two adjacent windows have an overlapping area, and each four windows are grouped into a group;

[0089] A prediction result acquisition module 202 is used to use a machine learning model to predict each group of windows and obtain a prediction result corresponding to each group of windows;

[0090] A conversion module 203, used to convert the prediction results corresponding to each group of windows into corresponding geometric objects, and store the geometric objects;

[0091] The truncated object identification module 204 is used to mark the geometric object corresponding to the boundary of two windows in each group of windows as a truncated object, check the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, and if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric object are saved in the final result;

[0092] The untruncated object identification module 205 is used to mark the non-edge-attached geometric objects corresponding to each window as untruncated objects, check the topological relationship between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group, and if the number of intersections between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group is greater than 0, then save the untruncated objects corresponding to the current window in the final result;

[0093] The final results are used for surface monitoring.

[0094] In some feasible embodiments, the truncated object identification module 204 includes:

[0095] a first truncated object identification unit, used for marking the geometric object corresponding to the lower boundary of the first window and the second window as a first truncated object, checking the topological relationship between all geometric objects corresponding to the third window and the fourth window and the first truncated object, and if it is determined that the first truncated object intersects with at least one geometric object in the third window and the fourth window respectively and the types are consistent, it is considered that the first truncated object is correctly identified, and the first truncated object and the intersecting geometric objects are saved in the final result;

[0096] a second truncated object identification unit, used for marking the geometric objects corresponding to the left borders of the second window and the fourth window as second truncated objects, checking the topological relationship between all geometric objects corresponding to the first window and the third window and the second truncated object, and if it is determined that the second truncated object intersects with at least one geometric object in the first window and the third window respectively and the types are consistent, it is considered that the second truncated object is correctly identified, and the second truncated object and the intersecting geometric objects are saved in the final result;

[0097] The first truncated object and the second truncated object both belong to the truncated object.

[0098] <Electronic equipment>

[0099] 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 the optimization method for remote sensing image object fusion and filtering based on overlapping blocks as described above are implemented.

[0100] <Non-transitory computer-readable storage medium>

[0101] 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 any of the above-described optimization methods for overlapping block-based remote sensing image object fusion and filtering.

[0102] 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.

[0103] 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.

[0104] 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.

[0105] 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.

[0106] 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. An optimization method for remote sensing image object fusion and filtering based on overlapping blocks, characterized in that: include: Dividing the remote sensing image into a plurality of windows, wherein each two adjacent windows have an overlapping area, and each four of the windows are grouped into one group; Use a machine learning model to predict each group of windows to obtain a prediction result corresponding to each group of windows, wherein the machine learning model includes a Swin Transformer and a deeplabv3 image semantic segmentation model, and the prediction operation is an image recognition operation, which is to use a trained model to recognize the remote sensing image to extract objects in the remote sensing image. The objects that can be recognized by the prediction operation include buildings, water bodies, bare soil, docks, and ships. The prediction result refers to the objects extracted from the remote sensing image, including buildings, water bodies, bare soil, docks, and ships, to generate geometric objects; Convert the prediction results corresponding to each group of windows into corresponding geometric objects, and store the geometric objects. The process of converting the prediction results into corresponding geometric objects is to extract the contour of the prediction results in the remote sensing image, obtain the coordinates of the contour and save them as polygon or multiple polygon object types; Mark the geometric object corresponding to the boundary of two windows in each group of windows as a truncated object, check the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, then it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric object are saved in the final result; Mark the non-edge-adjacent geometric objects corresponding to each window as untruncated objects, check the topological relationship between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group, and if the number of intersections between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group is greater than 0, save the untruncated objects corresponding to the current window in the final result; The final results are used for surface monitoring.

2. The optimization method for remote sensing image object fusion and filtering based on overlapping blocks according to claim 1 is characterized in that: The overlapping range of the overlapping area is 50%-70% of the window area.

3. The optimization method for remote sensing image object fusion and filtering based on overlapping blocks according to claim 2 is characterized in that: The multiple windows are distributed in multiple rows and columns, each group of windows includes a first window, a second window, a third window and a fourth window, the first window and the second window are current row windows, the third window and the fourth window are next row windows, the first window and the third window are current column windows, and the second window and the fourth window are next column windows.

4. The optimization method for remote sensing image object fusion and filtering based on overlapping blocks according to claim 3 is characterized in that: The step of marking the geometric object corresponding to the boundary of two windows in each group of windows as a truncated object, checking the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, and if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, then it is considered that the truncated object is correctly identified, and the step of saving the truncated object and the intersecting geometric object in the final result includes: Marking the geometric object corresponding to the lower boundary of the first window and the second window as a first truncated object, checking the topological relationship between all geometric objects corresponding to the third window and the fourth window and the first truncated object, if it is determined that the first truncated object intersects with at least one geometric object in the third window and the fourth window respectively and the types are consistent, it is considered that the first truncated object is correctly identified, and the first truncated object and the intersecting geometric objects are saved in the final result; Marking the geometric objects corresponding to the left borders of the second window and the fourth window as second truncated objects, checking the topological relationship between all geometric objects corresponding to the first window and the third window and the second truncated objects, if it is determined that the second truncated object intersects with at least one geometric object in the first window and the third window respectively and the types are consistent, it is considered that the second truncated object is correctly identified, and the second truncated object and the intersecting geometric objects are saved in the final result; The first truncated object and the second truncated object both belong to the truncated object.

5. An optimization device for remote sensing image object fusion and filtering based on overlapping blocks, characterized in that: include: A window division module is used to divide the remote sensing image into a plurality of windows, wherein each two adjacent windows have an overlapping area, and each four windows are grouped into a group; A prediction result acquisition module is used to use a machine learning model to predict each group of windows and obtain the prediction results corresponding to each group of windows. The machine learning model includes Swin Transformer and deeplabv3 image semantic segmentation model. The prediction operation is an image recognition operation, which is to use the trained model to recognize the remote sensing image to extract objects in the remote sensing image. The objects that can be recognized by the prediction operation include buildings, water bodies, bare soil, docks, and ships. The prediction results refer to the objects extracted from the remote sensing image, including buildings, water bodies, bare soil, docks, and ships, which are used to generate geometric objects. A conversion module, used for converting the prediction results corresponding to each group of windows into corresponding geometric objects and storing the geometric objects. The process of converting the prediction results into corresponding geometric objects is to extract the contour of the prediction results in the remote sensing image, obtain the coordinates of the contour and save them as polygon or multiple polygon object types; A truncated object identification module is used to mark the geometric object corresponding to the boundary of two windows in each group of windows as a truncated object, check the topological relationship between all geometric objects corresponding to the other two windows and the truncated object, and if it is determined that the truncated object intersects with at least one geometric object in the other two windows at the same time and the type is consistent, it is considered that the truncated object is correctly identified, and the truncated object and the intersecting geometric object are saved in the final result; The untruncated object identification module is used to mark the non-edge-adjacent geometric objects corresponding to each window as untruncated objects, check the topological relationship between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group, and if the number of intersections between the untruncated objects corresponding to the current window and the untruncated objects corresponding to other windows in the same group is greater than 0, save the untruncated objects corresponding to the current window in the final result; The final results are used for surface monitoring.

6. The optimization device for remote sensing image object fusion and filtering based on overlapping blocks according to claim 5, characterized in that: The multiple windows are distributed in multiple rows and columns, each group of windows includes a first window, a second window, a third window and a fourth window, the first window and the second window are windows in the current row, the third window and the fourth window are windows in the next row, the first window and the third window are windows in the current column, and the second window and the fourth window are windows in the next column, The truncated object recognition module comprises: a first truncated object identification unit, used to mark the geometric object corresponding to the lower boundary of the first window and the second window as the first truncated object, check the topological relationship between all geometric objects corresponding to the third window and the fourth window and the first truncated object, if it is determined that the first truncated object intersects with at least one geometric object in the third window and the fourth window respectively and the types are consistent, it is considered that the first truncated object is correctly identified, and the first truncated object and the intersecting geometric objects are saved in the final result; a second truncated object identification unit, used for marking the geometric objects corresponding to the left borders of the second window and the fourth window as second truncated objects, checking the topological relationship between all geometric objects corresponding to the first window and the third window and the second truncated object, and if it is determined that the second truncated object intersects with at least one geometric object in the first window and the third window respectively and the types are consistent, it is considered that the second truncated object is correctly identified, and the second truncated object and the intersecting geometric objects are saved in the final result; The first truncated object and the second truncated object both belong to the truncated object.

7. 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. When the processor executes the computer program, the steps of the optimization method for overlapping block-based remote sensing image object fusion and filtering as described in any one of claims 1 to 4 are implemented.

8. 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 optimization method for overlapping block-based remote sensing image object fusion and filtering as described in any one of claims 1 to 4 are implemented.

Images